tmakin/ack-tegra
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux into drm-next

Browse Source 
amdgpu and radeon changes for 4.3. Highlights:
- Fiji support for amdgpu.
- CGS support for amdgpu.  This is a new driver
  internal cross-component API.
- Initial GPU scheduler for amdgpu.  Still disabled
  by default.
- Lots of bug fixes and optimizations

* 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux: (130 commits)
  drm/amdgpu: wait on page directory changes. v2
  drm/amdgpu: Select BACKLIGHT_LCD_SUPPORT
  drm/radeon: Select BACKLIGHT_LCD_SUPPORT
  drm/amdgpu: cleanup sheduler rq handling v2
  drm/amdgpu: move prepare work out of scheduler to cs_ioctl
  drm/amdgpu: fix unnecessary wake up
  drm/amdgpu: fix duplicated mapping invoke bug
  drm/amdgpu: drop bo_list_clone when no scheduler
  drm/amdgpu: disable GPU reset by default
  drm/amdgpu: fix type mismatch error
  drm/amdgpu: add reference for **fence
  drm/amdgpu: fix waiting for all fences before flipping
  drm/amdgpu: fix UVD return code checking
  drm/amdgpu: remove scheduler fence list v2
  drm/amdgpu: remove amd_sched_wait_emit v2
  drm/amdgpu: remove unecessary scheduler fence callbacks
  drm/amdgpu: fix scheduler fence implementation
  drm/amdgpu: don't grab dev->struct_mutex in pm functions
  drm/amdgpu: Don't take dev->struct_mutex in bo_force_delete
  drm/radeon: Don't take dev->struct_mutex in pm functions
  ...
This commit is contained in:
Dave Airlie
2015-08-20 09:40:49 +10:00
parent 294947a5c7 05906dec7d
commit e2a8986f3e
74 changed files with 14135 additions and 1156 deletions
Download Patch File Download Diff File Expand all files Collapse all files
drivers/gpu/drm/Kconfig
+2
View File
@@ -128,6 +128,7 @@ config DRM_RADEON
select POWER_SUPPLY
select HWMON
select BACKLIGHT_CLASS_DEVICE
select BACKLIGHT_LCD_SUPPORT
select INTERVAL_TREE
help
Choose this option if you have an ATI Radeon graphics card. There
@@ -151,6 +152,7 @@ config DRM_AMDGPU
select POWER_SUPPLY
select HWMON
select BACKLIGHT_CLASS_DEVICE
select BACKLIGHT_LCD_SUPPORT
select INTERVAL_TREE
help
Choose this option if you have a recent AMD Radeon graphics card.
drivers/gpu/drm/amd/amdgpu/Makefile
+15 -3
View File
@@ -3,7 +3,9 @@
# Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/asic_reg \
-Idrivers/gpu/drm/amd/include
-Idrivers/gpu/drm/amd/include \
-Idrivers/gpu/drm/amd/amdgpu \
-Idrivers/gpu/drm/amd/scheduler
amdgpu-y := amdgpu_drv.o
@@ -21,7 +23,8 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
# add asic specific block
amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o gmc_v7_0.o cik_ih.o kv_smc.o kv_dpm.o \
ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o
ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o \
amdgpu_amdkfd_gfx_v7.o
amdgpu-y += \
vi.o
@@ -43,6 +46,7 @@ amdgpu-y += \
amdgpu_dpm.o \
cz_smc.o cz_dpm.o \
tonga_smc.o tonga_dpm.o \
fiji_smc.o fiji_dpm.o \
iceland_smc.o iceland_dpm.o
# add DCE block
@@ -74,9 +78,17 @@ amdgpu-y += \
# add amdkfd interfaces
amdgpu-y += \
amdgpu_amdkfd.o \
amdgpu_amdkfd_gfx_v7.o \
amdgpu_amdkfd_gfx_v8.o
# add cgs
amdgpu-y += amdgpu_cgs.o
# GPU scheduler
amdgpu-y += \
../scheduler/gpu_scheduler.o \
../scheduler/sched_fence.o \
amdgpu_sched.o
amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o
amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o
amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o
drivers/gpu/drm/amd/amdgpu/amdgpu.h
+107 -46
View File
@@ -42,17 +42,19 @@
#include <ttm/ttm_module.h>
#include <ttm/ttm_execbuf_util.h>
#include <drm/drmP.h>
#include <drm/drm_gem.h>
#include <drm/amdgpu_drm.h>
#include "amd_shared.h"
#include "amdgpu_family.h"
#include "amdgpu_mode.h"
#include "amdgpu_ih.h"
#include "amdgpu_irq.h"
#include "amdgpu_ucode.h"
#include "amdgpu_gds.h"
#include "gpu_scheduler.h"
/*
* Modules parameters.
*/
@@ -77,7 +79,11 @@ extern int amdgpu_bapm;
extern int amdgpu_deep_color;
extern int amdgpu_vm_size;
extern int amdgpu_vm_block_size;
extern int amdgpu_enable_scheduler;
extern int amdgpu_sched_jobs;
extern int amdgpu_sched_hw_submission;
#define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
#define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
#define AMDGPU_FENCE_JIFFIES_TIMEOUT (HZ / 2)
/* AMDGPU_IB_POOL_SIZE must be a power of 2 */
@@ -178,6 +184,7 @@ struct amdgpu_ring;
struct amdgpu_semaphore;
struct amdgpu_cs_parser;
struct amdgpu_irq_src;
struct amdgpu_fpriv;
enum amdgpu_cp_irq {
AMDGPU_CP_IRQ_GFX_EOP = 0,
@@ -381,10 +388,10 @@ struct amdgpu_fence_driver {
uint64_t sync_seq[AMDGPU_MAX_RINGS];
atomic64_t last_seq;
bool initialized;
bool delayed_irq;
struct amdgpu_irq_src *irq_src;
unsigned irq_type;
struct delayed_work lockup_work;
wait_queue_head_t fence_queue;
};
/* some special values for the owner field */
@@ -423,20 +430,18 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring);
int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
struct amdgpu_irq_src *irq_src,
unsigned irq_type);
void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
struct amdgpu_fence **fence);
int amdgpu_fence_recreate(struct amdgpu_ring *ring, void *owner,
uint64_t seq, struct amdgpu_fence **fence);
void amdgpu_fence_process(struct amdgpu_ring *ring);
int amdgpu_fence_wait_next(struct amdgpu_ring *ring);
int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
bool amdgpu_fence_signaled(struct amdgpu_fence *fence);
int amdgpu_fence_wait(struct amdgpu_fence *fence, bool interruptible);
int amdgpu_fence_wait_any(struct amdgpu_device *adev,
signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
struct amdgpu_fence **fences,
bool intr);
bool intr, long t);
struct amdgpu_fence *amdgpu_fence_ref(struct amdgpu_fence *fence);
void amdgpu_fence_unref(struct amdgpu_fence **fence);
@@ -481,7 +486,7 @@ static inline bool amdgpu_fence_is_earlier(struct amdgpu_fence *a,
return a->seq < b->seq;
}
int amdgpu_user_fence_emit(struct amdgpu_ring *ring, struct amdgpu_user_fence *user,
int amdgpu_user_fence_emit(struct amdgpu_ring *ring, struct amdgpu_user_fence *user,
void *owner, struct amdgpu_fence **fence);
/*
@@ -532,14 +537,16 @@ struct amdgpu_bo_va_mapping {
struct amdgpu_bo_va {
/* protected by bo being reserved */
struct list_head bo_list;
uint64_t addr;
struct amdgpu_fence *last_pt_update;
struct fence *last_pt_update;
unsigned ref_count;
/* protected by vm mutex */
struct list_head mappings;
/* protected by vm mutex and spinlock */
struct list_head vm_status;
/* mappings for this bo_va */
struct list_head invalids;
struct list_head valids;
/* constant after initialization */
struct amdgpu_vm *vm;
struct amdgpu_bo *bo;
@@ -697,8 +704,8 @@ struct amdgpu_sync {
};
void amdgpu_sync_create(struct amdgpu_sync *sync);
void amdgpu_sync_fence(struct amdgpu_sync *sync,
struct amdgpu_fence *fence);
int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
struct fence *f);
int amdgpu_sync_resv(struct amdgpu_device *adev,
struct amdgpu_sync *sync,
struct reservation_object *resv,
@@ -821,7 +828,9 @@ struct amdgpu_flip_work {
uint64_t base;
struct drm_pending_vblank_event *event;
struct amdgpu_bo *old_rbo;
struct fence *fence;
struct fence *excl;
unsigned shared_count;
struct fence **shared;
};
@@ -844,6 +853,8 @@ struct amdgpu_ib {
uint32_t gws_base, gws_size;
uint32_t oa_base, oa_size;
uint32_t flags;
/* resulting sequence number */
uint64_t sequence;
};
enum amdgpu_ring_type {
@@ -854,11 +865,23 @@ enum amdgpu_ring_type {
AMDGPU_RING_TYPE_VCE
};
extern struct amd_sched_backend_ops amdgpu_sched_ops;
int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
struct amdgpu_ring *ring,
struct amdgpu_ib *ibs,
unsigned num_ibs,
int (*free_job)(struct amdgpu_cs_parser *),
void *owner,
struct fence **fence);
struct amdgpu_ring {
struct amdgpu_device *adev;
const struct amdgpu_ring_funcs *funcs;
struct amdgpu_fence_driver fence_drv;
struct amd_gpu_scheduler *scheduler;
spinlock_t fence_lock;
struct mutex *ring_lock;
struct amdgpu_bo *ring_obj;
volatile uint32_t *ring;
@@ -892,6 +915,7 @@ struct amdgpu_ring {
struct amdgpu_ctx *current_ctx;
enum amdgpu_ring_type type;
char name[16];
bool is_pte_ring;
};
/*
@@ -943,18 +967,22 @@ struct amdgpu_vm {
struct rb_root va;
/* protecting invalidated and freed */
/* protecting invalidated */
spinlock_t status_lock;
/* BOs moved, but not yet updated in the PT */
struct list_head invalidated;
/* BOs freed, but not yet updated in the PT */
/* BOs cleared in the PT because of a move */
struct list_head cleared;
/* BO mappings freed, but not yet updated in the PT */
struct list_head freed;
/* contains the page directory */
struct amdgpu_bo *page_directory;
unsigned max_pde_used;
struct fence *page_directory_fence;
/* array of page tables, one for each page directory entry */
struct amdgpu_vm_pt *page_tables;
@@ -983,27 +1011,47 @@ struct amdgpu_vm_manager {
* context related structures
*/
struct amdgpu_ctx_state {
uint64_t flags;
uint32_t hangs;
#define AMDGPU_CTX_MAX_CS_PENDING 16
struct amdgpu_ctx_ring {
uint64_t sequence;
struct fence *fences[AMDGPU_CTX_MAX_CS_PENDING];
struct amd_sched_entity entity;
};
struct amdgpu_ctx {
/* call kref_get()before CS start and kref_put() after CS fence signaled */
struct kref refcount;
struct amdgpu_fpriv *fpriv;
struct amdgpu_ctx_state state;
uint32_t id;
unsigned reset_counter;
struct kref refcount;
struct amdgpu_device *adev;
unsigned reset_counter;
spinlock_t ring_lock;
struct amdgpu_ctx_ring rings[AMDGPU_MAX_RINGS];
};
struct amdgpu_ctx_mgr {
struct amdgpu_device *adev;
struct idr ctx_handles;
/* lock for IDR system */
struct mutex lock;
struct amdgpu_device *adev;
struct mutex lock;
/* protected by lock */
struct idr ctx_handles;
};
int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel,
struct amdgpu_ctx *ctx);
void amdgpu_ctx_fini(struct amdgpu_ctx *ctx);
struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
struct fence *fence, uint64_t queued_seq);
struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
struct amdgpu_ring *ring, uint64_t seq);
int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp);
void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
/*
* file private structure
*/
@@ -1012,7 +1060,7 @@ struct amdgpu_fpriv {
struct amdgpu_vm vm;
struct mutex bo_list_lock;
struct idr bo_list_handles;
struct amdgpu_ctx_mgr ctx_mgr;
struct amdgpu_ctx_mgr ctx_mgr;
};
/*
@@ -1029,6 +1077,8 @@ struct amdgpu_bo_list {
struct amdgpu_bo_list_entry *array;
};
struct amdgpu_bo_list *
amdgpu_bo_list_clone(struct amdgpu_bo_list *list);
struct amdgpu_bo_list *
amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id);
void amdgpu_bo_list_put(struct amdgpu_bo_list *list);
@@ -1205,6 +1255,14 @@ struct amdgpu_cs_parser {
/* user fence */
struct amdgpu_user_fence uf;
struct amdgpu_ring *ring;
struct mutex job_lock;
struct work_struct job_work;
int (*prepare_job)(struct amdgpu_cs_parser *sched_job);
int (*run_job)(struct amdgpu_cs_parser *sched_job);
int (*free_job)(struct amdgpu_cs_parser *sched_job);
struct amd_sched_fence *s_fence;
};
static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p, uint32_t ib_idx, int idx)
@@ -1849,17 +1907,12 @@ struct amdgpu_atcs {
struct amdgpu_atcs_functions functions;
};
int amdgpu_ctx_alloc(struct amdgpu_device *adev,struct amdgpu_fpriv *fpriv,
uint32_t *id,uint32_t flags);
int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv,
uint32_t id);
/*
* CGS
*/
void *amdgpu_cgs_create_device(struct amdgpu_device *adev);
void amdgpu_cgs_destroy_device(void *cgs_device);
void amdgpu_ctx_fini(struct amdgpu_fpriv *fpriv);
struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
extern int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp);
/*
* Core structure, functions and helpers.
@@ -1883,7 +1936,7 @@ struct amdgpu_device {
struct rw_semaphore exclusive_lock;
/* ASIC */
enum amdgpu_asic_type asic_type;
enum amd_asic_type asic_type;
uint32_t family;
uint32_t rev_id;
uint32_t external_rev_id;
@@ -1976,7 +2029,6 @@ struct amdgpu_device {
struct amdgpu_irq_src hpd_irq;
/* rings */
wait_queue_head_t fence_queue;
unsigned fence_context;
struct mutex ring_lock;
unsigned num_rings;
@@ -2028,6 +2080,9 @@ struct amdgpu_device {
/* amdkfd interface */
struct kfd_dev *kfd;
/* kernel conext for IB submission */
struct amdgpu_ctx kernel_ctx;
};
bool amdgpu_device_is_px(struct drm_device *dev);
@@ -2215,6 +2270,12 @@ void amdgpu_pci_config_reset(struct amdgpu_device *adev);
bool amdgpu_card_posted(struct amdgpu_device *adev);
void amdgpu_update_display_priority(struct amdgpu_device *adev);
bool amdgpu_boot_test_post_card(struct amdgpu_device *adev);
struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
struct drm_file *filp,
struct amdgpu_ctx *ctx,
struct amdgpu_ib *ibs,
uint32_t num_ibs);
int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data);
int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
u32 ip_instance, u32 ring,
@@ -2278,8 +2339,8 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
struct amdgpu_vm *vm,
struct list_head *head);
struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
struct amdgpu_vm *vm);
int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
struct amdgpu_sync *sync);
void amdgpu_vm_flush(struct amdgpu_ring *ring,
struct amdgpu_vm *vm,
struct amdgpu_fence *updates);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+3 -1
View File
@@ -21,7 +21,7 @@
*/
#include "amdgpu_amdkfd.h"
#include "amdgpu_family.h"
#include "amd_shared.h"
#include <drm/drmP.h>
#include "amdgpu.h"
#include <linux/module.h>
@@ -50,9 +50,11 @@ bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev)
#endif
switch (rdev->asic_type) {
#ifdef CONFIG_DRM_AMDGPU_CIK
case CHIP_KAVERI:
kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions();
break;
#endif
case CHIP_CARRIZO:
kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
break;
drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c
+2 -2
View File
@@ -897,7 +897,7 @@ bool amdgpu_atombios_get_asic_ss_info(struct amdgpu_device *adev,
if ((id == ASIC_INTERNAL_ENGINE_SS) ||
(id == ASIC_INTERNAL_MEMORY_SS))
ss->rate /= 100;
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
amdgpu_atombios_get_igp_ss_overrides(adev, ss, id);
return true;
}
@@ -1058,7 +1058,7 @@ void amdgpu_atombios_set_memory_clock(struct amdgpu_device *adev,
SET_MEMORY_CLOCK_PS_ALLOCATION args;
int index = GetIndexIntoMasterTable(COMMAND, SetMemoryClock);
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return;
args.ulTargetMemoryClock = cpu_to_le32(mem_clock); /* 10 khz */
drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
+1 -1
View File
@@ -42,7 +42,7 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence);
if (r)
goto exit_do_move;
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r)
goto exit_do_move;
amdgpu_fence_unref(&fence);
drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c
+3 -3
View File
@@ -48,7 +48,7 @@ static bool igp_read_bios_from_vram(struct amdgpu_device *adev)
resource_size_t vram_base;
resource_size_t size = 256 * 1024; /* ??? */
if (!(adev->flags & AMDGPU_IS_APU))
if (!(adev->flags & AMD_IS_APU))
if (!amdgpu_card_posted(adev))
return false;
@@ -184,7 +184,7 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
bool found = false;
/* ATRM is for the discrete card only */
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return false;
while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
@@ -246,7 +246,7 @@ static inline bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
static bool amdgpu_read_disabled_bios(struct amdgpu_device *adev)
{
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return igp_read_bios_from_vram(adev);
else
return amdgpu_asic_read_disabled_bios(adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+33
View File
@@ -62,6 +62,39 @@ static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv,
return 0;
}
struct amdgpu_bo_list *
amdgpu_bo_list_clone(struct amdgpu_bo_list *list)
{
struct amdgpu_bo_list *result;
unsigned i;
result = kmalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL);
if (!result)
return NULL;
result->array = drm_calloc_large(list->num_entries,
sizeof(struct amdgpu_bo_list_entry));
if (!result->array) {
kfree(result);
return NULL;
}
mutex_init(&result->lock);
result->gds_obj = list->gds_obj;
result->gws_obj = list->gws_obj;
result->oa_obj = list->oa_obj;
result->has_userptr = list->has_userptr;
result->num_entries = list->num_entries;
memcpy(result->array, list->array, list->num_entries *
sizeof(struct amdgpu_bo_list_entry));
for (i = 0; i < result->num_entries; ++i)
amdgpu_bo_ref(result->array[i].robj);
return result;
}
static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
{
struct amdgpu_bo_list *list;
drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
+838
View File
@@ -0,0 +1,838 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#include <linux/list.h>
#include <linux/slab.h>
#include <linux/pci.h>
#include <drm/drmP.h>
#include <linux/firmware.h>
#include <drm/amdgpu_drm.h>
#include "amdgpu.h"
#include "cgs_linux.h"
#include "atom.h"
#include "amdgpu_ucode.h"
struct amdgpu_cgs_device {
struct cgs_device base;
struct amdgpu_device *adev;
};
#define CGS_FUNC_ADEV \
struct amdgpu_device *adev = \
((struct amdgpu_cgs_device *)cgs_device)->adev
static int amdgpu_cgs_gpu_mem_info(void *cgs_device, enum cgs_gpu_mem_type type,
uint64_t *mc_start, uint64_t *mc_size,
uint64_t *mem_size)
{
CGS_FUNC_ADEV;
switch(type) {
case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__VISIBLE_FB:
*mc_start = 0;
*mc_size = adev->mc.visible_vram_size;
*mem_size = adev->mc.visible_vram_size - adev->vram_pin_size;
break;
case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
*mc_start = adev->mc.visible_vram_size;
*mc_size = adev->mc.real_vram_size - adev->mc.visible_vram_size;
*mem_size = *mc_size;
break;
case CGS_GPU_MEM_TYPE__GART_CACHEABLE:
case CGS_GPU_MEM_TYPE__GART_WRITECOMBINE:
*mc_start = adev->mc.gtt_start;
*mc_size = adev->mc.gtt_size;
*mem_size = adev->mc.gtt_size - adev->gart_pin_size;
break;
default:
return -EINVAL;
}
return 0;
}
static int amdgpu_cgs_gmap_kmem(void *cgs_device, void *kmem,
uint64_t size,
uint64_t min_offset, uint64_t max_offset,
cgs_handle_t *kmem_handle, uint64_t *mcaddr)
{
CGS_FUNC_ADEV;
int ret;
struct amdgpu_bo *bo;
struct page *kmem_page = vmalloc_to_page(kmem);
int npages = ALIGN(size, PAGE_SIZE) >> PAGE_SHIFT;
struct sg_table *sg = drm_prime_pages_to_sg(&kmem_page, npages);
ret = amdgpu_bo_create(adev, size, PAGE_SIZE, false,
AMDGPU_GEM_DOMAIN_GTT, 0, sg, &bo);
if (ret)
return ret;
ret = amdgpu_bo_reserve(bo, false);
if (unlikely(ret != 0))
return ret;
/* pin buffer into GTT */
ret = amdgpu_bo_pin_restricted(bo, AMDGPU_GEM_DOMAIN_GTT,
min_offset, max_offset, mcaddr);
amdgpu_bo_unreserve(bo);
*kmem_handle = (cgs_handle_t)bo;
return ret;
}
static int amdgpu_cgs_gunmap_kmem(void *cgs_device, cgs_handle_t kmem_handle)
{
struct amdgpu_bo *obj = (struct amdgpu_bo *)kmem_handle;
if (obj) {
int r = amdgpu_bo_reserve(obj, false);
if (likely(r == 0)) {
amdgpu_bo_unpin(obj);
amdgpu_bo_unreserve(obj);
}
amdgpu_bo_unref(&obj);
}
return 0;
}
static int amdgpu_cgs_alloc_gpu_mem(void *cgs_device,
enum cgs_gpu_mem_type type,
uint64_t size, uint64_t align,
uint64_t min_offset, uint64_t max_offset,
cgs_handle_t *handle)
{
CGS_FUNC_ADEV;
uint16_t flags = 0;
int ret = 0;
uint32_t domain = 0;
struct amdgpu_bo *obj;
struct ttm_placement placement;
struct ttm_place place;
if (min_offset > max_offset) {
BUG_ON(1);
return -EINVAL;
}
/* fail if the alignment is not a power of 2 */
if (((align != 1) && (align & (align - 1)))
|| size == 0 || align == 0)
return -EINVAL;
switch(type) {
case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__VISIBLE_FB:
flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
domain = AMDGPU_GEM_DOMAIN_VRAM;
if (max_offset > adev->mc.real_vram_size)
return -EINVAL;
place.fpfn = min_offset >> PAGE_SHIFT;
place.lpfn = max_offset >> PAGE_SHIFT;
place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
TTM_PL_FLAG_VRAM;
break;
case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
domain = AMDGPU_GEM_DOMAIN_VRAM;
if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
place.fpfn =
max(min_offset, adev->mc.visible_vram_size) >> PAGE_SHIFT;
place.lpfn =
min(max_offset, adev->mc.real_vram_size) >> PAGE_SHIFT;
place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
TTM_PL_FLAG_VRAM;
}
break;
case CGS_GPU_MEM_TYPE__GART_CACHEABLE:
domain = AMDGPU_GEM_DOMAIN_GTT;
place.fpfn = min_offset >> PAGE_SHIFT;
place.lpfn = max_offset >> PAGE_SHIFT;
place.flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
break;
case CGS_GPU_MEM_TYPE__GART_WRITECOMBINE:
flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
domain = AMDGPU_GEM_DOMAIN_GTT;
place.fpfn = min_offset >> PAGE_SHIFT;
place.lpfn = max_offset >> PAGE_SHIFT;
place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT |
TTM_PL_FLAG_UNCACHED;
break;
default:
return -EINVAL;
}
*handle = 0;
placement.placement = &place;
placement.num_placement = 1;
placement.busy_placement = &place;
placement.num_busy_placement = 1;
ret = amdgpu_bo_create_restricted(adev, size, PAGE_SIZE,
true, domain, flags,
NULL, &placement, &obj);
if (ret) {
DRM_ERROR("(%d) bo create failed\n", ret);
return ret;
}
*handle = (cgs_handle_t)obj;
return ret;
}
static int amdgpu_cgs_import_gpu_mem(void *cgs_device, int dmabuf_fd,
cgs_handle_t *handle)
{
CGS_FUNC_ADEV;
int r;
uint32_t dma_handle;
struct drm_gem_object *obj;
struct amdgpu_bo *bo;
struct drm_device *dev = adev->ddev;
struct drm_file *file_priv = NULL, *priv;
mutex_lock(&dev->struct_mutex);
list_for_each_entry(priv, &dev->filelist, lhead) {
rcu_read_lock();
if (priv->pid == get_pid(task_pid(current)))
file_priv = priv;
rcu_read_unlock();
if (file_priv)
break;
}
mutex_unlock(&dev->struct_mutex);
r = dev->driver->prime_fd_to_handle(dev,
file_priv, dmabuf_fd,
&dma_handle);
spin_lock(&file_priv->table_lock);
/* Check if we currently have a reference on the object */
obj = idr_find(&file_priv->object_idr, dma_handle);
if (obj == NULL) {
spin_unlock(&file_priv->table_lock);
return -EINVAL;
}
spin_unlock(&file_priv->table_lock);
bo = gem_to_amdgpu_bo(obj);
*handle = (cgs_handle_t)bo;
return 0;
}
static int amdgpu_cgs_free_gpu_mem(void *cgs_device, cgs_handle_t handle)
{
struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
if (obj) {
int r = amdgpu_bo_reserve(obj, false);
if (likely(r == 0)) {
amdgpu_bo_kunmap(obj);
amdgpu_bo_unpin(obj);
amdgpu_bo_unreserve(obj);
}
amdgpu_bo_unref(&obj);
}
return 0;
}
static int amdgpu_cgs_gmap_gpu_mem(void *cgs_device, cgs_handle_t handle,
uint64_t *mcaddr)
{
int r;
u64 min_offset, max_offset;
struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
WARN_ON_ONCE(obj->placement.num_placement > 1);
min_offset = obj->placements[0].fpfn << PAGE_SHIFT;
max_offset = obj->placements[0].lpfn << PAGE_SHIFT;
r = amdgpu_bo_reserve(obj, false);
if (unlikely(r != 0))
return r;
r = amdgpu_bo_pin_restricted(obj, AMDGPU_GEM_DOMAIN_GTT,
min_offset, max_offset, mcaddr);
amdgpu_bo_unreserve(obj);
return r;
}
static int amdgpu_cgs_gunmap_gpu_mem(void *cgs_device, cgs_handle_t handle)
{
int r;
struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
r = amdgpu_bo_reserve(obj, false);
if (unlikely(r != 0))
return r;
r = amdgpu_bo_unpin(obj);
amdgpu_bo_unreserve(obj);
return r;
}
static int amdgpu_cgs_kmap_gpu_mem(void *cgs_device, cgs_handle_t handle,
void **map)
{
int r;
struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
r = amdgpu_bo_reserve(obj, false);
if (unlikely(r != 0))
return r;
r = amdgpu_bo_kmap(obj, map);
amdgpu_bo_unreserve(obj);
return r;
}
static int amdgpu_cgs_kunmap_gpu_mem(void *cgs_device, cgs_handle_t handle)
{
int r;
struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
r = amdgpu_bo_reserve(obj, false);
if (unlikely(r != 0))
return r;
amdgpu_bo_kunmap(obj);
amdgpu_bo_unreserve(obj);
return r;
}
static uint32_t amdgpu_cgs_read_register(void *cgs_device, unsigned offset)
{
CGS_FUNC_ADEV;
return RREG32(offset);
}
static void amdgpu_cgs_write_register(void *cgs_device, unsigned offset,
uint32_t value)
{
CGS_FUNC_ADEV;
WREG32(offset, value);
}
static uint32_t amdgpu_cgs_read_ind_register(void *cgs_device,
enum cgs_ind_reg space,
unsigned index)
{
CGS_FUNC_ADEV;
switch (space) {
case CGS_IND_REG__MMIO:
return RREG32_IDX(index);
case CGS_IND_REG__PCIE:
return RREG32_PCIE(index);
case CGS_IND_REG__SMC:
return RREG32_SMC(index);
case CGS_IND_REG__UVD_CTX:
return RREG32_UVD_CTX(index);
case CGS_IND_REG__DIDT:
return RREG32_DIDT(index);
case CGS_IND_REG__AUDIO_ENDPT:
DRM_ERROR("audio endpt register access not implemented.\n");
return 0;
}
WARN(1, "Invalid indirect register space");
return 0;
}
static void amdgpu_cgs_write_ind_register(void *cgs_device,
enum cgs_ind_reg space,
unsigned index, uint32_t value)
{
CGS_FUNC_ADEV;
switch (space) {
case CGS_IND_REG__MMIO:
return WREG32_IDX(index, value);
case CGS_IND_REG__PCIE:
return WREG32_PCIE(index, value);
case CGS_IND_REG__SMC:
return WREG32_SMC(index, value);
case CGS_IND_REG__UVD_CTX:
return WREG32_UVD_CTX(index, value);
case CGS_IND_REG__DIDT:
return WREG32_DIDT(index, value);
case CGS_IND_REG__AUDIO_ENDPT:
DRM_ERROR("audio endpt register access not implemented.\n");
return;
}
WARN(1, "Invalid indirect register space");
}
static uint8_t amdgpu_cgs_read_pci_config_byte(void *cgs_device, unsigned addr)
{
CGS_FUNC_ADEV;
uint8_t val;
int ret = pci_read_config_byte(adev->pdev, addr, &val);
if (WARN(ret, "pci_read_config_byte error"))
return 0;
return val;
}
static uint16_t amdgpu_cgs_read_pci_config_word(void *cgs_device, unsigned addr)
{
CGS_FUNC_ADEV;
uint16_t val;
int ret = pci_read_config_word(adev->pdev, addr, &val);
if (WARN(ret, "pci_read_config_word error"))
return 0;
return val;
}
static uint32_t amdgpu_cgs_read_pci_config_dword(void *cgs_device,
unsigned addr)
{
CGS_FUNC_ADEV;
uint32_t val;
int ret = pci_read_config_dword(adev->pdev, addr, &val);
if (WARN(ret, "pci_read_config_dword error"))
return 0;
return val;
}
static void amdgpu_cgs_write_pci_config_byte(void *cgs_device, unsigned addr,
uint8_t value)
{
CGS_FUNC_ADEV;
int ret = pci_write_config_byte(adev->pdev, addr, value);
WARN(ret, "pci_write_config_byte error");
}
static void amdgpu_cgs_write_pci_config_word(void *cgs_device, unsigned addr,
uint16_t value)
{
CGS_FUNC_ADEV;
int ret = pci_write_config_word(adev->pdev, addr, value);
WARN(ret, "pci_write_config_word error");
}
static void amdgpu_cgs_write_pci_config_dword(void *cgs_device, unsigned addr,
uint32_t value)
{
CGS_FUNC_ADEV;
int ret = pci_write_config_dword(adev->pdev, addr, value);
WARN(ret, "pci_write_config_dword error");
}
static const void *amdgpu_cgs_atom_get_data_table(void *cgs_device,
unsigned table, uint16_t *size,
uint8_t *frev, uint8_t *crev)
{
CGS_FUNC_ADEV;
uint16_t data_start;
if (amdgpu_atom_parse_data_header(
adev->mode_info.atom_context, table, size,
frev, crev, &data_start))
return (uint8_t*)adev->mode_info.atom_context->bios +
data_start;
return NULL;
}
static int amdgpu_cgs_atom_get_cmd_table_revs(void *cgs_device, unsigned table,
uint8_t *frev, uint8_t *crev)
{
CGS_FUNC_ADEV;
if (amdgpu_atom_parse_cmd_header(
adev->mode_info.atom_context, table,
frev, crev))
return 0;
return -EINVAL;
}
static int amdgpu_cgs_atom_exec_cmd_table(void *cgs_device, unsigned table,
void *args)
{
CGS_FUNC_ADEV;
return amdgpu_atom_execute_table(
adev->mode_info.atom_context, table, args);
}
static int amdgpu_cgs_create_pm_request(void *cgs_device, cgs_handle_t *request)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_destroy_pm_request(void *cgs_device, cgs_handle_t request)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_set_pm_request(void *cgs_device, cgs_handle_t request,
int active)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_pm_request_clock(void *cgs_device, cgs_handle_t request,
enum cgs_clock clock, unsigned freq)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_pm_request_engine(void *cgs_device, cgs_handle_t request,
enum cgs_engine engine, int powered)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_pm_query_clock_limits(void *cgs_device,
enum cgs_clock clock,
struct cgs_clock_limits *limits)
{
/* TODO */
return 0;
}
static int amdgpu_cgs_set_camera_voltages(void *cgs_device, uint32_t mask,
const uint32_t *voltages)
{
DRM_ERROR("not implemented");
return -EPERM;
}
struct cgs_irq_params {
unsigned src_id;
cgs_irq_source_set_func_t set;
cgs_irq_handler_func_t handler;
void *private_data;
};
static int cgs_set_irq_state(struct amdgpu_device *adev,
struct amdgpu_irq_src *src,
unsigned type,
enum amdgpu_interrupt_state state)
{
struct cgs_irq_params *irq_params =
(struct cgs_irq_params *)src->data;
if (!irq_params)
return -EINVAL;
if (!irq_params->set)
return -EINVAL;
return irq_params->set(irq_params->private_data,
irq_params->src_id,
type,
(int)state);
}
static int cgs_process_irq(struct amdgpu_device *adev,
struct amdgpu_irq_src *source,
struct amdgpu_iv_entry *entry)
{
struct cgs_irq_params *irq_params =
(struct cgs_irq_params *)source->data;
if (!irq_params)
return -EINVAL;
if (!irq_params->handler)
return -EINVAL;
return irq_params->handler(irq_params->private_data,
irq_params->src_id,
entry->iv_entry);
}
static const struct amdgpu_irq_src_funcs cgs_irq_funcs = {
.set = cgs_set_irq_state,
.process = cgs_process_irq,
};
static int amdgpu_cgs_add_irq_source(void *cgs_device, unsigned src_id,
unsigned num_types,
cgs_irq_source_set_func_t set,
cgs_irq_handler_func_t handler,
void *private_data)
{
CGS_FUNC_ADEV;
int ret = 0;
struct cgs_irq_params *irq_params;
struct amdgpu_irq_src *source =
kzalloc(sizeof(struct amdgpu_irq_src), GFP_KERNEL);
if (!source)
return -ENOMEM;
irq_params =
kzalloc(sizeof(struct cgs_irq_params), GFP_KERNEL);
if (!irq_params) {
kfree(source);
return -ENOMEM;
}
source->num_types = num_types;
source->funcs = &cgs_irq_funcs;
irq_params->src_id = src_id;
irq_params->set = set;
irq_params->handler = handler;
irq_params->private_data = private_data;
source->data = (void *)irq_params;
ret = amdgpu_irq_add_id(adev, src_id, source);
if (ret) {
kfree(irq_params);
kfree(source);
}
return ret;
}
static int amdgpu_cgs_irq_get(void *cgs_device, unsigned src_id, unsigned type)
{
CGS_FUNC_ADEV;
return amdgpu_irq_get(adev, adev->irq.sources[src_id], type);
}
static int amdgpu_cgs_irq_put(void *cgs_device, unsigned src_id, unsigned type)
{
CGS_FUNC_ADEV;
return amdgpu_irq_put(adev, adev->irq.sources[src_id], type);
}
int amdgpu_cgs_set_clockgating_state(void *cgs_device,
enum amd_ip_block_type block_type,
enum amd_clockgating_state state)
{
CGS_FUNC_ADEV;
int i, r = -1;
for (i = 0; i < adev->num_ip_blocks; i++) {
if (!adev->ip_block_status[i].valid)
continue;
if (adev->ip_blocks[i].type == block_type) {
r = adev->ip_blocks[i].funcs->set_clockgating_state(
(void *)adev,
state);
break;
}
}
return r;
}
int amdgpu_cgs_set_powergating_state(void *cgs_device,
enum amd_ip_block_type block_type,
enum amd_powergating_state state)
{
CGS_FUNC_ADEV;
int i, r = -1;
for (i = 0; i < adev->num_ip_blocks; i++) {
if (!adev->ip_block_status[i].valid)
continue;
if (adev->ip_blocks[i].type == block_type) {
r = adev->ip_blocks[i].funcs->set_powergating_state(
(void *)adev,
state);
break;
}
}
return r;
}
static uint32_t fw_type_convert(void *cgs_device, uint32_t fw_type)
{
CGS_FUNC_ADEV;
enum AMDGPU_UCODE_ID result = AMDGPU_UCODE_ID_MAXIMUM;
switch (fw_type) {
case CGS_UCODE_ID_SDMA0:
result = AMDGPU_UCODE_ID_SDMA0;
break;
case CGS_UCODE_ID_SDMA1:
result = AMDGPU_UCODE_ID_SDMA1;
break;
case CGS_UCODE_ID_CP_CE:
result = AMDGPU_UCODE_ID_CP_CE;
break;
case CGS_UCODE_ID_CP_PFP:
result = AMDGPU_UCODE_ID_CP_PFP;
break;
case CGS_UCODE_ID_CP_ME:
result = AMDGPU_UCODE_ID_CP_ME;
break;
case CGS_UCODE_ID_CP_MEC:
case CGS_UCODE_ID_CP_MEC_JT1:
result = AMDGPU_UCODE_ID_CP_MEC1;
break;
case CGS_UCODE_ID_CP_MEC_JT2:
if (adev->asic_type == CHIP_TONGA)
result = AMDGPU_UCODE_ID_CP_MEC2;
else if (adev->asic_type == CHIP_CARRIZO)
result = AMDGPU_UCODE_ID_CP_MEC1;
break;
case CGS_UCODE_ID_RLC_G:
result = AMDGPU_UCODE_ID_RLC_G;
break;
default:
DRM_ERROR("Firmware type not supported\n");
}
return result;
}
static int amdgpu_cgs_get_firmware_info(void *cgs_device,
enum cgs_ucode_id type,
struct cgs_firmware_info *info)
{
CGS_FUNC_ADEV;
if (CGS_UCODE_ID_SMU != type) {
uint64_t gpu_addr;
uint32_t data_size;
const struct gfx_firmware_header_v1_0 *header;
enum AMDGPU_UCODE_ID id;
struct amdgpu_firmware_info *ucode;
id = fw_type_convert(cgs_device, type);
ucode = &adev->firmware.ucode[id];
if (ucode->fw == NULL)
return -EINVAL;
gpu_addr = ucode->mc_addr;
header = (const struct gfx_firmware_header_v1_0 *)ucode->fw->data;
data_size = le32_to_cpu(header->header.ucode_size_bytes);
if ((type == CGS_UCODE_ID_CP_MEC_JT1) ||
(type == CGS_UCODE_ID_CP_MEC_JT2)) {
gpu_addr += le32_to_cpu(header->jt_offset) << 2;
data_size = le32_to_cpu(header->jt_size) << 2;
}
info->mc_addr = gpu_addr;
info->image_size = data_size;
info->version = (uint16_t)le32_to_cpu(header->header.ucode_version);
info->feature_version = (uint16_t)le32_to_cpu(header->ucode_feature_version);
} else {
char fw_name[30] = {0};
int err = 0;
uint32_t ucode_size;
uint32_t ucode_start_address;
const uint8_t *src;
const struct smc_firmware_header_v1_0 *hdr;
switch (adev->asic_type) {
case CHIP_TONGA:
strcpy(fw_name, "amdgpu/tonga_smc.bin");
break;
default:
DRM_ERROR("SMC firmware not supported\n");
return -EINVAL;
}
err = request_firmware(&adev->pm.fw, fw_name, adev->dev);
if (err) {
DRM_ERROR("Failed to request firmware\n");
return err;
}
err = amdgpu_ucode_validate(adev->pm.fw);
if (err) {
DRM_ERROR("Failed to load firmware \"%s\"", fw_name);
release_firmware(adev->pm.fw);
adev->pm.fw = NULL;
return err;
}
hdr = (const struct smc_firmware_header_v1_0 *) adev->pm.fw->data;
adev->pm.fw_version = le32_to_cpu(hdr->header.ucode_version);
ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes);
ucode_start_address = le32_to_cpu(hdr->ucode_start_addr);
src = (const uint8_t *)(adev->pm.fw->data +
le32_to_cpu(hdr->header.ucode_array_offset_bytes));
info->version = adev->pm.fw_version;
info->image_size = ucode_size;
info->kptr = (void *)src;
}
return 0;
}
static const struct cgs_ops amdgpu_cgs_ops = {
amdgpu_cgs_gpu_mem_info,
amdgpu_cgs_gmap_kmem,
amdgpu_cgs_gunmap_kmem,
amdgpu_cgs_alloc_gpu_mem,
amdgpu_cgs_free_gpu_mem,
amdgpu_cgs_gmap_gpu_mem,
amdgpu_cgs_gunmap_gpu_mem,
amdgpu_cgs_kmap_gpu_mem,
amdgpu_cgs_kunmap_gpu_mem,
amdgpu_cgs_read_register,
amdgpu_cgs_write_register,
amdgpu_cgs_read_ind_register,
amdgpu_cgs_write_ind_register,
amdgpu_cgs_read_pci_config_byte,
amdgpu_cgs_read_pci_config_word,
amdgpu_cgs_read_pci_config_dword,
amdgpu_cgs_write_pci_config_byte,
amdgpu_cgs_write_pci_config_word,
amdgpu_cgs_write_pci_config_dword,
amdgpu_cgs_atom_get_data_table,
amdgpu_cgs_atom_get_cmd_table_revs,
amdgpu_cgs_atom_exec_cmd_table,
amdgpu_cgs_create_pm_request,
amdgpu_cgs_destroy_pm_request,
amdgpu_cgs_set_pm_request,
amdgpu_cgs_pm_request_clock,
amdgpu_cgs_pm_request_engine,
amdgpu_cgs_pm_query_clock_limits,
amdgpu_cgs_set_camera_voltages,
amdgpu_cgs_get_firmware_info,
amdgpu_cgs_set_powergating_state,
amdgpu_cgs_set_clockgating_state
};
static const struct cgs_os_ops amdgpu_cgs_os_ops = {
amdgpu_cgs_import_gpu_mem,
amdgpu_cgs_add_irq_source,
amdgpu_cgs_irq_get,
amdgpu_cgs_irq_put
};
void *amdgpu_cgs_create_device(struct amdgpu_device *adev)
{
struct amdgpu_cgs_device *cgs_device =
kmalloc(sizeof(*cgs_device), GFP_KERNEL);
if (!cgs_device) {
DRM_ERROR("Couldn't allocate CGS device structure\n");
return NULL;
}
cgs_device->base.ops = &amdgpu_cgs_ops;
cgs_device->base.os_ops = &amdgpu_cgs_os_ops;
cgs_device->adev = adev;
return cgs_device;
}
void amdgpu_cgs_destroy_device(void *cgs_device)
{
kfree(cgs_device);
}
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+246 -87
View File
@@ -126,12 +126,54 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
return 0;
}
static void amdgpu_job_work_func(struct work_struct *work)
{
struct amdgpu_cs_parser *sched_job =
container_of(work, struct amdgpu_cs_parser,
job_work);
mutex_lock(&sched_job->job_lock);
if (sched_job->free_job)
sched_job->free_job(sched_job);
mutex_unlock(&sched_job->job_lock);
/* after processing job, free memory */
fence_put(&sched_job->s_fence->base);
kfree(sched_job);
}
struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
struct drm_file *filp,
struct amdgpu_ctx *ctx,
struct amdgpu_ib *ibs,
uint32_t num_ibs)
{
struct amdgpu_cs_parser *parser;
int i;
parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL);
if (!parser)
return NULL;
parser->adev = adev;
parser->filp = filp;
parser->ctx = ctx;
parser->ibs = ibs;
parser->num_ibs = num_ibs;
if (amdgpu_enable_scheduler) {
mutex_init(&parser->job_lock);
INIT_WORK(&parser->job_work, amdgpu_job_work_func);
}
for (i = 0; i < num_ibs; i++)
ibs[i].ctx = ctx;
return parser;
}
int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
{
union drm_amdgpu_cs *cs = data;
uint64_t *chunk_array_user;
uint64_t *chunk_array = NULL;
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
struct amdgpu_bo_list *bo_list = NULL;
unsigned size, i;
int r = 0;
@@ -143,17 +185,30 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
r = -EINVAL;
goto out;
}
p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
if (!amdgpu_enable_scheduler)
p->bo_list = bo_list;
else {
if (bo_list && !bo_list->has_userptr) {
p->bo_list = amdgpu_bo_list_clone(bo_list);
amdgpu_bo_list_put(bo_list);
if (!p->bo_list)
return -ENOMEM;
} else if (bo_list && bo_list->has_userptr)
p->bo_list = bo_list;
else
p->bo_list = NULL;
}
/* get chunks */
INIT_LIST_HEAD(&p->validated);
chunk_array = kcalloc(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
if (chunk_array == NULL) {
r = -ENOMEM;
goto out;
}
chunk_array_user = (uint64_t *)(unsigned long)(cs->in.chunks);
chunk_array_user = (uint64_t __user *)(cs->in.chunks);
if (copy_from_user(chunk_array, chunk_array_user,
sizeof(uint64_t)*cs->in.num_chunks)) {
r = -EFAULT;
@@ -161,7 +216,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
}
p->nchunks = cs->in.num_chunks;
p->chunks = kcalloc(p->nchunks, sizeof(struct amdgpu_cs_chunk),
p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
GFP_KERNEL);
if (p->chunks == NULL) {
r = -ENOMEM;
@@ -173,7 +228,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
struct drm_amdgpu_cs_chunk user_chunk;
uint32_t __user *cdata;
chunk_ptr = (void __user *)(unsigned long)chunk_array[i];
chunk_ptr = (void __user *)chunk_array[i];
if (copy_from_user(&user_chunk, chunk_ptr,
sizeof(struct drm_amdgpu_cs_chunk))) {
r = -EFAULT;
@@ -183,7 +238,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
p->chunks[i].length_dw = user_chunk.length_dw;
size = p->chunks[i].length_dw;
cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
cdata = (void __user *)user_chunk.chunk_data;
p->chunks[i].user_ptr = cdata;
p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
@@ -235,11 +290,10 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
}
}
p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!p->ibs) {
p->ibs = kmalloc_array(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!p->ibs)
r = -ENOMEM;
goto out;
}
out:
kfree(chunk_array);
@@ -415,18 +469,8 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a,
return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
}
/**
* cs_parser_fini() - clean parser states
* @parser: parser structure holding parsing context.
* @error: error number
*
* If error is set than unvalidate buffer, otherwise just free memory
* used by parsing context.
**/
static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
unsigned i;
if (!error) {
/* Sort the buffer list from the smallest to largest buffer,
* which affects the order of buffers in the LRU list.
@@ -447,11 +491,19 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
ttm_eu_backoff_reservation(&parser->ticket,
&parser->validated);
}
}
static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
{
unsigned i;
if (parser->ctx)
amdgpu_ctx_put(parser->ctx);
if (parser->bo_list)
amdgpu_bo_list_put(parser->bo_list);
if (parser->bo_list) {
if (amdgpu_enable_scheduler && !parser->bo_list->has_userptr)
amdgpu_bo_list_free(parser->bo_list);
else
amdgpu_bo_list_put(parser->bo_list);
}
drm_free_large(parser->vm_bos);
for (i = 0; i < parser->nchunks; i++)
drm_free_large(parser->chunks[i].kdata);
@@ -462,6 +514,29 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
kfree(parser->ibs);
if (parser->uf.bo)
drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
if (!amdgpu_enable_scheduler)
kfree(parser);
}
/**
* cs_parser_fini() - clean parser states
* @parser: parser structure holding parsing context.
* @error: error number
*
* If error is set than unvalidate buffer, otherwise just free memory
* used by parsing context.
**/
static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
amdgpu_cs_parser_fini_early(parser, error, backoff);
amdgpu_cs_parser_fini_late(parser);
}
static int amdgpu_cs_parser_free_job(struct amdgpu_cs_parser *sched_job)
{
amdgpu_cs_parser_fini_late(sched_job);
return 0;
}
static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
@@ -476,12 +551,18 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
if (r)
return r;
r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
if (r)
return r;
r = amdgpu_vm_clear_freed(adev, vm);
if (r)
return r;
if (p->bo_list) {
for (i = 0; i < p->bo_list->num_entries; i++) {
struct fence *f;
/* ignore duplicates */
bo = p->bo_list->array[i].robj;
if (!bo)
@@ -495,7 +576,10 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
if (r)
return r;
amdgpu_sync_fence(&p->ibs[0].sync, bo_va->last_pt_update);
f = bo_va->last_pt_update;
r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
if (r)
return r;
}
}
@@ -529,9 +613,9 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
goto out;
}
amdgpu_cs_sync_rings(parser);
r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
parser->filp);
if (!amdgpu_enable_scheduler)
r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
parser->filp);
out:
mutex_unlock(&vm->mutex);
@@ -650,7 +734,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
ib->oa_size = amdgpu_bo_size(oa);
}
}
/* wrap the last IB with user fence */
if (parser->uf.bo) {
struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
@@ -693,9 +776,9 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
sizeof(struct drm_amdgpu_cs_chunk_dep);
for (j = 0; j < num_deps; ++j) {
struct amdgpu_fence *fence;
struct amdgpu_ring *ring;
struct amdgpu_ctx *ctx;
struct fence *fence;
r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
deps[j].ip_instance,
@@ -707,50 +790,34 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
if (ctx == NULL)
return -EINVAL;
r = amdgpu_fence_recreate(ring, p->filp,
deps[j].handle,
&fence);
if (r) {
fence = amdgpu_ctx_get_fence(ctx, ring,
deps[j].handle);
if (IS_ERR(fence)) {
r = PTR_ERR(fence);
amdgpu_ctx_put(ctx);
return r;
}
amdgpu_sync_fence(&ib->sync, fence);
amdgpu_fence_unref(&fence);
amdgpu_ctx_put(ctx);
} else if (fence) {
r = amdgpu_sync_fence(adev, &ib->sync, fence);
fence_put(fence);
amdgpu_ctx_put(ctx);
if (r)
return r;
}
}
}
return 0;
}
int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
static int amdgpu_cs_parser_prepare_job(struct amdgpu_cs_parser *sched_job)
{
struct amdgpu_device *adev = dev->dev_private;
union drm_amdgpu_cs *cs = data;
struct amdgpu_cs_parser parser;
int r, i;
struct amdgpu_cs_parser *parser = sched_job;
struct amdgpu_device *adev = sched_job->adev;
bool reserved_buffers = false;
down_read(&adev->exclusive_lock);
if (!adev->accel_working) {
up_read(&adev->exclusive_lock);
return -EBUSY;
}
/* initialize parser */
memset(&parser, 0, sizeof(struct amdgpu_cs_parser));
parser.filp = filp;
parser.adev = adev;
r = amdgpu_cs_parser_init(&parser, data);
if (r) {
DRM_ERROR("Failed to initialize parser !\n");
amdgpu_cs_parser_fini(&parser, r, false);
up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
r = amdgpu_cs_parser_relocs(&parser);
r = amdgpu_cs_parser_relocs(parser);
if (r) {
if (r != -ERESTARTSYS) {
if (r == -ENOMEM)
@@ -762,30 +829,114 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
if (!r) {
reserved_buffers = true;
r = amdgpu_cs_ib_fill(adev, &parser);
r = amdgpu_cs_ib_fill(adev, parser);
}
if (!r) {
r = amdgpu_cs_dependencies(adev, parser);
if (r)
DRM_ERROR("Failed in the dependencies handling %d!\n", r);
}
if (r) {
amdgpu_cs_parser_fini(parser, r, reserved_buffers);
return r;
}
if (!r)
r = amdgpu_cs_dependencies(adev, &parser);
for (i = 0; i < parser->num_ibs; i++)
trace_amdgpu_cs(parser, i);
r = amdgpu_cs_ib_vm_chunk(adev, parser);
return r;
}
static struct amdgpu_ring *amdgpu_cs_parser_get_ring(
struct amdgpu_device *adev,
struct amdgpu_cs_parser *parser)
{
int i, r;
struct amdgpu_cs_chunk *chunk;
struct drm_amdgpu_cs_chunk_ib *chunk_ib;
struct amdgpu_ring *ring;
for (i = 0; i < parser->nchunks; i++) {
chunk = &parser->chunks[i];
chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata;
if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB)
continue;
r = amdgpu_cs_get_ring(adev, chunk_ib->ip_type,
chunk_ib->ip_instance, chunk_ib->ring,
&ring);
if (r)
return NULL;
break;
}
return ring;
}
int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
{
struct amdgpu_device *adev = dev->dev_private;
union drm_amdgpu_cs *cs = data;
struct amdgpu_cs_parser *parser;
int r;
down_read(&adev->exclusive_lock);
if (!adev->accel_working) {
up_read(&adev->exclusive_lock);
return -EBUSY;
}
parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
if (!parser)
return -ENOMEM;
r = amdgpu_cs_parser_init(parser, data);
if (r) {
amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
DRM_ERROR("Failed to initialize parser !\n");
amdgpu_cs_parser_fini(parser, r, false);
up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
for (i = 0; i < parser.num_ibs; i++)
trace_amdgpu_cs(&parser, i);
if (amdgpu_enable_scheduler && parser->num_ibs) {
struct amdgpu_ring * ring =
amdgpu_cs_parser_get_ring(adev, parser);
r = amdgpu_cs_parser_prepare_job(parser);
if (r)
goto out;
parser->ring = ring;
parser->free_job = amdgpu_cs_parser_free_job;
mutex_lock(&parser->job_lock);
r = amd_sched_push_job(ring->scheduler,
&parser->ctx->rings[ring->idx].entity,
parser,
&parser->s_fence);
if (r) {
mutex_unlock(&parser->job_lock);
goto out;
}
parser->ibs[parser->num_ibs - 1].sequence =
amdgpu_ctx_add_fence(parser->ctx, ring,
&parser->s_fence->base,
parser->s_fence->v_seq);
cs->out.handle = parser->s_fence->v_seq;
list_sort(NULL, &parser->validated, cmp_size_smaller_first);
ttm_eu_fence_buffer_objects(&parser->ticket,
&parser->validated,
&parser->s_fence->base);
r = amdgpu_cs_ib_vm_chunk(adev, &parser);
if (r) {
goto out;
mutex_unlock(&parser->job_lock);
up_read(&adev->exclusive_lock);
return 0;
}
r = amdgpu_cs_parser_prepare_job(parser);
if (r)
goto out;
cs->out.handle = parser.ibs[parser.num_ibs - 1].fence->seq;
cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
out:
amdgpu_cs_parser_fini(&parser, r, true);
amdgpu_cs_parser_fini(parser, r, true);
up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
@@ -806,30 +957,29 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
union drm_amdgpu_wait_cs *wait = data;
struct amdgpu_device *adev = dev->dev_private;
unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
struct amdgpu_fence *fence = NULL;
struct amdgpu_ring *ring = NULL;
struct amdgpu_ctx *ctx;
struct fence *fence;
long r;
r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
wait->in.ring, &ring);
if (r)
return r;
ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
if (ctx == NULL)
return -EINVAL;
r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
wait->in.ring, &ring);
if (r) {
amdgpu_ctx_put(ctx);
return r;
}
fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
if (IS_ERR(fence))
r = PTR_ERR(fence);
else if (fence) {
r = fence_wait_timeout(fence, true, timeout);
fence_put(fence);
} else
r = 1;
r = amdgpu_fence_recreate(ring, filp, wait->in.handle, &fence);
if (r) {
amdgpu_ctx_put(ctx);
return r;
}
r = fence_wait_timeout(&fence->base, true, timeout);
amdgpu_fence_unref(&fence);
amdgpu_ctx_put(ctx);
if (r < 0)
return r;
@@ -864,7 +1014,16 @@ amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
if (!reloc->bo_va)
continue;
list_for_each_entry(mapping, &reloc->bo_va->mappings, list) {
list_for_each_entry(mapping, &reloc->bo_va->valids, list) {
if (mapping->it.start > addr ||
addr > mapping->it.last)
continue;
*bo = reloc->bo_va->bo;
return mapping;
}
list_for_each_entry(mapping, &reloc->bo_va->invalids, list) {
if (mapping->it.start > addr ||
addr > mapping->it.last)
continue;
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+173 -50
View File
@@ -25,54 +25,107 @@
#include <drm/drmP.h>
#include "amdgpu.h"
static void amdgpu_ctx_do_release(struct kref *ref)
int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel,
struct amdgpu_ctx *ctx)
{
struct amdgpu_ctx *ctx;
struct amdgpu_ctx_mgr *mgr;
unsigned i, j;
int r;
ctx = container_of(ref, struct amdgpu_ctx, refcount);
mgr = &ctx->fpriv->ctx_mgr;
memset(ctx, 0, sizeof(*ctx));
ctx->adev = adev;
kref_init(&ctx->refcount);
spin_lock_init(&ctx->ring_lock);
for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
ctx->rings[i].sequence = 1;
idr_remove(&mgr->ctx_handles, ctx->id);
kfree(ctx);
if (amdgpu_enable_scheduler) {
/* create context entity for each ring */
for (i = 0; i < adev->num_rings; i++) {
struct amd_sched_rq *rq;
if (kernel)
rq = &adev->rings[i]->scheduler->kernel_rq;
else
rq = &adev->rings[i]->scheduler->sched_rq;
r = amd_sched_entity_init(adev->rings[i]->scheduler,
&ctx->rings[i].entity,
rq, amdgpu_sched_jobs);
if (r)
break;
}
if (i < adev->num_rings) {
for (j = 0; j < i; j++)
amd_sched_entity_fini(adev->rings[j]->scheduler,
&ctx->rings[j].entity);
kfree(ctx);
return r;
}
}
return 0;
}
int amdgpu_ctx_alloc(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t *id, uint32_t flags)
void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
{
struct amdgpu_device *adev = ctx->adev;
unsigned i, j;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
for (j = 0; j < AMDGPU_CTX_MAX_CS_PENDING; ++j)
fence_put(ctx->rings[i].fences[j]);
if (amdgpu_enable_scheduler) {
for (i = 0; i < adev->num_rings; i++)
amd_sched_entity_fini(adev->rings[i]->scheduler,
&ctx->rings[i].entity);
}
}
static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
struct amdgpu_fpriv *fpriv,
uint32_t *id)
{
int r;
struct amdgpu_ctx *ctx;
struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
struct amdgpu_ctx *ctx;
int r;
ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
if (!ctx)
return -ENOMEM;
mutex_lock(&mgr->lock);
r = idr_alloc(&mgr->ctx_handles, ctx, 0, 0, GFP_KERNEL);
r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL);
if (r < 0) {
mutex_unlock(&mgr->lock);
kfree(ctx);
return r;
}
*id = (uint32_t)r;
memset(ctx, 0, sizeof(*ctx));
ctx->id = *id;
ctx->fpriv = fpriv;
kref_init(&ctx->refcount);
r = amdgpu_ctx_init(adev, false, ctx);
mutex_unlock(&mgr->lock);
return 0;
return r;
}
int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t id)
static void amdgpu_ctx_do_release(struct kref *ref)
{
struct amdgpu_ctx *ctx;
ctx = container_of(ref, struct amdgpu_ctx, refcount);
amdgpu_ctx_fini(ctx);
kfree(ctx);
}
static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id)
{
struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
struct amdgpu_ctx *ctx;
mutex_lock(&mgr->lock);
ctx = idr_find(&mgr->ctx_handles, id);
if (ctx) {
idr_remove(&mgr->ctx_handles, id);
kref_put(&ctx->refcount, amdgpu_ctx_do_release);
mutex_unlock(&mgr->lock);
return 0;
@@ -86,9 +139,13 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
union drm_amdgpu_ctx_out *out)
{
struct amdgpu_ctx *ctx;
struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
struct amdgpu_ctx_mgr *mgr;
unsigned reset_counter;
if (!fpriv)
return -EINVAL;
mgr = &fpriv->ctx_mgr;
mutex_lock(&mgr->lock);
ctx = idr_find(&mgr->ctx_handles, id);
if (!ctx) {
@@ -97,8 +154,8 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
}
/* TODO: these two are always zero */
out->state.flags = ctx->state.flags;
out->state.hangs = ctx->state.hangs;
out->state.flags = 0x0;
out->state.hangs = 0x0;
/* determine if a GPU reset has occured since the last call */
reset_counter = atomic_read(&adev->gpu_reset_counter);
@@ -113,28 +170,11 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
return 0;
}
void amdgpu_ctx_fini(struct amdgpu_fpriv *fpriv)
{
struct idr *idp;
struct amdgpu_ctx *ctx;
uint32_t id;
struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
idp = &mgr->ctx_handles;
idr_for_each_entry(idp,ctx,id) {
if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
DRM_ERROR("ctx (id=%ul) is still alive\n",ctx->id);
}
mutex_destroy(&mgr->lock);
}
int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp)
{
int r;
uint32_t id;
uint32_t flags;
union drm_amdgpu_ctx *args = data;
struct amdgpu_device *adev = dev->dev_private;
@@ -142,15 +182,14 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
r = 0;
id = args->in.ctx_id;
flags = args->in.flags;
switch (args->in.op) {
case AMDGPU_CTX_OP_ALLOC_CTX:
r = amdgpu_ctx_alloc(adev, fpriv, &id, flags);
r = amdgpu_ctx_alloc(adev, fpriv, &id);
args->out.alloc.ctx_id = id;
break;
case AMDGPU_CTX_OP_FREE_CTX:
r = amdgpu_ctx_free(adev, fpriv, id);
r = amdgpu_ctx_free(fpriv, id);
break;
case AMDGPU_CTX_OP_QUERY_STATE:
r = amdgpu_ctx_query(adev, fpriv, id, &args->out);
@@ -165,7 +204,12 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id)
{
struct amdgpu_ctx *ctx;
struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
struct amdgpu_ctx_mgr *mgr;
if (!fpriv)
return NULL;
mgr = &fpriv->ctx_mgr;
mutex_lock(&mgr->lock);
ctx = idr_find(&mgr->ctx_handles, id);
@@ -177,17 +221,96 @@ struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id)
int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
{
struct amdgpu_fpriv *fpriv;
struct amdgpu_ctx_mgr *mgr;
if (ctx == NULL)
return -EINVAL;
fpriv = ctx->fpriv;
mgr = &fpriv->ctx_mgr;
mutex_lock(&mgr->lock);
kref_put(&ctx->refcount, amdgpu_ctx_do_release);
mutex_unlock(&mgr->lock);
return 0;
}
uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
struct fence *fence, uint64_t queued_seq)
{
struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
uint64_t seq = 0;
unsigned idx = 0;
struct fence *other = NULL;
if (amdgpu_enable_scheduler)
seq = queued_seq;
else
seq = cring->sequence;
idx = seq % AMDGPU_CTX_MAX_CS_PENDING;
other = cring->fences[idx];
if (other) {
signed long r;
r = fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT);
if (r < 0)
DRM_ERROR("Error (%ld) waiting for fence!\n", r);
}
fence_get(fence);
spin_lock(&ctx->ring_lock);
cring->fences[idx] = fence;
if (!amdgpu_enable_scheduler)
cring->sequence++;
spin_unlock(&ctx->ring_lock);
fence_put(other);
return seq;
}
struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
struct amdgpu_ring *ring, uint64_t seq)
{
struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
struct fence *fence;
uint64_t queued_seq;
spin_lock(&ctx->ring_lock);
if (amdgpu_enable_scheduler)
queued_seq = amd_sched_next_queued_seq(&cring->entity);
else
queued_seq = cring->sequence;
if (seq >= queued_seq) {
spin_unlock(&ctx->ring_lock);
return ERR_PTR(-EINVAL);
}
if (seq + AMDGPU_CTX_MAX_CS_PENDING < queued_seq) {
spin_unlock(&ctx->ring_lock);
return NULL;
}
fence = fence_get(cring->fences[seq % AMDGPU_CTX_MAX_CS_PENDING]);
spin_unlock(&ctx->ring_lock);
return fence;
}
void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
{
mutex_init(&mgr->lock);
idr_init(&mgr->ctx_handles);
}
void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
{
struct amdgpu_ctx *ctx;
struct idr *idp;
uint32_t id;
idp = &mgr->ctx_handles;
idr_for_each_entry(idp, ctx, id) {
if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
DRM_ERROR("ctx %p is still alive\n", ctx);
}
idr_destroy(&mgr->ctx_handles);
mutex_destroy(&mgr->lock);
}
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+14 -19
View File
@@ -55,6 +55,7 @@ static const char *amdgpu_asic_name[] = {
"MULLINS",
"TOPAZ",
"TONGA",
"FIJI",
"CARRIZO",
"LAST",
};
@@ -63,7 +64,7 @@ bool amdgpu_device_is_px(struct drm_device *dev)
{
struct amdgpu_device *adev = dev->dev_private;
if (adev->flags & AMDGPU_IS_PX)
if (adev->flags & AMD_IS_PX)
return true;
return false;
}
@@ -1160,6 +1161,7 @@ static int amdgpu_early_init(struct amdgpu_device *adev)
switch (adev->asic_type) {
case CHIP_TOPAZ:
case CHIP_TONGA:
case CHIP_FIJI:
case CHIP_CARRIZO:
if (adev->asic_type == CHIP_CARRIZO)
adev->family = AMDGPU_FAMILY_CZ;
@@ -1377,7 +1379,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
adev->ddev = ddev;
adev->pdev = pdev;
adev->flags = flags;
adev->asic_type = flags & AMDGPU_ASIC_MASK;
adev->asic_type = flags & AMD_ASIC_MASK;
adev->is_atom_bios = false;
adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
adev->mc.gtt_size = 512 * 1024 * 1024;
@@ -1523,6 +1525,11 @@ int amdgpu_device_init(struct amdgpu_device *adev,
return r;
}
r = amdgpu_ctx_init(adev, true, &adev->kernel_ctx);
if (r) {
dev_err(adev->dev, "failed to create kernel context (%d).\n", r);
return r;
}
r = amdgpu_ib_ring_tests(adev);
if (r)
DRM_ERROR("ib ring test failed (%d).\n", r);
@@ -1584,6 +1591,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
adev->shutdown = true;
/* evict vram memory */
amdgpu_bo_evict_vram(adev);
amdgpu_ctx_fini(&adev->kernel_ctx);
amdgpu_ib_pool_fini(adev);
amdgpu_fence_driver_fini(adev);
amdgpu_fbdev_fini(adev);
@@ -1627,8 +1635,7 @@ int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
struct amdgpu_device *adev;
struct drm_crtc *crtc;
struct drm_connector *connector;
int i, r;
bool force_completion = false;
int r;
if (dev == NULL || dev->dev_private == NULL) {
return -ENODEV;
@@ -1667,21 +1674,7 @@ int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
/* evict vram memory */
amdgpu_bo_evict_vram(adev);
/* wait for gpu to finish processing current batch */
for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring)
continue;
r = amdgpu_fence_wait_empty(ring);
if (r) {
/* delay GPU reset to resume */
force_completion = true;
}
}
if (force_completion) {
amdgpu_fence_driver_force_completion(adev);
}
amdgpu_fence_driver_suspend(adev);
r = amdgpu_suspend(adev);
@@ -1739,6 +1732,8 @@ int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon)
r = amdgpu_resume(adev);
amdgpu_fence_driver_resume(adev);
r = amdgpu_ib_ring_tests(adev);
if (r)
DRM_ERROR("ib ring test failed (%d).\n", r);
drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
+53 -28
View File
@@ -35,6 +35,36 @@
#include <drm/drm_crtc_helper.h>
#include <drm/drm_edid.h>
static void amdgpu_flip_wait_fence(struct amdgpu_device *adev,
struct fence **f)
{
struct amdgpu_fence *fence;
long r;
if (*f == NULL)
return;
fence = to_amdgpu_fence(*f);
if (fence) {
r = fence_wait(&fence->base, false);
if (r == -EDEADLK) {
up_read(&adev->exclusive_lock);
r = amdgpu_gpu_reset(adev);
down_read(&adev->exclusive_lock);
}
} else
r = fence_wait(*f, false);
if (r)
DRM_ERROR("failed to wait on page flip fence (%ld)!\n", r);
/* We continue with the page flip even if we failed to wait on
* the fence, otherwise the DRM core and userspace will be
* confused about which BO the CRTC is scanning out
*/
fence_put(*f);
*f = NULL;
}
static void amdgpu_flip_work_func(struct work_struct *__work)
{
@@ -44,34 +74,13 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
struct amdgpu_crtc *amdgpuCrtc = adev->mode_info.crtcs[work->crtc_id];
struct drm_crtc *crtc = &amdgpuCrtc->base;
struct amdgpu_fence *fence;
unsigned long flags;
int r;
unsigned i;
down_read(&adev->exclusive_lock);
if (work->fence) {
fence = to_amdgpu_fence(work->fence);
if (fence) {
r = amdgpu_fence_wait(fence, false);
if (r == -EDEADLK) {
up_read(&adev->exclusive_lock);
r = amdgpu_gpu_reset(adev);
down_read(&adev->exclusive_lock);
}
} else
r = fence_wait(work->fence, false);
if (r)
DRM_ERROR("failed to wait on page flip fence (%d)!\n", r);
/* We continue with the page flip even if we failed to wait on
* the fence, otherwise the DRM core and userspace will be
* confused about which BO the CRTC is scanning out
*/
fence_put(work->fence);
work->fence = NULL;
}
amdgpu_flip_wait_fence(adev, &work->excl);
for (i = 0; i < work->shared_count; ++i)
amdgpu_flip_wait_fence(adev, &work->shared[i]);
/* We borrow the event spin lock for protecting flip_status */
spin_lock_irqsave(&crtc->dev->event_lock, flags);
@@ -108,6 +117,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work)
DRM_ERROR("failed to reserve buffer after flip\n");
drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
kfree(work->shared);
kfree(work);
}
@@ -127,7 +137,7 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc,
unsigned long flags;
u64 tiling_flags;
u64 base;
int r;
int i, r;
work = kzalloc(sizeof *work, GFP_KERNEL);
if (work == NULL)
@@ -167,7 +177,19 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc,
goto cleanup;
}
work->fence = fence_get(reservation_object_get_excl(new_rbo->tbo.resv));
r = reservation_object_get_fences_rcu(new_rbo->tbo.resv, &work->excl,
&work->shared_count,
&work->shared);
if (unlikely(r != 0)) {
amdgpu_bo_unreserve(new_rbo);
DRM_ERROR("failed to get fences for buffer\n");
goto cleanup;
}
fence_get(work->excl);
for (i = 0; i < work->shared_count; ++i)
fence_get(work->shared[i]);
amdgpu_bo_get_tiling_flags(new_rbo, &tiling_flags);
amdgpu_bo_unreserve(new_rbo);
@@ -212,7 +234,10 @@ pflip_cleanup:
cleanup:
drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
fence_put(work->fence);
fence_put(work->excl);
for (i = 0; i < work->shared_count; ++i)
fence_put(work->shared[i]);
kfree(work->shared);
kfree(work);
return r;
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+80 -66
View File
@@ -63,7 +63,7 @@ int amdgpu_disp_priority = 0;
int amdgpu_hw_i2c = 0;
int amdgpu_pcie_gen2 = -1;
int amdgpu_msi = -1;
int amdgpu_lockup_timeout = 10000;
int amdgpu_lockup_timeout = 0;
int amdgpu_dpm = -1;
int amdgpu_smc_load_fw = 1;
int amdgpu_aspm = -1;
@@ -75,6 +75,9 @@ int amdgpu_deep_color = 0;
int amdgpu_vm_size = 8;
int amdgpu_vm_block_size = -1;
int amdgpu_exp_hw_support = 0;
int amdgpu_enable_scheduler = 0;
int amdgpu_sched_jobs = 16;
int amdgpu_sched_hw_submission = 2;
MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);
@@ -103,7 +106,7 @@ module_param_named(pcie_gen2, amdgpu_pcie_gen2, int, 0444);
MODULE_PARM_DESC(msi, "MSI support (1 = enable, 0 = disable, -1 = auto)");
module_param_named(msi, amdgpu_msi, int, 0444);
MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (defaul 10000 = 10 seconds, 0 = disable)");
MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (default 0 = disable)");
module_param_named(lockup_timeout, amdgpu_lockup_timeout, int, 0444);
MODULE_PARM_DESC(dpm, "DPM support (1 = enable, 0 = disable, -1 = auto)");
@@ -139,36 +142,45 @@ module_param_named(vm_block_size, amdgpu_vm_block_size, int, 0444);
MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
MODULE_PARM_DESC(enable_scheduler, "enable SW GPU scheduler (1 = enable, 0 = disable ((default))");
module_param_named(enable_scheduler, amdgpu_enable_scheduler, int, 0444);
MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 16)");
module_param_named(sched_jobs, amdgpu_sched_jobs, int, 0444);
MODULE_PARM_DESC(sched_hw_submission, "the max number of HW submissions (default 2)");
module_param_named(sched_hw_submission, amdgpu_sched_hw_submission, int, 0444);
static struct pci_device_id pciidlist[] = {
#ifdef CONFIG_DRM_AMDGPU_CIK
/* Kaveri */
{0x1002, 0x1304, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x1305, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1306, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x1307, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1309, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x130F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1310, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1311, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1312, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1313, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1315, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1316, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1317, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x1318, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x131B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x131C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x131D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
{0x1002, 0x1304, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x1305, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1306, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x1307, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1309, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x130F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1310, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1311, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1312, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1313, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1315, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1316, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x1317, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x1318, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x131B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x131C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
{0x1002, 0x131D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
/* Bonaire */
{0x1002, 0x6640, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
{0x1002, 0x6641, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
{0x1002, 0x6646, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
{0x1002, 0x6647, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
{0x1002, 0x6640, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
{0x1002, 0x6641, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
{0x1002, 0x6646, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
{0x1002, 0x6647, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
{0x1002, 0x6649, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
{0x1002, 0x6650, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
{0x1002, 0x6651, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
@@ -190,39 +202,39 @@ static struct pci_device_id pciidlist[] = {
{0x1002, 0x67BA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAWAII},
{0x1002, 0x67BE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAWAII},
/* Kabini */
{0x1002, 0x9830, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9831, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x9832, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9833, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x9834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x9836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9837, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x9838, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9839, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x983a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x983b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x983c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x983d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x983e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x983f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
{0x1002, 0x9830, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9831, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x9832, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9833, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x9834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x9836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9837, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x9838, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9839, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x983a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x983b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x983c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x983d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x983e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
{0x1002, 0x983f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
/* mullins */
{0x1002, 0x9850, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9851, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9852, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9853, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9854, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9855, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9856, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9857, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9858, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9859, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
{0x1002, 0x9850, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9851, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9852, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9853, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9854, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9855, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9856, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9857, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9858, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x9859, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
{0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
#endif
/* topaz */
{0x1002, 0x6900, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ},
@@ -240,12 +252,14 @@ static struct pci_device_id pciidlist[] = {
{0x1002, 0x6930, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
{0x1002, 0x6938, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
{0x1002, 0x6939, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
/* fiji */
{0x1002, 0x7300, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_FIJI},
/* carrizo */
{0x1002, 0x9870, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
{0x1002, 0x9874, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
{0x1002, 0x9875, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
{0x1002, 0x9876, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
{0x1002, 0x9877, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
{0x1002, 0x9870, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
{0x1002, 0x9874, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
{0x1002, 0x9875, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
{0x1002, 0x9876, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
{0x1002, 0x9877, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
{0, 0, 0}
};
@@ -281,7 +295,7 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
unsigned long flags = ent->driver_data;
int ret;
if ((flags & AMDGPU_EXP_HW_SUPPORT) && !amdgpu_exp_hw_support) {
if ((flags & AMD_EXP_HW_SUPPORT) && !amdgpu_exp_hw_support) {
DRM_INFO("This hardware requires experimental hardware support.\n"
"See modparam exp_hw_support\n");
return -ENODEV;
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.h
+1 -1
View File
@@ -31,7 +31,7 @@
#include <linux/firmware.h>
#include <linux/platform_device.h>
#include "amdgpu_family.h"
#include "amd_shared.h"
/* General customization:
*/
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+169 -394
View File
@@ -126,7 +126,8 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
(*fence)->ring = ring;
(*fence)->owner = owner;
fence_init(&(*fence)->base, &amdgpu_fence_ops,
&adev->fence_queue.lock, adev->fence_context + ring->idx,
&ring->fence_drv.fence_queue.lock,
adev->fence_context + ring->idx,
(*fence)->seq);
amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
(*fence)->seq,
@@ -135,38 +136,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
return 0;
}
/**
* amdgpu_fence_recreate - recreate a fence from an user fence
*
* @ring: ring the fence is associated with
* @owner: creator of the fence
* @seq: user fence sequence number
* @fence: resulting amdgpu fence object
*
* Recreates a fence command from the user fence sequence number (all asics).
* Returns 0 on success, -ENOMEM on failure.
*/
int amdgpu_fence_recreate(struct amdgpu_ring *ring, void *owner,
uint64_t seq, struct amdgpu_fence **fence)
{
struct amdgpu_device *adev = ring->adev;
if (seq > ring->fence_drv.sync_seq[ring->idx])
return -EINVAL;
*fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
if ((*fence) == NULL)
return -ENOMEM;
(*fence)->seq = seq;
(*fence)->ring = ring;
(*fence)->owner = owner;
fence_init(&(*fence)->base, &amdgpu_fence_ops,
&adev->fence_queue.lock, adev->fence_context + ring->idx,
(*fence)->seq);
return 0;
}
/**
* amdgpu_fence_check_signaled - callback from fence_queue
*
@@ -196,9 +165,7 @@ static int amdgpu_fence_check_signaled(wait_queue_t *wait, unsigned mode, int fl
else
FENCE_TRACE(&fence->base, "was already signaled\n");
amdgpu_irq_put(adev, fence->ring->fence_drv.irq_src,
fence->ring->fence_drv.irq_type);
__remove_wait_queue(&adev->fence_queue, &fence->fence_wake);
__remove_wait_queue(&fence->ring->fence_drv.fence_queue, &fence->fence_wake);
fence_put(&fence->base);
} else
FENCE_TRACE(&fence->base, "pending\n");
@@ -299,14 +266,9 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
return;
}
if (fence_drv->delayed_irq && ring->adev->ddev->irq_enabled) {
fence_drv->delayed_irq = false;
amdgpu_irq_update(ring->adev, fence_drv->irq_src,
fence_drv->irq_type);
if (amdgpu_fence_activity(ring)) {
wake_up_all(&ring->fence_drv.fence_queue);
}
if (amdgpu_fence_activity(ring))
wake_up_all(&ring->adev->fence_queue);
else if (amdgpu_ring_is_lockup(ring)) {
/* good news we believe it's a lockup */
dev_warn(ring->adev->dev, "GPU lockup (current fence id "
@@ -316,7 +278,7 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
/* remember that we need an reset */
ring->adev->needs_reset = true;
wake_up_all(&ring->adev->fence_queue);
wake_up_all(&ring->fence_drv.fence_queue);
}
up_read(&ring->adev->exclusive_lock);
}
@@ -332,62 +294,8 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
*/
void amdgpu_fence_process(struct amdgpu_ring *ring)
{
uint64_t seq, last_seq, last_emitted;
unsigned count_loop = 0;
bool wake = false;
/* Note there is a scenario here for an infinite loop but it's
* very unlikely to happen. For it to happen, the current polling
* process need to be interrupted by another process and another
* process needs to update the last_seq btw the atomic read and
* xchg of the current process.
*
* More over for this to go in infinite loop there need to be
* continuously new fence signaled ie amdgpu_fence_read needs
* to return a different value each time for both the currently
* polling process and the other process that xchg the last_seq
* btw atomic read and xchg of the current process. And the
* value the other process set as last seq must be higher than
* the seq value we just read. Which means that current process
* need to be interrupted after amdgpu_fence_read and before
* atomic xchg.
*
* To be even more safe we count the number of time we loop and
* we bail after 10 loop just accepting the fact that we might
* have temporarly set the last_seq not to the true real last
* seq but to an older one.
*/
last_seq = atomic64_read(&ring->fence_drv.last_seq);
do {
last_emitted = ring->fence_drv.sync_seq[ring->idx];
seq = amdgpu_fence_read(ring);
seq |= last_seq & 0xffffffff00000000LL;
if (seq < last_seq) {
seq &= 0xffffffff;
seq |= last_emitted & 0xffffffff00000000LL;
}
if (seq <= last_seq || seq > last_emitted) {
break;
}
/* If we loop over we don't want to return without
* checking if a fence is signaled as it means that the
* seq we just read is different from the previous on.
*/
wake = true;
last_seq = seq;
if ((count_loop++) > 10) {
/* We looped over too many time leave with the
* fact that we might have set an older fence
* seq then the current real last seq as signaled
* by the hw.
*/
break;
}
} while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq);
if (wake)
wake_up_all(&ring->adev->fence_queue);
if (amdgpu_fence_activity(ring))
wake_up_all(&ring->fence_drv.fence_queue);
}
/**
@@ -447,284 +355,49 @@ static bool amdgpu_fence_enable_signaling(struct fence *f)
{
struct amdgpu_fence *fence = to_amdgpu_fence(f);
struct amdgpu_ring *ring = fence->ring;
struct amdgpu_device *adev = ring->adev;
if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
return false;
if (down_read_trylock(&adev->exclusive_lock)) {
amdgpu_irq_get(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
if (amdgpu_fence_activity(ring))
wake_up_all_locked(&adev->fence_queue);
/* did fence get signaled after we enabled the sw irq? */
if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) {
amdgpu_irq_put(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
up_read(&adev->exclusive_lock);
return false;
}
up_read(&adev->exclusive_lock);
} else {
/* we're probably in a lockup, lets not fiddle too much */
if (amdgpu_irq_get_delayed(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type))
ring->fence_drv.delayed_irq = true;
amdgpu_fence_schedule_check(ring);
}
fence->fence_wake.flags = 0;
fence->fence_wake.private = NULL;
fence->fence_wake.func = amdgpu_fence_check_signaled;
__add_wait_queue(&adev->fence_queue, &fence->fence_wake);
__add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake);
fence_get(f);
FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
return true;
}
/**
* amdgpu_fence_signaled - check if a fence has signaled
/*
* amdgpu_ring_wait_seq_timeout - wait for seq of the specific ring to signal
* @ring: ring to wait on for the seq number
* @seq: seq number wait for
*
* @fence: amdgpu fence object
*
* Check if the requested fence has signaled (all asics).
* Returns true if the fence has signaled or false if it has not.
* return value:
* 0: seq signaled, and gpu not hang
* -EDEADL: GPU hang detected
* -EINVAL: some paramter is not valid
*/
bool amdgpu_fence_signaled(struct amdgpu_fence *fence)
static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq)
{
if (!fence)
return true;
struct amdgpu_device *adev = ring->adev;
bool signaled = false;
if (amdgpu_fence_seq_signaled(fence->ring, fence->seq)) {
if (!fence_signal(&fence->base))
FENCE_TRACE(&fence->base, "signaled from amdgpu_fence_signaled\n");
return true;
}
BUG_ON(!ring);
if (seq > ring->fence_drv.sync_seq[ring->idx])
return -EINVAL;
return false;
}
if (atomic64_read(&ring->fence_drv.last_seq) >= seq)
return 0;
/**
* amdgpu_fence_any_seq_signaled - check if any sequence number is signaled
*
* @adev: amdgpu device pointer
* @seq: sequence numbers
*
* Check if the last signaled fence sequnce number is >= the requested
* sequence number (all asics).
* Returns true if any has signaled (current value is >= requested value)
* or false if it has not. Helper function for amdgpu_fence_wait_seq.
*/
static bool amdgpu_fence_any_seq_signaled(struct amdgpu_device *adev, u64 *seq)
{
unsigned i;
wait_event(ring->fence_drv.fence_queue, (
(signaled = amdgpu_fence_seq_signaled(ring, seq))
|| adev->needs_reset));
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
if (!adev->rings[i] || !seq[i])
continue;
if (amdgpu_fence_seq_signaled(adev->rings[i], seq[i]))
return true;
}
return false;
}
/**
* amdgpu_fence_wait_seq_timeout - wait for a specific sequence numbers
*
* @adev: amdgpu device pointer
* @target_seq: sequence number(s) we want to wait for
* @intr: use interruptable sleep
* @timeout: maximum time to wait, or MAX_SCHEDULE_TIMEOUT for infinite wait
*
* Wait for the requested sequence number(s) to be written by any ring
* (all asics). Sequnce number array is indexed by ring id.
* @intr selects whether to use interruptable (true) or non-interruptable
* (false) sleep when waiting for the sequence number. Helper function
* for amdgpu_fence_wait_*().
* Returns remaining time if the sequence number has passed, 0 when
* the wait timeout, or an error for all other cases.
* -EDEADLK is returned when a GPU lockup has been detected.
*/
static long amdgpu_fence_wait_seq_timeout(struct amdgpu_device *adev,
u64 *target_seq, bool intr,
long timeout)
{
uint64_t last_seq[AMDGPU_MAX_RINGS];
bool signaled;
int i;
long r;
if (timeout == 0) {
return amdgpu_fence_any_seq_signaled(adev, target_seq);
}
while (!amdgpu_fence_any_seq_signaled(adev, target_seq)) {
/* Save current sequence values, used to check for GPU lockups */
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring || !target_seq[i])
continue;
last_seq[i] = atomic64_read(&ring->fence_drv.last_seq);
trace_amdgpu_fence_wait_begin(adev->ddev, i, target_seq[i]);
amdgpu_irq_get(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
}
if (intr) {
r = wait_event_interruptible_timeout(adev->fence_queue, (
(signaled = amdgpu_fence_any_seq_signaled(adev, target_seq))
|| adev->needs_reset), AMDGPU_FENCE_JIFFIES_TIMEOUT);
} else {
r = wait_event_timeout(adev->fence_queue, (
(signaled = amdgpu_fence_any_seq_signaled(adev, target_seq))
|| adev->needs_reset), AMDGPU_FENCE_JIFFIES_TIMEOUT);
}
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring || !target_seq[i])
continue;
amdgpu_irq_put(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
trace_amdgpu_fence_wait_end(adev->ddev, i, target_seq[i]);
}
if (unlikely(r < 0))
return r;
if (unlikely(!signaled)) {
if (adev->needs_reset)
return -EDEADLK;
/* we were interrupted for some reason and fence
* isn't signaled yet, resume waiting */
if (r)
continue;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring || !target_seq[i])
continue;
if (last_seq[i] != atomic64_read(&ring->fence_drv.last_seq))
break;
}
if (i != AMDGPU_MAX_RINGS)
continue;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
if (!adev->rings[i] || !target_seq[i])
continue;
if (amdgpu_ring_is_lockup(adev->rings[i]))
break;
}
if (i < AMDGPU_MAX_RINGS) {
/* good news we believe it's a lockup */
dev_warn(adev->dev, "GPU lockup (waiting for "
"0x%016llx last fence id 0x%016llx on"
" ring %d)\n",
target_seq[i], last_seq[i], i);
/* remember that we need an reset */
adev->needs_reset = true;
wake_up_all(&adev->fence_queue);
return -EDEADLK;
}
if (timeout < MAX_SCHEDULE_TIMEOUT) {
timeout -= AMDGPU_FENCE_JIFFIES_TIMEOUT;
if (timeout <= 0) {
return 0;
}
}
}
}
return timeout;
}
/**
* amdgpu_fence_wait - wait for a fence to signal
*
* @fence: amdgpu fence object
* @intr: use interruptable sleep
*
* Wait for the requested fence to signal (all asics).
* @intr selects whether to use interruptable (true) or non-interruptable
* (false) sleep when waiting for the fence.
* Returns 0 if the fence has passed, error for all other cases.
*/
int amdgpu_fence_wait(struct amdgpu_fence *fence, bool intr)
{
uint64_t seq[AMDGPU_MAX_RINGS] = {};
long r;
seq[fence->ring->idx] = fence->seq;
r = amdgpu_fence_wait_seq_timeout(fence->ring->adev, seq, intr, MAX_SCHEDULE_TIMEOUT);
if (r < 0) {
return r;
}
r = fence_signal(&fence->base);
if (!r)
FENCE_TRACE(&fence->base, "signaled from fence_wait\n");
return 0;
}
/**
* amdgpu_fence_wait_any - wait for a fence to signal on any ring
*
* @adev: amdgpu device pointer
* @fences: amdgpu fence object(s)
* @intr: use interruptable sleep
*
* Wait for any requested fence to signal (all asics). Fence
* array is indexed by ring id. @intr selects whether to use
* interruptable (true) or non-interruptable (false) sleep when
* waiting for the fences. Used by the suballocator.
* Returns 0 if any fence has passed, error for all other cases.
*/
int amdgpu_fence_wait_any(struct amdgpu_device *adev,
struct amdgpu_fence **fences,
bool intr)
{
uint64_t seq[AMDGPU_MAX_RINGS];
unsigned i, num_rings = 0;
long r;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
seq[i] = 0;
if (!fences[i]) {
continue;
}
seq[i] = fences[i]->seq;
++num_rings;
}
/* nothing to wait for ? */
if (num_rings == 0)
return -ENOENT;
r = amdgpu_fence_wait_seq_timeout(adev, seq, intr, MAX_SCHEDULE_TIMEOUT);
if (r < 0) {
return r;
}
return 0;
if (signaled)
return 0;
else
return -EDEADLK;
}
/**
@@ -739,19 +412,12 @@ int amdgpu_fence_wait_any(struct amdgpu_device *adev,
*/
int amdgpu_fence_wait_next(struct amdgpu_ring *ring)
{
uint64_t seq[AMDGPU_MAX_RINGS] = {};
long r;
uint64_t seq = atomic64_read(&ring->fence_drv.last_seq) + 1ULL;
seq[ring->idx] = atomic64_read(&ring->fence_drv.last_seq) + 1ULL;
if (seq[ring->idx] >= ring->fence_drv.sync_seq[ring->idx]) {
/* nothing to wait for, last_seq is
already the last emited fence */
if (seq >= ring->fence_drv.sync_seq[ring->idx])
return -ENOENT;
}
r = amdgpu_fence_wait_seq_timeout(ring->adev, seq, false, MAX_SCHEDULE_TIMEOUT);
if (r < 0)
return r;
return 0;
return amdgpu_fence_ring_wait_seq(ring, seq);
}
/**
@@ -766,23 +432,12 @@ int amdgpu_fence_wait_next(struct amdgpu_ring *ring)
*/
int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
uint64_t seq[AMDGPU_MAX_RINGS] = {};
long r;
uint64_t seq = ring->fence_drv.sync_seq[ring->idx];
seq[ring->idx] = ring->fence_drv.sync_seq[ring->idx];
if (!seq[ring->idx])
if (!seq)
return 0;
r = amdgpu_fence_wait_seq_timeout(adev, seq, false, MAX_SCHEDULE_TIMEOUT);
if (r < 0) {
if (r == -EDEADLK)
return -EDEADLK;
dev_err(adev->dev, "error waiting for ring[%d] to become idle (%ld)\n",
ring->idx, r);
}
return 0;
return amdgpu_fence_ring_wait_seq(ring, seq);
}
/**
@@ -933,9 +588,12 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index;
}
amdgpu_fence_write(ring, atomic64_read(&ring->fence_drv.last_seq));
ring->fence_drv.initialized = true;
amdgpu_irq_get(adev, irq_src, irq_type);
ring->fence_drv.irq_src = irq_src;
ring->fence_drv.irq_type = irq_type;
ring->fence_drv.initialized = true;
dev_info(adev->dev, "fence driver on ring %d use gpu addr 0x%016llx, "
"cpu addr 0x%p\n", ring->idx,
ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr);
@@ -966,6 +624,16 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
INIT_DELAYED_WORK(&ring->fence_drv.lockup_work,
amdgpu_fence_check_lockup);
ring->fence_drv.ring = ring;
if (amdgpu_enable_scheduler) {
ring->scheduler = amd_sched_create((void *)ring->adev,
&amdgpu_sched_ops,
ring->idx, 5, 0,
amdgpu_sched_hw_submission);
if (!ring->scheduler)
DRM_ERROR("Failed to create scheduler on ring %d.\n",
ring->idx);
}
}
/**
@@ -982,7 +650,6 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
*/
int amdgpu_fence_driver_init(struct amdgpu_device *adev)
{
init_waitqueue_head(&adev->fence_queue);
if (amdgpu_debugfs_fence_init(adev))
dev_err(adev->dev, "fence debugfs file creation failed\n");
@@ -1011,12 +678,77 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
/* no need to trigger GPU reset as we are unloading */
amdgpu_fence_driver_force_completion(adev);
}
wake_up_all(&adev->fence_queue);
wake_up_all(&ring->fence_drv.fence_queue);
amdgpu_irq_put(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
if (ring->scheduler)
amd_sched_destroy(ring->scheduler);
ring->fence_drv.initialized = false;
}
mutex_unlock(&adev->ring_lock);
}
/**
* amdgpu_fence_driver_suspend - suspend the fence driver
* for all possible rings.
*
* @adev: amdgpu device pointer
*
* Suspend the fence driver for all possible rings (all asics).
*/
void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
{
int i, r;
mutex_lock(&adev->ring_lock);
for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring || !ring->fence_drv.initialized)
continue;
/* wait for gpu to finish processing current batch */
r = amdgpu_fence_wait_empty(ring);
if (r) {
/* delay GPU reset to resume */
amdgpu_fence_driver_force_completion(adev);
}
/* disable the interrupt */
amdgpu_irq_put(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
}
mutex_unlock(&adev->ring_lock);
}
/**
* amdgpu_fence_driver_resume - resume the fence driver
* for all possible rings.
*
* @adev: amdgpu device pointer
*
* Resume the fence driver for all possible rings (all asics).
* Not all asics have all rings, so each asic will only
* start the fence driver on the rings it has using
* amdgpu_fence_driver_start_ring().
* Returns 0 for success.
*/
void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
{
int i;
mutex_lock(&adev->ring_lock);
for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
struct amdgpu_ring *ring = adev->rings[i];
if (!ring || !ring->fence_drv.initialized)
continue;
/* enable the interrupt */
amdgpu_irq_get(adev, ring->fence_drv.irq_src,
ring->fence_drv.irq_type);
}
mutex_unlock(&adev->ring_lock);
}
/**
* amdgpu_fence_driver_force_completion - force all fence waiter to complete
*
@@ -1104,6 +836,22 @@ static inline bool amdgpu_test_signaled(struct amdgpu_fence *fence)
return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
}
static inline bool amdgpu_test_signaled_any(struct amdgpu_fence **fences)
{
int idx;
struct amdgpu_fence *fence;
idx = 0;
for (idx = 0; idx < AMDGPU_MAX_RINGS; ++idx) {
fence = fences[idx];
if (fence) {
if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags))
return true;
}
}
return false;
}
struct amdgpu_wait_cb {
struct fence_cb base;
struct task_struct *task;
@@ -1119,14 +867,35 @@ static void amdgpu_fence_wait_cb(struct fence *fence, struct fence_cb *cb)
static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
signed long t)
{
struct amdgpu_fence *array[AMDGPU_MAX_RINGS];
struct amdgpu_fence *fence = to_amdgpu_fence(f);
struct amdgpu_device *adev = fence->ring->adev;
struct amdgpu_wait_cb cb;
cb.task = current;
memset(&array[0], 0, sizeof(array));
array[0] = fence;
if (fence_add_callback(f, &cb.base, amdgpu_fence_wait_cb))
return t;
return amdgpu_fence_wait_any(adev, array, intr, t);
}
/* wait until any fence in array signaled */
signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
struct amdgpu_fence **array, bool intr, signed long t)
{
long idx = 0;
struct amdgpu_wait_cb cb[AMDGPU_MAX_RINGS];
struct amdgpu_fence *fence;
BUG_ON(!array);
for (idx = 0; idx < AMDGPU_MAX_RINGS; ++idx) {
fence = array[idx];
if (fence) {
cb[idx].task = current;
if (fence_add_callback(&fence->base,
&cb[idx].base, amdgpu_fence_wait_cb))
return t; /* return if fence is already signaled */
}
}
while (t > 0) {
if (intr)
@@ -1135,10 +904,10 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
set_current_state(TASK_UNINTERRUPTIBLE);
/*
* amdgpu_test_signaled must be called after
* amdgpu_test_signaled_any must be called after
* set_current_state to prevent a race with wake_up_process
*/
if (amdgpu_test_signaled(fence))
if (amdgpu_test_signaled_any(array))
break;
if (adev->needs_reset) {
@@ -1153,7 +922,13 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
}
__set_current_state(TASK_RUNNING);
fence_remove_callback(f, &cb.base);
idx = 0;
for (idx = 0; idx < AMDGPU_MAX_RINGS; ++idx) {
fence = array[idx];
if (fence)
fence_remove_callback(&fence->base, &cb[idx].base);
}
return t;
}
drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+15 -4
View File
@@ -88,6 +88,7 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
ib->fence = NULL;
ib->user = NULL;
ib->vm = vm;
ib->ctx = NULL;
ib->gds_base = 0;
ib->gds_size = 0;
ib->gws_base = 0;
@@ -142,6 +143,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
struct amdgpu_ring *ring;
struct amdgpu_ctx *ctx, *old_ctx;
struct amdgpu_vm *vm;
uint64_t sequence;
unsigned i;
int r = 0;
@@ -165,9 +167,11 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
if (vm) {
/* grab a vm id if necessary */
struct amdgpu_fence *vm_id_fence = NULL;
vm_id_fence = amdgpu_vm_grab_id(ibs->ring, ibs->vm);
amdgpu_sync_fence(&ibs->sync, vm_id_fence);
r = amdgpu_vm_grab_id(ibs->vm, ibs->ring, &ibs->sync);
if (r) {
amdgpu_ring_unlock_undo(ring);
return r;
}
}
r = amdgpu_sync_rings(&ibs->sync, ring);
@@ -212,11 +216,18 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
return r;
}
sequence = amdgpu_enable_scheduler ? ib->sequence : 0;
if (!amdgpu_enable_scheduler && ib->ctx)
ib->sequence = amdgpu_ctx_add_fence(ib->ctx, ring,
&ib->fence->base,
sequence);
/* wrap the last IB with fence */
if (ib->user) {
uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo);
addr += ib->user->offset;
amdgpu_ring_emit_fence(ring, addr, ib->fence->seq,
amdgpu_ring_emit_fence(ring, addr, ib->sequence,
AMDGPU_FENCE_FLAG_64BIT);
}
drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
+2
View File
@@ -206,6 +206,8 @@ restart_ih:
amdgpu_amdkfd_interrupt(adev,
(const void *) &adev->irq.ih.ring[ring_index]);
entry.iv_entry = (const uint32_t *)
&adev->irq.ih.ring[ring_index];
amdgpu_ih_decode_iv(adev, &entry);
adev->irq.ih.rptr &= adev->irq.ih.ptr_mask;
drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
+1
View File
@@ -52,6 +52,7 @@ struct amdgpu_iv_entry {
unsigned ring_id;
unsigned vm_id;
unsigned pas_id;
const uint32_t *iv_entry;
};
int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
+5
View File
@@ -272,6 +272,11 @@ void amdgpu_irq_fini(struct amdgpu_device *adev)
kfree(src->enabled_types);
src->enabled_types = NULL;
if (src->data) {
kfree(src->data);
kfree(src);
adev->irq.sources[i] = NULL;
}
}
}
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
+1
View File
@@ -40,6 +40,7 @@ struct amdgpu_irq_src {
unsigned num_types;
atomic_t *enabled_types;
const struct amdgpu_irq_src_funcs *funcs;
void *data;
};
/* provided by interrupt generating IP blocks */
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+6 -10
View File
@@ -96,8 +96,8 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
if ((amdgpu_runtime_pm != 0) &&
amdgpu_has_atpx() &&
((flags & AMDGPU_IS_APU) == 0))
flags |= AMDGPU_IS_PX;
((flags & AMD_IS_APU) == 0))
flags |= AMD_IS_PX;
/* amdgpu_device_init should report only fatal error
* like memory allocation failure or iomapping failure,
@@ -451,11 +451,11 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
dev_info.num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts;
dev_info._pad = 0;
dev_info.ids_flags = 0;
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
dev_info.ids_flags |= AMDGPU_IDS_FLAGS_FUSION;
dev_info.virtual_address_offset = AMDGPU_VA_RESERVED_SIZE;
dev_info.virtual_address_max = (uint64_t)adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
dev_info.virtual_address_alignment = max(PAGE_SIZE, 0x10000UL);
dev_info.virtual_address_alignment = max((int)PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
dev_info.pte_fragment_size = (1 << AMDGPU_LOG2_PAGES_PER_FRAG) *
AMDGPU_GPU_PAGE_SIZE;
dev_info.gart_page_size = AMDGPU_GPU_PAGE_SIZE;
@@ -527,10 +527,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
mutex_init(&fpriv->bo_list_lock);
idr_init(&fpriv->bo_list_handles);
/* init context manager */
mutex_init(&fpriv->ctx_mgr.lock);
idr_init(&fpriv->ctx_mgr.ctx_handles);
fpriv->ctx_mgr.adev = adev;
amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
file_priv->driver_priv = fpriv;
@@ -571,8 +568,7 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
idr_destroy(&fpriv->bo_list_handles);
mutex_destroy(&fpriv->bo_list_lock);
/* release context */
amdgpu_ctx_fini(fpriv);
amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
kfree(fpriv);
file_priv->driver_priv = NULL;
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+5 -19
View File
@@ -223,18 +223,6 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
size_t acc_size;
int r;
/* VI has a hw bug where VM PTEs have to be allocated in groups of 8.
* do this as a temporary workaround
*/
if (!(domain & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA))) {
if (adev->asic_type >= CHIP_TOPAZ) {
if (byte_align & 0x7fff)
byte_align = ALIGN(byte_align, 0x8000);
if (size & 0x7fff)
size = ALIGN(size, 0x8000);
}
}
page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
size = ALIGN(size, PAGE_SIZE);
@@ -462,7 +450,7 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo)
int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
{
/* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
if (0 && (adev->flags & AMDGPU_IS_APU)) {
if (0 && (adev->flags & AMD_IS_APU)) {
/* Useless to evict on IGP chips */
return 0;
}
@@ -478,7 +466,6 @@ void amdgpu_bo_force_delete(struct amdgpu_device *adev)
}
dev_err(adev->dev, "Userspace still has active objects !\n");
list_for_each_entry_safe(bo, n, &adev->gem.objects, list) {
mutex_lock(&adev->ddev->struct_mutex);
dev_err(adev->dev, "%p %p %lu %lu force free\n",
&bo->gem_base, bo, (unsigned long)bo->gem_base.size,
*((unsigned long *)&bo->gem_base.refcount));
@@ -486,8 +473,7 @@ void amdgpu_bo_force_delete(struct amdgpu_device *adev)
list_del_init(&bo->list);
mutex_unlock(&bo->adev->gem.mutex);
/* this should unref the ttm bo */
drm_gem_object_unreference(&bo->gem_base);
mutex_unlock(&adev->ddev->struct_mutex);
drm_gem_object_unreference_unlocked(&bo->gem_base);
}
}
@@ -658,13 +644,13 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
* @shared: true if fence should be added shared
*
*/
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct amdgpu_fence *fence,
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct fence *fence,
bool shared)
{
struct reservation_object *resv = bo->tbo.resv;
if (shared)
reservation_object_add_shared_fence(resv, &fence->base);
reservation_object_add_shared_fence(resv, fence);
else
reservation_object_add_excl_fence(resv, &fence->base);
reservation_object_add_excl_fence(resv, fence);
}
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+1 -1
View File
@@ -161,7 +161,7 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
struct ttm_mem_reg *new_mem);
int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct amdgpu_fence *fence,
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct fence *fence,
bool shared);
/*
drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
+2 -4
View File
@@ -82,7 +82,7 @@ static ssize_t amdgpu_set_dpm_state(struct device *dev,
mutex_unlock(&adev->pm.mutex);
/* Can't set dpm state when the card is off */
if (!(adev->flags & AMDGPU_IS_PX) ||
if (!(adev->flags & AMD_IS_PX) ||
(ddev->switch_power_state == DRM_SWITCH_POWER_ON))
amdgpu_pm_compute_clocks(adev);
fail:
@@ -538,7 +538,7 @@ static void amdgpu_dpm_change_power_state_locked(struct amdgpu_device *adev)
/* vce just modifies an existing state so force a change */
if (ps->vce_active != adev->pm.dpm.vce_active)
goto force;
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
/* for APUs if the num crtcs changed but state is the same,
* all we need to do is update the display configuration.
*/
@@ -580,7 +580,6 @@ force:
amdgpu_dpm_print_power_state(adev, adev->pm.dpm.requested_ps);
}
mutex_lock(&adev->ddev->struct_mutex);
mutex_lock(&adev->ring_lock);
/* update whether vce is active */
@@ -628,7 +627,6 @@ force:
done:
mutex_unlock(&adev->ring_lock);
mutex_unlock(&adev->ddev->struct_mutex);
}
void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable)
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+3 -1
View File
@@ -342,6 +342,8 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
amdgpu_fence_driver_init_ring(ring);
}
init_waitqueue_head(&ring->fence_drv.fence_queue);
r = amdgpu_wb_get(adev, &ring->rptr_offs);
if (r) {
dev_err(adev->dev, "(%d) ring rptr_offs wb alloc failed\n", r);
@@ -367,7 +369,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
}
ring->next_rptr_gpu_addr = adev->wb.gpu_addr + (ring->next_rptr_offs * 4);
ring->next_rptr_cpu_addr = &adev->wb.wb[ring->next_rptr_offs];
spin_lock_init(&ring->fence_lock);
r = amdgpu_fence_driver_start_ring(ring, irq_src, irq_type);
if (r) {
dev_err(adev->dev, "failed initializing fences (%d).\n", r);
drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c
+7 -4
View File
@@ -160,7 +160,8 @@ static void amdgpu_sa_bo_try_free(struct amdgpu_sa_manager *sa_manager)
sa_bo = list_entry(sa_manager->hole->next, struct amdgpu_sa_bo, olist);
list_for_each_entry_safe_from(sa_bo, tmp, &sa_manager->olist, olist) {
if (sa_bo->fence == NULL || !amdgpu_fence_signaled(sa_bo->fence)) {
if (sa_bo->fence == NULL ||
!fence_is_signaled(&sa_bo->fence->base)) {
return;
}
amdgpu_sa_bo_remove_locked(sa_bo);
@@ -274,7 +275,7 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
sa_bo = list_first_entry(&sa_manager->flist[i],
struct amdgpu_sa_bo, flist);
if (!amdgpu_fence_signaled(sa_bo->fence)) {
if (!fence_is_signaled(&sa_bo->fence->base)) {
fences[i] = sa_bo->fence;
continue;
}
@@ -317,6 +318,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
struct amdgpu_fence *fences[AMDGPU_MAX_RINGS];
unsigned tries[AMDGPU_MAX_RINGS];
int i, r;
signed long t;
BUG_ON(align > sa_manager->align);
BUG_ON(size > sa_manager->size);
@@ -350,7 +352,8 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
} while (amdgpu_sa_bo_next_hole(sa_manager, fences, tries));
spin_unlock(&sa_manager->wq.lock);
r = amdgpu_fence_wait_any(adev, fences, false);
t = amdgpu_fence_wait_any(adev, fences, false, MAX_SCHEDULE_TIMEOUT);
r = (t > 0) ? 0 : t;
spin_lock(&sa_manager->wq.lock);
/* if we have nothing to wait for block */
if (r == -ENOENT) {
@@ -379,7 +382,7 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo,
sa_manager = (*sa_bo)->manager;
spin_lock(&sa_manager->wq.lock);
if (fence && !amdgpu_fence_signaled(fence)) {
if (fence && !fence_is_signaled(&fence->base)) {
(*sa_bo)->fence = amdgpu_fence_ref(fence);
list_add_tail(&(*sa_bo)->flist,
&sa_manager->flist[fence->ring->idx]);
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+145
View File
@@ -0,0 +1,145 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#include <linux/kthread.h>
#include <linux/wait.h>
#include <linux/sched.h>
#include <drm/drmP.h>
#include "amdgpu.h"
static int amdgpu_sched_prepare_job(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity,
struct amd_sched_job *job)
{
int r = 0;
struct amdgpu_cs_parser *sched_job;
if (!job || !job->data) {
DRM_ERROR("job is null\n");
return -EINVAL;
}
sched_job = (struct amdgpu_cs_parser *)job->data;
if (sched_job->prepare_job) {
r = sched_job->prepare_job(sched_job);
if (r) {
DRM_ERROR("Prepare job error\n");
schedule_work(&sched_job->job_work);
}
}
return r;
}
static struct fence *amdgpu_sched_run_job(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity,
struct amd_sched_job *job)
{
int r = 0;
struct amdgpu_cs_parser *sched_job;
struct amdgpu_fence *fence;
if (!job || !job->data) {
DRM_ERROR("job is null\n");
return NULL;
}
sched_job = (struct amdgpu_cs_parser *)job->data;
mutex_lock(&sched_job->job_lock);
r = amdgpu_ib_schedule(sched_job->adev,
sched_job->num_ibs,
sched_job->ibs,
sched_job->filp);
if (r)
goto err;
fence = amdgpu_fence_ref(sched_job->ibs[sched_job->num_ibs - 1].fence);
if (sched_job->run_job) {
r = sched_job->run_job(sched_job);
if (r)
goto err;
}
mutex_unlock(&sched_job->job_lock);
return &fence->base;
err:
DRM_ERROR("Run job error\n");
mutex_unlock(&sched_job->job_lock);
schedule_work(&sched_job->job_work);
return NULL;
}
static void amdgpu_sched_process_job(struct amd_gpu_scheduler *sched,
struct amd_sched_job *job)
{
struct amdgpu_cs_parser *sched_job;
if (!job || !job->data) {
DRM_ERROR("job is null\n");
return;
}
sched_job = (struct amdgpu_cs_parser *)job->data;
schedule_work(&sched_job->job_work);
}
struct amd_sched_backend_ops amdgpu_sched_ops = {
.prepare_job = amdgpu_sched_prepare_job,
.run_job = amdgpu_sched_run_job,
.process_job = amdgpu_sched_process_job
};
int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
struct amdgpu_ring *ring,
struct amdgpu_ib *ibs,
unsigned num_ibs,
int (*free_job)(struct amdgpu_cs_parser *),
void *owner,
struct fence **f)
{
int r = 0;
if (amdgpu_enable_scheduler) {
struct amdgpu_cs_parser *sched_job =
amdgpu_cs_parser_create(adev, owner, &adev->kernel_ctx,
ibs, num_ibs);
if(!sched_job) {
return -ENOMEM;
}
sched_job->free_job = free_job;
mutex_lock(&sched_job->job_lock);
r = amd_sched_push_job(ring->scheduler,
&adev->kernel_ctx.rings[ring->idx].entity,
sched_job, &sched_job->s_fence);
if (r) {
mutex_unlock(&sched_job->job_lock);
kfree(sched_job);
return r;
}
ibs[num_ibs - 1].sequence = sched_job->s_fence->v_seq;
*f = fence_get(&sched_job->s_fence->base);
mutex_unlock(&sched_job->job_lock);
} else {
r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
if (r)
return r;
*f = fence_get(&ibs[num_ibs - 1].fence->base);
}
return 0;
}
drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
+36 -22
View File
@@ -53,20 +53,24 @@ void amdgpu_sync_create(struct amdgpu_sync *sync)
}
/**
* amdgpu_sync_fence - use the semaphore to sync to a fence
* amdgpu_sync_fence - remember to sync to this fence
*
* @sync: sync object to add fence to
* @fence: fence to sync to
*
* Sync to the fence using the semaphore objects
*/
void amdgpu_sync_fence(struct amdgpu_sync *sync,
struct amdgpu_fence *fence)
int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
struct fence *f)
{
struct amdgpu_fence *fence;
struct amdgpu_fence *other;
if (!fence)
return;
if (!f)
return 0;
fence = to_amdgpu_fence(f);
if (!fence || fence->ring->adev != adev)
return fence_wait(f, true);
other = sync->sync_to[fence->ring->idx];
sync->sync_to[fence->ring->idx] = amdgpu_fence_ref(
@@ -79,6 +83,8 @@ void amdgpu_sync_fence(struct amdgpu_sync *sync,
amdgpu_fence_later(fence, other));
amdgpu_fence_unref(&other);
}
return 0;
}
/**
@@ -106,11 +112,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
/* always sync to the exclusive fence */
f = reservation_object_get_excl(resv);
fence = f ? to_amdgpu_fence(f) : NULL;
if (fence && fence->ring->adev == adev)
amdgpu_sync_fence(sync, fence);
else if (f)
r = fence_wait(f, true);
r = amdgpu_sync_fence(adev, sync, f);
flist = reservation_object_get_list(resv);
if (!flist || r)
@@ -121,14 +123,26 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
reservation_object_held(resv));
fence = f ? to_amdgpu_fence(f) : NULL;
if (fence && fence->ring->adev == adev) {
if (fence->owner != owner ||
fence->owner == AMDGPU_FENCE_OWNER_UNDEFINED)
amdgpu_sync_fence(sync, fence);
} else if (f) {
r = fence_wait(f, true);
if (r)
break;
/* VM updates are only interesting
* for other VM updates and moves.
*/
if ((owner != AMDGPU_FENCE_OWNER_MOVE) &&
(fence->owner != AMDGPU_FENCE_OWNER_MOVE) &&
((owner == AMDGPU_FENCE_OWNER_VM) !=
(fence->owner == AMDGPU_FENCE_OWNER_VM)))
continue;
/* Ignore fence from the same owner as
* long as it isn't undefined.
*/
if (owner != AMDGPU_FENCE_OWNER_UNDEFINED &&
fence->owner == owner)
continue;
}
r = amdgpu_sync_fence(adev, sync, f);
if (r)
break;
}
return r;
}
@@ -164,9 +178,9 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
return -EINVAL;
}
if (count >= AMDGPU_NUM_SYNCS) {
if (amdgpu_enable_scheduler || (count >= AMDGPU_NUM_SYNCS)) {
/* not enough room, wait manually */
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r)
return r;
continue;
@@ -186,7 +200,7 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
if (!amdgpu_semaphore_emit_signal(other, semaphore)) {
/* signaling wasn't successful wait manually */
amdgpu_ring_undo(other);
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r)
return r;
continue;
@@ -196,7 +210,7 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
if (!amdgpu_semaphore_emit_wait(ring, semaphore)) {
/* waiting wasn't successful wait manually */
amdgpu_ring_undo(other);
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r)
return r;
continue;
drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
+22 -21
View File
@@ -116,7 +116,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
goto out_lclean_unpin;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_lclean_unpin;
@@ -161,7 +161,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
goto out_lclean_unpin;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(&fence->base, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_lclean_unpin;
@@ -238,7 +238,7 @@ void amdgpu_test_moves(struct amdgpu_device *adev)
static int amdgpu_test_create_and_emit_fence(struct amdgpu_device *adev,
struct amdgpu_ring *ring,
struct amdgpu_fence **fence)
struct fence **fence)
{
uint32_t handle = ring->idx ^ 0xdeafbeef;
int r;
@@ -269,15 +269,16 @@ static int amdgpu_test_create_and_emit_fence(struct amdgpu_device *adev,
DRM_ERROR("Failed to get dummy destroy msg\n");
return r;
}
} else {
struct amdgpu_fence *a_fence = NULL;
r = amdgpu_ring_lock(ring, 64);
if (r) {
DRM_ERROR("Failed to lock ring A %d\n", ring->idx);
return r;
}
amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_UNDEFINED, fence);
amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_UNDEFINED, &a_fence);
amdgpu_ring_unlock_commit(ring);
*fence = &a_fence->base;
}
return 0;
}
@@ -286,7 +287,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
struct amdgpu_ring *ringA,
struct amdgpu_ring *ringB)
{
struct amdgpu_fence *fence1 = NULL, *fence2 = NULL;
struct fence *fence1 = NULL, *fence2 = NULL;
struct amdgpu_semaphore *semaphore = NULL;
int r;
@@ -322,7 +323,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
mdelay(1000);
if (amdgpu_fence_signaled(fence1)) {
if (fence_is_signaled(fence1)) {
DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
@@ -335,7 +336,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
amdgpu_semaphore_emit_signal(ringB, semaphore);
amdgpu_ring_unlock_commit(ringB);
r = amdgpu_fence_wait(fence1, false);
r = fence_wait(fence1, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
@@ -343,7 +344,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
mdelay(1000);
if (amdgpu_fence_signaled(fence2)) {
if (fence_is_signaled(fence2)) {
DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
@@ -356,7 +357,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
amdgpu_semaphore_emit_signal(ringB, semaphore);
amdgpu_ring_unlock_commit(ringB);
r = amdgpu_fence_wait(fence2, false);
r = fence_wait(fence2, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
@@ -366,10 +367,10 @@ out_cleanup:
amdgpu_semaphore_free(adev, &semaphore, NULL);
if (fence1)
amdgpu_fence_unref(&fence1);
fence_put(fence1);
if (fence2)
amdgpu_fence_unref(&fence2);
fence_put(fence2);
if (r)
printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
@@ -380,7 +381,7 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
struct amdgpu_ring *ringB,
struct amdgpu_ring *ringC)
{
struct amdgpu_fence *fenceA = NULL, *fenceB = NULL;
struct fence *fenceA = NULL, *fenceB = NULL;
struct amdgpu_semaphore *semaphore = NULL;
bool sigA, sigB;
int i, r;
@@ -416,11 +417,11 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
mdelay(1000);
if (amdgpu_fence_signaled(fenceA)) {
if (fence_is_signaled(fenceA)) {
DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
goto out_cleanup;
}
if (amdgpu_fence_signaled(fenceB)) {
if (fence_is_signaled(fenceB)) {
DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
goto out_cleanup;
}
@@ -435,8 +436,8 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
for (i = 0; i < 30; ++i) {
mdelay(100);
sigA = amdgpu_fence_signaled(fenceA);
sigB = amdgpu_fence_signaled(fenceB);
sigA = fence_is_signaled(fenceA);
sigB = fence_is_signaled(fenceB);
if (sigA || sigB)
break;
}
@@ -461,12 +462,12 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
mdelay(1000);
r = amdgpu_fence_wait(fenceA, false);
r = fence_wait(fenceA, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence A\n");
goto out_cleanup;
}
r = amdgpu_fence_wait(fenceB, false);
r = fence_wait(fenceB, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence B\n");
goto out_cleanup;
@@ -476,10 +477,10 @@ out_cleanup:
amdgpu_semaphore_free(adev, &semaphore, NULL);
if (fenceA)
amdgpu_fence_unref(&fenceA);
fence_put(fenceA);
if (fenceB)
amdgpu_fence_unref(&fenceB);
fence_put(fenceB);
if (r)
printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+65 -37
View File
@@ -52,6 +52,7 @@
#endif
#define FIRMWARE_TONGA "amdgpu/tonga_uvd.bin"
#define FIRMWARE_CARRIZO "amdgpu/carrizo_uvd.bin"
#define FIRMWARE_FIJI "amdgpu/fiji_uvd.bin"
/**
* amdgpu_uvd_cs_ctx - Command submission parser context
@@ -81,6 +82,7 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS);
#endif
MODULE_FIRMWARE(FIRMWARE_TONGA);
MODULE_FIRMWARE(FIRMWARE_CARRIZO);
MODULE_FIRMWARE(FIRMWARE_FIJI);
static void amdgpu_uvd_note_usage(struct amdgpu_device *adev);
static void amdgpu_uvd_idle_work_handler(struct work_struct *work);
@@ -116,6 +118,9 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
case CHIP_TONGA:
fw_name = FIRMWARE_TONGA;
break;
case CHIP_FIJI:
fw_name = FIRMWARE_FIJI;
break;
case CHIP_CARRIZO:
fw_name = FIRMWARE_CARRIZO;
break;
@@ -283,7 +288,7 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
uint32_t handle = atomic_read(&adev->uvd.handles[i]);
if (handle != 0 && adev->uvd.filp[i] == filp) {
struct amdgpu_fence *fence;
struct fence *fence;
amdgpu_uvd_note_usage(adev);
@@ -293,8 +298,8 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
continue;
}
amdgpu_fence_wait(fence, false);
amdgpu_fence_unref(&fence);
fence_wait(fence, false);
fence_put(fence);
adev->uvd.filp[i] = NULL;
atomic_set(&adev->uvd.handles[i], 0);
@@ -374,7 +379,8 @@ static int amdgpu_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
unsigned height_in_mb = ALIGN(height / 16, 2);
unsigned fs_in_mb = width_in_mb * height_in_mb;
unsigned image_size, tmp, min_dpb_size, num_dpb_buffer, min_ctx_size;
unsigned image_size, tmp, min_dpb_size, num_dpb_buffer;
unsigned min_ctx_size = 0;
image_size = width * height;
image_size += image_size / 2;
@@ -507,28 +513,25 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
{
struct amdgpu_device *adev = ctx->parser->adev;
int32_t *msg, msg_type, handle;
struct fence *f;
void *ptr;
int i, r;
long r;
int i;
if (offset & 0x3F) {
DRM_ERROR("UVD messages must be 64 byte aligned!\n");
return -EINVAL;
}
f = reservation_object_get_excl(bo->tbo.resv);
if (f) {
r = amdgpu_fence_wait((struct amdgpu_fence *)f, false);
if (r) {
DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
return r;
}
r = reservation_object_wait_timeout_rcu(bo->tbo.resv, true, false,
MAX_SCHEDULE_TIMEOUT);
if (r < 0) {
DRM_ERROR("Failed waiting for UVD message (%ld)!\n", r);
return r;
}
r = amdgpu_bo_kmap(bo, &ptr);
if (r) {
DRM_ERROR("Failed mapping the UVD message (%d)!\n", r);
DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r);
return r;
}
@@ -803,14 +806,24 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
return 0;
}
static int amdgpu_uvd_free_job(
struct amdgpu_cs_parser *sched_job)
{
amdgpu_ib_free(sched_job->adev, sched_job->ibs);
kfree(sched_job->ibs);
return 0;
}
static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
struct amdgpu_bo *bo,
struct amdgpu_fence **fence)
struct fence **fence)
{
struct ttm_validate_buffer tv;
struct ww_acquire_ctx ticket;
struct list_head head;
struct amdgpu_ib ib;
struct amdgpu_ib *ib = NULL;
struct fence *f = NULL;
struct amdgpu_device *adev = ring->adev;
uint64_t addr;
int i, r;
@@ -832,34 +845,49 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
if (r)
goto err;
r = amdgpu_ib_get(ring, NULL, 64, &ib);
if (r)
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib) {
r = -ENOMEM;
goto err;
}
r = amdgpu_ib_get(ring, NULL, 64, ib);
if (r)
goto err1;
addr = amdgpu_bo_gpu_offset(bo);
ib.ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
ib.ptr[1] = addr;
ib.ptr[2] = PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0);
ib.ptr[3] = addr >> 32;
ib.ptr[4] = PACKET0(mmUVD_GPCOM_VCPU_CMD, 0);
ib.ptr[5] = 0;
ib->ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
ib->ptr[1] = addr;
ib->ptr[2] = PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0);
ib->ptr[3] = addr >> 32;
ib->ptr[4] = PACKET0(mmUVD_GPCOM_VCPU_CMD, 0);
ib->ptr[5] = 0;
for (i = 6; i < 16; ++i)
ib.ptr[i] = PACKET2(0);
ib.length_dw = 16;
ib->ptr[i] = PACKET2(0);
ib->length_dw = 16;
r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_uvd_free_job,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err;
ttm_eu_fence_buffer_objects(&ticket, &head, &ib.fence->base);
goto err2;
ttm_eu_fence_buffer_objects(&ticket, &head, f);
if (fence)
*fence = amdgpu_fence_ref(ib.fence);
amdgpu_ib_free(ring->adev, &ib);
*fence = fence_get(f);
amdgpu_bo_unref(&bo);
return 0;
fence_put(f);
if (amdgpu_enable_scheduler)
return 0;
amdgpu_ib_free(ring->adev, ib);
kfree(ib);
return 0;
err2:
amdgpu_ib_free(ring->adev, ib);
err1:
kfree(ib);
err:
ttm_eu_backoff_reservation(&ticket, &head);
return r;
@@ -869,7 +897,7 @@ err:
crash the vcpu so just try to emmit a dummy create/destroy msg to
avoid this */
int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence)
struct fence **fence)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_bo *bo;
@@ -916,7 +944,7 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
}
int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence)
struct fence **fence)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_bo *bo;
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h
+2 -2
View File
@@ -29,9 +29,9 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev);
int amdgpu_uvd_suspend(struct amdgpu_device *adev);
int amdgpu_uvd_resume(struct amdgpu_device *adev);
int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence);
struct fence **fence);
int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence);
struct fence **fence);
void amdgpu_uvd_free_handles(struct amdgpu_device *adev,
struct drm_file *filp);
int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx);
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+99 -66
View File
@@ -48,6 +48,7 @@
#endif
#define FIRMWARE_TONGA "amdgpu/tonga_vce.bin"
#define FIRMWARE_CARRIZO "amdgpu/carrizo_vce.bin"
#define FIRMWARE_FIJI "amdgpu/fiji_vce.bin"
#ifdef CONFIG_DRM_AMDGPU_CIK
MODULE_FIRMWARE(FIRMWARE_BONAIRE);
@@ -58,6 +59,7 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS);
#endif
MODULE_FIRMWARE(FIRMWARE_TONGA);
MODULE_FIRMWARE(FIRMWARE_CARRIZO);
MODULE_FIRMWARE(FIRMWARE_FIJI);
static void amdgpu_vce_idle_work_handler(struct work_struct *work);
@@ -101,6 +103,9 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
case CHIP_CARRIZO:
fw_name = FIRMWARE_CARRIZO;
break;
case CHIP_FIJI:
fw_name = FIRMWARE_FIJI;
break;
default:
return -EINVAL;
@@ -334,6 +339,14 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
}
}
static int amdgpu_vce_free_job(
struct amdgpu_cs_parser *sched_job)
{
amdgpu_ib_free(sched_job->adev, sched_job->ibs);
kfree(sched_job->ibs);
return 0;
}
/**
* amdgpu_vce_get_create_msg - generate a VCE create msg
*
@@ -345,59 +358,69 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
* Open up a stream for HW test
*/
int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence)
struct fence **fence)
{
const unsigned ib_size_dw = 1024;
struct amdgpu_ib ib;
struct amdgpu_ib *ib = NULL;
struct fence *f = NULL;
struct amdgpu_device *adev = ring->adev;
uint64_t dummy;
int i, r;
r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
if (r) {
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
kfree(ib);
return r;
}
dummy = ib.gpu_addr + 1024;
dummy = ib->gpu_addr + 1024;
/* stitch together an VCE create msg */
ib.length_dw = 0;
ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
ib.ptr[ib.length_dw++] = handle;
ib->length_dw = 0;
ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
ib->ptr[ib->length_dw++] = handle;
ib.ptr[ib.length_dw++] = 0x00000030; /* len */
ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */
ib.ptr[ib.length_dw++] = 0x00000000;
ib.ptr[ib.length_dw++] = 0x00000042;
ib.ptr[ib.length_dw++] = 0x0000000a;
ib.ptr[ib.length_dw++] = 0x00000001;
ib.ptr[ib.length_dw++] = 0x00000080;
ib.ptr[ib.length_dw++] = 0x00000060;
ib.ptr[ib.length_dw++] = 0x00000100;
ib.ptr[ib.length_dw++] = 0x00000100;
ib.ptr[ib.length_dw++] = 0x0000000c;
ib.ptr[ib.length_dw++] = 0x00000000;
ib->ptr[ib->length_dw++] = 0x00000030; /* len */
ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
ib->ptr[ib->length_dw++] = 0x00000000;
ib->ptr[ib->length_dw++] = 0x00000042;
ib->ptr[ib->length_dw++] = 0x0000000a;
ib->ptr[ib->length_dw++] = 0x00000001;
ib->ptr[ib->length_dw++] = 0x00000080;
ib->ptr[ib->length_dw++] = 0x00000060;
ib->ptr[ib->length_dw++] = 0x00000100;
ib->ptr[ib->length_dw++] = 0x00000100;
ib->ptr[ib->length_dw++] = 0x0000000c;
ib->ptr[ib->length_dw++] = 0x00000000;
ib.ptr[ib.length_dw++] = 0x00000014; /* len */
ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
ib.ptr[ib.length_dw++] = dummy;
ib.ptr[ib.length_dw++] = 0x00000001;
ib->ptr[ib->length_dw++] = 0x00000014; /* len */
ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
ib->ptr[ib->length_dw++] = dummy;
ib->ptr[ib->length_dw++] = 0x00000001;
for (i = ib.length_dw; i < ib_size_dw; ++i)
ib.ptr[i] = 0x0;
r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
if (r) {
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
}
for (i = ib->length_dw; i < ib_size_dw; ++i)
ib->ptr[i] = 0x0;
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_vce_free_job,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err;
if (fence)
*fence = amdgpu_fence_ref(ib.fence);
amdgpu_ib_free(ring->adev, &ib);
*fence = fence_get(f);
fence_put(f);
if (amdgpu_enable_scheduler)
return 0;
err:
amdgpu_ib_free(adev, ib);
kfree(ib);
return r;
}
@@ -412,49 +435,59 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
* Close up a stream for HW test or if userspace failed to do so
*/
int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence)
struct fence **fence)
{
const unsigned ib_size_dw = 1024;
struct amdgpu_ib ib;
struct amdgpu_ib *ib = NULL;
struct fence *f = NULL;
struct amdgpu_device *adev = ring->adev;
uint64_t dummy;
int i, r;
r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
if (r) {
kfree(ib);
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
return r;
}
dummy = ib.gpu_addr + 1024;
dummy = ib->gpu_addr + 1024;
/* stitch together an VCE destroy msg */
ib.length_dw = 0;
ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
ib.ptr[ib.length_dw++] = handle;
ib->length_dw = 0;
ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
ib->ptr[ib->length_dw++] = handle;
ib.ptr[ib.length_dw++] = 0x00000014; /* len */
ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
ib.ptr[ib.length_dw++] = dummy;
ib.ptr[ib.length_dw++] = 0x00000001;
ib->ptr[ib->length_dw++] = 0x00000014; /* len */
ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
ib->ptr[ib->length_dw++] = dummy;
ib->ptr[ib->length_dw++] = 0x00000001;
ib.ptr[ib.length_dw++] = 0x00000008; /* len */
ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */
for (i = ib.length_dw; i < ib_size_dw; ++i)
ib.ptr[i] = 0x0;
r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
if (r) {
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
}
ib->ptr[ib->length_dw++] = 0x00000008; /* len */
ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */
for (i = ib->length_dw; i < ib_size_dw; ++i)
ib->ptr[i] = 0x0;
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_vce_free_job,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err;
if (fence)
*fence = amdgpu_fence_ref(ib.fence);
amdgpu_ib_free(ring->adev, &ib);
*fence = fence_get(f);
fence_put(f);
if (amdgpu_enable_scheduler)
return 0;
err:
amdgpu_ib_free(adev, ib);
kfree(ib);
return r;
}
@@ -800,7 +833,7 @@ int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
*/
int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_fence *fence = NULL;
struct fence *fence = NULL;
int r;
r = amdgpu_vce_get_create_msg(ring, 1, NULL);
@@ -815,13 +848,13 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
goto error;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(fence, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
} else {
DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
}
error:
amdgpu_fence_unref(&fence);
fence_put(fence);
return r;
}
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h
+2 -2
View File
@@ -29,9 +29,9 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev);
int amdgpu_vce_suspend(struct amdgpu_device *adev);
int amdgpu_vce_resume(struct amdgpu_device *adev);
int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence);
struct fence **fence);
int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
struct amdgpu_fence **fence);
struct fence **fence);
void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp);
int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx);
bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring,
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+177 -95
View File
@@ -127,16 +127,16 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
/**
* amdgpu_vm_grab_id - allocate the next free VMID
*
* @ring: ring we want to submit job to
* @vm: vm to allocate id for
* @ring: ring we want to submit job to
* @sync: sync object where we add dependencies
*
* Allocate an id for the vm (cayman+).
* Returns the fence we need to sync to (if any).
* Allocate an id for the vm, adding fences to the sync obj as necessary.
*
* Global and local mutex must be locked!
* Global mutex must be locked!
*/
struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
struct amdgpu_vm *vm)
int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
struct amdgpu_sync *sync)
{
struct amdgpu_fence *best[AMDGPU_MAX_RINGS] = {};
struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
@@ -148,7 +148,7 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
/* check if the id is still valid */
if (vm_id->id && vm_id->last_id_use &&
vm_id->last_id_use == adev->vm_manager.active[vm_id->id])
return NULL;
return 0;
/* we definately need to flush */
vm_id->pd_gpu_addr = ~0ll;
@@ -161,7 +161,7 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
/* found a free one */
vm_id->id = i;
trace_amdgpu_vm_grab_id(i, ring->idx);
return NULL;
return 0;
}
if (amdgpu_fence_is_earlier(fence, best[fence->ring->idx])) {
@@ -172,15 +172,19 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
for (i = 0; i < 2; ++i) {
if (choices[i]) {
struct amdgpu_fence *fence;
fence = adev->vm_manager.active[choices[i]];
vm_id->id = choices[i];
trace_amdgpu_vm_grab_id(choices[i], ring->idx);
return adev->vm_manager.active[choices[i]];
return amdgpu_sync_fence(ring->adev, sync, &fence->base);
}
}
/* should never happen */
BUG();
return NULL;
return -EINVAL;
}
/**
@@ -200,13 +204,15 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
{
uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory);
struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
struct amdgpu_fence *flushed_updates = vm_id->flushed_updates;
if (pd_addr != vm_id->pd_gpu_addr || !vm_id->flushed_updates ||
amdgpu_fence_is_earlier(vm_id->flushed_updates, updates)) {
if (pd_addr != vm_id->pd_gpu_addr || !flushed_updates ||
(updates && amdgpu_fence_is_earlier(flushed_updates, updates))) {
trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id);
amdgpu_fence_unref(&vm_id->flushed_updates);
vm_id->flushed_updates = amdgpu_fence_ref(updates);
vm_id->flushed_updates = amdgpu_fence_ref(
amdgpu_fence_later(flushed_updates, updates));
amdgpu_fence_unref(&flushed_updates);
vm_id->pd_gpu_addr = pd_addr;
amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr);
}
@@ -300,6 +306,16 @@ static void amdgpu_vm_update_pages(struct amdgpu_device *adev,
}
}
static int amdgpu_vm_free_job(
struct amdgpu_cs_parser *sched_job)
{
int i;
for (i = 0; i < sched_job->num_ibs; i++)
amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
kfree(sched_job->ibs);
return 0;
}
/**
* amdgpu_vm_clear_bo - initially clear the page dir/table
*
@@ -310,7 +326,8 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
struct amdgpu_bo *bo)
{
struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
struct amdgpu_ib ib;
struct fence *fence = NULL;
struct amdgpu_ib *ib;
unsigned entries;
uint64_t addr;
int r;
@@ -330,24 +347,33 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
addr = amdgpu_bo_gpu_offset(bo);
entries = amdgpu_bo_size(bo) / 8;
r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, &ib);
if (r)
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib)
goto error_unreserve;
ib.length_dw = 0;
amdgpu_vm_update_pages(adev, &ib, addr, 0, entries, 0, 0, 0);
amdgpu_vm_pad_ib(adev, &ib);
WARN_ON(ib.length_dw > 64);
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, ib);
if (r)
goto error_free;
amdgpu_bo_fence(bo, ib.fence, true);
ib->length_dw = 0;
amdgpu_vm_update_pages(adev, ib, addr, 0, entries, 0, 0, 0);
amdgpu_vm_pad_ib(adev, ib);
WARN_ON(ib->length_dw > 64);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_vm_free_job,
AMDGPU_FENCE_OWNER_VM,
&fence);
if (!r)
amdgpu_bo_fence(bo, fence, true);
fence_put(fence);
if (amdgpu_enable_scheduler) {
amdgpu_bo_unreserve(bo);
return 0;
}
error_free:
amdgpu_ib_free(adev, &ib);
amdgpu_ib_free(adev, ib);
kfree(ib);
error_unreserve:
amdgpu_bo_unreserve(bo);
@@ -400,7 +426,9 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
uint32_t incr = AMDGPU_VM_PTE_COUNT * 8;
uint64_t last_pde = ~0, last_pt = ~0;
unsigned count = 0, pt_idx, ndw;
struct amdgpu_ib ib;
struct amdgpu_ib *ib;
struct fence *fence = NULL;
int r;
/* padding, etc. */
@@ -413,10 +441,14 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
if (ndw > 0xfffff)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ndw * 4, &ib);
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
if (r)
return r;
ib.length_dw = 0;
ib->length_dw = 0;
/* walk over the address space and update the page directory */
for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) {
@@ -436,7 +468,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
((last_pt + incr * count) != pt)) {
if (count) {
amdgpu_vm_update_pages(adev, &ib, last_pde,
amdgpu_vm_update_pages(adev, ib, last_pde,
last_pt, count, incr,
AMDGPU_PTE_VALID, 0);
}
@@ -450,23 +482,37 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
}
if (count)
amdgpu_vm_update_pages(adev, &ib, last_pde, last_pt, count,
amdgpu_vm_update_pages(adev, ib, last_pde, last_pt, count,
incr, AMDGPU_PTE_VALID, 0);
if (ib.length_dw != 0) {
amdgpu_vm_pad_ib(adev, &ib);
amdgpu_sync_resv(adev, &ib.sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
WARN_ON(ib.length_dw > ndw);
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
if (r) {
amdgpu_ib_free(adev, &ib);
return r;
}
amdgpu_bo_fence(pd, ib.fence, true);
if (ib->length_dw != 0) {
amdgpu_vm_pad_ib(adev, ib);
amdgpu_sync_resv(adev, &ib->sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
WARN_ON(ib->length_dw > ndw);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_vm_free_job,
AMDGPU_FENCE_OWNER_VM,
&fence);
if (r)
goto error_free;
amdgpu_bo_fence(pd, fence, true);
fence_put(vm->page_directory_fence);
vm->page_directory_fence = fence_get(fence);
fence_put(fence);
}
if (!amdgpu_enable_scheduler || ib->length_dw == 0) {
amdgpu_ib_free(adev, ib);
kfree(ib);
}
amdgpu_ib_free(adev, &ib);
return 0;
error_free:
amdgpu_ib_free(adev, ib);
kfree(ib);
return r;
}
/**
@@ -640,7 +686,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
*/
static void amdgpu_vm_fence_pts(struct amdgpu_vm *vm,
uint64_t start, uint64_t end,
struct amdgpu_fence *fence)
struct fence *fence)
{
unsigned i;
@@ -670,12 +716,13 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
struct amdgpu_vm *vm,
struct amdgpu_bo_va_mapping *mapping,
uint64_t addr, uint32_t gtt_flags,
struct amdgpu_fence **fence)
struct fence **fence)
{
struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
unsigned nptes, ncmds, ndw;
uint32_t flags = gtt_flags;
struct amdgpu_ib ib;
struct amdgpu_ib *ib;
struct fence *f = NULL;
int r;
/* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here
@@ -722,46 +769,65 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
if (ndw > 0xfffff)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ndw * 4, &ib);
if (r)
ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
if (!ib)
return -ENOMEM;
r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
if (r) {
kfree(ib);
return r;
ib.length_dw = 0;
}
ib->length_dw = 0;
if (!(flags & AMDGPU_PTE_VALID)) {
unsigned i;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
struct amdgpu_fence *f = vm->ids[i].last_id_use;
amdgpu_sync_fence(&ib.sync, f);
r = amdgpu_sync_fence(adev, &ib->sync, &f->base);
if (r)
return r;
}
}
r = amdgpu_vm_update_ptes(adev, vm, &ib, mapping->it.start,
r = amdgpu_vm_update_ptes(adev, vm, ib, mapping->it.start,
mapping->it.last + 1, addr + mapping->offset,
flags, gtt_flags);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_ib_free(adev, ib);
kfree(ib);
return r;
}
amdgpu_vm_pad_ib(adev, &ib);
WARN_ON(ib.length_dw > ndw);
amdgpu_vm_pad_ib(adev, ib);
WARN_ON(ib->length_dw > ndw);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
&amdgpu_vm_free_job,
AMDGPU_FENCE_OWNER_VM,
&f);
if (r)
goto error_free;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
if (r) {
amdgpu_ib_free(adev, &ib);
return r;
}
amdgpu_vm_fence_pts(vm, mapping->it.start,
mapping->it.last + 1, ib.fence);
mapping->it.last + 1, f);
if (fence) {
amdgpu_fence_unref(fence);
*fence = amdgpu_fence_ref(ib.fence);
fence_put(*fence);
*fence = fence_get(f);
}
fence_put(f);
if (!amdgpu_enable_scheduler) {
amdgpu_ib_free(adev, ib);
kfree(ib);
}
amdgpu_ib_free(adev, &ib);
return 0;
error_free:
amdgpu_ib_free(adev, ib);
kfree(ib);
return r;
}
/**
@@ -794,21 +860,25 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
addr = 0;
}
if (addr == bo_va->addr)
return 0;
flags = amdgpu_ttm_tt_pte_flags(adev, bo_va->bo->tbo.ttm, mem);
list_for_each_entry(mapping, &bo_va->mappings, list) {
spin_lock(&vm->status_lock);
if (!list_empty(&bo_va->vm_status))
list_splice_init(&bo_va->valids, &bo_va->invalids);
spin_unlock(&vm->status_lock);
list_for_each_entry(mapping, &bo_va->invalids, list) {
r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, addr,
flags, &bo_va->last_pt_update);
if (r)
return r;
}
bo_va->addr = addr;
spin_lock(&vm->status_lock);
list_splice_init(&bo_va->invalids, &bo_va->valids);
list_del_init(&bo_va->vm_status);
if (!mem)
list_add(&bo_va->vm_status, &vm->cleared);
spin_unlock(&vm->status_lock);
return 0;
@@ -861,7 +931,7 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
struct amdgpu_vm *vm, struct amdgpu_sync *sync)
{
struct amdgpu_bo_va *bo_va = NULL;
int r;
int r = 0;
spin_lock(&vm->status_lock);
while (!list_empty(&vm->invalidated)) {
@@ -878,8 +948,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
spin_unlock(&vm->status_lock);
if (bo_va)
amdgpu_sync_fence(sync, bo_va->last_pt_update);
return 0;
r = amdgpu_sync_fence(adev, sync, bo_va->last_pt_update);
return r;
}
/**
@@ -907,10 +978,10 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
}
bo_va->vm = vm;
bo_va->bo = bo;
bo_va->addr = 0;
bo_va->ref_count = 1;
INIT_LIST_HEAD(&bo_va->bo_list);
INIT_LIST_HEAD(&bo_va->mappings);
INIT_LIST_HEAD(&bo_va->valids);
INIT_LIST_HEAD(&bo_va->invalids);
INIT_LIST_HEAD(&bo_va->vm_status);
mutex_lock(&vm->mutex);
@@ -999,12 +1070,10 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
mapping->offset = offset;
mapping->flags = flags;
list_add(&mapping->list, &bo_va->mappings);
list_add(&mapping->list, &bo_va->invalids);
interval_tree_insert(&mapping->it, &vm->va);
trace_amdgpu_vm_bo_map(bo_va, mapping);
bo_va->addr = 0;
/* Make sure the page tables are allocated */
saddr >>= amdgpu_vm_block_size;
eaddr >>= amdgpu_vm_block_size;
@@ -1085,17 +1154,27 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
{
struct amdgpu_bo_va_mapping *mapping;
struct amdgpu_vm *vm = bo_va->vm;
bool valid = true;
saddr /= AMDGPU_GPU_PAGE_SIZE;
list_for_each_entry(mapping, &bo_va->mappings, list) {
list_for_each_entry(mapping, &bo_va->valids, list) {
if (mapping->it.start == saddr)
break;
}
if (&mapping->list == &bo_va->mappings) {
amdgpu_bo_unreserve(bo_va->bo);
return -ENOENT;
if (&mapping->list == &bo_va->valids) {
valid = false;
list_for_each_entry(mapping, &bo_va->invalids, list) {
if (mapping->it.start == saddr)
break;
}
if (&mapping->list == &bo_va->invalids) {
amdgpu_bo_unreserve(bo_va->bo);
return -ENOENT;
}
}
mutex_lock(&vm->mutex);
@@ -1103,12 +1182,10 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
interval_tree_remove(&mapping->it, &vm->va);
trace_amdgpu_vm_bo_unmap(bo_va, mapping);
if (bo_va->addr) {
/* clear the old address */
if (valid)
list_add(&mapping->list, &vm->freed);
} else {
else
kfree(mapping);
}
mutex_unlock(&vm->mutex);
amdgpu_bo_unreserve(bo_va->bo);
@@ -1139,16 +1216,19 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
list_del(&bo_va->vm_status);
spin_unlock(&vm->status_lock);
list_for_each_entry_safe(mapping, next, &bo_va->mappings, list) {
list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
list_del(&mapping->list);
interval_tree_remove(&mapping->it, &vm->va);
trace_amdgpu_vm_bo_unmap(bo_va, mapping);
if (bo_va->addr)
list_add(&mapping->list, &vm->freed);
else
kfree(mapping);
list_add(&mapping->list, &vm->freed);
}
amdgpu_fence_unref(&bo_va->last_pt_update);
list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
list_del(&mapping->list);
interval_tree_remove(&mapping->it, &vm->va);
kfree(mapping);
}
fence_put(bo_va->last_pt_update);
kfree(bo_va);
mutex_unlock(&vm->mutex);
@@ -1169,12 +1249,10 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
struct amdgpu_bo_va *bo_va;
list_for_each_entry(bo_va, &bo->va, bo_list) {
if (bo_va->addr) {
spin_lock(&bo_va->vm->status_lock);
list_del(&bo_va->vm_status);
spin_lock(&bo_va->vm->status_lock);
if (list_empty(&bo_va->vm_status))
list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
spin_unlock(&bo_va->vm->status_lock);
}
spin_unlock(&bo_va->vm->status_lock);
}
}
@@ -1202,6 +1280,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
vm->va = RB_ROOT;
spin_lock_init(&vm->status_lock);
INIT_LIST_HEAD(&vm->invalidated);
INIT_LIST_HEAD(&vm->cleared);
INIT_LIST_HEAD(&vm->freed);
pd_size = amdgpu_vm_directory_size(adev);
@@ -1215,6 +1294,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
return -ENOMEM;
}
vm->page_directory_fence = NULL;
r = amdgpu_bo_create(adev, pd_size, align, true,
AMDGPU_GEM_DOMAIN_VRAM, 0,
NULL, &vm->page_directory);
@@ -1263,6 +1344,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
kfree(vm->page_tables);
amdgpu_bo_unref(&vm->page_directory);
fence_put(vm->page_directory_fence);
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
amdgpu_fence_unref(&vm->ids[i].flushed_updates);
drivers/gpu/drm/amd/amdgpu/atombios_encoders.c
+1 -1
View File
@@ -812,7 +812,7 @@ amdgpu_atombios_encoder_setup_dig_transmitter(struct drm_encoder *encoder, int a
else
args.v1.ucConfig |= ATOM_TRANSMITTER_CONFIG_DIG1_ENCODER;
if ((adev->flags & AMDGPU_IS_APU) &&
if ((adev->flags & AMD_IS_APU) &&
(amdgpu_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_UNIPHY)) {
if (is_dp ||
!amdgpu_dig_monitor_is_duallink(encoder, amdgpu_encoder->pixel_clock)) {
drivers/gpu/drm/amd/amdgpu/cik.c
+6 -6
View File
@@ -838,7 +838,7 @@ static u32 cik_get_xclk(struct amdgpu_device *adev)
{
u32 reference_clock = adev->clock.spll.reference_freq;
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
if (RREG32_SMC(ixGENERAL_PWRMGT) & GENERAL_PWRMGT__GPU_COUNTER_CLK_MASK)
return reference_clock / 2;
} else {
@@ -1235,7 +1235,7 @@ static void cik_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask)
if (reset_mask & AMDGPU_RESET_VMC)
srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_VMC_MASK;
if (!(adev->flags & AMDGPU_IS_APU)) {
if (!(adev->flags & AMD_IS_APU)) {
if (reset_mask & AMDGPU_RESET_MC)
srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_MC_MASK;
}
@@ -1411,7 +1411,7 @@ static void cik_gpu_pci_config_reset(struct amdgpu_device *adev)
dev_warn(adev->dev, "Wait for MC idle timed out !\n");
}
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
kv_save_regs_for_reset(adev, &kv_save);
/* disable BM */
@@ -1429,7 +1429,7 @@ static void cik_gpu_pci_config_reset(struct amdgpu_device *adev)
}
/* does asic init need to be run first??? */
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
kv_restore_regs_for_reset(adev, &kv_save);
}
@@ -1570,7 +1570,7 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev)
if (amdgpu_pcie_gen2 == 0)
return;
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return;
ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask);
@@ -1730,7 +1730,7 @@ static void cik_program_aspm(struct amdgpu_device *adev)
return;
/* XXX double check APUs */
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return;
orig = data = RREG32_PCIE(ixPCIE_LC_N_FTS_CNTL);
drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+16 -15
View File
@@ -614,6 +614,7 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct fence *f = NULL;
unsigned i;
unsigned index;
int r;
@@ -629,12 +630,10 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
return r;
goto err0;
}
ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
@@ -643,20 +642,16 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
ib.ptr[3] = 1;
ib.ptr[4] = 0xDEADBEEF;
ib.length_dw = 5;
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err1;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
r = fence_wait(f, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
return r;
goto err1;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -666,12 +661,17 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
ring->idx, i);
goto err1;
} else {
DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
err1:
fence_put(f);
amdgpu_ib_free(adev, &ib);
err0:
amdgpu_wb_free(adev, index);
return r;
}
@@ -1404,5 +1404,6 @@ static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev)
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs;
adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
+25 -2
View File
@@ -126,9 +126,31 @@ static const u32 tonga_mgcg_cgcg_init[] =
mmXDMA_MEM_POWER_CNTL, 0x00000101, 0x00000000,
};
static const u32 golden_settings_fiji_a10[] =
{
mmDCI_CLK_CNTL, 0x00000080, 0x00000000,
mmFBC_DEBUG_COMP, 0x000000f0, 0x00000070,
mmFBC_MISC, 0x1f311fff, 0x12300000,
mmHDMI_CONTROL, 0x31000111, 0x00000011,
};
static const u32 fiji_mgcg_cgcg_init[] =
{
mmXDMA_CLOCK_GATING_CNTL, 0xffffffff, 0x00000100,
mmXDMA_MEM_POWER_CNTL, 0x00000101, 0x00000000,
};
static void dce_v10_0_init_golden_registers(struct amdgpu_device *adev)
{
switch (adev->asic_type) {
case CHIP_FIJI:
amdgpu_program_register_sequence(adev,
fiji_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
amdgpu_program_register_sequence(adev,
golden_settings_fiji_a10,
(const u32)ARRAY_SIZE(golden_settings_fiji_a10));
break;
case CHIP_TONGA:
amdgpu_program_register_sequence(adev,
tonga_mgcg_cgcg_init,
@@ -803,11 +825,11 @@ static u32 dce_v10_0_line_buffer_adjust(struct amdgpu_device *adev,
buffer_alloc = 2;
} else if (mode->crtc_hdisplay < 4096) {
mem_cfg = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
} else {
DRM_DEBUG_KMS("Mode too big for LB!\n");
mem_cfg = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
}
} else {
mem_cfg = 1;
@@ -2888,6 +2910,7 @@ static int dce_v10_0_early_init(void *handle)
dce_v10_0_set_irq_funcs(adev);
switch (adev->asic_type) {
case CHIP_FIJI:
case CHIP_TONGA:
adev->mode_info.num_crtc = 6; /* XXX 7??? */
adev->mode_info.num_hpd = 6;
drivers/gpu/drm/amd/amdgpu/dce_v11_0.c
+2 -2
View File
@@ -801,11 +801,11 @@ static u32 dce_v11_0_line_buffer_adjust(struct amdgpu_device *adev,
buffer_alloc = 2;
} else if (mode->crtc_hdisplay < 4096) {
mem_cfg = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
} else {
DRM_DEBUG_KMS("Mode too big for LB!\n");
mem_cfg = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
}
} else {
mem_cfg = 1;
drivers/gpu/drm/amd/amdgpu/dce_v8_0.c
+2 -2
View File
@@ -770,11 +770,11 @@ static u32 dce_v8_0_line_buffer_adjust(struct amdgpu_device *adev,
buffer_alloc = 2;
} else if (mode->crtc_hdisplay < 4096) {
tmp = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
} else {
DRM_DEBUG_KMS("Mode too big for LB!\n");
tmp = 0;
buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
}
} else {
tmp = 1;
drivers/gpu/drm/amd/amdgpu/fiji_dpm.c
+181
View File
@@ -0,0 +1,181 @@
/*
* Copyright 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*/
#include <linux/firmware.h>
#include "drmP.h"
#include "amdgpu.h"
#include "fiji_smumgr.h"
MODULE_FIRMWARE("amdgpu/fiji_smc.bin");
static void fiji_dpm_set_funcs(struct amdgpu_device *adev);
static int fiji_dpm_early_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
fiji_dpm_set_funcs(adev);
return 0;
}
static int fiji_dpm_init_microcode(struct amdgpu_device *adev)
{
char fw_name[30] = "amdgpu/fiji_smc.bin";
int err;
err = request_firmware(&adev->pm.fw, fw_name, adev->dev);
if (err)
goto out;
err = amdgpu_ucode_validate(adev->pm.fw);
out:
if (err) {
DRM_ERROR("Failed to load firmware \"%s\"", fw_name);
release_firmware(adev->pm.fw);
adev->pm.fw = NULL;
}
return err;
}
static int fiji_dpm_sw_init(void *handle)
{
int ret;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
ret = fiji_dpm_init_microcode(adev);
if (ret)
return ret;
return 0;
}
static int fiji_dpm_sw_fini(void *handle)
{
return 0;
}
static int fiji_dpm_hw_init(void *handle)
{
int ret;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
mutex_lock(&adev->pm.mutex);
ret = fiji_smu_init(adev);
if (ret) {
DRM_ERROR("SMU initialization failed\n");
goto fail;
}
ret = fiji_smu_start(adev);
if (ret) {
DRM_ERROR("SMU start failed\n");
goto fail;
}
mutex_unlock(&adev->pm.mutex);
return 0;
fail:
adev->firmware.smu_load = false;
mutex_unlock(&adev->pm.mutex);
return -EINVAL;
}
static int fiji_dpm_hw_fini(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
mutex_lock(&adev->pm.mutex);
fiji_smu_fini(adev);
mutex_unlock(&adev->pm.mutex);
return 0;
}
static int fiji_dpm_suspend(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
fiji_dpm_hw_fini(adev);
return 0;
}
static int fiji_dpm_resume(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
fiji_dpm_hw_init(adev);
return 0;
}
static int fiji_dpm_set_clockgating_state(void *handle,
enum amd_clockgating_state state)
{
return 0;
}
static int fiji_dpm_set_powergating_state(void *handle,
enum amd_powergating_state state)
{
return 0;
}
const struct amd_ip_funcs fiji_dpm_ip_funcs = {
.early_init = fiji_dpm_early_init,
.late_init = NULL,
.sw_init = fiji_dpm_sw_init,
.sw_fini = fiji_dpm_sw_fini,
.hw_init = fiji_dpm_hw_init,
.hw_fini = fiji_dpm_hw_fini,
.suspend = fiji_dpm_suspend,
.resume = fiji_dpm_resume,
.is_idle = NULL,
.wait_for_idle = NULL,
.soft_reset = NULL,
.print_status = NULL,
.set_clockgating_state = fiji_dpm_set_clockgating_state,
.set_powergating_state = fiji_dpm_set_powergating_state,
};
static const struct amdgpu_dpm_funcs fiji_dpm_funcs = {
.get_temperature = NULL,
.pre_set_power_state = NULL,
.set_power_state = NULL,
.post_set_power_state = NULL,
.display_configuration_changed = NULL,
.get_sclk = NULL,
.get_mclk = NULL,
.print_power_state = NULL,
.debugfs_print_current_performance_level = NULL,
.force_performance_level = NULL,
.vblank_too_short = NULL,
.powergate_uvd = NULL,
};
static void fiji_dpm_set_funcs(struct amdgpu_device *adev)
{
if (NULL == adev->pm.funcs)
adev->pm.funcs = &fiji_dpm_funcs;
}
drivers/gpu/drm/amd/amdgpu/fiji_ppsmc.h
+182
View File
@@ -0,0 +1,182 @@
/*
* Copyright 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*/
#ifndef FIJI_PP_SMC_H
#define FIJI_PP_SMC_H
#pragma pack(push, 1)
#define PPSMC_SWSTATE_FLAG_DC 0x01
#define PPSMC_SWSTATE_FLAG_UVD 0x02
#define PPSMC_SWSTATE_FLAG_VCE 0x04
#define PPSMC_THERMAL_PROTECT_TYPE_INTERNAL 0x00
#define PPSMC_THERMAL_PROTECT_TYPE_EXTERNAL 0x01
#define PPSMC_THERMAL_PROTECT_TYPE_NONE 0xff
#define PPSMC_SYSTEMFLAG_GPIO_DC 0x01
#define PPSMC_SYSTEMFLAG_STEPVDDC 0x02
#define PPSMC_SYSTEMFLAG_GDDR5 0x04
#define PPSMC_SYSTEMFLAG_DISABLE_BABYSTEP 0x08
#define PPSMC_SYSTEMFLAG_REGULATOR_HOT 0x10
#define PPSMC_SYSTEMFLAG_REGULATOR_HOT_ANALOG 0x20
#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_MASK 0x07
#define PPSMC_EXTRAFLAGS_AC2DC_DONT_WAIT_FOR_VBLANK 0x08
#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTODPMLOWSTATE 0x00
#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTOINITIALSTATE 0x01
#define PPSMC_DPM2FLAGS_TDPCLMP 0x01
#define PPSMC_DPM2FLAGS_PWRSHFT 0x02
#define PPSMC_DPM2FLAGS_OCP 0x04
#define PPSMC_DISPLAY_WATERMARK_LOW 0
#define PPSMC_DISPLAY_WATERMARK_HIGH 1
#define PPSMC_STATEFLAG_AUTO_PULSE_SKIP 0x01
#define PPSMC_STATEFLAG_POWERBOOST 0x02
#define PPSMC_STATEFLAG_PSKIP_ON_TDP_FAULT 0x04
#define PPSMC_STATEFLAG_POWERSHIFT 0x08
#define PPSMC_STATEFLAG_SLOW_READ_MARGIN 0x10
#define PPSMC_STATEFLAG_DEEPSLEEP_THROTTLE 0x20
#define PPSMC_STATEFLAG_DEEPSLEEP_BYPASS 0x40
#define FDO_MODE_HARDWARE 0
#define FDO_MODE_PIECE_WISE_LINEAR 1
enum FAN_CONTROL {
FAN_CONTROL_FUZZY,
FAN_CONTROL_TABLE
};
//Gemini Modes
#define PPSMC_GeminiModeNone 0 //Single GPU board
#define PPSMC_GeminiModeMaster 1 //Master GPU on a Gemini board
#define PPSMC_GeminiModeSlave 2 //Slave GPU on a Gemini board
#define PPSMC_Result_OK ((uint16_t)0x01)
#define PPSMC_Result_NoMore ((uint16_t)0x02)
#define PPSMC_Result_NotNow ((uint16_t)0x03)
#define PPSMC_Result_Failed ((uint16_t)0xFF)
#define PPSMC_Result_UnknownCmd ((uint16_t)0xFE)
#define PPSMC_Result_UnknownVT ((uint16_t)0xFD)
typedef uint16_t PPSMC_Result;
#define PPSMC_isERROR(x) ((uint16_t)0x80 & (x))
#define PPSMC_MSG_Halt ((uint16_t)0x10)
#define PPSMC_MSG_Resume ((uint16_t)0x11)
#define PPSMC_MSG_EnableDPMLevel ((uint16_t)0x12)
#define PPSMC_MSG_ZeroLevelsDisabled ((uint16_t)0x13)
#define PPSMC_MSG_OneLevelsDisabled ((uint16_t)0x14)
#define PPSMC_MSG_TwoLevelsDisabled ((uint16_t)0x15)
#define PPSMC_MSG_EnableThermalInterrupt ((uint16_t)0x16)
#define PPSMC_MSG_RunningOnAC ((uint16_t)0x17)
#define PPSMC_MSG_LevelUp ((uint16_t)0x18)
#define PPSMC_MSG_LevelDown ((uint16_t)0x19)
#define PPSMC_MSG_ResetDPMCounters ((uint16_t)0x1a)
#define PPSMC_MSG_SwitchToSwState ((uint16_t)0x20)
#define PPSMC_MSG_SwitchToSwStateLast ((uint16_t)0x3f)
#define PPSMC_MSG_SwitchToInitialState ((uint16_t)0x40)
#define PPSMC_MSG_NoForcedLevel ((uint16_t)0x41)
#define PPSMC_MSG_ForceHigh ((uint16_t)0x42)
#define PPSMC_MSG_ForceMediumOrHigh ((uint16_t)0x43)
#define PPSMC_MSG_SwitchToMinimumPower ((uint16_t)0x51)
#define PPSMC_MSG_ResumeFromMinimumPower ((uint16_t)0x52)
#define PPSMC_MSG_EnableCac ((uint16_t)0x53)
#define PPSMC_MSG_DisableCac ((uint16_t)0x54)
#define PPSMC_DPMStateHistoryStart ((uint16_t)0x55)
#define PPSMC_DPMStateHistoryStop ((uint16_t)0x56)
#define PPSMC_CACHistoryStart ((uint16_t)0x57)
#define PPSMC_CACHistoryStop ((uint16_t)0x58)
#define PPSMC_TDPClampingActive ((uint16_t)0x59)
#define PPSMC_TDPClampingInactive ((uint16_t)0x5A)
#define PPSMC_StartFanControl ((uint16_t)0x5B)
#define PPSMC_StopFanControl ((uint16_t)0x5C)
#define PPSMC_NoDisplay ((uint16_t)0x5D)
#define PPSMC_HasDisplay ((uint16_t)0x5E)
#define PPSMC_MSG_UVDPowerOFF ((uint16_t)0x60)
#define PPSMC_MSG_UVDPowerON ((uint16_t)0x61)
#define PPSMC_MSG_EnableULV ((uint16_t)0x62)
#define PPSMC_MSG_DisableULV ((uint16_t)0x63)
#define PPSMC_MSG_EnterULV ((uint16_t)0x64)
#define PPSMC_MSG_ExitULV ((uint16_t)0x65)
#define PPSMC_PowerShiftActive ((uint16_t)0x6A)
#define PPSMC_PowerShiftInactive ((uint16_t)0x6B)
#define PPSMC_OCPActive ((uint16_t)0x6C)
#define PPSMC_OCPInactive ((uint16_t)0x6D)
#define PPSMC_CACLongTermAvgEnable ((uint16_t)0x6E)
#define PPSMC_CACLongTermAvgDisable ((uint16_t)0x6F)
#define PPSMC_MSG_InferredStateSweep_Start ((uint16_t)0x70)
#define PPSMC_MSG_InferredStateSweep_Stop ((uint16_t)0x71)
#define PPSMC_MSG_SwitchToLowestInfState ((uint16_t)0x72)
#define PPSMC_MSG_SwitchToNonInfState ((uint16_t)0x73)
#define PPSMC_MSG_AllStateSweep_Start ((uint16_t)0x74)
#define PPSMC_MSG_AllStateSweep_Stop ((uint16_t)0x75)
#define PPSMC_MSG_SwitchNextLowerInfState ((uint16_t)0x76)
#define PPSMC_MSG_SwitchNextHigherInfState ((uint16_t)0x77)
#define PPSMC_MSG_MclkRetrainingTest ((uint16_t)0x78)
#define PPSMC_MSG_ForceTDPClamping ((uint16_t)0x79)
#define PPSMC_MSG_CollectCAC_PowerCorreln ((uint16_t)0x7A)
#define PPSMC_MSG_CollectCAC_WeightCalib ((uint16_t)0x7B)
#define PPSMC_MSG_CollectCAC_SQonly ((uint16_t)0x7C)
#define PPSMC_MSG_CollectCAC_TemperaturePwr ((uint16_t)0x7D)
#define PPSMC_MSG_ExtremitiesTest_Start ((uint16_t)0x7E)
#define PPSMC_MSG_ExtremitiesTest_Stop ((uint16_t)0x7F)
#define PPSMC_FlushDataCache ((uint16_t)0x80)
#define PPSMC_FlushInstrCache ((uint16_t)0x81)
#define PPSMC_MSG_SetEnabledLevels ((uint16_t)0x82)
#define PPSMC_MSG_SetForcedLevels ((uint16_t)0x83)
#define PPSMC_MSG_ResetToDefaults ((uint16_t)0x84)
#define PPSMC_MSG_SetForcedLevelsAndJump ((uint16_t)0x85)
#define PPSMC_MSG_SetCACHistoryMode ((uint16_t)0x86)
#define PPSMC_MSG_EnableDTE ((uint16_t)0x87)
#define PPSMC_MSG_DisableDTE ((uint16_t)0x88)
#define PPSMC_MSG_SmcSpaceSetAddress ((uint16_t)0x89)
#define PPSMC_MSG_SmcSpaceWriteDWordInc ((uint16_t)0x8A)
#define PPSMC_MSG_SmcSpaceWriteWordInc ((uint16_t)0x8B)
#define PPSMC_MSG_SmcSpaceWriteByteInc ((uint16_t)0x8C)
#define PPSMC_MSG_BREAK ((uint16_t)0xF8)
#define PPSMC_MSG_Test ((uint16_t)0x100)
#define PPSMC_MSG_DRV_DRAM_ADDR_HI ((uint16_t)0x250)
#define PPSMC_MSG_DRV_DRAM_ADDR_LO ((uint16_t)0x251)
#define PPSMC_MSG_SMU_DRAM_ADDR_HI ((uint16_t)0x252)
#define PPSMC_MSG_SMU_DRAM_ADDR_LO ((uint16_t)0x253)
#define PPSMC_MSG_LoadUcodes ((uint16_t)0x254)
typedef uint16_t PPSMC_Msg;
#define PPSMC_EVENT_STATUS_THERMAL 0x00000001
#define PPSMC_EVENT_STATUS_REGULATORHOT 0x00000002
#define PPSMC_EVENT_STATUS_DC 0x00000004
#define PPSMC_EVENT_STATUS_GPIO17 0x00000008
#pragma pack(pop)
#endif
drivers/gpu/drm/amd/amdgpu/fiji_smc.c
+853
View File
@@ -0,0 +1,853 @@
/*
* Copyright 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*/
#include <linux/firmware.h>
#include "drmP.h"
#include "amdgpu.h"
#include "fiji_ppsmc.h"
#include "fiji_smumgr.h"
#include "smu_ucode_xfer_vi.h"
#include "amdgpu_ucode.h"
#include "smu/smu_7_1_3_d.h"
#include "smu/smu_7_1_3_sh_mask.h"
#define FIJI_SMC_SIZE 0x20000
static int fiji_set_smc_sram_address(struct amdgpu_device *adev, uint32_t smc_address, uint32_t limit)
{
uint32_t val;
if (smc_address & 3)
return -EINVAL;
if ((smc_address + 3) > limit)
return -EINVAL;
WREG32(mmSMC_IND_INDEX_0, smc_address);
val = RREG32(mmSMC_IND_ACCESS_CNTL);
val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 0);
WREG32(mmSMC_IND_ACCESS_CNTL, val);
return 0;
}
static int fiji_copy_bytes_to_smc(struct amdgpu_device *adev, uint32_t smc_start_address, const uint8_t *src, uint32_t byte_count, uint32_t limit)
{
uint32_t addr;
uint32_t data, orig_data;
int result = 0;
uint32_t extra_shift;
unsigned long flags;
if (smc_start_address & 3)
return -EINVAL;
if ((smc_start_address + byte_count) > limit)
return -EINVAL;
addr = smc_start_address;
spin_lock_irqsave(&adev->smc_idx_lock, flags);
while (byte_count >= 4) {
/* Bytes are written into the SMC addres space with the MSB first */
data = (src[0] << 24) + (src[1] << 16) + (src[2] << 8) + src[3];
result = fiji_set_smc_sram_address(adev, addr, limit);
if (result)
goto out;
WREG32(mmSMC_IND_DATA_0, data);
src += 4;
byte_count -= 4;
addr += 4;
}
if (0 != byte_count) {
/* Now write odd bytes left, do a read modify write cycle */
data = 0;
result = fiji_set_smc_sram_address(adev, addr, limit);
if (result)
goto out;
orig_data = RREG32(mmSMC_IND_DATA_0);
extra_shift = 8 * (4 - byte_count);
while (byte_count > 0) {
data = (data << 8) + *src++;
byte_count--;
}
data <<= extra_shift;
data |= (orig_data & ~((~0UL) << extra_shift));
result = fiji_set_smc_sram_address(adev, addr, limit);
if (result)
goto out;
WREG32(mmSMC_IND_DATA_0, data);
}
out:
spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
return result;
}
static int fiji_program_jump_on_start(struct amdgpu_device *adev)
{
static unsigned char data[] = {0xE0, 0x00, 0x80, 0x40};
fiji_copy_bytes_to_smc(adev, 0x0, data, 4, sizeof(data)+1);
return 0;
}
static bool fiji_is_smc_ram_running(struct amdgpu_device *adev)
{
uint32_t val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
val = REG_GET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable);
return ((0 == val) && (0x20100 <= RREG32_SMC(ixSMC_PC_C)));
}
static int wait_smu_response(struct amdgpu_device *adev)
{
int i;
uint32_t val;
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32(mmSMC_RESP_0);
if (REG_GET_FIELD(val, SMC_RESP_0, SMC_RESP))
break;
udelay(1);
}
if (i == adev->usec_timeout)
return -EINVAL;
return 0;
}
static int fiji_send_msg_to_smc_offset(struct amdgpu_device *adev)
{
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send previous message\n");
return -EINVAL;
}
WREG32(mmSMC_MSG_ARG_0, 0x20000);
WREG32(mmSMC_MESSAGE_0, PPSMC_MSG_Test);
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send message\n");
return -EINVAL;
}
return 0;
}
static int fiji_send_msg_to_smc(struct amdgpu_device *adev, PPSMC_Msg msg)
{
if (!fiji_is_smc_ram_running(adev))
{
return -EINVAL;;
}
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send previous message\n");
return -EINVAL;
}
WREG32(mmSMC_MESSAGE_0, msg);
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send message\n");
return -EINVAL;
}
return 0;
}
static int fiji_send_msg_to_smc_without_waiting(struct amdgpu_device *adev,
PPSMC_Msg msg)
{
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send previous message\n");
return -EINVAL;
}
WREG32(mmSMC_MESSAGE_0, msg);
return 0;
}
static int fiji_send_msg_to_smc_with_parameter(struct amdgpu_device *adev,
PPSMC_Msg msg,
uint32_t parameter)
{
if (!fiji_is_smc_ram_running(adev))
return -EINVAL;
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send previous message\n");
return -EINVAL;
}
WREG32(mmSMC_MSG_ARG_0, parameter);
return fiji_send_msg_to_smc(adev, msg);
}
static int fiji_send_msg_to_smc_with_parameter_without_waiting(
struct amdgpu_device *adev,
PPSMC_Msg msg, uint32_t parameter)
{
if (wait_smu_response(adev)) {
DRM_ERROR("Failed to send previous message\n");
return -EINVAL;
}
WREG32(mmSMC_MSG_ARG_0, parameter);
return fiji_send_msg_to_smc_without_waiting(adev, msg);
}
#if 0 /* not used yet */
static int fiji_wait_for_smc_inactive(struct amdgpu_device *adev)
{
int i;
uint32_t val;
if (!fiji_is_smc_ram_running(adev))
return -EINVAL;
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
if (REG_GET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, cken) == 0)
break;
udelay(1);
}
if (i == adev->usec_timeout)
return -EINVAL;
return 0;
}
#endif
static int fiji_smu_upload_firmware_image(struct amdgpu_device *adev)
{
const struct smc_firmware_header_v1_0 *hdr;
uint32_t ucode_size;
uint32_t ucode_start_address;
const uint8_t *src;
uint32_t val;
uint32_t byte_count;
uint32_t *data;
unsigned long flags;
if (!adev->pm.fw)
return -EINVAL;
hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data;
amdgpu_ucode_print_smc_hdr(&hdr->header);
adev->pm.fw_version = le32_to_cpu(hdr->header.ucode_version);
ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes);
ucode_start_address = le32_to_cpu(hdr->ucode_start_addr);
src = (const uint8_t *)
(adev->pm.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
if (ucode_size & 3) {
DRM_ERROR("SMC ucode is not 4 bytes aligned\n");
return -EINVAL;
}
if (ucode_size > FIJI_SMC_SIZE) {
DRM_ERROR("SMC address is beyond the SMC RAM area\n");
return -EINVAL;
}
spin_lock_irqsave(&adev->smc_idx_lock, flags);
WREG32(mmSMC_IND_INDEX_0, ucode_start_address);
val = RREG32(mmSMC_IND_ACCESS_CNTL);
val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 1);
WREG32(mmSMC_IND_ACCESS_CNTL, val);
byte_count = ucode_size;
data = (uint32_t *)src;
for (; byte_count >= 4; data++, byte_count -= 4)
WREG32(mmSMC_IND_DATA_0, data[0]);
val = RREG32(mmSMC_IND_ACCESS_CNTL);
val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 0);
WREG32(mmSMC_IND_ACCESS_CNTL, val);
spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
return 0;
}
#if 0 /* not used yet */
static int fiji_read_smc_sram_dword(struct amdgpu_device *adev,
uint32_t smc_address,
uint32_t *value,
uint32_t limit)
{
int result;
unsigned long flags;
spin_lock_irqsave(&adev->smc_idx_lock, flags);
result = fiji_set_smc_sram_address(adev, smc_address, limit);
if (result == 0)
*value = RREG32(mmSMC_IND_DATA_0);
spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
return result;
}
static int fiji_write_smc_sram_dword(struct amdgpu_device *adev,
uint32_t smc_address,
uint32_t value,
uint32_t limit)
{
int result;
unsigned long flags;
spin_lock_irqsave(&adev->smc_idx_lock, flags);
result = fiji_set_smc_sram_address(adev, smc_address, limit);
if (result == 0)
WREG32(mmSMC_IND_DATA_0, value);
spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
return result;
}
static int fiji_smu_stop_smc(struct amdgpu_device *adev)
{
uint32_t val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 1);
WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
return 0;
}
#endif
static enum AMDGPU_UCODE_ID fiji_convert_fw_type(uint32_t fw_type)
{
switch (fw_type) {
case UCODE_ID_SDMA0:
return AMDGPU_UCODE_ID_SDMA0;
case UCODE_ID_SDMA1:
return AMDGPU_UCODE_ID_SDMA1;
case UCODE_ID_CP_CE:
return AMDGPU_UCODE_ID_CP_CE;
case UCODE_ID_CP_PFP:
return AMDGPU_UCODE_ID_CP_PFP;
case UCODE_ID_CP_ME:
return AMDGPU_UCODE_ID_CP_ME;
case UCODE_ID_CP_MEC:
case UCODE_ID_CP_MEC_JT1:
case UCODE_ID_CP_MEC_JT2:
return AMDGPU_UCODE_ID_CP_MEC1;
case UCODE_ID_RLC_G:
return AMDGPU_UCODE_ID_RLC_G;
default:
DRM_ERROR("ucode type is out of range!\n");
return AMDGPU_UCODE_ID_MAXIMUM;
}
}
static int fiji_smu_populate_single_firmware_entry(struct amdgpu_device *adev,
uint32_t fw_type,
struct SMU_Entry *entry)
{
enum AMDGPU_UCODE_ID id = fiji_convert_fw_type(fw_type);
struct amdgpu_firmware_info *ucode = &adev->firmware.ucode[id];
const struct gfx_firmware_header_v1_0 *header = NULL;
uint64_t gpu_addr;
uint32_t data_size;
if (ucode->fw == NULL)
return -EINVAL;
gpu_addr = ucode->mc_addr;
header = (const struct gfx_firmware_header_v1_0 *)ucode->fw->data;
data_size = le32_to_cpu(header->header.ucode_size_bytes);
if ((fw_type == UCODE_ID_CP_MEC_JT1) ||
(fw_type == UCODE_ID_CP_MEC_JT2)) {
gpu_addr += le32_to_cpu(header->jt_offset) << 2;
data_size = le32_to_cpu(header->jt_size) << 2;
}
entry->version = (uint16_t)le32_to_cpu(header->header.ucode_version);
entry->id = (uint16_t)fw_type;
entry->image_addr_high = upper_32_bits(gpu_addr);
entry->image_addr_low = lower_32_bits(gpu_addr);
entry->meta_data_addr_high = 0;
entry->meta_data_addr_low = 0;
entry->data_size_byte = data_size;
entry->num_register_entries = 0;
if (fw_type == UCODE_ID_RLC_G)
entry->flags = 1;
else
entry->flags = 0;
return 0;
}
static int fiji_smu_request_load_fw(struct amdgpu_device *adev)
{
struct fiji_smu_private_data *private = (struct fiji_smu_private_data *)adev->smu.priv;
struct SMU_DRAMData_TOC *toc;
uint32_t fw_to_load;
WREG32_SMC(ixSOFT_REGISTERS_TABLE_28, 0);
fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_SMU_DRAM_ADDR_HI, private->smu_buffer_addr_high);
fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_SMU_DRAM_ADDR_LO, private->smu_buffer_addr_low);
toc = (struct SMU_DRAMData_TOC *)private->header;
toc->num_entries = 0;
toc->structure_version = 1;
if (!adev->firmware.smu_load)
return 0;
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_RLC_G,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for RLC\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_CE,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for CE\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_PFP,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for PFP\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_ME,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for ME\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for MEC\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC_JT1,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for MEC_JT1\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC_JT2,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for MEC_JT2\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_SDMA0,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for SDMA0\n");
return -EINVAL;
}
if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_SDMA1,
&toc->entry[toc->num_entries++])) {
DRM_ERROR("Failed to get firmware entry for SDMA1\n");
return -EINVAL;
}
fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_DRV_DRAM_ADDR_HI, private->header_addr_high);
fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_DRV_DRAM_ADDR_LO, private->header_addr_low);
fw_to_load = UCODE_ID_RLC_G_MASK |
UCODE_ID_SDMA0_MASK |
UCODE_ID_SDMA1_MASK |
UCODE_ID_CP_CE_MASK |
UCODE_ID_CP_ME_MASK |
UCODE_ID_CP_PFP_MASK |
UCODE_ID_CP_MEC_MASK;
if (fiji_send_msg_to_smc_with_parameter_without_waiting(adev, PPSMC_MSG_LoadUcodes, fw_to_load)) {
DRM_ERROR("Fail to request SMU load ucode\n");
return -EINVAL;
}
return 0;
}
static uint32_t fiji_smu_get_mask_for_fw_type(uint32_t fw_type)
{
switch (fw_type) {
case AMDGPU_UCODE_ID_SDMA0:
return UCODE_ID_SDMA0_MASK;
case AMDGPU_UCODE_ID_SDMA1:
return UCODE_ID_SDMA1_MASK;
case AMDGPU_UCODE_ID_CP_CE:
return UCODE_ID_CP_CE_MASK;
case AMDGPU_UCODE_ID_CP_PFP:
return UCODE_ID_CP_PFP_MASK;
case AMDGPU_UCODE_ID_CP_ME:
return UCODE_ID_CP_ME_MASK;
case AMDGPU_UCODE_ID_CP_MEC1:
return UCODE_ID_CP_MEC_MASK;
case AMDGPU_UCODE_ID_CP_MEC2:
return UCODE_ID_CP_MEC_MASK;
case AMDGPU_UCODE_ID_RLC_G:
return UCODE_ID_RLC_G_MASK;
default:
DRM_ERROR("ucode type is out of range!\n");
return 0;
}
}
static int fiji_smu_check_fw_load_finish(struct amdgpu_device *adev,
uint32_t fw_type)
{
uint32_t fw_mask = fiji_smu_get_mask_for_fw_type(fw_type);
int i;
for (i = 0; i < adev->usec_timeout; i++) {
if (fw_mask == (RREG32_SMC(ixSOFT_REGISTERS_TABLE_28) & fw_mask))
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("check firmware loading failed\n");
return -EINVAL;
}
return 0;
}
static int fiji_smu_start_in_protection_mode(struct amdgpu_device *adev)
{
int result;
uint32_t val;
int i;
/* Assert reset */
val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
result = fiji_smu_upload_firmware_image(adev);
if (result)
return result;
/* Clear status */
WREG32_SMC(ixSMU_STATUS, 0);
/* Enable clock */
val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 0);
WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
/* De-assert reset */
val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 0);
WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
/* Set SMU Auto Start */
val = RREG32_SMC(ixSMU_INPUT_DATA);
val = REG_SET_FIELD(val, SMU_INPUT_DATA, AUTO_START, 1);
WREG32_SMC(ixSMU_INPUT_DATA, val);
/* Clear firmware interrupt enable flag */
WREG32_SMC(ixFIRMWARE_FLAGS, 0);
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixRCU_UC_EVENTS);
if (REG_GET_FIELD(val, RCU_UC_EVENTS, INTERRUPTS_ENABLED))
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("Interrupt is not enabled by firmware\n");
return -EINVAL;
}
/* Call Test SMU message with 0x20000 offset
* to trigger SMU start
*/
fiji_send_msg_to_smc_offset(adev);
DRM_INFO("[FM]try triger smu start\n");
/* Wait for done bit to be set */
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixSMU_STATUS);
if (REG_GET_FIELD(val, SMU_STATUS, SMU_DONE))
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("Timeout for SMU start\n");
return -EINVAL;
}
/* Check pass/failed indicator */
val = RREG32_SMC(ixSMU_STATUS);
if (!REG_GET_FIELD(val, SMU_STATUS, SMU_PASS)) {
DRM_ERROR("SMU Firmware start failed\n");
return -EINVAL;
}
DRM_INFO("[FM]smu started\n");
/* Wait for firmware to initialize */
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixFIRMWARE_FLAGS);
if(REG_GET_FIELD(val, FIRMWARE_FLAGS, INTERRUPTS_ENABLED))
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("SMU firmware initialization failed\n");
return -EINVAL;
}
DRM_INFO("[FM]smu initialized\n");
return 0;
}
static int fiji_smu_start_in_non_protection_mode(struct amdgpu_device *adev)
{
int i, result;
uint32_t val;
/* wait for smc boot up */
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixRCU_UC_EVENTS);
val = REG_GET_FIELD(val, RCU_UC_EVENTS, boot_seq_done);
if (val)
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("SMC boot sequence is not completed\n");
return -EINVAL;
}
/* Clear firmware interrupt enable flag */
WREG32_SMC(ixFIRMWARE_FLAGS, 0);
/* Assert reset */
val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
result = fiji_smu_upload_firmware_image(adev);
if (result)
return result;
/* Set smc instruct start point at 0x0 */
fiji_program_jump_on_start(adev);
/* Enable clock */
val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 0);
WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
/* De-assert reset */
val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 0);
WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
/* Wait for firmware to initialize */
for (i = 0; i < adev->usec_timeout; i++) {
val = RREG32_SMC(ixFIRMWARE_FLAGS);
if (REG_GET_FIELD(val, FIRMWARE_FLAGS, INTERRUPTS_ENABLED))
break;
udelay(1);
}
if (i == adev->usec_timeout) {
DRM_ERROR("Timeout for SMC firmware initialization\n");
return -EINVAL;
}
return 0;
}
int fiji_smu_start(struct amdgpu_device *adev)
{
int result;
uint32_t val;
if (!fiji_is_smc_ram_running(adev)) {
val = RREG32_SMC(ixSMU_FIRMWARE);
if (!REG_GET_FIELD(val, SMU_FIRMWARE, SMU_MODE)) {
DRM_INFO("[FM]start smu in nonprotection mode\n");
result = fiji_smu_start_in_non_protection_mode(adev);
if (result)
return result;
} else {
DRM_INFO("[FM]start smu in protection mode\n");
result = fiji_smu_start_in_protection_mode(adev);
if (result)
return result;
}
}
return fiji_smu_request_load_fw(adev);
}
static const struct amdgpu_smumgr_funcs fiji_smumgr_funcs = {
.check_fw_load_finish = fiji_smu_check_fw_load_finish,
.request_smu_load_fw = NULL,
.request_smu_specific_fw = NULL,
};
int fiji_smu_init(struct amdgpu_device *adev)
{
struct fiji_smu_private_data *private;
uint32_t image_size = ((sizeof(struct SMU_DRAMData_TOC) / 4096) + 1) * 4096;
uint32_t smu_internal_buffer_size = 200*4096;
struct amdgpu_bo **toc_buf = &adev->smu.toc_buf;
struct amdgpu_bo **smu_buf = &adev->smu.smu_buf;
uint64_t mc_addr;
void *toc_buf_ptr;
void *smu_buf_ptr;
int ret;
private = kzalloc(sizeof(struct fiji_smu_private_data), GFP_KERNEL);
if (NULL == private)
return -ENOMEM;
/* allocate firmware buffers */
if (adev->firmware.smu_load)
amdgpu_ucode_init_bo(adev);
adev->smu.priv = private;
adev->smu.fw_flags = 0;
/* Allocate FW image data structure and header buffer */
ret = amdgpu_bo_create(adev, image_size, PAGE_SIZE,
true, AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, toc_buf);
if (ret) {
DRM_ERROR("Failed to allocate memory for TOC buffer\n");
return -ENOMEM;
}
/* Allocate buffer for SMU internal buffer */
ret = amdgpu_bo_create(adev, smu_internal_buffer_size, PAGE_SIZE,
true, AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, smu_buf);
if (ret) {
DRM_ERROR("Failed to allocate memory for SMU internal buffer\n");
return -ENOMEM;
}
/* Retrieve GPU address for header buffer and internal buffer */
ret = amdgpu_bo_reserve(adev->smu.toc_buf, false);
if (ret) {
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to reserve the TOC buffer\n");
return -EINVAL;
}
ret = amdgpu_bo_pin(adev->smu.toc_buf, AMDGPU_GEM_DOMAIN_VRAM, &mc_addr);
if (ret) {
amdgpu_bo_unreserve(adev->smu.toc_buf);
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to pin the TOC buffer\n");
return -EINVAL;
}
ret = amdgpu_bo_kmap(*toc_buf, &toc_buf_ptr);
if (ret) {
amdgpu_bo_unreserve(adev->smu.toc_buf);
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to map the TOC buffer\n");
return -EINVAL;
}
amdgpu_bo_unreserve(adev->smu.toc_buf);
private->header_addr_low = lower_32_bits(mc_addr);
private->header_addr_high = upper_32_bits(mc_addr);
private->header = toc_buf_ptr;
ret = amdgpu_bo_reserve(adev->smu.smu_buf, false);
if (ret) {
amdgpu_bo_unref(&adev->smu.smu_buf);
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to reserve the SMU internal buffer\n");
return -EINVAL;
}
ret = amdgpu_bo_pin(adev->smu.smu_buf, AMDGPU_GEM_DOMAIN_VRAM, &mc_addr);
if (ret) {
amdgpu_bo_unreserve(adev->smu.smu_buf);
amdgpu_bo_unref(&adev->smu.smu_buf);
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to pin the SMU internal buffer\n");
return -EINVAL;
}
ret = amdgpu_bo_kmap(*smu_buf, &smu_buf_ptr);
if (ret) {
amdgpu_bo_unreserve(adev->smu.smu_buf);
amdgpu_bo_unref(&adev->smu.smu_buf);
amdgpu_bo_unref(&adev->smu.toc_buf);
DRM_ERROR("Failed to map the SMU internal buffer\n");
return -EINVAL;
}
amdgpu_bo_unreserve(adev->smu.smu_buf);
private->smu_buffer_addr_low = lower_32_bits(mc_addr);
private->smu_buffer_addr_high = upper_32_bits(mc_addr);
adev->smu.smumgr_funcs = &fiji_smumgr_funcs;
return 0;
}
int fiji_smu_fini(struct amdgpu_device *adev)
{
amdgpu_bo_unref(&adev->smu.toc_buf);
amdgpu_bo_unref(&adev->smu.smu_buf);
kfree(adev->smu.priv);
adev->smu.priv = NULL;
if (adev->firmware.fw_buf)
amdgpu_ucode_fini_bo(adev);
return 0;
}
drivers/gpu/drm/amd/amdgpu/amdgpu_family.h → drivers/gpu/drm/amd/amdgpu/fiji_smumgr.h
+16 -36
View File
@@ -1,7 +1,5 @@
/*
* Copyright 2008 Advanced Micro Devices, Inc.
* Copyright 2008 Red Hat Inc.
* Copyright 2009 Jerome Glisse.
* Copyright 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -21,42 +19,24 @@
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
* Authors: Dave Airlie
* Alex Deucher
* Jerome Glisse
*/
/* this file defines the CHIP_ and family flags used in the pciids,
* its is common between kms and non-kms because duplicating it and
* changing one place is fail.
*/
#ifndef AMDGPU_FAMILY_H
#define AMDGPU_FAMILY_H
/*
* Supported ASIC types
*/
enum amdgpu_asic_type {
CHIP_BONAIRE = 0,
CHIP_KAVERI,
CHIP_KABINI,
CHIP_HAWAII,
CHIP_MULLINS,
CHIP_TOPAZ,
CHIP_TONGA,
CHIP_CARRIZO,
CHIP_LAST,
};
#ifndef FIJI_SMUMGR_H
#define FIJI_SMUMGR_H
/*
* Chip flags
*/
enum amdgpu_chip_flags {
AMDGPU_ASIC_MASK = 0x0000ffffUL,
AMDGPU_FLAGS_MASK = 0xffff0000UL,
AMDGPU_IS_MOBILITY = 0x00010000UL,
AMDGPU_IS_APU = 0x00020000UL,
AMDGPU_IS_PX = 0x00040000UL,
AMDGPU_EXP_HW_SUPPORT = 0x00080000UL,
#include "fiji_ppsmc.h"
int fiji_smu_init(struct amdgpu_device *adev);
int fiji_smu_fini(struct amdgpu_device *adev);
int fiji_smu_start(struct amdgpu_device *adev);
struct fiji_smu_private_data
{
uint8_t *header;
uint32_t smu_buffer_addr_high;
uint32_t smu_buffer_addr_low;
uint32_t header_addr_high;
uint32_t header_addr_low;
};
#endif
drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+20 -17
View File
@@ -2173,7 +2173,7 @@ static void gfx_v7_0_gpu_init(struct amdgpu_device *adev)
adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
adev->gfx.config.mem_max_burst_length_bytes = 256;
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
/* Get memory bank mapping mode. */
tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
@@ -2648,6 +2648,7 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct fence *f = NULL;
uint32_t scratch;
uint32_t tmp = 0;
unsigned i;
@@ -2662,26 +2663,23 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
amdgpu_gfx_scratch_free(adev, scratch);
return r;
goto err1;
}
ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
ib.ptr[2] = 0xDEADBEEF;
ib.length_dw = 3;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
if (r) {
amdgpu_gfx_scratch_free(adev, scratch);
amdgpu_ib_free(adev, &ib);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err2;
r = fence_wait(f, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
amdgpu_gfx_scratch_free(adev, scratch);
amdgpu_ib_free(adev, &ib);
return r;
goto err2;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = RREG32(scratch);
@@ -2691,14 +2689,19 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
ring->idx, i);
goto err2;
} else {
DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
scratch, tmp);
r = -EINVAL;
}
amdgpu_gfx_scratch_free(adev, scratch);
err2:
fence_put(f);
amdgpu_ib_free(adev, &ib);
err1:
amdgpu_gfx_scratch_free(adev, scratch);
return r;
}
@@ -3758,7 +3761,7 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
int r;
/* allocate rlc buffers */
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
if (adev->asic_type == CHIP_KAVERI) {
adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
adev->gfx.rlc.reg_list_size =
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+126 -17
View File
@@ -87,6 +87,13 @@ MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
MODULE_FIRMWARE("amdgpu/topaz_mec2.bin");
MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
MODULE_FIRMWARE("amdgpu/fiji_me.bin");
MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
{
{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
@@ -217,6 +224,71 @@ static const u32 tonga_mgcg_cgcg_init[] =
mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
};
static const u32 fiji_golden_common_all[] =
{
mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
mmGB_ADDR_CONFIG, 0xffffffff, 0x12011003,
mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00007FBF,
mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00007FAF
};
static const u32 golden_settings_fiji_a10[] =
{
mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
mmDB_DEBUG2, 0xf00fffff, 0x00000400,
mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x00000100,
mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
mmTCC_CTRL, 0x00100000, 0xf30fff7f,
mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
mmTCP_CHAN_STEER_HI, 0xffffffff, 0x7d6cf5e4,
mmTCP_CHAN_STEER_LO, 0xffffffff, 0x3928b1a0,
};
static const u32 fiji_mgcg_cgcg_init[] =
{
mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffc0,
mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
};
static const u32 golden_settings_iceland_a11[] =
{
mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
@@ -439,6 +511,18 @@ static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
iceland_golden_common_all,
(const u32)ARRAY_SIZE(iceland_golden_common_all));
break;
case CHIP_FIJI:
amdgpu_program_register_sequence(adev,
fiji_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
amdgpu_program_register_sequence(adev,
golden_settings_fiji_a10,
(const u32)ARRAY_SIZE(golden_settings_fiji_a10));
amdgpu_program_register_sequence(adev,
fiji_golden_common_all,
(const u32)ARRAY_SIZE(fiji_golden_common_all));
break;
case CHIP_TONGA:
amdgpu_program_register_sequence(adev,
tonga_mgcg_cgcg_init,
@@ -526,6 +610,7 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct fence *f = NULL;
uint32_t scratch;
uint32_t tmp = 0;
unsigned i;
@@ -540,26 +625,23 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
amdgpu_gfx_scratch_free(adev, scratch);
return r;
goto err1;
}
ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
ib.ptr[2] = 0xDEADBEEF;
ib.length_dw = 3;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
if (r) {
amdgpu_gfx_scratch_free(adev, scratch);
amdgpu_ib_free(adev, &ib);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err2;
r = fence_wait(f, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
amdgpu_gfx_scratch_free(adev, scratch);
amdgpu_ib_free(adev, &ib);
return r;
goto err2;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = RREG32(scratch);
@@ -569,14 +651,18 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
ring->idx, i);
goto err2;
} else {
DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
scratch, tmp);
r = -EINVAL;
}
amdgpu_gfx_scratch_free(adev, scratch);
err2:
fence_put(f);
amdgpu_ib_free(adev, &ib);
err1:
amdgpu_gfx_scratch_free(adev, scratch);
return r;
}
@@ -601,6 +687,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
case CHIP_CARRIZO:
chip_name = "carrizo";
break;
case CHIP_FIJI:
chip_name = "fiji";
break;
default:
BUG();
}
@@ -1236,6 +1325,7 @@ static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden;
WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden);
}
case CHIP_FIJI:
case CHIP_TONGA:
for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
switch (reg_offset) {
@@ -1984,6 +2074,23 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
break;
case CHIP_FIJI:
adev->gfx.config.max_shader_engines = 4;
adev->gfx.config.max_tile_pipes = 16;
adev->gfx.config.max_cu_per_sh = 16;
adev->gfx.config.max_sh_per_se = 1;
adev->gfx.config.max_backends_per_se = 4;
adev->gfx.config.max_texture_channel_caches = 8;
adev->gfx.config.max_gprs = 256;
adev->gfx.config.max_gs_threads = 32;
adev->gfx.config.max_hw_contexts = 8;
adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
break;
case CHIP_TONGA:
adev->gfx.config.max_shader_engines = 4;
adev->gfx.config.max_tile_pipes = 8;
@@ -2078,7 +2185,7 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
adev->gfx.config.mem_max_burst_length_bytes = 256;
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
/* Get memory bank mapping mode. */
tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
@@ -2490,6 +2597,7 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
switch (adev->asic_type) {
case CHIP_TONGA:
case CHIP_FIJI:
amdgpu_ring_write(ring, 0x16000012);
amdgpu_ring_write(ring, 0x0000002A);
break;
@@ -3875,7 +3983,8 @@ static bool gfx_v8_0_ring_emit_semaphore(struct amdgpu_ring *ring,
unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
if (ring->adev->asic_type == CHIP_TOPAZ ||
ring->adev->asic_type == CHIP_TONGA)
ring->adev->asic_type == CHIP_TONGA ||
ring->adev->asic_type == CHIP_FIJI)
/* we got a hw semaphore bug in VI TONGA, return false to switch back to sw fence wait */
return false;
else {
drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+5 -5
View File
@@ -636,7 +636,7 @@ static int gmc_v7_0_vm_init(struct amdgpu_device *adev)
adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
/* base offset of vram pages */
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
tmp <<= 22;
adev->vm_manager.vram_base_offset = tmp;
@@ -841,7 +841,7 @@ static int gmc_v7_0_early_init(void *handle)
gmc_v7_0_set_gart_funcs(adev);
gmc_v7_0_set_irq_funcs(adev);
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
} else {
u32 tmp = RREG32(mmMC_SEQ_MISC0);
@@ -957,7 +957,7 @@ static int gmc_v7_0_hw_init(void *handle)
gmc_v7_0_mc_program(adev);
if (!(adev->flags & AMDGPU_IS_APU)) {
if (!(adev->flags & AMD_IS_APU)) {
r = gmc_v7_0_mc_load_microcode(adev);
if (r) {
DRM_ERROR("Failed to load MC firmware!\n");
@@ -1172,7 +1172,7 @@ static int gmc_v7_0_soft_reset(void *handle)
if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
if (!(adev->flags & AMDGPU_IS_APU))
if (!(adev->flags & AMD_IS_APU))
srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
}
@@ -1282,7 +1282,7 @@ static int gmc_v7_0_set_clockgating_state(void *handle,
if (state == AMD_CG_STATE_GATE)
gate = true;
if (!(adev->flags & AMDGPU_IS_APU)) {
if (!(adev->flags & AMD_IS_APU)) {
gmc_v7_0_enable_mc_mgcg(adev, gate);
gmc_v7_0_enable_mc_ls(adev, gate);
}
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+29 -4
View File
@@ -44,6 +44,7 @@ static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
MODULE_FIRMWARE("amdgpu/topaz_mc.bin");
MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
MODULE_FIRMWARE("amdgpu/fiji_mc.bin");
static const u32 golden_settings_tonga_a11[] =
{
@@ -61,6 +62,19 @@ static const u32 tonga_mgcg_cgcg_init[] =
mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
};
static const u32 golden_settings_fiji_a10[] =
{
mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
};
static const u32 fiji_mgcg_cgcg_init[] =
{
mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
};
static const u32 golden_settings_iceland_a11[] =
{
mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
@@ -90,6 +104,14 @@ static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
golden_settings_iceland_a11,
(const u32)ARRAY_SIZE(golden_settings_iceland_a11));
break;
case CHIP_FIJI:
amdgpu_program_register_sequence(adev,
fiji_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
amdgpu_program_register_sequence(adev,
golden_settings_fiji_a10,
(const u32)ARRAY_SIZE(golden_settings_fiji_a10));
break;
case CHIP_TONGA:
amdgpu_program_register_sequence(adev,
tonga_mgcg_cgcg_init,
@@ -202,6 +224,9 @@ static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
case CHIP_TONGA:
chip_name = "tonga";
break;
case CHIP_FIJI:
chip_name = "fiji";
break;
case CHIP_CARRIZO:
return 0;
default: BUG();
@@ -737,7 +762,7 @@ static int gmc_v8_0_vm_init(struct amdgpu_device *adev)
adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
/* base offset of vram pages */
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
tmp <<= 22;
adev->vm_manager.vram_base_offset = tmp;
@@ -816,7 +841,7 @@ static int gmc_v8_0_early_init(void *handle)
gmc_v8_0_set_gart_funcs(adev);
gmc_v8_0_set_irq_funcs(adev);
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
} else {
u32 tmp = RREG32(mmMC_SEQ_MISC0);
@@ -934,7 +959,7 @@ static int gmc_v8_0_hw_init(void *handle)
gmc_v8_0_mc_program(adev);
if (!(adev->flags & AMDGPU_IS_APU)) {
if (!(adev->flags & AMD_IS_APU)) {
r = gmc_v8_0_mc_load_microcode(adev);
if (r) {
DRM_ERROR("Failed to load MC firmware!\n");
@@ -1147,7 +1172,7 @@ static int gmc_v8_0_soft_reset(void *handle)
if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
if (!(adev->flags & AMDGPU_IS_APU))
if (!(adev->flags & AMD_IS_APU))
srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
}
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+17 -15
View File
@@ -673,6 +673,7 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct fence *f = NULL;
unsigned i;
unsigned index;
int r;
@@ -688,12 +689,10 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
return r;
goto err0;
}
ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
@@ -707,19 +706,16 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
ib.ptr[7] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
ib.length_dw = 8;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err1;
r = fence_wait(f, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
return r;
goto err1;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -729,12 +725,17 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
ring->idx, i);
goto err1;
} else {
DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
err1:
fence_put(f);
amdgpu_ib_free(adev, &ib);
err0:
amdgpu_wb_free(adev, index);
return r;
}
@@ -1415,5 +1416,6 @@ static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev)
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs;
adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+47 -15
View File
@@ -53,6 +53,8 @@ MODULE_FIRMWARE("amdgpu/tonga_sdma.bin");
MODULE_FIRMWARE("amdgpu/tonga_sdma1.bin");
MODULE_FIRMWARE("amdgpu/carrizo_sdma.bin");
MODULE_FIRMWARE("amdgpu/carrizo_sdma1.bin");
MODULE_FIRMWARE("amdgpu/fiji_sdma.bin");
MODULE_FIRMWARE("amdgpu/fiji_sdma1.bin");
static const u32 sdma_offsets[SDMA_MAX_INSTANCE] =
{
@@ -80,6 +82,24 @@ static const u32 tonga_mgcg_cgcg_init[] =
mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100
};
static const u32 golden_settings_fiji_a10[] =
{
mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007,
mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100,
mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100,
mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100,
mmSDMA1_CHICKEN_BITS, 0xfc910007, 0x00810007,
mmSDMA1_GFX_IB_CNTL, 0x800f0111, 0x00000100,
mmSDMA1_RLC0_IB_CNTL, 0x800f0111, 0x00000100,
mmSDMA1_RLC1_IB_CNTL, 0x800f0111, 0x00000100,
};
static const u32 fiji_mgcg_cgcg_init[] =
{
mmSDMA0_CLK_CTRL, 0xff000ff0, 0x00000100,
mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100
};
static const u32 cz_golden_settings_a11[] =
{
mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007,
@@ -122,6 +142,14 @@ static const u32 cz_mgcg_cgcg_init[] =
static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev)
{
switch (adev->asic_type) {
case CHIP_FIJI:
amdgpu_program_register_sequence(adev,
fiji_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
amdgpu_program_register_sequence(adev,
golden_settings_fiji_a10,
(const u32)ARRAY_SIZE(golden_settings_fiji_a10));
break;
case CHIP_TONGA:
amdgpu_program_register_sequence(adev,
tonga_mgcg_cgcg_init,
@@ -167,6 +195,9 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
case CHIP_TONGA:
chip_name = "tonga";
break;
case CHIP_FIJI:
chip_name = "fiji";
break;
case CHIP_CARRIZO:
chip_name = "carrizo";
break;
@@ -763,6 +794,7 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct fence *f = NULL;
unsigned i;
unsigned index;
int r;
@@ -778,12 +810,10 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
return r;
goto err0;
}
ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
@@ -797,19 +827,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
ib.length_dw = 8;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
AMDGPU_FENCE_OWNER_UNDEFINED,
&f);
if (r)
goto err1;
r = fence_wait(f, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
return r;
goto err1;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -819,12 +846,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
ring->idx, i);
goto err1;
} else {
DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
err1:
fence_put(f);
amdgpu_ib_free(adev, &ib);
err0:
amdgpu_wb_free(adev, index);
return r;
}
@@ -1509,5 +1540,6 @@ static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev)
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs;
adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c
+3 -3
View File
@@ -534,7 +534,7 @@ static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_fence *fence = NULL;
struct fence *fence = NULL;
int r;
r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
@@ -555,14 +555,14 @@ static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring)
goto error;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(fence, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
goto error;
}
DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
error:
amdgpu_fence_unref(&fence);
fence_put(fence);
amdgpu_asic_set_uvd_clocks(adev, 0, 0);
return r;
}
drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c
+3 -3
View File
@@ -580,7 +580,7 @@ static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_fence *fence = NULL;
struct fence *fence = NULL;
int r;
r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
@@ -601,14 +601,14 @@ static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring)
goto error;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(fence, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
goto error;
}
DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
error:
amdgpu_fence_unref(&fence);
fence_put(fence);
amdgpu_asic_set_uvd_clocks(adev, 0, 0);
return r;
}
drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
+3 -3
View File
@@ -575,7 +575,7 @@ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
*/
static int uvd_v6_0_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_fence *fence = NULL;
struct fence *fence = NULL;
int r;
r = amdgpu_uvd_get_create_msg(ring, 1, NULL);
@@ -590,14 +590,14 @@ static int uvd_v6_0_ring_test_ib(struct amdgpu_ring *ring)
goto error;
}
r = amdgpu_fence_wait(fence, false);
r = fence_wait(fence, false);
if (r) {
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
goto error;
}
DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
error:
amdgpu_fence_unref(&fence);
fence_put(fence);
return r;
}
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
+8 -1
View File
@@ -205,7 +205,14 @@ static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
u32 tmp;
unsigned ret;
if (adev->flags & AMDGPU_IS_APU)
/* Fiji is single pipe */
if (adev->asic_type == CHIP_FIJI) {
ret = AMDGPU_VCE_HARVEST_VCE1;
return ret;
}
/* Tonga and CZ are dual or single pipe */
if (adev->flags & AMD_IS_APU)
tmp = (RREG32_SMC(ixVCE_HARVEST_FUSE_MACRO__ADDRESS) &
VCE_HARVEST_FUSE_MACRO__MASK) >>
VCE_HARVEST_FUSE_MACRO__SHIFT;
drivers/gpu/drm/amd/amdgpu/vi.c
+115 -5
View File
@@ -203,6 +203,17 @@ static const u32 tonga_mgcg_cgcg_init[] =
mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104,
};
static const u32 fiji_mgcg_cgcg_init[] =
{
mmCGTT_DRM_CLK_CTRL0, 0xffffffff, 0x00600100,
mmPCIE_INDEX, 0xffffffff, 0x0140001c,
mmPCIE_DATA, 0x000f0000, 0x00000000,
mmSMC_IND_INDEX_4, 0xffffffff, 0xC060000C,
mmSMC_IND_DATA_4, 0xc0000fff, 0x00000100,
mmCGTT_DRM_CLK_CTRL0, 0xff000fff, 0x00000100,
mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104,
};
static const u32 iceland_mgcg_cgcg_init[] =
{
mmPCIE_INDEX, 0xffffffff, ixPCIE_CNTL2,
@@ -232,6 +243,11 @@ static void vi_init_golden_registers(struct amdgpu_device *adev)
iceland_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(iceland_mgcg_cgcg_init));
break;
case CHIP_FIJI:
amdgpu_program_register_sequence(adev,
fiji_mgcg_cgcg_init,
(const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
break;
case CHIP_TONGA:
amdgpu_program_register_sequence(adev,
tonga_mgcg_cgcg_init,
@@ -261,7 +277,7 @@ static u32 vi_get_xclk(struct amdgpu_device *adev)
u32 reference_clock = adev->clock.spll.reference_freq;
u32 tmp;
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return reference_clock;
tmp = RREG32_SMC(ixCG_CLKPIN_CNTL_2);
@@ -362,6 +378,26 @@ static struct amdgpu_allowed_register_entry cz_allowed_read_registers[] = {
static struct amdgpu_allowed_register_entry vi_allowed_read_registers[] = {
{mmGRBM_STATUS, false},
{mmGRBM_STATUS2, false},
{mmGRBM_STATUS_SE0, false},
{mmGRBM_STATUS_SE1, false},
{mmGRBM_STATUS_SE2, false},
{mmGRBM_STATUS_SE3, false},
{mmSRBM_STATUS, false},
{mmSRBM_STATUS2, false},
{mmSRBM_STATUS3, false},
{mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET, false},
{mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET, false},
{mmCP_STAT, false},
{mmCP_STALLED_STAT1, false},
{mmCP_STALLED_STAT2, false},
{mmCP_STALLED_STAT3, false},
{mmCP_CPF_BUSY_STAT, false},
{mmCP_CPF_STALLED_STAT1, false},
{mmCP_CPF_STATUS, false},
{mmCP_CPC_BUSY_STAT, false},
{mmCP_CPC_STALLED_STAT1, false},
{mmCP_CPC_STATUS, false},
{mmGB_ADDR_CONFIG, false},
{mmMC_ARB_RAMCFG, false},
{mmGB_TILE_MODE0, false},
@@ -449,6 +485,7 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num,
asic_register_table = tonga_allowed_read_registers;
size = ARRAY_SIZE(tonga_allowed_read_registers);
break;
case CHIP_FIJI:
case CHIP_TONGA:
case CHIP_CARRIZO:
asic_register_table = cz_allowed_read_registers;
@@ -751,7 +788,7 @@ static void vi_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask)
srbm_soft_reset =
REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
if (!(adev->flags & AMDGPU_IS_APU)) {
if (!(adev->flags & AMD_IS_APU)) {
if (reset_mask & AMDGPU_RESET_MC)
srbm_soft_reset =
REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
@@ -971,7 +1008,7 @@ static void vi_pcie_gen3_enable(struct amdgpu_device *adev)
if (amdgpu_pcie_gen2 == 0)
return;
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return;
ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask);
@@ -999,7 +1036,7 @@ static void vi_enable_doorbell_aperture(struct amdgpu_device *adev,
u32 tmp;
/* not necessary on CZ */
if (adev->flags & AMDGPU_IS_APU)
if (adev->flags & AMD_IS_APU)
return;
tmp = RREG32(mmBIF_DOORBELL_APER_EN);
@@ -1127,6 +1164,74 @@ static const struct amdgpu_ip_block_version tonga_ip_blocks[] =
},
};
static const struct amdgpu_ip_block_version fiji_ip_blocks[] =
{
/* ORDER MATTERS! */
{
.type = AMD_IP_BLOCK_TYPE_COMMON,
.major = 2,
.minor = 0,
.rev = 0,
.funcs = &vi_common_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_GMC,
.major = 8,
.minor = 5,
.rev = 0,
.funcs = &gmc_v8_0_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_IH,
.major = 3,
.minor = 0,
.rev = 0,
.funcs = &tonga_ih_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_SMC,
.major = 7,
.minor = 1,
.rev = 0,
.funcs = &fiji_dpm_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_DCE,
.major = 10,
.minor = 1,
.rev = 0,
.funcs = &dce_v10_0_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_GFX,
.major = 8,
.minor = 0,
.rev = 0,
.funcs = &gfx_v8_0_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_SDMA,
.major = 3,
.minor = 0,
.rev = 0,
.funcs = &sdma_v3_0_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_UVD,
.major = 6,
.minor = 0,
.rev = 0,
.funcs = &uvd_v6_0_ip_funcs,
},
{
.type = AMD_IP_BLOCK_TYPE_VCE,
.major = 3,
.minor = 0,
.rev = 0,
.funcs = &vce_v3_0_ip_funcs,
},
};
static const struct amdgpu_ip_block_version cz_ip_blocks[] =
{
/* ORDER MATTERS! */
@@ -1202,6 +1307,10 @@ int vi_set_ip_blocks(struct amdgpu_device *adev)
adev->ip_blocks = topaz_ip_blocks;
adev->num_ip_blocks = ARRAY_SIZE(topaz_ip_blocks);
break;
case CHIP_FIJI:
adev->ip_blocks = fiji_ip_blocks;
adev->num_ip_blocks = ARRAY_SIZE(fiji_ip_blocks);
break;
case CHIP_TONGA:
adev->ip_blocks = tonga_ip_blocks;
adev->num_ip_blocks = ARRAY_SIZE(tonga_ip_blocks);
@@ -1248,7 +1357,7 @@ static int vi_common_early_init(void *handle)
bool smc_enabled = false;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
if (adev->flags & AMDGPU_IS_APU) {
if (adev->flags & AMD_IS_APU) {
adev->smc_rreg = &cz_smc_rreg;
adev->smc_wreg = &cz_smc_wreg;
} else {
@@ -1279,6 +1388,7 @@ static int vi_common_early_init(void *handle)
if (amdgpu_smc_load_fw && smc_enabled)
adev->firmware.smu_load = true;
break;
case CHIP_FIJI:
case CHIP_TONGA:
adev->has_uvd = true;
adev->cg_flags = 0;
drivers/gpu/drm/amd/amdgpu/vi_dpm.h
+1 -1
View File
@@ -30,7 +30,7 @@ int cz_smu_start(struct amdgpu_device *adev);
int cz_smu_fini(struct amdgpu_device *adev);
extern const struct amd_ip_funcs tonga_dpm_ip_funcs;
extern const struct amd_ip_funcs fiji_dpm_ip_funcs;
extern const struct amd_ip_funcs iceland_dpm_ip_funcs;
#endif
drivers/gpu/drm/amd/include/amd_shared.h
+39
View File
@@ -23,6 +23,45 @@
#ifndef __AMD_SHARED_H__
#define __AMD_SHARED_H__
#define AMD_MAX_USEC_TIMEOUT 100000 /* 100 ms */
/*
* Supported GPU families (aligned with amdgpu_drm.h)
*/
#define AMD_FAMILY_UNKNOWN 0
#define AMD_FAMILY_CI 120 /* Bonaire, Hawaii */
#define AMD_FAMILY_KV 125 /* Kaveri, Kabini, Mullins */
#define AMD_FAMILY_VI 130 /* Iceland, Tonga */
#define AMD_FAMILY_CZ 135 /* Carrizo */
/*
* Supported ASIC types
*/
enum amd_asic_type {
CHIP_BONAIRE = 0,
CHIP_KAVERI,
CHIP_KABINI,
CHIP_HAWAII,
CHIP_MULLINS,
CHIP_TOPAZ,
CHIP_TONGA,
CHIP_FIJI,
CHIP_CARRIZO,
CHIP_LAST,
};
/*
* Chip flags
*/
enum amd_chip_flags {
AMD_ASIC_MASK = 0x0000ffffUL,
AMD_FLAGS_MASK = 0xffff0000UL,
AMD_IS_MOBILITY = 0x00010000UL,
AMD_IS_APU = 0x00020000UL,
AMD_IS_PX = 0x00040000UL,
AMD_EXP_HW_SUPPORT = 0x00080000UL,
};
enum amd_ip_block_type {
AMD_IP_BLOCK_TYPE_COMMON,
AMD_IP_BLOCK_TYPE_GMC,
drivers/gpu/drm/amd/include/asic_reg/smu/smu_7_1_3_d.h
+1246
View File
File diff suppressed because it is too large Load Diff
drivers/gpu/drm/amd/include/asic_reg/smu/smu_7_1_3_enum.h
+1282
View File
File diff suppressed because it is too large Load Diff
drivers/gpu/drm/amd/include/asic_reg/smu/smu_7_1_3_sh_mask.h
+6080
View File
File diff suppressed because it is too large Load Diff
drivers/gpu/drm/amd/amdgpu/atom-bits.h → drivers/gpu/drm/amd/include/atom-bits.h
View File
drivers/gpu/drm/amd/amdgpu/atom-names.h → drivers/gpu/drm/amd/include/atom-names.h
View File
drivers/gpu/drm/amd/amdgpu/atom-types.h → drivers/gpu/drm/amd/include/atom-types.h
View File
drivers/gpu/drm/amd/amdgpu/atombios.h → drivers/gpu/drm/amd/include/atombios.h
View File
drivers/gpu/drm/amd/include/cgs_common.h
+624
View File
@@ -0,0 +1,624 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#ifndef _CGS_COMMON_H
#define _CGS_COMMON_H
#include "amd_shared.h"
/**
* enum cgs_gpu_mem_type - GPU memory types
*/
enum cgs_gpu_mem_type {
CGS_GPU_MEM_TYPE__VISIBLE_FB,
CGS_GPU_MEM_TYPE__INVISIBLE_FB,
CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB,
CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB,
CGS_GPU_MEM_TYPE__GART_CACHEABLE,
CGS_GPU_MEM_TYPE__GART_WRITECOMBINE
};
/**
* enum cgs_ind_reg - Indirect register spaces
*/
enum cgs_ind_reg {
CGS_IND_REG__MMIO,
CGS_IND_REG__PCIE,
CGS_IND_REG__SMC,
CGS_IND_REG__UVD_CTX,
CGS_IND_REG__DIDT,
CGS_IND_REG__AUDIO_ENDPT
};
/**
* enum cgs_clock - Clocks controlled by the SMU
*/
enum cgs_clock {
CGS_CLOCK__SCLK,
CGS_CLOCK__MCLK,
CGS_CLOCK__VCLK,
CGS_CLOCK__DCLK,
CGS_CLOCK__ECLK,
CGS_CLOCK__ACLK,
CGS_CLOCK__ICLK,
/* ... */
};
/**
* enum cgs_engine - Engines that can be statically power-gated
*/
enum cgs_engine {
CGS_ENGINE__UVD,
CGS_ENGINE__VCE,
CGS_ENGINE__VP8,
CGS_ENGINE__ACP_DMA,
CGS_ENGINE__ACP_DSP0,
CGS_ENGINE__ACP_DSP1,
CGS_ENGINE__ISP,
/* ... */
};
/**
* enum cgs_voltage_planes - Voltage planes for external camera HW
*/
enum cgs_voltage_planes {
CGS_VOLTAGE_PLANE__SENSOR0,
CGS_VOLTAGE_PLANE__SENSOR1,
/* ... */
};
/*
* enum cgs_ucode_id - Firmware types for different IPs
*/
enum cgs_ucode_id {
CGS_UCODE_ID_SMU = 0,
CGS_UCODE_ID_SDMA0,
CGS_UCODE_ID_SDMA1,
CGS_UCODE_ID_CP_CE,
CGS_UCODE_ID_CP_PFP,
CGS_UCODE_ID_CP_ME,
CGS_UCODE_ID_CP_MEC,
CGS_UCODE_ID_CP_MEC_JT1,
CGS_UCODE_ID_CP_MEC_JT2,
CGS_UCODE_ID_GMCON_RENG,
CGS_UCODE_ID_RLC_G,
CGS_UCODE_ID_MAXIMUM,
};
/**
* struct cgs_clock_limits - Clock limits
*
* Clocks are specified in 10KHz units.
*/
struct cgs_clock_limits {
unsigned min; /**< Minimum supported frequency */
unsigned max; /**< Maxumim supported frequency */
unsigned sustainable; /**< Thermally sustainable frequency */
};
/**
* struct cgs_firmware_info - Firmware information
*/
struct cgs_firmware_info {
uint16_t version;
uint16_t feature_version;
uint32_t image_size;
uint64_t mc_addr;
void *kptr;
};
typedef unsigned long cgs_handle_t;
/**
* cgs_gpu_mem_info() - Return information about memory heaps
* @cgs_device: opaque device handle
* @type: memory type
* @mc_start: Start MC address of the heap (output)
* @mc_size: MC address space size (output)
* @mem_size: maximum amount of memory available for allocation (output)
*
* This function returns information about memory heaps. The type
* parameter is used to select the memory heap. The mc_start and
* mc_size for GART heaps may be bigger than the memory available for
* allocation.
*
* mc_start and mc_size are undefined for non-contiguous FB memory
* types, since buffers allocated with these types may or may not be
* GART mapped.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_gpu_mem_info_t)(void *cgs_device, enum cgs_gpu_mem_type type,
uint64_t *mc_start, uint64_t *mc_size,
uint64_t *mem_size);
/**
* cgs_gmap_kmem() - map kernel memory to GART aperture
* @cgs_device: opaque device handle
* @kmem: pointer to kernel memory
* @size: size to map
* @min_offset: minimum offset from start of GART aperture
* @max_offset: maximum offset from start of GART aperture
* @kmem_handle: kernel memory handle (output)
* @mcaddr: MC address (output)
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_gmap_kmem_t)(void *cgs_device, void *kmem, uint64_t size,
uint64_t min_offset, uint64_t max_offset,
cgs_handle_t *kmem_handle, uint64_t *mcaddr);
/**
* cgs_gunmap_kmem() - unmap kernel memory
* @cgs_device: opaque device handle
* @kmem_handle: kernel memory handle returned by gmap_kmem
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_gunmap_kmem_t)(void *cgs_device, cgs_handle_t kmem_handle);
/**
* cgs_alloc_gpu_mem() - Allocate GPU memory
* @cgs_device: opaque device handle
* @type: memory type
* @size: size in bytes
* @align: alignment in bytes
* @min_offset: minimum offset from start of heap
* @max_offset: maximum offset from start of heap
* @handle: memory handle (output)
*
* The memory types CGS_GPU_MEM_TYPE_*_CONTIG_FB force contiguous
* memory allocation. This guarantees that the MC address returned by
* cgs_gmap_gpu_mem is not mapped through the GART. The non-contiguous
* FB memory types may be GART mapped depending on memory
* fragmentation and memory allocator policies.
*
* If min/max_offset are non-0, the allocation will be forced to
* reside between these offsets in its respective memory heap. The
* base address that the offset relates to, depends on the memory
* type.
*
* - CGS_GPU_MEM_TYPE__*_CONTIG_FB: FB MC base address
* - CGS_GPU_MEM_TYPE__GART_*: GART aperture base address
* - others: undefined, don't use with max_offset
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_alloc_gpu_mem_t)(void *cgs_device, enum cgs_gpu_mem_type type,
uint64_t size, uint64_t align,
uint64_t min_offset, uint64_t max_offset,
cgs_handle_t *handle);
/**
* cgs_free_gpu_mem() - Free GPU memory
* @cgs_device: opaque device handle
* @handle: memory handle returned by alloc or import
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_free_gpu_mem_t)(void *cgs_device, cgs_handle_t handle);
/**
* cgs_gmap_gpu_mem() - GPU-map GPU memory
* @cgs_device: opaque device handle
* @handle: memory handle returned by alloc or import
* @mcaddr: MC address (output)
*
* Ensures that a buffer is GPU accessible and returns its MC address.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_gmap_gpu_mem_t)(void *cgs_device, cgs_handle_t handle,
uint64_t *mcaddr);
/**
* cgs_gunmap_gpu_mem() - GPU-unmap GPU memory
* @cgs_device: opaque device handle
* @handle: memory handle returned by alloc or import
*
* Allows the buffer to be migrated while it's not used by the GPU.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_gunmap_gpu_mem_t)(void *cgs_device, cgs_handle_t handle);
/**
* cgs_kmap_gpu_mem() - Kernel-map GPU memory
*
* @cgs_device: opaque device handle
* @handle: memory handle returned by alloc or import
* @map: Kernel virtual address the memory was mapped to (output)
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_kmap_gpu_mem_t)(void *cgs_device, cgs_handle_t handle,
void **map);
/**
* cgs_kunmap_gpu_mem() - Kernel-unmap GPU memory
* @cgs_device: opaque device handle
* @handle: memory handle returned by alloc or import
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_kunmap_gpu_mem_t)(void *cgs_device, cgs_handle_t handle);
/**
* cgs_read_register() - Read an MMIO register
* @cgs_device: opaque device handle
* @offset: register offset
*
* Return: register value
*/
typedef uint32_t (*cgs_read_register_t)(void *cgs_device, unsigned offset);
/**
* cgs_write_register() - Write an MMIO register
* @cgs_device: opaque device handle
* @offset: register offset
* @value: register value
*/
typedef void (*cgs_write_register_t)(void *cgs_device, unsigned offset,
uint32_t value);
/**
* cgs_read_ind_register() - Read an indirect register
* @cgs_device: opaque device handle
* @offset: register offset
*
* Return: register value
*/
typedef uint32_t (*cgs_read_ind_register_t)(void *cgs_device, enum cgs_ind_reg space,
unsigned index);
/**
* cgs_write_ind_register() - Write an indirect register
* @cgs_device: opaque device handle
* @offset: register offset
* @value: register value
*/
typedef void (*cgs_write_ind_register_t)(void *cgs_device, enum cgs_ind_reg space,
unsigned index, uint32_t value);
/**
* cgs_read_pci_config_byte() - Read byte from PCI configuration space
* @cgs_device: opaque device handle
* @addr: address
*
* Return: Value read
*/
typedef uint8_t (*cgs_read_pci_config_byte_t)(void *cgs_device, unsigned addr);
/**
* cgs_read_pci_config_word() - Read word from PCI configuration space
* @cgs_device: opaque device handle
* @addr: address, must be word-aligned
*
* Return: Value read
*/
typedef uint16_t (*cgs_read_pci_config_word_t)(void *cgs_device, unsigned addr);
/**
* cgs_read_pci_config_dword() - Read dword from PCI configuration space
* @cgs_device: opaque device handle
* @addr: address, must be dword-aligned
*
* Return: Value read
*/
typedef uint32_t (*cgs_read_pci_config_dword_t)(void *cgs_device,
unsigned addr);
/**
* cgs_write_pci_config_byte() - Write byte to PCI configuration space
* @cgs_device: opaque device handle
* @addr: address
* @value: value to write
*/
typedef void (*cgs_write_pci_config_byte_t)(void *cgs_device, unsigned addr,
uint8_t value);
/**
* cgs_write_pci_config_word() - Write byte to PCI configuration space
* @cgs_device: opaque device handle
* @addr: address, must be word-aligned
* @value: value to write
*/
typedef void (*cgs_write_pci_config_word_t)(void *cgs_device, unsigned addr,
uint16_t value);
/**
* cgs_write_pci_config_dword() - Write byte to PCI configuration space
* @cgs_device: opaque device handle
* @addr: address, must be dword-aligned
* @value: value to write
*/
typedef void (*cgs_write_pci_config_dword_t)(void *cgs_device, unsigned addr,
uint32_t value);
/**
* cgs_atom_get_data_table() - Get a pointer to an ATOM BIOS data table
* @cgs_device: opaque device handle
* @table: data table index
* @size: size of the table (output, may be NULL)
* @frev: table format revision (output, may be NULL)
* @crev: table content revision (output, may be NULL)
*
* Return: Pointer to start of the table, or NULL on failure
*/
typedef const void *(*cgs_atom_get_data_table_t)(
void *cgs_device, unsigned table,
uint16_t *size, uint8_t *frev, uint8_t *crev);
/**
* cgs_atom_get_cmd_table_revs() - Get ATOM BIOS command table revisions
* @cgs_device: opaque device handle
* @table: data table index
* @frev: table format revision (output, may be NULL)
* @crev: table content revision (output, may be NULL)
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_atom_get_cmd_table_revs_t)(void *cgs_device, unsigned table,
uint8_t *frev, uint8_t *crev);
/**
* cgs_atom_exec_cmd_table() - Execute an ATOM BIOS command table
* @cgs_device: opaque device handle
* @table: command table index
* @args: arguments
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_atom_exec_cmd_table_t)(void *cgs_device,
unsigned table, void *args);
/**
* cgs_create_pm_request() - Create a power management request
* @cgs_device: opaque device handle
* @request: handle of created PM request (output)
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_create_pm_request_t)(void *cgs_device, cgs_handle_t *request);
/**
* cgs_destroy_pm_request() - Destroy a power management request
* @cgs_device: opaque device handle
* @request: handle of created PM request
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_destroy_pm_request_t)(void *cgs_device, cgs_handle_t request);
/**
* cgs_set_pm_request() - Activate or deactiveate a PM request
* @cgs_device: opaque device handle
* @request: PM request handle
* @active: 0 = deactivate, non-0 = activate
*
* While a PM request is active, its minimum clock requests are taken
* into account as the requested engines are powered up. When the
* request is inactive, the engines may be powered down and clocks may
* be lower, depending on other PM requests by other driver
* components.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_set_pm_request_t)(void *cgs_device, cgs_handle_t request,
int active);
/**
* cgs_pm_request_clock() - Request a minimum frequency for a specific clock
* @cgs_device: opaque device handle
* @request: PM request handle
* @clock: which clock?
* @freq: requested min. frequency in 10KHz units (0 to clear request)
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_pm_request_clock_t)(void *cgs_device, cgs_handle_t request,
enum cgs_clock clock, unsigned freq);
/**
* cgs_pm_request_engine() - Request an engine to be powered up
* @cgs_device: opaque device handle
* @request: PM request handle
* @engine: which engine?
* @powered: 0 = powered down, non-0 = powered up
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_pm_request_engine_t)(void *cgs_device, cgs_handle_t request,
enum cgs_engine engine, int powered);
/**
* cgs_pm_query_clock_limits() - Query clock frequency limits
* @cgs_device: opaque device handle
* @clock: which clock?
* @limits: clock limits
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_pm_query_clock_limits_t)(void *cgs_device,
enum cgs_clock clock,
struct cgs_clock_limits *limits);
/**
* cgs_set_camera_voltages() - Apply specific voltages to PMIC voltage planes
* @cgs_device: opaque device handle
* @mask: bitmask of voltages to change (1<<CGS_VOLTAGE_PLANE__xyz|...)
* @voltages: pointer to array of voltage values in 1mV units
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_set_camera_voltages_t)(void *cgs_device, uint32_t mask,
const uint32_t *voltages);
/**
* cgs_get_firmware_info - Get the firmware information from core driver
* @cgs_device: opaque device handle
* @type: the firmware type
* @info: returend firmware information
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_get_firmware_info)(void *cgs_device,
enum cgs_ucode_id type,
struct cgs_firmware_info *info);
typedef int(*cgs_set_powergating_state)(void *cgs_device,
enum amd_ip_block_type block_type,
enum amd_powergating_state state);
typedef int(*cgs_set_clockgating_state)(void *cgs_device,
enum amd_ip_block_type block_type,
enum amd_clockgating_state state);
struct cgs_ops {
/* memory management calls (similar to KFD interface) */
cgs_gpu_mem_info_t gpu_mem_info;
cgs_gmap_kmem_t gmap_kmem;
cgs_gunmap_kmem_t gunmap_kmem;
cgs_alloc_gpu_mem_t alloc_gpu_mem;
cgs_free_gpu_mem_t free_gpu_mem;
cgs_gmap_gpu_mem_t gmap_gpu_mem;
cgs_gunmap_gpu_mem_t gunmap_gpu_mem;
cgs_kmap_gpu_mem_t kmap_gpu_mem;
cgs_kunmap_gpu_mem_t kunmap_gpu_mem;
/* MMIO access */
cgs_read_register_t read_register;
cgs_write_register_t write_register;
cgs_read_ind_register_t read_ind_register;
cgs_write_ind_register_t write_ind_register;
/* PCI configuration space access */
cgs_read_pci_config_byte_t read_pci_config_byte;
cgs_read_pci_config_word_t read_pci_config_word;
cgs_read_pci_config_dword_t read_pci_config_dword;
cgs_write_pci_config_byte_t write_pci_config_byte;
cgs_write_pci_config_word_t write_pci_config_word;
cgs_write_pci_config_dword_t write_pci_config_dword;
/* ATOM BIOS */
cgs_atom_get_data_table_t atom_get_data_table;
cgs_atom_get_cmd_table_revs_t atom_get_cmd_table_revs;
cgs_atom_exec_cmd_table_t atom_exec_cmd_table;
/* Power management */
cgs_create_pm_request_t create_pm_request;
cgs_destroy_pm_request_t destroy_pm_request;
cgs_set_pm_request_t set_pm_request;
cgs_pm_request_clock_t pm_request_clock;
cgs_pm_request_engine_t pm_request_engine;
cgs_pm_query_clock_limits_t pm_query_clock_limits;
cgs_set_camera_voltages_t set_camera_voltages;
/* Firmware Info */
cgs_get_firmware_info get_firmware_info;
/* cg pg interface*/
cgs_set_powergating_state set_powergating_state;
cgs_set_clockgating_state set_clockgating_state;
/* ACPI (TODO) */
};
struct cgs_os_ops; /* To be define in OS-specific CGS header */
struct cgs_device
{
const struct cgs_ops *ops;
const struct cgs_os_ops *os_ops;
/* to be embedded at the start of driver private structure */
};
/* Convenience macros that make CGS indirect function calls look like
* normal function calls */
#define CGS_CALL(func,dev,...) \
(((struct cgs_device *)dev)->ops->func(dev, ##__VA_ARGS__))
#define CGS_OS_CALL(func,dev,...) \
(((struct cgs_device *)dev)->os_ops->func(dev, ##__VA_ARGS__))
#define cgs_gpu_mem_info(dev,type,mc_start,mc_size,mem_size) \
CGS_CALL(gpu_mem_info,dev,type,mc_start,mc_size,mem_size)
#define cgs_gmap_kmem(dev,kmem,size,min_off,max_off,kmem_handle,mcaddr) \
CGS_CALL(gmap_kmem,dev,kmem,size,min_off,max_off,kmem_handle,mcaddr)
#define cgs_gunmap_kmem(dev,kmem_handle) \
CGS_CALL(gunmap_kmem,dev,keme_handle)
#define cgs_alloc_gpu_mem(dev,type,size,align,min_off,max_off,handle) \
CGS_CALL(alloc_gpu_mem,dev,type,size,align,min_off,max_off,handle)
#define cgs_free_gpu_mem(dev,handle) \
CGS_CALL(free_gpu_mem,dev,handle)
#define cgs_gmap_gpu_mem(dev,handle,mcaddr) \
CGS_CALL(gmap_gpu_mem,dev,handle,mcaddr)
#define cgs_gunmap_gpu_mem(dev,handle) \
CGS_CALL(gunmap_gpu_mem,dev,handle)
#define cgs_kmap_gpu_mem(dev,handle,map) \
CGS_CALL(kmap_gpu_mem,dev,handle,map)
#define cgs_kunmap_gpu_mem(dev,handle) \
CGS_CALL(kunmap_gpu_mem,dev,handle)
#define cgs_read_register(dev,offset) \
CGS_CALL(read_register,dev,offset)
#define cgs_write_register(dev,offset,value) \
CGS_CALL(write_register,dev,offset,value)
#define cgs_read_ind_register(dev,space,index) \
CGS_CALL(read_ind_register,dev,space,index)
#define cgs_write_ind_register(dev,space,index,value) \
CGS_CALL(write_ind_register,dev,space,index,value)
#define cgs_read_pci_config_byte(dev,addr) \
CGS_CALL(read_pci_config_byte,dev,addr)
#define cgs_read_pci_config_word(dev,addr) \
CGS_CALL(read_pci_config_word,dev,addr)
#define cgs_read_pci_config_dword(dev,addr) \
CGS_CALL(read_pci_config_dword,dev,addr)
#define cgs_write_pci_config_byte(dev,addr,value) \
CGS_CALL(write_pci_config_byte,dev,addr,value)
#define cgs_write_pci_config_word(dev,addr,value) \
CGS_CALL(write_pci_config_word,dev,addr,value)
#define cgs_write_pci_config_dword(dev,addr,value) \
CGS_CALL(write_pci_config_dword,dev,addr,value)
#define cgs_atom_get_data_table(dev,table,size,frev,crev) \
CGS_CALL(atom_get_data_table,dev,table,size,frev,crev)
#define cgs_atom_get_cmd_table_revs(dev,table,frev,crev) \
CGS_CALL(atom_get_cmd_table_revs,dev,table,frev,crev)
#define cgs_atom_exec_cmd_table(dev,table,args) \
CGS_CALL(atom_exec_cmd_table,dev,table,args)
#define cgs_create_pm_request(dev,request) \
CGS_CALL(create_pm_request,dev,request)
#define cgs_destroy_pm_request(dev,request) \
CGS_CALL(destroy_pm_request,dev,request)
#define cgs_set_pm_request(dev,request,active) \
CGS_CALL(set_pm_request,dev,request,active)
#define cgs_pm_request_clock(dev,request,clock,freq) \
CGS_CALL(pm_request_clock,dev,request,clock,freq)
#define cgs_pm_request_engine(dev,request,engine,powered) \
CGS_CALL(pm_request_engine,dev,request,engine,powered)
#define cgs_pm_query_clock_limits(dev,clock,limits) \
CGS_CALL(pm_query_clock_limits,dev,clock,limits)
#define cgs_set_camera_voltages(dev,mask,voltages) \
CGS_CALL(set_camera_voltages,dev,mask,voltages)
#define cgs_get_firmware_info(dev, type, info) \
CGS_CALL(get_firmware_info, dev, type, info)
#define cgs_set_powergating_state(dev, block_type, state) \
CGS_CALL(set_powergating_state, dev, block_type, state)
#define cgs_set_clockgating_state(dev, block_type, state) \
CGS_CALL(set_clockgating_state, dev, block_type, state)
#endif /* _CGS_COMMON_H */
drivers/gpu/drm/amd/include/cgs_linux.h
+135
View File
@@ -0,0 +1,135 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#ifndef _CGS_LINUX_H
#define _CGS_LINUX_H
#include "cgs_common.h"
/**
* cgs_import_gpu_mem() - Import dmabuf handle
* @cgs_device: opaque device handle
* @dmabuf_fd: DMABuf file descriptor
* @handle: memory handle (output)
*
* Must be called in the process context that dmabuf_fd belongs to.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_import_gpu_mem_t)(void *cgs_device, int dmabuf_fd,
cgs_handle_t *handle);
/**
* cgs_irq_source_set_func() - Callback for enabling/disabling interrupt sources
* @private_data: private data provided to cgs_add_irq_source
* @src_id: interrupt source ID
* @type: interrupt type
* @enabled: 0 = disable source, non-0 = enable source
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_irq_source_set_func_t)(void *private_data,
unsigned src_id, unsigned type,
int enabled);
/**
* cgs_irq_handler_func() - Interrupt handler callback
* @private_data: private data provided to cgs_add_irq_source
* @src_id: interrupt source ID
* @iv_entry: pointer to raw ih ring entry
*
* This callback runs in interrupt context.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_irq_handler_func_t)(void *private_data,
unsigned src_id, const uint32_t *iv_entry);
/**
* cgs_add_irq_source() - Add an IRQ source
* @cgs_device: opaque device handle
* @src_id: interrupt source ID
* @num_types: number of interrupt types that can be independently enabled
* @set: callback function to enable/disable an interrupt type
* @handler: interrupt handler callback
* @private_data: private data to pass to callback functions
*
* The same IRQ source can be added only once. Adding an IRQ source
* indicates ownership of that IRQ source and all its IRQ types.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_add_irq_source_t)(void *cgs_device, unsigned src_id,
unsigned num_types,
cgs_irq_source_set_func_t set,
cgs_irq_handler_func_t handler,
void *private_data);
/**
* cgs_irq_get() - Request enabling an IRQ source and type
* @cgs_device: opaque device handle
* @src_id: interrupt source ID
* @type: interrupt type
*
* cgs_irq_get and cgs_irq_put calls must be balanced. They count
* "references" to IRQ sources.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_irq_get_t)(void *cgs_device, unsigned src_id, unsigned type);
/**
* cgs_irq_put() - Indicate IRQ source is no longer needed
* @cgs_device: opaque device handle
* @src_id: interrupt source ID
* @type: interrupt type
*
* cgs_irq_get and cgs_irq_put calls must be balanced. They count
* "references" to IRQ sources. Even after cgs_irq_put is called, the
* IRQ handler may still be called if there are more refecences to
* the IRQ source.
*
* Return: 0 on success, -errno otherwise
*/
typedef int (*cgs_irq_put_t)(void *cgs_device, unsigned src_id, unsigned type);
struct cgs_os_ops {
cgs_import_gpu_mem_t import_gpu_mem;
/* IRQ handling */
cgs_add_irq_source_t add_irq_source;
cgs_irq_get_t irq_get;
cgs_irq_put_t irq_put;
};
#define cgs_import_gpu_mem(dev,dmabuf_fd,handle) \
CGS_OS_CALL(import_gpu_mem,dev,dmabuf_fd,handle)
#define cgs_add_irq_source(dev,src_id,num_types,set,handler,private_data) \
CGS_OS_CALL(add_irq_source,dev,src_id,num_types,set,handler, \
private_data)
#define cgs_irq_get(dev,src_id,type) \
CGS_OS_CALL(irq_get,dev,src_id,type)
#define cgs_irq_put(dev,src_id,type) \
CGS_OS_CALL(irq_put,dev,src_id,type)
#endif /* _CGS_LINUX_H */
drivers/gpu/drm/amd/amdgpu/pptable.h → drivers/gpu/drm/amd/include/pptable.h
+5 -1
View File
@@ -146,6 +146,9 @@ typedef struct _ATOM_PPLIB_EXTENDEDHEADER
#define ATOM_PP_PLATFORM_CAP_VRHOT_GPIO_CONFIGURABLE 0x00200000 // Does the driver supports VR HOT GPIO Configurable.
#define ATOM_PP_PLATFORM_CAP_TEMP_INVERSION 0x00400000 // Does the driver supports Temp Inversion feature.
#define ATOM_PP_PLATFORM_CAP_EVV 0x00800000
#define ATOM_PP_PLATFORM_COMBINE_PCC_WITH_THERMAL_SIGNAL 0x01000000
#define ATOM_PP_PLATFORM_LOAD_POST_PRODUCTION_FIRMWARE 0x02000000
#define ATOM_PP_PLATFORM_CAP_DISABLE_USING_ACTUAL_TEMPERATURE_FOR_POWER_CALC 0x04000000
typedef struct _ATOM_PPLIB_POWERPLAYTABLE
{
@@ -673,7 +676,8 @@ typedef struct _ATOM_PPLIB_POWERTUNE_Table_V1
UCHAR revid;
ATOM_PowerTune_Table power_tune_table;
USHORT usMaximumPowerDeliveryLimit;
USHORT usReserve[7];
USHORT usTjMax;
USHORT usReserve[6];
} ATOM_PPLIB_POWERTUNE_Table_V1;
#define ATOM_PPM_A_A 1
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+462
View File
@@ -0,0 +1,462 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#include <linux/kthread.h>
#include <linux/wait.h>
#include <linux/sched.h>
#include <drm/drmP.h>
#include "gpu_scheduler.h"
/* Initialize a given run queue struct */
static void amd_sched_rq_init(struct amd_sched_rq *rq)
{
INIT_LIST_HEAD(&rq->entities);
mutex_init(&rq->lock);
rq->current_entity = NULL;
}
static void amd_sched_rq_add_entity(struct amd_sched_rq *rq,
struct amd_sched_entity *entity)
{
mutex_lock(&rq->lock);
list_add_tail(&entity->list, &rq->entities);
mutex_unlock(&rq->lock);
}
static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq,
struct amd_sched_entity *entity)
{
mutex_lock(&rq->lock);
list_del_init(&entity->list);
if (rq->current_entity == entity)
rq->current_entity = NULL;
mutex_unlock(&rq->lock);
}
/**
* Select next entity from a specified run queue with round robin policy.
* It could return the same entity as current one if current is the only
* available one in the queue. Return NULL if nothing available.
*/
static struct amd_sched_entity *
amd_sched_rq_select_entity(struct amd_sched_rq *rq)
{
struct amd_sched_entity *entity = rq->current_entity;
if (entity) {
list_for_each_entry_continue(entity, &rq->entities, list) {
if (!kfifo_is_empty(&entity->job_queue)) {
rq->current_entity = entity;
return rq->current_entity;
}
}
}
list_for_each_entry(entity, &rq->entities, list) {
if (!kfifo_is_empty(&entity->job_queue)) {
rq->current_entity = entity;
return rq->current_entity;
}
if (entity == rq->current_entity)
break;
}
return NULL;
}
/**
* Note: This function should only been called inside scheduler main
* function for thread safety, there is no other protection here.
* return ture if scheduler has something ready to run.
*
* For active_hw_rq, there is only one producer(scheduler thread) and
* one consumer(ISR). It should be safe to use this function in scheduler
* main thread to decide whether to continue emit more IBs.
*/
static bool is_scheduler_ready(struct amd_gpu_scheduler *sched)
{
unsigned long flags;
bool full;
spin_lock_irqsave(&sched->queue_lock, flags);
full = atomic64_read(&sched->hw_rq_count) <
sched->hw_submission_limit ? true : false;
spin_unlock_irqrestore(&sched->queue_lock, flags);
return full;
}
/**
* Select next entity from the kernel run queue, if not available,
* return null.
*/
static struct amd_sched_entity *
kernel_rq_select_context(struct amd_gpu_scheduler *sched)
{
struct amd_sched_entity *sched_entity;
struct amd_sched_rq *rq = &sched->kernel_rq;
mutex_lock(&rq->lock);
sched_entity = amd_sched_rq_select_entity(rq);
mutex_unlock(&rq->lock);
return sched_entity;
}
/**
* Select next entity containing real IB submissions
*/
static struct amd_sched_entity *
select_context(struct amd_gpu_scheduler *sched)
{
struct amd_sched_entity *wake_entity = NULL;
struct amd_sched_entity *tmp;
struct amd_sched_rq *rq;
if (!is_scheduler_ready(sched))
return NULL;
/* Kernel run queue has higher priority than normal run queue*/
tmp = kernel_rq_select_context(sched);
if (tmp != NULL)
goto exit;
rq = &sched->sched_rq;
mutex_lock(&rq->lock);
tmp = amd_sched_rq_select_entity(rq);
mutex_unlock(&rq->lock);
exit:
if (sched->current_entity && (sched->current_entity != tmp))
wake_entity = sched->current_entity;
sched->current_entity = tmp;
if (wake_entity && wake_entity->need_wakeup)
wake_up(&wake_entity->wait_queue);
return tmp;
}
/**
* Init a context entity used by scheduler when submit to HW ring.
*
* @sched The pointer to the scheduler
* @entity The pointer to a valid amd_sched_entity
* @rq The run queue this entity belongs
* @kernel If this is an entity for the kernel
* @jobs The max number of jobs in the job queue
*
* return 0 if succeed. negative error code on failure
*/
int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity,
struct amd_sched_rq *rq,
uint32_t jobs)
{
uint64_t seq_ring = 0;
char name[20];
if (!(sched && entity && rq))
return -EINVAL;
memset(entity, 0, sizeof(struct amd_sched_entity));
seq_ring = ((uint64_t)sched->ring_id) << 60;
spin_lock_init(&entity->lock);
entity->belongto_rq = rq;
entity->scheduler = sched;
init_waitqueue_head(&entity->wait_queue);
init_waitqueue_head(&entity->wait_emit);
entity->fence_context = fence_context_alloc(1);
snprintf(name, sizeof(name), "c_entity[%llu]", entity->fence_context);
memcpy(entity->name, name, 20);
entity->need_wakeup = false;
if(kfifo_alloc(&entity->job_queue,
jobs * sizeof(void *),
GFP_KERNEL))
return -EINVAL;
spin_lock_init(&entity->queue_lock);
atomic64_set(&entity->last_queued_v_seq, seq_ring);
atomic64_set(&entity->last_signaled_v_seq, seq_ring);
/* Add the entity to the run queue */
amd_sched_rq_add_entity(rq, entity);
return 0;
}
/**
* Query if entity is initialized
*
* @sched Pointer to scheduler instance
* @entity The pointer to a valid scheduler entity
*
* return true if entity is initialized, false otherwise
*/
static bool is_context_entity_initialized(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity)
{
return entity->scheduler == sched &&
entity->belongto_rq != NULL;
}
static bool is_context_entity_idle(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity)
{
/**
* Idle means no pending IBs, and the entity is not
* currently being used.
*/
barrier();
if ((sched->current_entity != entity) &&
kfifo_is_empty(&entity->job_queue))
return true;
return false;
}
/**
* Destroy a context entity
*
* @sched Pointer to scheduler instance
* @entity The pointer to a valid scheduler entity
*
* return 0 if succeed. negative error code on failure
*/
int amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity)
{
int r = 0;
struct amd_sched_rq *rq = entity->belongto_rq;
if (!is_context_entity_initialized(sched, entity))
return 0;
entity->need_wakeup = true;
/**
* The client will not queue more IBs during this fini, consume existing
* queued IBs
*/
r = wait_event_timeout(
entity->wait_queue,
is_context_entity_idle(sched, entity),
msecs_to_jiffies(AMD_GPU_WAIT_IDLE_TIMEOUT_IN_MS)
) ? 0 : -1;
if (r) {
if (entity->is_pending)
DRM_INFO("Entity %p is in waiting state during fini,\
all pending ibs will be canceled.\n",
entity);
}
amd_sched_rq_remove_entity(rq, entity);
kfifo_free(&entity->job_queue);
return r;
}
/**
* Submit a normal job to the job queue
*
* @sched The pointer to the scheduler
* @c_entity The pointer to amd_sched_entity
* @job The pointer to job required to submit
* return 0 if succeed. -1 if failed.
* -2 indicate queue is full for this client, client should wait untill
* scheduler consum some queued command.
* -1 other fail.
*/
int amd_sched_push_job(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *c_entity,
void *data,
struct amd_sched_fence **fence)
{
struct amd_sched_job *job;
if (!fence)
return -EINVAL;
job = kzalloc(sizeof(struct amd_sched_job), GFP_KERNEL);
if (!job)
return -ENOMEM;
job->sched = sched;
job->s_entity = c_entity;
job->data = data;
*fence = amd_sched_fence_create(c_entity);
if ((*fence) == NULL) {
kfree(job);
return -EINVAL;
}
fence_get(&(*fence)->base);
job->s_fence = *fence;
while (kfifo_in_spinlocked(&c_entity->job_queue, &job, sizeof(void *),
&c_entity->queue_lock) != sizeof(void *)) {
/**
* Current context used up all its IB slots
* wait here, or need to check whether GPU is hung
*/
schedule();
}
/* first job wake up scheduler */
if ((kfifo_len(&c_entity->job_queue) / sizeof(void *)) == 1)
wake_up_interruptible(&sched->wait_queue);
return 0;
}
static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
{
struct amd_sched_job *sched_job =
container_of(cb, struct amd_sched_job, cb);
struct amd_gpu_scheduler *sched;
unsigned long flags;
sched = sched_job->sched;
atomic64_set(&sched_job->s_entity->last_signaled_v_seq,
sched_job->s_fence->v_seq);
amd_sched_fence_signal(sched_job->s_fence);
spin_lock_irqsave(&sched->queue_lock, flags);
list_del(&sched_job->list);
atomic64_dec(&sched->hw_rq_count);
spin_unlock_irqrestore(&sched->queue_lock, flags);
sched->ops->process_job(sched, sched_job);
fence_put(&sched_job->s_fence->base);
kfree(sched_job);
wake_up_interruptible(&sched->wait_queue);
}
static int amd_sched_main(void *param)
{
int r;
struct amd_sched_job *job;
struct sched_param sparam = {.sched_priority = 1};
struct amd_sched_entity *c_entity = NULL;
struct amd_gpu_scheduler *sched = (struct amd_gpu_scheduler *)param;
sched_setscheduler(current, SCHED_FIFO, &sparam);
while (!kthread_should_stop()) {
struct fence *fence;
wait_event_interruptible(sched->wait_queue,
is_scheduler_ready(sched) &&
(c_entity = select_context(sched)));
r = kfifo_out(&c_entity->job_queue, &job, sizeof(void *));
if (r != sizeof(void *))
continue;
r = sched->ops->prepare_job(sched, c_entity, job);
if (!r) {
unsigned long flags;
spin_lock_irqsave(&sched->queue_lock, flags);
list_add_tail(&job->list, &sched->active_hw_rq);
atomic64_inc(&sched->hw_rq_count);
spin_unlock_irqrestore(&sched->queue_lock, flags);
}
mutex_lock(&sched->sched_lock);
fence = sched->ops->run_job(sched, c_entity, job);
if (fence) {
r = fence_add_callback(fence, &job->cb,
amd_sched_process_job);
if (r == -ENOENT)
amd_sched_process_job(fence, &job->cb);
else if (r)
DRM_ERROR("fence add callback failed (%d)\n", r);
fence_put(fence);
}
mutex_unlock(&sched->sched_lock);
}
return 0;
}
/**
* Create a gpu scheduler
*
* @device The device context for this scheduler
* @ops The backend operations for this scheduler.
* @id The scheduler is per ring, here is ring id.
* @granularity The minumum ms unit the scheduler will scheduled.
* @preemption Indicate whether this ring support preemption, 0 is no.
*
* return the pointer to scheduler for success, otherwise return NULL
*/
struct amd_gpu_scheduler *amd_sched_create(void *device,
struct amd_sched_backend_ops *ops,
unsigned ring,
unsigned granularity,
unsigned preemption,
unsigned hw_submission)
{
struct amd_gpu_scheduler *sched;
char name[20];
sched = kzalloc(sizeof(struct amd_gpu_scheduler), GFP_KERNEL);
if (!sched)
return NULL;
sched->device = device;
sched->ops = ops;
sched->granularity = granularity;
sched->ring_id = ring;
sched->preemption = preemption;
sched->hw_submission_limit = hw_submission;
snprintf(name, sizeof(name), "gpu_sched[%d]", ring);
mutex_init(&sched->sched_lock);
spin_lock_init(&sched->queue_lock);
amd_sched_rq_init(&sched->sched_rq);
amd_sched_rq_init(&sched->kernel_rq);
init_waitqueue_head(&sched->wait_queue);
INIT_LIST_HEAD(&sched->active_hw_rq);
atomic64_set(&sched->hw_rq_count, 0);
/* Each scheduler will run on a seperate kernel thread */
sched->thread = kthread_create(amd_sched_main, sched, name);
if (sched->thread) {
wake_up_process(sched->thread);
return sched;
}
DRM_ERROR("Failed to create scheduler for id %d.\n", ring);
kfree(sched);
return NULL;
}
/**
* Destroy a gpu scheduler
*
* @sched The pointer to the scheduler
*
* return 0 if succeed. -1 if failed.
*/
int amd_sched_destroy(struct amd_gpu_scheduler *sched)
{
kthread_stop(sched->thread);
kfree(sched);
return 0;
}
/**
* Get next queued sequence number
*
* @entity The context entity
*
* return the next queued sequence number
*/
uint64_t amd_sched_next_queued_seq(struct amd_sched_entity *c_entity)
{
return atomic64_read(&c_entity->last_queued_v_seq) + 1;
}
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+162
View File
@@ -0,0 +1,162 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*/
#ifndef _GPU_SCHEDULER_H_
#define _GPU_SCHEDULER_H_
#include <linux/kfifo.h>
#include <linux/fence.h>
#define AMD_GPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
struct amd_gpu_scheduler;
struct amd_sched_rq;
/**
* A scheduler entity is a wrapper around a job queue or a group
* of other entities. Entities take turns emitting jobs from their
* job queues to corresponding hardware ring based on scheduling
* policy.
*/
struct amd_sched_entity {
struct list_head list;
struct amd_sched_rq *belongto_rq;
spinlock_t lock;
/* the virtual_seq is unique per context per ring */
atomic64_t last_queued_v_seq;
atomic64_t last_signaled_v_seq;
/* the job_queue maintains the jobs submitted by clients */
struct kfifo job_queue;
spinlock_t queue_lock;
struct amd_gpu_scheduler *scheduler;
wait_queue_head_t wait_queue;
wait_queue_head_t wait_emit;
bool is_pending;
uint64_t fence_context;
char name[20];
bool need_wakeup;
};
/**
* Run queue is a set of entities scheduling command submissions for
* one specific ring. It implements the scheduling policy that selects
* the next entity to emit commands from.
*/
struct amd_sched_rq {
struct mutex lock;
struct list_head entities;
struct amd_sched_entity *current_entity;
};
struct amd_sched_fence {
struct fence base;
struct fence_cb cb;
struct amd_sched_entity *entity;
uint64_t v_seq;
spinlock_t lock;
};
struct amd_sched_job {
struct list_head list;
struct fence_cb cb;
struct amd_gpu_scheduler *sched;
struct amd_sched_entity *s_entity;
void *data;
struct amd_sched_fence *s_fence;
};
extern const struct fence_ops amd_sched_fence_ops;
static inline struct amd_sched_fence *to_amd_sched_fence(struct fence *f)
{
struct amd_sched_fence *__f = container_of(f, struct amd_sched_fence, base);
if (__f->base.ops == &amd_sched_fence_ops)
return __f;
return NULL;
}
/**
* Define the backend operations called by the scheduler,
* these functions should be implemented in driver side
*/
struct amd_sched_backend_ops {
int (*prepare_job)(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *c_entity,
struct amd_sched_job *job);
struct fence *(*run_job)(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *c_entity,
struct amd_sched_job *job);
void (*process_job)(struct amd_gpu_scheduler *sched,
struct amd_sched_job *job);
};
/**
* One scheduler is implemented for each hardware ring
*/
struct amd_gpu_scheduler {
void *device;
struct task_struct *thread;
struct amd_sched_rq sched_rq;
struct amd_sched_rq kernel_rq;
struct list_head active_hw_rq;
atomic64_t hw_rq_count;
struct amd_sched_backend_ops *ops;
uint32_t ring_id;
uint32_t granularity; /* in ms unit */
uint32_t preemption;
wait_queue_head_t wait_queue;
struct amd_sched_entity *current_entity;
struct mutex sched_lock;
spinlock_t queue_lock;
uint32_t hw_submission_limit;
};
struct amd_gpu_scheduler *amd_sched_create(void *device,
struct amd_sched_backend_ops *ops,
uint32_t ring,
uint32_t granularity,
uint32_t preemption,
uint32_t hw_submission);
int amd_sched_destroy(struct amd_gpu_scheduler *sched);
int amd_sched_push_job(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *c_entity,
void *data,
struct amd_sched_fence **fence);
int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity,
struct amd_sched_rq *rq,
uint32_t jobs);
int amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity);
uint64_t amd_sched_next_queued_seq(struct amd_sched_entity *c_entity);
struct amd_sched_fence *amd_sched_fence_create(
struct amd_sched_entity *s_entity);
void amd_sched_fence_signal(struct amd_sched_fence *fence);
#endif
drivers/gpu/drm/amd/scheduler/sched_fence.c
+78
View File
@@ -0,0 +1,78 @@
/*
* Copyright 2015 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*
*
*/
#include <linux/kthread.h>
#include <linux/wait.h>
#include <linux/sched.h>
#include <drm/drmP.h>
#include "gpu_scheduler.h"
struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity)
{
struct amd_sched_fence *fence = NULL;
fence = kzalloc(sizeof(struct amd_sched_fence), GFP_KERNEL);
if (fence == NULL)
return NULL;
fence->v_seq = atomic64_inc_return(&s_entity->last_queued_v_seq);
fence->entity = s_entity;
spin_lock_init(&fence->lock);
fence_init(&fence->base, &amd_sched_fence_ops,
&fence->lock,
s_entity->fence_context,
fence->v_seq);
return fence;
}
void amd_sched_fence_signal(struct amd_sched_fence *fence)
{
int ret = fence_signal(&fence->base);
if (!ret)
FENCE_TRACE(&fence->base, "signaled from irq context\n");
else
FENCE_TRACE(&fence->base, "was already signaled\n");
}
static const char *amd_sched_fence_get_driver_name(struct fence *fence)
{
return "amd_sched";
}
static const char *amd_sched_fence_get_timeline_name(struct fence *f)
{
struct amd_sched_fence *fence = to_amd_sched_fence(f);
return (const char *)fence->entity->name;
}
static bool amd_sched_fence_enable_signaling(struct fence *f)
{
return true;
}
const struct fence_ops amd_sched_fence_ops = {
.get_driver_name = amd_sched_fence_get_driver_name,
.get_timeline_name = amd_sched_fence_get_timeline_name,
.enable_signaling = amd_sched_fence_enable_signaling,
.signaled = NULL,
.wait = fence_default_wait,
.release = NULL,
};
drivers/gpu/drm/radeon/dce6_afmt.c
+24 -5
View File
@@ -76,16 +76,35 @@ static void dce6_afmt_get_connected_pins(struct radeon_device *rdev)
struct r600_audio_pin *dce6_audio_get_pin(struct radeon_device *rdev)
{
int i;
struct drm_encoder *encoder;
struct radeon_encoder *radeon_encoder;
struct radeon_encoder_atom_dig *dig;
struct r600_audio_pin *pin = NULL;
int i, pin_count;
dce6_afmt_get_connected_pins(rdev);
for (i = 0; i < rdev->audio.num_pins; i++) {
if (rdev->audio.pin[i].connected)
return &rdev->audio.pin[i];
if (rdev->audio.pin[i].connected) {
pin = &rdev->audio.pin[i];
pin_count = 0;
list_for_each_entry(encoder, &rdev->ddev->mode_config.encoder_list, head) {
if (radeon_encoder_is_digital(encoder)) {
radeon_encoder = to_radeon_encoder(encoder);
dig = radeon_encoder->enc_priv;
if (dig->pin == pin)
pin_count++;
}
}
if (pin_count == 0)
return pin;
}
}
DRM_ERROR("No connected audio pins found!\n");
return NULL;
if (!pin)
DRM_ERROR("No connected audio pins found!\n");
return pin;
}
void dce6_afmt_select_pin(struct drm_encoder *encoder)
drivers/gpu/drm/radeon/radeon_object.c
+1 -3
View File
@@ -419,7 +419,6 @@ void radeon_bo_force_delete(struct radeon_device *rdev)
}
dev_err(rdev->dev, "Userspace still has active objects !\n");
list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
mutex_lock(&rdev->ddev->struct_mutex);
dev_err(rdev->dev, "%p %p %lu %lu force free\n",
&bo->gem_base, bo, (unsigned long)bo->gem_base.size,
*((unsigned long *)&bo->gem_base.refcount));
@@ -427,8 +426,7 @@ void radeon_bo_force_delete(struct radeon_device *rdev)
list_del_init(&bo->list);
mutex_unlock(&bo->rdev->gem.mutex);
/* this should unref the ttm bo */
drm_gem_object_unreference(&bo->gem_base);
mutex_unlock(&rdev->ddev->struct_mutex);
drm_gem_object_unreference_unlocked(&bo->gem_base);
}
}
drivers/gpu/drm/radeon/radeon_pm.c
-5
View File
@@ -253,7 +253,6 @@ static void radeon_pm_set_clocks(struct radeon_device *rdev)
(rdev->pm.requested_power_state_index == rdev->pm.current_power_state_index))
return;
mutex_lock(&rdev->ddev->struct_mutex);
down_write(&rdev->pm.mclk_lock);
mutex_lock(&rdev->ring_lock);
@@ -268,7 +267,6 @@ static void radeon_pm_set_clocks(struct radeon_device *rdev)
/* needs a GPU reset dont reset here */
mutex_unlock(&rdev->ring_lock);
up_write(&rdev->pm.mclk_lock);
mutex_unlock(&rdev->ddev->struct_mutex);
return;
}
}
@@ -304,7 +302,6 @@ static void radeon_pm_set_clocks(struct radeon_device *rdev)
mutex_unlock(&rdev->ring_lock);
up_write(&rdev->pm.mclk_lock);
mutex_unlock(&rdev->ddev->struct_mutex);
}
static void radeon_pm_print_states(struct radeon_device *rdev)
@@ -1062,7 +1059,6 @@ force:
radeon_dpm_print_power_state(rdev, rdev->pm.dpm.requested_ps);
}
mutex_lock(&rdev->ddev->struct_mutex);
down_write(&rdev->pm.mclk_lock);
mutex_lock(&rdev->ring_lock);
@@ -1113,7 +1109,6 @@ force:
done:
mutex_unlock(&rdev->ring_lock);
up_write(&rdev->pm.mclk_lock);
mutex_unlock(&rdev->ddev->struct_mutex);
}
void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable)