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

NVIDIA: SAUCE: cpufreq: tegra194: fix policy initialization

Browse Source 
Fix the policy-initialization scheme to correctly be dependent on cores
sharing a clusterid or a clock-source via connected cluster-ids.
This solves an issue where we floorsweep only one core in a cluster,
causing logical IDs to fall out of order with physical IDs and so
causing cores to be tied together incorrectly in the cpufreq framework.

http://nvbugs/5213905

Signed-off-by: Ishan Shah <ishah@nvidia.com>
Reviewed-by: Sumit Gupta <sumitg@nvidia.com>
Reviewed-by: Nathan Hartman <nhartman@nvidia.com>
Signed-off-by: Kartik Rajput <kkartik@nvidia.com>
Acked-by: Noah Wager <noah.wager@canonical.com>
Acked-by: Jacob Martin <jacob.martin@canonical.com>
Signed-off-by: Noah Wager <noah.wager@canonical.com>
This commit is contained in:
Ishan Shah
2025-04-11 22:23:31 +00:00
committed by Noah Wager
parent 3bd1742fc8
commit 3abe156a4e
1 changed files with 12 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
drivers/cpufreq/tegra194-cpufreq.c
+12 -5
View File
@@ -620,17 +620,24 @@ static int tegra194_cpufreq_init(struct cpufreq_policy *policy)
u32 clusterid = data->cpu_data[policy->cpu].clusterid;
struct cpufreq_frequency_table *freq_table;
struct cpufreq_frequency_table *bpmp_lut;
u32 start_cpu, cpu;
u32 cpu;
int ret;
if (clusterid >= data->soc->num_clusters || !data->bpmp_luts[clusterid])
return -EINVAL;
start_cpu = rounddown(policy->cpu, maxcpus_per_clock);
/* set same policy for all cpus in a cluster */
for (cpu = start_cpu; cpu < (start_cpu + maxcpus_per_clock); cpu++) {
if (cpu_possible(cpu))
/* Set same policy for all possible cpus in a clocking cluster */
for_each_possible_cpu(cpu) {
if (data->cpu_data[cpu].clusterid == clusterid) {
cpumask_set_cpu(cpu, policy->cpus);
} else if (data->soc->clusters_per_clk > 1) {
u32 policy_clk_id = clusterid / data->soc->clusters_per_clk;
u32 clk_id = data->cpu_data[cpu].clusterid /
data->soc->clusters_per_clk;
if (policy_clk_id == clk_id) {
cpumask_set_cpu(cpu, policy->cpus);
}
}
}
policy->cpuinfo.transition_latency = TEGRA_CPUFREQ_TRANSITION_LATENCY;