[adding Ionela]
On Wed, Oct 25, 2023 at 05:38:46PM +0800, Zeng Heng wrote:
As ARM AMU's document says, all counters are subject to any changes
in clock frequency, including clock stopping caused by the WFI and WFE
instructions.
Therefore, using smp_call_on_cpu() to trigger target CPU to
read self's AMU counters, which ensures the counters are working
properly while cstate feature is enabled.
IIUC there's a pretty deliberate split with all the actual reading of the AMU
living in arch/arm64/kernel/topolgy.c, and the driver code being (relatively)
generic.
We already have code in arch/arm64/kernel/topolgy.c to read counters on a
specific CPU; why can't e reuse that (and avoid exporting cpu_has_amu_feat())?
This patch seems mostly based on my previous patch [1] and discussed
here [2] already. Beata [CCed] shared an alternate approach [3]
leveraging existing code from 'topology.c' to get the average freq for
last tick period.
Beata,
Could you share v2 of [3] with the request to merge. We can try to solve
the issue with CPU IDLE case later on top?
Additionally, also please include the fix in [4] if it looks fine.
Best Regards,
Sumit Gupta
[1] https://lore.kernel.org/all/20230418113459.12860-7-sumitg@xxxxxxxxxx/
[2]
https://lore.kernel.org/lkml/cde1d8a9-3a21-e82b-7895-40603a14d898@xxxxxxxxxx/T/#m2174305de4706006e0bd9c103a0e5ff61cea7a12
[3]
https://lore.kernel.org/lkml/20230606155754.245998-1-beata.michalska@xxxxxxx/
[4]
https://lore.kernel.org/lkml/6a5710f6-bfbb-5dfd-11cd-0cd02220cee7@xxxxxxxxxx/
Reported-by: Sumit Gupta <sumitg@xxxxxxxxxx>
Link: https://lore.kernel.org/all/20230418113459.12860-7-sumitg@xxxxxxxxxx/
Signed-off-by: Zeng Heng <zengheng4@xxxxxxxxxx>
---
drivers/cpufreq/cppc_cpufreq.c | 39 ++++++++++++++++++++++++++--------
1 file changed, 30 insertions(+), 9 deletions(-)
diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
index fe08ca419b3d..321a9dc9484d 100644
--- a/drivers/cpufreq/cppc_cpufreq.c
+++ b/drivers/cpufreq/cppc_cpufreq.c
@@ -90,6 +90,12 @@ static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
struct cppc_perf_fb_ctrs *fb_ctrs_t0,
struct cppc_perf_fb_ctrs *fb_ctrs_t1);
+struct fb_ctr_pair {
+ u32 cpu;
+ struct cppc_perf_fb_ctrs fb_ctrs_t0;
+ struct cppc_perf_fb_ctrs fb_ctrs_t1;
+};
+
/**
* cppc_scale_freq_workfn - CPPC arch_freq_scale updater for frequency invariance
* @work: The work item.
@@ -840,9 +846,24 @@ static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
return (reference_perf * delta_delivered) / delta_reference;
}
+static int cppc_get_perf_ctrs_pair(void *val)
+{
+ struct fb_ctr_pair *fb_ctrs = val;
+ int cpu = fb_ctrs->cpu;
+ int ret;
+
+ ret = cppc_get_perf_ctrs(cpu, &fb_ctrs->fb_ctrs_t0);
+ if (ret)
+ return ret;
+
+ udelay(2); /* 2usec delay between sampling */
+
+ return cppc_get_perf_ctrs(cpu, &fb_ctrs->fb_ctrs_t1);
+}
+
static unsigned int cppc_cpufreq_get_rate(unsigned int cpu)
{
- struct cppc_perf_fb_ctrs fb_ctrs_t0 = {0}, fb_ctrs_t1 = {0};
+ struct fb_ctr_pair fb_ctrs = { .cpu = cpu, };
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
struct cppc_cpudata *cpu_data = policy->driver_data;
u64 delivered_perf;
@@ -850,18 +871,18 @@ static unsigned int cppc_cpufreq_get_rate(unsigned int cpu)
cpufreq_cpu_put(policy);
- ret = cppc_get_perf_ctrs(cpu, &fb_ctrs_t0);
- if (ret)
- return 0;
-
- udelay(2); /* 2usec delay between sampling */
+ if (cpu_has_amu_feat(cpu))
+ ret = smp_call_on_cpu(cpu, cppc_get_perf_ctrs_pair,
+ &fb_ctrs, false);
+ else
+ ret = cppc_get_perf_ctrs_pair(&fb_ctrs);
- ret = cppc_get_perf_ctrs(cpu, &fb_ctrs_t1);
if (ret)
return 0;
- delivered_perf = cppc_perf_from_fbctrs(cpu_data, &fb_ctrs_t0,
- &fb_ctrs_t1);
+ delivered_perf = cppc_perf_from_fbctrs(cpu_data,
+ &fb_ctrs.fb_ctrs_t0,
+ &fb_ctrs.fb_ctrs_t1);
return cppc_cpufreq_perf_to_khz(cpu_data, delivered_perf);
}
--
2.25.1
