Re: [PATCH v2] cpufreq: intel_pstate: set stale CPU frequency to minimum

From: Keyon Jie
Date: Mon Aug 28 2023 - 19:38:14 EST




On 8/22/23 04:46, Rafael J. Wysocki wrote:
On Sun, Aug 20, 2023 at 10:46 PM Doug Smythies <dsmythies@xxxxxxxxx> wrote:

The intel_pstate CPU frequency scaling driver does not
use policy->cur and it is 0.
When the CPU frequency is outdated arch_freq_get_on_cpu()
will default to the nominal clock frequency when its call to
cpufreq_quick_getpolicy_cur returns the never updated 0.
Thus, the listed frequency might be outside of currently
set limits. Some users are complaining about the high
reported frequency, albeit stale, when their system is
idle and/or it is above the reduced maximum they have set.

This patch will maintain policy_cur for the intel_pstate
driver at the current minimum CPU frequency.

Reported-by: Yang Jie <yang.jie@xxxxxxxxxxxxxxx>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217597
Signed-off-by: Doug Smythies <dsmythies@xxxxxxxxx>
---

v1 -> v2:
* v1 was a completely different approach, programming around
the issue rather than fixing it at the source.
reference:
https://patchwork.kernel.org/project/linux-pm/patch/006901d9be8c$f4439930$dccacb90$@telus.net/
* v2 does not fix an issue with the intel_cpufreq CPU scaling
driver (A.K.A. the intel_pstate driver in passive mode) and
the schedutil CPU frequency scaling governor when HWP is enabled
where limit changes are not reflected in the stale listed frequencies.
A fix for that will be some future patch.

---
drivers/cpufreq/intel_pstate.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index 8ca2bce4341a..08284dee583a 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -2609,6 +2609,11 @@ static int intel_pstate_set_policy(struct cpufreq_policy *policy)
intel_pstate_clear_update_util_hook(policy->cpu);
intel_pstate_hwp_set(policy->cpu);
}
+ /* policy current is never updated with the intel_pstate driver
+ * but it is used as a stale frequency value. So, keep it within
+ * limits.
+ */
+ policy->cur = policy->min;

mutex_unlock(&intel_pstate_limits_lock);

--

Applied as 6.6 material, with some mailer-induced white space damage
fixed and the new comment adjusted to the kernel coding style.

Thanks!

Hi Doug and Rafael,

Thank you for making the fix happen.

Hi Greg,

Will this be picked to the stable linux-6.1.y and linux-6.4.y kernel, it could benefit to users there.

Thanks,
~Keyon