| Message ID | OS3P286MB249076187B3497D1EDD70988B1F32@OS3P286MB2490.JPNP286.PROD.OUTLOOK.COM (mailing list archive) |
|---|---|
| State | New |
| Delegated to: | viresh kumar |
| Headers | show |
| Series | [v3] cpufreq/cppc: Remove the desired_perf compare when set target | expand |
Ionela, Beata, On 30-05-24, 19:08, Riwen Lu wrote: > From: Riwen Lu <luriwen@kylinos.cn> > > There is a case that desired_perf is exactly the same with the old perf, > but the actual current freq is not. > > This happened in S3 while the cpufreq governor is set to powersave. > During cpufreq resume process, the booting CPU's new_freq obtained via > .get() is the highest frequency, while the policy->cur and > cpu->perf_ctrls.desired_perf are in the lowest level(powersave > governor). Causing the warning: "CPU frequency out of sync:", and set > policy->cur to new_freq. Then the governor->limits() calls > cppc_cpufreq_set_target() to configures the CPU frequency and returns > directly because the desired_perf converted from target_freq is the > same with cpu->perf_ctrls.desired_perf and both are the lowest_perf. > Since target_freq and policy->cur have been compared in > __cpufreq_driver_target(), there's no need to compare desired_perf > and cpu->perf_ctrls.desired_perf again in cppc_cpufreq_set_target() > to ensure that the CPU frequency is properly configured. > > Signed-off-by: Riwen Lu <luriwen@kylinos.cn> > > --- > v1 -> v2: > - Update commit message and email. > v2 -> v3: > - Update patch subject and commit message. > - Remove the desired_perf compare logic. > --- > drivers/cpufreq/cppc_cpufreq.c | 3 --- > 1 file changed, 3 deletions(-) > > diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c > index 15f1d41920a3..337cece61ab5 100644 > --- a/drivers/cpufreq/cppc_cpufreq.c > +++ b/drivers/cpufreq/cppc_cpufreq.c > @@ -295,9 +295,6 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, > int ret = 0; > > desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); > - /* Return if it is exactly the same perf */ > - if (desired_perf == cpu_data->perf_ctrls.desired_perf) > - return ret; > > cpu_data->perf_ctrls.desired_perf = desired_perf; > freqs.old = policy->cur; Any objections to this change ?
Hey, On Thursday 06 Jun 2024 at 14:37:37 (+0530), Viresh Kumar wrote: > Ionela, Beata, > > On 30-05-24, 19:08, Riwen Lu wrote: > > From: Riwen Lu <luriwen@kylinos.cn> > > > > There is a case that desired_perf is exactly the same with the old perf, > > but the actual current freq is not. > > > > This happened in S3 while the cpufreq governor is set to powersave. > > During cpufreq resume process, the booting CPU's new_freq obtained via > > .get() is the highest frequency, while the policy->cur and > > cpu->perf_ctrls.desired_perf are in the lowest level(powersave > > governor). Causing the warning: "CPU frequency out of sync:", and set > > policy->cur to new_freq. Then the governor->limits() calls > > cppc_cpufreq_set_target() to configures the CPU frequency and returns > > directly because the desired_perf converted from target_freq is the > > same with cpu->perf_ctrls.desired_perf and both are the lowest_perf. > > Since target_freq and policy->cur have been compared in ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ [note] below > > __cpufreq_driver_target(), there's no need to compare desired_perf > > and cpu->perf_ctrls.desired_perf again in cppc_cpufreq_set_target() > > to ensure that the CPU frequency is properly configured. > > > > Signed-off-by: Riwen Lu <luriwen@kylinos.cn> > > > > --- > > v1 -> v2: > > - Update commit message and email. > > v2 -> v3: > > - Update patch subject and commit message. > > - Remove the desired_perf compare logic. > > --- > > drivers/cpufreq/cppc_cpufreq.c | 3 --- > > 1 file changed, 3 deletions(-) > > > > diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c > > index 15f1d41920a3..337cece61ab5 100644 > > --- a/drivers/cpufreq/cppc_cpufreq.c > > +++ b/drivers/cpufreq/cppc_cpufreq.c > > @@ -295,9 +295,6 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, > > int ret = 0; > > > > desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); > > - /* Return if it is exactly the same perf */ > > - if (desired_perf == cpu_data->perf_ctrls.desired_perf) > > - return ret; > > > > cpu_data->perf_ctrls.desired_perf = desired_perf; > > freqs.old = policy->cur; > > Any objections to this change ? It's alright with me. Some "nits": - the "desired_perf" local variable could be removed in this case. - [note] while this change helps, we'd still need policy->cur to always have the latest request value (see details at [1]) for this check to be made obsolete by the comparison between target_freq and policy->cur, as mentioned in the commit message. But this is/can be a separate matter. [1] https://lore.kernel.org/lkml/ZmB1qKucR5fXk100@arm.com/ Thanks, Ionela. > > -- > viresh
On 11-06-24, 09:54, Ionela Voinescu wrote: > It's alright with me. Great. > Some "nits": > - the "desired_perf" local variable could be removed in this case. Riwen, please fix this and resend. > - [note] while this change helps, we'd still need policy->cur to always > have the latest request value (see details at [1]) for this check to > be made obsolete by the comparison between target_freq and policy->cur, > as mentioned in the commit message. But this is/can be a separate > matter. > > [1] https://lore.kernel.org/lkml/ZmB1qKucR5fXk100@arm.com/ Yeah, lets discuss that in the other thread only.
在 2024/6/11 17:10, Viresh Kumar 写道: > On 11-06-24, 09:54, Ionela Voinescu wrote: >> It's alright with me. > > Great. > >> Some "nits": >> - the "desired_perf" local variable could be removed in this case. > > Riwen, please fix this and resend. I think it's no need to remove the "desired_perf" local variable here, the code is seems more readable. > >> - [note] while this change helps, we'd still need policy->cur to always >> have the latest request value (see details at [1]) for this check to >> be made obsolete by the comparison between target_freq and policy->cur, >> as mentioned in the commit message. But this is/can be a separate >> matter. >> >> [1] https://lore.kernel.org/lkml/ZmB1qKucR5fXk100@arm.com/ > > Yeah, lets discuss that in the other thread only. >
On 12-06-24, 10:52, Riwen Lu wrote: > I think it's no need to remove the "desired_perf" local variable here, the > code is seems more readable. There is no need of a temporary variable and I don't think it increases readability at all. Moreover, same needs to be done in cppc_cpufreq_fast_switch() too.
On 12-06-24, 11:54, Viresh Kumar wrote: > On 12-06-24, 10:52, Riwen Lu wrote: > > I think it's no need to remove the "desired_perf" local variable here, the > > code is seems more readable. > > There is no need of a temporary variable and I don't think it increases > readability at all. Moreover, same needs to be done in > cppc_cpufreq_fast_switch() too. Also please don't send the new patch in-reply-to the earlier one. Send a fresh thread. Thanks.
Hello Riwen, This function seems to be the only cpufreq function saving and comparing the requested frequency with the last requested frequency. This seems to be more the task of the cpufreq framework than the cpufreq driver. So FYIW, the patch looks good to me. On 5/30/24 13:08, Riwen Lu wrote: > From: Riwen Lu <luriwen@kylinos.cn> > > There is a case that desired_perf is exactly the same with the old perf, > but the actual current freq is not. > > This happened in S3 while the cpufreq governor is set to powersave. > During cpufreq resume process, the booting CPU's new_freq obtained via > .get() is the highest frequency, while the policy->cur and > cpu->perf_ctrls.desired_perf are in the lowest level(powersave > governor). Causing the warning: "CPU frequency out of sync:", and set > policy->cur to new_freq. (new paragraph) Then the governor->limits() calls > cppc_cpufreq_set_target() to configures the CPU frequency and returns > directly because the desired_perf converted from target_freq is the > same with cpu->perf_ctrls.desired_perf and both are the lowest_perf. (new paragraph) > Since target_freq and policy->cur have been compared in > __cpufreq_driver_target(), there's no need to compare desired_perf > and cpu->perf_ctrls.desired_perf again in cppc_cpufreq_set_target() > to ensure that the CPU frequency is properly configured. NIT: Would it be possible to make distinct paragraphs ? > > Signed-off-by: Riwen Lu <luriwen@kylinos.cn> > > --- > v1 -> v2: > - Update commit message and email. > v2 -> v3: > - Update patch subject and commit message. > - Remove the desired_perf compare logic. > --- > drivers/cpufreq/cppc_cpufreq.c | 3 --- > 1 file changed, 3 deletions(-) > > diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c > index 15f1d41920a3..337cece61ab5 100644 > --- a/drivers/cpufreq/cppc_cpufreq.c > +++ b/drivers/cpufreq/cppc_cpufreq.c > @@ -295,9 +295,6 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, > int ret = 0; > > desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); > - /* Return if it is exactly the same perf */ > - if (desired_perf == cpu_data->perf_ctrls.desired_perf) > - return ret; > > cpu_data->perf_ctrls.desired_perf = desired_perf; > freqs.old = policy->cur;
在 2024/6/12 14:24, Viresh Kumar 写道: > On 12-06-24, 10:52, Riwen Lu wrote: >> I think it's no need to remove the "desired_perf" local variable here, the >> code is seems more readable. > > There is no need of a temporary variable and I don't think it increases > readability at all. Moreover, same needs to be done in > cppc_cpufreq_fast_switch() too. > OK. I'll push a patch that remove the desired_perf temporary in cppc_cpufreq_fast_switch().
diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c index 15f1d41920a3..337cece61ab5 100644 --- a/drivers/cpufreq/cppc_cpufreq.c +++ b/drivers/cpufreq/cppc_cpufreq.c @@ -295,9 +295,6 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, int ret = 0; desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); - /* Return if it is exactly the same perf */ - if (desired_perf == cpu_data->perf_ctrls.desired_perf) - return ret; cpu_data->perf_ctrls.desired_perf = desired_perf; freqs.old = policy->cur;