[Public]
Hi,
> -----Original Message-----
> From: Jan Beulich <jbeul...@suse.com>
> Sent: Tuesday, April 1, 2025 9:10 PM
> To: Penny, Zheng <penny.zh...@amd.com>
> Cc: Huang, Ray <ray.hu...@amd.com>; Andrew Cooper
> <andrew.coop...@citrix.com>; Roger Pau Monné <roger....@citrix.com>;
> Anthony PERARD <anthony.per...@vates.tech>; Orzel, Michal
> <michal.or...@amd.com>; Julien Grall <jul...@xen.org>; Stefano Stabellini
> <sstabell...@kernel.org>; xen-devel@lists.xenproject.org
> Subject: Re: [PATCH v2 10/19] xen/sysctl: introduce CONFIG_PM_STATS
>
> On 26.03.2025 06:50, Penny Zheng wrote:
> > We intend to introduce CONFIG_PM_STATS for wrapping all operations
> > regarding performance management statistics.
> > The major codes reside in xen/drivers/acpi/pmstat.c, including two
> > main pm-related sysctl op: do_get_pm_info() and do_pm_op().
> > So This commit also makes CONFIG_PM_STATS depend on
> CONFIG_SYSCTL
> >
> > Signed-off-by: Penny Zheng <penny.zh...@amd.com>
> > ---
> > v1 -> v2:
> > - rename to CONFIG_PM_STATS
> > - fix indention and stray semicolon
> > - make code movements into a new commit
> > - No need to wrap inline functions and declarations
> > ---
> > xen/arch/x86/acpi/cpu_idle.c | 2 ++
> > xen/arch/x86/acpi/cpufreq/hwp.c | 6 ++++++
> > xen/arch/x86/acpi/cpufreq/powernow.c | 4 ++++
> > xen/common/Kconfig | 5 +++++
> > xen/common/sysctl.c | 4 ++--
> > xen/drivers/acpi/Makefile | 2 +-
> > xen/drivers/cpufreq/cpufreq_misc_governors.c | 2 ++
> > xen/drivers/cpufreq/cpufreq_ondemand.c | 2 ++
> > xen/include/acpi/cpufreq/processor_perf.h | 14 ++++++++++++++
> > 9 files changed, 38 insertions(+), 3 deletions(-)
> >
> > diff --git a/xen/arch/x86/acpi/cpu_idle.c
> > b/xen/arch/x86/acpi/cpu_idle.c index 420198406d..b537ac4cd6 100644
> > --- a/xen/arch/x86/acpi/cpu_idle.c
> > +++ b/xen/arch/x86/acpi/cpu_idle.c
> > @@ -1487,6 +1487,7 @@ static void amd_cpuidle_init(struct
> acpi_processor_power *power)
> > vendor_override = -1;
> > }
> >
> > +#ifdef CONFIG_PM_STATS
> > uint32_t pmstat_get_cx_nr(unsigned int cpu) {
> > return processor_powers[cpu] ? processor_powers[cpu]->count : 0;
> > @@ -1606,6 +1607,7 @@ int pmstat_reset_cx_stat(unsigned int cpu) {
> > return 0;
> > }
> > +#endif /* CONFIG_PM_STATS */
> >
> > void cpuidle_disable_deep_cstate(void) { diff --git
> > a/xen/arch/x86/acpi/cpufreq/hwp.c b/xen/arch/x86/acpi/cpufreq/hwp.c
> > index d5fa3d47ca..98e9d46890 100644
> > --- a/xen/arch/x86/acpi/cpufreq/hwp.c
> > +++ b/xen/arch/x86/acpi/cpufreq/hwp.c
> > @@ -466,6 +466,7 @@ static int cf_check hwp_cpufreq_cpu_exit(struct
> cpufreq_policy *policy)
> > return 0;
> > }
> >
> > +#ifdef CONFIG_PM_STATS
> > /*
> > * The SDM reads like turbo should be disabled with MSR_IA32_PERF_CTL
> and
> > * PERF_CTL_TURBO_DISENGAGE, but that does not seem to actually work,
> > at least @@ -508,6 +509,7 @@ static int cf_check
> > hwp_cpufreq_update(unsigned int cpu, struct cpufreq_policy *
> >
> > return per_cpu(hwp_drv_data, cpu)->ret; }
> > +#endif /* CONFIG_PM_STATS */
> >
> > static const struct cpufreq_driver __initconst_cf_clobber
> > hwp_cpufreq_driver = { @@ -516,9 +518,12 @@ hwp_cpufreq_driver = {
> > .target = hwp_cpufreq_target,
> > .init = hwp_cpufreq_cpu_init,
> > .exit = hwp_cpufreq_cpu_exit,
> > +#ifdef CONFIG_PM_STATS
> > .update = hwp_cpufreq_update,
> > +#endif
> > };
>
> Something's wrong here: The .update hook is actually making changes, so is
> definitely not (only) about statistics. Same for the powernow driver.
>
> > +#ifdef CONFIG_PM_STATS
> > int get_hwp_para(unsigned int cpu,
> > struct xen_cppc_para *cppc_para) { @@ -639,6 +644,7
> > @@ int set_hwp_para(struct cpufreq_policy *policy,
> >
> > return hwp_cpufreq_target(policy, 0, 0); }
> > +#endif /* CONFIG_PM_STATS */
>
> This also isn't about statistics, but about getting / setting parameters.
>
> > --- a/xen/common/Kconfig
> > +++ b/xen/common/Kconfig
> > @@ -557,4 +557,9 @@ config SYSCTL
> > to reduce Xen footprint.
> > endmenu
> >
> > +config PM_STATS
> > + bool "Enable Performance Management Statistics"
> > + depends on ACPI && HAS_CPUFREQ && SYSCTL
> > + default y
>
> As per above - either name, prompt and the description that Stefano suggested
> are wrong, or it is too much that is being covered by this new control.
>
We have two sysctl-op on performance, do_get_pm_info() and do_pm_op().
I think do_get_pm_info() is to collect PM statistic info, which could be
wrapped with
CONFIG_PM_STATS, while maybe do_pm_op() is more focusing on performance tuning.
How about we introduce another Kconfig CONFIG_PM_TUNE to wrap do_pm_op()
and related helpers? I suggest to introduce a new file pmtune.c to contain.
Or any better suggestion?
>
> Jan
