AMD Pstate

AMD Pstate

[Mike Lothian]

Hi

I'm not subscribed to the mailing list so apologies for emailing you directly

It looks like something isn't quite right with the kconfig selects

SCHED_MC_PRIO can't be enabled without CPU_SUP_INTEL



│ Symbol: X86_AMD_PSTATE [=y]

                                                              │
 │ Type  : bool

                                                               │
 │ Defined at drivers/cpufreq/Kconfig.x86:37

                                                               │
 │   Prompt: AMD Processor P-State driver

                                                               │
 │   Depends on: CPU_FREQ [=y] && X86 [=y]

                                                               │
 │   Location:

                                                               │
 │     Main menu

                                                               │
 │       -> Power management and ACPI options

                                                               │
 │         -> CPU Frequency scaling

                                                               │
 │           -> CPU Frequency scaling (CPU_FREQ [=y])

                                                               │
 │ Selects: ACPI_PROCESSOR [=y] && ACPI_CPPC_LIB [=n] &&
CPU_FREQ_GOV_SCHEDUTIL [=y]



│ Symbol: ACPI_CPPC_LIB [=n]

                                                              │
 │ Type  : bool

                                                               │
 │ Defined at drivers/acpi/Kconfig:268

                                                               │
 │   Depends on: ACPI [=y] && ACPI_PROCESSOR [=y]

                                                               │
 │ Selects: MAILBOX [=n] && PCC [=n]

                                                               │
 │ Selected by [n]:

                                                               │
 │   - X86_INTEL_PSTATE [=n] && CPU_FREQ [=y] && X86 [=y] && X86_64
[=y] && ACPI [=y] && SCHED_MC_PRIO [=n]
                                                                  │
 │   - X86_AMD_PSTATE [=y] && CPU_FREQ [=y] && X86 [=y] && X86_64 [=y]
&& ACPI [=y] && SCHED_MC_PRIO [=n]
                                                               │
 │   - ACPI_CPPC_CPUFREQ [=n] && CPU_FREQ [=y] && (ARM || ARM64) &&
ACPI_PROCESSOR [=y]



│ Symbol: SCHED_MC_PRIO [=n]

                                                              │
 │ Type  : bool

                                                               │
 │ Defined at arch/x86/Kconfig:1029

                                                               │
 │   Prompt: CPU core priorities scheduler support

                                                               │
 │   Depends on: SCHED_MC [=y] && CPU_SUP_INTEL [=n]

                                                               │
 │   Location:

                                                               │
 │     Main menu

                                                               │
 │       -> Processor type and features

                                                               │
 │ (1)     -> Multi-core scheduler support (SCHED_MC [=y])

                                                               │
 │ Selects: X86_INTEL_PSTATE [=n] && CPU_FREQ [=y]



I've used this patch to work around the issue:

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 95dd1ee01546..bdfcf155f48e 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1028,8 +1028,8 @@ config SCHED_MC

config SCHED_MC_PRIO
       bool "CPU core priorities scheduler support"
-       depends on SCHED_MC && CPU_SUP_INTEL
-       select X86_INTEL_PSTATE
+       depends on SCHED_MC && CPU_SUP_AMD
+       select X86_AMD_PSTATE
       select CPU_FREQ
       default y
       help

Which gets things working here

Re: AMD Pstate

[Huang Rui]

On Sun, Nov 21, 2021 at 06:37:56PM +0800, Mike Lothian wrote:
> Hi
> 
> I'm not subscribed to the mailing list so apologies for emailing you directly
> 
> It looks like something isn't quite right with the kconfig selects
> 
> SCHED_MC_PRIO can't be enabled without CPU_SUP_INTEL
> 

Yes, right. I missed the update in this version, sorry. This issue was
actually reported from Stefano Picascia before in below history:

https://lore.kernel.org/linux-kernel/YYqDhaXDIqJyjGjg@hr-amd/

I will correct it in V5. Thank you again to raise this problem!

Thanks,
Ray

Re: AMD Pstate

[Huang Rui]

On Sun, Nov 21, 2021 at 07:52:30PM +0800, Huang Rui wrote:
> On Sun, Nov 21, 2021 at 06:37:56PM +0800, Mike Lothian wrote:
> > Hi
> > 
> > I'm not subscribed to the mailing list so apologies for emailing you directly
> > 
> > It looks like something isn't quite right with the kconfig selects
> > 
> > SCHED_MC_PRIO can't be enabled without CPU_SUP_INTEL
> > 
> 
> Yes, right. I missed the update in this version, sorry. This issue was
> actually reported from Stefano Picascia before in below history:
> 
> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Flinux-kernel%2FYYqDhaXDIqJyjGjg%40hr-amd%2F&data=04%7C01%7Cray.huang%40amd.com%7Cd921714980b241b6fbb208d9ace5794c%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637730923851487352%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=3wqjeuesO0bZ79G%2BOpQmyPlQ3sDvAWBV2RGydgWjfTY%3D&reserved=0
> 
> I will correct it in V5. Thank you again to raise this problem!
> 

Just read back patches again. I had addressed the comment in V4:

https://lore.kernel.org/linux-pm/20211119103102.88124-7-ray.huang@amd.com/

Thanks,
Ray

Re: AMD Pstate

[Mike Lothian]

Hi again

Do I need to enable something else for shedutil to work, I've tried
with and without shared_mem? Compiling a kernel with -j16 has the freq
stuck at 1GHz on all cores - ondemand works fine

That's on a AMD Ryzen 9 5900HX

Also compiling with Clang-13 gives the following warnings:

drivers/cpufreq/amd-pstate.c:257:6: warning: variable 'amd_des_perf'
is used uninitialized whenever 'if' condition is false
[-Wsometimes-uninitialized]

        if (target_perf < capacity)

            ^~~~~~~~~~~~~~~~~~~~~~

drivers/cpufreq/amd-pstate.c:272:40: note: uninitialized use occurs here

        amd_des_perf = clamp_t(unsigned long, amd_des_perf,

                                              ^~~~~~~~~~~~

./include/linux/minmax.h:124:60: note: expanded from macro 'clamp_t'

#define clamp_t(type, val, lo, hi) min_t(type, max_t(type, val, lo), hi)

                                                           ^~~

./include/linux/minmax.h:112:48: note: expanded from macro 'max_t'

#define max_t(type, x, y)       __careful_cmp((type)(x), (type)(y), >)

                                                     ^

./include/linux/minmax.h:38:14: note: expanded from macro '__careful_cmp'

                __cmp_once(x, y, __UNIQUE_ID(__x), __UNIQUE_ID(__y), op))

                           ^

note: (skipping 1 expansions in backtrace; use
-fmacro-backtrace-limit=0 to see all)

./include/linux/minmax.h:104:48: note: expanded from macro 'min_t'

#define min_t(type, x, y)       __careful_cmp((type)(x), (type)(y), <)

                                                     ^

./include/linux/minmax.h:38:14: note: expanded from macro '__careful_cmp'

                __cmp_once(x, y, __UNIQUE_ID(__x), __UNIQUE_ID(__y), op))

                           ^

./include/linux/minmax.h:31:25: note: expanded from macro '__cmp_once'

                typeof(x) unique_x = (x);               \

                                      ^

drivers/cpufreq/amd-pstate.c:257:2: note: remove the 'if' if its
condition is always true

        if (target_perf < capacity)

        ^~~~~~~~~~~~~~~~~~~~~~~~~~~

drivers/cpufreq/amd-pstate.c:249:56: note: initialize the variable
'amd_des_perf' to silence this warning

        unsigned long amd_max_perf, amd_min_perf, amd_des_perf,

                                                              ^

                                                               = 0

On Sun, 21 Nov 2021 at 10:37, Mike Lothian <mike@fireburn.co.uk> wrote:
>
> Hi
>
> I'm not subscribed to the mailing list so apologies for emailing you directly
>
> It looks like something isn't quite right with the kconfig selects
>
> SCHED_MC_PRIO can't be enabled without CPU_SUP_INTEL
>
>
>
> │ Symbol: X86_AMD_PSTATE [=y]
>
>                                                               │
>  │ Type  : bool
>
>                                                                │
>  │ Defined at drivers/cpufreq/Kconfig.x86:37
>
>                                                                │
>  │   Prompt: AMD Processor P-State driver
>
>                                                                │
>  │   Depends on: CPU_FREQ [=y] && X86 [=y]
>
>                                                                │
>  │   Location:
>
>                                                                │
>  │     Main menu
>
>                                                                │
>  │       -> Power management and ACPI options
>
>                                                                │
>  │         -> CPU Frequency scaling
>
>                                                                │
>  │           -> CPU Frequency scaling (CPU_FREQ [=y])
>
>                                                                │
>  │ Selects: ACPI_PROCESSOR [=y] && ACPI_CPPC_LIB [=n] &&
> CPU_FREQ_GOV_SCHEDUTIL [=y]
>
>
>
> │ Symbol: ACPI_CPPC_LIB [=n]
>
>                                                               │
>  │ Type  : bool
>
>                                                                │
>  │ Defined at drivers/acpi/Kconfig:268
>
>                                                                │
>  │   Depends on: ACPI [=y] && ACPI_PROCESSOR [=y]
>
>                                                                │
>  │ Selects: MAILBOX [=n] && PCC [=n]
>
>                                                                │
>  │ Selected by [n]:
>
>                                                                │
>  │   - X86_INTEL_PSTATE [=n] && CPU_FREQ [=y] && X86 [=y] && X86_64
> [=y] && ACPI [=y] && SCHED_MC_PRIO [=n]
>                                                                   │
>  │   - X86_AMD_PSTATE [=y] && CPU_FREQ [=y] && X86 [=y] && X86_64 [=y]
> && ACPI [=y] && SCHED_MC_PRIO [=n]
>                                                                │
>  │   - ACPI_CPPC_CPUFREQ [=n] && CPU_FREQ [=y] && (ARM || ARM64) &&
> ACPI_PROCESSOR [=y]
>
>
>
> │ Symbol: SCHED_MC_PRIO [=n]
>
>                                                               │
>  │ Type  : bool
>
>                                                                │
>  │ Defined at arch/x86/Kconfig:1029
>
>                                                                │
>  │   Prompt: CPU core priorities scheduler support
>
>                                                                │
>  │   Depends on: SCHED_MC [=y] && CPU_SUP_INTEL [=n]
>
>                                                                │
>  │   Location:
>
>                                                                │
>  │     Main menu
>
>                                                                │
>  │       -> Processor type and features
>
>                                                                │
>  │ (1)     -> Multi-core scheduler support (SCHED_MC [=y])
>
>                                                                │
>  │ Selects: X86_INTEL_PSTATE [=n] && CPU_FREQ [=y]
>
>
>
> I've used this patch to work around the issue:
>
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index 95dd1ee01546..bdfcf155f48e 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -1028,8 +1028,8 @@ config SCHED_MC
>
> config SCHED_MC_PRIO
>        bool "CPU core priorities scheduler support"
> -       depends on SCHED_MC && CPU_SUP_INTEL
> -       select X86_INTEL_PSTATE
> +       depends on SCHED_MC && CPU_SUP_AMD
> +       select X86_AMD_PSTATE
>        select CPU_FREQ
>        default y
>        help
>
> Which gets things working here

Re: AMD Pstate

[Huang Rui]

On Tue, Nov 23, 2021 at 06:47:34PM +0800, Mike Lothian wrote:
> Hi again
> 
> Do I need to enable something else for shedutil to work, I've tried
> with and without shared_mem? Compiling a kernel with -j16 has the freq
> stuck at 1GHz on all cores - ondemand works fine
> 

If you set the CONFIG_CPU_FREQ_GOV_SCHEDUTIL, then no more configurations.

Did you use the V4 version?

https://git.kernel.org/pub/scm/linux/kernel/git/rui/linux.git/log/?h=amd-pstate-dev-v4

> That's on a AMD Ryzen 9 5900HX
> 
> Also compiling with Clang-13 gives the following warnings:
> 
> drivers/cpufreq/amd-pstate.c:257:6: warning: variable 'amd_des_perf'
> is used uninitialized whenever 'if' condition is false
> [-Wsometimes-uninitialized]
> 
>         if (target_perf < capacity)
> 
>             ^~~~~~~~~~~~~~~~~~~~~~
> 
> drivers/cpufreq/amd-pstate.c:272:40: note: uninitialized use occurs here
> 
>         amd_des_perf = clamp_t(unsigned long, amd_des_perf,
> 
>                                               ^~~~~~~~~~~~

Ah, miss one line. Thanks for the info. I will correct it.

Thanks,
Ray

> 
> ./include/linux/minmax.h:124:60: note: expanded from macro 'clamp_t'
> 
> #define clamp_t(type, val, lo, hi) min_t(type, max_t(type, val, lo), hi)
> 
>                                                            ^~~
> 
> ./include/linux/minmax.h:112:48: note: expanded from macro 'max_t'
> 
> #define max_t(type, x, y)       __careful_cmp((type)(x), (type)(y), >)
> 
>                                                      ^
> 
> ./include/linux/minmax.h:38:14: note: expanded from macro '__careful_cmp'
> 
>                 __cmp_once(x, y, __UNIQUE_ID(__x), __UNIQUE_ID(__y), op))
> 
>                            ^
> 
> note: (skipping 1 expansions in backtrace; use
> -fmacro-backtrace-limit=0 to see all)
> 
> ./include/linux/minmax.h:104:48: note: expanded from macro 'min_t'
> 
> #define min_t(type, x, y)       __careful_cmp((type)(x), (type)(y), <)
> 
>                                                      ^
> 
> ./include/linux/minmax.h:38:14: note: expanded from macro '__careful_cmp'
> 
>                 __cmp_once(x, y, __UNIQUE_ID(__x), __UNIQUE_ID(__y), op))
> 
>                            ^
> 
> ./include/linux/minmax.h:31:25: note: expanded from macro '__cmp_once'
> 
>                 typeof(x) unique_x = (x);               \
> 
>                                       ^
> 
> drivers/cpufreq/amd-pstate.c:257:2: note: remove the 'if' if its
> condition is always true
> 
>         if (target_perf < capacity)
> 
>         ^~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
> drivers/cpufreq/amd-pstate.c:249:56: note: initialize the variable
> 'amd_des_perf' to silence this warning
> 
>         unsigned long amd_max_perf, amd_min_perf, amd_des_perf,
> 
>                                                               ^
> 
>                                                                = 0
> 
> On Sun, 21 Nov 2021 at 10:37, Mike Lothian <mike@fireburn.co.uk> wrote:
> >
> > Hi
> >
> > I'm not subscribed to the mailing list so apologies for emailing you directly
> >
> > It looks like something isn't quite right with the kconfig selects
> >
> > SCHED_MC_PRIO can't be enabled without CPU_SUP_INTEL
> >
> >
> >
> > │ Symbol: X86_AMD_PSTATE [=y]
> >
> >                                                               │
> >  │ Type  : bool
> >
> >                                                                │
> >  │ Defined at drivers/cpufreq/Kconfig.x86:37
> >
> >                                                                │
> >  │   Prompt: AMD Processor P-State driver
> >
> >                                                                │
> >  │   Depends on: CPU_FREQ [=y] && X86 [=y]
> >
> >                                                                │
> >  │   Location:
> >
> >                                                                │
> >  │     Main menu
> >
> >                                                                │
> >  │       -> Power management and ACPI options
> >
> >                                                                │
> >  │         -> CPU Frequency scaling
> >
> >                                                                │
> >  │           -> CPU Frequency scaling (CPU_FREQ [=y])
> >
> >                                                                │
> >  │ Selects: ACPI_PROCESSOR [=y] && ACPI_CPPC_LIB [=n] &&
> > CPU_FREQ_GOV_SCHEDUTIL [=y]
> >
> >
> >
> > │ Symbol: ACPI_CPPC_LIB [=n]
> >
> >                                                               │
> >  │ Type  : bool
> >
> >                                                                │
> >  │ Defined at drivers/acpi/Kconfig:268
> >
> >                                                                │
> >  │   Depends on: ACPI [=y] && ACPI_PROCESSOR [=y]
> >
> >                                                                │
> >  │ Selects: MAILBOX [=n] && PCC [=n]
> >
> >                                                                │
> >  │ Selected by [n]:
> >
> >                                                                │
> >  │   - X86_INTEL_PSTATE [=n] && CPU_FREQ [=y] && X86 [=y] && X86_64
> > [=y] && ACPI [=y] && SCHED_MC_PRIO [=n]
> >                                                                   │
> >  │   - X86_AMD_PSTATE [=y] && CPU_FREQ [=y] && X86 [=y] && X86_64 [=y]
> > && ACPI [=y] && SCHED_MC_PRIO [=n]
> >                                                                │
> >  │   - ACPI_CPPC_CPUFREQ [=n] && CPU_FREQ [=y] && (ARM || ARM64) &&
> > ACPI_PROCESSOR [=y]
> >
> >
> >
> > │ Symbol: SCHED_MC_PRIO [=n]
> >
> >                                                               │
> >  │ Type  : bool
> >
> >                                                                │
> >  │ Defined at arch/x86/Kconfig:1029
> >
> >                                                                │
> >  │   Prompt: CPU core priorities scheduler support
> >
> >                                                                │
> >  │   Depends on: SCHED_MC [=y] && CPU_SUP_INTEL [=n]
> >
> >                                                                │
> >  │   Location:
> >
> >                                                                │
> >  │     Main menu
> >
> >                                                                │
> >  │       -> Processor type and features
> >
> >                                                                │
> >  │ (1)     -> Multi-core scheduler support (SCHED_MC [=y])
> >
> >                                                                │
> >  │ Selects: X86_INTEL_PSTATE [=n] && CPU_FREQ [=y]
> >
> >
> >
> > I've used this patch to work around the issue:
> >
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index 95dd1ee01546..bdfcf155f48e 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -1028,8 +1028,8 @@ config SCHED_MC
> >
> > config SCHED_MC_PRIO
> >        bool "CPU core priorities scheduler support"
> > -       depends on SCHED_MC && CPU_SUP_INTEL
> > -       select X86_INTEL_PSTATE
> > +       depends on SCHED_MC && CPU_SUP_AMD
> > +       select X86_AMD_PSTATE
> >        select CPU_FREQ
> >        default y
> >        help
> >
> > Which gets things working here

Re: AMD Pstate

[Mike Lothian]

On Tue, 23 Nov 2021 at 11:18, Huang Rui <ray.huang@amd.com> wrote:
>
> On Tue, Nov 23, 2021 at 06:47:34PM +0800, Mike Lothian wrote:
> > Hi again
> >
> > Do I need to enable something else for shedutil to work, I've tried
> > with and without shared_mem? Compiling a kernel with -j16 has the freq
> > stuck at 1GHz on all cores - ondemand works fine
> >
>
> If you set the CONFIG_CPU_FREQ_GOV_SCHEDUTIL, then no more configurations.
>
> Did you use the V4 version?
>
> https://git.kernel.org/pub/scm/linux/kernel/git/rui/linux.git/log/?h=amd-pstate-dev-v4
>
> > That's on a AMD Ryzen 9 5900HX
> >
> > Also compiling with Clang-13 gives the following warnings:
> >
> > drivers/cpufreq/amd-pstate.c:257:6: warning: variable 'amd_des_perf'
> > is used uninitialized whenever 'if' condition is false
> > [-Wsometimes-uninitialized]
> >
> >         if (target_perf < capacity)
> >
> >             ^~~~~~~~~~~~~~~~~~~~~~
> >
> > drivers/cpufreq/amd-pstate.c:272:40: note: uninitialized use occurs here
> >
> >         amd_des_perf = clamp_t(unsigned long, amd_des_perf,
> >
> >                                               ^~~~~~~~~~~~
>
> Ah, miss one line. Thanks for the info. I will correct it.
>
> Thanks,
> Ray
>

Hi

Yes I used v4 and CONFIG_CPU_FREQ_GOV_SCHEDUTIL was set as the
default, is there any other info that would be useful, I can switch
back to schedutil from ondemand and collect for you

Cheers

Mike

Re: AMD Pstate

[Huang Rui]

On Tue, Nov 23, 2021 at 07:22:24PM +0800, Mike Lothian wrote:
> On Tue, 23 Nov 2021 at 11:18, Huang Rui <ray.huang@amd.com> wrote:
> >
> > On Tue, Nov 23, 2021 at 06:47:34PM +0800, Mike Lothian wrote:
> > > Hi again
> > >
> > > Do I need to enable something else for shedutil to work, I've tried
> > > with and without shared_mem? Compiling a kernel with -j16 has the freq
> > > stuck at 1GHz on all cores - ondemand works fine
> > >
> >
> > If you set the CONFIG_CPU_FREQ_GOV_SCHEDUTIL, then no more configurations.
> >
> > Did you use the V4 version?
> >
> > https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit.kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Frui%2Flinux.git%2Flog%2F%3Fh%3Damd-pstate-dev-v4&amp;data=04%7C01%7Cray.huang%40amd.com%7C3006dbf33ae54ffaa1aa08d9ae738d2d%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637732633874328299%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=L1H7yy9UCweIH9CaSYTrfo4DZUDok1wgQxjtNCNLC9g%3D&amp;reserved=0
> >
> > > That's on a AMD Ryzen 9 5900HX
> > >
> > > Also compiling with Clang-13 gives the following warnings:
> > >
> > > drivers/cpufreq/amd-pstate.c:257:6: warning: variable 'amd_des_perf'
> > > is used uninitialized whenever 'if' condition is false
> > > [-Wsometimes-uninitialized]
> > >
> > >         if (target_perf < capacity)
> > >
> > >             ^~~~~~~~~~~~~~~~~~~~~~
> > >
> > > drivers/cpufreq/amd-pstate.c:272:40: note: uninitialized use occurs here
> > >
> > >         amd_des_perf = clamp_t(unsigned long, amd_des_perf,
> > >
> > >                                               ^~~~~~~~~~~~
> >
> > Ah, miss one line. Thanks for the info. I will correct it.
> >
> > Thanks,
> > Ray
> >
> 
> Hi
> 
> Yes I used v4 and CONFIG_CPU_FREQ_GOV_SCHEDUTIL was set as the
> default, is there any other info that would be useful, I can switch
> back to schedutil from ondemand and collect for you
> 

> > > That's on a AMD Ryzen 9 5900HX

May I know the CPU family/model id? (you can dump it from lscpu)

But you may need to add "amd_pstate.shared_mem=1" in your grub to enable
the shared memory manually. Because we are debugging the performance issue
on shared memory solution, so it is disabled by default.

> > >         amd_des_perf = clamp_t(unsigned long, amd_des_perf,

It's not the V4's codes. I renamed this variate in the V4.

Thanks,
Ray

Re: AMD Pstate

[Mike Lothian]

On Tue, 23 Nov 2021 at 11:33, Huang Rui <ray.huang@amd.com> wrote:
>
> >
> > Hi
> >
> > Yes I used v4 and CONFIG_CPU_FREQ_GOV_SCHEDUTIL was set as the
> > default, is there any other info that would be useful, I can switch
> > back to schedutil from ondemand and collect for you
> >
>
> > > > That's on a AMD Ryzen 9 5900HX
>
> May I know the CPU family/model id? (you can dump it from lscpu)
>
> But you may need to add "amd_pstate.shared_mem=1" in your grub to enable
> the shared memory manually. Because we are debugging the performance issue
> on shared memory solution, so it is disabled by default.
>
> > > >         amd_des_perf = clamp_t(unsigned long, amd_des_perf,
>
> It's not the V4's codes. I renamed this variate in the V4.
>
> Thanks,
> Ray

Hmm very strange, can you confirm this download from patchwork (which
is what's applied) is correct or not
https://patchwork.kernel.org/series/582973/mbox/

Architecture:            x86_64
 CPU op-mode(s):        32-bit, 64-bit
 Address sizes:         48 bits physical, 48 bits virtual
 Byte Order:            Little Endian
CPU(s):                  16
 On-line CPU(s) list:   0-15
Vendor ID:               AuthenticAMD
 Model name:            AMD Ryzen 9 5900HX with Radeon Graphics
   CPU family:          25
   Model:               80
   Thread(s) per core:  2
   Core(s) per socket:  8
   Socket(s):           1
   Stepping:            0
   Frequency boost:     enabled
   CPU max MHz:         4890.0000
   CPU min MHz:         400.0000
   BogoMIPS:            6587.55
   Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep
mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx
mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl
nonstop_tsc
                        cpuid extd_apicid aperfmperf rapl pni
pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave
avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a
misaligns
                        se 3dnowprefetch osvw ibs skinit wdt tce
topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3
cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2
smep bm
                        i2 erms invpcid cqm rdt_a rdseed adx
clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc
cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr
rdpru wbnoinvd cppc
                         arat npt lbrv svm_lock nrip_save tsc_scale
vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic
v_vmsave_vmload vgif v_spec_ctrl umip pku vaes vpclmulqdq rdpid
overflow_re
                        cov succor smca fsrm
Virtualization features:
 Virtualization:        AMD-V
Caches (sum of all):
 L1d:                   256 KiB (8 instances)
 L1i:                   256 KiB (8 instances)
 L2:                    4 MiB (8 instances)
 L3:                    16 MiB (1 instance)
Vulnerabilities:
 Itlb multihit:         Not affected
 L1tf:                  Not affected
 Mds:                   Not affected
 Meltdown:              Not affected
 Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl
 Spectre v1:            Mitigation; usercopy/swapgs barriers and
__user pointer sanitization
 Spectre v2:            Mitigation; Full AMD retpoline, IBPB
conditional, IBRS_FW, STIBP always-on, RSB filling
 Srbds:                 Not affected
 Tsx async abort:       Not affected

Re: AMD Pstate

[Mike Lothian]

Is there any extra info that might be useful to debug this?

RE: AMD Pstate

[Du, Xiaojian]

[AMD Official Use Only]

Hi Mike

Could you check the smc firmware version?
AMD-Pstate driver will communicate with SMU, so we need to confirm your current smc firmware version is not older than ours.

Our test platform is B550, smc firmware version is:
SMC feature version: 0, firmware version: 0x00283300

You can fetch this info like this:

$sudo su (switch to root)
$cd  /sys/kernel/debug/dri/   (enter the sysfs node folder)
$ll (to list the GPU device(s))
$cd 0 (if your platform has only one GPU, the "0" means this GPU device exactly; but if you have one iGPU and one dGPU on your platform, you need to have a try, like "0" or "1")
$cat amdgpu_firmware_info (if you have two AMD GPUs, you need to capture both of them)

It will print like this:
VCE feature version: 0, firmware version: 0x00000000
UVD feature version: 0, firmware version: 0x00000000
MC feature version: 0, firmware version: 0x00000000
ME feature version: 31, firmware version: 0x0000005e
PFP feature version: 31, firmware version: 0x0000008f
CE feature version: 31, firmware version: 0x00000025
RLC feature version: 1, firmware version: 0x0000004c
RLC SRLC feature version: 0, firmware version: 0x00000000
RLC SRLG feature version: 0, firmware version: 0x00000000
RLC SRLS feature version: 0, firmware version: 0x00000000
MEC feature version: 31, firmware version: 0x00000087
MEC2 feature version: 31, firmware version: 0x00000087
SOS feature version: 0, firmware version: 0x00160052
ASD feature version: 0, firmware version: 0x21000036
TA XGMI feature version: 0, firmware version: 0x00000000
TA RAS feature version: 0, firmware version: 0x00000000
SMC feature version: 0, firmware version: 0x00351f00
SDMA0 feature version: 50, firmware version: 0x00000024
SDMA1 feature version: 50, firmware version: 0x00000024
VCN feature version: 0, firmware version: 0x0510a00d
DMCU feature version: 0, firmware version: 0x00000000
VBIOS version: 113-D3231000-101

And we need the "smc firmware version" and "VBIOS version".
BTW, thank for your test and feedback.

Thanks,
Xiaojian


-----Original Message-----
From: Mike Lothian <mike@fireburn.co.uk>
Sent: 2021年11月26日 18:50
To: Huang, Ray <Ray.Huang@amd.com>
Cc: Linux PM list <linux-pm@vger.kernel.org>; Su, Jinzhou (Joe) <Jinzhou.Su@amd.com>; Du, Xiaojian <Xiaojian.Du@amd.com>
Subject: Re: AMD Pstate

Is there any extra info that might be useful to debug this?

Re: AMD Pstate

[Mike Lothian]

[-- Attachment #1: Type: text/plain, Size: 6050 bytes --]

So

0 is my 6800M

VCE feature version: 0, firmware version: 0x00000000
UVD feature version: 0, firmware version: 0x00000000
MC feature version: 0, firmware version: 0x00000000
ME feature version: 40, firmware version: 0x00000040
PFP feature version: 40, firmware version: 0x00000058
CE feature version: 40, firmware version: 0x00000024
RLC feature version: 1, firmware version: 0x00000044
RLC SRLC feature version: 0, firmware version: 0x00000000
RLC SRLG feature version: 0, firmware version: 0x00000000
RLC SRLS feature version: 0, firmware version: 0x00000000
MEC feature version: 40, firmware version: 0x0000005c
MEC2 feature version: 40, firmware version: 0x0000005c
SOS feature version: 0, firmware version: 0x0022020a
ASD feature version: 553648226, firmware version: 0x21000062
TA XGMI feature version: 0x00000000, firmware version: 0x00000000
TA RAS feature version: 0x00000000, firmware version: 0x00000000
TA HDCP feature version: 0x00000000, firmware version: 0x17000024
TA DTM feature version: 0x00000000, firmware version: 0x1200000b
TA RAP feature version: 0x00000000, firmware version: 0x0700000e
TA SECUREDISPLAY feature version: 0x00000000, firmware version: 0x00000000
SMC feature version: 0, firmware version: 0x00412e00 (65.46.0)
SDMA0 feature version: 52, firmware version: 0x00000045
SDMA1 feature version: 52, firmware version: 0x00000045
VCN feature version: 0, firmware version: 0x02110001
DMCU feature version: 0, firmware version: 0x00000000
DMCUB feature version: 0, firmware version: 0x02020007
TOC feature version: 0, firmware version: 0x00000000
VBIOS version: SWBRT77321.001

And 1 is the onboard graphics for the 5900HX

VCE feature version: 0, firmware version: 0x00000000
UVD feature version: 0, firmware version: 0x00000000
MC feature version: 0, firmware version: 0x00000000
ME feature version: 53, firmware version: 0x000000a6
PFP feature version: 53, firmware version: 0x000000c0
CE feature version: 53, firmware version: 0x0000004f
RLC feature version: 1, firmware version: 0x0000003c
RLC SRLC feature version: 1, firmware version: 0x00000001
RLC SRLG feature version: 1, firmware version: 0x00000001
RLC SRLS feature version: 1, firmware version: 0x00000001
MEC feature version: 53, firmware version: 0x000001ce
SOS feature version: 0, firmware version: 0x00000000
ASD feature version: 0, firmware version: 0x21000064
TA XGMI feature version: 0x00000000, firmware version: 0x00000000
TA RAS feature version: 0x00000000, firmware version: 0x00000000
TA HDCP feature version: 0x00000000, firmware version: 0x17000025
TA DTM feature version: 0x00000000, firmware version: 0x1200000c
TA RAP feature version: 0x00000000, firmware version: 0x00000000
TA SECUREDISPLAY feature version: 0x00000000, firmware version: 0x00000000
SMC feature version: 0, firmware version: 0x00402d00 (64.45.0)
SDMA0 feature version: 41, firmware version: 0x00000028
VCN feature version: 0, firmware version: 0x05110003
DMCU feature version: 0, firmware version: 0x00000000
DMCUB feature version: 0, firmware version: 0x0101001c
TOC feature version: 0, firmware version: 0x00000000
VBIOS version: 113-CEZANNE-018

That's using the latest firmware from
git://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git

I've attached my dmesg too in case its useful

I've switched back to ondemand to make my system performant again

On Fri, 26 Nov 2021 at 11:32, Du, Xiaojian <Xiaojian.Du@amd.com> wrote:
>
> [AMD Official Use Only]
>
> Hi Mike
>
> Could you check the smc firmware version?
> AMD-Pstate driver will communicate with SMU, so we need to confirm your current smc firmware version is not older than ours.
>
> Our test platform is B550, smc firmware version is:
> SMC feature version: 0, firmware version: 0x00283300
>
> You can fetch this info like this:
>
> $sudo su (switch to root)
> $cd  /sys/kernel/debug/dri/   (enter the sysfs node folder)
> $ll (to list the GPU device(s))
> $cd 0 (if your platform has only one GPU, the "0" means this GPU device exactly; but if you have one iGPU and one dGPU on your platform, you need to have a try, like "0" or "1")
> $cat amdgpu_firmware_info (if you have two AMD GPUs, you need to capture both of them)
>
> It will print like this:
> VCE feature version: 0, firmware version: 0x00000000
> UVD feature version: 0, firmware version: 0x00000000
> MC feature version: 0, firmware version: 0x00000000
> ME feature version: 31, firmware version: 0x0000005e
> PFP feature version: 31, firmware version: 0x0000008f
> CE feature version: 31, firmware version: 0x00000025
> RLC feature version: 1, firmware version: 0x0000004c
> RLC SRLC feature version: 0, firmware version: 0x00000000
> RLC SRLG feature version: 0, firmware version: 0x00000000
> RLC SRLS feature version: 0, firmware version: 0x00000000
> MEC feature version: 31, firmware version: 0x00000087
> MEC2 feature version: 31, firmware version: 0x00000087
> SOS feature version: 0, firmware version: 0x00160052
> ASD feature version: 0, firmware version: 0x21000036
> TA XGMI feature version: 0, firmware version: 0x00000000
> TA RAS feature version: 0, firmware version: 0x00000000
> SMC feature version: 0, firmware version: 0x00351f00
> SDMA0 feature version: 50, firmware version: 0x00000024
> SDMA1 feature version: 50, firmware version: 0x00000024
> VCN feature version: 0, firmware version: 0x0510a00d
> DMCU feature version: 0, firmware version: 0x00000000
> VBIOS version: 113-D3231000-101
>
> And we need the "smc firmware version" and "VBIOS version".
> BTW, thank for your test and feedback.
>
> Thanks,
> Xiaojian
>
>
> -----Original Message-----
> From: Mike Lothian <mike@fireburn.co.uk>
> Sent: 2021年11月26日 18:50
> To: Huang, Ray <Ray.Huang@amd.com>
> Cc: Linux PM list <linux-pm@vger.kernel.org>; Su, Jinzhou (Joe) <Jinzhou.Su@amd.com>; Du, Xiaojian <Xiaojian.Du@amd.com>
> Subject: Re: AMD Pstate
>
> Is there any extra info that might be useful to debug this?

[-- Attachment #2: dmesg.ondemand.xz --]
[-- Type: application/x-xz, Size: 19068 bytes --]

Re: AMD Pstate

[Huang Rui]

On Fri, Nov 26, 2021 at 09:24:43PM +0800, Mike Lothian wrote:
> So
> 
> 0 is my 6800M
> 
> VCE feature version: 0, firmware version: 0x00000000
> UVD feature version: 0, firmware version: 0x00000000
> MC feature version: 0, firmware version: 0x00000000
> ME feature version: 40, firmware version: 0x00000040
> PFP feature version: 40, firmware version: 0x00000058
> CE feature version: 40, firmware version: 0x00000024
> RLC feature version: 1, firmware version: 0x00000044
> RLC SRLC feature version: 0, firmware version: 0x00000000
> RLC SRLG feature version: 0, firmware version: 0x00000000
> RLC SRLS feature version: 0, firmware version: 0x00000000
> MEC feature version: 40, firmware version: 0x0000005c
> MEC2 feature version: 40, firmware version: 0x0000005c
> SOS feature version: 0, firmware version: 0x0022020a
> ASD feature version: 553648226, firmware version: 0x21000062
> TA XGMI feature version: 0x00000000, firmware version: 0x00000000
> TA RAS feature version: 0x00000000, firmware version: 0x00000000
> TA HDCP feature version: 0x00000000, firmware version: 0x17000024
> TA DTM feature version: 0x00000000, firmware version: 0x1200000b
> TA RAP feature version: 0x00000000, firmware version: 0x0700000e
> TA SECUREDISPLAY feature version: 0x00000000, firmware version: 0x00000000
> SMC feature version: 0, firmware version: 0x00412e00 (65.46.0)
> SDMA0 feature version: 52, firmware version: 0x00000045
> SDMA1 feature version: 52, firmware version: 0x00000045
> VCN feature version: 0, firmware version: 0x02110001
> DMCU feature version: 0, firmware version: 0x00000000
> DMCUB feature version: 0, firmware version: 0x02020007
> TOC feature version: 0, firmware version: 0x00000000
> VBIOS version: SWBRT77321.001
> 
> And 1 is the onboard graphics for the 5900HX
> 
> VCE feature version: 0, firmware version: 0x00000000
> UVD feature version: 0, firmware version: 0x00000000
> MC feature version: 0, firmware version: 0x00000000
> ME feature version: 53, firmware version: 0x000000a6
> PFP feature version: 53, firmware version: 0x000000c0
> CE feature version: 53, firmware version: 0x0000004f
> RLC feature version: 1, firmware version: 0x0000003c
> RLC SRLC feature version: 1, firmware version: 0x00000001
> RLC SRLG feature version: 1, firmware version: 0x00000001
> RLC SRLS feature version: 1, firmware version: 0x00000001
> MEC feature version: 53, firmware version: 0x000001ce
> SOS feature version: 0, firmware version: 0x00000000
> ASD feature version: 0, firmware version: 0x21000064
> TA XGMI feature version: 0x00000000, firmware version: 0x00000000
> TA RAS feature version: 0x00000000, firmware version: 0x00000000
> TA HDCP feature version: 0x00000000, firmware version: 0x17000025
> TA DTM feature version: 0x00000000, firmware version: 0x1200000c
> TA RAP feature version: 0x00000000, firmware version: 0x00000000
> TA SECUREDISPLAY feature version: 0x00000000, firmware version: 0x00000000
> SMC feature version: 0, firmware version: 0x00402d00 (64.45.0)

Thanks for information. Could you please enable the trace event of
amd_pstate_perf before you run the test.

echo 1 > /sys/kernel/tracing/events/amd_cpu/enable

Then dump the traces with "cat /sys/kernel/debug/tracing/trace". We can
double confirm whether the amd-pstate driver program the correct
performance value in your processor.

Thanks,
Ray

Re: AMD Pstate

[Huang Rui]

On Wed, Dec 01, 2021 at 01:40:52PM +0800, Mike Lothian wrote:
>    I've just tested v5 of the patchset, the warnings are gone, kconfig
>    looks good and shedutil seems to be behaving
>    Would it still be useful to get the above trace?

Is the previous issue that the cpu clock stuck on 1 GHz still existed on
V5?

Thanks,
Ray

> 
>    On Tue, 30 Nov 2021 at 10:50, Huang Rui <[1]ray.huang@amd.com> wrote:
> 
>      On Fri, Nov 26, 2021 at 09:24:43PM +0800, Mike Lothian wrote:
>      > So
>      >
>      > 0 is my 6800M
>      >
>      > VCE feature version: 0, firmware version: 0x00000000
>      > UVD feature version: 0, firmware version: 0x00000000
>      > MC feature version: 0, firmware version: 0x00000000
>      > ME feature version: 40, firmware version: 0x00000040
>      > PFP feature version: 40, firmware version: 0x00000058
>      > CE feature version: 40, firmware version: 0x00000024
>      > RLC feature version: 1, firmware version: 0x00000044
>      > RLC SRLC feature version: 0, firmware version: 0x00000000
>      > RLC SRLG feature version: 0, firmware version: 0x00000000
>      > RLC SRLS feature version: 0, firmware version: 0x00000000
>      > MEC feature version: 40, firmware version: 0x0000005c
>      > MEC2 feature version: 40, firmware version: 0x0000005c
>      > SOS feature version: 0, firmware version: 0x0022020a
>      > ASD feature version: 553648226, firmware version: 0x21000062
>      > TA XGMI feature version: 0x00000000, firmware version: 0x00000000
>      > TA RAS feature version: 0x00000000, firmware version: 0x00000000
>      > TA HDCP feature version: 0x00000000, firmware version: 0x17000024
>      > TA DTM feature version: 0x00000000, firmware version: 0x1200000b
>      > TA RAP feature version: 0x00000000, firmware version: 0x0700000e
>      > TA SECUREDISPLAY feature version: 0x00000000, firmware version:
>      0x00000000
>      > SMC feature version: 0, firmware version: 0x00412e00 (65.46.0)
>      > SDMA0 feature version: 52, firmware version: 0x00000045
>      > SDMA1 feature version: 52, firmware version: 0x00000045
>      > VCN feature version: 0, firmware version: 0x02110001
>      > DMCU feature version: 0, firmware version: 0x00000000
>      > DMCUB feature version: 0, firmware version: 0x02020007
>      > TOC feature version: 0, firmware version: 0x00000000
>      > VBIOS version: SWBRT77321.001
>      >
>      > And 1 is the onboard graphics for the 5900HX
>      >
>      > VCE feature version: 0, firmware version: 0x00000000
>      > UVD feature version: 0, firmware version: 0x00000000
>      > MC feature version: 0, firmware version: 0x00000000
>      > ME feature version: 53, firmware version: 0x000000a6
>      > PFP feature version: 53, firmware version: 0x000000c0
>      > CE feature version: 53, firmware version: 0x0000004f
>      > RLC feature version: 1, firmware version: 0x0000003c
>      > RLC SRLC feature version: 1, firmware version: 0x00000001
>      > RLC SRLG feature version: 1, firmware version: 0x00000001
>      > RLC SRLS feature version: 1, firmware version: 0x00000001
>      > MEC feature version: 53, firmware version: 0x000001ce
>      > SOS feature version: 0, firmware version: 0x00000000
>      > ASD feature version: 0, firmware version: 0x21000064
>      > TA XGMI feature version: 0x00000000, firmware version: 0x00000000
>      > TA RAS feature version: 0x00000000, firmware version: 0x00000000
>      > TA HDCP feature version: 0x00000000, firmware version: 0x17000025
>      > TA DTM feature version: 0x00000000, firmware version: 0x1200000c
>      > TA RAP feature version: 0x00000000, firmware version: 0x00000000
>      > TA SECUREDISPLAY feature version: 0x00000000, firmware version:
>      0x00000000
>      > SMC feature version: 0, firmware version: 0x00402d00 (64.45.0)
>      Thanks for information. Could you please enable the trace event of
>      amd_pstate_perf before you run the test.
>      echo 1 > /sys/kernel/tracing/events/amd_cpu/enable
>      Then dump the traces with "cat /sys/kernel/debug/tracing/trace". We
>      can
>      double confirm whether the amd-pstate driver program the correct
>      performance value in your processor.
>      Thanks,
>      Ray
> 
> References
> 
>    1. mailto:ray.huang@amd.com

Re: AMD Pstate

[Mike Lothian]

On Wed, 1 Dec 2021 at 06:51, Huang Rui <ray.huang@amd.com> wrote:
>
> On Wed, Dec 01, 2021 at 01:40:52PM +0800, Mike Lothian wrote:
> >    I've just tested v5 of the patchset, the warnings are gone, kconfig
> >    looks good and shedutil seems to be behaving
> >    Would it still be useful to get the above trace?
>
> Is the previous issue that the cpu clock stuck on 1 GHz still existed on
> V5?
>
> Thanks,
> Ray
>

 Hi

No the issue appears to be resolved, schedutil is now going to the
maximum frequency - it appears to be behaving now

Thanks

Re: AMD Pstate

[Huang Rui]

On Fri, Dec 03, 2021 at 09:08:14AM +0800, Mike Lothian wrote:
> On Wed, 1 Dec 2021 at 06:51, Huang Rui <ray.huang@amd.com> wrote:
> >
> > On Wed, Dec 01, 2021 at 01:40:52PM +0800, Mike Lothian wrote:
> > >    I've just tested v5 of the patchset, the warnings are gone, kconfig
> > >    looks good and shedutil seems to be behaving
> > >    Would it still be useful to get the above trace?
> >
> > Is the previous issue that the cpu clock stuck on 1 GHz still existed on
> > V5?
> >
> > Thanks,
> > Ray
> >
> 
>  Hi
> 
> No the issue appears to be resolved, schedutil is now going to the
> maximum frequency - it appears to be behaving now
> 

Good to hear that, thanks!

Ray