sched/doc: supplement CPU capacity with RISC-V

Commit Message

Song Shuai Feb. 27, 2023, 10:59 a.m. UTC

This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
shared loation") updates some references about capacity-dmips-mhz
property in this document.

The list of architectures using capacity-dmips-mhz omits RISC-V, so
supplements it here.

Signed-off-by: Song Shuai <[e-mail]>
---
 Documentation/scheduler/sched-capacity.rst                    | 2 +-
 Documentation/translations/zh_CN/scheduler/sched-capacity.rst | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

Comments

Conor Dooley Feb. 27, 2023, 11:56 a.m. UTC | #1

On Mon, Feb 27, 2023 at 06:59:41PM +0800, Song Shuai wrote:
> This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
> shared loation") updates some references about capacity-dmips-mhz

Not requesting a respin for this, but mentioning commit 991994509ee9
("dt-bindings: riscv: add a capacity-dmips-mhz cpu property") is
probably more relevant as a justification for this change.

> property in this document.
> 
> The list of architectures using capacity-dmips-mhz omits RISC-V, so
> supplements it here.
> 
> Signed-off-by: Song Shuai <[e-mail]>
> ---
>  Documentation/scheduler/sched-capacity.rst                    | 2 +-
>  Documentation/translations/zh_CN/scheduler/sched-capacity.rst | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/scheduler/sched-capacity.rst b/Documentation/scheduler/sched-capacity.rst
> index 8e2b8538bc2b..e2c1cf743158 100644
> --- a/Documentation/scheduler/sched-capacity.rst
> +++ b/Documentation/scheduler/sched-capacity.rst
> @@ -258,7 +258,7 @@ Linux cannot currently figure out CPU capacity on its own, this information thus
>  needs to be handed to it. Architectures must define arch_scale_cpu_capacity()
>  for that purpose.
>  
> -The arm and arm64 architectures directly map this to the arch_topology driver
> +The arm, arm64, and RISC-V architectures directly map this to the arch_topology driver

My grep clearly missed this change, thanks.
For the only part of the patch I understand:
Reviewed-by: Conor Dooley <[e-mail]>

Thanks,
Conor.

Song Shuai Feb. 27, 2023, 12:40 p.m. UTC | #2

Conor Dooley <[e-mail]> 于2023年2月27日周一 11:57写道：
>
> On Mon, Feb 27, 2023 at 06:59:41PM +0800, Song Shuai wrote:
> > This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
> > shared loation") updates some references about capacity-dmips-mhz
>
> Not requesting a respin for this, but mentioning commit 991994509ee9
> ("dt-bindings: riscv: add a capacity-dmips-mhz cpu property") is
> probably more relevant as a justification for this change.
>
Thanks for your correction, I'll pay attention next time.

I have a question about the patch you mentioned:
The patch uses cpu_scale per_cpu variable to store the CPU capacity
through arch_topology,
But arch_scale_cpu_capacity() interface seems not defined to deliver
the cpu_scale to the scheduler
In contrast, arm64 defines it as the topology_get_cpu_scale() in its
arch/arm64/include/asm/topology.h.
Is this an oversight or a particular purpose?

> > property in this document.
> >
> > The list of architectures using capacity-dmips-mhz omits RISC-V, so
> > supplements it here.
> >
> > Signed-off-by: Song Shuai <[e-mail]>
> > ---
> >  Documentation/scheduler/sched-capacity.rst                    | 2 +-
> >  Documentation/translations/zh_CN/scheduler/sched-capacity.rst | 2 +-
> >  2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/Documentation/scheduler/sched-capacity.rst b/Documentation/scheduler/sched-capacity.rst
> > index 8e2b8538bc2b..e2c1cf743158 100644
> > --- a/Documentation/scheduler/sched-capacity.rst
> > +++ b/Documentation/scheduler/sched-capacity.rst
> > @@ -258,7 +258,7 @@ Linux cannot currently figure out CPU capacity on its own, this information thus
> >  needs to be handed to it. Architectures must define arch_scale_cpu_capacity()
> >  for that purpose.
> >
> > -The arm and arm64 architectures directly map this to the arch_topology driver
> > +The arm, arm64, and RISC-V architectures directly map this to the arch_topology driver
>
> My grep clearly missed this change, thanks.
> For the only part of the patch I understand:
> Reviewed-by: Conor Dooley <[e-mail]>
>
> Thanks,
> Conor.
>

Conor Dooley Feb. 27, 2023, 12:55 p.m. UTC | #3

On Mon, Feb 27, 2023 at 12:40:45PM +0000, Song Shuai wrote:
> Conor Dooley <[e-mail]> 于2023年2月27日周一 11:57写道：
> >
> > On Mon, Feb 27, 2023 at 06:59:41PM +0800, Song Shuai wrote:
> > > This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
> > > shared loation") updates some references about capacity-dmips-mhz
> >
> > Not requesting a respin for this, but mentioning commit 991994509ee9
> > ("dt-bindings: riscv: add a capacity-dmips-mhz cpu property") is
> > probably more relevant as a justification for this change.
> >
> Thanks for your correction, I'll pay attention next time.
> 
> I have a question about the patch you mentioned:
> The patch uses cpu_scale per_cpu variable to store the CPU capacity
> through arch_topology,
> But arch_scale_cpu_capacity() interface seems not defined to deliver
> the cpu_scale to the scheduler
> In contrast, arm64 defines it as the topology_get_cpu_scale() in its
> arch/arm64/include/asm/topology.h.
> Is this an oversight or a particular purpose?

Intentional oversight I suppose? It wasn't my intention to do anything
other than document the property that people were already using in
their devicetrees (and finding bugs with!).
In retrospect, perhaps it is better if I un-review this patch until
we know it is plumbed into the scheduler properly?

Ley Foon Tan is the one that found the RISC-V bugs with this property in
their devicetree, so perhaps they've already done the work here?

Thanks,
Conor.

Ley Foon Tan Feb. 28, 2023, 2 a.m. UTC | #4

> -----Original Message-----
> From: Conor Dooley <[e-mail]>
> Sent: Monday, February 27, 2023 8:56 PM
> To: Song Shuai <[e-mail]>
> Cc: [e-mail]; [e-mail]; [e-mail];
> [e-mail]; [e-mail]; Leyfoon Tan
> <[e-mail]>; [e-mail]; linux-
> [e-mail]; [e-mail]
> Subject: Re: [PATCH] sched/doc: supplement CPU capacity with RISC-V
> 
> On Mon, Feb 27, 2023 at 12:40:45PM +0000, Song Shuai wrote:
> > Conor Dooley <[e-mail]> 于2023年2月27日周一
> 11:57写道：
> > >
> > > On Mon, Feb 27, 2023 at 06:59:41PM +0800, Song Shuai wrote:
> > > > This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to
> > > > a shared loation") updates some references about
> > > > capacity-dmips-mhz
> > >
> > > Not requesting a respin for this, but mentioning commit 991994509ee9
> > > ("dt-bindings: riscv: add a capacity-dmips-mhz cpu property") is
> > > probably more relevant as a justification for this change.
> > >
> > Thanks for your correction, I'll pay attention next time.
> >
> > I have a question about the patch you mentioned:
> > The patch uses cpu_scale per_cpu variable to store the CPU capacity
> > through arch_topology, But arch_scale_cpu_capacity() interface seems
> > not defined to deliver the cpu_scale to the scheduler In contrast,
> > arm64 defines it as the topology_get_cpu_scale() in its
> > arch/arm64/include/asm/topology.h.
> > Is this an oversight or a particular purpose?
> 
> Intentional oversight I suppose? It wasn't my intention to do anything other
> than document the property that people were already using in their
> devicetrees (and finding bugs with!).
> In retrospect, perhaps it is better if I un-review this patch until we know it is
> plumbed into the scheduler properly?
> 
> Ley Foon Tan is the one that found the RISC-V bugs with this property in their
> devicetree, so perhaps they've already done the work here?

I have few patches to wire up few arch_*_topology functions for RISC-V architecture. But haven't send out to ML.

Regards
Ley Foon

Alex Shi Feb. 28, 2023, 7:11 a.m. UTC | #5

On Mon, Feb 27, 2023 at 6:59 PM Song Shuai <[e-mail]> wrote:
>
> This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
> shared loation") updates some references about capacity-dmips-mhz
> property in this document.
>
> The list of architectures using capacity-dmips-mhz omits RISC-V, so
> supplements it here.
>
> Signed-off-by: Song Shuai <[e-mail]>
> ---
>  Documentation/scheduler/sched-capacity.rst                    | 2 +-
>  Documentation/translations/zh_CN/scheduler/sched-capacity.rst | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/Documentation/scheduler/sched-capacity.rst b/Documentation/scheduler/sched-capacity.rst
> index 8e2b8538bc2b..e2c1cf743158 100644
> --- a/Documentation/scheduler/sched-capacity.rst
> +++ b/Documentation/scheduler/sched-capacity.rst
> @@ -258,7 +258,7 @@ Linux cannot currently figure out CPU capacity on its own, this information thus
>  needs to be handed to it. Architectures must define arch_scale_cpu_capacity()
>  for that purpose.
>
> -The arm and arm64 architectures directly map this to the arch_topology driver
> +The arm, arm64, and RISC-V architectures directly map this to the arch_topology driver
>  CPU scaling data, which is derived from the capacity-dmips-mhz CPU binding; see
>  Documentation/devicetree/bindings/cpu/cpu-capacity.txt.
>
> diff --git a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> index e07ffdd391d3..8cba135dcd1a 100644
> --- a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> +++ b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> @@ -231,7 +231,7 @@ CFS调度类基于实体负载跟踪机制（Per-Entity Load Tracking, PELT）
>  当前，Linux无法凭自身算出CPU算力，因此必须要有把这个信息传递给Linux的方式。每个架构必须为此
>  定义arch_scale_cpu_capacity()函数。
>
> -arm和arm64架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
> +arm、arm64和RISC-V架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
>  arch_topology.h的percpu变量cpu_scale），它是从capacity-dmips-mhz CPU binding中衍生计算
>  出来的。参见Documentation/devicetree/bindings/cpu/cpu-capacity.txt。

LGTM,
Reviewed-by: Alex Shi <[e-mail]>

>
> --
> 2.20.1
>

Palmer Dabbelt Feb. 28, 2023, 9:59 p.m. UTC | #6

On Mon, 27 Feb 2023 02:59:41 PST (-0800), [e-mail] wrote:
> This commit 7d2078310cbf ("dt-bindings: arm: move cpu-capacity to a
> shared loation") updates some references about capacity-dmips-mhz
> property in this document.
>
> The list of architectures using capacity-dmips-mhz omits RISC-V, so
> supplements it here.
>
> Signed-off-by: Song Shuai <[e-mail]>
> ---
>  Documentation/scheduler/sched-capacity.rst                    | 2 +-
>  Documentation/translations/zh_CN/scheduler/sched-capacity.rst | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/Documentation/scheduler/sched-capacity.rst b/Documentation/scheduler/sched-capacity.rst
> index 8e2b8538bc2b..e2c1cf743158 100644
> --- a/Documentation/scheduler/sched-capacity.rst
> +++ b/Documentation/scheduler/sched-capacity.rst
> @@ -258,7 +258,7 @@ Linux cannot currently figure out CPU capacity on its own, this information thus
>  needs to be handed to it. Architectures must define arch_scale_cpu_capacity()
>  for that purpose.
>
> -The arm and arm64 architectures directly map this to the arch_topology driver
> +The arm, arm64, and RISC-V architectures directly map this to the arch_topology driver
>  CPU scaling data, which is derived from the capacity-dmips-mhz CPU binding; see
>  Documentation/devicetree/bindings/cpu/cpu-capacity.txt.
>
> diff --git a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> index e07ffdd391d3..8cba135dcd1a 100644
> --- a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> +++ b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
> @@ -231,7 +231,7 @@ CFS调度类基于实体负载跟踪机制（Per-Entity Load Tracking, PELT）
>  当前，Linux无法凭自身算出CPU算力，因此必须要有把这个信息传递给Linux的方式。每个架构必须为此
>  定义arch_scale_cpu_capacity()函数。
>
> -arm和arm64架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
> +arm、arm64和RISC-V架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
>  arch_topology.h的percpu变量cpu_scale），它是从capacity-dmips-mhz CPU binding中衍生计算
>  出来的。参见Documentation/devicetree/bindings/cpu/cpu-capacity.txt。

Reviewed-by: Palmer Dabbelt <[e-mail]> # English
Acked-by: Palmer Dabbelt <[e-mail]>

Happy to take this through the RISC-V tree, but I'll hold off for a bit 
and let the doc folks have a chance to chime in (or just pick it up).  
I'm planning on sending something later this week either way.

Jonathan Corbet Feb. 28, 2023, 10:13 p.m. UTC | #7

Palmer Dabbelt <[e-mail]> writes:

> Happy to take this through the RISC-V tree, but I'll hold off for a bit 
> and let the doc folks have a chance to chime in (or just pick it up).  
> I'm planning on sending something later this week either way.

"Doc folks" are fine if you just include this one in the pull you're
doing.

Thanks,

jon

Jonathan Corbet March 7, 2023, 5:18 p.m. UTC | #8

Jonathan Corbet <[e-mail]> writes:

> Palmer Dabbelt <[e-mail]> writes:
>
>> Happy to take this through the RISC-V tree, but I'll hold off for a bit 
>> and let the doc folks have a chance to chime in (or just pick it up).  
>> I'm planning on sending something later this week either way.
>
> "Doc folks" are fine if you just include this one in the pull you're
> doing.

So it looks like that didn't happen...so I'll go ahead and apply this to
my fixes branch.

Thanks,

jon

Patch

diff --git a/Documentation/scheduler/sched-capacity.rst b/Documentation/scheduler/sched-capacity.rst
index 8e2b8538bc2b..e2c1cf743158 100644
--- a/Documentation/scheduler/sched-capacity.rst
+++ b/Documentation/scheduler/sched-capacity.rst
@@ -258,7 +258,7 @@  Linux cannot currently figure out CPU capacity on its own, this information thus
 needs to be handed to it. Architectures must define arch_scale_cpu_capacity()
 for that purpose.
 
-The arm and arm64 architectures directly map this to the arch_topology driver
+The arm, arm64, and RISC-V architectures directly map this to the arch_topology driver
 CPU scaling data, which is derived from the capacity-dmips-mhz CPU binding; see
 Documentation/devicetree/bindings/cpu/cpu-capacity.txt.
 
diff --git a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
index e07ffdd391d3..8cba135dcd1a 100644
--- a/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
+++ b/Documentation/translations/zh_CN/scheduler/sched-capacity.rst
@@ -231,7 +231,7 @@  CFS调度类基于实体负载跟踪机制（Per-Entity Load Tracking, PELT）
 当前，Linux无法凭自身算出CPU算力，因此必须要有把这个信息传递给Linux的方式。每个架构必须为此
 定义arch_scale_cpu_capacity()函数。
 
-arm和arm64架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
+arm、arm64和RISC-V架构直接把这个信息映射到arch_topology驱动的CPU scaling数据中（译注：参考
 arch_topology.h的percpu变量cpu_scale），它是从capacity-dmips-mhz CPU binding中衍生计算
 出来的。参见Documentation/devicetree/bindings/cpu/cpu-capacity.txt。