Re: [PATCH v2] dt-bingdings: perf: Support uncore ARM NI-700 PMU

[Robin Murphy]

On 31/01/2024 10:18 am, Yang Jialong 杨佳龙 wrote:
> 在 2024/1/31 18:16, Krzysztof Kozlowski 写道:
> > On 31/01/2024 11:13, Yang Jialong 杨佳龙 wrote:
> > > 在 2024/1/31 17:30, Krzysztof Kozlowski 写道:
> > > > On 31/01/2024 10:26, Yang Jialong 杨佳龙 wrote:
> > > > > 在 2024/1/31 15:49, Krzysztof Kozlowski 写道:
> > > > > > On 31/01/2024 07:59, JiaLong.Yang wrote:
> > > > > > > Add file corresponding to hx_arm_ni.c introducing ARM NI-700 PMU 
> > > > > > > driver
> > > > > > > for HX.
> > > > > > >
> > > > > > > Signed-off-by: JiaLong.Yang <jialong.yang@xxxxxxxxxxxx>
> > > > > > > ---
> > > > > > > v1 --> v2:
> > > > > > > 1. Submit dt-bindings file Seperately.
> > > > > >
> > > > > > Where is the driver?
> > > > > >
> > > > > > Please read:
> > > > > > https://elixir.bootlin.com/linux/v6.8-rc2/source/Documentation/process/submitting-patches.rst
> > > > > > before posting.
> > > >
> > > > Keep all discussions public.
> > >
> > > Get.
> > >
> > > > > > > +  pccs-id:
> > > > > > > +    $ref: /schemas/types.yaml#/definitions/uint32
> > > > > > > +    description: Used to identify NIs in system which has more than
> > > > > > > +      one NI.
> > > > > >
> > > > > > No, reg does it. Drop the property. Anyway you miss here vendor 
> > > > > > prefix
> > > > > > and proper explanation.
> > > > >
> > > > > reg will tell phy address. Phy address is too long. I just want a id.
> > > > > example: perf stat -a -e ni_pmu_${pccs-id}/cycles/
> > > > > I will use it in user space. Not only in driver.
> > > >
> > > > Custom vendor property is not for that purpose. Use for example IDR, DT
> > > > aliases or something else.
> > >
> > > I have considered TD aliases. It's not very easy. Two places...
> > > IDR.. I have tested. But it chouldn't correspond to the HWs.
> > > I need it to identify NIs.
> > > DT aliases is reachable. But no very easy.
> >
> > Except that "you want" I did not see any rationale, any argument
> > explaining why this is needed and why this has to be present.
>
> OK. DT aliases it good.

The real way to address that particular issue is to fix perf to properly 
associate the PMU device with the underlying hardware device. Jonathan 
had a series doing that[1], but I'm not sure what its status is now.

Thanks,
Robin.

[1] 
https://lore.kernel.org/linux-arm-kernel/20230404134225.13408-1-Jonathan.Cameron@xxxxxxxxxx/