linux-kernel - Re: [PATCH 3/4] drm/msm: Add SYSPROF param

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 3 Mar 2022 12:47:10 -0800
From:   Stephen Boyd <swboyd@...omium.org>
To:     Rob Clark <robdclark@...il.com>, dri-devel@...ts.freedesktop.org
Cc:     freedreno@...ts.freedesktop.org, linux-arm-msm@...r.kernel.org,
        Jordan Crouse <jordan@...micpenguin.net>,
        Akhil P Oommen <quic_akhilpo@...cinc.com>,
        Rob Clark <robdclark@...omium.org>,
        Sean Paul <sean@...rly.run>,
        Abhinav Kumar <quic_abhinavk@...cinc.com>,
        David Airlie <airlied@...ux.ie>,
        Daniel Vetter <daniel@...ll.ch>,
        Jonathan Marek <jonathan@...ek.ca>,
        Emma Anholt <emma@...olt.net>,
        Yangtao Li <tiny.windzz@...il.com>,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/4] drm/msm: Add SYSPROF param

Quoting Rob Clark (2022-03-03 11:46:47)
> diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
> index fde9a29f884e..0ba1dbd4e50f 100644
> --- a/drivers/gpu/drm/msm/msm_gpu.h
> +++ b/drivers/gpu/drm/msm/msm_gpu.h
> @@ -330,6 +337,24 @@ struct msm_file_private {
>         struct kref ref;
>         int seqno;
>
> +       /**
> +        * sysprof:
> +        *
> +        * The value of MSM_PARAM_SYSPROF set by userspace.  This is
> +        * intended to be used by system profiling tools like Mesa's
> +        * pps-producer (perfetto), and restricted to CAP_SYS_ADMIN.
> +        *
> +        * Setting a value of 1 will preserve performance counters across
> +        * context switches.  Setting a value of 2 will in addition
> +        * suppress suspend.  (Performance counters loose  state across

s/loose  /lose/

> +        * power collapse, which is undesirable for profiling in some
> +        * cases.)
> +        *
> +        * The value automatically reverts to zero when the drm device
> +        * file is closed.
> +        */
> +       int sysprof;
> +
>         /**
>          * elapsed:
>          *
> diff --git a/drivers/gpu/drm/msm/msm_submitqueue.c b/drivers/gpu/drm/msm/msm_submitqueue.c
> index 7cb158bcbcf6..4179db54ac93 100644
> --- a/drivers/gpu/drm/msm/msm_submitqueue.c
> +++ b/drivers/gpu/drm/msm/msm_submitqueue.c
> @@ -7,6 +7,40 @@
>
>  #include "msm_gpu.h"
>
> +int msm_file_private_set_sysprof(struct msm_file_private *ctx,
> +                                struct msm_gpu *gpu, int sysprof)
> +{
> +       /* unwind old value first: */
> +       switch (ctx->sysprof) {
> +       case 2:
> +               pm_runtime_put_autosuspend(&gpu->pdev->dev);
> +               fallthrough;
> +       case 1:
> +               refcount_dec(&gpu->sysprof_active);
> +               fallthrough;
> +       case 0:
> +               break;
> +       }
> +
> +       /* then apply new value: */

It would be safer to swap this. Otherwise a set when the values are at
"1" would drop to "zero" here and potentially trigger some glitch,
whereas incrementing one more time and then dropping the previous state
would avoid that short blip.

> +       switch (sysprof) {
> +       default:
> +               return -EINVAL;

This will become more complicated though.

> +       case 2:
> +               pm_runtime_get_sync(&gpu->pdev->dev);
> +               fallthrough;
> +       case 1:
> +               refcount_inc(&gpu->sysprof_active);
> +               fallthrough;
> +       case 0:
> +               break;
> +       }
> +
> +       ctx->sysprof = sysprof;
> +
> +       return 0;
> +}