All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: "Xu, Feifei" <Feifei.Xu@amd.com>,
	"Wang, Yang(Kevin)" <KevinYang.Wang@amd.com>,
	"amd-gfx@lists.freedesktop.org" <amd-gfx@lists.freedesktop.org>
Cc: "Koenig, Christian" <Christian.Koenig@amd.com>,
	"Zhang, Hawking" <Hawking.Zhang@amd.com>
Subject: Re: [PATCH] drm/amdgpu:Check gfx poweron when skip flush_gpu_tlb
Date: Mon, 9 Oct 2023 10:51:31 +0200	[thread overview]
Message-ID: <10a568bd-e514-4ce1-b3d5-3c7534a8f3e7@gmail.com> (raw)
In-Reply-To: <CH2PR12MB4152A458EC29B83672A600FCFECEA@CH2PR12MB4152.namprd12.prod.outlook.com>

Am 09.10.23 um 03:50 schrieb Xu, Feifei:
> [AMD Official Use Only - General]
>
> Hi,
>
>>> Based on your description, the above code should use "||" instead of "&&",
> && is to add more restriction here.  To avoid skipping necessary TLB flush by return.
> For Asics < GFX11, !adev->gfx.is_poweron is always true (this paremeter is intrudoced from GFX11), only depends on reset_domain->sem;
> For Asics = GFX11, !adev->gfx.is_poweron might be false (which gfx might already poweron in the reset), this will make the if () not ture, return will not be executed, thus flush TLB.

First of all the patch is broken because you only handle the locking, 
but not the unlocking part.

Then a TLB flush shouldn't be necessary on reset. A reset implies that 
the TLB is cleared as well.

We discussed the possibility to avoid that, but this is not supposed to 
be happening at the moment.

Regards,
Christian.

>
>>> And after merging code into one line may result in the lock not being released if the lock can be acquired success.
> If !adev->gfx.is_poweron is true, the reset_domin->sem will not be down_read_trylock, thus could avoid this deadlock.

>
> Thanks,
> Feifei
>
> -----Original Message-----
> From: Wang, Yang(Kevin) <KevinYang.Wang@amd.com>
> Sent: Sunday, October 8, 2023 9:36 PM
> To: Xu, Feifei <Feifei.Xu@amd.com>; amd-gfx@lists.freedesktop.org
> Cc: Xu, Feifei <Feifei.Xu@amd.com>; Xu, Feifei <Feifei.Xu@amd.com>; Koenig, Christian <Christian.Koenig@amd.com>; Zhang, Hawking <Hawking.Zhang@amd.com>
> Subject: RE: [PATCH] drm/amdgpu:Check gfx poweron when skip flush_gpu_tlb
>
>
> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces@lists.freedesktop.org> On Behalf Of Feifei Xu
> Sent: Sunday, October 8, 2023 6:07 PM
> To: amd-gfx@lists.freedesktop.org
> Cc: Xu, Feifei <Feifei.Xu@amd.com>; Xu, Feifei <Feifei.Xu@amd.com>; Koenig, Christian <Christian.Koenig@amd.com>; Zhang, Hawking <Hawking.Zhang@amd.com>
> Subject: [PATCH] drm/amdgpu:Check gfx poweron when skip flush_gpu_tlb
>
> To fix the gpu recovery failed on GFX11 with gfxhub pagefault, flush gpu tlb after reset on GFX11.
> Gfxhub tlb flush need check if adev->gfx.is_poweron set.
>
> Fixes: d0c860f33553 ("drm/amdgpu: rework lock handling for flush_tlb v2")
>
> Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> index 2f9bb86edd71..048d32edee88 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> @@ -611,8 +611,9 @@ void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
>                  /*
>                   * A GPU reset should flush all TLBs anyway, so no need to do
>                   * this while one is ongoing.
> +                * For GFX11, gfxhub flush check if adev->gfx.is_poweron is set.
>                   */
> -               if (!down_read_trylock(&adev->reset_domain->sem))
> +               if (!down_read_trylock(&adev->reset_domain->sem) &&
> +!adev->gfx.is_poweron)
>                          return;
>
> [Kevin]:
> Based on your description, the above code should use "||" instead of "&&",
> And after merging code into one line may result in the lock not being released if the lock can be acquired success.
>
> Best Regards,
> Kevin
>
>                  if (adev->gmc.flush_tlb_needs_extra_type_2)
> --
> 2.34.1
>


  reply	other threads:[~2023-10-09  8:51 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-10-08 10:06 [PATCH] drm/amdgpu:Check gfx poweron when skip flush_gpu_tlb Feifei Xu
2023-10-08 13:35 ` Wang, Yang(Kevin)
2023-10-09  1:50   ` Xu, Feifei
2023-10-09  8:51     ` Christian König [this message]
2023-10-10  9:34       ` Xu, Feifei
2023-10-10  9:44       ` Xu, Feifei
2023-10-10 10:07         ` Xu, Feifei
2023-10-09  8:58     ` Zhang, Hawking
2023-10-10  2:27       ` Xu, Feifei
2023-10-10  9:06         ` Christian König
2023-10-10 10:14           ` Xu, Feifei
2023-10-12  1:19             ` Xu, Feifei
2023-10-10  9:25         ` Xu, Feifei

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=10a568bd-e514-4ce1-b3d5-3c7534a8f3e7@gmail.com \
    --to=ckoenig.leichtzumerken@gmail.com \
    --cc=Christian.Koenig@amd.com \
    --cc=Feifei.Xu@amd.com \
    --cc=Hawking.Zhang@amd.com \
    --cc=KevinYang.Wang@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.