All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Kim, Jonathan" <Jonathan.Kim@amd.com>
To: "Yang, Philip" <Philip.Yang@amd.com>,
	"Sierra Guiza, Alejandro (Alex)" <Alex.Sierra@amd.com>,
	"amd-gfx@lists.freedesktop.org" <amd-gfx@lists.freedesktop.org>
Subject: RE: [PATCH] drm/amdkfd: drop process ref count when xnack disable
Date: Wed, 1 Sep 2021 16:59:18 +0000	[thread overview]
Message-ID: <DM5PR12MB4680F859F8CA637B9BB9B35185CD9@DM5PR12MB4680.namprd12.prod.outlook.com> (raw)
In-Reply-To: <813b0a09-a81e-3d0d-0342-eaad81f67f8a@amd.com>

[-- Attachment #1: Type: text/plain, Size: 3333 bytes --]

[Public]

I wouldn’t know if it was another bug elsewhere.
From what I was seeing, the leak was coming from !p->xnack_enable on the svm_range_restore_pages call.

If it helps, I saw this on Aldebaran where a shader does some bad memory access on purpose on a debugged ptraced child process.
The vm fault prompt pops up in dmesgs and a stale KFD process appends per run without this fix.
I’m just assuming at this point that the IV retry bit is set but I never confirmed that.

Thanks,

Jon
From: Yang, Philip <Philip.Yang@amd.com>
Sent: Wednesday, September 1, 2021 12:30 PM
To: Kim, Jonathan <Jonathan.Kim@amd.com>; Yang, Philip <Philip.Yang@amd.com>; Sierra Guiza, Alejandro (Alex) <Alex.Sierra@amd.com>; amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH] drm/amdkfd: drop process ref count when xnack disable



On 2021-09-01 9:45 a.m., Kim, Jonathan wrote:

[AMD Official Use Only]

We were seeing process leaks on a couple of machines running certain tests that triggered vm faults on purpose.
I think svm_range_restore_pages gets called unconditionally on vm fault handling (unless the retry interrupt payload bit is supposed to be clear with xnack off)?


yes, with xnack off, sh_mem_config retry should be off, retry bit is supposed to be clear in fault interrupt vector, we should not try to recover vm fault, just report the vm fault back to application and evict user queues. Maybe it is another bug cause p->xnack_enabled and sh_mem_config retry mismatch under specific condition?

Regards,

Philip
Either way, this patch prevents the process leaks we seeing and is also:
Reviewed-by: Jonathan Kim <jonathan.kim@amd.com><mailto:jonathan.kim@amd.com>

Thanks,

Jon


From: amd-gfx <amd-gfx-bounces@lists.freedesktop.org><mailto:amd-gfx-bounces@lists.freedesktop.org> On Behalf Of philip yang
Sent: Wednesday, September 1, 2021 7:30 AM
To: Sierra Guiza, Alejandro (Alex) <Alex.Sierra@amd.com><mailto:Alex.Sierra@amd.com>; amd-gfx@lists.freedesktop.org<mailto:amd-gfx@lists.freedesktop.org>
Subject: Re: [PATCH] drm/amdkfd: drop process ref count when xnack disable

[CAUTION: External Email]


On 2021-08-31 10:41 p.m., Alex Sierra wrote:

During svm restore pages interrupt handler, kfd_process ref count was

never dropped when xnack was disabled. Therefore, the object was never

released.

Good catch, but if xnack is off, we should not get here to recover fault.

The fix looks good to me.

Reviewed-by: Philip Yang <philip.yang@amd.com><mailto:philip.yang@amd.com>



Signed-off-by: Alex Sierra <alex.sierra@amd.com><mailto:alex.sierra@amd.com>

---

 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 3 ++-

 1 file changed, 2 insertions(+), 1 deletion(-)



diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c

index 8f9b5b53dab5..110c46cd7fac 100644

--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c

+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c

@@ -2484,7 +2484,8 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,

         }

         if (!p->xnack_enabled) {

                pr_debug("XNACK not enabled for pasid 0x%x\n", pasid);

-               return -EFAULT;

+               r = -EFAULT;

+               goto out;

         }

         svms = &p->svms;



[-- Attachment #2: Type: text/html, Size: 8742 bytes --]

  reply	other threads:[~2021-09-01 16:59 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-01  2:41 [PATCH] drm/amdkfd: drop process ref count when xnack disable Alex Sierra
2021-09-01 11:29 ` philip yang
2021-09-01 13:45   ` Kim, Jonathan
2021-09-01 16:29     ` philip yang
2021-09-01 16:59       ` Kim, Jonathan [this message]
2021-09-01 17:54         ` Felix Kuehling
2021-09-01 19:43           ` Felix Kuehling

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=DM5PR12MB4680F859F8CA637B9BB9B35185CD9@DM5PR12MB4680.namprd12.prod.outlook.com \
    --to=jonathan.kim@amd.com \
    --cc=Alex.Sierra@amd.com \
    --cc=Philip.Yang@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.