Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Raag Jadav <raag.jadav@intel.com>
To: "Tauro, Riana" <riana.tauro@intel.com>
Cc: intel-xe@lists.freedesktop.org, anshuman.gupta@intel.com,
	rodrigo.vivi@intel.com, aravind.iddamsetty@linux.intel.com,
	mallesh.koujalagi@intel.com, badal.nilawar@intel.com
Subject: Re: [PATCH v4 3/8] drm/xe/survivability: Guard runtime survivability enable against unbind
Date: Wed, 11 Nov 2026 08:09:16 +0100	[thread overview]
Message-ID: <avQUnHig9Ign-njH@black.igk.intel.com> (raw)
In-Reply-To: <524ae4f5-d484-4af4-a7ac-68bdaef970c5@intel.com>

On Wed, Nov 11, 2026 at 11:20:02AM +0530, Tauro, Riana wrote:
> On 11-11-2026 11:09, Raag Jadav wrote:
> > On Wed, Nov 11, 2026 at 10:13:44AM +0530, Tauro, Riana wrote:
> > > On 10-11-2026 18:21, Raag Jadav wrote:
> > > > On Fri, Nov 06, 2026 at 12:49:48PM +0530, Riana Tauro wrote:
> > > > > Runtime surivability mode creates sysfs and declares the device as
> > > > > wedged in a worker thread. This can race with unbind.
> > > > > 
> > > > > During unbind, devres actions such as disabling the worker run as part of
> > > > > teardown. A runtime survivability enable that is still in flight or gets
> > > > > scheduled in that window can create the survivability sysfs entries while
> > > > > the device is being removed. This is incorrect, as runtime survivability mode
> > > > > must not be entered once during devres teardown.
> > > > > 
> > > > > Use the DRM device enter/exit protection so that the enable path
> > > > > synchronizes with device unplug and returns early if the device is
> > > > > already being unbound.
> > > > I'm a bit lost here. Doesn't disable_work_sync() already guarantee this?
> > > If the worker is already queued and there is a unbind in progress. The
> > > disable work sync from devm
> > > action waits for the worker to complete. This causes the device to enter
> > > survivability mode during unbind
> > > which is not intended.
> > Wouldn't the survivability stuff still land on devres and get cleaned
> > up anyway?
> 
> devres_release_all running as part of unbind copies all the devres nodes
> into a todo list and then
> releases them. If during release, we add a new devres that will not be
> cleared during unbind
> which can cause stale sysfs.

Hm, I thought all the devm actions are protected by devres_lock but
you're right.

A more robust fix would be to disable all the workers in ->remove()
callback itself and avoid such races to begin with, but I'll leave it
upto you.

Raag

  reply	other threads:[~2026-11-11  7:09 UTC|newest]

Thread overview: 32+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-06  7:19 [PATCH v4 0/8] Handle non-fatal errors routed via IRQ Riana Tauro
2026-11-06  7:19 ` [PATCH v4 1/8] drm/xe/hw_error: Reuse CSC worker for generic error handling Riana Tauro
2026-11-06  7:19 ` [PATCH v4 2/8] drm/xe/hw_error: Initialize hw_error_work per tile Riana Tauro
2026-11-10  5:48   ` Mallesh, Koujalagi
2026-11-10 12:40   ` Raag Jadav
2026-11-11  4:45     ` Tauro, Riana
2026-11-11  5:40       ` Raag Jadav
2026-11-11  5:51         ` Tauro, Riana
2026-11-06  7:19 ` [PATCH v4 3/8] drm/xe/survivability: Guard runtime survivability enable against unbind Riana Tauro
2026-11-10  6:19   ` Mallesh, Koujalagi
2026-11-10 12:51   ` Raag Jadav
2026-11-11  4:43     ` Tauro, Riana
2026-11-11  5:39       ` Raag Jadav
2026-11-11  5:50         ` Tauro, Riana
2026-11-11  7:09           ` Raag Jadav [this message]
2026-11-11  7:17             ` Tauro, Riana
2026-11-06  7:19 ` [PATCH v4 4/8] drm/xe/ras: Treat non-fatal PCIe errors as recovered Riana Tauro
2026-11-06  7:19 ` [PATCH v4 5/8] drm/xe/pci: Add upstream switch port lookup helper Riana Tauro
2026-11-06  7:19 ` [PATCH v4 6/8] drm/xe/xe_ras: Add a flag for uncorrectable internal error Riana Tauro
2026-11-10 13:07   ` Raag Jadav
2026-11-11  4:47     ` Tauro, Riana
2026-11-06  7:19 ` [PATCH v4 7/8] drm/xe: Add support for wa_14028434957 Riana Tauro
2026-11-10  6:41   ` Mallesh, Koujalagi
2026-11-10 13:18   ` Raag Jadav
2026-11-11  4:50     ` Tauro, Riana
2026-11-11  6:15       ` Raag Jadav
2026-11-11  6:22         ` Tauro, Riana
2026-11-06  7:19 ` [PATCH v4 8/8] drm/xe/xe_ras: Skip double-bit ECC page offlining on AER path Riana Tauro
2026-11-06  7:27 ` ✗ CI.checkpatch: warning for Handle non-fatal errors routed via IRQ (rev4) Patchwork
2026-11-06  7:29 ` ✓ CI.KUnit: success " Patchwork
2026-11-06  8:13 ` ✓ Xe.CI.BAT: " Patchwork
2026-11-07  7:38 ` ✗ Xe.CI.FULL: failure " Patchwork

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=avQUnHig9Ign-njH@black.igk.intel.com \
    --to=raag.jadav@intel.com \
    --cc=anshuman.gupta@intel.com \
    --cc=aravind.iddamsetty@linux.intel.com \
    --cc=badal.nilawar@intel.com \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=mallesh.koujalagi@intel.com \
    --cc=riana.tauro@intel.com \
    --cc=rodrigo.vivi@intel.com \
    /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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox