ALSA: hda: fix general protection fault in azx_runtime_idle
authorKai Vehmanen <kai.vehmanen@linux.intel.com>
Wed, 10 Nov 2021 21:03:07 +0000 (23:03 +0200)
committerTakashi Iwai <tiwai@suse.de>
Wed, 10 Nov 2021 21:48:31 +0000 (22:48 +0100)
commite81478bbe7a1a062eeb5cd97cf00152a13ab8652
treef7f23da18a72eb605c8d4017450fbd70e31b51c9
parent39173303c83859723dab32c2abfb97296d6af3bf
ALSA: hda: fix general protection fault in azx_runtime_idle

Fix a corner case between PCI device driver remove callback and
runtime PM idle callback.

Following sequence of events can happen:
  - at azx_create, context is allocated with devm_kzalloc() and
    stored as pci_set_drvdata()
  - user-space requests to unbind audio driver
  - dd.c:__device_release_driver() calls PCI remove
  - pci-driver.c:pci_device_remove() calls the audio
    driver azx_remove() callback and this is completed
  - pci-driver.c:pm_runtime_put_sync() leads to a call
    to rpm_idle() which again calls azx_runtime_idle()
  - the azx context object, as returned by dev_get_drvdata(),
    is no longer valid
  -> access fault in azx_runtime_idle when executing
struct snd_card *card = dev_get_drvdata(dev);
chip = card->private_data;
if (chip->disabled || hda->init_failed)

This was discovered by i915_module_load test with 5.15.0 based
linux-next tree.

Example log caught by i915_module_load test with linux-next
https://intel-gfx-ci.01.org/tree/linux-next/

<4> [264.038232] general protection fault, probably for non-canonical address 0x6b6b6b6b6b6b73f0: 0000 [#1] PREEMPT SMP NOPTI
<4> [264.038248] CPU: 0 PID: 5374 Comm: i915_module_loa Not tainted 5.15.0-next-20211109-gc8109c2ba35e-next-20211109 #1
[...]
<4> [264.038267] RIP: 0010:azx_runtime_idle+0x12/0x60 [snd_hda_intel]
[...]
<4> [264.038355] Call Trace:
<4> [264.038359]  <TASK>
<4> [264.038362]  __rpm_callback+0x3d/0x110
<4> [264.038371]  rpm_idle+0x27f/0x380
<4> [264.038376]  __pm_runtime_idle+0x3b/0x100
<4> [264.038382]  pci_device_remove+0x6d/0xa0
<4> [264.038388]  device_release_driver_internal+0xef/0x1e0
<4> [264.038395]  unbind_store+0xeb/0x120
<4> [264.038400]  kernfs_fop_write_iter+0x11a/0x1c0

Fix the issue by setting drvdata to NULL at end of azx_remove().

Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Link: https://lore.kernel.org/r/20211110210307.1172004-1-kai.vehmanen@linux.intel.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
sound/pci/hda/hda_intel.c