ALSA: hda - Workaround for unbalanced i915 power refcount by concurrent probe
authorTakashi Iwai <tiwai@suse.de>
Sat, 19 Mar 2016 09:40:21 +0000 (10:40 +0100)
committerTakashi Iwai <tiwai@suse.de>
Sun, 20 Mar 2016 07:38:32 +0000 (08:38 +0100)
The recent addition of on-demand i915 audio component binding in the
codec driver seems leading to the unbalanced i915 power refcount,
according to Intel CI tests.  Typically, it gets a kernel WARNING
like:
  WARNING: CPU: 3 PID: 173 at sound/hda/hdac_i915.c:91 snd_hdac_display_power+0xf1/0x110 [snd_hda_core]()
  Call Trace:
   [<ffffffff813fef15>] dump_stack+0x67/0x92
   [<ffffffff81078a21>] warn_slowpath_common+0x81/0xc0
   [<ffffffff81078b15>] warn_slowpath_null+0x15/0x20
   [<ffffffffa00f77e1>] snd_hdac_display_power+0xf1/0x110 [snd_hda_core]
   [<ffffffffa015039d>] azx_intel_link_power+0xd/0x10 [snd_hda_intel]
   [<ffffffffa011e32a>] azx_link_power+0x1a/0x30 [snd_hda_codec]
   [<ffffffffa00f21f9>] snd_hdac_link_power+0x29/0x40 [snd_hda_core]
   [<ffffffffa01192a6>] hda_codec_runtime_suspend+0x76/0xa0 [snd_hda_codec]
   .....

The scenario is like below:
- HD-audio driver and i915 driver are probed concurrently at the
  (almost) same time; HDA bus tries to bind with i915, but it fails
  because i915 initialization is still being processed.
- Later on, HD-audio probes the HDMI codec, where it again tries to
  bind with i915.  At this time, it succeeds.
- At finishing the probe of HDA, it decreases the refcount as if it
  were already bound at the bus probe, since the component is bound
  now.  This triggers a kernel WARNING due to the unbalance.

As a workaround, in this patch, we just disable the on-demand i915
component binding in the codec driver.  This essentially reverts back
to the state of 4.4 kernel.

We know that this is no real solution, but it's a minimalistic simple
change that can be applied to 4.5.x kernel as stable.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94566
Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org> # v4.5
Signed-off-by: Takashi Iwai <tiwai@suse.de>
sound/pci/hda/patch_hdmi.c

index 56d3575ee6cc0d4e9d599caacb926b49445f3644..7ae614d27954f83ed99711db6f31aaf174261ffb 100644 (file)
@@ -2258,9 +2258,15 @@ static int patch_generic_hdmi(struct hda_codec *codec)
        /* Try to bind with i915 for Intel HSW+ codecs (if not done yet) */
        if ((codec->core.vendor_id >> 16) == 0x8086 &&
            is_haswell_plus(codec)) {
+#if 0
+               /* on-demand binding leads to an unbalanced refcount when
+                * both i915 and hda drivers are probed concurrently;
+                * disabled temporarily for now
+                */
                if (!codec->bus->core.audio_component)
                        if (!snd_hdac_i915_init(&codec->bus->core))
                                spec->i915_bound = true;
+#endif
                /* use i915 audio component notifier for hotplug */
                if (codec->bus->core.audio_component)
                        spec->use_acomp_notifier = true;