[PATCH] ALSA: hda/hdmi: fix incorrect locking in hdmi_pcm_close
A race exists between closing a PCM and update of ELD data. In hdmi_pcm_close(), hinfo->nid value is modified without taking spec->pcm_lock. If this happens concurrently while processing an ELD update in hdmi_pcm_setup_pin(), converter assignment may be done incorrectly.
This bug was found by hitting a WARN_ON in snd_hda_spdif_ctls_assign() in a HDMI receiver connection stress test:
[2739.684569] WARNING: CPU: 5 PID: 2090 at sound/pci/hda/patch_hdmi.c:1898 check_non_pcm_per_cvt+0x41/0x50 [snd_hda_codec_hdmi] ... [2739.684707] Call Trace: [2739.684720] update_eld+0x121/0x5a0 [snd_hda_codec_hdmi] [2739.684736] hdmi_present_sense+0x21e/0x3b0 [snd_hda_codec_hdmi] [2739.684750] check_presence_and_report+0x81/0xd0 [snd_hda_codec_hdmi] [2739.684842] intel_audio_codec_enable+0x122/0x190 [i915]
Fixes: 42b2987079ec ("ALSA: hda - hdmi playback without monitor in dynamic pcm bind mode") Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com --- sound/pci/hda/patch_hdmi.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-)
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index 0ffbfcb91256..ccd1df059654 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -2046,22 +2046,25 @@ static int hdmi_pcm_close(struct hda_pcm_stream *hinfo, int pinctl; int err = 0;
+ mutex_lock(&spec->pcm_lock); if (hinfo->nid) { pcm_idx = hinfo_to_pcm_index(codec, hinfo); - if (snd_BUG_ON(pcm_idx < 0)) - return -EINVAL; + if (snd_BUG_ON(pcm_idx < 0)) { + err = -EINVAL; + goto unlock; + } cvt_idx = cvt_nid_to_cvt_index(codec, hinfo->nid); - if (snd_BUG_ON(cvt_idx < 0)) - return -EINVAL; + if (snd_BUG_ON(cvt_idx < 0)) { + err = -EINVAL; + goto unlock; + } per_cvt = get_cvt(spec, cvt_idx); - snd_BUG_ON(!per_cvt->assigned); per_cvt->assigned = 0; hinfo->nid = 0;
azx_stream(get_azx_dev(substream))->stripe = 0;
- mutex_lock(&spec->pcm_lock); snd_hda_spdif_ctls_unassign(codec, pcm_idx); clear_bit(pcm_idx, &spec->pcm_in_use); pin_idx = hinfo_to_pin_index(codec, hinfo); @@ -2091,10 +2094,11 @@ static int hdmi_pcm_close(struct hda_pcm_stream *hinfo, per_pin->setup = false; per_pin->channels = 0; mutex_unlock(&per_pin->lock); - unlock: - mutex_unlock(&spec->pcm_lock); }
+unlock: + mutex_unlock(&spec->pcm_lock); + return err; }
On Tue, 13 Oct 2020 17:26:28 +0200, Kai Vehmanen wrote:
A race exists between closing a PCM and update of ELD data. In hdmi_pcm_close(), hinfo->nid value is modified without taking spec->pcm_lock. If this happens concurrently while processing an ELD update in hdmi_pcm_setup_pin(), converter assignment may be done incorrectly.
This bug was found by hitting a WARN_ON in snd_hda_spdif_ctls_assign() in a HDMI receiver connection stress test:
[2739.684569] WARNING: CPU: 5 PID: 2090 at sound/pci/hda/patch_hdmi.c:1898 check_non_pcm_per_cvt+0x41/0x50 [snd_hda_codec_hdmi] ... [2739.684707] Call Trace: [2739.684720] update_eld+0x121/0x5a0 [snd_hda_codec_hdmi] [2739.684736] hdmi_present_sense+0x21e/0x3b0 [snd_hda_codec_hdmi] [2739.684750] check_presence_and_report+0x81/0xd0 [snd_hda_codec_hdmi] [2739.684842] intel_audio_codec_enable+0x122/0x190 [i915]
Fixes: 42b2987079ec ("ALSA: hda - hdmi playback without monitor in dynamic pcm bind mode") Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com
Applied now. Thanks.
Takashi
participants (2)
-
Kai Vehmanen
-
Takashi Iwai