ASoC: SOF: Intel: hda-dsp: Make sure that no irq handler is pending before suspend

In the existing IPC support, the reply to each IPC message is handled in
an IRQ thread. The assumption is that the IRQ thread is scheduled without
significant delays.

On an experimental (iow, buggy) kernel, the IRQ thread dealing with the
reply to the last IPC message before powering-down the DSP can be delayed
by several seconds. The IRQ thread will proceed with register accesses
after the DSP is powered-down which results in a kernel crash.

While the bug which causes the delay is not in the audio stack, we must
handle such cases with defensive programming to avoid such crashes.

Call synchronize_irq() before proceeding to power down the DSP to make
sure that no irq thread is pending execution.

Closes: https://github.com/thesofproject/linux/issues/4608
Reviewed-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20231012191850.147140-3-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>

diff --git a/sound/soc/sof/intel/hda-dsp.c b/sound/soc/sof/intel/hda-dsp.c
index 44f39a520bb39cc44e068231a817846c77b8b0a9..2445ae7f6b2e924fa0f192b2322dbd77eeee1b29 100644 (file)
--- a/sound/soc/sof/intel/hda-dsp.c
+++ b/sound/soc/sof/intel/hda-dsp.c
@@ -699,6 +699,9 @@ static int hda_suspend(struct snd_sof_dev *sdev, bool runtime_suspend)
        if (ret < 0)
                return ret;
 
+       /* make sure that no irq handler is pending before shutdown */
+       synchronize_irq(sdev->ipc_irq);
+
        hda_codec_jack_wake_enable(sdev, runtime_suspend);
 
        /* power down all hda links */