[BUG] IO transfer timed out in kmod test on TGLU-SDW

[keqiaozhang]

Describe the bug
This error happens when removing audio modules, but it won't impact audio functionality and sof can work after reloading.
Daily report ID:1867
Device:TGLU_SKU0A32_SDCA

To Reproduce
$ /sof-test/test-case/check-kmod-load-unload.sh

Reproduction Rate
100%.

Expected behavior
no errors during the kmod test

Impact
no functionality impact

Environment
Kernel Branch: topic/sof-dev
Kernel Commit: ba1a2f8-1
SOF Branch: master
SOF Commit: 0b2876b

Screenshots or console output

kernel: [ 6579.975684] sof-audio-pci 0000:00:1f.3: ipc tx succeeded: 0x50010000: GLB_COMP_MSG: SET_VALUE
kernel: [ 6579.975689] sof-audio-pci 0000:00:1f.3: ipc tx: 0x50010000: GLB_COMP_MSG: SET_VALUE
kernel: [ 6579.976035] sof-audio-pci 0000:00:1f.3: ipc tx succeeded: 0x50010000: GLB_COMP_MSG: SET_VALUE
kernel: [ 6579.976045] sof-audio-pci 0000:00:1f.3: ipc tx: 0x40020000: GLB_PM_MSG: CTX_RESTORE
kernel: [ 6579.976229] sof-audio-pci 0000:00:1f.3: ipc tx succeeded: 0x40020000: GLB_PM_MSG: CTX_RESTORE
kernel: [ 6579.976253] sof-audio-pci 0000:00:1f.3: ipc tx: 0x40010000: GLB_PM_MSG: CTX_SAVE
kernel: [ 6579.986613] sof-audio-pci 0000:00:1f.3: ipc tx succeeded: 0x40010000: GLB_PM_MSG: CTX_SAVE
kernel: [ 6579.987336] sof-audio-pci 0000:00:1f.3: FW Poll Status: reg=0x20140000 successful
kernel: [ 6580.547024] intel-sdw intel-sdw.0: IO transfer timed out, cmd 3 device 1 addr 5c len 1
kernel: [ 6580.547047] soundwire sdw-master-0: trf on Slave 1 failed:-110 write addr 5c count 0
kernel: [ 6581.051058] intel-sdw intel-sdw.0: IO transfer timed out, cmd 3 device 1 addr 5d len 1
kernel: [ 6581.051079] soundwire sdw-master-0: trf on Slave 1 failed:-110 write addr 5d count 0
kernel: [ 6581.051091] rt711-sdca sdw:0:25d:711:1: in rt711_sdca_jack_init disable
kernel: [ 6581.052090] sof-audio-pci 0000:00:1f.3: FW Poll Status: reg=0x9010f0f successful
kernel: [ 6581.052215] sof-audio-pci 0000:00:1f.3: FW Poll Status: reg=0xf0f successful
kernel: [ 6581.052223] sof-audio-pci 0000:00:1f.3: DSP core(s) enabled? 0 : core_mask 1

[bardliao]

@shumingfan @plbossart This issue was fixed by #2640. We use regcache_cache_only to avoid to access SDW_SCP_SDCA_INTMASK1 and SDW_SCP_SDCA_INTMASK2. However, after #2642 uses sdw_write_no_pm directly, regcache_cache_only can't protect it anymore. What is your idea here?

[shumingfan]

@bardliao @plbossart Could we remove the sdw_write_no_pm when the jack disable?
The SDW_SCP_SDCA_INTMASK1/2 will be cleared by any reset anyway.

@@ -477,10 +477,6 @@ static void rt711_sdca_jack_init(struct rt711_sdca_priv *rt711)
 rt711_sdca_index_update_bits(rt711, RT711_VENDOR_HDA_CTL,
 RT711_GE_MODE_RELATED_CTL, 0x0c00, 0x0000);

- /* set SCP_SDCA_IntMask1[0]=0 */
- sdw_write_no_pm(rt711->slave, SDW_SCP_SDCA_INTMASK1, 0);
- /* set SCP_SDCA_IntMask2[0]=0 */
- sdw_write_no_pm(rt711->slave, SDW_SCP_SDCA_INTMASK2, 0);
 dev_dbg(&rt711->slave->dev, "in %s disable\n", __func__);
 }

[plbossart]

I don't understand the issue, did we fix the problem or not? It seems we have the same report but a different error? Completely lost on what the problem is.

[bardliao]

@plbossart The issue is that snd_soc_component_set_jack(component, NULL, NULL) will be called when asoc component is removed. SoundWire bus is suspended at that moment, and we use sdw_write_no_pm() which was regmap_write() to set the SDW_SCP_SDCA_INTMASK in rt711's set_jack callback function. In other words, we call sdw_write_no_pm() when sdw bus is suspended which result in IO transfer timed out.
The issue was fixed by setting regcache_cache_only() before calling regmap_write(), but now it is back due to we use sdw_write_no_pm() instead.

[keqiaozhang]

This issue is fixed by #2728