On 2/6/23 18:42, David Rau wrote:
-----Original Message----- From: Guenter Roeck groeck7@gmail.com On Behalf Of Guenter Roeck Sent: Monday, February 6, 2023 22:05 To: David Rau david.rau.zg@renesas.com; Mark Brown broonie@kernel.org Cc: perex@perex.cz; lgirdwood@gmail.com; tiwai@suse.com; support.opensource@diasemi.com; alsa-devel@alsa-project.org; linux-kernel@vger.kernel.org Subject: Re: [PATCH] ASoC: da7219: Fix pole orientation detection on OMTP headsets when playing music
On 2/5/23 21:38, David Rau wrote:
-----Original Message----- From: Guenter Roeck groeck7@gmail.com On Behalf Of Guenter Roeck Sent: Saturday, February 4, 2023 23:42 To: Mark Brown broonie@kernel.org Cc: David Rau david.rau.zg@renesas.com; perex@perex.cz; lgirdwood@gmail.com; tiwai@suse.com; support.opensource@diasemi.com; alsa-devel@alsa-project.org; linux-kernel@vger.kernel.org Subject: Re: [PATCH] ASoC: da7219: Fix pole orientation detection on OMTP headsets when playing music
On Thu, Feb 02, 2023 at 07:36:42PM +0000, Mark Brown wrote:
they have the potential to actually lock up are the cancel_work_sync() calls but they were unchanged and the backtrace you showed was showing the thread in the msleep(). My guess would be that you've got systems where there are very frequent jack detection events (potentiallly with broken accessories, or possibly due to the ground switch putting things into the wrong priority) and that the interrupt is firing again as soon as the thread unmasks the primary interrupt which means it never actually stops running.
That is what I strongly suspect is happening. I don't know why exactly the interrupt is firing continuously, but the hang is always in msleep(). One possibility might be that the event is actually a disconnect event, and that enabling and immediately disabling the ground switch causes another interrupt, which is then handled immediately, causing the hang.
Could be. I'd be willing to guess that it's not just one event but rather a stream of events of some kind. Possibly if it's due to the ground switch it's spuriously detecting a constant stream of button presses for the affected systems, which don't produce any UI visible result which would cause users to pull the accessory for whatever reason? Whatever's going on I bet it's broken accessories triggering it.
That seems to be unlikely. The average number of crashes per affected system is 1.92, which points to something the users are doing and less to a broken accessory.
We do observe crashes due to broken accessories, but in those cases the number of crashes per system tends to be much > higher.
Anyway, below is a patch with a possible fix. Of course, I still don't know what the patch originally tried to fix, so it might not do much if anything good.
I added the software debouncing before insertion task to ensue the better compatibility of OMTP Jack.
For example, it keeps button detection in the interrupt handler to avoid dropping button events, so if spurious button detection as you suspected is indeed (part of) the problem we might still see a large number of interrupts.
Guenter
Thanks a lot for your big efforts to implement the temporary fix and verifications. Would you please let me know the average number of crashes per affected system if you rollback to the pervious fix? Ref: https://jpn01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit. kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Ftorvalds%2Flinux.git%2 Fcommit%2Fsound%2Fsoc%2Fcodecs%3Fid%3D2d969e8f35b1849a43156029a7a6e294 3b89d0c0&data=05%7C01%7Cdavid.rau.zg%40renesas.com%7Cae6910f8ff4e4e299 bc408db084b1a2a%7C53d82571da1947e49cb4625a166a4a2a%7C0%7C0%7C638112890 873388020%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIi LCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=8KgHP%2FOD%2BTDcr rUVSATFkDCDDmhiCu7d5%2FKhyOszThA%3D&reserved=0 https://jpn01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit. kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Ftorvalds%2Flinux.git%2 Fcommit%2Fsound%2Fsoc%2Fcodecs%3Fid%3D06f5882122e3faa183d76c4ec2c92f4c 38e2c7bb&data=05%7C01%7Cdavid.rau.zg%40renesas.com%7Cae6910f8ff4e4e299 bc408db084b1a2a%7C53d82571da1947e49cb4625a166a4a2a%7C0%7C0%7C638112890 873388020%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIi LCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=WosfvANk0YxeJD5PG %2FnAuAWVqt7m4U3mMaYXefLLdS4%3D&reserved=0
You mean just keep the above two patches and revert 969357ec94e6 ? Sure, I can do that, but feedback from the field would take some 2-3 months. Is that what you recommend to do for now ?
Thanks, Guenter
Thanks for the feedback. What I mean is just do rollback to remove the "sleep" patch I did in your repository.
After the rollback, could you please let me know the average number of crashes per affected system with the same test conditions? Will it still take some 2-3 months?
Yes, due to our rollout schedules. Those are crashes observed in the field, after all.
Guenter