Re: [PATCH] scsi: core: ufs: Fix a hang in the error handler

["Peter Wang (王信友)" <peter.wang@mediatek.com>]

On Tue, 2025-05-27 at 15:01 -0700, Bart Van Assche wrote:
> 
> 
> Hi Peter,
> 
> I have not yet tried to perform a full root-cause analysis. I decided
> to share this patch since this issue has been observed not only by
> the
> author of the above patch but also by my colleagues.
> 
> Regarding your question, are you referring to the
> ufshcd_link_recovery()
> call in ufshcd_eh_host_reset_handler()?
> ufshcd_eh_host_reset_handler()
> is not the only function that can trigger the UFS error handler.
> There
> are several ufshcd_schedule_eh_work() calls outside the
> ufshcd_eh_host_reset_handler() function.
> 
> Bart.


Hi Bart,


Yes, in my opinion, if an error occurs during runtime suspend,
it can trigger ufshcd_err_handler and then directly return busy 
to break this suspend attempt.
But for a runtime resume error, can only use ufshcd_link_recovery; 
otherwise, ufshcd_err_handler will still get stuck waiting 
for runtime PM to become active.
The previous patch was specifically for this resume case.
https://patchwork.kernel.org/project/linux-scsi/patch/20230927033557.13801-1-peter.wang@mediatek.com/

And for this patch, I feel that it's like what you mentioned, other 
functions triggered the UFS error handler, such as a UIC error. 
Since the SSU can be sent normally, it means it's an error that 
the hardware can recover from automatically. After the hardware 
automatically recovers, there's no issue during the runtime 
suspend process, and the suspend isn't break.
Or, if there's no issue during the runtime resume process, 
ufshcd_link_recovery isn't used directly either.

But this kind of patch assumes that the hardware can recover 
automatically. If the hardware can't recover, ufshcd_err_handler 
might still get stuck waiting for PM resume.
In other words, this patch solves part of the problem, but it may 
not be able to address all issues that occur during suspend/resume.
However, before we can identify the root cause, it seems there 
isn't a better solution for now. What do you think?

Thanks.
Peter