Re: [PATCH v1] ufs: core: complete wl runtime resume after SCSI EH

[Bart Van Assche <bvanassche@acm.org>]

On 5/27/26 4:23 AM, Fang Hongjie(方洪杰) wrote:
> I looked closer at the direct ufshcd_link_recovery() approach from
> ufshcd_eh_timed_out(). There is one detail that I think needs to be
> handled.
> 
> For the legacy single-doorbell path, force_compl=true currently
> still calls ufshcd_transfer_req_compl(), which only completes requests for
> which the doorbell bit has already been cleared:
>    completed_reqs = ~tr_doorbell & hba->outstanding_reqs;
> 
> So if the timed-out SSU is still marked in both hba->outstanding_reqs and
> the transfer request doorbell after ufshcd_hba_stop(), it will not be
> completed by ufshcd_complete_requests(hba, true). Returning
> SCSI_EH_RESET_TIMER in that state would only restart the request timer and
> would not wake the blk_execute_rq() waiter.

Feel free to add force_compl support to the legacy single doorbell code
path.

Thanks,

Bart.