RE: [PATCH v1] scsi: ufs: core: Remove the ufshcd_release in ufshcd_err_handling_prepare

[hoyoung seo]

> -----Original Message-----
> From: Bart Van Assche <bvanassche@xxxxxxx>
> Sent: Tuesday, January 23, 2024 5:37 AM
> To: SEO HOYOUNG <hy50.seo@xxxxxxxxxxx>; linux-scsi@xxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; alim.akhtar@xxxxxxxxxxx; avri.altman@xxxxxxx;
> jejb@xxxxxxxxxxxxx; martin.petersen@xxxxxxxxxx; beanhuo@xxxxxxxxxx;
> kwangwon.min@xxxxxxxxxxx; kwmad.kim@xxxxxxxxxxx; sh425.lee@xxxxxxxxxxx;
> sc.suh@xxxxxxxxxxx; quic_nguyenb@xxxxxxxxxxx; cpgs@xxxxxxxxxxx;
> grant.jung@xxxxxxxxxxx; junwoo80.lee@xxxxxxxxxxx
> Subject: Re: [PATCH v1] scsi: ufs: core: Remove the ufshcd_release in
> ufshcd_err_handling_prepare
> 
> On 1/22/24 00:33, SEO HOYOUNG wrote:
> > If err_handler is performed in the suspend/resume situation,
> > ufs_release can be called twice and active_reqs valid can be negative.
> > This is because ufshcd_errhandling_prepare() and
> > ufshcd_err_handling_unprepare() repeatedly release calls.
> > Eventually, active_reqs have a value different from the intention.
> > To prevent this, release duplication processing was removed.
> >
> > Signed-off-by: SEO HOYOUNG <hy50.seo@xxxxxxxxxxx>
> > ---
> >   drivers/ufs/core/ufshcd.c | 1 -
> >   1 file changed, 1 deletion(-)
> >
> > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> > index 7c59d7a02243..423e83074a20 100644
> > --- a/drivers/ufs/core/ufshcd.c
> > +++ b/drivers/ufs/core/ufshcd.c
> > @@ -6351,7 +6351,6 @@ static void ufshcd_err_handling_prepare(struct
> ufs_hba *hba)
> >   		ufshcd_hold(hba);
> >   		if (!ufshcd_is_clkgating_allowed(hba))
> >   			ufshcd_setup_clocks(hba, true);
> > -		ufshcd_release(hba);
> >   		pm_op = hba->is_sys_suspended ? UFS_SYSTEM_PM :
> UFS_RUNTIME_PM;
> >   		ufshcd_vops_resume(hba, pm_op);
> >   	} else {
> 
> I think that the above ufshcd_release() call pairs with the ufshcd_hold()
> call three lines above it and hence that removing that call would be wrong.
> 
> Thanks,
> 
> Bart.

Hi,

It was a different when I tested it.
If __ufshcd_wl_resume() is called active_reqs is 1.
Because ufshcd_hold() is called in __ufshcd_wl_suspend().
If occurred hibern8_exit failed in __ufschd_wl_resume(), ufshcd_release()
is called in the :out syntax, and active_reqs becomes 0.
After that, active_reqs becomes 0 because ufshcd_hold() is called 
from ufshcd_err_handling_repare()and ufshcd_release() is called again while
err_handler is operating.
When err_handler is completed, active_reqs becomes negative because 
ufshcd_release() is called again in ufshcd_err_handling_unprepare().
I tested it while printing the log, and if I misanalyzed it, let me know.

Thanks,

SEO.