Re: [RFC PATCH v2 0/2] Fix deadlock in ufs

From: Asutosh Das <asutoshd@codeaurora.org>
To: Avri Altman <Avri.Altman@wdc.com>
Cc: Alan Stern <stern@rowland.harvard.edu>,
	"cang@codeaurora.org" <cang@codeaurora.org>,
	"martin.petersen@oracle.com" <martin.petersen@oracle.com>,
	Bart Van Assche <bvanassche@acm.org>,
	"linux-arm-msm@vger.kernel.org" <linux-arm-msm@vger.kernel.org>,
	"linux-scsi@vger.kernel.org" <linux-scsi@vger.kernel.org>
Subject: Re: [RFC PATCH v2 0/2] Fix deadlock in ufs
Date: Mon, 8 Feb 2021 08:24:22 -0800	[thread overview]
Message-ID: <20210208162422.GL37557@stor-presley.qualcomm.com> (raw)
In-Reply-To: <DM6PR04MB65759417507BA054F8CAE86AFCB19@DM6PR04MB6575.namprd04.prod.outlook.com>

On Sat, Feb 06 2021 at 11:24 -0800, Avri Altman wrote:
>> >Regardless of your above proposal, as for the issues you were witnessing with
>> rpmb,
>> >That started this RFC in the first place, and the whole clearing uac series for
>> that matter:
>> > "In order to conduct FFU or RPMB operations, UFS needs to clear UNIT
>> ATTENTION condition...."
>> >
>> >Functionally, This was already done for the device wlun, and only added the
>> rpmb wlun.
>> >
>> >Now you are trying to solve stuff because the rpmb is not provisioned.
>> >a) There should be no relation between response to request-sense command,
>> > and if the key is programmed or not. And
>> >b) rpmb is accessed from user-space.  If it is not provisioned, it should
>> processed the error (-7)
>> >    and realize that by itself.  And also, It only makes sense that if needed,
>> >    the access sequence will include  the request-sense command.
>> >
>> >Therefore, IMHO, just reverting Randall commit (1918651f2d7e) and fixing
>> the user-space code
>> >Should suffice.
>> >
>> >Thanks,
>> >Avri
>> >
>> Hi Avri
>>
>> Thanks.
>>
>> I don't think reverting 1918651f2d7e would fix this.
>>
>> [   12.182750] ufshcd-qcom 1d84000.ufshc: ufshcd_suspend: Setting power
>> mode
>> [   12.190467] ufshcd-qcom 1d84000.ufshc: wlun_dev_clr_ua: 0 <-- uac wasn't
>> sent
>> [   12.196735] ufshcd-qcom 1d84000.ufshc: Sending ssu
>> [   12.202412] scsi 0:0:0:49488: Queue rpm status b4 ssu: 2 <- sdev_ufs_device
>> queue is suspended
>> [   12.208613] ufshcd-qcom 1d84000.ufshc: Wait for resume - <-- deadlock!
>>
>> The issue is sending any command to any lun which is registered for blk
>> runtime-pm in ufs host's suspend path would deadlock; since, it'd try to resume
>> the ufs host in the same suspend calling sequence.
>Did you managed to bisect the commit that caused the regression?
No - I didn't bisect.

>Is it in the series that Bart referred to?
>
Yes - the debug points to that.

>Thanks,
>Avri