linux-kernel - Re: [EXT] [PATCH v1 5/5] scsi: ufs: Complete pending requests in host reset and restore path

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 14 Nov 2019 09:18:16 +0800
From:   cang@...eaurora.org
To:     "Bean Huo (beanhuo)" <beanhuo@...ron.com>
Cc:     asutoshd@...eaurora.org, nguyenb@...eaurora.org,
        rnayak@...eaurora.org, linux-scsi@...r.kernel.org,
        kernel-team@...roid.com, saravanak@...gle.com, salyzyn@...gle.com,
        Alim Akhtar <alim.akhtar@...sung.com>,
        Avri Altman <avri.altman@....com>,
        Pedro Sousa <pedrom.sousa@...opsys.com>,
        "James E.J. Bottomley" <jejb@...ux.ibm.com>,
        "Martin K. Petersen" <martin.petersen@...cle.com>,
        Stanley Chu <stanley.chu@...iatek.com>,
        Subhash Jadavani <subhashj@...eaurora.org>,
        Tomas Winkler <tomas.winkler@...el.com>,
        open list <linux-kernel@...r.kernel.org>,
        Can Guo <cang@...eaurora.org>
Subject: Re: [EXT] [PATCH v1 5/5] scsi: ufs: Complete pending requests in host
 reset and restore path

On 2019-11-14 09:03, cang@...eaurora.org wrote:
> On 2019-11-14 06:04, Bean Huo (beanhuo) wrote:
>>> 
>>> In UFS host reset and restore path, before probe, we stop and start 
>>> the host
>>> controller once. After host controller is stopped, the pending 
>>> requests, if any,
>>> are cleared from the doorbell, but no completion IRQ would be raised 
>>> due to the
>>> hba is stopped.
>>> These pending requests shall be completed along with the first 
>>> NOP_OUT
>>> command(as it is the first command which can raise a transfer 
>>> completion
>>> IRQ) sent during probe.
>> 
>> Hi, Can
>> I am not sure for this point, because there is HW/SW device reset
>> before or after host reset/restore.
>> Device HW/SW reset also will clear the pended tasks in device side.
>> That will be better.
>> I think Qcom platform already enabled HW reset.
>> 
>> //Bean
>> 
> 
> Hi Bean,
> 
> By pending tasks here, it means the requests sent down from scsi/block 
> layer,
> but have not yet been handled by ufs driver(cmd->scsi_done() have not
> been called yet for these requests).
> For these requests, although removed by host and UFS device in their
> HW queues(doorbell),
> UFS driver still needs to complete them from SW side(call
> cmd->scsi_done() for each one of them) to
> let upper layer know that they are finished(although not successfully)
> to avoid hitting
> timeout of these pending tasks. I hope I make my explanation clearly.
> 
> Best Regards,
> Can Guo.
> 

Hi Bean,

Just want to add up more phrases. We do have HW/SW reset.
Sorry about below lines which make you confused. Here I am just 
describing what
is like with previous code. Since these pending requests does not have
a chance to be handled in their IRQ handler after hba is stopped, and as
they have been cleared from doorbell already, then once there is an 
available
transfer completion IRQ, these requests will be handled in the IRQ 
handler,
no matter what is the transfer completion IRQ fired for. And NOP_OUT is 
just
the first command that can fire a transer completion IRQ.

Can Guo.

These pending requests shall be completed along with the first NOP_OUT
command(as it is the first command which can raise a transfer completion
IRQ) sent during probe.

>>> Since the OCSs of these pending requests are not SUCCESS(because they 
>>> are not
>>> yet literally finished), their UPIUs shall be dumped. When there are 
>>> multiple
>>> pending requests, the UPIU dump can be overwhelming and may lead to 
>>> stability
>>> issues because it is in atomic context.
>>> Therefore, before probe, complete these pending requests right after 
>>> host
>>> controller is stopped.