linux-kernel - Re: [PATCH] scsi: ufs: core: move some irq handling back to hardirq (with time limit)

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]

Message-ID: <754d5c83-ba83-40d2-9309-8eafcb885b9f@linaro.org>
Date: Mon, 28 Jul 2025 16:41:41 +0200
From: Neil Armstrong <neil.armstrong@...aro.org>
To: Manivannan Sadhasivam <mani@...nel.org>,
 André Draszik <andre.draszik@...aro.org>
Cc: Alim Akhtar <alim.akhtar@...sung.com>, Avri Altman <avri.altman@....com>,
 Bart Van Assche <bvanassche@....org>,
 "James E.J. Bottomley" <James.Bottomley@...senpartnership.com>,
 "Martin K. Petersen" <martin.petersen@...cle.com>,
 Peter Griffin <peter.griffin@...aro.org>,
 Tudor Ambarus <tudor.ambarus@...aro.org>,
 Will McVicker <willmcvicker@...gle.com>, kernel-team@...roid.com,
 linux-arm-msm@...r.kernel.org, linux-scsi@...r.kernel.org,
 linux-kernel@...r.kernel.org, stable@...r.kernel.org,
 quic_nitirawa@...cinc.com
Subject: Re: [PATCH] scsi: ufs: core: move some irq handling back to hardirq
 (with time limit)

On 28/07/2025 16:39, Manivannan Sadhasivam wrote:
> On Mon, Jul 28, 2025 at 08:06:21PM GMT, Manivannan Sadhasivam wrote:
>> + Nitin
>>
> 
> Really added Nitin now.

BTW what about MCQ on SM8650 ? it's probably the real fix here...

Neil

> 
>> On Thu, Jul 24, 2025 at 02:38:30PM GMT, André Draszik wrote:
>>> On Thu, 2025-07-24 at 13:54 +0200, Neil Armstrong wrote:
>>>> On 24/07/2025 13:44, André Draszik wrote:
>>>>> On Thu, 2025-07-24 at 10:54 +0100, André Draszik wrote:
>>>>>> fio results on Pixel 6:
>>>>>>     read / 1 job     original    after    this commit
>>>>>>       min IOPS        4,653.60   2,704.40    3,902.80
>>>>>>       max IOPS        6,151.80   4,847.60    6,103.40
>>>>>>       avg IOPS        5,488.82   4,226.61    5,314.89
>>>>>>       cpu % usr           1.85       1.72        1.97
>>>>>>       cpu % sys          32.46      28.88       33.29
>>>>>>       bw MB/s            21.46      16.50       20.76
>>>>>>
>>>>>>     read / 8 jobs    original    after    this commit
>>>>>>       min IOPS       18,207.80  11,323.00   17,911.80
>>>>>>       max IOPS       25,535.80  14,477.40   24,373.60
>>>>>>       avg IOPS       22,529.93  13,325.59   21,868.85
>>>>>>       cpu % usr           1.70       1.41        1.67
>>>>>>       cpu % sys          27.89      21.85       27.23
>>>>>>       bw MB/s            88.10      52.10       84.48
>>>>>>
>>>>>>     write / 1 job    original    after    this commit
>>>>>>       min IOPS        6,524.20   3,136.00    5,988.40
>>>>>>       max IOPS        7,303.60   5,144.40    7,232.40
>>>>>>       avg IOPS        7,169.80   4,608.29    7,014.66
>>>>>>       cpu % usr           2.29       2.34        2.23
>>>>>>       cpu % sys          41.91      39.34       42.48
>>>>>>       bw MB/s            28.02      18.00       27.42
>>>>>>
>>>>>>     write / 8 jobs   original    after    this commit
>>>>>>       min IOPS       12,685.40  13,783.00   12,622.40
>>>>>>       max IOPS       30,814.20  22,122.00   29,636.00
>>>>>>       avg IOPS       21,539.04  18,552.63   21,134.65
>>>>>>       cpu % usr           2.08       1.61        2.07
>>>>>>       cpu % sys          30.86      23.88       30.64
>>>>>>       bw MB/s            84.18      72.54       82.62
>>>>>
>>>>> Given the severe performance drop introduced by the culprit
>>>>> commit, it might make sense to instead just revert it for
>>>>> 6.16 now, while this patch here can mature and be properly
>>>>> reviewed. At least then 6.16 will not have any performance
>>>>> regression of such a scale.
>>>>
>>>> The original change was designed to stop the interrupt handler
>>>> to starve the system and create display artifact and cause
>>>> timeouts on system controller submission. While imperfect,
>>>> it would require some fine tuning for smaller controllers
>>>> like on the Pixel 6 that when less queues.
>>>
>>> Well, the patch has solved one problem by creating another problem.
>>> I don't think that's how things are normally done. A 40% bandwidth
>>> and IOPS drop is not negligible.
>>>
>>> And while I am referencing Pixel 6 above as it's the only device
>>> I have available to test, I suspect all < v4 controllers / devices
>>> are affected in a similar way, given the nature of the change.
>>>
>>
>> IMO we should just revert the offending commit for 6.16 and see how to properly
>> implement it in the next release. Even with this series, we are not on par with
>> the original IOPS, which is bad for everyone.
>>
>> - Mani
>>
>> -- 
>> மணிவண்ணன் சதாசிவம்
>