linux-kernel - Re: [PATCH] net/mlx5: Added cond

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: <5fc5d450-b77f-40eb-b15d-33939719a124@nvidia.com>
Date: Fri, 23 Aug 2024 08:16:32 +0300
From: Moshe Shemesh <moshe@...dia.com>
To: Przemek Kitszel <przemyslaw.kitszel@...el.com>, Mohamed Khalfella
	<mkhalfella@...estorage.com>
CC: Yuanyuan Zhong <yzhong@...estorage.com>, Saeed Mahameed
	<saeedm@...dia.com>, Leon Romanovsky <leon@...nel.org>, Tariq Toukan
	<tariqt@...dia.com>, "David S. Miller" <davem@...emloft.net>, Eric Dumazet
	<edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>, Paolo Abeni
	<pabeni@...hat.com>, Shay Drori <shayd@...dia.com>, <netdev@...r.kernel.org>,
	<linux-rdma@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] net/mlx5: Added cond_resched() to crdump collection



On 8/23/2024 7:08 AM, Przemek Kitszel wrote:
> 
> On 8/22/24 19:08, Mohamed Khalfella wrote:
>> On 2024-08-22 09:40:21 +0300, Moshe Shemesh wrote:
>>>
>>>
>>> On 8/21/2024 1:27 AM, Mohamed Khalfella wrote:
>>>>
>>>> On 2024-08-20 12:09:37 +0200, Przemek Kitszel wrote:
>>>>> On 8/19/24 23:42, Mohamed Khalfella wrote:
> 
> 
>>>>
>>>> Putting a cond_resched() every 16 register reads, similar to
>>>> mlx5_vsc_wait_on_flag(), should be okay. With the numbers above, this
>>>> will result in cond_resched() every ~0.56ms, which is okay IMO.
>>>
>>> Sorry for the late response, I just got back from vacation.
>>> All your measures looks right.
>>> crdump is the devlink health dump of mlx5 FW fatal health reporter.
>>> In the common case since auto-dump and auto-recover are default for this
>>> health reporter, the crdump will be collected on fatal error of the mlx5
>>> device and the recovery flow waits for it and run right after crdump
>>> finished.
>>> I agree with adding cond_resched(), but I would reduce the frequency,
>>> like once in 1024 iterations of register read.
>>> mlx5_vsc_wait_on_flag() is a bit different case as the usleep there is
>>> after 16 retries waiting for the value to change.
>>> Thanks.
>>
>> Thanks for taking a look. Once in every 1024 iterations approximately
>> translates to 35284.4ns * 1024 ~= 36.1ms, which is relatively long time
>> IMO. How about any power-of-two <= 128 (~4.51ms)?

OK
> 
> Such tune-up would matter for interactive use of the machine with very
> little cores, is that the case? Otherwise I see no point [to make it
> overall a little slower, as that is the tradeoff].

Yes, as I see it, the point here is host with very few cores.