linux-kernel - Re: [PATCH v4 6/6] media: uvcvideo: Move decode processing to process context

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:   Wed, 8 Aug 2018 12:50:37 +0900
From:   Tomasz Figa <tfiga@...omium.org>
To:     Laurent Pinchart <laurent.pinchart@...asonboard.com>
Cc:     Kieran Bingham <kieran.bingham@...asonboard.com>,
        Linux Media Mailing List <linux-media@...r.kernel.org>,
        Guennadi Liakhovetski <g.liakhovetski@....de>,
        olivier.braun@...reolabs.com, troy.kisky@...ndarydevices.com,
        Randy Dunlap <rdunlap@...radead.org>,
        Philipp Zabel <philipp.zabel@...il.com>,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v4 6/6] media: uvcvideo: Move decode processing to process context

Hi Laurent,

On Wed, Aug 8, 2018 at 8:12 AM Laurent Pinchart
<laurent.pinchart@...asonboard.com> wrote:
>
> Hi Tomasz,
>
> On Tuesday, 7 August 2018 12:54:02 EEST Tomasz Figa wrote:
> > On Wed, Mar 28, 2018 at 1:47 AM Kieran Bingham wrote:
>
> [snip]
>
> > In our testing, this function ends up being called twice
>
> In your testing, has this patch series brought noticeable performance
> improvements ? Is there a particular reason you tested it, beside general
> support of UVC devices in ChromeOS ?

Some of our older ARM devices have external USB ports wired to a low
end dwc2 controller, which puts quite strict timing requirements on
interrupt handling. For some cameras that produce bigger payloads (in
our testing that was Logitech BRIO, running at 1080p), almost half of
every frame would be dropped, due to the memcpy from uncached memory
taking too much time. With this series, it goes down to bottom ~10% of
only a part of the frames. With one more optimization from Keiichi
[1], the problem disappears almost completely.

[1] https://lore.kernel.org/patchwork/patch/956388/

Best regards,
Tomasz