linux-kernel - Re: [PATCH v5 00/39] i.MX Media Driver

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:   Sun, 12 Mar 2017 13:39:37 -0700
From:   Steve Longerbeam <slongerbeam@...il.com>
To:     Russell King - ARM Linux <linux@...linux.org.uk>
Cc:     mark.rutland@....com, andrew-ct.chen@...iatek.com,
        minghsiu.tsai@...iatek.com, sakari.ailus@...ux.intel.com,
        nick@...anahar.org, songjun.wu@...rochip.com, hverkuil@...all.nl,
        Steve Longerbeam <steve_longerbeam@...tor.com>, pavel@....cz,
        robert.jarzmik@...e.fr, devel@...verdev.osuosl.org,
        markus.heiser@...marIT.de,
        laurent.pinchart+renesas@...asonboard.com, shuah@...nel.org,
        geert@...ux-m68k.org, linux-media@...r.kernel.org,
        devicetree@...r.kernel.org, kernel@...gutronix.de, arnd@...db.de,
        mchehab@...nel.org, bparrot@...com, robh+dt@...nel.org,
        horms+renesas@...ge.net.au, tiffany.lin@...iatek.com,
        linux-arm-kernel@...ts.infradead.org,
        niklas.soderlund+renesas@...natech.se, gregkh@...uxfoundation.org,
        linux-kernel@...r.kernel.org, jean-christophe.trotin@...com,
        p.zabel@...gutronix.de, fabio.estevam@....com, shawnguo@...nel.org,
        sudipm.mukherjee@...il.com
Subject: Re: [PATCH v5 00/39] i.MX Media Driver



On 03/12/2017 01:36 PM, Steve Longerbeam wrote:
>
>
> On 03/12/2017 01:16 PM, Steve Longerbeam wrote:
>>
>>
>> On 03/12/2017 12:44 PM, Steve Longerbeam wrote:
>>>
>>>
>>> On 03/12/2017 12:29 PM, Russell King - ARM Linux wrote:
>>>> On Sun, Mar 12, 2017 at 12:21:45PM -0700, Steve Longerbeam wrote:
>>>>> There's actually nothing preventing userland from disabling a link
>>>>> multiple times, and imx_media_link_notify() complies, and so
>>>>> csi_s_power(OFF) gets called multiple times, and so that WARN_ON()
>>>>> in there is silly, I borrowed this from other MC driver examples,
>>>>> but it makes no sense to me, I'll remove it and prevent the power
>>>>> count from going negative.
>>>>
>>>> Hmm.  So what happens if one of the CSI's links is enabled, and we
>>>> disable a different link from the CSI several times?  Doesn't that
>>>> mean the power count will go to zero despite there being an enabled
>>>> link?
>>>
>>> Yes, the CSI will be powered off even if it still has an enabled link.
>>> But one of its other links has been disabled, meaning the pipeline as
>>> a whole is disabled. So I think it makes sense to power down the CSI,
>>> the pipeline isn't usable at that point.
>>>
>>> And remember that the CSI does not allow both output pads to be enabled
>>> at the same time. If that were so then indeed there would be a problem,
>>> because it would mean there is another active pipeline that requires the
>>> CSI being powered on, but that's not the case.
>>>
>>> I think this is consistent with the other entities as well, but I will
>>> double check.
>>
>>
>> At first I thought this could be a problem for one entity, the csi-2
>> receiver.
>>
>> It can enable all four of its output pads at once (if the input stream
>> contains all 4 virtual channels, the csi-2 receiver must support
>> demuxing all of them onto all 4 of its output pads).
>>
>> But after more review, this should not be an issue. If a csi-2 sink
>> (a CSI or a CSI mux) link is disabled, the csi-2 receiver is no longer
>> reachable from that sink, so attempts to disable the csi-2 via that
>> path again is not possible. The other potential problem is disabling
>> from the csi-2's own sink pad, but in that case the csi-2 no longer
>> has a source, so again it makes sense to power off the csi-2 even
>> if it has enabled output pads.
>>
>
> But hold on, if my logic is correct, then why did the CSI power-off
> get reached in your case, multiple times? Yes I think there is a bug,
> link_notify() is not checking if the link has already been disabled.
> I will fix this. But I'm surprised media core's link_notify handling
> doesn't do this.

but it does:

int __media_entity_setup_link(struct media_link *link, u32 flags)
{
...
     if (link->flags == flags)
         return 0;
...
}

What the heck. Anyway, I'll track this down.

Steve

>
> Steve
>