[PATCH xorg-server] Fix xf86EdidModes.c: array subscript is above array bounds

Torsten Kaiser just.for.lkml at googlemail.com
Fri Jul 6 10:24:29 PDT 2012

On Fri, Jul 6, 2012 at 9:44 AM, Keith Packard <keithp at keithp.com> wrote:
> Torsten Kaiser <just.for.lkml at googlemail.com> writes:
>> My guess would be, that these bits where reserved in the spec and
>> should never been set.
> These bits come from external sources, and often buggy ones at that. We

Yes, because buggy devices can trigger this, I like to see it fixed
and not only making gcc no longer notice this.

> should probably just compute the index and skip it if it's out of

There was an earlier patch that did that:

But I found that adding these calculations to the loop made it rather
difficult to read.
The solution with the dummy entries looked better to me, because its
obvious what they are at the first glance.

> range. Sending a mode of all zeros along to the rest of the server is
> likely to cause chaos (and divide-by-zero errors).

Thats why I added "if (EstIIIModes[m].w)" into the loop to skip over
them, even if the device sends the strange bits to enable them.

Just the current code could cause chaos, if these bits ever get
enabled, because it would try to use whatever comes after EstIIIModes
as mode information.

>> The two issues where just located so near each other, that I fixed
>> them in one go.
> It is nice to have separate patches anyways; one per bug. That can make
> back-porting changes easier to manage.

Not arguing with that. Thats why I immediately redid patch(es) after
your request.

I just couldn't resist slipping the single = into the original patch,
after I noticed the second bug. ;-)


More information about the xorg-devel mailing list