[FFmpeg-cvslog] r29509 - trunk/libswscale/swscale.c

Reimar Döffinger Reimar.Doeffinger
Fri Aug 14 13:49:27 CEST 2009


On Fri, Aug 14, 2009 at 08:31:43AM -0300, Ramiro Polla wrote:
> On Fri, Aug 14, 2009 at 8:18 AM, Reimar
> D?ffinger<Reimar.Doeffinger at gmx.de> wrote:
> > It is what it says. What you read into it is "If an implementation
> > cannot support the _exact_ combination" - which I admit is _one_
> > reasonable way to read it, but not in the context...
> > E.g. by supporting READ+WRITE+EXEC at once an architecture also
> > supports READ or READ+WRITE or READ+EXEC etc., it just does not
> > deny _all_ other types of access.
> > Whereas an architecture where e.g. all code is in ROM indeed really
> > and in no way can support anything with WRITE+EXEC.
> 
> Hmm, I just remembered this code is strictly x86+mmx2. It has worked
> up to now. We could remove READ and wait for a possible regression, or
> leave it the way we know it works. I prefer just leaving it as is.

Uh, what "has worked up to now" is READ+WRITE+EXEC.
Now it is READ+WRITE which is supported by x86 but until NX was
introduced (and then only when PAE is used) it is always
READ+WRITE+EXEC.
So if any implementor went with M?ns interpretation the current code
couldn't work on any system that doesn't support NX+PAE.
Now I admit that removing read in difference to EXEC does probably
not provide any relevant security advantage though, I still consider
it more correct anyway.



More information about the ffmpeg-cvslog mailing list