Re: [Qemu-devel] Re: [PATCH][ARM] Fix wrong destination register for smuad, smusd, smlad, smlsd

["Laurent Desnogues" <laurent.desnogues@gmail.com>]

On Wed, Jun 25, 2008 at 6:58 PM, Paul Brook <paul@codesourcery.com> wrote:
> On Wednesday 25 June 2008, Laurent Desnogues wrote:
>> 2008/6/23 Laurent Desnogues <laurent.desnogues@gmail.com>:
>> > smuad, smusd, smlad, smlsd write the wrong register, resulting in
>> > PC corruption.
>>
>> Here is a better patch.
>>
>> I will stop posting patches as I have the feeling the maintainer
>> doesn't care.  He might be rewriting everything and so my
>> patches are useless and a waste of time for everyone :-)
>
> TBH It's hard to tell if these are well tested patches, or just quick hacks
> that you're throwing over the wall.
>
> As above where it took you two tries. This isn't bad per se, but compete lack
> of explanation about what's different doesn't help. Some of your other
> patches have been prefixed with "I did not check the correctness of that
> instruction in general, I only made a change that looked logical" and "These
> are all *wild guesses*". As we know from the recent Debian SSL debacle
> making "a change that looked logical" can be fairly disastrous :-)

I am certainly over cautious as everyone should be when he doesn't
master all of a software :-)  So indeed when I propose some "wild
guess" it's all related to parts of qemu I am not sure to understand
and generally doesn't come with a patch. On the other hand when I
post a fix for the behaviour of an instruction I am sure it's OK except
when it contains multiple bugs (as was the case for that instruction).
I am now doing explicit tests to check each instruction does what
it should, so this mistake should not happen anymore.

> This means I'm unwilling to accept the patches a face value, and need to go
> through them with a fine tooth comb. This takes time, and you go in the queue
> with the dozens of other of patches, bugs and new features that need my
> attention.

That's enough of a feedback to me, I am not expecting you to
commit anything within a minute or even within days.  The lack of
any feedback from you made me wonder if you were not within a
complete rewrite of the ARM target that would have made my
patches useless and just some noise on the list :)

Anyway, are you interested in a patch that corrects enough of
v6 instructions that FFmpeg for ARMv6 works?  Would FFmpeg
be considered as a good enough test?


Laurent