spca5xx freezes system

[Lonni J Friedman]

It took me a month, but I finally found time to build a 2.6.16 kernel
with FC5's gcc32, and I confirmed that this does workaround the
problem.  I emailed the developer of spca5xx about this, and this was
his reply:
######
People report crash when the kernel is configured with a 4ko stack
CONFIG_4KSTACKS
you should never set this, the jpeg decoder inside the spca5xx module need a
8ko stack.
People report crash with FC5 and gcc 4.1, setting in the Makefile this option
for CFLAGS fix the problem
-fno-unit-at-a-time
I am using OpenSuse set with 2.6.16 and gcc 4.1 without any problem
#####

I also tested his -fno-unit-at-a-time workaround, and it made no
difference for me.  His requirement of having an 8K stack seems
ridiculous to me.  This is a webcam driver, it shouldn't need an 8k
stack, especially since no others in the kernel have that requirement.

On 4/18/06, D. Hugh Redelmeier <hugh at mimosa.com> wrote:
> | From: Lonni J Friedman <netllama at gmail.com>
>
> | I couldn't even find such a list on the spca5xx website.  If you've
> | got a pointer, I'll gladly take the discussion there.
>
> Well, you can wander around from the links I posted.  They are from
> the SourceForge site for the project.  The home page on the site
> mentions the mailing list (and an IRC channel).
>
> http://sourceforge.net/mail/?group_id=28498
>
> | I wouldn't be building a Fedora kernel, I'd be building a kernel.org
> | kernel, which most definitely is qualified against gcc versions
> | earlier than 4.x.  But I do see your point.
>
> I'm a chicken.  I have no idea whether things break if you use a
> kernel.org kernel with FC5.
>
> Each of us has our own way to get to a goal.  I'm fairly knowledgeable
> about C and C compilers, so I'd try to isolate the bug myself.  Given
> that the developer seems reluctant to do so.
>
> On the other hand, your approach is good in that it gives the
> developer what he is asking for.
>
> | Well sure, but since the spca5xx developer said that he thought
> | gcc-4.x was the problem, I figured it would be a better use of my time
> | to not spend time on gcc-4.x, and isolate whether the version of gcc
> | that the spca5xx dev claims is 'good' truly is.
>
> My suspicion was what I described: new GCC4 optimizations expose bugs
> in the driver.  My path would get to that sooner.
>
> But a quick glance at the list suggests that there are other problems
> with the driver.  Your approach would more quickly reveal the problem
> if it were device-dependent.


-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
L. Friedman                                    netllama at gmail.com
LlamaLand                       http://netllama.linux-sxs.org