[ECOS] Memory Problems Under eCos

[ECOS] Memory Problems Under eCos

[Colin Ford]

Hello,

I'm still finding that my GoAhead web server runs out of
memory. I've traced it a bit further and found that when
a request comes through memory is trying to be freed but
the bp->flags have been corrupted and the integrity
not valid anymore.

It seems that the following gets corrupted:

  ../sockGen.c:712
  ../webs.c:2108
  ../sockGen.c:716
  ../sockGen.c:714

These then mount up by one each time and eventually
I run out of memory.......Doh!

So I reckon that its eCos. Ho well I'll carry on looking.

Thanks,
Col.

--
===========================================
Colin Ford               PipingHot Networks
Software Engineer        +44 (0)1364 655510

Re: [ECOS] Memory Problems Under eCos

[Hugo Tyson]

Colin Ford <colin.ford@pipinghotnetworks.com> writes:

> I'm still finding that my GoAhead web server runs out of
> memory. I've traced it a bit further and found that when
> a request comes through memory is trying to be freed but
> the bp->flags have been corrupted and the integrity
> not valid anymore.
> 
> It seems that the following gets corrupted:
> 
>   ../sockGen.c:712
>   ../webs.c:2108
>   ../sockGen.c:716
>   ../sockGen.c:714
> 
> These then mount up by one each time and eventually
> I run out of memory.......Doh!
> 
> So I reckon that its eCos. Ho well I'll carry on looking.

I know 0.0 about the context here, but when I'm debugging and I see stuff
like that, what I do is severalfold:

a) run with asserts enabled if you're not already
b) make your stacks much bigger - eCos does NOT do stack checking
c) make your malloc heap much bigger (if that applies to your platform) -
   a failed malloc returns NULL, which can go undetected.
[Sorry if this is preaching to the converted...]

What's a bp-> ?  Is it an eCos data structure or part of GoAhead?
Those source files are references to allocation calls or pointers?
(If anyone else has more context here and can help properly, ignore me)

	- Huge

Re: [ECOS] Memory Problems Under eCos

[Jonathan Larmour]

Hugo Tyson wrote:
> 
> Colin Ford <colin.ford@pipinghotnetworks.com> writes:
> 
> > I'm still finding that my GoAhead web server runs out of
> > memory. I've traced it a bit further and found that when
> > a request comes through memory is trying to be freed but
> > the bp->flags have been corrupted and the integrity
> > not valid anymore.
> >
[snip] 
> I know 0.0 about the context here, but when I'm debugging and I see stuff
> like that, what I do is severalfold:
> 
> a) run with asserts enabled if you're not already
> b) make your stacks much bigger - eCos does NOT do stack checking
> c) make your malloc heap much bigger (if that applies to your platform) -
>    a failed malloc returns NULL, which can go undetected.
> [Sorry if this is preaching to the converted...]
> 
> What's a bp-> ?  Is it an eCos data structure or part of GoAhead?

GoAhead

To add some more context, this is something we've seen as well, but we
haven't worked out where it is happening - we'd be grateful if Colin could
tell us if he finds the problem! GoAhead have their own memory allocation
stuff, and don't use eCos malloc. I imagine Gary T has some more state but
I happen to know he's busy right now :).

Jifl
-- 
Red Hat, 35 Cambridge Place, Cambridge, UK. CB2 1NS  Tel: +44 (1223) 728762
"Plan to be spontaneous tomorrow."  ||  These opinions are all my own fault