Re: [ECOS] Doug Lea's malloc implementation?

[Jonathan Larmour <jifl at eCosCentric dot com>]

Bart Veer wrote:
> > > > > > "Nikolay" == Nikolay V Pyatkov <npyatkov@usa.net> writes:
>     Nikolay> I was debugging GoAhead web server on linux synthetic
>     Nikolay> target. It was crashing in the middle of the first page
>     Nikolay> transfer. I traced the problem to free() in the "Doug
>     Nikolay> Lea's malloc implementation". The values to the free
>     Nikolay> looks fine. At the top of the stack was allocation of 4
>     Nikolay> bytes and free of the same 4 bytes. I could not get any
>     Nikolay> useful information from the dbg. I removed -O2. The same
>     Nikolay> result.
>
>     Nikolay> But after selecting "Simple variable block
>     Nikolay> implementation" instead of " Doug Lea's malloc
>     Nikolay> implementation" and recompiling the kernel everything
>     Nikolay> works just fine.
>
>     Nikolay> Did anyone see any problems with this package or it is
>     Nikolay> just my setup?
Make sure you are using recent-ish anoncvs. Some bugs were fixed a year 
and a bit ago.

> The Doug Lea malloc implementation is known to be pretty solid, and is
> further stress-tested by the various tests in the services/memalloc
> package (especially malloc4). However if there is a bug somewhere in
> the application, for example a double free on a region of memory, then
> you may well see different behaviour from the different allocators.
Or similarly easily, running off the end of an allocated bit of memory. A 
double free would cause an obvious assert failure for the variable 
allocator (something like "Couldn't free!" in free() in malloc.cxx). 
dlmalloc would probably also assert, but maybe not in so obvious a way.

> With one allocator such a bug might go unnoticed (assuming the freed
> memory has not yet been reallocated), while with another allocator it
> might cause immediate memory corruption. Make sure you are running
> with CYGDBG_MEMALLOC_ALLOCATOR_DLMALLOC_DEBUG enabled, that may help 
> find the problem.
Indeed. That should be on by default when asserts are enabled. And for 
development, asserts should always be enabled :).

Jifl
--
--[ "You can complain because roses have thorns, or you ]--
--[  can rejoice because thorns have roses." -Lincoln   ]-- Opinions==mine


--
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss