Re: [PATCH] calloc should not duplicate malloc logic.

[OndÅej BÃlka <neleai at seznam dot cz>]

On Wed, Feb 26, 2014 at 11:10:50PM +0530, Siddhesh Poyarekar wrote:
> On Wed, Feb 26, 2014 at 06:24:12PM +0100, OndÅej BÃlka wrote:
> > I did not said so, I said there could be minor regression and asked if
> > that is problem. Also there is problem of synchronization as independent
> > modifications would conflict and I want minimize these issue before they
> > happen
> 
> Right, and I opined that it is a problem.  There need to be more
> opinions on this so that a consensus can be achieved.
> 
> > > You haven't understood why this is a bad idea.  The pages will have
> > > been zeroed twice with your patch, once by mmap and again with memset.
> > > 
> > I understood you perfecly and you are wrong here. Current implementation
> > also zeroes memory twice as this example demonstrates. Your point is
> > invalid here.
> 
> Your example does not demonstrate that.  Your example only
> demonstrates that the allocated memory is being touched, which is
> obvious since the metadata is written to the chunk.  I am surprised
> that the kernel decided to page in the entire block for it though.
>
Its not kernel, calloc does that. After debugging a while I found its
because we installed a malloc_hook_ini that skip this path, so I assumed
that calloc design is broken. I will send revised patch.
 
> If a chunk is mmapped, it is only memset if perturb_byte is set:
> 
>   /* Two optional cases in which clearing not necessary */
>   if (chunk_is_mmapped (p))
>     {
>       if (__builtin_expect (perturb_byte, 0))
>         return memset (mem, 0, sz);
> 
>       return mem;
>     }
>