Re: [PATCH v2 1/1] memalign: Support scanning for aligned chunks.

[Florian Weimer <fweimer@redhat.com>]

* DJ Delorie via Libc-alpha:

> diff --git a/malloc/malloc.c b/malloc/malloc.c
> index 12908b8f97..14ee98dfa2 100644
> --- a/malloc/malloc.c
> +++ b/malloc/malloc.c
> @@ -3557,6 +3557,30 @@ _mid_memalign (size_t alignment, size_t bytes, void *address)
>        alignment = a;
>      }
>  
> +#if USE_TCACHE
> +  {
> +    size_t tbytes;
> +    tbytes = checked_request2size (bytes);
> +    if (tbytes == 0)
> +      {
> +	__set_errno (ENOMEM);
> +	return NULL;
> +      }
> +    size_t tc_idx = csize2tidx (tbytes);
> +
> +    MAYBE_INIT_TCACHE ();
> +
> +    if (tc_idx < mp_.tcache_bins
> +	&& tcache
> +	&& tcache->counts[tc_idx] > 0
> +	&& ((intptr_t)tcache->entries[tc_idx] & (alignment - 1)) == 0)
> +      {
> +	void *victim = tcache_get (tc_idx);
> +	return tag_new_usable (victim);
> +      }
> +  }
> +#endif

MAYBE_INIT_TCACHE does not seem necessary.  If the tcache is empty,
there's no need to initialize it and scane it.  And why not scan the
entire chain here?  It's another source of missed chunk reuse.

The tcache null check should be written as tcache != NULL.

Thanks,
Florian