On 06/27/2012 09:36 AM, Gavin Shan wrote:

With CONFIG_SPARSEMEM_EXTREME, the root memory section descriptors
are allocated by slab or bootmem allocator. Also, the descriptors
might have been allocated and initialized during the hotplug path.
However, the memory chunk allocated in current implementation wouldn't
be put into the available pool if that has been allocated. The situation
will lead to memory leak.

I've read this changelog about ten times and I'm still not really clear
what the bug is here.

--

sparse_index_init() is designed to be safe if two copies of it race.  It
uses "index_init_lock" to ensure that, even in the case of a race, only
one CPU will manage to do:

	mem_section[root] = section;

However, in the case where two copies of sparse_index_init() _do_ race,
the one that loses the race will leak the "section" that
sparse_index_alloc() allocated for it.  This patch fixes that leak.

--

Technically, I'm not sure that we can race during the time when we'd be
using bootmem.  I think we do all those initializations single-threaded
at the moment, and we'd finish them before we turn the slab on.  So,
technically, we probably don't need the bootmem stuff in
sparse_index_free().  But, I guess it doesn't hurt, and it's fine for
completeness.

Gavin, have you actually tested this in some way?  It looks OK to me,
but I worry that you've just added a block of code that's exceedingly
unlikely to get run.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>