Make slab allocator work with SLAB_MUST_HWCACHE_ALIGN
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 4 lines · Manfred Spraul
But back to the patch that started this thread: Do you still need the ability to set an explicit alignment for slab allocations? If yes, then I'd polish my patch, double check all
kmem_cache_createcallers and then send the patch to akpm. Otherwise I'd wait - the patch is not a bugfix.
The explicit alignment would be my preferred way to fix slab debugging on s390. We still have the problem that practically all s390 device drivers need 64-bit alignment on slab allocations.
Our current hack is to redefine BYTES_PER_WORD in slab.c to 8, but what I'd like to see is a per-architecture alignment in kmem_cache_init that would default to 8 on s390 and to sizeof(long) otherwise.
Using the current SLAB_MUST_HWCACHE_ALIGN is not an option since it wastes a lot of memory with our 2048-bit L1 cache lines. I'd also like to avoid creating special slab caches for anything that needs 8-byte alignment.
Arnd <><