Blackfin: use atomic kmalloc in L1 alloc so it too can be atomic

Some drivers allocate L1 SRAM in atomic contexts, so make sure these
functions also use GFP_ATOMIC to avoid BUG()'s.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>

arch/blackfin/mm/sram-alloc.c

@@ -256,7 +256,8 @@ static void *_sram_alloc(size_t size, struct sram_piece *pfree_head,
 		plast->next = pslot->next;
 		pavail = pslot;
 	} else {
-		pavail = kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
+		/* use atomic so our L1 allocator can be used atomically */
+		pavail = kmem_cache_alloc(sram_piece_cache, GFP_ATOMIC);
 
 		if (!pavail)
 			return NULL;