mm: make minimum slab alignment a runtime property

When CONFIG_KASAN_HW_TAGS is enabled we currently increase the minimum
slab alignment to 16.  This happens even if MTE is not supported in
hardware or disabled via kasan=off, which creates an unnecessary memory
overhead in those cases.  Eliminate this overhead by making the minimum
slab alignment a runtime property and only aligning to 16 if KASAN is
enabled at runtime.

On a DragonBoard 845c (non-MTE hardware) with a kernel built with
CONFIG_KASAN_HW_TAGS, waiting for quiescence after a full Android boot I
see the following Slab measurements in /proc/meminfo (median of 3
reboots):

Before: 169020 kB
After:  167304 kB

[[email protected]: make slab alignment type `unsigned int' to avoid casting]
Link: https://linux-review.googlesource.com/id/I752e725179b43b144153f4b6f584ceb646473ead
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Peter Collingbourne <[email protected]>
Reviewed-by: Andrey Konovalov <[email protected]>
Reviewed-by: Hyeonggon Yoo <[email protected]>
Tested-by: Hyeonggon Yoo <[email protected]>
Acked-by: David Rientjes <[email protected]>
Reviewed-by: Catalin Marinas <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Cc: Pekka Enberg <[email protected]>
Cc: Roman Gushchin <[email protected]>
Cc: Joonsoo Kim <[email protected]>
Cc: Herbert Xu <[email protected]>
Cc: Andrey Ryabinin <[email protected]>
Cc: Alexander Potapenko <[email protected]>
Cc: Dmitry Vyukov <[email protected]>
Cc: Eric W. Biederman <[email protected]>
Cc: Kees Cook <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

Author: pcc

arch/arm64/include/asm/cache.h

@@ -6,6 +6,7 @@
 #define __ASM_CACHE_H
 
 #include <asm/cputype.h>
+#include <asm/mte-def.h>
 
 #define CTR_L1IP_SHIFT		14
 #define CTR_L1IP_MASK		3
@@ -49,16 +50,22 @@
  */
 #define ARCH_DMA_MINALIGN	(128)
 
+#ifndef __ASSEMBLY__
+
+#include <linux/bitops.h>
+#include <linux/kasan-enabled.h>
+
 #ifdef CONFIG_KASAN_SW_TAGS
 #define ARCH_SLAB_MINALIGN	(1ULL << KASAN_SHADOW_SCALE_SHIFT)
 #elif defined(CONFIG_KASAN_HW_TAGS)
-#define ARCH_SLAB_MINALIGN	MTE_GRANULE_SIZE
+static inline unsigned int arch_slab_minalign(void)
+{
+	return kasan_hw_tags_enabled() ? MTE_GRANULE_SIZE :
+					 __alignof__(unsigned long long);
+}
+#define arch_slab_minalign() arch_slab_minalign()
 #endif
 
-#ifndef __ASSEMBLY__
-
-#include <linux/bitops.h>
-
 #define ICACHEF_ALIASING	0
 #define ICACHEF_VPIPT		1
 extern unsigned long __icache_flags;

include/linux/slab.h

@@ -209,6 +209,18 @@ void kmem_dump_obj(void *object);
 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
 #endif
 
+/*
+ * Arches can define this function if they want to decide the minimum slab
+ * alignment at runtime. The value returned by the function must be a power
+ * of two and >= ARCH_SLAB_MINALIGN.
+ */
+#ifndef arch_slab_minalign
+static inline unsigned int arch_slab_minalign(void)
+{
+	return ARCH_SLAB_MINALIGN;
+}
+#endif
+
 /*
  * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
  * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN

mm/slab.c

@@ -3009,10 +3009,9 @@ static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
 	objp += obj_offset(cachep);
 	if (cachep->ctor && cachep->flags & SLAB_POISON)
 		cachep->ctor(objp);
-	if (ARCH_SLAB_MINALIGN &&
-	    ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
-		pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
-		       objp, (int)ARCH_SLAB_MINALIGN);
+	if ((unsigned long)objp & (arch_slab_minalign() - 1)) {
+		pr_err("0x%px: not aligned to arch_slab_minalign()=%u\n", objp,
+		       arch_slab_minalign());
 	}
 	return objp;
 }

mm/slab_common.c

@@ -154,8 +154,7 @@ static unsigned int calculate_alignment(slab_flags_t flags,
 		align = max(align, ralign);
 	}
 
-	if (align < ARCH_SLAB_MINALIGN)
-		align = ARCH_SLAB_MINALIGN;
+	align = max(align, arch_slab_minalign());
 
 	return ALIGN(align, sizeof(void *));
 }

mm/slob.c

@@ -478,9 +478,11 @@ static __always_inline void *
 __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller)
 {
 	unsigned int *m;
-	int minalign = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
+	unsigned int minalign;
 	void *ret;
 
+	minalign = max_t(unsigned int, ARCH_KMALLOC_MINALIGN,
+			 arch_slab_minalign());
 	gfp &= gfp_allowed_mask;
 
 	might_alloc(gfp);
@@ -493,7 +495,7 @@ __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller)
 		 * kmalloc()'d objects.
 		 */
 		if (is_power_of_2(size))
-			align = max(minalign, (int) size);
+			align = max_t(unsigned int, minalign, size);
 
 		if (!size)
 			return ZERO_SIZE_PTR;
@@ -555,8 +557,11 @@ void kfree(const void *block)
 
 	sp = virt_to_folio(block);
 	if (folio_test_slab(sp)) {
-		int align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
+		unsigned int align = max_t(unsigned int,
+					   ARCH_KMALLOC_MINALIGN,
+					   arch_slab_minalign());
 		unsigned int *m = (unsigned int *)(block - align);
+
 		slob_free(m, *m + align);
 	} else {
 		unsigned int order = folio_order(sp);
@@ -573,7 +578,7 @@ EXPORT_SYMBOL(kfree);
 size_t __ksize(const void *block)
 {
 	struct folio *folio;
-	int align;
+	unsigned int align;
 	unsigned int *m;
 
 	BUG_ON(!block);
@@ -584,7 +589,8 @@ size_t __ksize(const void *block)
 	if (unlikely(!folio_test_slab(folio)))
 		return folio_size(folio);
 
-	align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
+	align = max_t(unsigned int, ARCH_KMALLOC_MINALIGN,
+		      arch_slab_minalign());
 	m = (unsigned int *)(block - align);
 	return SLOB_UNITS(*m) * SLOB_UNIT;
 }