mm/slab: use IS_ENABLED() instead of ZONE_DMA_FLAG

The Kconfig symbol ZONE_DMA_FLAG probably predates the introduction of IS_ENABLED(). Remove it and replace its two uses with the equivalent IS_ENABLED(CONFIG_ZONE_DMA).

The Kconfig symbol ZONE_DMA_FLAG probably predates the introduction of IS_ENABLED(). Remove it and replace its two uses with the equivalent IS_ENABLED(CONFIG_ZONE_DMA).

Signed-off-by Paul Bolle <[email protected]>

---
Build tested on x86_64 (on top of next-20140925).

Run tested on i686 (on top of v3.17-rc6). That test required me to switch from SLUB (Fedora's default) to SLAB. That makes running this patch both more scary and less informative. Besides, I have no idea how to hit the codepaths I just changed. You'd expect this to not actually change slab.o, but I'm not sure how to check that. So, in short: review very much appreciated.

 mm/Kconfig | 5 -----
 mm/slab.c  | 4 ++--
 2 files changed, 2 insertions(+), 7 deletions(-)

mm/Kconfig +0 -5

index 886db21..8e860c7 100644
@@ -273,11 +273,6 @@ config ARCH_ENABLE_HUGEPAGE_MIGRATION
 config PHYS_ADDR_T_64BIT
 	def_bool 64BIT || ARCH_PHYS_ADDR_T_64BIT
 
-config ZONE_DMA_FLAG
-	int
-	default "0" if !ZONE_DMA
-	default "1"
-
 config BOUNCE
 	bool "Enable bounce buffers"
 	default y

mm/slab.c +0 -0

index 628f2b5..766c90e 100644

Quoted 40 lines · Paul Bolle

The Kconfig symbol ZONE_DMA_FLAG probably predates the introduction of IS_ENABLED(). Remove it and replace its two uses with the equivalent IS_ENABLED(CONFIG_ZONE_DMA).

Signed-off-by Paul Bolle <[email protected]>

---
Build tested on x86_64 (on top of next-20140925).

Run tested on i686 (on top of v3.17-rc6). That test required me to switch from SLUB (Fedora's default) to SLAB. That makes running this patch both more scary and less informative. Besides, I have no idea how to hit the codepaths I just changed. You'd expect this to not actually change slab.o, but I'm not sure how to check that. So, in short: review very much appreciated.

 mm/Kconfig | 5 -----
 mm/slab.c  | 4 ++--
 2 files changed, 2 insertions(+), 7 deletions(-)

mm/Kconfig +0 -5

index 886db21..8e860c7 100644
@@ -273,11 +273,6 @@ config ARCH_ENABLE_HUGEPAGE_MIGRATION
 config PHYS_ADDR_T_64BIT
 	def_bool 64BIT || ARCH_PHYS_ADDR_T_64BIT
 
-config ZONE_DMA_FLAG
-	int
-	default "0" if !ZONE_DMA
-	default "1"
-
 config BOUNCE
 	bool "Enable bounce buffers"
 	default y

mm/slab.c +0 -0

index 628f2b5..766c90e 100644

mm/slab.c __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) · line 2247

Quoted 6 lines

mm/slab.c +1 -1

@@ -2243,7 +2243,7 @@ __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags)
 	cachep->freelist_size = freelist_size;
 	cachep->flags = flags;
 	cachep->allocflags = __GFP_COMP;
-	if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA))
+	if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & SLAB_CACHE_DMA))
 		cachep->allocflags |= GFP_DMA;

GFP_DMA is actually safe to use even without CONFIG_ZONE_DMA, so you only need to check for SLAB_CACHE_DMA here.

Quoted 2 lines

mm/slab.c +0 -0

@@ -2516,7 +2516,7 @@ static void cache_init_objs(struct kmem_cache *cachep,
 

mm/slab.c static void cache_init_objs(struct kmem_cache *cachep, · line 2522

Quoted 7 lines

mm/slab.c +1 -1

 static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags)
 {
-	if (CONFIG_ZONE_DMA_FLAG) {
+	if (IS_ENABLED(CONFIG_ZONE_DMA)) {
 		if (flags & GFP_DMA)
 			BUG_ON(!(cachep->allocflags & GFP_DMA));
 		else

I think this assertion can be removed altogether and replaced by ORing the passed in flags with the cache gfp flags. The page allocator will catch any contradictions, but the 3 callsites that actually do use DMA caches are well-behaved as of now.

Do your comments require the patch to be redone (partially or entirely)? In that case someone else should probably take it and improve it, as I hardly understand the issues you raise. Or is the patch already queued somewhere, with Cristoph's Ack attached?