Allow arch override for kmem_bufctl_t

Hi Andrew,

The slab allocator keeps track of the free objects in a slab with a linked list of integers (typedef'ed to kmem_bufctl_t). Right now unsigned int is used for kmem_bufctl_t, i.e. 4 bytes per-object overhead.
The attached patch allows an arch override for this type: Theoretically, unsigned short is sufficient for kmem_bufctl_t and this would reduce the per-object overhead to 2 bytes. But some archs cannot operate on 16-bit values efficiently, thus it's not possible to switch everyone to ushort. The patch switches i386, the port maintainers should switch their arch if there are no problems with 16-bit quantities. All accesses to kmem_bufctl_t are under a per-cache spinlock and not in the hottest path.

Could you apply it to -mm?

--
Manfred

Attached is the patch - sorry.

--
Manfred

// $Header$
// Kernel Version:

//  VERSION = 2
//  PATCHLEVEL = 6
//  SUBLEVEL = 6
//  EXTRAVERSION = -mm2

@@ -161,10 +161,9 @@
  * is less than 512 (PAGE_SIZE<<3), but greater than 256.
  */
 
-#define BUFCTL_END	0xffffFFFF
-#define BUFCTL_FREE	0xffffFFFE
-#define	SLAB_LIMIT	0xffffFFFD
-typedef unsigned int kmem_bufctl_t;
+#define BUFCTL_END	(((kmem_bufctl_t)(~0U))-0)
+#define BUFCTL_FREE	(((kmem_bufctl_t)(~0U))-1)
+#define	SLAB_LIMIT	(((kmem_bufctl_t)(~0U))-2)
 
 /* Max number of objs-per-slab for caches which use off-slab slabs.
  * Needed to avoid a possible looping condition in cache_grow().

@@ -140,6 +140,9 @@
 #define pgoff_t unsigned long
 #endif
 
+#ifndef HAVE_KMEM_BUFCTL_T
+typedef unsigned int kmem_bufctl_t;
+#endif
 #endif /* __KERNEL_STRICT_NAMES */
 
 /*

@@ -63,6 +63,9 @@
 #define HAVE_SECTOR_T
 #endif
 
+#define HAVE_KMEM_BUFCTL_T 1
+typedef unsigned short kmem_bufctl_t;
+
 #endif /* __ASSEMBLY__ */
 
 #endif /* __KERNEL__ */

16-bit is no problem for m68k, so feel free to switch it.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

No problem for MIPS either.

Ralf

Ditto for Alpha.

Ivan.

x86_64 has no problems with 16 bit neither.

-Andi

Manfred> Hi Andrew, The slab allocator keeps track of the free Manfred> objects in a slab with a linked list of integers Manfred> (typedef'ed to kmem_bufctl_t). Right now unsigned int is Manfred> used for kmem_bufctl_t, i.e. 4 bytes per-object overhead. Manfred> The attached patch allows an arch override for this type:
Manfred> Theoretically, unsigned short is sufficient for Manfred> kmem_bufctl_t and this would reduce the per-object overhead Manfred> to 2 bytes. But some archs cannot operate on 16-bit values Manfred> efficiently, thus it's not possible to switch everyone to Manfred> ushort. The patch switches i386, the port maintainers Manfred> should switch their arch if there are no problems with Manfred> 16-bit quantities. All accesses to kmem_bufctl_t are under Manfred> a per-cache spinlock and not in the hottest path.

Manfred> Could you apply it to -mm?

There was no patch attached. I suspect we'd want to turn this on for ia64 as well. Do you have a good (worst-case) benchmark to evaluate this change?

	--david

Just out of curiosity, which arch knowingly has the problem?

I think it's ARM ...

"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

I thought it was sparc and perhaps alpha, but I was wrong: it was a general comment from Mark Hemment (May 2000):

--
Manfred