[patch] pae-2.4.3-C3

[Ingo Molnar <mingo@elte.hu>]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 857 bytes --]


the attached pae-2.4.3-C3 patch fixes the PAE code to work with SLAB
FORCED_DEBUG (which enables redzoning) too.

the problem is that redzoning is enabled unconditionally, and SLAB has no
information about how crutial alignment is in the case of any particular
SLAB cache. The CPU generates a general protection fault if in PAE mode a
non-16-byte aligned pgd is loaded into %cr3.

Redzoning is very useful and most caches use HWCACHE_ALIGN, so i didnt
want to disable redzoning if SLAB_HWCACHE_ALIGN is set. The patch adds
SLAB_MUST_HWCACHE_ALIGN, which makes the distinction.

the patch is against 2.4.3-pre8 (applies cleanly to ac26 as well), and
compiles & boot cleanly in PAE mode, with or without CONFIG_DEBUG_SLAB.
The kernel also compiles cleanly in non-PAE mode.

(the patch also introduces the pae_pgd SLAB cache, which is a tiny
speedup.)

	Ingo

[-- Attachment #2: Type: TEXT/PLAIN, Size: 4322 bytes --]

--- linux/init/main.c.orig	Wed Mar 28 14:41:59 2001
+++ linux/init/main.c	Wed Mar 28 14:42:02 2001
@@ -570,6 +570,9 @@
 #endif
 	mem_init();
 	kmem_cache_sizes_init();
+#if CONFIG_X86_PAE
+	init_pae_pgd_cache();
+#endif
 #ifdef CONFIG_3215_CONSOLE
         con3215_activate();
 #endif
--- linux/mm/slab.c.orig	Wed Mar 28 14:41:59 2001
+++ linux/mm/slab.c	Wed Mar 28 14:42:02 2001
@@ -102,9 +102,11 @@
 #if DEBUG
 # define CREATE_MASK	(SLAB_DEBUG_INITIAL | SLAB_RED_ZONE | \
 			 SLAB_POISON | SLAB_HWCACHE_ALIGN | \
-			 SLAB_NO_REAP | SLAB_CACHE_DMA)
+			 SLAB_NO_REAP | SLAB_CACHE_DMA | \
+			 SLAB_MUST_HWCACHE_ALIGN)
 #else
-# define CREATE_MASK	(SLAB_HWCACHE_ALIGN | SLAB_NO_REAP | SLAB_CACHE_DMA)
+# define CREATE_MASK	(SLAB_HWCACHE_ALIGN | SLAB_NO_REAP | \
+			 SLAB_CACHE_DMA | SLAB_MUST_HWCACHE_ALIGN)
 #endif
 
 /*
@@ -641,7 +643,7 @@
 		flags &= ~SLAB_POISON;
 	}
 #if FORCED_DEBUG
-	if (size < (PAGE_SIZE>>3))
+	if ((size < (PAGE_SIZE>>3)) && !(flags & SLAB_MUST_HWCACHE_ALIGN))
 		/*
 		 * do not red zone large object, causes severe
 		 * fragmentation.
--- linux/include/linux/slab.h.orig	Sun Dec 31 20:10:17 2000
+++ linux/include/linux/slab.h	Wed Mar 28 14:44:25 2001
@@ -36,6 +36,7 @@
 #define	SLAB_NO_REAP		0x00001000UL	/* never reap from the cache */
 #define	SLAB_HWCACHE_ALIGN	0x00002000UL	/* align objs on a h/w cache lines */
 #define SLAB_CACHE_DMA		0x00004000UL	/* use GFP_DMA memory */
+#define	SLAB_MUST_HWCACHE_ALIGN	0x00008000UL	/* force alignment */
 
 /* flags passed to a constructor func */
 #define	SLAB_CTOR_CONSTRUCTOR	0x001UL		/* if not set, then deconstructor */
--- linux/include/asm-i386/pgalloc.h.orig	Wed Mar 28 14:41:59 2001
+++ linux/include/asm-i386/pgalloc.h	Wed Mar 28 14:44:25 2001
@@ -20,13 +20,19 @@
 
 #if CONFIG_X86_PAE
 
-extern void *kmalloc(size_t, int);
-extern void kfree(const void *);
+struct kmem_cache_s;
+
+extern struct kmem_cache_s *pae_pgd_cachep;
+
+extern void *kmem_cache_alloc(struct kmem_cache_s *, int);
+extern void kmem_cache_free(struct kmem_cache_s *, void *);
+
+extern void init_pae_pgd_cache(void);
 
 extern __inline__ pgd_t *get_pgd_slow(void)
 {
 	int i;
-	pgd_t *pgd = kmalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
+	pgd_t *pgd = kmem_cache_alloc(pae_pgd_cachep, GFP_KERNEL);
 
 	if (pgd) {
 		for (i = 0; i < USER_PTRS_PER_PGD; i++) {
@@ -42,7 +48,7 @@
 out_oom:
 	for (i--; i >= 0; i--)
 		free_page((unsigned long)__va(pgd_val(pgd[i])-1));
-	kfree(pgd);
+	kmem_cache_free(pae_pgd_cachep, pgd);
 	return NULL;
 }
 
@@ -88,7 +94,7 @@
 
 	for (i = 0; i < USER_PTRS_PER_PGD; i++)
 		free_page((unsigned long)__va(pgd_val(pgd[i])-1));
-	kfree(pgd);
+	kmem_cache_free(pae_pgd_cachep, pgd);
 #else
 	free_page((unsigned long)pgd);
 #endif
--- linux/arch/i386/mm/init.c.orig	Wed Mar 28 14:41:56 2001
+++ linux/arch/i386/mm/init.c	Wed Mar 28 14:42:02 2001
@@ -15,6 +15,9 @@
 #include <linux/types.h>
 #include <linux/ptrace.h>
 #include <linux/mman.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/fs.h>
 #include <linux/mm.h>
 #include <linux/swap.h>
 #include <linux/smp.h>
@@ -131,15 +134,11 @@
 	pte_t *pte;
 
 	pgd = swapper_pg_dir + __pgd_offset(vaddr);
-	if (pgd_none(*pgd)) {
-		printk("PAE BUG #00!\n");
-		return;
-	}
+	if (pgd_none(*pgd))
+		BUG();
 	pmd = pmd_offset(pgd, vaddr);
-	if (pmd_none(*pmd)) {
-		printk("PAE BUG #01!\n");
-		return;
-	}
+	if (pmd_none(*pmd))
+		BUG();
 	pte = pte_offset(pmd, vaddr);
 	if (pte_val(*pte))
 		pte_ERROR(*pte);
@@ -183,7 +182,7 @@
 			pmd = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
 			set_pgd(pgd, __pgd(__pa(pmd) + 0x1));
 			if (pmd != pmd_offset(pgd, 0))
-				printk("PAE BUG #02!\n");
+				BUG();
 		}
 		pmd = pmd_offset(pgd, vaddr);
 #else
@@ -581,3 +580,23 @@
 	val->mem_unit = PAGE_SIZE;
 	return;
 }
+
+#if CONFIG_X86_PAE
+
+#include <linux/slab.h>
+
+struct kmem_cache_s *pae_pgd_cachep;
+
+void __init init_pae_pgd_cache(void)
+{
+	/*
+	 * PAE pgds must be 16-byte aligned:
+	 */
+	pae_pgd_cachep = kmem_cache_create("pae_pgd", 32, 0,
+		SLAB_HWCACHE_ALIGN | SLAB_MUST_HWCACHE_ALIGN, NULL, NULL);
+	if (!pae_pgd_cachep)
+		panic("init_pae(): Cannot alloc pae_pgd SLAB cache");
+}
+
+#endif
+