[PATCH] ia64: make hugetlbfs page size a boot-time option

This patch makes it possible to select the hugetlbfs page size at boot-time,
rather than at compile-time.  We believe we have resolved all the remaining
issues.  All critical speed path has been taken care of, i.e., vhpt handler
and context switch.  There should be no performance penalty with this dynamic
hugetlb page size feature.

We would like to thank Jack Steiner for his initiative on this feature
and his initial cool working patch.

arch/ia64/Kconfig

@@ -288,39 +288,6 @@ config FORCE_MAX_ZONEORDER
 	int
 	default "18"
 
-choice
-	prompt "Huge TLB page size"
-	depends on HUGETLB_PAGE
-	default HUGETLB_PAGE_SIZE_16MB
-
-config HUGETLB_PAGE_SIZE_4GB
-	depends on MCKINLEY
-	bool "4GB"
-
-config HUGETLB_PAGE_SIZE_1GB
-	depends on MCKINLEY
-	bool "1GB"
-
-config HUGETLB_PAGE_SIZE_256MB
-	bool "256MB"
-
-config HUGETLB_PAGE_SIZE_64MB
-	bool "64MB"
-
-config HUGETLB_PAGE_SIZE_16MB
-	bool "16MB"
-
-config HUGETLB_PAGE_SIZE_4MB
-	bool "4MB"
-
-config HUGETLB_PAGE_SIZE_1MB
-	bool "1MB"
-
-config HUGETLB_PAGE_SIZE_256KB
-	bool "256KB"
-
-endchoice
-
 config IA64_PAL_IDLE
 	bool "Use PAL_HALT_LIGHT in idle loop"
 	help

arch/ia64/kernel/ivt.S

@@ -118,10 +118,11 @@ ENTRY(vhpt_miss)
 #ifdef CONFIG_HUGETLB_PAGE
 	extr.u r26=r25,2,6
 	;;
-	cmp.eq p8,p0=HPAGE_SHIFT,r26
+	cmp.ne p8,p0=r18,r26
+	sub r27=r26,r18
 	;;
 (p8)	dep r25=r18,r25,2,6
-(p8)	shr r22=r22,HPAGE_SHIFT-PAGE_SHIFT
+(p8)	shr r22=r22,r27
 #endif
 	;;
 	cmp.eq p6,p7=5,r17			// is IFA pointing into to region 5?

arch/ia64/mm/hugetlbpage.c

 /*
  * IA-64 Huge TLB Page Support for Kernel.
  *
- * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
+ * Copyright (C) 2002-2004 Rohit Seth <rohit.seth@intel.com>
+ * Copyright (C) 2003-2004 Ken Chen <kenneth.w.chen@intel.com>
+ *
+ * Sep, 2003: add numa support
+ * Feb, 2004: dynamic hugetlb page size via boot parameter
  */
 
 #include <linux/config.h>
@@ -18,11 +22,10 @@
 #include <asm/tlb.h>
 #include <asm/tlbflush.h>
 
-#define TASK_HPAGE_BASE (REGION_HPAGE << REGION_SHIFT)
-
 static long	htlbpagemem;
 int		htlbpage_max;
 static long	htlbzone_pages;
+unsigned int	hpage_shift=HPAGE_SHIFT_DEFAULT;
 
 static struct list_head hugepage_freelists[MAX_NUMNODES];
 static spinlock_t htlbpage_lock = SPIN_LOCK_UNLOCKED;
@@ -407,7 +410,7 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, u
 		return -EINVAL;
 	/* This code assumes that REGION_HPAGE != 0. */
 	if ((REGION_NUMBER(addr) != REGION_HPAGE) || (addr & (HPAGE_SIZE - 1)))
-		addr = TASK_HPAGE_BASE;
+		addr = HPAGE_REGION_BASE;
 	else
 		addr = ALIGN(addr, HPAGE_SIZE);
 	for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
@@ -520,6 +523,35 @@ static int __init hugetlb_setup(char *s)
 }
 __setup("hugepages=", hugetlb_setup);
 
+static int __init hugetlb_setup_sz(char *str)
+{
+	u64 tr_pages;
+	unsigned long long size;
+
+	if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0)
+		/*
+		 * shouldn't happen, but just in case.
+		 */
+		tr_pages = 0x15557000UL;
+
+	size = memparse(str, &str);
+	if (*str || (size & (size-1)) || !(tr_pages & size) ||
+		size <= PAGE_SIZE ||
+		size >= (1UL << PAGE_SHIFT << MAX_ORDER)) {
+		printk(KERN_WARNING "Invalid huge page size specified\n");
+		return 1;
+	}
+
+	hpage_shift = __ffs(size);
+	/*
+	 * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
+	 * override here with new page shift.
+	 */
+	ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2);
+	return 1;
+}
+__setup("hugepagesz=", hugetlb_setup_sz);
+
 static int __init hugetlb_init(void)
 {
 	int i;
@@ -540,7 +572,7 @@ static int __init hugetlb_init(void)
 	printk("Total HugeTLB memory allocated, %ld\n", htlbpagemem);
 	return 0;
 }
-module_init(hugetlb_init);
+__initcall(hugetlb_init);
 
 int hugetlb_report_meminfo(char *buf)
 {

arch/ia64/mm/init.c

@@ -342,6 +342,10 @@ ia64_mmu_init (void *my_cpu_data)
 
 	ia64_tlb_init();
 
+#ifdef	CONFIG_HUGETLB_PAGE
+	ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
+#endif
+
 #ifdef	CONFIG_IA64_MCA
 	cpu = smp_processor_id();
 

include/asm-ia64/mmu_context.h

@@ -140,8 +140,9 @@ reload_context (mm_context_t context)
 {
 	unsigned long rid;
 	unsigned long rid_incr = 0;
-	unsigned long rr0, rr1, rr2, rr3, rr4;
+	unsigned long rr0, rr1, rr2, rr3, rr4, old_rr4;
 
+	old_rr4 = ia64_get_rr(0x8000000000000000);
 	rid = context << 3;	/* make space for encoding the region number */
 	rid_incr = 1 << 8;
 
@@ -152,7 +153,7 @@ reload_context (mm_context_t context)
 	rr3 = rr0 + 3*rid_incr;
 	rr4 = rr0 + 4*rid_incr;
 #ifdef  CONFIG_HUGETLB_PAGE
-	rr4 = (rr4 & (~(0xfcUL))) | (HPAGE_SHIFT << 2);
+	rr4 = (rr4 & (~(0xfcUL))) | (old_rr4 & 0xfc);
 #endif
 
 	ia64_set_rr(0x0000000000000000, rr0);

include/asm-ia64/page.h

@@ -37,31 +37,14 @@
 #define RGN_MAP_LIMIT	((1UL << (4*PAGE_SHIFT - 12)) - PAGE_SIZE)	/* per region addr limit */
 
 #ifdef CONFIG_HUGETLB_PAGE
+# define REGION_HPAGE		(4UL)	/* note: this is hardcoded in reload_context()!*/
+# define REGION_SHIFT		61
+# define HPAGE_REGION_BASE	(REGION_HPAGE << REGION_SHIFT)
+# define HPAGE_SHIFT		hpage_shift
+# define HPAGE_SHIFT_DEFAULT	28	/* check ia64 SDM for architecture supported size */
+# define HPAGE_SIZE		(__IA64_UL_CONST(1) << HPAGE_SHIFT)
+# define HPAGE_MASK		(~(HPAGE_SIZE - 1))
 
-# if defined(CONFIG_HUGETLB_PAGE_SIZE_4GB)
-#  define HPAGE_SHIFT	32
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_1GB)
-#  define HPAGE_SHIFT	30
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_256MB)
-#  define HPAGE_SHIFT	28
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_64MB)
-#  define HPAGE_SHIFT	26
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_16MB)
-#  define HPAGE_SHIFT	24
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_4MB)
-#  define HPAGE_SHIFT	22
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_1MB)
-#  define HPAGE_SHIFT	20
-# elif defined(CONFIG_HUGETLB_PAGE_SIZE_256KB)
-#  define HPAGE_SHIFT	18
-# else
-#  error Unsupported IA-64 HugeTLB Page Size!
-# endif
-
-# define REGION_HPAGE	(4UL)	/* note: this is hardcoded in mmu_context.h:reload_context()!*/
-# define REGION_SHIFT	61
-# define HPAGE_SIZE	(__IA64_UL_CONST(1) << HPAGE_SHIFT)
-# define HPAGE_MASK	(~(HPAGE_SIZE - 1))
 # define HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 # define ARCH_HAS_HUGEPAGE_ONLY_RANGE
 #endif /* CONFIG_HUGETLB_PAGE */
@@ -140,6 +123,7 @@ typedef union ia64_va {
 # define is_hugepage_only_range(addr, len)		\
 	 (REGION_NUMBER(addr) == REGION_HPAGE &&	\
 	  REGION_NUMBER((addr)+(len)) == REGION_HPAGE)
+extern unsigned int hpage_shift;
 #endif
 
 static __inline__ int