From be5a0c126ad1dea2128dc5aef12c87083518d1ab Mon Sep 17 00:00:00 2001 From: "venkatesh.pallipadi@intel.com" Date: Wed, 10 Feb 2010 11:57:06 -0800 Subject: x86, pat: Preparatory changes in pat.c for bigger rbtree change Minor changes in pat.c to cleanup code and make it smoother to introduce bigger rbtree only change in the following patch. The changes are cleaup only and should not have any functional impact. Signed-off-by: Venkatesh Pallipadi LKML-Reference: <20100210195909.792781000@intel.com> Signed-off-by: Suresh Siddha Signed-off-by: H. Peter Anvin --- arch/x86/mm/pat.c | 170 +++++++++++++++++++++++---------------------- arch/x86/mm/pat_internal.h | 28 ++++++++ 2 files changed, 116 insertions(+), 82 deletions(-) create mode 100644 arch/x86/mm/pat_internal.h (limited to 'arch/x86/mm') diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c index ae9648eb1c7f..628e507b7936 100644 --- a/arch/x86/mm/pat.c +++ b/arch/x86/mm/pat.c @@ -30,6 +30,8 @@ #include #include +#include "pat_internal.h" + #ifdef CONFIG_X86_PAT int __read_mostly pat_enabled = 1; @@ -53,19 +55,15 @@ static inline void pat_disable(const char *reason) #endif -static int debug_enable; +int pat_debug_enable; static int __init pat_debug_setup(char *str) { - debug_enable = 1; + pat_debug_enable = 1; return 0; } __setup("debugpat", pat_debug_setup); -#define dprintk(fmt, arg...) \ - do { if (debug_enable) printk(KERN_INFO fmt, ##arg); } while (0) - - static u64 __read_mostly boot_pat_state; enum { @@ -132,17 +130,6 @@ void pat_init(void) #undef PAT -static char *cattr_name(unsigned long flags) -{ - switch (flags & _PAGE_CACHE_MASK) { - case _PAGE_CACHE_UC: return "uncached"; - case _PAGE_CACHE_UC_MINUS: return "uncached-minus"; - case _PAGE_CACHE_WB: return "write-back"; - case _PAGE_CACHE_WC: return "write-combining"; - default: return "broken"; - } -} - /* * The global memtype list keeps track of memory type for specific * physical memory areas. Conflicting memory types in different @@ -159,14 +146,6 @@ static char *cattr_name(unsigned long flags) * memtype_lock protects both the linear list and rbtree. */ -struct memtype { - u64 start; - u64 end; - unsigned long type; - struct list_head nd; - struct rb_node rb; -}; - static struct rb_root memtype_rbroot = RB_ROOT; static LIST_HEAD(memtype_list); static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ @@ -349,6 +328,64 @@ static int free_ram_pages_type(u64 start, u64 end) return 0; } +static int memtype_check_insert(struct memtype *new, unsigned long *new_type) +{ + struct memtype *entry; + u64 start, end; + unsigned long actual_type; + struct list_head *where; + int err = 0; + + start = new->start; + end = new->end; + actual_type = new->type; + + /* Search for existing mapping that overlaps the current range */ + where = NULL; + list_for_each_entry(entry, &memtype_list, nd) { + if (end <= entry->start) { + where = entry->nd.prev; + break; + } else if (start <= entry->start) { /* end > entry->start */ + err = chk_conflict(new, entry, new_type); + if (!err) { + dprintk("Overlap at 0x%Lx-0x%Lx\n", + entry->start, entry->end); + where = entry->nd.prev; + } + break; + } else if (start < entry->end) { /* start > entry->start */ + err = chk_conflict(new, entry, new_type); + if (!err) { + dprintk("Overlap at 0x%Lx-0x%Lx\n", + entry->start, entry->end); + + /* + * Move to right position in the linked + * list to add this new entry + */ + list_for_each_entry_continue(entry, + &memtype_list, nd) { + if (start <= entry->start) { + where = entry->nd.prev; + break; + } + } + } + break; + } + } + if (!err) { + if (where) + list_add(&new->nd, where); + else + list_add_tail(&new->nd, &memtype_list); + + memtype_rb_insert(&memtype_rbroot, new); + } + return err; +} + /* * req_type typically has one of the: * - _PAGE_CACHE_WB @@ -364,9 +401,8 @@ static int free_ram_pages_type(u64 start, u64 end) int reserve_memtype(u64 start, u64 end, unsigned long req_type, unsigned long *new_type) { - struct memtype *new, *entry; + struct memtype *new; unsigned long actual_type; - struct list_head *where; int is_range_ram; int err = 0; @@ -423,42 +459,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, spin_lock(&memtype_lock); - /* Search for existing mapping that overlaps the current range */ - where = NULL; - list_for_each_entry(entry, &memtype_list, nd) { - if (end <= entry->start) { - where = entry->nd.prev; - break; - } else if (start <= entry->start) { /* end > entry->start */ - err = chk_conflict(new, entry, new_type); - if (!err) { - dprintk("Overlap at 0x%Lx-0x%Lx\n", - entry->start, entry->end); - where = entry->nd.prev; - } - break; - } else if (start < entry->end) { /* start > entry->start */ - err = chk_conflict(new, entry, new_type); - if (!err) { - dprintk("Overlap at 0x%Lx-0x%Lx\n", - entry->start, entry->end); - - /* - * Move to right position in the linked - * list to add this new entry - */ - list_for_each_entry_continue(entry, - &memtype_list, nd) { - if (start <= entry->start) { - where = entry->nd.prev; - break; - } - } - } - break; - } - } - + err = memtype_check_insert(new, new_type); if (err) { printk(KERN_INFO "reserve_memtype failed 0x%Lx-0x%Lx, " "track %s, req %s\n", @@ -469,13 +470,6 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, return err; } - if (where) - list_add(&new->nd, where); - else - list_add_tail(&new->nd, &memtype_list); - - memtype_rb_insert(&memtype_rbroot, new); - spin_unlock(&memtype_lock); dprintk("reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n", @@ -937,28 +931,40 @@ EXPORT_SYMBOL_GPL(pgprot_writecombine); #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) /* get Nth element of the linked list */ -static struct memtype *memtype_get_idx(loff_t pos) +static int copy_memtype_nth_element(struct memtype *out, loff_t pos) { - struct memtype *list_node, *print_entry; + struct memtype *list_node; int i = 1; - print_entry = kmalloc(sizeof(struct memtype), GFP_KERNEL); - if (!print_entry) - return NULL; - - spin_lock(&memtype_lock); list_for_each_entry(list_node, &memtype_list, nd) { if (pos == i) { - *print_entry = *list_node; - spin_unlock(&memtype_lock); - return print_entry; + *out = *list_node; + return 0; } ++i; } + return 1; +} + +static struct memtype *memtype_get_idx(loff_t pos) +{ + struct memtype *print_entry; + int ret; + + print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL); + if (!print_entry) + return NULL; + + spin_lock(&memtype_lock); + ret = copy_memtype_nth_element(print_entry, pos); spin_unlock(&memtype_lock); - kfree(print_entry); - return NULL; + if (!ret) { + return print_entry; + } else { + kfree(print_entry); + return NULL; + } } static void *memtype_seq_start(struct seq_file *seq, loff_t *pos) diff --git a/arch/x86/mm/pat_internal.h b/arch/x86/mm/pat_internal.h new file mode 100644 index 000000000000..6c98780eb731 --- /dev/null +++ b/arch/x86/mm/pat_internal.h @@ -0,0 +1,28 @@ +#ifndef __PAT_INTERNAL_H_ +#define __PAT_INTERNAL_H_ + +extern int pat_debug_enable; + +#define dprintk(fmt, arg...) \ + do { if (pat_debug_enable) printk(KERN_INFO fmt, ##arg); } while (0) + +struct memtype { + u64 start; + u64 end; + unsigned long type; + struct list_head nd; + struct rb_node rb; +}; + +static inline char *cattr_name(unsigned long flags) +{ + switch (flags & _PAGE_CACHE_MASK) { + case _PAGE_CACHE_UC: return "uncached"; + case _PAGE_CACHE_UC_MINUS: return "uncached-minus"; + case _PAGE_CACHE_WB: return "write-back"; + case _PAGE_CACHE_WC: return "write-combining"; + default: return "broken"; + } +} + +#endif /* __PAT_INTERNAL_H_ */ -- cgit v1.2.3 From 9e41a49aab88a5a6c8f4875bf10a5543bc321f2d Mon Sep 17 00:00:00 2001 From: "Pallipadi, Venkatesh" Date: Wed, 10 Feb 2010 15:26:07 -0800 Subject: x86, pat: Migrate to rbtree only backend for pat memtype management Move pat backend to fully rbtree based implementation from the existing rbtree and linked list hybrid. New rbtree based solution uses interval trees (augmented rbtrees) in order to store the PAT ranges. The new code seprates out the pat backend to pat_rbtree.c file, making is cleaner. The change also makes the PAT lookup, reserve and free operations more optimal, as we don't have to traverse linear linked list of few tens of entries in normal case. Signed-off-by: Venkatesh Pallipadi LKML-Reference: <20100210232607.GB11465@linux-os.sc.intel.com> Signed-off-by: Suresh Siddha Signed-off-by: H. Peter Anvin --- arch/x86/mm/Makefile | 1 + arch/x86/mm/pat.c | 209 +--------------------------------- arch/x86/mm/pat_internal.h | 20 +++- arch/x86/mm/pat_rbtree.c | 271 +++++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 296 insertions(+), 205 deletions(-) create mode 100644 arch/x86/mm/pat_rbtree.c (limited to 'arch/x86/mm') diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile index 06630d26e56d..a4c768397baa 100644 --- a/arch/x86/mm/Makefile +++ b/arch/x86/mm/Makefile @@ -6,6 +6,7 @@ nostackp := $(call cc-option, -fno-stack-protector) CFLAGS_physaddr.o := $(nostackp) CFLAGS_setup_nx.o := $(nostackp) +obj-$(CONFIG_X86_PAT) += pat_rbtree.o obj-$(CONFIG_SMP) += tlb.o obj-$(CONFIG_X86_32) += pgtable_32.o iomap_32.o diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c index 628e507b7936..951011166ef5 100644 --- a/arch/x86/mm/pat.c +++ b/arch/x86/mm/pat.c @@ -130,65 +130,7 @@ void pat_init(void) #undef PAT -/* - * The global memtype list keeps track of memory type for specific - * physical memory areas. Conflicting memory types in different - * mappings can cause CPU cache corruption. To avoid this we keep track. - * - * The list is sorted based on starting address and can contain multiple - * entries for each address (this allows reference counting for overlapping - * areas). All the aliases have the same cache attributes of course. - * Zero attributes are represented as holes. - * - * The data structure is a list that is also organized as an rbtree - * sorted on the start address of memtype range. - * - * memtype_lock protects both the linear list and rbtree. - */ - -static struct rb_root memtype_rbroot = RB_ROOT; -static LIST_HEAD(memtype_list); -static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ - -static struct memtype *memtype_rb_search(struct rb_root *root, u64 start) -{ - struct rb_node *node = root->rb_node; - struct memtype *last_lower = NULL; - - while (node) { - struct memtype *data = container_of(node, struct memtype, rb); - - if (data->start < start) { - last_lower = data; - node = node->rb_right; - } else if (data->start > start) { - node = node->rb_left; - } else - return data; - } - - /* Will return NULL if there is no entry with its start <= start */ - return last_lower; -} - -static void memtype_rb_insert(struct rb_root *root, struct memtype *data) -{ - struct rb_node **new = &(root->rb_node); - struct rb_node *parent = NULL; - - while (*new) { - struct memtype *this = container_of(*new, struct memtype, rb); - - parent = *new; - if (data->start <= this->start) - new = &((*new)->rb_left); - else if (data->start > this->start) - new = &((*new)->rb_right); - } - - rb_link_node(&data->rb, parent, new); - rb_insert_color(&data->rb, root); -} +static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */ /* * Does intersection of PAT memory type and MTRR memory type and returns @@ -216,33 +158,6 @@ static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type) return req_type; } -static int -chk_conflict(struct memtype *new, struct memtype *entry, unsigned long *type) -{ - if (new->type != entry->type) { - if (type) { - new->type = entry->type; - *type = entry->type; - } else - goto conflict; - } - - /* check overlaps with more than one entry in the list */ - list_for_each_entry_continue(entry, &memtype_list, nd) { - if (new->end <= entry->start) - break; - else if (new->type != entry->type) - goto conflict; - } - return 0; - - conflict: - printk(KERN_INFO "%s:%d conflicting memory types " - "%Lx-%Lx %s<->%s\n", current->comm, current->pid, new->start, - new->end, cattr_name(new->type), cattr_name(entry->type)); - return -EBUSY; -} - static int pat_pagerange_is_ram(unsigned long start, unsigned long end) { int ram_page = 0, not_rampage = 0; @@ -328,64 +243,6 @@ static int free_ram_pages_type(u64 start, u64 end) return 0; } -static int memtype_check_insert(struct memtype *new, unsigned long *new_type) -{ - struct memtype *entry; - u64 start, end; - unsigned long actual_type; - struct list_head *where; - int err = 0; - - start = new->start; - end = new->end; - actual_type = new->type; - - /* Search for existing mapping that overlaps the current range */ - where = NULL; - list_for_each_entry(entry, &memtype_list, nd) { - if (end <= entry->start) { - where = entry->nd.prev; - break; - } else if (start <= entry->start) { /* end > entry->start */ - err = chk_conflict(new, entry, new_type); - if (!err) { - dprintk("Overlap at 0x%Lx-0x%Lx\n", - entry->start, entry->end); - where = entry->nd.prev; - } - break; - } else if (start < entry->end) { /* start > entry->start */ - err = chk_conflict(new, entry, new_type); - if (!err) { - dprintk("Overlap at 0x%Lx-0x%Lx\n", - entry->start, entry->end); - - /* - * Move to right position in the linked - * list to add this new entry - */ - list_for_each_entry_continue(entry, - &memtype_list, nd) { - if (start <= entry->start) { - where = entry->nd.prev; - break; - } - } - } - break; - } - } - if (!err) { - if (where) - list_add(&new->nd, where); - else - list_add_tail(&new->nd, &memtype_list); - - memtype_rb_insert(&memtype_rbroot, new); - } - return err; -} - /* * req_type typically has one of the: * - _PAGE_CACHE_WB @@ -459,7 +316,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, spin_lock(&memtype_lock); - err = memtype_check_insert(new, new_type); + err = rbt_memtype_check_insert(new, new_type); if (err) { printk(KERN_INFO "reserve_memtype failed 0x%Lx-0x%Lx, " "track %s, req %s\n", @@ -481,7 +338,6 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, int free_memtype(u64 start, u64 end) { - struct memtype *entry, *saved_entry; int err = -EINVAL; int is_range_ram; @@ -505,46 +361,7 @@ int free_memtype(u64 start, u64 end) } spin_lock(&memtype_lock); - - entry = memtype_rb_search(&memtype_rbroot, start); - if (unlikely(entry == NULL)) - goto unlock_ret; - - /* - * Saved entry points to an entry with start same or less than what - * we searched for. Now go through the list in both directions to look - * for the entry that matches with both start and end, with list stored - * in sorted start address - */ - saved_entry = entry; - list_for_each_entry_from(entry, &memtype_list, nd) { - if (entry->start == start && entry->end == end) { - rb_erase(&entry->rb, &memtype_rbroot); - list_del(&entry->nd); - kfree(entry); - err = 0; - break; - } else if (entry->start > start) { - break; - } - } - - if (!err) - goto unlock_ret; - - entry = saved_entry; - list_for_each_entry_reverse(entry, &memtype_list, nd) { - if (entry->start == start && entry->end == end) { - rb_erase(&entry->rb, &memtype_rbroot); - list_del(&entry->nd); - kfree(entry); - err = 0; - break; - } else if (entry->start < start) { - break; - } - } -unlock_ret: + err = rbt_memtype_erase(start, end); spin_unlock(&memtype_lock); if (err) { @@ -593,7 +410,7 @@ static unsigned long lookup_memtype(u64 paddr) spin_lock(&memtype_lock); - entry = memtype_rb_search(&memtype_rbroot, paddr); + entry = rbt_memtype_lookup(paddr); if (entry != NULL) rettype = entry->type; else @@ -930,22 +747,6 @@ EXPORT_SYMBOL_GPL(pgprot_writecombine); #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT) -/* get Nth element of the linked list */ -static int copy_memtype_nth_element(struct memtype *out, loff_t pos) -{ - struct memtype *list_node; - int i = 1; - - list_for_each_entry(list_node, &memtype_list, nd) { - if (pos == i) { - *out = *list_node; - return 0; - } - ++i; - } - return 1; -} - static struct memtype *memtype_get_idx(loff_t pos) { struct memtype *print_entry; @@ -956,7 +757,7 @@ static struct memtype *memtype_get_idx(loff_t pos) return NULL; spin_lock(&memtype_lock); - ret = copy_memtype_nth_element(print_entry, pos); + ret = rbt_memtype_copy_nth_element(print_entry, pos); spin_unlock(&memtype_lock); if (!ret) { diff --git a/arch/x86/mm/pat_internal.h b/arch/x86/mm/pat_internal.h index 6c98780eb731..4f39eefa3e61 100644 --- a/arch/x86/mm/pat_internal.h +++ b/arch/x86/mm/pat_internal.h @@ -9,8 +9,8 @@ extern int pat_debug_enable; struct memtype { u64 start; u64 end; + u64 subtree_max_end; unsigned long type; - struct list_head nd; struct rb_node rb; }; @@ -25,4 +25,22 @@ static inline char *cattr_name(unsigned long flags) } } +#ifdef CONFIG_X86_PAT +extern int rbt_memtype_check_insert(struct memtype *new, + unsigned long *new_type); +extern int rbt_memtype_erase(u64 start, u64 end); +extern struct memtype *rbt_memtype_lookup(u64 addr); +extern int rbt_memtype_copy_nth_element(struct memtype *out, loff_t pos); +#else +static inline int rbt_memtype_check_insert(struct memtype *new, + unsigned long *new_type) +{ return 0; } +static inline int rbt_memtype_erase(u64 start, u64 end) +{ return 0; } +static inline struct memtype *rbt_memtype_lookup(u64 addr) +{ return NULL; } +static inline int rbt_memtype_copy_nth_element(struct memtype *out, loff_t pos) +{ return 0; } +#endif + #endif /* __PAT_INTERNAL_H_ */ diff --git a/arch/x86/mm/pat_rbtree.c b/arch/x86/mm/pat_rbtree.c new file mode 100644 index 000000000000..9063f40b638b --- /dev/null +++ b/arch/x86/mm/pat_rbtree.c @@ -0,0 +1,271 @@ +/* + * Handle caching attributes in page tables (PAT) + * + * Authors: Venkatesh Pallipadi + * Suresh B Siddha + * + * Interval tree (augmented rbtree) used to store the PAT memory type + * reservations. + */ + +#include +#include +#include +#include +#include +#include +#include + +#include +#include + +#include "pat_internal.h" + +/* + * The memtype tree keeps track of memory type for specific + * physical memory areas. Without proper tracking, conflicting memory + * types in different mappings can cause CPU cache corruption. + * + * The tree is an interval tree (augmented rbtree) with tree ordered + * on starting address. Tree can contain multiple entries for + * different regions which overlap. All the aliases have the same + * cache attributes of course. + * + * memtype_lock protects the rbtree. + */ + +static void memtype_rb_augment_cb(struct rb_node *node); +static struct rb_root memtype_rbroot = RB_AUGMENT_ROOT(&memtype_rb_augment_cb); + +static int is_node_overlap(struct memtype *node, u64 start, u64 end) +{ + if (node->start >= end || node->end <= start) + return 0; + + return 1; +} + +static u64 get_subtree_max_end(struct rb_node *node) +{ + u64 ret = 0; + if (node) { + struct memtype *data = container_of(node, struct memtype, rb); + ret = data->subtree_max_end; + } + return ret; +} + +/* Update 'subtree_max_end' for a node, based on node and its children */ +static void update_node_max_end(struct rb_node *node) +{ + struct memtype *data; + u64 max_end, child_max_end; + + if (!node) + return; + + data = container_of(node, struct memtype, rb); + max_end = data->end; + + child_max_end = get_subtree_max_end(node->rb_right); + if (child_max_end > max_end) + max_end = child_max_end; + + child_max_end = get_subtree_max_end(node->rb_left); + if (child_max_end > max_end) + max_end = child_max_end; + + data->subtree_max_end = max_end; +} + +/* Update 'subtree_max_end' for a node and all its ancestors */ +static void update_path_max_end(struct rb_node *node) +{ + u64 old_max_end, new_max_end; + + while (node) { + struct memtype *data = container_of(node, struct memtype, rb); + + old_max_end = data->subtree_max_end; + update_node_max_end(node); + new_max_end = data->subtree_max_end; + + if (new_max_end == old_max_end) + break; + + node = rb_parent(node); + } +} + +/* Find the first (lowest start addr) overlapping range from rb tree */ +static struct memtype *memtype_rb_lowest_match(struct rb_root *root, + u64 start, u64 end) +{ + struct rb_node *node = root->rb_node; + struct memtype *last_lower = NULL; + + while (node) { + struct memtype *data = container_of(node, struct memtype, rb); + + if (get_subtree_max_end(node->rb_left) > start) { + /* Lowest overlap if any must be on left side */ + node = node->rb_left; + } else if (is_node_overlap(data, start, end)) { + last_lower = data; + break; + } else if (start >= data->start) { + /* Lowest overlap if any must be on right side */ + node = node->rb_right; + } else { + break; + } + } + return last_lower; /* Returns NULL if there is no overlap */ +} + +static struct memtype *memtype_rb_exact_match(struct rb_root *root, + u64 start, u64 end) +{ + struct memtype *match; + + match = memtype_rb_lowest_match(root, start, end); + while (match != NULL && match->start < end) { + struct rb_node *node; + + if (match->start == start && match->end == end) + return match; + + node = rb_next(&match->rb); + if (node) + match = container_of(node, struct memtype, rb); + else + match = NULL; + } + + return NULL; /* Returns NULL if there is no exact match */ +} + +static int memtype_rb_check_conflict(struct rb_root *root, + u64 start, u64 end, + unsigned long reqtype, unsigned long *newtype) +{ + struct rb_node *node; + struct memtype *match; + int found_type = reqtype; + + match = memtype_rb_lowest_match(&memtype_rbroot, start, end); + if (match == NULL) + goto success; + + if (match->type != found_type && newtype == NULL) + goto failure; + + dprintk("Overlap at 0x%Lx-0x%Lx\n", match->start, match->end); + found_type = match->type; + + node = rb_next(&match->rb); + while (node) { + match = container_of(node, struct memtype, rb); + + if (match->start >= end) /* Checked all possible matches */ + goto success; + + if (is_node_overlap(match, start, end) && + match->type != found_type) { + goto failure; + } + + node = rb_next(&match->rb); + } +success: + if (newtype) + *newtype = found_type; + + return 0; + +failure: + printk(KERN_INFO "%s:%d conflicting memory types " + "%Lx-%Lx %s<->%s\n", current->comm, current->pid, start, + end, cattr_name(found_type), cattr_name(match->type)); + return -EBUSY; +} + +static void memtype_rb_augment_cb(struct rb_node *node) +{ + if (node) + update_path_max_end(node); +} + +static void memtype_rb_insert(struct rb_root *root, struct memtype *newdata) +{ + struct rb_node **node = &(root->rb_node); + struct rb_node *parent = NULL; + + while (*node) { + struct memtype *data = container_of(*node, struct memtype, rb); + + parent = *node; + if (newdata->start <= data->start) + node = &((*node)->rb_left); + else if (newdata->start > data->start) + node = &((*node)->rb_right); + } + + rb_link_node(&newdata->rb, parent, node); + rb_insert_color(&newdata->rb, root); +} + +int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type) +{ + int err = 0; + + err = memtype_rb_check_conflict(&memtype_rbroot, new->start, new->end, + new->type, ret_type); + + if (!err) { + new->type = *ret_type; + memtype_rb_insert(&memtype_rbroot, new); + } + return err; +} + +int rbt_memtype_erase(u64 start, u64 end) +{ + struct memtype *data; + + data = memtype_rb_exact_match(&memtype_rbroot, start, end); + if (!data) + return -EINVAL; + + rb_erase(&data->rb, &memtype_rbroot); + return 0; +} + +struct memtype *rbt_memtype_lookup(u64 addr) +{ + struct memtype *data; + data = memtype_rb_lowest_match(&memtype_rbroot, addr, addr + PAGE_SIZE); + return data; +} + +#if defined(CONFIG_DEBUG_FS) +int rbt_memtype_copy_nth_element(struct memtype *out, loff_t pos) +{ + struct rb_node *node; + int i = 1; + + node = rb_first(&memtype_rbroot); + while (node && pos != i) { + node = rb_next(node); + i++; + } + + if (node) { /* pos == i */ + struct memtype *this = container_of(node, struct memtype, rb); + *out = *this; + return 0; + } else { + return 1; + } +} +#endif -- cgit v1.2.3 From 4daa2a8093ecd1148270a1fc64e99f072b8c2901 Mon Sep 17 00:00:00 2001 From: "Pallipadi, Venkatesh" Date: Wed, 24 Feb 2010 13:43:55 -0800 Subject: x86, pat: In rbt_memtype_check_insert(), update new->type only if valid new->type should only change when there is a valid ret_type. Otherwise the requested type and return type should be same. Signed-off-by: Venkatesh Pallipadi LKML-Reference: <20100224214355.GA16431@linux-os.sc.intel.com> Tested-by: Jack Steiner Signed-off-by: H. Peter Anvin --- arch/x86/mm/pat_rbtree.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'arch/x86/mm') diff --git a/arch/x86/mm/pat_rbtree.c b/arch/x86/mm/pat_rbtree.c index 9063f40b638b..07de4cb8cc30 100644 --- a/arch/x86/mm/pat_rbtree.c +++ b/arch/x86/mm/pat_rbtree.c @@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type) new->type, ret_type); if (!err) { - new->type = *ret_type; + if (ret_type) + new->type = *ret_type; + memtype_rb_insert(&memtype_rbroot, new); } return err; -- cgit v1.2.3 From 1f9cc3cb6a27521edfe0a21abf97d2bb11c4d237 Mon Sep 17 00:00:00 2001 From: Robin Holt Date: Fri, 23 Apr 2010 10:36:22 -0500 Subject: x86, pat: Update the page flags for memtype atomically instead of using memtype_lock While testing an application using the xpmem (out of kernel) driver, we noticed a significant page fault rate reduction of x86_64 with respect to ia64. For one test running with 32 cpus, one thread per cpu, it took 01:08 for each of the threads to vm_insert_pfn 2GB worth of pages. For the same test running on 256 cpus, one thread per cpu, it took 14:48 to vm_insert_pfn 2 GB worth of pages. The slowdown was tracked to lookup_memtype which acquires the spinlock memtype_lock. This heavily contended lock was slowing down vm_insert_pfn(). With the cmpxchg on page->flags method, both the 32 cpu and 256 cpu cases take approx 00:01.3 seconds to complete. Signed-off-by: Robin Holt LKML-Reference: <20100423153627.751194346@gulag1.americas.sgi.com> Cc: Venkatesh Pallipadi Cc: Rafael Wysocki Reviewed-by: Suresh Siddha Signed-off-by: H. Peter Anvin --- arch/x86/include/asm/cacheflush.h | 44 ++++++++++++++++++++++----------------- arch/x86/mm/pat.c | 8 ------- 2 files changed, 25 insertions(+), 27 deletions(-) (limited to 'arch/x86/mm') diff --git a/arch/x86/include/asm/cacheflush.h b/arch/x86/include/asm/cacheflush.h index 634c40a739a6..c70068d05f70 100644 --- a/arch/x86/include/asm/cacheflush.h +++ b/arch/x86/include/asm/cacheflush.h @@ -44,9 +44,6 @@ static inline void copy_from_user_page(struct vm_area_struct *vma, memcpy(dst, src, len); } -#define PG_WC PG_arch_1 -PAGEFLAG(WC, WC) - #ifdef CONFIG_X86_PAT /* * X86 PAT uses page flags WC and Uncached together to keep track of @@ -55,16 +52,24 @@ PAGEFLAG(WC, WC) * _PAGE_CACHE_UC_MINUS and fourth state where page's memory type has not * been changed from its default (value of -1 used to denote this). * Note we do not support _PAGE_CACHE_UC here. - * - * Caller must hold memtype_lock for atomicity. */ + +#define _PGMT_DEFAULT 0 +#define _PGMT_WC (1UL << PG_arch_1) +#define _PGMT_UC_MINUS (1UL << PG_uncached) +#define _PGMT_WB (1UL << PG_uncached | 1UL << PG_arch_1) +#define _PGMT_MASK (1UL << PG_uncached | 1UL << PG_arch_1) +#define _PGMT_CLEAR_MASK (~_PGMT_MASK) + static inline unsigned long get_page_memtype(struct page *pg) { - if (!PageUncached(pg) && !PageWC(pg)) + unsigned long pg_flags = pg->flags & _PGMT_MASK; + + if (pg_flags == _PGMT_DEFAULT) return -1; - else if (!PageUncached(pg) && PageWC(pg)) + else if (pg_flags == _PGMT_WC) return _PAGE_CACHE_WC; - else if (PageUncached(pg) && !PageWC(pg)) + else if (pg_flags == _PGMT_UC_MINUS) return _PAGE_CACHE_UC_MINUS; else return _PAGE_CACHE_WB; @@ -72,25 +77,26 @@ static inline unsigned long get_page_memtype(struct page *pg) static inline void set_page_memtype(struct page *pg, unsigned long memtype) { + unsigned long memtype_flags = _PGMT_DEFAULT; + unsigned long old_flags; + unsigned long new_flags; + switch (memtype) { case _PAGE_CACHE_WC: - ClearPageUncached(pg); - SetPageWC(pg); + memtype_flags = _PGMT_WC; break; case _PAGE_CACHE_UC_MINUS: - SetPageUncached(pg); - ClearPageWC(pg); + memtype_flags = _PGMT_UC_MINUS; break; case _PAGE_CACHE_WB: - SetPageUncached(pg); - SetPageWC(pg); - break; - default: - case -1: - ClearPageUncached(pg); - ClearPageWC(pg); + memtype_flags = _PGMT_WB; break; } + + do { + old_flags = pg->flags; + new_flags = (old_flags & _PGMT_CLEAR_MASK) | memtype_flags; + } while (cmpxchg(&pg->flags, old_flags, new_flags) != old_flags); } #else static inline unsigned long get_page_memtype(struct page *pg) { return -1; } diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c index 951011166ef5..501fc60e5e4d 100644 --- a/arch/x86/mm/pat.c +++ b/arch/x86/mm/pat.c @@ -190,8 +190,6 @@ static int pat_pagerange_is_ram(unsigned long start, unsigned long end) * Here we do two pass: * - Find the memtype of all the pages in the range, look for any conflicts * - In case of no conflicts, set the new memtype for pages in the range - * - * Caller must hold memtype_lock for atomicity. */ static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type, unsigned long *new_type) @@ -297,9 +295,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type, is_range_ram = pat_pagerange_is_ram(start, end); if (is_range_ram == 1) { - spin_lock(&memtype_lock); err = reserve_ram_pages_type(start, end, req_type, new_type); - spin_unlock(&memtype_lock); return err; } else if (is_range_ram < 0) { @@ -351,9 +347,7 @@ int free_memtype(u64 start, u64 end) is_range_ram = pat_pagerange_is_ram(start, end); if (is_range_ram == 1) { - spin_lock(&memtype_lock); err = free_ram_pages_type(start, end); - spin_unlock(&memtype_lock); return err; } else if (is_range_ram < 0) { @@ -394,10 +388,8 @@ static unsigned long lookup_memtype(u64 paddr) if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) { struct page *page; - spin_lock(&memtype_lock); page = pfn_to_page(paddr >> PAGE_SHIFT); rettype = get_page_memtype(page); - spin_unlock(&memtype_lock); /* * -1 from get_page_memtype() implies RAM page is in its * default state and not reserved, and hence of type WB -- cgit v1.2.3