From 60c371bc753495f36d3a71338b46030f7fffce3b Mon Sep 17 00:00:00 2001 From: Andrew Morton Date: Sat, 5 Aug 2006 12:14:25 -0700 Subject: [PATCH] fadvise() make POSIX_FADV_NOREUSE a no-op The POSIX_FADV_NOREUSE hint means "the application will use this range of the file a single time". It seems to be intended that the implementation will use this hint to perform drop-behind of that part of the file when the application gets around to reading or writing it. However for reasons which aren't obvious (or sane?) I mapped POSIX_FADV_NOREUSE onto POSIX_FADV_WILLNEED. ie: it does readahead. That's daft. So for now, make POSIX_FADV_NOREUSE a no-op. This is a non-back-compatible change. If someone was using POSIX_FADV_NOREUSE to perform readahead, they lose. The likelihood is low. If/when we later implement POSIX_FADV_NOREUSE things will get interesting - to do it fully we'll need to maintain file offset/length ranges and peform all sorts of complex tricks, and managing the lifetime of those ranges' data structures will be interesting.. A sensible implementation would probably ignore the file range and would simply mark the entire file as needing some form of drop-behind treatment. Cc: Michael Kerrisk Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/fadvise.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'mm') diff --git a/mm/fadvise.c b/mm/fadvise.c index 60a5d55e51d9..168c78a121bb 100644 --- a/mm/fadvise.c +++ b/mm/fadvise.c @@ -73,7 +73,6 @@ asmlinkage long sys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice) file->f_ra.ra_pages = bdi->ra_pages * 2; break; case POSIX_FADV_WILLNEED: - case POSIX_FADV_NOREUSE: if (!mapping->a_ops->readpage) { ret = -EINVAL; break; @@ -94,6 +93,8 @@ asmlinkage long sys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice) if (ret > 0) ret = 0; break; + case POSIX_FADV_NOREUSE: + break; case POSIX_FADV_DONTNEED: if (!bdi_write_congested(mapping->backing_dev_info)) filemap_flush(mapping); -- cgit v1.2.3 From 6f712711dbd180aa3777efe5ae3b9b0e915b9471 Mon Sep 17 00:00:00 2001 From: KAMEZAWA Hiroyuki Date: Sat, 5 Aug 2006 12:14:58 -0700 Subject: [PATCH] memory hotadd fixes: not-aligned memory hotadd handling fix ioresouce handling code in memory hotplug allows not-aligned memory hot add. But when memmap and other memory structures are initialized, parameters should be aligned. (if not aligned, initialization of mem_map will do wrong, it assumes parameters are aligned.) This patch fix it. And this patch allows ioresource collision check to handle -EEXIST. Signed-off-by: KAMEZAWA Hiroyuki Cc: Keith Mannthey Cc: Yasunori Goto Cc: Dave Hansen Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/memory_hotplug.c | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-) (limited to 'mm') diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index 01c9fb97c619..7d25cc12235f 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -76,15 +76,22 @@ int __add_pages(struct zone *zone, unsigned long phys_start_pfn, { unsigned long i; int err = 0; + int start_sec, end_sec; + /* during initialize mem_map, align hot-added range to section */ + start_sec = pfn_to_section_nr(phys_start_pfn); + end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); - for (i = 0; i < nr_pages; i += PAGES_PER_SECTION) { - err = __add_section(zone, phys_start_pfn + i); + for (i = start_sec; i <= end_sec; i++) { + err = __add_section(zone, i << PFN_SECTION_SHIFT); - /* We want to keep adding the rest of the - * sections if the first ones already exist + /* + * EEXIST is finally dealed with by ioresource collision + * check. see add_memory() => register_memory_resource() + * Warning will be printed if there is collision. */ if (err && (err != -EEXIST)) break; + err = 0; } return err; @@ -213,10 +220,10 @@ static void rollback_node_hotadd(int nid, pg_data_t *pgdat) } /* add this memory to iomem resource */ -static void register_memory_resource(u64 start, u64 size) +static int register_memory_resource(u64 start, u64 size) { struct resource *res; - + int ret = 0; res = kzalloc(sizeof(struct resource), GFP_KERNEL); BUG_ON(!res); @@ -228,7 +235,9 @@ static void register_memory_resource(u64 start, u64 size) printk("System RAM resource %llx - %llx cannot be added\n", (unsigned long long)res->start, (unsigned long long)res->end); kfree(res); + ret = -EEXIST; } + return ret; } @@ -269,7 +278,7 @@ int add_memory(int nid, u64 start, u64 size) } /* register this memory as resource */ - register_memory_resource(start, size); + ret = register_memory_resource(start, size); return ret; error: -- cgit v1.2.3 From 58c1b5b079071d82b2f924000b7e8fb5585ce7d8 Mon Sep 17 00:00:00 2001 From: KAMEZAWA Hiroyuki Date: Sat, 5 Aug 2006 12:15:01 -0700 Subject: [PATCH] memory hotadd fixes: find_next_system_ram catch range fix find_next_system_ram() is used to find available memory resource at onlining newly added memory. This patch fixes following problem. find_next_system_ram() cannot catch this case. Resource: (start)-------------(end) Section : (start)-------------(end) Signed-off-by: KAMEZAWA Hiroyuki Cc: Keith Mannthey Cc: Yasunori Goto Cc: Dave Hansen Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/resource.c | 3 ++- mm/memory_hotplug.c | 2 +- 2 files changed, 3 insertions(+), 2 deletions(-) (limited to 'mm') diff --git a/kernel/resource.c b/kernel/resource.c index 63e879379dbd..46286434af80 100644 --- a/kernel/resource.c +++ b/kernel/resource.c @@ -244,6 +244,7 @@ int find_next_system_ram(struct resource *res) start = res->start; end = res->end; + BUG_ON(start >= end); read_lock(&resource_lock); for (p = iomem_resource.child; p ; p = p->sibling) { @@ -254,7 +255,7 @@ int find_next_system_ram(struct resource *res) p = NULL; break; } - if (p->start >= start) + if ((p->end >= start) && (p->start < end)) break; } read_unlock(&resource_lock); diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index 7d25cc12235f..26f1840879d6 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -163,7 +163,7 @@ int online_pages(unsigned long pfn, unsigned long nr_pages) res.flags = IORESOURCE_MEM; /* we just need system ram */ section_end = res.end; - while (find_next_system_ram(&res) >= 0) { + while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) { start_pfn = (unsigned long)(res.start >> PAGE_SHIFT); nr_pages = (unsigned long) ((res.end + 1 - res.start) >> PAGE_SHIFT); -- cgit v1.2.3 From ebd15302dc0ba1b8761600c20854f5371e7bae1e Mon Sep 17 00:00:00 2001 From: KAMEZAWA Hiroyuki Date: Sat, 5 Aug 2006 12:15:06 -0700 Subject: [PATCH] memory hotadd fixes: enhance collision check This patch is for collision check enhancement for memory hot add. It's better to do resouce collision check before doing memory hot add, which will touch memory management structures. And add_section() should check section exists or not before calling sparse_add_one_section(). (sparse_add_one_section() will do another check anyway. but checking in memory_hotplug.c will be easy to understand.) Signed-off-by: KAMEZAWA Hiroyuki Cc: keith mannthey Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/memory_hotplug.c | 29 ++++++++++++++++++++++------- 1 file changed, 22 insertions(+), 7 deletions(-) (limited to 'mm') diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index 26f1840879d6..c37319542b70 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -52,6 +52,9 @@ static int __add_section(struct zone *zone, unsigned long phys_start_pfn) int nr_pages = PAGES_PER_SECTION; int ret; + if (pfn_valid(phys_start_pfn)) + return -EEXIST; + ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); if (ret < 0) @@ -220,10 +223,9 @@ static void rollback_node_hotadd(int nid, pg_data_t *pgdat) } /* add this memory to iomem resource */ -static int register_memory_resource(u64 start, u64 size) +static struct resource *register_memory_resource(u64 start, u64 size) { struct resource *res; - int ret = 0; res = kzalloc(sizeof(struct resource), GFP_KERNEL); BUG_ON(!res); @@ -235,9 +237,18 @@ static int register_memory_resource(u64 start, u64 size) printk("System RAM resource %llx - %llx cannot be added\n", (unsigned long long)res->start, (unsigned long long)res->end); kfree(res); - ret = -EEXIST; + res = NULL; } - return ret; + return res; +} + +static void release_memory_resource(struct resource *res) +{ + if (!res) + return; + release_resource(res); + kfree(res); + return; } @@ -246,8 +257,13 @@ int add_memory(int nid, u64 start, u64 size) { pg_data_t *pgdat = NULL; int new_pgdat = 0; + struct resource *res; int ret; + res = register_memory_resource(start, size); + if (!res) + return -EEXIST; + if (!node_online(nid)) { pgdat = hotadd_new_pgdat(nid, start); if (!pgdat) @@ -277,14 +293,13 @@ int add_memory(int nid, u64 start, u64 size) BUG_ON(ret); } - /* register this memory as resource */ - ret = register_memory_resource(start, size); - return ret; error: /* rollback pgdat allocation and others */ if (new_pgdat) rollback_node_hotadd(nid, pgdat); + if (res) + release_memory_resource(res); return ret; } -- cgit v1.2.3