]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
ring_buffer: allocate buffer page pointer
authorSteven Rostedt <rostedt@goodmis.org>
Wed, 1 Oct 2008 15:14:54 +0000 (11:14 -0400)
committerIngo Molnar <mingo@elte.hu>
Tue, 14 Oct 2008 08:39:08 +0000 (10:39 +0200)
The current method of overlaying the page frame as the buffer page pointer
can be very dangerous and limits our ability to do other things with
a page from the buffer, like send it off to disk.

This patch allocates the buffer_page instead of overlaying the page's
page frame. The use of the buffer_page has hardly changed due to this.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/trace/ring_buffer.c

index 9631abf2ae291122760016d05a44120d0105c87e..98145718988d1145a54fdcd8e5df7e97d8bb5322 100644 (file)
@@ -115,16 +115,10 @@ void *ring_buffer_event_data(struct ring_buffer_event *event)
  * Thanks to Peter Zijlstra for suggesting this idea.
  */
 struct buffer_page {
-       union {
-               struct {
-                       unsigned long    flags;         /* mandatory */
-                       atomic_t         _count;        /* mandatory */
-                       u64              time_stamp;    /* page time stamp */
-                       unsigned         size;          /* size of page data */
-                       struct list_head list;          /* list of free pages */
-               };
-               struct page page;
-       };
+       u64              time_stamp;    /* page time stamp */
+       unsigned         size;          /* size of page data */
+       struct list_head list;          /* list of free pages */
+       void *page;                     /* Actual data page */
 };
 
 /*
@@ -133,9 +127,9 @@ struct buffer_page {
  */
 static inline void free_buffer_page(struct buffer_page *bpage)
 {
-       reset_page_mapcount(&bpage->page);
-       bpage->page.mapping = NULL;
-       __free_page(&bpage->page);
+       if (bpage->page)
+               __free_page(bpage->page);
+       kfree(bpage);
 }
 
 /*
@@ -237,11 +231,16 @@ static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
        unsigned i;
 
        for (i = 0; i < nr_pages; i++) {
+               page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
+                                   GFP_KERNEL, cpu_to_node(cpu));
+               if (!page)
+                       goto free_pages;
+               list_add(&page->list, &pages);
+
                addr = __get_free_page(GFP_KERNEL);
                if (!addr)
                        goto free_pages;
-               page = (struct buffer_page *)virt_to_page(addr);
-               list_add(&page->list, &pages);
+               page->page = (void *)addr;
        }
 
        list_splice(&pages, head);
@@ -262,6 +261,7 @@ static struct ring_buffer_per_cpu *
 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
 {
        struct ring_buffer_per_cpu *cpu_buffer;
+       struct buffer_page *page;
        unsigned long addr;
        int ret;
 
@@ -275,10 +275,17 @@ rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
        spin_lock_init(&cpu_buffer->lock);
        INIT_LIST_HEAD(&cpu_buffer->pages);
 
+       page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
+                           GFP_KERNEL, cpu_to_node(cpu));
+       if (!page)
+               goto fail_free_buffer;
+
+       cpu_buffer->reader_page = page;
        addr = __get_free_page(GFP_KERNEL);
        if (!addr)
-               goto fail_free_buffer;
-       cpu_buffer->reader_page = (struct buffer_page *)virt_to_page(addr);
+               goto fail_free_reader;
+       page->page = (void *)addr;
+
        INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
        cpu_buffer->reader_page->size = 0;
 
@@ -523,11 +530,16 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
 
        for_each_buffer_cpu(buffer, cpu) {
                for (i = 0; i < new_pages; i++) {
+                       page = kzalloc_node(ALIGN(sizeof(*page),
+                                                 cache_line_size()),
+                                           GFP_KERNEL, cpu_to_node(cpu));
+                       if (!page)
+                               goto free_pages;
+                       list_add(&page->list, &pages);
                        addr = __get_free_page(GFP_KERNEL);
                        if (!addr)
                                goto free_pages;
-                       page = (struct buffer_page *)virt_to_page(addr);
-                       list_add(&page->list, &pages);
+                       page->page = (void *)addr;
                }
        }
 
@@ -567,9 +579,7 @@ static inline int rb_null_event(struct ring_buffer_event *event)
 
 static inline void *rb_page_index(struct buffer_page *page, unsigned index)
 {
-       void *addr = page_address(&page->page);
-
-       return addr + index;
+       return page->page + index;
 }
 
 static inline struct ring_buffer_event *