mm: introduce get_user_pages_fast

Introduce a new get_user_pages_fast mm API, which is basically a
get_user_pages with a less general API (but still tends to be suited to
the common case):

- task and mm are always current and current->mm
- force is always 0
- pages is always non-NULL
- don't pass back vmas

This restricted API can be implemented in a much more scalable way on many
architectures when the ptes are present, by walking the page tables
locklessly (no mmap_sem or page table locks).  When the ptes are not
populated, get_user_pages_fast() could be slower.

This is implemented locklessly on x86, and used in some key direct IO call
sites, in later patches, which provides nearly 10% performance improvement
on a threaded database workload.

Lots of other code could use this too, depending on use cases (eg.  grep
drivers/).  And it might inspire some new and clever ways to use it.

[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Dave Kleikamp <shaggy@austin.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Dave Kleikamp <shaggy@austin.ibm.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Reviewed-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/include/linux/mm.h b/include/linux/mm.h
index d87a5a5..f3fd70d 100644 (file)
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -833,6 +833,39 @@ extern int mprotect_fixup(struct vm_area_struct *vma,
                          struct vm_area_struct **pprev, unsigned long start,
                          unsigned long end, unsigned long newflags);
 
+#ifdef CONFIG_HAVE_GET_USER_PAGES_FAST
+/*
+ * get_user_pages_fast provides equivalent functionality to get_user_pages,
+ * operating on current and current->mm (force=0 and doesn't return any vmas).
+ *
+ * get_user_pages_fast may take mmap_sem and page tables, so no assumptions
+ * can be made about locking. get_user_pages_fast is to be implemented in a
+ * way that is advantageous (vs get_user_pages()) when the user memory area is
+ * already faulted in and present in ptes. However if the pages have to be
+ * faulted in, it may turn out to be slightly slower).
+ */
+int get_user_pages_fast(unsigned long start, int nr_pages, int write,
+                       struct page **pages);
+
+#else
+/*
+ * Should probably be moved to asm-generic, and architectures can include it if
+ * they don't implement their own get_user_pages_fast.
+ */
+#define get_user_pages_fast(start, nr_pages, write, pages)     \
+({                                                             \
+       struct mm_struct *mm = current->mm;                     \
+       int ret;                                                \
+                                                               \
+       down_read(&mm->mmap_sem);                               \
+       ret = get_user_pages(current, mm, start, nr_pages,      \
+                                       write, 0, pages, NULL); \
+       up_read(&mm->mmap_sem);                                 \
+                                                               \
+       ret;                                                    \
+})
+#endif
+
 /*
  * A callback you can register to apply pressure to ageable caches.
  *