[RFC][PATCH 5/12] x86: introduce __set_bit() like function for bitmaps in user space

Takuya Yoshikawa takuya.yoshikawa at gmail.com
Tue May 4 23:02:47 EST 2010


During the work of KVM's dirty page logging optimization, we encountered
the need of manipulating bitmaps in user space efficiantly. To achive this,
we introduce a uaccess function for setting a bit in user space following
Avi's suggestion.

  KVM is now using dirty bitmaps for live-migration and VGA. Although we need
  to update them from kernel side, copying them every time for updating the
  dirty log is a big bottleneck. Especially, we tested that zero-copy bitmap
  manipulation improves responses of GUI manipulations a lot.

We also found one similar need in drivers/vhost/vhost.c in which the author
implemented set_bit_to_user() locally using inefficient functions: see TODO
at the top of that.

Probably, this kind of need would be common for virtualization area.

So we introduce a macro set_bit_user_non_atomic() following the implementation
style of x86's uaccess functions.

Note: there is a one restriction to this macro: bitmaps must be 64-bit
aligned (see the comment in this patch).

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya at oss.ntt.co.jp>
Signed-off-by: Fernando Luis Vazquez Cao <fernando at oss.ntt.co.jp>
CC: Avi Kivity <avi at redhat.com>
Cc: Thomas Gleixner <tglx at linutronix.de>
CC: Ingo Molnar <mingo at redhat.com>
Cc: "H. Peter Anvin" <hpa at zytor.com>
---
 arch/x86/include/asm/uaccess.h |   39 +++++++++++++++++++++++++++++++++++++++
 1 files changed, 39 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
index abd3e0e..3138e65 100644
--- a/arch/x86/include/asm/uaccess.h
+++ b/arch/x86/include/asm/uaccess.h
@@ -98,6 +98,45 @@ struct exception_table_entry {
 
 extern int fixup_exception(struct pt_regs *regs);
 
+/**
+ * set_bit_user_non_atomic: - set a bit of a bitmap in user space.
+ * @nr:   Bit offset.
+ * @addr: Base address of a bitmap in user space.
+ *
+ * Context: User context only.  This function may sleep.
+ *
+ * This macro set a bit of a bitmap in user space.
+ *
+ * Restriction: the bitmap pointed to by @addr must be 64-bit aligned:
+ * the kernel accesses the bitmap by its own word length, so bitmaps
+ * allocated by 32-bit processes may cause fault.
+ *
+ * Returns zero on success, or -EFAULT on error.
+ */
+#define __set_bit_user_non_atomic_asm(nr, addr, err, errret)		\
+	asm volatile("1:	bts %1,%2\n"				\
+		     "2:\n"						\
+		     ".section .fixup,\"ax\"\n"				\
+		     "3:	mov %3,%0\n"				\
+		     "	jmp 2b\n"					\
+		     ".previous\n"					\
+		     _ASM_EXTABLE(1b, 3b)				\
+		     : "=r"(err)					\
+		     : "r" (nr), "m" (__m(addr)), "i" (errret), "0" (err))
+
+#define set_bit_user_non_atomic(nr, addr)				\
+({									\
+	int __ret_sbu;							\
+									\
+	might_fault();							\
+	if (access_ok(VERIFY_WRITE, addr, nr/8 + 1))			\
+		__set_bit_user_non_atomic_asm(nr, addr,	__ret_sbu, -EFAULT);\
+	else								\
+		__ret_sbu = -EFAULT;					\
+									\
+	__ret_sbu;							\
+})
+
 /*
  * These are the main single-value transfer routines.  They automatically
  * use the right size if we just have the right pointer type.
-- 
1.7.0.4



More information about the Linuxppc-dev mailing list