x86: inline the 'rep movs' in user copies for the FSRM case

This does the same thing for the user copies as commit 0db7058e8e23
("x86/clear_user: Make it faster") did for clear_user().  In other
words, it inlines the "rep movs" case when X86_FEATURE_FSRM is set,
avoiding the function call entirely.

In order to do that, it makes the calling convention for the out-of-line
case ("copy_user_generic_unrolled") match the 'rep movs' calling
convention, although it does also end up clobbering a number of
additional registers.

Also, to simplify code sharing in the low-level assembly with the
__copy_user_nocache() function (that uses the normal C calling
convention), we end up with a kind of mixed return value for the
low-level asm code: it will return the result in both %rcx (to work as
an alternative for the 'rep movs' case), _and_ in %rax (for the nocache
case).

We could avoid this by wrapping __copy_user_nocache() callers in an
inline asm, but since the cost is just an extra register copy, it's
probably not worth it.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2 files changed, 31 insertions, 47 deletions

diff --git a/arch/x86/include/asm/uaccess_64.h b/arch/x86/include/asm/uaccess_64.h
index 339883729065..8cc918acbabc 100644
--- a/arch/x86/include/asm/uaccess_64.h
+++ b/arch/x86/include/asm/uaccess_64.h
@@ -18,29 +18,26 @@
 
 /* Handles exceptions in both to and from, but doesn't do access_ok */
 __must_check unsigned long
-copy_user_fast_string(void *to, const void *from, unsigned len);
-__must_check unsigned long
 copy_user_generic_unrolled(void *to, const void *from, unsigned len);
 
 static __always_inline __must_check unsigned long
-copy_user_generic(void *to, const void *from, unsigned len)
+copy_user_generic(void *to, const void *from, unsigned long len)
 {
-	unsigned ret;
-
 	stac();
 	/*
 	 * If CPU has FSRM feature, use 'rep movs'.
 	 * Otherwise, use copy_user_generic_unrolled.
 	 */
-	alternative_call(copy_user_generic_unrolled,
-			 copy_user_fast_string,
-			 X86_FEATURE_FSRM,
-			 ASM_OUTPUT2("=a" (ret), "=D" (to), "=S" (from),
-				     "=d" (len)),
-			 "1" (to), "2" (from), "3" (len)
-			 : "memory", "rcx", "r8", "r9", "r10", "r11");
+	asm volatile(
+		"1:\n\t"
+		ALTERNATIVE("rep movsb",
+			    "call copy_user_generic_unrolled", ALT_NOT(X86_FEATURE_FSRM))
+		"2:\n"
+		_ASM_EXTABLE_UA(1b, 2b)
+		:"+c" (len), "+D" (to), "+S" (from), ASM_CALL_CONSTRAINT
+		: : "memory", "rax", "rdx", "r8", "r9", "r10", "r11");
 	clac();
-	return ret;
+	return len;
 }
 
 static __always_inline __must_check unsigned long
diff --git a/arch/x86/lib/copy_user_64.S b/arch/x86/lib/copy_user_64.S
index 818f2f728294..16a743f11b11 100644
--- a/arch/x86/lib/copy_user_64.S
+++ b/arch/x86/lib/copy_user_64.S
@@ -45,13 +45,29 @@
  * Input:
  * rdi destination
  * rsi source
- * rdx count
+ * rcx count
  *
  * Output:
- * eax uncopied bytes or 0 if successful.
+ * rcx uncopied bytes or 0 if successful.
+ *
+ * NOTE! The calling convention is very intentionally the same as
+ * for 'rep movs', so that we can rewrite the function call with
+ * just a plain 'rep movs' on machines that have FSRM.
+ *
+ * HOWEVER! This function ends up having a lot of the code common
+ * with __copy_user_nocache(), which is a normal C function, and
+ * has a similar calling convention, but gets the 'count' in %rdx,
+ * and returns the result in %rax.
+ *
+ * To share as much code as possible, we end up returning the
+ * result in *both* %rcx/%rax, and we also move the initial count
+ * into %rdx.
+ *
+ * We can clobber rdx/rsi/rdi and r8-r11
  */
 SYM_FUNC_START(copy_user_generic_unrolled)
-	cmpl $8,%edx
+	movl %ecx,%edx
+	cmpl $8,%ecx
 	jb .Lcopy_user_short_string_bytes
 	ALIGN_DESTINATION
 	movl %edx,%ecx
@@ -104,37 +120,6 @@ SYM_FUNC_END(copy_user_generic_unrolled)
 EXPORT_SYMBOL(copy_user_generic_unrolled)
 
 /*
- * Some CPUs support FSRM for Fast Short REP MOVS.
- *
- * Only 4GB of copy is supported. This shouldn't be a problem
- * because the kernel normally only writes from/to page sized chunks
- * even if user space passed a longer buffer.
- * And more would be dangerous because both Intel and AMD have
- * errata with rep movsq > 4GB. If someone feels the need to fix
- * this please consider this.
- *
- * Input:
- * rdi destination
- * rsi source
- * rdx count
- *
- * Output:
- * eax uncopied bytes or 0 if successful.
- */
-SYM_FUNC_START(copy_user_fast_string)
-	movl %edx,%ecx
-1:	rep movsb
-	xorl %eax,%eax
-	RET
-
-12:	movl %ecx,%eax		/* ecx is zerorest also */
-	RET
-
-	_ASM_EXTABLE_CPY(1b, 12b)
-SYM_FUNC_END(copy_user_fast_string)
-EXPORT_SYMBOL(copy_user_fast_string)
-
-/*
  * Try to copy last bytes and clear the rest if needed.
  * Since protection fault in copy_from/to_user is not a normal situation,
  * it is not necessary to optimize tail handling.
@@ -160,6 +145,7 @@ SYM_CODE_START_LOCAL(.Lcopy_user_handle_tail)
 
 3:
 	movl %edx,%eax
+	movl %edx,%ecx
 	RET
 
 	_ASM_EXTABLE_CPY(1b, 2b)
@@ -203,6 +189,7 @@ SYM_CODE_START_LOCAL(copy_user_short_string)
 	decl %ecx
 	jnz 21b
 23:	xor %eax,%eax
+	xor %ecx,%ecx
 	RET
 
 40:	leal (%rdx,%rcx,8),%edx