arm64: Rewrite __arch_clear_user()

Now that we're always using STTR variants rather than abstracting two
different addressing modes, the user_ldst macro here is frankly more
obfuscating than helpful. Rewrite __arch_clear_user() with regular
USER() annotations so that it's clearer what's going on, and take the
opportunity to minimise the branchiness in the most common paths, while
also allowing the exception fixup to return an accurate result.

Apparently some folks examine large reads from /dev/zero closely enough
to notice the loop being hot, so align it per the other critical loops
(presumably around a typical instruction fetch granularity).
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/1cbd78b12c076a8ad4656a345811cfb9425df0b3.1622128527.git.robin.murphy@arm.comSigned-off-by: Will Deacon <will@kernel.org>

arch/arm64/lib/clear_user.S

 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Based on arch/arm/lib/clear_user.S
- *
- * Copyright (C) 2012 ARM Ltd.
+ * Copyright (C) 2021 Arm Ltd.
  */
-#include <linux/linkage.h>
 
-#include <asm/asm-uaccess.h>
+#include <linux/linkage.h>
 #include <asm/assembler.h>
 
 	.text
@@ -19,25 +16,33 @@
  *
  * Alignment fixed up by hardware.
  */
+
+	.p2align 4
+	// Alignment is for the loop, but since the prologue (including BTI)
+	// is also 16 bytes we can keep any padding outside the function
 SYM_FUNC_START(__arch_clear_user)
-	mov	x2, x1			// save the size for fixup return
+	add	x2, x0, x1
 	subs	x1, x1, #8
 	b.mi	2f
 1:
-user_ldst 9f, sttr, xzr, x0, 8
+USER(9f, sttr	xzr, [x0])
+	add	x0, x0, #8
 	subs	x1, x1, #8
-	b.pl	1b
-2:	adds	x1, x1, #4
-	b.mi	3f
-user_ldst 9f, sttr, wzr, x0, 4
-	sub	x1, x1, #4
-3:	adds	x1, x1, #2
-	b.mi	4f
-user_ldst 9f, sttrh, wzr, x0, 2
-	sub	x1, x1, #2
-4:	adds	x1, x1, #1
-	b.mi	5f
-user_ldst 9f, sttrb, wzr, x0, 0
+	b.hi	1b
+USER(9f, sttr	xzr, [x2, #-8])
+	mov	x0, #0
+	ret
+
+2:	tbz	x1, #2, 3f
+USER(9f, sttr	wzr, [x0])
+USER(8f, sttr	wzr, [x2, #-4])
+	mov	x0, #0
+	ret
+
+3:	tbz	x1, #1, 4f
+USER(9f, sttrh	wzr, [x0])
+4:	tbz	x1, #0, 5f
+USER(7f, sttrb	wzr, [x2, #-1])
 5:	mov	x0, #0
 	ret
 SYM_FUNC_END(__arch_clear_user)
@@ -45,6 +50,8 @@ EXPORT_SYMBOL(__arch_clear_user)
 
 	.section .fixup,"ax"
 	.align	2
-9:	mov	x0, x2			// return the original size
+7:	sub	x0, x2, #5	// Adjust for faulting on the final byte...
+8:	add	x0, x0, #4	// ...or the second word of the 4-7 byte case
+9:	sub	x0, x2, x0
 	ret
 	.previous