arm64: lib: Use MOPS for usercopy routines

Similarly to what was done with the memcpy() routines, make
copy_to_user(), copy_from_user() and clear_user() also use the Armv8.8
FEAT_MOPS instructions.

Both MOPS implementation options (A and B) are supported, including
asymmetric systems. The exception fixup code fixes up the registers
according to the option used.

In case of a fault the routines return precisely how much was not copied
(as required by the comment in include/linux/uaccess.h), as unprivileged
versions of CPY/SET are guaranteed not to have written past the
addresses reported in the GPRs.

The MOPS instructions could possibly be inlined into callers (and
patched to branch to the generic implementation if not detected;
similarly to what x86 does), but as a first step this patch just uses
them in the out-of-line routines.

Signed-off-by: Kristina Martšenko <kristina.martsenko@arm.com>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20250228170006.390100-4-kristina.martsenko@arm.com
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>

Author: kristina-martsenko

arch/arm64/include/asm/asm-uaccess.h

@@ -61,6 +61,10 @@ alternative_else_nop_endif
 9999:	x;					\
 	_asm_extable_uaccess	9999b, l
 
+#define USER_CPY(l, uaccess_is_write, x...)	\
+9999:	x;					\
+	_asm_extable_uaccess_cpy 9999b, l, uaccess_is_write
+
 /*
  * Generate the assembly for LDTR/STTR with exception table entries.
  * This is complicated as there is no post-increment or pair versions of the

arch/arm64/lib/clear_user.S

@@ -17,14 +17,27 @@
  * 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)
 	add	x2, x0, x1
+
+#ifdef CONFIG_AS_HAS_MOPS
+	.arch_extension mops
+alternative_if_not ARM64_HAS_MOPS
+	b	.Lno_mops
+alternative_else_nop_endif
+
+USER(9f, setpt	[x0]!, x1!, xzr)
+USER(6f, setmt	[x0]!, x1!, xzr)
+USER(6f, setet	[x0]!, x1!, xzr)
+	mov	x0, #0
+	ret
+.Lno_mops:
+#endif
+
 	subs	x1, x1, #8
 	b.mi	2f
-1:
+
+1:	.p2align 4
 USER(9f, sttr	xzr, [x0])
 	add	x0, x0, #8
 	subs	x1, x1, #8
@@ -47,6 +60,10 @@ USER(7f, sttrb	wzr, [x2, #-1])
 	ret
 
 	// Exception fixups
+6:	b.cs	9f
+	// Registers are in Option A format
+	add	x0, x0, x1
+	b	9f
 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

arch/arm64/lib/copy_from_user.S

@@ -52,6 +52,13 @@
 	stp \reg1, \reg2, [\ptr], \val
 	.endm
 
+	.macro cpy1 dst, src, count
+	.arch_extension mops
+	USER_CPY(9997f, 0, cpyfprt [\dst]!, [\src]!, \count!)
+	USER_CPY(9996f, 0, cpyfmrt [\dst]!, [\src]!, \count!)
+	USER_CPY(9996f, 0, cpyfert [\dst]!, [\src]!, \count!)
+	.endm
+
 end	.req	x5
 srcin	.req	x15
 SYM_FUNC_START(__arch_copy_from_user)
@@ -62,6 +69,9 @@ SYM_FUNC_START(__arch_copy_from_user)
 	ret
 
 	// Exception fixups
+9996:	b.cs	9997f
+	// Registers are in Option A format
+	add	dst, dst, count
 9997:	cmp	dst, dstin
 	b.ne	9998f
 	// Before being absolutely sure we couldn't copy anything, try harder

arch/arm64/lib/copy_template.S

@@ -40,6 +40,16 @@ D_l	.req	x13
 D_h	.req	x14
 
 	mov	dst, dstin
+
+#ifdef CONFIG_AS_HAS_MOPS
+alternative_if_not ARM64_HAS_MOPS
+	b	.Lno_mops
+alternative_else_nop_endif
+	cpy1	dst, src, count
+	b	.Lexitfunc
+.Lno_mops:
+#endif
+
 	cmp	count, #16
 	/*When memory length is less than 16, the accessed are not aligned.*/
 	b.lo	.Ltiny15

arch/arm64/lib/copy_to_user.S

@@ -51,6 +51,13 @@
 	user_stp 9997f, \reg1, \reg2, \ptr, \val
 	.endm
 
+	.macro cpy1 dst, src, count
+	.arch_extension mops
+	USER_CPY(9997f, 1, cpyfpwt [\dst]!, [\src]!, \count!)
+	USER_CPY(9996f, 1, cpyfmwt [\dst]!, [\src]!, \count!)
+	USER_CPY(9996f, 1, cpyfewt [\dst]!, [\src]!, \count!)
+	.endm
+
 end	.req	x5
 srcin	.req	x15
 SYM_FUNC_START(__arch_copy_to_user)
@@ -61,6 +68,9 @@ SYM_FUNC_START(__arch_copy_to_user)
 	ret
 
 	// Exception fixups
+9996:	b.cs	9997f
+	// Registers are in Option A format
+	add	dst, dst, count
 9997:	cmp	dst, dstin
 	b.ne	9998f
 	// Before being absolutely sure we couldn't copy anything, try harder