diff --git a/arch/arm64/lib/copy_from_user.S b/arch/arm64/lib/copy_from_user.S
index 204b2e72966790..351e853fd28dde 100644
--- a/arch/arm64/lib/copy_from_user.S
+++ b/arch/arm64/lib/copy_from_user.S
@@ -63,11 +63,46 @@
 	USER_CPY(9996f, 0, cpyfert [\dst]!, [\src]!, \count!)
 	.endm
 
+	.macro ldpc1 reg1, reg2, ptr, val
+	user_ldp 9997f, \reg1, \reg2, \ptr, \val, #16
+	.endm
+
+	.macro stpc1 reg1, reg2, ptr, val
+	stp \reg1, \reg2, [\ptr], \val
+	.endm
+
 end	.req	x5
-srcin	.req	x15
+dstin	.req	x6
+req_reg_pcuabi srcin, c15, x15
 SYM_FUNC_START(COPY_FUNC_NAME)
 	add	end, x0, x2
+	mov	dstin, x0
+#ifdef CONFIG_CHERI_PURECAP_UABI
+.arch morello+c64
+	bx	#4
+	/*
+	 * Having switched to C64, argumentless RET is equivalent to RET CLR.
+	 * Because we have been called from A64, only LR is set. We therefore
+	 * set CLR to a valid capability, derived from PCC (as if we had been
+	 * called from C64). Conveniently this will also automatically switch
+	 * us back to A64 when returning (as the LSB of LR should be unset).
+	 */
+	cvtp	clr, lr
+	/*
+	 * Accessing memory via X registers in C64 requires using
+	 * alternate-base loads and stores; unfortunately most loads and stores
+	 * used in copy_template.S do not have an alternate-base counterpart.
+	 * The most straightforward solution is to access memory via C
+	 * registers only. We therefore need to create a valid capability for
+	 * the kernel buffer too, which is done by deriving it from DDC. Since
+	 * X-based accesses are validated against DDC, this is functionally
+	 * equivalent.
+	 */
+	cvtd	c0, x0
+	mov	srcin, c1
+#else
 	mov	srcin, x1
+#endif
 #include "copy_template.S"
 	mov	x0, #0				// Nothing to copy
 	ret
@@ -75,13 +110,13 @@ SYM_FUNC_START(COPY_FUNC_NAME)
 	// Exception fixups
 9996:	b.cs	9997f
 	// Registers are in Option A format
-	add	dst, dst, count
-9997:	cmp	dst, dstin
+	add	dstx, dstx, count
+9997:	cmp	dstx, dstin
 	b.ne	9998f
 	// Before being absolutely sure we couldn't copy anything, try harder
 USER(9998f, ldtrb tmp1w, [srcin])
 	strb	tmp1w, [dst], #1
-9998:	sub	x0, end, dst			// bytes not copied
+9998:	sub	x0, end, dstx			// bytes not copied
 	ret
 SYM_FUNC_END(COPY_FUNC_NAME)
 EXPORT_SYMBOL(COPY_FUNC_NAME)
diff --git a/arch/arm64/lib/copy_template.S b/arch/arm64/lib/copy_template.S
index 7f2f5a0e2fb9f0..459983994cfb66 100644
--- a/arch/arm64/lib/copy_template.S
+++ b/arch/arm64/lib/copy_template.S
@@ -18,17 +18,16 @@
  *	x0 - dest
  *	x1 - src
  *	x2 - n
- * Returns:
- *	x0 - dest
  */
-dstin	.req	x0
-src	.req	x1
+req_reg_pcuabi dst, c0, x0
+dstx	.req	x0
+req_reg_pcuabi src, c1, x1
+srcx	.req	x1
 count	.req	x2
 tmp1	.req	x3
 tmp1w	.req	w3
 tmp2	.req	x4
 tmp2w	.req	w4
-dst	.req	x6
 
 A_l	.req	x7
 A_h	.req	x8
@@ -39,7 +38,15 @@ C_h	.req	x12
 D_l	.req	x13
 D_h	.req	x14
 
-	mov	dst, dstin
+#ifdef COPY_CAPTAGS
+tmp1c	.req	c3
+tmp2c	.req	c4
+
+Ac_l	.req	c7
+Ac_h	.req	c8
+Bc_l	.req	c9
+Bc_h	.req	c10
+#endif
 
 #ifdef CONFIG_AS_HAS_MOPS
 alternative_if_not ARM64_HAS_MOPS
@@ -54,7 +61,7 @@ alternative_else_nop_endif
 	/*When memory length is less than 16, the accessed are not aligned.*/
 	b.lo	.Ltiny15
 
-	neg	tmp2, src
+	neg	tmp2, srcx
 	ands	tmp2, tmp2, #15/* Bytes to reach alignment. */
 	b.eq	.LSrcAligned
 	sub	count, count, tmp2
@@ -81,6 +88,11 @@ alternative_else_nop_endif
 	str1	tmp1, dst, #8
 
 .LSrcAligned:
+#ifdef COPY_CAPTAGS
+	/* src now 16-byte aligned, copy capability tags if dst also aligned */
+	tst	dstx, #15
+	b.eq	.LSrcAligned_cpycaps
+#endif
 	cmp	count, #64
 	b.ge	.Lcpy_over64
 	/*
@@ -188,4 +200,68 @@ alternative_else_nop_endif
 
 	tst	count, #0x3f
 	b.ne	.Ltail63
+#ifdef COPY_CAPTAGS
+	b	.Lexitfunc
+
+	/*
+	 * The .L*_cpycaps instruction sequences below are copies of the
+	 * sequences above.
+	 * The only functional difference is that they use capability
+	 * loads/stores, such that capability tags are copied from the source to
+	 * the destination. For that reason, they require both src and dst to be
+	 * 16-byte aligned.
+	 * Because C registers are twice as bigs as X registers, we only need
+	 * half the L/S instructions to transfer the same amount of data.
+	 */
+.LSrcAligned_cpycaps:
+	cmp	count, #64
+	b.ge	.Lcpy_over64_cpycaps
+
+.Ltail63_cpycaps:
+	ands	tmp1, count, #0x30
+	b.eq	.Ltiny15
+	cmp	tmp1w, #0x20
+	b.eq	1f
+	b.lt	2f
+	ldr1	tmp2c, src, #16
+	str1	tmp2c, dst, #16
+1:
+	ldr1	tmp2c, src, #16
+	str1	tmp2c, dst, #16
+2:
+	ldr1	tmp2c, src, #16
+	str1	tmp2c, dst, #16
+	b	.Ltiny15
+
+.Lcpy_over64_cpycaps:
+	subs	count, count, #128
+	b.ge	.Lcpy_body_large_cpycaps
+
+	ldpc1	Ac_l, Ac_h, src, #32
+	stpc1	Ac_l, Ac_h, dst, #32
+	ldpc1	Bc_l, Bc_h, src, #32
+	stpc1	Bc_l, Bc_h, dst, #32
+
+	tst	count, #0x3f
+	b.ne	.Ltail63_cpycaps
+	b	.Lexitfunc
+
+	.p2align	L1_CACHE_SHIFT
+.Lcpy_body_large_cpycaps:
+	ldpc1	Ac_l, Ac_h, src, #32
+	ldpc1	Bc_l, Bc_h, src, #32
+1:
+	/* attempt to keep 64-byte blocks of loads and stores interlaced */
+	stpc1	Ac_l, Ac_h, dst, #32
+	ldpc1	Ac_l, Ac_h, src, #32
+	stpc1	Bc_l, Bc_h, dst, #32
+	ldpc1	Bc_l, Bc_h, src, #32
+	subs	count, count, #64
+	b.ge	1b
+	stpc1	Ac_l, Ac_h, dst, #32
+	stpc1	Bc_l, Bc_h, dst, #32
+
+	tst	count, #0x3f
+	b.ne	.Ltail63_cpycaps
+#endif /* COPY_CAPTAGS */
 .Lexitfunc:
diff --git a/arch/arm64/lib/copy_to_user.S b/arch/arm64/lib/copy_to_user.S
index f4ac5fec80fe49..32e3a6b8fe94db 100644
--- a/arch/arm64/lib/copy_to_user.S
+++ b/arch/arm64/lib/copy_to_user.S
@@ -62,11 +62,30 @@
 	USER_CPY(9996f, 1, cpyfewt [\dst]!, [\src]!, \count!)
 	.endm
 
+	.macro ldpc1 reg1, reg2, ptr, val
+	ldp \reg1, \reg2, [\ptr], \val
+	.endm
+
+	.macro stpc1 reg1, reg2, ptr, val
+	user_stp 9997f, \reg1, \reg2, \ptr, \val, #16
+	.endm
+
 end	.req	x5
-srcin	.req	x15
+dstin	.req	x6
+req_reg_pcuabi srcin, c15, x15
 SYM_FUNC_START(COPY_FUNC_NAME)
 	add	end, x0, x2
+	mov	dstin, x0
+#ifdef CONFIG_CHERI_PURECAP_UABI
+.arch morello+c64
+	bx	#4
+	/* See comments in copy_from_user.S */
+	cvtp	clr, lr
+	cvtd	c1, x1
+	mov	srcin, c1
+#else
 	mov	srcin, x1
+#endif
 #include "copy_template.S"
 	mov	x0, #0
 	ret
@@ -74,14 +93,14 @@ SYM_FUNC_START(COPY_FUNC_NAME)
 	// Exception fixups
 9996:	b.cs	9997f
 	// Registers are in Option A format
-	add	dst, dst, count
-9997:	cmp	dst, dstin
+	add	dstx, dstx, count
+9997:	cmp	dstx, dstin
 	b.ne	9998f
 	// Before being absolutely sure we couldn't copy anything, try harder
 	ldrb	tmp1w, [srcin]
 USER(9998f, sttrb tmp1w, [dst])
 	add	dst, dst, #1
-9998:	sub	x0, end, dst			// bytes not copied
+9998:	sub	x0, end, dstx			// bytes not copied
 	ret
 SYM_FUNC_END(COPY_FUNC_NAME)
 EXPORT_SYMBOL(COPY_FUNC_NAME)
diff --git a/tools/testing/selftests/arm64/morello/freestanding_init_globals.c b/tools/testing/selftests/arm64/morello/freestanding_init_globals.c
index 12eb0af97887b6..2fa2aa643fd60d 100644
--- a/tools/testing/selftests/arm64/morello/freestanding_init_globals.c
+++ b/tools/testing/selftests/arm64/morello/freestanding_init_globals.c
@@ -38,6 +38,9 @@
 #ifndef R_MORELLO_RELATIVE
 #define R_MORELLO_RELATIVE		59395
 #endif
+#ifndef R_MORELLO_FUNC_RELATIVE
+#define R_MORELLO_FUNC_RELATIVE		59400
+#endif
 
 struct cap_reloc {
 	size_t capability_location;
@@ -181,7 +184,8 @@ void __morello_process_dynamic_relocs(void *auxv)
 	for (reloc = rela_dyn_start; reloc < rela_dyn_end; ++reloc) {
 		uintptr_t *reloc_addr, value;
 
-		if (reloc->r_info != R_MORELLO_RELATIVE)
+		if (reloc->r_info != R_MORELLO_RELATIVE &&
+		    reloc->r_info != R_MORELLO_FUNC_RELATIVE)
 			continue;
 		reloc_addr = (uintptr_t *)cheri_address_set(cap_rw, reloc->r_offset);
 		value = morello_relative(0, cap_rx, cap_rw, reloc, reloc_addr);