crypto: arm64/crc-t10dif - move NEON yield to C code

Instead of yielding from the bowels of the asm routine if a reschedule
is needed, divide up the input into 4 KB chunks in the C glue. This
simplifies the code substantially, and avoids scheduling out the task
with the asm routine on the call stack, which is undesirable from a
CFI/instrumentation point of view.

Signed-off-by: Ard Biesheuvel <[e-mail]>
Signed-off-by: Herbert Xu <[e-mail]>

arch/arm64/crypto/crct10dif-ce-core.S

···
 	.text
 	.arch		armv8-a+crypto
 
-	init_crc	.req	w19
-	buf		.req	x20
-	len		.req	x21
-	fold_consts_ptr	.req	x22
+	init_crc	.req	w0
+	buf		.req	x1
+	len		.req	x2
+	fold_consts_ptr	.req	x3
 
 	fold_consts	.req	v10
 
···
 	.endm
 
 	.macro		crc_t10dif_pmull, p
-	frame_push	4, 128
-
-	mov		init_crc, w0
-	mov		buf, x1
-	mov		len, x2
-
 	__pmull_init_\p
 
 	// For sizes less than 256 bytes, we can't fold 128 bytes at a time.
···
 	fold_32_bytes	\p, v6, v7
 
 	subs		len, len, #128
-	b.lt		.Lfold_128_bytes_loop_done_\@
-
-	if_will_cond_yield_neon
-	stp		q0, q1, [sp, #.Lframe_local_offset]
-	stp		q2, q3, [sp, #.Lframe_local_offset + 32]
-	stp		q4, q5, [sp, #.Lframe_local_offset + 64]
-	stp		q6, q7, [sp, #.Lframe_local_offset + 96]
-	do_cond_yield_neon
-	ldp		q0, q1, [sp, #.Lframe_local_offset]
-	ldp		q2, q3, [sp, #.Lframe_local_offset + 32]
-	ldp		q4, q5, [sp, #.Lframe_local_offset + 64]
-	ldp		q6, q7, [sp, #.Lframe_local_offset + 96]
-	ld1		{fold_consts.2d}, [fold_consts_ptr]
-	__pmull_init_\p
-	__pmull_pre_\p	fold_consts
-	endif_yield_neon
-
-	b		.Lfold_128_bytes_loop_\@
-
-.Lfold_128_bytes_loop_done_\@:
+	b.ge		.Lfold_128_bytes_loop_\@
 
 	// Now fold the 112 bytes in v0-v6 into the 16 bytes in v7.
 
···
 	// Final CRC value (x^16 * M(x)) mod G(x) is in low 16 bits of v0.
 
 	umov		w0, v0.h[0]
-	frame_pop
+	.ifc		\p, p8
+	ldp		x29, x30, [sp], #16
+	.endif
 	ret
 
 .Lless_than_256_bytes_\@:
···
 // Assumes len >= 16.
 //
 SYM_FUNC_START(crc_t10dif_pmull_p8)
-	crc_t10dif_pmull	p8
+	stp		x29, x30, [sp, #-16]!
+	mov		x29, sp
+	crc_t10dif_pmull p8
 SYM_FUNC_END(crc_t10dif_pmull_p8)
 
 	.align		5

arch/arm64/crypto/crct10dif-ce-glue.c

···
 	u16 *crc = shash_desc_ctx(desc);
 
 	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && crypto_simd_usable()) {
-		kernel_neon_begin();
-		*crc = crc_t10dif_pmull_p8(*crc, data, length);
-		kernel_neon_end();
+		do {
+			unsigned int chunk = length;
+
+			if (chunk > SZ_4K + CRC_T10DIF_PMULL_CHUNK_SIZE)
+				chunk = SZ_4K;
+
+			kernel_neon_begin();
+			*crc = crc_t10dif_pmull_p8(*crc, data, chunk);
+			kernel_neon_end();
+			data += chunk;
+			length -= chunk;
+		} while (length);
 	} else {
 		*crc = crc_t10dif_generic(*crc, data, length);
 	}
···
 	u16 *crc = shash_desc_ctx(desc);
 
 	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && crypto_simd_usable()) {
-		kernel_neon_begin();
-		*crc = crc_t10dif_pmull_p64(*crc, data, length);
-		kernel_neon_end();
+		do {
+			unsigned int chunk = length;
+
+			if (chunk > SZ_4K + CRC_T10DIF_PMULL_CHUNK_SIZE)
+				chunk = SZ_4K;
+
+			kernel_neon_begin();
+			*crc = crc_t10dif_pmull_p64(*crc, data, chunk);
+			kernel_neon_end();
+			data += chunk;
+			length -= chunk;
+		} while (length);
 	} else {
 		*crc = crc_t10dif_generic(*crc, data, length);
 	}