Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6

Pull crypto update from Herbert Xu:
 "API:
   - Restrict crypto_cipher to internal API users only.

  Algorithms:
   - Add x86 aesni acceleration for cts.
   - Improve x86 aesni acceleration for xts.
   - Remove x86 acceleration of some uncommon algorithms.
   - Remove RIPE-MD, Tiger and Salsa20.
   - Remove tnepres.
   - Add ARM acceleration for BLAKE2s and BLAKE2b.

  Drivers:
   - Add Keem Bay OCS HCU driver.
   - Add Marvell OcteonTX2 CPT PF driver.
   - Remove PicoXcell driver.
   - Remove mediatek driver"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (154 commits)
  hwrng: timeriomem - Use device-managed registration API
  crypto: hisilicon/qm - fix printing format issue
  crypto: hisilicon/qm - do not reset hardware when CE happens
  crypto: hisilicon/qm - update irqflag
  crypto: hisilicon/qm - fix the value of 'QM_SQC_VFT_BASE_MASK_V2'
  crypto: hisilicon/qm - fix request missing error
  crypto: hisilicon/qm - removing driver after reset
  crypto: octeontx2 - fix -Wpointer-bool-conversion warning
  crypto: hisilicon/hpre - enable Elliptic curve cryptography
  crypto: hisilicon - PASID fixed on Kunpeng 930
  crypto: hisilicon/qm - fix use of 'dma_map_single'
  crypto: hisilicon/hpre - tiny fix
  crypto: hisilicon/hpre - adapt the number of clusters
  crypto: cpt - remove casting dma_alloc_coherent
  crypto: keembay-ocs-aes - Fix 'q' assignment during CCM B0 generation
  crypto: xor - Fix typo of optimization
  hwrng: optee - Use device-managed registration API
  crypto: arm64/crc-t10dif - move NEON yield to C code
  crypto: arm64/aes-ce-mac - simplify NEON yield
  crypto: arm64/aes-neonbs - remove NEON yield calls
  ...

.mailmap

@@ -174,7 +174,6 @@ Juha Yrjola <at solidboot.com>
 Juha Yrjola <juha.yrjola@nokia.com>
 Juha Yrjola <juha.yrjola@solidboot.com>
 Julien Thierry <julien.thierry.kdev@gmail.com> <julien.thierry@arm.com>
-Kamil Konieczny <k.konieczny@samsung.com> <k.konieczny@partner.samsung.com>
 Kay Sievers <kay.sievers@vrfy.org>
 Kees Cook <keescook@chromium.org> <kees.cook@canonical.com>
 Kees Cook <keescook@chromium.org> <keescook@google.com>

Documentation/admin-guide/device-mapper/dm-integrity.rst

@@ -143,8 +143,8 @@ recalculate
 journal_crypt:algorithm(:key)	(the key is optional)
 	Encrypt the journal using given algorithm to make sure that the
 	attacker can't read the journal. You can use a block cipher here
-	(such as "cbc(aes)") or a stream cipher (for example "chacha20",
-	"salsa20" or "ctr(aes)").
+	(such as "cbc(aes)") or a stream cipher (for example "chacha20"
+	or "ctr(aes)").
 
 	The journal contains history of last writes to the block device,
 	an attacker reading the journal could see the last sector numbers

Documentation/crypto/api-skcipher.rst

@@ -28,8 +28,8 @@ Symmetric Key Cipher Request Handle
 Single Block Cipher API
 -----------------------
 
-.. kernel-doc:: include/linux/crypto.h
+.. kernel-doc:: include/crypto/internal/cipher.h
    :doc: Single Block Cipher API
 
-.. kernel-doc:: include/linux/crypto.h
+.. kernel-doc:: include/crypto/internal/cipher.h
    :functions: crypto_alloc_cipher crypto_free_cipher crypto_has_cipher crypto_cipher_blocksize crypto_cipher_setkey crypto_cipher_encrypt_one crypto_cipher_decrypt_one

Documentation/devicetree/bindings/crypto/intel,keembay-ocs-hcu.yaml

@@ -0,0 +1,46 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/crypto/intel,keembay-ocs-hcu.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Intel Keem Bay OCS HCU Device Tree Bindings
+
+maintainers:
+  - Declan Murphy <declan.murphy@intel.com>
+  - Daniele Alessandrelli <daniele.alessandrelli@intel.com>
+
+description:
+  The Intel Keem Bay Offload and Crypto Subsystem (OCS) Hash Control Unit (HCU)
+  provides hardware-accelerated hashing and HMAC.
+
+properties:
+  compatible:
+    const: intel,keembay-ocs-hcu
+
+  reg:
+    maxItems: 1
+
+  interrupts:
+    maxItems: 1
+
+  clocks:
+    maxItems: 1
+
+required:
+  - compatible
+  - reg
+  - interrupts
+  - clocks
+
+additionalProperties: false
+
+examples:
+  - |
+    #include <dt-bindings/interrupt-controller/arm-gic.h>
+    crypto@3000b000 {
+      compatible = "intel,keembay-ocs-hcu";
+      reg = <0x3000b000 0x1000>;
+      interrupts = <GIC_SPI 121 IRQ_TYPE_LEVEL_HIGH>;
+      clocks = <&scmi_clk 94>;
+    };

Documentation/devicetree/bindings/crypto/samsung-slimsss.yaml

@@ -8,7 +8,6 @@ title: Samsung Exynos SoC SlimSSS (Slim Security SubSystem) module
 
 maintainers:
   - Krzysztof Kozlowski <krzk@kernel.org>
-  - Kamil Konieczny <k.konieczny@partner.samsung.com>
 
 description: |+
   The SlimSSS module in Exynos5433 SoC supports the following:

Documentation/devicetree/bindings/crypto/samsung-sss.yaml

@@ -8,7 +8,6 @@ title: Samsung Exynos SoC SSS (Security SubSystem) module
 
 maintainers:
   - Krzysztof Kozlowski <krzk@kernel.org>
-  - Kamil Konieczny <k.konieczny@partner.samsung.com>
 
 description: |+
   The SSS module in S5PV210 SoC supports the following:

MAINTAINERS

@@ -9032,6 +9032,17 @@ F:	drivers/crypto/keembay/keembay-ocs-aes-core.c
 F:	drivers/crypto/keembay/ocs-aes.c
 F:	drivers/crypto/keembay/ocs-aes.h
 
+INTEL KEEM BAY OCS HCU CRYPTO DRIVER
+M:	Daniele Alessandrelli <daniele.alessandrelli@intel.com>
+M:	Declan Murphy <declan.murphy@intel.com>
+S:	Maintained
+F:	Documentation/devicetree/bindings/crypto/intel,keembay-ocs-hcu.yaml
+F:	drivers/crypto/keembay/Kconfig
+F:	drivers/crypto/keembay/Makefile
+F:	drivers/crypto/keembay/keembay-ocs-hcu-core.c
+F:	drivers/crypto/keembay/ocs-hcu.c
+F:	drivers/crypto/keembay/ocs-hcu.h
+
 INTEL MANAGEMENT ENGINE (mei)
 M:	Tomas Winkler <tomas.winkler@intel.com>
 L:	linux-kernel@vger.kernel.org
@@ -15683,7 +15694,6 @@ F:	drivers/media/i2c/s5k5baf.c
 SAMSUNG S5P Security SubSystem (SSS) DRIVER
 M:	Krzysztof Kozlowski <krzk@kernel.org>
 M:	Vladimir Zapolskiy <vz@mleia.com>
-M:	Kamil Konieczny <k.konieczny@samsung.com>
 L:	linux-crypto@vger.kernel.org
 L:	linux-samsung-soc@vger.kernel.org
 S:	Maintained

arch/arm/crypto/Kconfig

@@ -62,6 +62,25 @@ config CRYPTO_SHA512_ARM
 	  SHA-512 secure hash standard (DFIPS 180-2) implemented
 	  using optimized ARM assembler and NEON, when available.
 
+config CRYPTO_BLAKE2S_ARM
+	tristate "BLAKE2s digest algorithm (ARM)"
+	select CRYPTO_ARCH_HAVE_LIB_BLAKE2S
+	help
+	  BLAKE2s digest algorithm optimized with ARM scalar instructions.  This
+	  is faster than the generic implementations of BLAKE2s and BLAKE2b, but
+	  slower than the NEON implementation of BLAKE2b.  (There is no NEON
+	  implementation of BLAKE2s, since NEON doesn't really help with it.)
+
+config CRYPTO_BLAKE2B_NEON
+	tristate "BLAKE2b digest algorithm (ARM NEON)"
+	depends on KERNEL_MODE_NEON
+	select CRYPTO_BLAKE2B
+	help
+	  BLAKE2b digest algorithm optimized with ARM NEON instructions.
+	  On ARM processors that have NEON support but not the ARMv8
+	  Crypto Extensions, typically this BLAKE2b implementation is
+	  much faster than SHA-2 and slightly faster than SHA-1.
+
 config CRYPTO_AES_ARM
 	tristate "Scalar AES cipher for ARM"
 	select CRYPTO_ALGAPI

arch/arm/crypto/Makefile

@@ -9,6 +9,8 @@ obj-$(CONFIG_CRYPTO_SHA1_ARM) += sha1-arm.o
 obj-$(CONFIG_CRYPTO_SHA1_ARM_NEON) += sha1-arm-neon.o
 obj-$(CONFIG_CRYPTO_SHA256_ARM) += sha256-arm.o
 obj-$(CONFIG_CRYPTO_SHA512_ARM) += sha512-arm.o
+obj-$(CONFIG_CRYPTO_BLAKE2S_ARM) += blake2s-arm.o
+obj-$(CONFIG_CRYPTO_BLAKE2B_NEON) += blake2b-neon.o
 obj-$(CONFIG_CRYPTO_CHACHA20_NEON) += chacha-neon.o
 obj-$(CONFIG_CRYPTO_POLY1305_ARM) += poly1305-arm.o
 obj-$(CONFIG_CRYPTO_NHPOLY1305_NEON) += nhpoly1305-neon.o
@@ -29,6 +31,8 @@ sha256-arm-neon-$(CONFIG_KERNEL_MODE_NEON) := sha256_neon_glue.o
 sha256-arm-y	:= sha256-core.o sha256_glue.o $(sha256-arm-neon-y)
 sha512-arm-neon-$(CONFIG_KERNEL_MODE_NEON) := sha512-neon-glue.o
 sha512-arm-y	:= sha512-core.o sha512-glue.o $(sha512-arm-neon-y)
+blake2s-arm-y   := blake2s-core.o blake2s-glue.o
+blake2b-neon-y  := blake2b-neon-core.o blake2b-neon-glue.o
 sha1-arm-ce-y	:= sha1-ce-core.o sha1-ce-glue.o
 sha2-arm-ce-y	:= sha2-ce-core.o sha2-ce-glue.o
 aes-arm-ce-y	:= aes-ce-core.o aes-ce-glue.o

arch/arm/crypto/aes-neonbs-glue.c

@@ -9,6 +9,7 @@
 #include <asm/simd.h>
 #include <crypto/aes.h>
 #include <crypto/ctr.h>
+#include <crypto/internal/cipher.h>
 #include <crypto/internal/simd.h>
 #include <crypto/internal/skcipher.h>
 #include <crypto/scatterwalk.h>
@@ -23,6 +24,8 @@ MODULE_ALIAS_CRYPTO("cbc(aes)-all");
 MODULE_ALIAS_CRYPTO("ctr(aes)");
 MODULE_ALIAS_CRYPTO("xts(aes)");
 
+MODULE_IMPORT_NS(CRYPTO_INTERNAL);
+
 asmlinkage void aesbs_convert_key(u8 out[], u32 const rk[], int rounds);
 
 asmlinkage void aesbs_ecb_encrypt(u8 out[], u8 const in[], u8 const rk[],

arch/arm/crypto/blake2b-neon-core.S

@@ -0,0 +1,347 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * BLAKE2b digest algorithm, NEON accelerated
+ *
+ * Copyright 2020 Google LLC
+ *
+ * Author: Eric Biggers <ebiggers@google.com>
+ */
+
+#include <linux/linkage.h>
+
+	.text
+	.fpu		neon
+
+	// The arguments to blake2b_compress_neon()
+	STATE		.req	r0
+	BLOCK		.req	r1
+	NBLOCKS		.req	r2
+	INC		.req	r3
+
+	// Pointers to the rotation tables
+	ROR24_TABLE	.req	r4
+	ROR16_TABLE	.req	r5
+
+	// The original stack pointer
+	ORIG_SP		.req	r6
+
+	// NEON registers which contain the message words of the current block.
+	// M_0-M_3 are occasionally used for other purposes too.
+	M_0		.req	d16
+	M_1		.req	d17
+	M_2		.req	d18
+	M_3		.req	d19
+	M_4		.req	d20
+	M_5		.req	d21
+	M_6		.req	d22
+	M_7		.req	d23
+	M_8		.req	d24
+	M_9		.req	d25
+	M_10		.req	d26
+	M_11		.req	d27
+	M_12		.req	d28
+	M_13		.req	d29
+	M_14		.req	d30
+	M_15		.req	d31
+
+	.align		4
+	// Tables for computing ror64(x, 24) and ror64(x, 16) using the vtbl.8
+	// instruction.  This is the most efficient way to implement these
+	// rotation amounts with NEON.  (On Cortex-A53 it's the same speed as
+	// vshr.u64 + vsli.u64, while on Cortex-A7 it's faster.)
+.Lror24_table:
+	.byte		3, 4, 5, 6, 7, 0, 1, 2
+.Lror16_table:
+	.byte		2, 3, 4, 5, 6, 7, 0, 1
+	// The BLAKE2b initialization vector
+.Lblake2b_IV:
+	.quad		0x6a09e667f3bcc908, 0xbb67ae8584caa73b
+	.quad		0x3c6ef372fe94f82b, 0xa54ff53a5f1d36f1
+	.quad		0x510e527fade682d1, 0x9b05688c2b3e6c1f
+	.quad		0x1f83d9abfb41bd6b, 0x5be0cd19137e2179
+
+// Execute one round of BLAKE2b by updating the state matrix v[0..15] in the
+// NEON registers q0-q7.  The message block is in q8..q15 (M_0-M_15).  The stack
+// pointer points to a 32-byte aligned buffer containing a copy of q8 and q9
+// (M_0-M_3), so that they can be reloaded if they are used as temporary
+// registers.  The macro arguments s0-s15 give the order in which the message
+// words are used in this round.  'final' is 1 if this is the final round.
+.macro	_blake2b_round	s0, s1, s2, s3, s4, s5, s6, s7, \
+			s8, s9, s10, s11, s12, s13, s14, s15, final=0
+
+	// Mix the columns:
+	// (v[0], v[4], v[8], v[12]), (v[1], v[5], v[9], v[13]),
+	// (v[2], v[6], v[10], v[14]), and (v[3], v[7], v[11], v[15]).
+
+	// a += b + m[blake2b_sigma[r][2*i + 0]];
+	vadd.u64	q0, q0, q2
+	vadd.u64	q1, q1, q3
+	vadd.u64	d0, d0, M_\s0
+	vadd.u64	d1, d1, M_\s2
+	vadd.u64	d2, d2, M_\s4
+	vadd.u64	d3, d3, M_\s6
+
+	// d = ror64(d ^ a, 32);
+	veor		q6, q6, q0
+	veor		q7, q7, q1
+	vrev64.32	q6, q6
+	vrev64.32	q7, q7
+
+	// c += d;
+	vadd.u64	q4, q4, q6
+	vadd.u64	q5, q5, q7
+
+	// b = ror64(b ^ c, 24);
+	vld1.8		{M_0}, [ROR24_TABLE, :64]
+	veor		q2, q2, q4
+	veor		q3, q3, q5
+	vtbl.8		d4, {d4}, M_0
+	vtbl.8		d5, {d5}, M_0
+	vtbl.8		d6, {d6}, M_0
+	vtbl.8		d7, {d7}, M_0
+
+	// a += b + m[blake2b_sigma[r][2*i + 1]];
+	//
+	// M_0 got clobbered above, so we have to reload it if any of the four
+	// message words this step needs happens to be M_0.  Otherwise we don't
+	// need to reload it here, as it will just get clobbered again below.
+.if \s1 == 0 || \s3 == 0 || \s5 == 0 || \s7 == 0
+	vld1.8		{M_0}, [sp, :64]
+.endif
+	vadd.u64	q0, q0, q2
+	vadd.u64	q1, q1, q3
+	vadd.u64	d0, d0, M_\s1
+	vadd.u64	d1, d1, M_\s3
+	vadd.u64	d2, d2, M_\s5
+	vadd.u64	d3, d3, M_\s7
+
+	// d = ror64(d ^ a, 16);
+	vld1.8		{M_0}, [ROR16_TABLE, :64]
+	veor		q6, q6, q0
+	veor		q7, q7, q1
+	vtbl.8		d12, {d12}, M_0
+	vtbl.8		d13, {d13}, M_0
+	vtbl.8		d14, {d14}, M_0
+	vtbl.8		d15, {d15}, M_0
+
+	// c += d;
+	vadd.u64	q4, q4, q6
+	vadd.u64	q5, q5, q7
+
+	// b = ror64(b ^ c, 63);
+	//
+	// This rotation amount isn't a multiple of 8, so it has to be
+	// implemented using a pair of shifts, which requires temporary
+	// registers.  Use q8-q9 (M_0-M_3) for this, and reload them afterwards.
+	veor		q8, q2, q4
+	veor		q9, q3, q5
+	vshr.u64	q2, q8, #63
+	vshr.u64	q3, q9, #63
+	vsli.u64	q2, q8, #1
+	vsli.u64	q3, q9, #1
+	vld1.8		{q8-q9}, [sp, :256]
+
+	// Mix the diagonals:
+	// (v[0], v[5], v[10], v[15]), (v[1], v[6], v[11], v[12]),
+	// (v[2], v[7], v[8], v[13]), and (v[3], v[4], v[9], v[14]).
+	//
+	// There are two possible ways to do this: use 'vext' instructions to
+	// shift the rows of the matrix so that the diagonals become columns,
+	// and undo it afterwards; or just use 64-bit operations on 'd'
+	// registers instead of 128-bit operations on 'q' registers.  We use the
+	// latter approach, as it performs much better on Cortex-A7.
+
+	// a += b + m[blake2b_sigma[r][2*i + 0]];
+	vadd.u64	d0, d0, d5
+	vadd.u64	d1, d1, d6
+	vadd.u64	d2, d2, d7
+	vadd.u64	d3, d3, d4
+	vadd.u64	d0, d0, M_\s8
+	vadd.u64	d1, d1, M_\s10
+	vadd.u64	d2, d2, M_\s12
+	vadd.u64	d3, d3, M_\s14
+
+	// d = ror64(d ^ a, 32);
+	veor		d15, d15, d0
+	veor		d12, d12, d1
+	veor		d13, d13, d2
+	veor		d14, d14, d3
+	vrev64.32	d15, d15
+	vrev64.32	d12, d12
+	vrev64.32	d13, d13
+	vrev64.32	d14, d14
+
+	// c += d;
+	vadd.u64	d10, d10, d15
+	vadd.u64	d11, d11, d12
+	vadd.u64	d8, d8, d13
+	vadd.u64	d9, d9, d14
+
+	// b = ror64(b ^ c, 24);
+	vld1.8		{M_0}, [ROR24_TABLE, :64]
+	veor		d5, d5, d10
+	veor		d6, d6, d11
+	veor		d7, d7, d8
+	veor		d4, d4, d9
+	vtbl.8		d5, {d5}, M_0
+	vtbl.8		d6, {d6}, M_0
+	vtbl.8		d7, {d7}, M_0
+	vtbl.8		d4, {d4}, M_0
+
+	// a += b + m[blake2b_sigma[r][2*i + 1]];
+.if \s9 == 0 || \s11 == 0 || \s13 == 0 || \s15 == 0
+	vld1.8		{M_0}, [sp, :64]
+.endif
+	vadd.u64	d0, d0, d5
+	vadd.u64	d1, d1, d6
+	vadd.u64	d2, d2, d7
+	vadd.u64	d3, d3, d4
+	vadd.u64	d0, d0, M_\s9
+	vadd.u64	d1, d1, M_\s11
+	vadd.u64	d2, d2, M_\s13
+	vadd.u64	d3, d3, M_\s15
+
+	// d = ror64(d ^ a, 16);
+	vld1.8		{M_0}, [ROR16_TABLE, :64]
+	veor		d15, d15, d0
+	veor		d12, d12, d1
+	veor		d13, d13, d2
+	veor		d14, d14, d3
+	vtbl.8		d12, {d12}, M_0
+	vtbl.8		d13, {d13}, M_0
+	vtbl.8		d14, {d14}, M_0
+	vtbl.8		d15, {d15}, M_0
+
+	// c += d;
+	vadd.u64	d10, d10, d15
+	vadd.u64	d11, d11, d12
+	vadd.u64	d8, d8, d13
+	vadd.u64	d9, d9, d14
+
+	// b = ror64(b ^ c, 63);
+	veor		d16, d4, d9
+	veor		d17, d5, d10
+	veor		d18, d6, d11
+	veor		d19, d7, d8
+	vshr.u64	q2, q8, #63
+	vshr.u64	q3, q9, #63
+	vsli.u64	q2, q8, #1
+	vsli.u64	q3, q9, #1
+	// Reloading q8-q9 can be skipped on the final round.
+.if ! \final
+	vld1.8		{q8-q9}, [sp, :256]
+.endif
+.endm
+
+//
+// void blake2b_compress_neon(struct blake2b_state *state,
+//			      const u8 *block, size_t nblocks, u32 inc);
+//
+// Only the first three fields of struct blake2b_state are used:
+//	u64 h[8];	(inout)
+//	u64 t[2];	(inout)
+//	u64 f[2];	(in)
+//
+	.align		5
+ENTRY(blake2b_compress_neon)
+	push		{r4-r10}
+
+	// Allocate a 32-byte stack buffer that is 32-byte aligned.
+	mov		ORIG_SP, sp
+	sub		ip, sp, #32
+	bic		ip, ip, #31
+	mov		sp, ip
+
+	adr		ROR24_TABLE, .Lror24_table
+	adr		ROR16_TABLE, .Lror16_table
+
+	mov		ip, STATE
+	vld1.64		{q0-q1}, [ip]!		// Load h[0..3]
+	vld1.64		{q2-q3}, [ip]!		// Load h[4..7]
+.Lnext_block:
+	  adr		r10, .Lblake2b_IV
+	vld1.64		{q14-q15}, [ip]		// Load t[0..1] and f[0..1]
+	vld1.64		{q4-q5}, [r10]!		// Load IV[0..3]
+	  vmov		r7, r8, d28		// Copy t[0] to (r7, r8)
+	vld1.64		{q6-q7}, [r10]		// Load IV[4..7]
+	  adds		r7, r7, INC		// Increment counter
+	bcs		.Lslow_inc_ctr
+	vmov.i32	d28[0], r7
+	vst1.64		{d28}, [ip]		// Update t[0]
+.Linc_ctr_done:
+
+	// Load the next message block and finish initializing the state matrix
+	// 'v'.  Fortunately, there are exactly enough NEON registers to fit the
+	// entire state matrix in q0-q7 and the entire message block in q8-15.
+	//
+	// However, _blake2b_round also needs some extra registers for rotates,
+	// so we have to spill some registers.  It's better to spill the message
+	// registers than the state registers, as the message doesn't change.
+	// Therefore we store a copy of the first 32 bytes of the message block
+	// (q8-q9) in an aligned buffer on the stack so that they can be
+	// reloaded when needed.  (We could just reload directly from the
+	// message buffer, but it's faster to use aligned loads.)
+	vld1.8		{q8-q9}, [BLOCK]!
+	  veor		q6, q6, q14	// v[12..13] = IV[4..5] ^ t[0..1]
+	vld1.8		{q10-q11}, [BLOCK]!
+	  veor		q7, q7, q15	// v[14..15] = IV[6..7] ^ f[0..1]
+	vld1.8		{q12-q13}, [BLOCK]!
+	vst1.8		{q8-q9}, [sp, :256]
+	  mov		ip, STATE
+	vld1.8		{q14-q15}, [BLOCK]!
+
+	// Execute the rounds.  Each round is provided the order in which it
+	// needs to use the message words.
+	_blake2b_round	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
+	_blake2b_round	14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3
+	_blake2b_round	11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4
+	_blake2b_round	7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8
+	_blake2b_round	9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13
+	_blake2b_round	2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9
+	_blake2b_round	12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11
+	_blake2b_round	13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10
+	_blake2b_round	6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5
+	_blake2b_round	10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0
+	_blake2b_round	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
+	_blake2b_round	14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 \
+			final=1
+
+	// Fold the final state matrix into the hash chaining value:
+	//
+	//	for (i = 0; i < 8; i++)
+	//		h[i] ^= v[i] ^ v[i + 8];
+	//
+	  vld1.64	{q8-q9}, [ip]!		// Load old h[0..3]
+	veor		q0, q0, q4		// v[0..1] ^= v[8..9]
+	veor		q1, q1, q5		// v[2..3] ^= v[10..11]
+	  vld1.64	{q10-q11}, [ip]		// Load old h[4..7]
+	veor		q2, q2, q6		// v[4..5] ^= v[12..13]
+	veor		q3, q3, q7		// v[6..7] ^= v[14..15]
+	veor		q0, q0, q8		// v[0..1] ^= h[0..1]
+	veor		q1, q1, q9		// v[2..3] ^= h[2..3]
+	  mov		ip, STATE
+	  subs		NBLOCKS, NBLOCKS, #1	// nblocks--
+	  vst1.64	{q0-q1}, [ip]!		// Store new h[0..3]
+	veor		q2, q2, q10		// v[4..5] ^= h[4..5]
+	veor		q3, q3, q11		// v[6..7] ^= h[6..7]
+	  vst1.64	{q2-q3}, [ip]!		// Store new h[4..7]
+
+	// Advance to the next block, if there is one.
+	bne		.Lnext_block		// nblocks != 0?
+
+	mov		sp, ORIG_SP
+	pop		{r4-r10}
+	mov		pc, lr
+
+.Lslow_inc_ctr:
+	// Handle the case where the counter overflowed its low 32 bits, by
+	// carrying the overflow bit into the full 128-bit counter.
+	vmov		r9, r10, d29
+	adcs		r8, r8, #0
+	adcs		r9, r9, #0
+	adc		r10, r10, #0
+	vmov		d28, r7, r8
+	vmov		d29, r9, r10
+	vst1.64		{q14}, [ip]		// Update t[0] and t[1]
+	b		.Linc_ctr_done
+ENDPROC(blake2b_compress_neon)

arch/arm/crypto/blake2b-neon-glue.c

@@ -0,0 +1,105 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * BLAKE2b digest algorithm, NEON accelerated
+ *
+ * Copyright 2020 Google LLC
+ */
+
+#include <crypto/internal/blake2b.h>
+#include <crypto/internal/hash.h>
+#include <crypto/internal/simd.h>
+
+#include <linux/module.h>
+#include <linux/sizes.h>
+
+#include <asm/neon.h>
+#include <asm/simd.h>
+
+asmlinkage void blake2b_compress_neon(struct blake2b_state *state,
+				      const u8 *block, size_t nblocks, u32 inc);
+
+static void blake2b_compress_arch(struct blake2b_state *state,
+				  const u8 *block, size_t nblocks, u32 inc)
+{
+	if (!crypto_simd_usable()) {
+		blake2b_compress_generic(state, block, nblocks, inc);
+		return;
+	}
+
+	do {
+		const size_t blocks = min_t(size_t, nblocks,
+					    SZ_4K / BLAKE2B_BLOCK_SIZE);
+
+		kernel_neon_begin();
+		blake2b_compress_neon(state, block, blocks, inc);
+		kernel_neon_end();
+
+		nblocks -= blocks;
+		block += blocks * BLAKE2B_BLOCK_SIZE;
+	} while (nblocks);
+}
+
+static int crypto_blake2b_update_neon(struct shash_desc *desc,
+				      const u8 *in, unsigned int inlen)
+{
+	return crypto_blake2b_update(desc, in, inlen, blake2b_compress_arch);
+}
+
+static int crypto_blake2b_final_neon(struct shash_desc *desc, u8 *out)
+{
+	return crypto_blake2b_final(desc, out, blake2b_compress_arch);
+}
+
+#define BLAKE2B_ALG(name, driver_name, digest_size)			\
+	{								\
+		.base.cra_name		= name,				\
+		.base.cra_driver_name	= driver_name,			\
+		.base.cra_priority	= 200,				\
+		.base.cra_flags		= CRYPTO_ALG_OPTIONAL_KEY,	\
+		.base.cra_blocksize	= BLAKE2B_BLOCK_SIZE,		\
+		.base.cra_ctxsize	= sizeof(struct blake2b_tfm_ctx), \
+		.base.cra_module	= THIS_MODULE,			\
+		.digestsize		= digest_size,			\
+		.setkey			= crypto_blake2b_setkey,	\
+		.init			= crypto_blake2b_init,		\
+		.update			= crypto_blake2b_update_neon,	\
+		.final			= crypto_blake2b_final_neon,	\
+		.descsize		= sizeof(struct blake2b_state),	\
+	}
+
+static struct shash_alg blake2b_neon_algs[] = {
+	BLAKE2B_ALG("blake2b-160", "blake2b-160-neon", BLAKE2B_160_HASH_SIZE),
+	BLAKE2B_ALG("blake2b-256", "blake2b-256-neon", BLAKE2B_256_HASH_SIZE),
+	BLAKE2B_ALG("blake2b-384", "blake2b-384-neon", BLAKE2B_384_HASH_SIZE),
+	BLAKE2B_ALG("blake2b-512", "blake2b-512-neon", BLAKE2B_512_HASH_SIZE),
+};
+
+static int __init blake2b_neon_mod_init(void)
+{
+	if (!(elf_hwcap & HWCAP_NEON))
+		return -ENODEV;
+
+	return crypto_register_shashes(blake2b_neon_algs,
+				       ARRAY_SIZE(blake2b_neon_algs));
+}
+
+static void __exit blake2b_neon_mod_exit(void)
+{
+	return crypto_unregister_shashes(blake2b_neon_algs,
+					 ARRAY_SIZE(blake2b_neon_algs));
+}
+
+module_init(blake2b_neon_mod_init);
+module_exit(blake2b_neon_mod_exit);
+
+MODULE_DESCRIPTION("BLAKE2b digest algorithm, NEON accelerated");
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Eric Biggers <ebiggers@google.com>");
+MODULE_ALIAS_CRYPTO("blake2b-160");
+MODULE_ALIAS_CRYPTO("blake2b-160-neon");
+MODULE_ALIAS_CRYPTO("blake2b-256");
+MODULE_ALIAS_CRYPTO("blake2b-256-neon");
+MODULE_ALIAS_CRYPTO("blake2b-384");
+MODULE_ALIAS_CRYPTO("blake2b-384-neon");
+MODULE_ALIAS_CRYPTO("blake2b-512");
+MODULE_ALIAS_CRYPTO("blake2b-512-neon");

arch/arm/crypto/blake2s-core.S

@@ -0,0 +1,285 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * BLAKE2s digest algorithm, ARM scalar implementation
+ *
+ * Copyright 2020 Google LLC
+ *
+ * Author: Eric Biggers <ebiggers@google.com>
+ */
+
+#include <linux/linkage.h>
+
+	// Registers used to hold message words temporarily.  There aren't
+	// enough ARM registers to hold the whole message block, so we have to
+	// load the words on-demand.
+	M_0		.req	r12
+	M_1		.req	r14
+
+// The BLAKE2s initialization vector
+.Lblake2s_IV:
+	.word	0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A
+	.word	0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
+
+.macro __ldrd		a, b, src, offset
+#if __LINUX_ARM_ARCH__ >= 6
+	ldrd		\a, \b, [\src, #\offset]
+#else
+	ldr		\a, [\src, #\offset]
+	ldr		\b, [\src, #\offset + 4]
+#endif
+.endm
+
+.macro __strd		a, b, dst, offset
+#if __LINUX_ARM_ARCH__ >= 6
+	strd		\a, \b, [\dst, #\offset]
+#else
+	str		\a, [\dst, #\offset]
+	str		\b, [\dst, #\offset + 4]
+#endif
+.endm
+
+// Execute a quarter-round of BLAKE2s by mixing two columns or two diagonals.
+// (a0, b0, c0, d0) and (a1, b1, c1, d1) give the registers containing the two
+// columns/diagonals.  s0-s1 are the word offsets to the message words the first
+// column/diagonal needs, and likewise s2-s3 for the second column/diagonal.
+// M_0 and M_1 are free to use, and the message block can be found at sp + 32.
+//
+// Note that to save instructions, the rotations don't happen when the
+// pseudocode says they should, but rather they are delayed until the values are
+// used.  See the comment above _blake2s_round().
+.macro _blake2s_quarterround  a0, b0, c0, d0,  a1, b1, c1, d1,  s0, s1, s2, s3
+
+	ldr		M_0, [sp, #32 + 4 * \s0]
+	ldr		M_1, [sp, #32 + 4 * \s2]
+
+	// a += b + m[blake2s_sigma[r][2*i + 0]];
+	add		\a0, \a0, \b0, ror #brot
+	add		\a1, \a1, \b1, ror #brot
+	add		\a0, \a0, M_0
+	add		\a1, \a1, M_1
+
+	// d = ror32(d ^ a, 16);
+	eor		\d0, \a0, \d0, ror #drot
+	eor		\d1, \a1, \d1, ror #drot
+
+	// c += d;
+	add		\c0, \c0, \d0, ror #16
+	add		\c1, \c1, \d1, ror #16
+
+	// b = ror32(b ^ c, 12);
+	eor		\b0, \c0, \b0, ror #brot
+	eor		\b1, \c1, \b1, ror #brot
+
+	ldr		M_0, [sp, #32 + 4 * \s1]
+	ldr		M_1, [sp, #32 + 4 * \s3]
+
+	// a += b + m[blake2s_sigma[r][2*i + 1]];
+	add		\a0, \a0, \b0, ror #12
+	add		\a1, \a1, \b1, ror #12
+	add		\a0, \a0, M_0
+	add		\a1, \a1, M_1
+
+	// d = ror32(d ^ a, 8);
+	eor		\d0, \a0, \d0, ror#16
+	eor		\d1, \a1, \d1, ror#16
+
+	// c += d;
+	add		\c0, \c0, \d0, ror#8
+	add		\c1, \c1, \d1, ror#8
+
+	// b = ror32(b ^ c, 7);
+	eor		\b0, \c0, \b0, ror#12
+	eor		\b1, \c1, \b1, ror#12
+.endm
+
+// Execute one round of BLAKE2s by updating the state matrix v[0..15].  v[0..9]
+// are in r0..r9.  The stack pointer points to 8 bytes of scratch space for
+// spilling v[8..9], then to v[9..15], then to the message block.  r10-r12 and
+// r14 are free to use.  The macro arguments s0-s15 give the order in which the
+// message words are used in this round.
+//
+// All rotates are performed using the implicit rotate operand accepted by the
+// 'add' and 'eor' instructions.  This is faster than using explicit rotate
+// instructions.  To make this work, we allow the values in the second and last
+// rows of the BLAKE2s state matrix (rows 'b' and 'd') to temporarily have the
+// wrong rotation amount.  The rotation amount is then fixed up just in time
+// when the values are used.  'brot' is the number of bits the values in row 'b'
+// need to be rotated right to arrive at the correct values, and 'drot'
+// similarly for row 'd'.  (brot, drot) start out as (0, 0) but we make it such
+// that they end up as (7, 8) after every round.
+.macro	_blake2s_round	s0, s1, s2, s3, s4, s5, s6, s7, \
+			s8, s9, s10, s11, s12, s13, s14, s15
+
+	// Mix first two columns:
+	// (v[0], v[4], v[8], v[12]) and (v[1], v[5], v[9], v[13]).
+	__ldrd		r10, r11, sp, 16	// load v[12] and v[13]
+	_blake2s_quarterround	r0, r4, r8, r10,  r1, r5, r9, r11, \
+				\s0, \s1, \s2, \s3
+	__strd		r8, r9, sp, 0
+	__strd		r10, r11, sp, 16
+
+	// Mix second two columns:
+	// (v[2], v[6], v[10], v[14]) and (v[3], v[7], v[11], v[15]).
+	__ldrd		r8, r9, sp, 8		// load v[10] and v[11]
+	__ldrd		r10, r11, sp, 24	// load v[14] and v[15]
+	_blake2s_quarterround	r2, r6, r8, r10,  r3, r7, r9, r11, \
+				\s4, \s5, \s6, \s7
+	str		r10, [sp, #24]		// store v[14]
+	// v[10], v[11], and v[15] are used below, so no need to store them yet.
+
+	.set brot, 7
+	.set drot, 8
+
+	// Mix first two diagonals:
+	// (v[0], v[5], v[10], v[15]) and (v[1], v[6], v[11], v[12]).
+	ldr		r10, [sp, #16]		// load v[12]
+	_blake2s_quarterround	r0, r5, r8, r11,  r1, r6, r9, r10, \
+				\s8, \s9, \s10, \s11
+	__strd		r8, r9, sp, 8
+	str		r11, [sp, #28]
+	str		r10, [sp, #16]
+
+	// Mix second two diagonals:
+	// (v[2], v[7], v[8], v[13]) and (v[3], v[4], v[9], v[14]).
+	__ldrd		r8, r9, sp, 0		// load v[8] and v[9]
+	__ldrd		r10, r11, sp, 20	// load v[13] and v[14]
+	_blake2s_quarterround	r2, r7, r8, r10,  r3, r4, r9, r11, \
+				\s12, \s13, \s14, \s15
+	__strd		r10, r11, sp, 20
+.endm
+
+//
+// void blake2s_compress_arch(struct blake2s_state *state,
+//			      const u8 *block, size_t nblocks, u32 inc);
+//
+// Only the first three fields of struct blake2s_state are used:
+//	u32 h[8];	(inout)
+//	u32 t[2];	(inout)
+//	u32 f[2];	(in)
+//
+	.align		5
+ENTRY(blake2s_compress_arch)
+	push		{r0-r2,r4-r11,lr}	// keep this an even number
+
+.Lnext_block:
+	// r0 is 'state'
+	// r1 is 'block'
+	// r3 is 'inc'
+
+	// Load and increment the counter t[0..1].
+	__ldrd		r10, r11, r0, 32
+	adds		r10, r10, r3
+	adc		r11, r11, #0
+	__strd		r10, r11, r0, 32
+
+	// _blake2s_round is very short on registers, so copy the message block
+	// to the stack to save a register during the rounds.  This also has the
+	// advantage that misalignment only needs to be dealt with in one place.
+	sub		sp, sp, #64
+	mov		r12, sp
+	tst		r1, #3
+	bne		.Lcopy_block_misaligned
+	ldmia		r1!, {r2-r9}
+	stmia		r12!, {r2-r9}
+	ldmia		r1!, {r2-r9}
+	stmia		r12, {r2-r9}
+.Lcopy_block_done:
+	str		r1, [sp, #68]		// Update message pointer
+
+	// Calculate v[8..15].  Push v[9..15] onto the stack, and leave space
+	// for spilling v[8..9].  Leave v[8..9] in r8-r9.
+	mov		r14, r0			// r14 = state
+	adr		r12, .Lblake2s_IV
+	ldmia		r12!, {r8-r9}		// load IV[0..1]
+	__ldrd		r0, r1, r14, 40		// load f[0..1]
+	ldm		r12, {r2-r7}		// load IV[3..7]
+	eor		r4, r4, r10		// v[12] = IV[4] ^ t[0]
+	eor		r5, r5, r11		// v[13] = IV[5] ^ t[1]
+	eor		r6, r6, r0		// v[14] = IV[6] ^ f[0]
+	eor		r7, r7, r1		// v[15] = IV[7] ^ f[1]
+	push		{r2-r7}			// push v[9..15]
+	sub		sp, sp, #8		// leave space for v[8..9]
+
+	// Load h[0..7] == v[0..7].
+	ldm		r14, {r0-r7}
+
+	// Execute the rounds.  Each round is provided the order in which it
+	// needs to use the message words.
+	.set brot, 0
+	.set drot, 0
+	_blake2s_round	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
+	_blake2s_round	14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3
+	_blake2s_round	11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4
+	_blake2s_round	7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8
+	_blake2s_round	9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13
+	_blake2s_round	2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9
+	_blake2s_round	12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11
+	_blake2s_round	13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10
+	_blake2s_round	6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5
+	_blake2s_round	10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0
+
+	// Fold the final state matrix into the hash chaining value:
+	//
+	//	for (i = 0; i < 8; i++)
+	//		h[i] ^= v[i] ^ v[i + 8];
+	//
+	ldr		r14, [sp, #96]		// r14 = &h[0]
+	add		sp, sp, #8		// v[8..9] are already loaded.
+	pop		{r10-r11}		// load v[10..11]
+	eor		r0, r0, r8
+	eor		r1, r1, r9
+	eor		r2, r2, r10
+	eor		r3, r3, r11
+	ldm		r14, {r8-r11}		// load h[0..3]
+	eor		r0, r0, r8
+	eor		r1, r1, r9
+	eor		r2, r2, r10
+	eor		r3, r3, r11
+	stmia		r14!, {r0-r3}		// store new h[0..3]
+	ldm		r14, {r0-r3}		// load old h[4..7]
+	pop		{r8-r11}		// load v[12..15]
+	eor		r0, r0, r4, ror #brot
+	eor		r1, r1, r5, ror #brot
+	eor		r2, r2, r6, ror #brot
+	eor		r3, r3, r7, ror #brot
+	eor		r0, r0, r8, ror #drot
+	eor		r1, r1, r9, ror #drot
+	eor		r2, r2, r10, ror #drot
+	eor		r3, r3, r11, ror #drot
+	  add		sp, sp, #64		// skip copy of message block
+	stm		r14, {r0-r3}		// store new h[4..7]
+
+	// Advance to the next block, if there is one.  Note that if there are
+	// multiple blocks, then 'inc' (the counter increment amount) must be
+	// 64.  So we can simply set it to 64 without re-loading it.
+	ldm		sp, {r0, r1, r2}	// load (state, block, nblocks)
+	mov		r3, #64			// set 'inc'
+	subs		r2, r2, #1		// nblocks--
+	str		r2, [sp, #8]
+	bne		.Lnext_block		// nblocks != 0?
+
+	pop		{r0-r2,r4-r11,pc}
+
+	// The next message block (pointed to by r1) isn't 4-byte aligned, so it
+	// can't be loaded using ldmia.  Copy it to the stack buffer (pointed to
+	// by r12) using an alternative method.  r2-r9 are free to use.
+.Lcopy_block_misaligned:
+	mov		r2, #64
+1:
+#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+	ldr		r3, [r1], #4
+#else
+	ldrb		r3, [r1, #0]
+	ldrb		r4, [r1, #1]
+	ldrb		r5, [r1, #2]
+	ldrb		r6, [r1, #3]
+	add		r1, r1, #4
+	orr		r3, r3, r4, lsl #8
+	orr		r3, r3, r5, lsl #16
+	orr		r3, r3, r6, lsl #24
+#endif
+	subs		r2, r2, #4
+	str		r3, [r12], #4
+	bne		1b
+	b		.Lcopy_block_done
+ENDPROC(blake2s_compress_arch)

arch/arm/crypto/blake2s-glue.c

@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * BLAKE2s digest algorithm, ARM scalar implementation
+ *
+ * Copyright 2020 Google LLC
+ */
+
+#include <crypto/internal/blake2s.h>
+#include <crypto/internal/hash.h>
+
+#include <linux/module.h>
+
+/* defined in blake2s-core.S */
+EXPORT_SYMBOL(blake2s_compress_arch);
+
+static int crypto_blake2s_update_arm(struct shash_desc *desc,
+				     const u8 *in, unsigned int inlen)
+{
+	return crypto_blake2s_update(desc, in, inlen, blake2s_compress_arch);
+}
+
+static int crypto_blake2s_final_arm(struct shash_desc *desc, u8 *out)
+{
+	return crypto_blake2s_final(desc, out, blake2s_compress_arch);
+}
+
+#define BLAKE2S_ALG(name, driver_name, digest_size)			\
+	{								\
+		.base.cra_name		= name,				\
+		.base.cra_driver_name	= driver_name,			\
+		.base.cra_priority	= 200,				\
+		.base.cra_flags		= CRYPTO_ALG_OPTIONAL_KEY,	\
+		.base.cra_blocksize	= BLAKE2S_BLOCK_SIZE,		\
+		.base.cra_ctxsize	= sizeof(struct blake2s_tfm_ctx), \
+		.base.cra_module	= THIS_MODULE,			\
+		.digestsize		= digest_size,			\
+		.setkey			= crypto_blake2s_setkey,	\
+		.init			= crypto_blake2s_init,		\
+		.update			= crypto_blake2s_update_arm,	\
+		.final			= crypto_blake2s_final_arm,	\
+		.descsize		= sizeof(struct blake2s_state),	\
+	}
+
+static struct shash_alg blake2s_arm_algs[] = {
+	BLAKE2S_ALG("blake2s-128", "blake2s-128-arm", BLAKE2S_128_HASH_SIZE),
+	BLAKE2S_ALG("blake2s-160", "blake2s-160-arm", BLAKE2S_160_HASH_SIZE),
+	BLAKE2S_ALG("blake2s-224", "blake2s-224-arm", BLAKE2S_224_HASH_SIZE),
+	BLAKE2S_ALG("blake2s-256", "blake2s-256-arm", BLAKE2S_256_HASH_SIZE),
+};
+
+static int __init blake2s_arm_mod_init(void)
+{
+	return IS_REACHABLE(CONFIG_CRYPTO_HASH) ?
+		crypto_register_shashes(blake2s_arm_algs,
+					ARRAY_SIZE(blake2s_arm_algs)) : 0;
+}
+
+static void __exit blake2s_arm_mod_exit(void)
+{
+	if (IS_REACHABLE(CONFIG_CRYPTO_HASH))
+		crypto_unregister_shashes(blake2s_arm_algs,
+					  ARRAY_SIZE(blake2s_arm_algs));
+}
+
+module_init(blake2s_arm_mod_init);
+module_exit(blake2s_arm_mod_exit);
+
+MODULE_DESCRIPTION("BLAKE2s digest algorithm, ARM scalar implementation");
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Eric Biggers <ebiggers@google.com>");
+MODULE_ALIAS_CRYPTO("blake2s-128");
+MODULE_ALIAS_CRYPTO("blake2s-128-arm");
+MODULE_ALIAS_CRYPTO("blake2s-160");
+MODULE_ALIAS_CRYPTO("blake2s-160-arm");
+MODULE_ALIAS_CRYPTO("blake2s-224");
+MODULE_ALIAS_CRYPTO("blake2s-224-arm");
+MODULE_ALIAS_CRYPTO("blake2s-256");
+MODULE_ALIAS_CRYPTO("blake2s-256-arm");

arch/arm64/crypto/aes-glue.c

@@ -24,6 +24,7 @@
 #ifdef USE_V8_CRYPTO_EXTENSIONS
 #define MODE			"ce"
 #define PRIO			300
+#define STRIDE			5
 #define aes_expandkey		ce_aes_expandkey
 #define aes_ecb_encrypt		ce_aes_ecb_encrypt
 #define aes_ecb_decrypt		ce_aes_ecb_decrypt
@@ -41,6 +42,7 @@ MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS using ARMv8 Crypto Extensions");
 #else
 #define MODE			"neon"
 #define PRIO			200
+#define STRIDE			4
 #define aes_ecb_encrypt		neon_aes_ecb_encrypt
 #define aes_ecb_decrypt		neon_aes_ecb_decrypt
 #define aes_cbc_encrypt		neon_aes_cbc_encrypt
@@ -55,7 +57,7 @@ MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS using ARMv8 Crypto Extensions");
 #define aes_mac_update		neon_aes_mac_update
 MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS using ARMv8 NEON");
 #endif
-#if defined(USE_V8_CRYPTO_EXTENSIONS) || !defined(CONFIG_CRYPTO_AES_ARM64_BS)
+#if defined(USE_V8_CRYPTO_EXTENSIONS) || !IS_ENABLED(CONFIG_CRYPTO_AES_ARM64_BS)
 MODULE_ALIAS_CRYPTO("ecb(aes)");
 MODULE_ALIAS_CRYPTO("cbc(aes)");
 MODULE_ALIAS_CRYPTO("ctr(aes)");
@@ -87,7 +89,7 @@ asmlinkage void aes_cbc_cts_decrypt(u8 out[], u8 const in[], u32 const rk[],
 				int rounds, int bytes, u8 const iv[]);
 
 asmlinkage void aes_ctr_encrypt(u8 out[], u8 const in[], u32 const rk[],
-				int rounds, int blocks, u8 ctr[]);
+				int rounds, int bytes, u8 ctr[], u8 finalbuf[]);
 
 asmlinkage void aes_xts_encrypt(u8 out[], u8 const in[], u32 const rk1[],
 				int rounds, int bytes, u32 const rk2[], u8 iv[],
@@ -103,9 +105,9 @@ asmlinkage void aes_essiv_cbc_decrypt(u8 out[], u8 const in[], u32 const rk1[],
 				      int rounds, int blocks, u8 iv[],
 				      u32 const rk2[]);
 
-asmlinkage void aes_mac_update(u8 const in[], u32 const rk[], int rounds,
-			       int blocks, u8 dg[], int enc_before,
-			       int enc_after);
+asmlinkage int aes_mac_update(u8 const in[], u32 const rk[], int rounds,
+			      int blocks, u8 dg[], int enc_before,
+			      int enc_after);
 
 struct crypto_aes_xts_ctx {
 	struct crypto_aes_ctx key1;
@@ -448,34 +450,36 @@ static int ctr_encrypt(struct skcipher_request *req)
 	struct crypto_aes_ctx *ctx = crypto_skcipher_ctx(tfm);
 	int err, rounds = 6 + ctx->key_length / 4;
 	struct skcipher_walk walk;
-	int blocks;
 
 	err = skcipher_walk_virt(&walk, req, false);
 
-	while ((blocks = (walk.nbytes / AES_BLOCK_SIZE))) {
-		kernel_neon_begin();
-		aes_ctr_encrypt(walk.dst.virt.addr, walk.src.virt.addr,
-				ctx->key_enc, rounds, blocks, walk.iv);
-		kernel_neon_end();
-		err = skcipher_walk_done(&walk, walk.nbytes % AES_BLOCK_SIZE);
-	}
-	if (walk.nbytes) {
-		u8 __aligned(8) tail[AES_BLOCK_SIZE];
+	while (walk.nbytes > 0) {
+		const u8 *src = walk.src.virt.addr;
 		unsigned int nbytes = walk.nbytes;
-		u8 *tdst = walk.dst.virt.addr;
-		u8 *tsrc = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+		u8 buf[AES_BLOCK_SIZE];
+		unsigned int tail;
 
-		/*
-		 * Tell aes_ctr_encrypt() to process a tail block.
-		 */
-		blocks = -1;
+		if (unlikely(nbytes < AES_BLOCK_SIZE))
+			src = memcpy(buf, src, nbytes);
+		else if (nbytes < walk.total)
+			nbytes &= ~(AES_BLOCK_SIZE - 1);
 
 		kernel_neon_begin();
-		aes_ctr_encrypt(tail, NULL, ctx->key_enc, rounds,
-				blocks, walk.iv);
+		aes_ctr_encrypt(dst, src, ctx->key_enc, rounds, nbytes,
+				walk.iv, buf);
 		kernel_neon_end();
-		crypto_xor_cpy(tdst, tsrc, tail, nbytes);
-		err = skcipher_walk_done(&walk, 0);
+
+		tail = nbytes % (STRIDE * AES_BLOCK_SIZE);
+		if (tail > 0 && tail < AES_BLOCK_SIZE)
+			/*
+			 * The final partial block could not be returned using
+			 * an overlapping store, so it was passed via buf[]
+			 * instead.
+			 */
+			memcpy(dst + nbytes - tail, buf, tail);
+
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
 	}
 
 	return err;
@@ -650,7 +654,7 @@ static int __maybe_unused xts_decrypt(struct skcipher_request *req)
 }
 
 static struct skcipher_alg aes_algs[] = { {
-#if defined(USE_V8_CRYPTO_EXTENSIONS) || !defined(CONFIG_CRYPTO_AES_ARM64_BS)
+#if defined(USE_V8_CRYPTO_EXTENSIONS) || !IS_ENABLED(CONFIG_CRYPTO_AES_ARM64_BS)
 	.base = {
 		.cra_name		= "__ecb(aes)",
 		.cra_driver_name	= "__ecb-aes-" MODE,
@@ -852,10 +856,17 @@ static void mac_do_update(struct crypto_aes_ctx *ctx, u8 const in[], int blocks,
 	int rounds = 6 + ctx->key_length / 4;
 
 	if (crypto_simd_usable()) {
-		kernel_neon_begin();
-		aes_mac_update(in, ctx->key_enc, rounds, blocks, dg, enc_before,
-			       enc_after);
-		kernel_neon_end();
+		int rem;
+
+		do {
+			kernel_neon_begin();
+			rem = aes_mac_update(in, ctx->key_enc, rounds, blocks,
+					     dg, enc_before, enc_after);
+			kernel_neon_end();
+			in += (blocks - rem) * AES_BLOCK_SIZE;
+			blocks = rem;
+			enc_before = 0;
+		} while (blocks);
 	} else {
 		if (enc_before)
 			aes_encrypt(ctx, dg, dg);

arch/arm64/crypto/aes-modes.S

@@ -321,42 +321,76 @@ AES_FUNC_END(aes_cbc_cts_decrypt)
 
 	/*
 	 * aes_ctr_encrypt(u8 out[], u8 const in[], u8 const rk[], int rounds,
-	 *		   int blocks, u8 ctr[])
+	 *		   int bytes, u8 ctr[], u8 finalbuf[])
 	 */
 
 AES_FUNC_START(aes_ctr_encrypt)
 	stp		x29, x30, [sp, #-16]!
 	mov		x29, sp
 
-	enc_prepare	w3, x2, x6
+	enc_prepare	w3, x2, x12
 	ld1		{vctr.16b}, [x5]
 
-	umov		x6, vctr.d[1]		/* keep swabbed ctr in reg */
-	rev		x6, x6
-	cmn		w6, w4			/* 32 bit overflow? */
-	bcs		.Lctrloop
+	umov		x12, vctr.d[1]		/* keep swabbed ctr in reg */
+	rev		x12, x12
+
 .LctrloopNx:
-	subs		w4, w4, #MAX_STRIDE
-	bmi		.Lctr1x
-	add		w7, w6, #1
+	add		w7, w4, #15
+	sub		w4, w4, #MAX_STRIDE << 4
+	lsr		w7, w7, #4
+	mov		w8, #MAX_STRIDE
+	cmp		w7, w8
+	csel		w7, w7, w8, lt
+	adds		x12, x12, x7
+
 	mov		v0.16b, vctr.16b
-	add		w8, w6, #2
 	mov		v1.16b, vctr.16b
-	add		w9, w6, #3
 	mov		v2.16b, vctr.16b
-	add		w9, w6, #3
-	rev		w7, w7
 	mov		v3.16b, vctr.16b
-	rev		w8, w8
 ST5(	mov		v4.16b, vctr.16b		)
-	mov		v1.s[3], w7
-	rev		w9, w9
-ST5(	add		w10, w6, #4			)
-	mov		v2.s[3], w8
-ST5(	rev		w10, w10			)
-	mov		v3.s[3], w9
-ST5(	mov		v4.s[3], w10			)
-	ld1		{v5.16b-v7.16b}, [x1], #48	/* get 3 input blocks */
+	bcs		0f
+
+	.subsection	1
+	/* apply carry to outgoing counter */
+0:	umov		x8, vctr.d[0]
+	rev		x8, x8
+	add		x8, x8, #1
+	rev		x8, x8
+	ins		vctr.d[0], x8
+
+	/* apply carry to N counter blocks for N := x12 */
+	adr		x16, 1f
+	sub		x16, x16, x12, lsl #3
+	br		x16
+	hint		34			// bti c
+	mov		v0.d[0], vctr.d[0]
+	hint		34			// bti c
+	mov		v1.d[0], vctr.d[0]
+	hint		34			// bti c
+	mov		v2.d[0], vctr.d[0]
+	hint		34			// bti c
+	mov		v3.d[0], vctr.d[0]
+ST5(	hint		34				)
+ST5(	mov		v4.d[0], vctr.d[0]		)
+1:	b		2f
+	.previous
+
+2:	rev		x7, x12
+	ins		vctr.d[1], x7
+	sub		x7, x12, #MAX_STRIDE - 1
+	sub		x8, x12, #MAX_STRIDE - 2
+	sub		x9, x12, #MAX_STRIDE - 3
+	rev		x7, x7
+	rev		x8, x8
+	mov		v1.d[1], x7
+	rev		x9, x9
+ST5(	sub		x10, x12, #MAX_STRIDE - 4	)
+	mov		v2.d[1], x8
+ST5(	rev		x10, x10			)
+	mov		v3.d[1], x9
+ST5(	mov		v4.d[1], x10			)
+	tbnz		w4, #31, .Lctrtail
+	ld1		{v5.16b-v7.16b}, [x1], #48
 ST4(	bl		aes_encrypt_block4x		)
 ST5(	bl		aes_encrypt_block5x		)
 	eor		v0.16b, v5.16b, v0.16b
@@ -368,47 +402,72 @@ ST5(	ld1		{v5.16b-v6.16b}, [x1], #32	)
 ST5(	eor		v4.16b, v6.16b, v4.16b		)
 	st1		{v0.16b-v3.16b}, [x0], #64
 ST5(	st1		{v4.16b}, [x0], #16		)
-	add		x6, x6, #MAX_STRIDE
-	rev		x7, x6
-	ins		vctr.d[1], x7
 	cbz		w4, .Lctrout
 	b		.LctrloopNx
-.Lctr1x:
-	adds		w4, w4, #MAX_STRIDE
-	beq		.Lctrout
-.Lctrloop:
-	mov		v0.16b, vctr.16b
-	encrypt_block	v0, w3, x2, x8, w7
-
-	adds		x6, x6, #1		/* increment BE ctr */
-	rev		x7, x6
-	ins		vctr.d[1], x7
-	bcs		.Lctrcarry		/* overflow? */
-
-.Lctrcarrydone:
-	subs		w4, w4, #1
-	bmi		.Lctrtailblock		/* blocks <0 means tail block */
-	ld1		{v3.16b}, [x1], #16
-	eor		v3.16b, v0.16b, v3.16b
-	st1		{v3.16b}, [x0], #16
-	bne		.Lctrloop
 
 .Lctrout:
 	st1		{vctr.16b}, [x5]	/* return next CTR value */
 	ldp		x29, x30, [sp], #16
 	ret
 
-.Lctrtailblock:
-	st1		{v0.16b}, [x0]
+.Lctrtail:
+	/* XOR up to MAX_STRIDE * 16 - 1 bytes of in/output with v0 ... v3/v4 */
+	mov		x16, #16
+	ands		x13, x4, #0xf
+	csel		x13, x13, x16, ne
+
+ST5(	cmp		w4, #64 - (MAX_STRIDE << 4)	)
+ST5(	csel		x14, x16, xzr, gt		)
+	cmp		w4, #48 - (MAX_STRIDE << 4)
+	csel		x15, x16, xzr, gt
+	cmp		w4, #32 - (MAX_STRIDE << 4)
+	csel		x16, x16, xzr, gt
+	cmp		w4, #16 - (MAX_STRIDE << 4)
+	ble		.Lctrtail1x
+
+	adr_l		x12, .Lcts_permute_table
+	add		x12, x12, x13
+
+ST5(	ld1		{v5.16b}, [x1], x14		)
+	ld1		{v6.16b}, [x1], x15
+	ld1		{v7.16b}, [x1], x16
+
+ST4(	bl		aes_encrypt_block4x		)
+ST5(	bl		aes_encrypt_block5x		)
+
+	ld1		{v8.16b}, [x1], x13
+	ld1		{v9.16b}, [x1]
+	ld1		{v10.16b}, [x12]
+
+ST4(	eor		v6.16b, v6.16b, v0.16b		)
+ST4(	eor		v7.16b, v7.16b, v1.16b		)
+ST4(	tbl		v3.16b, {v3.16b}, v10.16b	)
+ST4(	eor		v8.16b, v8.16b, v2.16b		)
+ST4(	eor		v9.16b, v9.16b, v3.16b		)
+
+ST5(	eor		v5.16b, v5.16b, v0.16b		)
+ST5(	eor		v6.16b, v6.16b, v1.16b		)
+ST5(	tbl		v4.16b, {v4.16b}, v10.16b	)
+ST5(	eor		v7.16b, v7.16b, v2.16b		)
+ST5(	eor		v8.16b, v8.16b, v3.16b		)
+ST5(	eor		v9.16b, v9.16b, v4.16b		)
+
+ST5(	st1		{v5.16b}, [x0], x14		)
+	st1		{v6.16b}, [x0], x15
+	st1		{v7.16b}, [x0], x16
+	add		x13, x13, x0
+	st1		{v9.16b}, [x13]		// overlapping stores
+	st1		{v8.16b}, [x0]
 	b		.Lctrout
 
-.Lctrcarry:
-	umov		x7, vctr.d[0]		/* load upper word of ctr  */
-	rev		x7, x7			/* ... to handle the carry */
-	add		x7, x7, #1
-	rev		x7, x7
-	ins		vctr.d[0], x7
-	b		.Lctrcarrydone
+.Lctrtail1x:
+	csel		x0, x0, x6, eq		// use finalbuf if less than a full block
+	ld1		{v5.16b}, [x1]
+ST5(	mov		v3.16b, v4.16b			)
+	encrypt_block	v3, w3, x2, x8, w7
+	eor		v5.16b, v5.16b, v3.16b
+	st1		{v5.16b}, [x0]
+	b		.Lctrout
 AES_FUNC_END(aes_ctr_encrypt)
 
 
@@ -619,61 +678,47 @@ AES_FUNC_END(aes_xts_decrypt)
 	 *		  int blocks, u8 dg[], int enc_before, int enc_after)
 	 */
 AES_FUNC_START(aes_mac_update)
-	frame_push	6
-
-	mov		x19, x0
-	mov		x20, x1
-	mov		x21, x2
-	mov		x22, x3
-	mov		x23, x4
-	mov		x24, x6
-
-	ld1		{v0.16b}, [x23]			/* get dg */
+	ld1		{v0.16b}, [x4]			/* get dg */
 	enc_prepare	w2, x1, x7
 	cbz		w5, .Lmacloop4x
 
 	encrypt_block	v0, w2, x1, x7, w8
 
 .Lmacloop4x:
-	subs		w22, w22, #4
+	subs		w3, w3, #4
 	bmi		.Lmac1x
-	ld1		{v1.16b-v4.16b}, [x19], #64	/* get next pt block */
+	ld1		{v1.16b-v4.16b}, [x0], #64	/* get next pt block */
 	eor		v0.16b, v0.16b, v1.16b		/* ..and xor with dg */
-	encrypt_block	v0, w21, x20, x7, w8
+	encrypt_block	v0, w2, x1, x7, w8
 	eor		v0.16b, v0.16b, v2.16b
-	encrypt_block	v0, w21, x20, x7, w8
+	encrypt_block	v0, w2, x1, x7, w8
 	eor		v0.16b, v0.16b, v3.16b
-	encrypt_block	v0, w21, x20, x7, w8
+	encrypt_block	v0, w2, x1, x7, w8
 	eor		v0.16b, v0.16b, v4.16b
-	cmp		w22, wzr
-	csinv		x5, x24, xzr, eq
+	cmp		w3, wzr
+	csinv		x5, x6, xzr, eq
 	cbz		w5, .Lmacout
-	encrypt_block	v0, w21, x20, x7, w8
-	st1		{v0.16b}, [x23]			/* return dg */
-	cond_yield_neon	.Lmacrestart
+	encrypt_block	v0, w2, x1, x7, w8
+	st1		{v0.16b}, [x4]			/* return dg */
+	cond_yield	.Lmacout, x7
 	b		.Lmacloop4x
 .Lmac1x:
-	add		w22, w22, #4
+	add		w3, w3, #4
 .Lmacloop:
-	cbz		w22, .Lmacout
-	ld1		{v1.16b}, [x19], #16		/* get next pt block */
+	cbz		w3, .Lmacout
+	ld1		{v1.16b}, [x0], #16		/* get next pt block */
 	eor		v0.16b, v0.16b, v1.16b		/* ..and xor with dg */
 
-	subs		w22, w22, #1
-	csinv		x5, x24, xzr, eq
+	subs		w3, w3, #1
+	csinv		x5, x6, xzr, eq
 	cbz		w5, .Lmacout
 
 .Lmacenc:
-	encrypt_block	v0, w21, x20, x7, w8
+	encrypt_block	v0, w2, x1, x7, w8
 	b		.Lmacloop
 
 .Lmacout:
-	st1		{v0.16b}, [x23]			/* return dg */
-	frame_pop
+	st1		{v0.16b}, [x4]			/* return dg */
+	mov		w0, w3
 	ret
-
-.Lmacrestart:
-	ld1		{v0.16b}, [x23]			/* get dg */
-	enc_prepare	w21, x20, x0
-	b		.Lmacloop4x
 AES_FUNC_END(aes_mac_update)

arch/arm64/crypto/aes-neonbs-core.S

@@ -613,7 +613,6 @@ SYM_FUNC_END(aesbs_decrypt8)
 	st1		{\o7\().16b}, [x19], #16
 
 	cbz		x23, 1f
-	cond_yield_neon
 	b		99b
 
 1:	frame_pop
@@ -715,7 +714,6 @@ SYM_FUNC_START(aesbs_cbc_decrypt)
 1:	st1		{v24.16b}, [x24]		// store IV
 
 	cbz		x23, 2f
-	cond_yield_neon
 	b		99b
 
 2:	frame_pop
@@ -801,7 +799,7 @@ SYM_FUNC_END(__xts_crypt8)
 	mov		x23, x4
 	mov		x24, x5
 
-0:	movi		v30.2s, #0x1
+	movi		v30.2s, #0x1
 	movi		v25.2s, #0x87
 	uzp1		v30.4s, v30.4s, v25.4s
 	ld1		{v25.16b}, [x24]
@@ -846,7 +844,6 @@ SYM_FUNC_END(__xts_crypt8)
 	cbz		x23, 1f
 	st1		{v25.16b}, [x24]
 
-	cond_yield_neon	0b
 	b		99b
 
 1:	st1		{v25.16b}, [x24]
@@ -889,7 +886,7 @@ SYM_FUNC_START(aesbs_ctr_encrypt)
 	cset		x26, ne
 	add		x23, x23, x26		// do one extra block if final
 
-98:	ldp		x7, x8, [x24]
+	ldp		x7, x8, [x24]
 	ld1		{v0.16b}, [x24]
 CPU_LE(	rev		x7, x7		)
 CPU_LE(	rev		x8, x8		)
@@ -967,7 +964,6 @@ CPU_LE(	rev		x8, x8		)
 	st1		{v0.16b}, [x24]
 	cbz		x23, .Lctr_done
 
-	cond_yield_neon	98b
 	b		99b
 
 .Lctr_done:

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

@@ -68,10 +68,10 @@
 	.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
 
@@ -257,12 +257,6 @@ CPU_LE(	ext		v12.16b, v12.16b, v12.16b, #8	)
 	.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.
@@ -317,26 +311,7 @@ CPU_LE(	ext		v7.16b, v7.16b, v7.16b, #8	)
 	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.
 
@@ -453,7 +428,9 @@ CPU_LE(	ext		v0.16b, v0.16b, v0.16b, #8	)
 	// 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_\@:
@@ -489,7 +466,9 @@ CPU_LE(	ext		v7.16b, v7.16b, v7.16b, #8	)
 // 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

@@ -37,9 +37,18 @@ static int crct10dif_update_pmull_p8(struct shash_desc *desc, const u8 *data,
 	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);
 	}
@@ -53,9 +62,18 @@ static int crct10dif_update_pmull_p64(struct shash_desc *desc, const u8 *data,
 	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);
 	}

arch/arm64/crypto/sha1-ce-core.S

@@ -62,40 +62,34 @@
 	.endm
 
 	/*
-	 * void sha1_ce_transform(struct sha1_ce_state *sst, u8 const *src,
-	 *			  int blocks)
+	 * int sha1_ce_transform(struct sha1_ce_state *sst, u8 const *src,
+	 *			 int blocks)
 	 */
 SYM_FUNC_START(sha1_ce_transform)
-	frame_push	3
-
-	mov		x19, x0
-	mov		x20, x1
-	mov		x21, x2
-
 	/* load round constants */
-0:	loadrc		k0.4s, 0x5a827999, w6
+	loadrc		k0.4s, 0x5a827999, w6
 	loadrc		k1.4s, 0x6ed9eba1, w6
 	loadrc		k2.4s, 0x8f1bbcdc, w6
 	loadrc		k3.4s, 0xca62c1d6, w6
 
 	/* load state */
-	ld1		{dgav.4s}, [x19]
-	ldr		dgb, [x19, #16]
+	ld1		{dgav.4s}, [x0]
+	ldr		dgb, [x0, #16]
 
 	/* load sha1_ce_state::finalize */
 	ldr_l		w4, sha1_ce_offsetof_finalize, x4
-	ldr		w4, [x19, x4]
+	ldr		w4, [x0, x4]
 
 	/* load input */
-1:	ld1		{v8.4s-v11.4s}, [x20], #64
-	sub		w21, w21, #1
+0:	ld1		{v8.4s-v11.4s}, [x1], #64
+	sub		w2, w2, #1
 
 CPU_LE(	rev32		v8.16b, v8.16b		)
 CPU_LE(	rev32		v9.16b, v9.16b		)
 CPU_LE(	rev32		v10.16b, v10.16b	)
 CPU_LE(	rev32		v11.16b, v11.16b	)
 
-2:	add		t0.4s, v8.4s, k0.4s
+1:	add		t0.4s, v8.4s, k0.4s
 	mov		dg0v.16b, dgav.16b
 
 	add_update	c, ev, k0,  8,  9, 10, 11, dgb
@@ -126,25 +120,18 @@ CPU_LE(	rev32		v11.16b, v11.16b	)
 	add		dgbv.2s, dgbv.2s, dg1v.2s
 	add		dgav.4s, dgav.4s, dg0v.4s
 
-	cbz		w21, 3f
-
-	if_will_cond_yield_neon
-	st1		{dgav.4s}, [x19]
-	str		dgb, [x19, #16]
-	do_cond_yield_neon
+	cbz		w2, 2f
+	cond_yield	3f, x5
 	b		0b
-	endif_yield_neon
-
-	b		1b
 
 	/*
 	 * Final block: add padding and total bit count.
 	 * Skip if the input size was not a round multiple of the block size,
 	 * the padding is handled by the C code in that case.
 	 */
-3:	cbz		x4, 4f
+2:	cbz		x4, 3f
 	ldr_l		w4, sha1_ce_offsetof_count, x4
-	ldr		x4, [x19, x4]
+	ldr		x4, [x0, x4]
 	movi		v9.2d, #0
 	mov		x8, #0x80000000
 	movi		v10.2d, #0
@@ -153,11 +140,11 @@ CPU_LE(	rev32		v11.16b, v11.16b	)
 	mov		x4, #0
 	mov		v11.d[0], xzr
 	mov		v11.d[1], x7
-	b		2b
+	b		1b
 
 	/* store new state */
-4:	st1		{dgav.4s}, [x19]
-	str		dgb, [x19, #16]
-	frame_pop
+3:	st1		{dgav.4s}, [x0]
+	str		dgb, [x0, #16]
+	mov		w0, w2
 	ret
 SYM_FUNC_END(sha1_ce_transform)