summaryrefslogtreecommitdiff
path: root/lib/crypto/arm
diff options
context:
space:
mode:
Diffstat (limited to 'lib/crypto/arm')
-rw-r--r--lib/crypto/arm/blake2b-neon-core.S350
-rw-r--r--lib/crypto/arm/blake2b.h40
-rw-r--r--lib/crypto/arm/blake2s-core.S22
-rw-r--r--lib/crypto/arm/blake2s.h4
-rw-r--r--lib/crypto/arm/chacha.h11
-rw-r--r--lib/crypto/arm/curve25519.h5
-rw-r--r--lib/crypto/arm/poly1305.h6
-rw-r--r--lib/crypto/arm/sha1-armv7-neon.S2
-rw-r--r--lib/crypto/arm/sha1-ce-core.S2
-rw-r--r--lib/crypto/arm/sha1.h13
-rw-r--r--lib/crypto/arm/sha256-ce.S2
-rw-r--r--lib/crypto/arm/sha256.h12
-rw-r--r--lib/crypto/arm/sha512.h5
13 files changed, 428 insertions, 46 deletions
diff --git a/lib/crypto/arm/blake2b-neon-core.S b/lib/crypto/arm/blake2b-neon-core.S
new file mode 100644
index 000000000000..b55c37f0b88f
--- /dev/null
+++ b/lib/crypto/arm/blake2b-neon-core.S
@@ -0,0 +1,350 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * 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 the SHA-2 family
+ * and slightly faster than SHA-1.
+ *
+ * Copyright 2020 Google LLC
+ *
+ * Author: Eric Biggers <ebiggers@google.com>
+ */
+
+#include <linux/linkage.h>
+
+ .text
+ .fpu neon
+
+ // The arguments to blake2b_compress_neon()
+ CTX .req r0
+ DATA .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_ctx *ctx,
+// const u8 *data, size_t nblocks, u32 inc);
+//
+// Only the first three fields of struct blake2b_ctx 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, CTX
+ 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}, [DATA]!
+ veor q6, q6, q14 // v[12..13] = IV[4..5] ^ t[0..1]
+ vld1.8 {q10-q11}, [DATA]!
+ veor q7, q7, q15 // v[14..15] = IV[6..7] ^ f[0..1]
+ vld1.8 {q12-q13}, [DATA]!
+ vst1.8 {q8-q9}, [sp, :256]
+ mov ip, CTX
+ vld1.8 {q14-q15}, [DATA]!
+
+ // 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, CTX
+ 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)
diff --git a/lib/crypto/arm/blake2b.h b/lib/crypto/arm/blake2b.h
new file mode 100644
index 000000000000..5c76498521e6
--- /dev/null
+++ b/lib/crypto/arm/blake2b.h
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * BLAKE2b digest algorithm, NEON accelerated
+ *
+ * Copyright 2020 Google LLC
+ */
+
+#include <asm/neon.h>
+#include <asm/simd.h>
+
+static __ro_after_init DEFINE_STATIC_KEY_FALSE(have_neon);
+
+asmlinkage void blake2b_compress_neon(struct blake2b_ctx *ctx,
+ const u8 *data, size_t nblocks, u32 inc);
+
+static void blake2b_compress(struct blake2b_ctx *ctx,
+ const u8 *data, size_t nblocks, u32 inc)
+{
+ if (!static_branch_likely(&have_neon) || !may_use_simd()) {
+ blake2b_compress_generic(ctx, data, nblocks, inc);
+ return;
+ }
+ do {
+ const size_t blocks = min_t(size_t, nblocks,
+ SZ_4K / BLAKE2B_BLOCK_SIZE);
+
+ scoped_ksimd()
+ blake2b_compress_neon(ctx, data, blocks, inc);
+
+ data += blocks * BLAKE2B_BLOCK_SIZE;
+ nblocks -= blocks;
+ } while (nblocks);
+}
+
+#define blake2b_mod_init_arch blake2b_mod_init_arch
+static void blake2b_mod_init_arch(void)
+{
+ if (elf_hwcap & HWCAP_NEON)
+ static_branch_enable(&have_neon);
+}
diff --git a/lib/crypto/arm/blake2s-core.S b/lib/crypto/arm/blake2s-core.S
index 293f44fa8f31..933f0558b7cd 100644
--- a/lib/crypto/arm/blake2s-core.S
+++ b/lib/crypto/arm/blake2s-core.S
@@ -115,7 +115,7 @@
// 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
+// spilling v[8..9], then to v[10..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.
//
@@ -170,10 +170,10 @@
.endm
//
-// void blake2s_compress(struct blake2s_state *state,
-// const u8 *block, size_t nblocks, u32 inc);
+// void blake2s_compress(struct blake2s_ctx *ctx,
+// const u8 *data, size_t nblocks, u32 inc);
//
-// Only the first three fields of struct blake2s_state are used:
+// Only the first three fields of struct blake2s_ctx are used:
// u32 h[8]; (inout)
// u32 t[2]; (inout)
// u32 f[2]; (in)
@@ -183,8 +183,8 @@ ENTRY(blake2s_compress)
push {r0-r2,r4-r11,lr} // keep this an even number
.Lnext_block:
- // r0 is 'state'
- // r1 is 'block'
+ // r0 is 'ctx'
+ // r1 is 'data'
// r3 is 'inc'
// Load and increment the counter t[0..1].
@@ -209,18 +209,18 @@ ENTRY(blake2s_compress)
.Lcopy_block_done:
str r1, [sp, #68] // Update message pointer
- // Calculate v[8..15]. Push v[9..15] onto the stack, and leave space
+ // Calculate v[8..15]. Push v[10..15] onto the stack, and leave space
// for spilling v[8..9]. Leave v[8..9] in r8-r9.
- mov r14, r0 // r14 = state
+ mov r14, r0 // r14 = ctx
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]
+ ldm r12, {r2-r7} // load IV[2..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]
+ push {r2-r7} // push v[10..15]
sub sp, sp, #8 // leave space for v[8..9]
// Load h[0..7] == v[0..7].
@@ -275,7 +275,7 @@ ENTRY(blake2s_compress)
// 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)
+ ldm sp, {r0, r1, r2} // load (ctx, data, nblocks)
mov r3, #64 // set 'inc'
subs r2, r2, #1 // nblocks--
str r2, [sp, #8]
diff --git a/lib/crypto/arm/blake2s.h b/lib/crypto/arm/blake2s.h
index aa7a97139ea7..42c04440c191 100644
--- a/lib/crypto/arm/blake2s.h
+++ b/lib/crypto/arm/blake2s.h
@@ -1,5 +1,5 @@
/* SPDX-License-Identifier: GPL-2.0-or-later */
/* defined in blake2s-core.S */
-void blake2s_compress(struct blake2s_state *state, const u8 *block,
- size_t nblocks, u32 inc);
+void blake2s_compress(struct blake2s_ctx *ctx,
+ const u8 *data, size_t nblocks, u32 inc);
diff --git a/lib/crypto/arm/chacha.h b/lib/crypto/arm/chacha.h
index 0cae30f8ee5d..836e49088e98 100644
--- a/lib/crypto/arm/chacha.h
+++ b/lib/crypto/arm/chacha.h
@@ -12,7 +12,6 @@
#include <asm/cputype.h>
#include <asm/hwcap.h>
-#include <asm/neon.h>
#include <asm/simd.h>
asmlinkage void chacha_block_xor_neon(const struct chacha_state *state,
@@ -68,9 +67,8 @@ static void hchacha_block_arch(const struct chacha_state *state,
if (!IS_ENABLED(CONFIG_KERNEL_MODE_NEON) || !neon_usable()) {
hchacha_block_arm(state, out, nrounds);
} else {
- kernel_neon_begin();
- hchacha_block_neon(state, out, nrounds);
- kernel_neon_end();
+ scoped_ksimd()
+ hchacha_block_neon(state, out, nrounds);
}
}
@@ -87,9 +85,8 @@ static void chacha_crypt_arch(struct chacha_state *state, u8 *dst,
do {
unsigned int todo = min_t(unsigned int, bytes, SZ_4K);
- kernel_neon_begin();
- chacha_doneon(state, dst, src, todo, nrounds);
- kernel_neon_end();
+ scoped_ksimd()
+ chacha_doneon(state, dst, src, todo, nrounds);
bytes -= todo;
src += todo;
diff --git a/lib/crypto/arm/curve25519.h b/lib/crypto/arm/curve25519.h
index f6d66494eb8f..b1a566885e95 100644
--- a/lib/crypto/arm/curve25519.h
+++ b/lib/crypto/arm/curve25519.h
@@ -25,9 +25,8 @@ static void curve25519_arch(u8 out[CURVE25519_KEY_SIZE],
const u8 point[CURVE25519_KEY_SIZE])
{
if (static_branch_likely(&have_neon) && crypto_simd_usable()) {
- kernel_neon_begin();
- curve25519_neon(out, scalar, point);
- kernel_neon_end();
+ scoped_ksimd()
+ curve25519_neon(out, scalar, point);
} else {
curve25519_generic(out, scalar, point);
}
diff --git a/lib/crypto/arm/poly1305.h b/lib/crypto/arm/poly1305.h
index 0021cf368307..0fe903d8de55 100644
--- a/lib/crypto/arm/poly1305.h
+++ b/lib/crypto/arm/poly1305.h
@@ -6,7 +6,6 @@
*/
#include <asm/hwcap.h>
-#include <asm/neon.h>
#include <asm/simd.h>
#include <linux/cpufeature.h>
#include <linux/jump_label.h>
@@ -32,9 +31,8 @@ static void poly1305_blocks(struct poly1305_block_state *state, const u8 *src,
do {
unsigned int todo = min_t(unsigned int, len, SZ_4K);
- kernel_neon_begin();
- poly1305_blocks_neon(state, src, todo, padbit);
- kernel_neon_end();
+ scoped_ksimd()
+ poly1305_blocks_neon(state, src, todo, padbit);
len -= todo;
src += todo;
diff --git a/lib/crypto/arm/sha1-armv7-neon.S b/lib/crypto/arm/sha1-armv7-neon.S
index 6edba3ab62e8..a0323fa5c58a 100644
--- a/lib/crypto/arm/sha1-armv7-neon.S
+++ b/lib/crypto/arm/sha1-armv7-neon.S
@@ -1,5 +1,5 @@
/* SPDX-License-Identifier: GPL-2.0-or-later */
-/* sha1-armv7-neon.S - ARM/NEON accelerated SHA-1 transform function
+/* ARM/NEON accelerated SHA-1 transform function
*
* Copyright © 2013-2014 Jussi Kivilinna <jussi.kivilinna@iki.fi>
*/
diff --git a/lib/crypto/arm/sha1-ce-core.S b/lib/crypto/arm/sha1-ce-core.S
index 2de40dd25e47..7d6b2631ca8d 100644
--- a/lib/crypto/arm/sha1-ce-core.S
+++ b/lib/crypto/arm/sha1-ce-core.S
@@ -1,6 +1,6 @@
/* SPDX-License-Identifier: GPL-2.0-only */
/*
- * sha1-ce-core.S - SHA-1 secure hash using ARMv8 Crypto Extensions
+ * SHA-1 secure hash using ARMv8 Crypto Extensions
*
* Copyright (C) 2015 Linaro Ltd.
* Author: Ard Biesheuvel <ard.biesheuvel@linaro.org>
diff --git a/lib/crypto/arm/sha1.h b/lib/crypto/arm/sha1.h
index 29f8bcad0447..3e2d8c7cab9f 100644
--- a/lib/crypto/arm/sha1.h
+++ b/lib/crypto/arm/sha1.h
@@ -4,7 +4,6 @@
*
* Copyright 2025 Google LLC
*/
-#include <asm/neon.h>
#include <asm/simd.h>
static __ro_after_init DEFINE_STATIC_KEY_FALSE(have_neon);
@@ -22,12 +21,12 @@ static void sha1_blocks(struct sha1_block_state *state,
{
if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) &&
static_branch_likely(&have_neon) && likely(may_use_simd())) {
- kernel_neon_begin();
- if (static_branch_likely(&have_ce))
- sha1_ce_transform(state, data, nblocks);
- else
- sha1_transform_neon(state, data, nblocks);
- kernel_neon_end();
+ scoped_ksimd() {
+ if (static_branch_likely(&have_ce))
+ sha1_ce_transform(state, data, nblocks);
+ else
+ sha1_transform_neon(state, data, nblocks);
+ }
} else {
sha1_block_data_order(state, data, nblocks);
}
diff --git a/lib/crypto/arm/sha256-ce.S b/lib/crypto/arm/sha256-ce.S
index 7481ac8e6c0d..144ee805f64a 100644
--- a/lib/crypto/arm/sha256-ce.S
+++ b/lib/crypto/arm/sha256-ce.S
@@ -1,6 +1,6 @@
/* SPDX-License-Identifier: GPL-2.0-only */
/*
- * sha256-ce.S - SHA-224/256 secure hash using ARMv8 Crypto Extensions
+ * SHA-224/256 secure hash using ARMv8 Crypto Extensions
*
* Copyright (C) 2015 Linaro Ltd.
* Author: Ard Biesheuvel <ard.biesheuvel@linaro.org>
diff --git a/lib/crypto/arm/sha256.h b/lib/crypto/arm/sha256.h
index 7556457b3094..ae7e52dd6e3b 100644
--- a/lib/crypto/arm/sha256.h
+++ b/lib/crypto/arm/sha256.h
@@ -22,12 +22,12 @@ static void sha256_blocks(struct sha256_block_state *state,
{
if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) &&
static_branch_likely(&have_neon) && likely(may_use_simd())) {
- kernel_neon_begin();
- if (static_branch_likely(&have_ce))
- sha256_ce_transform(state, data, nblocks);
- else
- sha256_block_data_order_neon(state, data, nblocks);
- kernel_neon_end();
+ scoped_ksimd() {
+ if (static_branch_likely(&have_ce))
+ sha256_ce_transform(state, data, nblocks);
+ else
+ sha256_block_data_order_neon(state, data, nblocks);
+ }
} else {
sha256_block_data_order(state, data, nblocks);
}
diff --git a/lib/crypto/arm/sha512.h b/lib/crypto/arm/sha512.h
index d1b485dd275d..ed9bd81d6d78 100644
--- a/lib/crypto/arm/sha512.h
+++ b/lib/crypto/arm/sha512.h
@@ -19,9 +19,8 @@ static void sha512_blocks(struct sha512_block_state *state,
{
if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) &&
static_branch_likely(&have_neon) && likely(may_use_simd())) {
- kernel_neon_begin();
- sha512_block_data_order_neon(state, data, nblocks);
- kernel_neon_end();
+ scoped_ksimd()
+ sha512_block_data_order_neon(state, data, nblocks);
} else {
sha512_block_data_order(state, data, nblocks);
}
generated by cgit v1.2.3 (git 2.46.0) at 2025-12-03 16:31:42 +0000