lib/crypto: arm64/sha512-ce: Drop compatibility macros for older binutils
Subsystems (2)
Eric Biggers
maintainerPATCH 1/1 lib/crypto: arm64/sha512-ce: Drop compatibility macros for older binutils
Now that the oldest supported binutils version is 2.30, the macros that emit the SHA-512 instructions as '.inst' words are no longer needed. So drop them. No change in the generated machine code.
Changed from the original patch by Ard Biesheuvel:
(https://lore.kernel.org/r/[email protected]):
- Reduced scope to just SHA-512
- Added comment that explains why "sha3" is used instead of "sha2"
---
This patch is targeting libcrypto-next
lib/crypto/arm64/sha512-ce-core.S | 27 +++++++-------------------- 1 file changed, 7 insertions(+), 20 deletions(-)
lib/crypto/arm64/sha512-ce-core.S +7 -20
@@ -10,30 +10,17 @@ */ #include <linux/linkage.h> #include <asm/assembler.h> - .irp b,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 - .set .Lq\b, \b - .set .Lv\b\().2d, \b - .endr - - .macro sha512h, rd, rn, rm - .inst 0xce608000 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512h2, rd, rn, rm - .inst 0xce608400 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512su0, rd, rn - .inst 0xcec08000 | .L\rd | (.L\rn << 5) - .endm - - .macro sha512su1, rd, rn, rm - .inst 0xce608800 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm + /* + * While SHA-512 is part of the SHA-2 family of algorithms, the + * corresponding arm64 instructions are actually part of the "sha3" CPU + * feature. (Except in binutils 2.30 through 2.42, which used "sha2". + * But "sha3" implies "sha2", so "sha3" still works in those versions.) + */ + .arch armv8-a+sha3 /* * The SHA-512 round constants */ .section ".rodata", "a"
base-commit: 66be847cc4c2e82fb50190b52b05b3bb0ef57999
Signature · 1 line
2.50.1
Quoted 11 lines · Eric Biggers
Now that the oldest supported binutils version is 2.30, the macros that emit the SHA-512 instructions as '.inst' words are no longer needed. So drop them. No change in the generated machine code.
Changed from the original patch by Ard Biesheuvel:
(https://lore.kernel.org/r/[email protected]):
- Reduced scope to just SHA-512
- Added comment that explains why "sha3" is used instead of "sha2"Signed-off-by Eric Biggers <[email protected]>
Nit below
Quoted 31 lines
---
This patch is targeting libcrypto-next
lib/crypto/arm64/sha512-ce-core.S | 27 +++++++-------------------- 1 file changed, 7 insertions(+), 20 deletions(-)lib/crypto/arm64/sha512-ce-core.S +0
-14@@ -10,30 +10,17 @@ */ #include <linux/linkage.h> #include <asm/assembler.h> - .irp b,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 - .set .Lq\b, \b - .set .Lv\b\().2d, \b - .endr - - .macro sha512h, rd, rn, rm - .inst 0xce608000 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512h2, rd, rn, rm - .inst 0xce608400 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512su0, rd, rn
lib/crypto/arm64/sha512-ce-core.S line 18
Quoted 10 lines
lib/crypto/arm64/sha512-ce-core.S +4
-6- .inst 0xcec08000 | .L\rd | (.L\rn << 5) - .endm - - .macro sha512su1, rd, rn, rm - .inst 0xce608800 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm + /* + * While SHA-512 is part of the SHA-2 family of algorithms, the + * corresponding arm64 instructions are actually part of the "sha3" CPU + * feature. (Except in binutils 2.30 through 2.42, which used "sha2".
Nit: the ARM ARM describes these features as FEAT_SHA256, FEAT_SHA512 and FEAT_SHA3, and the latter two happen to have appeared in the same architecture revision. So this is likely just the GCC/binutils devs getting confused, and assuming a) that SHA-3 implies SHA-2 (which is silly if you know the difference) and b) SHA512 has anything to do with SHA-3.
Quoted 13 lines
lib/crypto/arm64/sha512-ce-core.S +3
-0+ * But "sha3" implies "sha2", so "sha3" still works in those versions.) + */ + .arch armv8-a+sha3 /* * The SHA-512 round constants */ .section ".rodata", "a"base-commit:
66be847cc4c2e82fb50190b52b05b3bb0ef57999--
2.50.1
Eric Biggers
maintainerQuoted 57 lines · Ard Biesheuvel
Quoted 11 lines · Eric Biggers
Now that the oldest supported binutils version is 2.30, the macros that emit the SHA-512 instructions as '.inst' words are no longer needed. So drop them. No change in the generated machine code.
Changed from the original patch by Ard Biesheuvel:
(https://lore.kernel.org/r/[email protected]):
- Reduced scope to just SHA-512
- Added comment that explains why "sha3" is used instead of "sha2"Signed-off-by Eric Biggers <[email protected]>Acked-by Ard Biesheuvel <[email protected]>Nit below
Quoted 40 lines
---
This patch is targeting libcrypto-next
lib/crypto/arm64/sha512-ce-core.S | 27 +++++++-------------------- 1 file changed, 7 insertions(+), 20 deletions(-)lib/crypto/arm64/sha512-ce-core.S +3
-20@@ -10,30 +10,17 @@ */ #include <linux/linkage.h> #include <asm/assembler.h> - .irp b,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 - .set .Lq\b, \b - .set .Lv\b\().2d, \b - .endr - - .macro sha512h, rd, rn, rm - .inst 0xce608000 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512h2, rd, rn, rm - .inst 0xce608400 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512su0, rd, rn - .inst 0xcec08000 | .L\rd | (.L\rn << 5) - .endm - - .macro sha512su1, rd, rn, rm - .inst 0xce608800 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm + /* + * While SHA-512 is part of the SHA-2 family of algorithms, the + * corresponding arm64 instructions are actually part of the "sha3" CPU
Quoted 8 lines
lib/crypto/arm64/sha512-ce-core.S line 18
Quoted 1 line
lib/crypto/arm64/sha512-ce-core.S +1
-0+ * feature. (Except in binutils 2.30 through 2.42, which used "sha2".Nit: the ARM ARM describes these features as
FEAT_SHA256,FEAT_SHA512andFEAT_SHA3, and the latter two happen to have appeared in the same architecture revision. So this is likely just the GCC/binutils devs getting confused, and assuming a) that SHA-3 implies SHA-2 (which is silly if you know the difference) and b) SHA512 has anything to do with SHA-3.
How does the following look?
/*
* We have to specify the "sha3" feature here, since the GNU and clang
* assemblers both consider the SHA-512 instructions to be part of the
* "sha3" feature. (Except binutils 2.30 through 2.42, which used
* "sha2". But "sha3" implies "sha2", so "sha3" still works in those
* versions.) "sha3" doesn't make a lot of sense, since SHA-512 is part
* of the SHA-2 family of algorithms, and also the Arm Architecture
* Reference Manual defines FEAT_SHA512 and FEAT_SHA3 separately.
* Regardless, we must use "sha3" to be compatible with the assemblers. */
By the way, the ARM ARM does actually have the following:
If FEAT_SHA256 is implemented, then FEAT_SHA1 is implemented.
If FEAT_SHA512 is implemented, then FEAT_SHA256 and FEAT_SHA1 are implemented. If FEAT_SHA3 is implemented, then FEAT_SHA256 and FEAT_SHA1 are implemented.
So some of the SHAs do imply other ones. But notably absent is FEAT_SHA3 implying FEAT_SHA512...
- Eric
Ard Biesheuvel
maintainerQuoted 70 lines · Eric Biggers
Quoted 57 lines · Ard Biesheuvel
Quoted 11 lines · Eric Biggers
Now that the oldest supported binutils version is 2.30, the macros that emit the SHA-512 instructions as '.inst' words are no longer needed. So drop them. No change in the generated machine code.
Changed from the original patch by Ard Biesheuvel:
(https://lore.kernel.org/r/[email protected]):
- Reduced scope to just SHA-512
- Added comment that explains why "sha3" is used instead of "sha2"Signed-off-by Eric Biggers <[email protected]>Acked-by Ard Biesheuvel <[email protected]>Nit below
Quoted 40 lines
---
This patch is targeting libcrypto-next
lib/crypto/arm64/sha512-ce-core.S | 27 +++++++-------------------- 1 file changed, 7 insertions(+), 20 deletions(-)lib/crypto/arm64/sha512-ce-core.S +3
-20@@ -10,30 +10,17 @@ */ #include <linux/linkage.h> #include <asm/assembler.h> - .irp b,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 - .set .Lq\b, \b - .set .Lv\b\().2d, \b - .endr - - .macro sha512h, rd, rn, rm - .inst 0xce608000 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512h2, rd, rn, rm - .inst 0xce608400 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm - - .macro sha512su0, rd, rn - .inst 0xcec08000 | .L\rd | (.L\rn << 5) - .endm - - .macro sha512su1, rd, rn, rm - .inst 0xce608800 | .L\rd | (.L\rn << 5) | (.L\rm << 16) - .endm + /* + * While SHA-512 is part of the SHA-2 family of algorithms, the + * corresponding arm64 instructions are actually part of the "sha3" CPUQuoted 8 lines
lib/crypto/arm64/sha512-ce-core.S line 18
Quoted 1 line
lib/crypto/arm64/sha512-ce-core.S +1
-0+ * feature. (Except in binutils 2.30 through 2.42, which used "sha2".Nit: the ARM ARM describes these features as
FEAT_SHA256,FEAT_SHA512andFEAT_SHA3, and the latter two happen to have appeared in the same architecture revision. So this is likely just the GCC/binutils devs getting confused, and assuming a) that SHA-3 implies SHA-2 (which is silly if you know the difference) and b) SHA512 has anything to do with SHA-3.How does the following look?
Quoted 11 lines
/*
* We have to specify the "sha3" feature here, since the GNU and clang
* assemblers both consider the SHA-512 instructions to be part of the
* "sha3" feature. (Except binutils 2.30 through 2.42, which used
* "sha2". But "sha3" implies "sha2", so "sha3" still works in those
* versions.) "sha3" doesn't make a lot of sense, since SHA-512 is part
* of the SHA-2 family of algorithms, and also the Arm Architecture
* Reference Manual definesFEAT_SHA512andFEAT_SHA3separately.
* Regardless, we must use "sha3" to be compatible with the assemblers. */
LGTM
Quoted 8 lines
By the way, the ARM ARM does actually have the following:
If
FEAT_SHA256is implemented, thenFEAT_SHA1is implemented.
IfFEAT_SHA512is implemented, thenFEAT_SHA256andFEAT_SHA1are implemented. IfFEAT_SHA3is implemented, thenFEAT_SHA256andFEAT_SHA1are implemented.So some of the SHAs do imply other ones. But notably absent is
FEAT_SHA3implyingFEAT_SHA512...
Yeah, and such policies usually evaporate into thin air as soon as Apple decides to implement something different, so they tend to be rather meaningless. (E.g, FEAT_SME used to imply FEAT_SVE but it no longer does)