| Message ID | 20201207172625.2888810-1-dann.frazier@canonical.com |
|---|---|
| State | New |
| Headers | show |
| Series | [4.4] Revert "crypto: arm64/sha - avoid non-standard inline asm tricks" | expand |
On Mon, 7 Dec 2020 at 18:26, dann frazier <dann.frazier@canonical.com> wrote: > > This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686. > > This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce > modules. This relocation type is not supported by the linux-4.4.y kernel > module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so > these modules now fail to load: > > [ 37.866250] module sha1_ce: unsupported RELA relocation: 275 > > This issue does not exist with the backport to 4.9+. Bisection shows that > this is due to those kernels also having a backport of > commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR") Hi Dann, Would it be an option to backport 41c066f as well? > > Signed-off-by: dann frazier <dann.frazier@canonical.com> > --- > arch/arm64/crypto/sha1-ce-core.S | 6 ++---- > arch/arm64/crypto/sha1-ce-glue.c | 11 ++++++++--- > arch/arm64/crypto/sha2-ce-core.S | 6 ++---- > arch/arm64/crypto/sha2-ce-glue.c | 13 ++++++++----- > 4 files changed, 20 insertions(+), 16 deletions(-) > > diff --git a/arch/arm64/crypto/sha1-ce-core.S b/arch/arm64/crypto/sha1-ce-core.S > index 8550408735a0..c98e7e849f06 100644 > --- a/arch/arm64/crypto/sha1-ce-core.S > +++ b/arch/arm64/crypto/sha1-ce-core.S > @@ -82,8 +82,7 @@ ENTRY(sha1_ce_transform) > ldr dgb, [x0, #16] > > /* load sha1_ce_state::finalize */ > - ldr_l w4, sha1_ce_offsetof_finalize, x4 > - ldr w4, [x0, x4] > + ldr w4, [x0, #:lo12:sha1_ce_offsetof_finalize] > > /* load input */ > 0: ld1 {v8.4s-v11.4s}, [x1], #64 > @@ -133,8 +132,7 @@ CPU_LE( rev32 v11.16b, v11.16b ) > * the padding is handled by the C code in that case. > */ > cbz x4, 3f > - ldr_l w4, sha1_ce_offsetof_count, x4 > - ldr x4, [x0, x4] > + ldr x4, [x0, #:lo12:sha1_ce_offsetof_count] > movi v9.2d, #0 > mov x8, #0x80000000 > movi v10.2d, #0 > diff --git a/arch/arm64/crypto/sha1-ce-glue.c b/arch/arm64/crypto/sha1-ce-glue.c > index 1b7b4684c35b..01e48b8970b1 100644 > --- a/arch/arm64/crypto/sha1-ce-glue.c > +++ b/arch/arm64/crypto/sha1-ce-glue.c > @@ -17,6 +17,9 @@ > #include <linux/crypto.h> > #include <linux/module.h> > > +#define ASM_EXPORT(sym, val) \ > + asm(".globl " #sym "; .set " #sym ", %0" :: "I"(val)); > + > MODULE_DESCRIPTION("SHA1 secure hash using ARMv8 Crypto Extensions"); > MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); > MODULE_LICENSE("GPL v2"); > @@ -29,9 +32,6 @@ struct sha1_ce_state { > asmlinkage void sha1_ce_transform(struct sha1_ce_state *sst, u8 const *src, > int blocks); > > -const u32 sha1_ce_offsetof_count = offsetof(struct sha1_ce_state, sst.count); > -const u32 sha1_ce_offsetof_finalize = offsetof(struct sha1_ce_state, finalize); > - > static int sha1_ce_update(struct shash_desc *desc, const u8 *data, > unsigned int len) > { > @@ -52,6 +52,11 @@ static int sha1_ce_finup(struct shash_desc *desc, const u8 *data, > struct sha1_ce_state *sctx = shash_desc_ctx(desc); > bool finalize = !sctx->sst.count && !(len % SHA1_BLOCK_SIZE) && len; > > + ASM_EXPORT(sha1_ce_offsetof_count, > + offsetof(struct sha1_ce_state, sst.count)); > + ASM_EXPORT(sha1_ce_offsetof_finalize, > + offsetof(struct sha1_ce_state, finalize)); > + > /* > * Allow the asm code to perform the finalization if there is no > * partial data and the input is a round multiple of the block size. > diff --git a/arch/arm64/crypto/sha2-ce-core.S b/arch/arm64/crypto/sha2-ce-core.S > index 679c6c002f4f..01cfee066837 100644 > --- a/arch/arm64/crypto/sha2-ce-core.S > +++ b/arch/arm64/crypto/sha2-ce-core.S > @@ -88,8 +88,7 @@ ENTRY(sha2_ce_transform) > ld1 {dgav.4s, dgbv.4s}, [x0] > > /* load sha256_ce_state::finalize */ > - ldr_l w4, sha256_ce_offsetof_finalize, x4 > - ldr w4, [x0, x4] > + ldr w4, [x0, #:lo12:sha256_ce_offsetof_finalize] > > /* load input */ > 0: ld1 {v16.4s-v19.4s}, [x1], #64 > @@ -137,8 +136,7 @@ CPU_LE( rev32 v19.16b, v19.16b ) > * the padding is handled by the C code in that case. > */ > cbz x4, 3f > - ldr_l w4, sha256_ce_offsetof_count, x4 > - ldr x4, [x0, x4] > + ldr x4, [x0, #:lo12:sha256_ce_offsetof_count] > movi v17.2d, #0 > mov x8, #0x80000000 > movi v18.2d, #0 > diff --git a/arch/arm64/crypto/sha2-ce-glue.c b/arch/arm64/crypto/sha2-ce-glue.c > index 356ca9397a86..7a7f95b94869 100644 > --- a/arch/arm64/crypto/sha2-ce-glue.c > +++ b/arch/arm64/crypto/sha2-ce-glue.c > @@ -17,6 +17,9 @@ > #include <linux/crypto.h> > #include <linux/module.h> > > +#define ASM_EXPORT(sym, val) \ > + asm(".globl " #sym "; .set " #sym ", %0" :: "I"(val)); > + > MODULE_DESCRIPTION("SHA-224/SHA-256 secure hash using ARMv8 Crypto Extensions"); > MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); > MODULE_LICENSE("GPL v2"); > @@ -29,11 +32,6 @@ struct sha256_ce_state { > asmlinkage void sha2_ce_transform(struct sha256_ce_state *sst, u8 const *src, > int blocks); > > -const u32 sha256_ce_offsetof_count = offsetof(struct sha256_ce_state, > - sst.count); > -const u32 sha256_ce_offsetof_finalize = offsetof(struct sha256_ce_state, > - finalize); > - > static int sha256_ce_update(struct shash_desc *desc, const u8 *data, > unsigned int len) > { > @@ -54,6 +52,11 @@ static int sha256_ce_finup(struct shash_desc *desc, const u8 *data, > struct sha256_ce_state *sctx = shash_desc_ctx(desc); > bool finalize = !sctx->sst.count && !(len % SHA256_BLOCK_SIZE) && len; > > + ASM_EXPORT(sha256_ce_offsetof_count, > + offsetof(struct sha256_ce_state, sst.count)); > + ASM_EXPORT(sha256_ce_offsetof_finalize, > + offsetof(struct sha256_ce_state, finalize)); > + > /* > * Allow the asm code to perform the finalization if there is no > * partial data and the input is a round multiple of the block size. > -- > 2.29.2 >
On Mon, 7 Dec 2020 at 19:08, dann frazier <dann.frazier@canonical.com> wrote: > > On Mon, Dec 7, 2020 at 10:50 AM Ard Biesheuvel <ardb@kernel.org> wrote: > > > > On Mon, 7 Dec 2020 at 18:26, dann frazier <dann.frazier@canonical.com> wrote: > > > > > > This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686. > > > > > > This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce > > > modules. This relocation type is not supported by the linux-4.4.y kernel > > > module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so > > > these modules now fail to load: > > > > > > [ 37.866250] module sha1_ce: unsupported RELA relocation: 275 > > > > > > This issue does not exist with the backport to 4.9+. Bisection shows that > > > this is due to those kernels also having a backport of > > > commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR") > > > > Hi Dann, > > > > Would it be an option to backport 41c066f as well? > > Hi Ard, > > That was attempted before, but caused a build failure which would > still happen today: > https://www.spinics.net/lists/stable/msg179709.html > Specifically, head.S still has a 3 argument usage of adr_l. I'm not > sure how to safely fix that up myself. > Given that the original reason for reverting the backport of 41c066f no longer holds (as there are other users of adr_l in v4.4 now), I think the best solution is to backport it again, but with the hunk below folded in. (This just replaces the macro invocation with its output when called with the 3 arguments in question, so the generated code is identical) --- a/arch/arm64/kernel/head.S +++ b/arch/arm64/kernel/head.S @@ -424,7 +424,8 @@ __mmap_switched: str xzr, [x6], #8 // Clear BSS b 1b 2: - adr_l sp, initial_sp, x4 + adrp x4, initial_sp + add sp, x4, :lo12:initial_sp str_l x21, __fdt_pointer, x5 // Save FDT pointer str_l x24, memstart_addr, x6 // Save PHYS_OFFSET mov x29, #0
On Mon, Dec 7, 2020 at 11:29 AM Ard Biesheuvel <ardb@kernel.org> wrote: > > On Mon, 7 Dec 2020 at 19:08, dann frazier <dann.frazier@canonical.com> wrote: > > > > On Mon, Dec 7, 2020 at 10:50 AM Ard Biesheuvel <ardb@kernel.org> wrote: > > > > > > On Mon, 7 Dec 2020 at 18:26, dann frazier <dann.frazier@canonical.com> wrote: > > > > > > > > This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686. > > > > > > > > This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce > > > > modules. This relocation type is not supported by the linux-4.4.y kernel > > > > module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so > > > > these modules now fail to load: > > > > > > > > [ 37.866250] module sha1_ce: unsupported RELA relocation: 275 > > > > > > > > This issue does not exist with the backport to 4.9+. Bisection shows that > > > > this is due to those kernels also having a backport of > > > > commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR") > > > > > > Hi Dann, > > > > > > Would it be an option to backport 41c066f as well? > > > > Hi Ard, > > > > That was attempted before, but caused a build failure which would > > still happen today: > > https://www.spinics.net/lists/stable/msg179709.html > > Specifically, head.S still has a 3 argument usage of adr_l. I'm not > > sure how to safely fix that up myself. > > > > Given that the original reason for reverting the backport of 41c066f > no longer holds (as there are other users of adr_l in v4.4 now), I > think the best solution is to backport it again, but with the hunk > below folded in. (This just replaces the macro invocation with its > output when called with the 3 arguments in question, so the generated > code is identical) > > --- a/arch/arm64/kernel/head.S > +++ b/arch/arm64/kernel/head.S > @@ -424,7 +424,8 @@ __mmap_switched: > str xzr, [x6], #8 // Clear BSS > b 1b > 2: > - adr_l sp, initial_sp, x4 > + adrp x4, initial_sp > + add sp, x4, :lo12:initial_sp > str_l x21, __fdt_pointer, x5 // Save FDT pointer > str_l x24, memstart_addr, x6 // Save PHYS_OFFSET > mov x29, #0 Thanks Ard - that works. I'll follow-up with a backport patch. -dann
On Mon, 7 Dec 2020 at 21:36, dann frazier <dann.frazier@canonical.com> wrote: > > On Mon, Dec 7, 2020 at 11:29 AM Ard Biesheuvel <ardb@kernel.org> wrote: > > > > On Mon, 7 Dec 2020 at 19:08, dann frazier <dann.frazier@canonical.com> wrote: > > > > > > On Mon, Dec 7, 2020 at 10:50 AM Ard Biesheuvel <ardb@kernel.org> wrote: > > > > > > > > On Mon, 7 Dec 2020 at 18:26, dann frazier <dann.frazier@canonical.com> wrote: > > > > > > > > > > This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686. > > > > > > > > > > This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce > > > > > modules. This relocation type is not supported by the linux-4.4.y kernel > > > > > module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so > > > > > these modules now fail to load: > > > > > > > > > > [ 37.866250] module sha1_ce: unsupported RELA relocation: 275 > > > > > > > > > > This issue does not exist with the backport to 4.9+. Bisection shows that > > > > > this is due to those kernels also having a backport of > > > > > commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR") > > > > > > > > Hi Dann, > > > > > > > > Would it be an option to backport 41c066f as well? > > > > > > Hi Ard, > > > > > > That was attempted before, but caused a build failure which would > > > still happen today: > > > https://www.spinics.net/lists/stable/msg179709.html > > > Specifically, head.S still has a 3 argument usage of adr_l. I'm not > > > sure how to safely fix that up myself. > > > > > > > Given that the original reason for reverting the backport of 41c066f > > no longer holds (as there are other users of adr_l in v4.4 now), I > > think the best solution is to backport it again, but with the hunk > > below folded in. (This just replaces the macro invocation with its > > output when called with the 3 arguments in question, so the generated > > code is identical) > > > > --- a/arch/arm64/kernel/head.S > > +++ b/arch/arm64/kernel/head.S > > @@ -424,7 +424,8 @@ __mmap_switched: > > str xzr, [x6], #8 // Clear BSS > > b 1b > > 2: > > - adr_l sp, initial_sp, x4 > > + adrp x4, initial_sp > > + add sp, x4, :lo12:initial_sp > > str_l x21, __fdt_pointer, x5 // Save FDT pointer > > str_l x24, memstart_addr, x6 // Save PHYS_OFFSET > > mov x29, #0 > > Thanks Ard - that works. I'll follow-up with a backport patch. > Excellent.
diff --git a/arch/arm64/crypto/sha1-ce-core.S b/arch/arm64/crypto/sha1-ce-core.S index 8550408735a0..c98e7e849f06 100644 --- a/arch/arm64/crypto/sha1-ce-core.S +++ b/arch/arm64/crypto/sha1-ce-core.S @@ -82,8 +82,7 @@ ENTRY(sha1_ce_transform) ldr dgb, [x0, #16] /* load sha1_ce_state::finalize */ - ldr_l w4, sha1_ce_offsetof_finalize, x4 - ldr w4, [x0, x4] + ldr w4, [x0, #:lo12:sha1_ce_offsetof_finalize] /* load input */ 0: ld1 {v8.4s-v11.4s}, [x1], #64 @@ -133,8 +132,7 @@ CPU_LE( rev32 v11.16b, v11.16b ) * the padding is handled by the C code in that case. */ cbz x4, 3f - ldr_l w4, sha1_ce_offsetof_count, x4 - ldr x4, [x0, x4] + ldr x4, [x0, #:lo12:sha1_ce_offsetof_count] movi v9.2d, #0 mov x8, #0x80000000 movi v10.2d, #0 diff --git a/arch/arm64/crypto/sha1-ce-glue.c b/arch/arm64/crypto/sha1-ce-glue.c index 1b7b4684c35b..01e48b8970b1 100644 --- a/arch/arm64/crypto/sha1-ce-glue.c +++ b/arch/arm64/crypto/sha1-ce-glue.c @@ -17,6 +17,9 @@ #include <linux/crypto.h> #include <linux/module.h> +#define ASM_EXPORT(sym, val) \ + asm(".globl " #sym "; .set " #sym ", %0" :: "I"(val)); + MODULE_DESCRIPTION("SHA1 secure hash using ARMv8 Crypto Extensions"); MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); MODULE_LICENSE("GPL v2"); @@ -29,9 +32,6 @@ struct sha1_ce_state { asmlinkage void sha1_ce_transform(struct sha1_ce_state *sst, u8 const *src, int blocks); -const u32 sha1_ce_offsetof_count = offsetof(struct sha1_ce_state, sst.count); -const u32 sha1_ce_offsetof_finalize = offsetof(struct sha1_ce_state, finalize); - static int sha1_ce_update(struct shash_desc *desc, const u8 *data, unsigned int len) { @@ -52,6 +52,11 @@ static int sha1_ce_finup(struct shash_desc *desc, const u8 *data, struct sha1_ce_state *sctx = shash_desc_ctx(desc); bool finalize = !sctx->sst.count && !(len % SHA1_BLOCK_SIZE) && len; + ASM_EXPORT(sha1_ce_offsetof_count, + offsetof(struct sha1_ce_state, sst.count)); + ASM_EXPORT(sha1_ce_offsetof_finalize, + offsetof(struct sha1_ce_state, finalize)); + /* * Allow the asm code to perform the finalization if there is no * partial data and the input is a round multiple of the block size. diff --git a/arch/arm64/crypto/sha2-ce-core.S b/arch/arm64/crypto/sha2-ce-core.S index 679c6c002f4f..01cfee066837 100644 --- a/arch/arm64/crypto/sha2-ce-core.S +++ b/arch/arm64/crypto/sha2-ce-core.S @@ -88,8 +88,7 @@ ENTRY(sha2_ce_transform) ld1 {dgav.4s, dgbv.4s}, [x0] /* load sha256_ce_state::finalize */ - ldr_l w4, sha256_ce_offsetof_finalize, x4 - ldr w4, [x0, x4] + ldr w4, [x0, #:lo12:sha256_ce_offsetof_finalize] /* load input */ 0: ld1 {v16.4s-v19.4s}, [x1], #64 @@ -137,8 +136,7 @@ CPU_LE( rev32 v19.16b, v19.16b ) * the padding is handled by the C code in that case. */ cbz x4, 3f - ldr_l w4, sha256_ce_offsetof_count, x4 - ldr x4, [x0, x4] + ldr x4, [x0, #:lo12:sha256_ce_offsetof_count] movi v17.2d, #0 mov x8, #0x80000000 movi v18.2d, #0 diff --git a/arch/arm64/crypto/sha2-ce-glue.c b/arch/arm64/crypto/sha2-ce-glue.c index 356ca9397a86..7a7f95b94869 100644 --- a/arch/arm64/crypto/sha2-ce-glue.c +++ b/arch/arm64/crypto/sha2-ce-glue.c @@ -17,6 +17,9 @@ #include <linux/crypto.h> #include <linux/module.h> +#define ASM_EXPORT(sym, val) \ + asm(".globl " #sym "; .set " #sym ", %0" :: "I"(val)); + MODULE_DESCRIPTION("SHA-224/SHA-256 secure hash using ARMv8 Crypto Extensions"); MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); MODULE_LICENSE("GPL v2"); @@ -29,11 +32,6 @@ struct sha256_ce_state { asmlinkage void sha2_ce_transform(struct sha256_ce_state *sst, u8 const *src, int blocks); -const u32 sha256_ce_offsetof_count = offsetof(struct sha256_ce_state, - sst.count); -const u32 sha256_ce_offsetof_finalize = offsetof(struct sha256_ce_state, - finalize); - static int sha256_ce_update(struct shash_desc *desc, const u8 *data, unsigned int len) { @@ -54,6 +52,11 @@ static int sha256_ce_finup(struct shash_desc *desc, const u8 *data, struct sha256_ce_state *sctx = shash_desc_ctx(desc); bool finalize = !sctx->sst.count && !(len % SHA256_BLOCK_SIZE) && len; + ASM_EXPORT(sha256_ce_offsetof_count, + offsetof(struct sha256_ce_state, sst.count)); + ASM_EXPORT(sha256_ce_offsetof_finalize, + offsetof(struct sha256_ce_state, finalize)); + /* * Allow the asm code to perform the finalization if there is no * partial data and the input is a round multiple of the block size.
This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686. This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce modules. This relocation type is not supported by the linux-4.4.y kernel module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so these modules now fail to load: [ 37.866250] module sha1_ce: unsupported RELA relocation: 275 This issue does not exist with the backport to 4.9+. Bisection shows that this is due to those kernels also having a backport of commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR") Signed-off-by: dann frazier <dann.frazier@canonical.com> --- arch/arm64/crypto/sha1-ce-core.S | 6 ++---- arch/arm64/crypto/sha1-ce-glue.c | 11 ++++++++--- arch/arm64/crypto/sha2-ce-core.S | 6 ++---- arch/arm64/crypto/sha2-ce-glue.c | 13 ++++++++----- 4 files changed, 20 insertions(+), 16 deletions(-)