ShuriZma
/
xemu
mirror of https://github.com/xemu-project/xemu.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

target/arm: Convert sha512 and sm3 to gvec helpers 

Do not yet convert the helpers to loop over opr_sz, but the
descriptor allows the vector tail to be cleared.  Which fixes
an existing bug vs SVE.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200514212831.31248-4-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Richard Henderson 2020-05-14 14:28:28 -07:00 committed by Peter Maydell
parent 1738860d7e
commit aaffebd6d3
3 changed files with 55 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
target/arm/crypto_helper.c
View File

@ -31,6 +31,19 @@ union CRYPTO_STATE {
#define CR_ST_WORD(state, i) (state.words[i]) #define CR_ST_WORD(state, i) (state.words[i])
#endif #endif
/*
* The caller has not been converted to full gvec, and so only
* modifies the low 16 bytes of the vector register.
*/
static void clear_tail_16(void *vd, uint32_t desc)
{
int opr_sz = simd_oprsz(desc);
int max_sz = simd_maxsz(desc);
assert(opr_sz == 16);
clear_tail(vd, opr_sz, max_sz);
}
static void do_crypto_aese(uint64_t *rd, uint64_t *rn, static void do_crypto_aese(uint64_t *rd, uint64_t *rn,
uint64_t *rm, bool decrypt) uint64_t *rm, bool decrypt)
{ {
@ -470,7 +483,7 @@ static uint64_t s1_512(uint64_t x)
return ror64(x, 19) ^ ror64(x, 61) ^ (x >> 6); return ror64(x, 19) ^ ror64(x, 61) ^ (x >> 6);
} }
void HELPER(crypto_sha512h)(void *vd, void *vn, void *vm) void HELPER(crypto_sha512h)(void *vd, void *vn, void *vm, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -483,9 +496,11 @@ void HELPER(crypto_sha512h)(void *vd, void *vn, void *vm)
rd[0] = d0; rd[0] = d0;
rd[1] = d1; rd[1] = d1;
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sha512h2)(void *vd, void *vn, void *vm) void HELPER(crypto_sha512h2)(void *vd, void *vn, void *vm, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -498,9 +513,11 @@ void HELPER(crypto_sha512h2)(void *vd, void *vn, void *vm)
rd[0] = d0; rd[0] = d0;
rd[1] = d1; rd[1] = d1;
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sha512su0)(void *vd, void *vn) void HELPER(crypto_sha512su0)(void *vd, void *vn, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -512,9 +529,11 @@ void HELPER(crypto_sha512su0)(void *vd, void *vn)
rd[0] = d0; rd[0] = d0;
rd[1] = d1; rd[1] = d1;
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sha512su1)(void *vd, void *vn, void *vm) void HELPER(crypto_sha512su1)(void *vd, void *vn, void *vm, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -522,9 +541,11 @@ void HELPER(crypto_sha512su1)(void *vd, void *vn, void *vm)
rd[0] += s1_512(rn[0]) + rm[0]; rd[0] += s1_512(rn[0]) + rm[0];
rd[1] += s1_512(rn[1]) + rm[1]; rd[1] += s1_512(rn[1]) + rm[1];
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sm3partw1)(void *vd, void *vn, void *vm) void HELPER(crypto_sm3partw1)(void *vd, void *vn, void *vm, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -548,9 +569,11 @@ void HELPER(crypto_sm3partw1)(void *vd, void *vn, void *vm)
rd[0] = d.l[0]; rd[0] = d.l[0];
rd[1] = d.l[1]; rd[1] = d.l[1];
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sm3partw2)(void *vd, void *vn, void *vm) void HELPER(crypto_sm3partw2)(void *vd, void *vn, void *vm, uint32_t desc)
{ {
uint64_t *rd = vd; uint64_t *rd = vd;
uint64_t *rn = vn; uint64_t *rn = vn;
@ -568,6 +591,8 @@ void HELPER(crypto_sm3partw2)(void *vd, void *vn, void *vm)
rd[0] = d.l[0]; rd[0] = d.l[0];
rd[1] = d.l[1]; rd[1] = d.l[1];
clear_tail_16(vd, desc);
} }
void HELPER(crypto_sm3tt)(void *vd, void *vn, void *vm, uint32_t imm2, void HELPER(crypto_sm3tt)(void *vd, void *vn, void *vm, uint32_t imm2,

15
target/arm/helper.h
View File

@ -522,14 +522,17 @@ DEF_HELPER_FLAGS_3(crypto_sha256h2, TCG_CALL_NO_RWG, void, ptr, ptr, ptr)
DEF_HELPER_FLAGS_2(crypto_sha256su0, TCG_CALL_NO_RWG, void, ptr, ptr) DEF_HELPER_FLAGS_2(crypto_sha256su0, TCG_CALL_NO_RWG, void, ptr, ptr)
DEF_HELPER_FLAGS_3(crypto_sha256su1, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) DEF_HELPER_FLAGS_3(crypto_sha256su1, TCG_CALL_NO_RWG, void, ptr, ptr, ptr)
DEF_HELPER_FLAGS_3(crypto_sha512h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(crypto_sha512h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(crypto_sha512h2, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(crypto_sha512h2, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_2(crypto_sha512su0, TCG_CALL_NO_RWG, void, ptr, ptr) DEF_HELPER_FLAGS_3(crypto_sha512su0, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(crypto_sha512su1, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(crypto_sha512su1, TCG_CALL_NO_RWG,
void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_5(crypto_sm3tt, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32) DEF_HELPER_FLAGS_5(crypto_sm3tt, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32)
DEF_HELPER_FLAGS_3(crypto_sm3partw1, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(crypto_sm3partw1, TCG_CALL_NO_RWG,
DEF_HELPER_FLAGS_3(crypto_sm3partw2, TCG_CALL_NO_RWG, void, ptr, ptr, ptr) void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(crypto_sm3partw2, TCG_CALL_NO_RWG,
void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(crypto_sm4e, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(crypto_sm4e, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(crypto_sm4ekey, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(crypto_sm4ekey, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)

50
target/arm/translate-a64.c
View File

@ -13624,7 +13624,6 @@ static void disas_crypto_three_reg_sha512(DisasContext *s, uint32_t insn)
int rn = extract32(insn, 5, 5); int rn = extract32(insn, 5, 5);
int rd = extract32(insn, 0, 5); int rd = extract32(insn, 0, 5);
bool feature; bool feature;
CryptoThreeOpFn *genfn = NULL;
gen_helper_gvec_3 *oolfn = NULL; gen_helper_gvec_3 *oolfn = NULL;
GVecGen3Fn *gvecfn = NULL; GVecGen3Fn *gvecfn = NULL;
@ -13632,15 +13631,15 @@ static void disas_crypto_three_reg_sha512(DisasContext *s, uint32_t insn)
switch (opcode) { switch (opcode) {
case 0: /* SHA512H */ case 0: /* SHA512H */
feature = dc_isar_feature(aa64_sha512, s); feature = dc_isar_feature(aa64_sha512, s);
genfn = gen_helper_crypto_sha512h; oolfn = gen_helper_crypto_sha512h;
break; break;
case 1: /* SHA512H2 */ case 1: /* SHA512H2 */
feature = dc_isar_feature(aa64_sha512, s); feature = dc_isar_feature(aa64_sha512, s);
genfn = gen_helper_crypto_sha512h2; oolfn = gen_helper_crypto_sha512h2;
break; break;
case 2: /* SHA512SU1 */ case 2: /* SHA512SU1 */
feature = dc_isar_feature(aa64_sha512, s); feature = dc_isar_feature(aa64_sha512, s);
genfn = gen_helper_crypto_sha512su1; oolfn = gen_helper_crypto_sha512su1;
break; break;
case 3: /* RAX1 */ case 3: /* RAX1 */
feature = dc_isar_feature(aa64_sha3, s); feature = dc_isar_feature(aa64_sha3, s);
@ -13653,11 +13652,11 @@ static void disas_crypto_three_reg_sha512(DisasContext *s, uint32_t insn)
switch (opcode) { switch (opcode) {
case 0: /* SM3PARTW1 */ case 0: /* SM3PARTW1 */
feature = dc_isar_feature(aa64_sm3, s); feature = dc_isar_feature(aa64_sm3, s);
genfn = gen_helper_crypto_sm3partw1; oolfn = gen_helper_crypto_sm3partw1;
break; break;
case 1: /* SM3PARTW2 */ case 1: /* SM3PARTW2 */
feature = dc_isar_feature(aa64_sm3, s); feature = dc_isar_feature(aa64_sm3, s);
genfn = gen_helper_crypto_sm3partw2; oolfn = gen_helper_crypto_sm3partw2;
break; break;
case 2: /* SM4EKEY */ case 2: /* SM4EKEY */
feature = dc_isar_feature(aa64_sm4, s); feature = dc_isar_feature(aa64_sm4, s);
@ -13680,20 +13679,8 @@ static void disas_crypto_three_reg_sha512(DisasContext *s, uint32_t insn)
if (oolfn) { if (oolfn) {
gen_gvec_op3_ool(s, true, rd, rn, rm, 0, oolfn); gen_gvec_op3_ool(s, true, rd, rn, rm, 0, oolfn);
} else if (gvecfn) {
gen_gvec_fn3(s, true, rd, rn, rm, gvecfn, MO_64);
} else { } else {
TCGv_ptr tcg_rd_ptr, tcg_rn_ptr, tcg_rm_ptr; gen_gvec_fn3(s, true, rd, rn, rm, gvecfn, MO_64);
tcg_rd_ptr = vec_full_reg_ptr(s, rd);
tcg_rn_ptr = vec_full_reg_ptr(s, rn);
tcg_rm_ptr = vec_full_reg_ptr(s, rm);
genfn(tcg_rd_ptr, tcg_rn_ptr, tcg_rm_ptr);
tcg_temp_free_ptr(tcg_rd_ptr);
tcg_temp_free_ptr(tcg_rn_ptr);
tcg_temp_free_ptr(tcg_rm_ptr);
} }
} }
@ -13708,19 +13695,14 @@ static void disas_crypto_two_reg_sha512(DisasContext *s, uint32_t insn)
int opcode = extract32(insn, 10, 2); int opcode = extract32(insn, 10, 2);
int rn = extract32(insn, 5, 5); int rn = extract32(insn, 5, 5);
int rd = extract32(insn, 0, 5); int rd = extract32(insn, 0, 5);
TCGv_ptr tcg_rd_ptr, tcg_rn_ptr;
bool feature; bool feature;
CryptoTwoOpFn *genfn;
gen_helper_gvec_3 *oolfn = NULL;
switch (opcode) { switch (opcode) {
case 0: /* SHA512SU0 */ case 0: /* SHA512SU0 */
feature = dc_isar_feature(aa64_sha512, s); feature = dc_isar_feature(aa64_sha512, s);
genfn = gen_helper_crypto_sha512su0;
break; break;
case 1: /* SM4E */ case 1: /* SM4E */
feature = dc_isar_feature(aa64_sm4, s); feature = dc_isar_feature(aa64_sm4, s);
oolfn = gen_helper_crypto_sm4e;
break; break;
default: default:
unallocated_encoding(s); unallocated_encoding(s);
@ -13736,18 +13718,16 @@ static void disas_crypto_two_reg_sha512(DisasContext *s, uint32_t insn)
return; return;
} }
if (oolfn) { switch (opcode) {
gen_gvec_op3_ool(s, true, rd, rd, rn, 0, oolfn); case 0: /* SHA512SU0 */
return; gen_gvec_op2_ool(s, true, rd, rn, 0, gen_helper_crypto_sha512su0);
break;
case 1: /* SM4E */
gen_gvec_op3_ool(s, true, rd, rd, rn, 0, gen_helper_crypto_sm4e);
break;
default:
g_assert_not_reached();
} }
tcg_rd_ptr = vec_full_reg_ptr(s, rd);
tcg_rn_ptr = vec_full_reg_ptr(s, rn);
genfn(tcg_rd_ptr, tcg_rn_ptr);
tcg_temp_free_ptr(tcg_rd_ptr);
tcg_temp_free_ptr(tcg_rn_ptr);
} }
/* Crypto four-register /* Crypto four-register