busybox-w32/networking/tls_sp_c32.c, branch zstd

tls: make ECDHE_RSA work against our client (openssl s_client not yet)

2026-02-15T14:16:26+00:00

function old new delta tls_handshake_as_server 1601 2033 +432 sp_ecc_make_key_256 - 103 +103 curve_P256_compute_premaster - 65 +65 .rodata 108023 108079 +56 curve_x25519_generate_keypair - 44 +44 tls_get_zeroed_outbuf - 28 +28 curve_P256_generate_keypair - 27 +27 sp_256_from_bin_8 - 26 +26 curve_x25519_compute_premaster - 15 +15 tls_xread_record 708 704 -4 tls_handshake 1530 1519 -11 get_outbuf_fill_handshake_record 51 37 -14 sp_256_point_from_bin2x32 70 43 -27 curve_x25519_compute_pubkey_and_premaster 71 39 -32 curve_P256_compute_pubkey_and_premaster 167 65 -102 ------------------------------------------------------------------------------ (add/remove: 7/0 grow/shrink: 2/6 up/down: 796/-190) Total: 606 bytes Signed-off-by: Denys Vlasenko

tls: P256: improve x86_64 multiplication asm code

2024-07-12T09:10:24+00:00

gcc is being rather silly. Usues suboptimal registers, and does not realize that i and j are never negative, thus usese even _more_ registers for temporaries to sign-extend i/j to 64-bit offsets. function old new delta sp_256_mont_mul_8 155 132 -23 Signed-off-by: Denys Vlasenko

tls: P256: fix obscure x86_64 asm misbehavior, closes 15679

2024-07-11T21:53:47+00:00

gcc does not necessarily clear upper bits in 64-bit regs if you ask it to load a 32-bit constant. Cast it to unsigned long. Better yet, hand-write loading of the constant with a smaller instruction. Signed-off-by: Denys Vlasenko

tls: P256: remove NOP macro sp_256_norm_8()

2022-07-13T14:11:17+00:00

Signed-off-by: Denys Vlasenko

tls: P256: factor out "multiply then reduce" operation

2021-12-11T22:27:40+00:00

function old new delta sp_256_mont_mul_and_reduce_8 - 44 +44 sp_256_ecc_mulmod_8 517 442 -75 ------------------------------------------------------------------------------ (add/remove: 1/0 grow/shrink: 0/1 up/down: 44/-75) Total: -31 bytes Signed-off-by: Denys Vlasenko

tls: P256: enable 64-bit version of montgomery reduction

2021-11-28T20:43:51+00:00

After more testing, (1) I'm more sure it is indeed correct, and (2) it is a significant speedup - we do a lot of those multiplications. function old new delta sp_512to256_mont_reduce_8 191 223 +32 Signed-off-by: Denys Vlasenko

tls: P256: add 64-bit montgomery reduce (disabled), small optimization in 32-bit code

2021-11-28T14:44:08+00:00

function old new delta sp_512to256_mont_reduce_8 191 185 -6 Signed-off-by: Denys Vlasenko

tls: P256: add comment on logic in sp_512to256_mont_reduce_8, no code changes

2021-11-28T11:55:20+00:00

Signed-off-by: Denys Vlasenko

tls: P256: simplify sp_256_mont_inv_8 (no need for a temporary)

2021-11-28T10:15:34+00:00

function old new delta sp_256_ecc_mulmod_8 543 517 -26 Signed-off-by: Denys Vlasenko

tls: P256: pad struct sp_point to 64 bits (on 64-bit arches)

2021-11-28T01:56:02+00:00

function old new delta curve_P256_compute_pubkey_and_premaster 198 190 -8 Signed-off-by: Denys Vlasenko