On 6/7/19 9:09 AM, Laurent Vivier wrote: > On 07/06/2019 11:29, Laurent Vivier wrote: >> On 29/05/2019 08:49, David Gibson wrote: >>> From: Richard Henderson <richard.hender...@linaro.org> >>> >>> The gvec expanders take care of masking the shift amount >>> against the element width. >>> >>> Signed-off-by: Richard Henderson <richard.hender...@linaro.org> >>> Message-Id: <20190518191430.21686-2-richard.hender...@linaro.org> >>> Signed-off-by: David Gibson <da...@gibson.dropbear.id.au> >>> --- >>> target/ppc/helper.h | 12 ---------- >>> target/ppc/int_helper.c | 37 ----------------------------- >>> target/ppc/translate/vmx-impl.inc.c | 24 +++++++++---------- >>> 3 files changed, 12 insertions(+), 61 deletions(-) >> >> This patch introduces a regressions >> with Fedora 29 guest: >> >> - during kernel boot: >> >> [ 40.397876] crypto_register_alg 'aes' = 0 >> [ 40.577517] crypto_register_alg 'cbc(aes)' = 0 >> [ 40.743576] crypto_register_alg 'ctr(aes)' = 0 >> [ 41.061379] alg: skcipher: Test 1 failed (invalid result) on encryption >> for p8_aes_xts >> [ 41.062054] 00000000: 91 7c f6 9e bd 68 b2 ec 9b 9f e9 a3 ea dd a6 92 >> [ 41.062163] 00000010: 98 10 35 57 5e dc 36 1e 9a f7 bc ba 39 f2 5c eb >> [ 41.062834] crypto_register_alg 'xts(aes)' = 0 >> [ 41.077358] alg: hash: Test 2 failed for p8_ghash >> [ 41.077553] 00000000: 5f 89 ab f7 20 57 20 57 20 57 20 57 20 57 20 57 >> >> - with libssl: >> >> # curl -o /dev/null https://www.google.com >> % Total % Received % Xferd Average Speed Time Time Time >> Current >> Dload Upload Total Spent Left Speed >> 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- >> 0 >> curl: (35) error:1408F119:SSL routines:ssl3_get_record:decryption failed or >> bad record mac >> >> [before, this one fails with: >> curl: (35) error:04091068:rsa routines:int_rsa_verify:bad signature ] >> >> If I revert this patch on top of 0d74f3b427 + "target/ppc: Fix lxvw4x, >> lxvh8x and lxvb16x", all works fine. >> >> Thanks, >> Laurent >> > > This seems to fix the problem: > > diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c > index 3b6052fe97..6f0709b307 100644 > --- a/accel/tcg/tcg-runtime-gvec.c > +++ b/accel/tcg/tcg-runtime-gvec.c > @@ -874,7 +874,7 @@ void HELPER(gvec_sar8v)(void *d, void *a, void *b, > uint32_t desc) > intptr_t oprsz = simd_oprsz(desc); > intptr_t i; > > - for (i = 0; i < oprsz; i += sizeof(vec8)) { > + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { > uint8_t sh = *(uint8_t *)(b + i) & 7; > *(int8_t *)(d + i) = *(int8_t *)(a + i) >> sh; > }
Grr. I really really need to come up with a solution for testing that allows me to test paths that the host cpu would not ordinarily take. This bug is hidden on a host with AVX2. Thanks for the digging. r~
