From b17ab4705c78484d1bdd86cd9c1f7d42338cc1f1 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Mon, 7 Mar 2022 17:16:55 -1000 Subject: [PATCH] target/arm: Fix sve2 ldnt1 and stnt1 For both ldnt1 and stnt1, the meaning of the Rn and Rm are different from ld1 and st1: the vector and integer registers are reversed, and the integer register 31 refers to XZR instead of SP. Secondly, the 64-bit version of ldnt1 was being interpreted as 32-bit unpacked unscaled offset instead of 64-bit unscaled offset, which discarded the upper 32 bits of the address coming from the vector argument. Thirdly, validate that the memory element size is in range for the vector element size for ldnt1. For ld1, we do this via independent decode patterns, but for ldnt1 we need to do it manually. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/826 Signed-off-by: Richard Henderson Reviewed-by: Peter Maydell Message-id: 20220308031655.240710-1-richard.henderson@linaro.org Signed-off-by: Peter Maydell --- target/arm/sve.decode | 5 ++- target/arm/translate-sve.c | 51 +++++++++++++++++++++++++++++-- tests/tcg/aarch64/Makefile.target | 4 +++ tests/tcg/aarch64/test-826.c | 50 ++++++++++++++++++++++++++++++ tests/tcg/configure.sh | 4 +++ 5 files changed, 109 insertions(+), 5 deletions(-) create mode 100644 tests/tcg/aarch64/test-826.c diff --git a/target/arm/sve.decode b/target/arm/sve.decode index c60b9f0fec..0388cce3bd 100644 --- a/target/arm/sve.decode +++ b/target/arm/sve.decode @@ -1575,10 +1575,9 @@ USDOT_zzzz 01000100 .. 0 ..... 011 110 ..... ..... @rda_rn_rm ### SVE2 Memory Gather Load Group -# SVE2 64-bit gather non-temporal load -# (scalar plus unpacked 32-bit unscaled offsets) +# SVE2 64-bit gather non-temporal load (scalar plus 64-bit unscaled offsets) LDNT1_zprz 1100010 msz:2 00 rm:5 1 u:1 0 pg:3 rn:5 rd:5 \ - &rprr_gather_load xs=0 esz=3 scale=0 ff=0 + &rprr_gather_load xs=2 esz=3 scale=0 ff=0 # SVE2 32-bit gather non-temporal load (scalar plus 32-bit unscaled offsets) LDNT1_zprz 1000010 msz:2 00 rm:5 10 u:1 pg:3 rn:5 rd:5 \ diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c index 33ca1bcfac..2c23459e76 100644 --- a/target/arm/translate-sve.c +++ b/target/arm/translate-sve.c @@ -6487,10 +6487,33 @@ static bool trans_LD1_zpiz(DisasContext *s, arg_LD1_zpiz *a) static bool trans_LDNT1_zprz(DisasContext *s, arg_LD1_zprz *a) { + gen_helper_gvec_mem_scatter *fn = NULL; + bool be = s->be_data == MO_BE; + bool mte = s->mte_active[0]; + + if (a->esz < a->msz + !a->u) { + return false; + } if (!dc_isar_feature(aa64_sve2, s)) { return false; } - return trans_LD1_zprz(s, a); + if (!sve_access_check(s)) { + return true; + } + + switch (a->esz) { + case MO_32: + fn = gather_load_fn32[mte][be][0][0][a->u][a->msz]; + break; + case MO_64: + fn = gather_load_fn64[mte][be][0][2][a->u][a->msz]; + break; + } + assert(fn != NULL); + + do_mem_zpz(s, a->rd, a->pg, a->rn, 0, + cpu_reg(s, a->rm), a->msz, false, fn); + return true; } /* Indexed by [mte][be][xs][msz]. */ @@ -6647,10 +6670,34 @@ static bool trans_ST1_zpiz(DisasContext *s, arg_ST1_zpiz *a) static bool trans_STNT1_zprz(DisasContext *s, arg_ST1_zprz *a) { + gen_helper_gvec_mem_scatter *fn; + bool be = s->be_data == MO_BE; + bool mte = s->mte_active[0]; + + if (a->esz < a->msz) { + return false; + } if (!dc_isar_feature(aa64_sve2, s)) { return false; } - return trans_ST1_zprz(s, a); + if (!sve_access_check(s)) { + return true; + } + + switch (a->esz) { + case MO_32: + fn = scatter_store_fn32[mte][be][0][a->msz]; + break; + case MO_64: + fn = scatter_store_fn64[mte][be][2][a->msz]; + break; + default: + g_assert_not_reached(); + } + + do_mem_zpz(s, a->rd, a->pg, a->rn, 0, + cpu_reg(s, a->rm), a->msz, true, fn); + return true; } /* diff --git a/tests/tcg/aarch64/Makefile.target b/tests/tcg/aarch64/Makefile.target index ac07acde66..f7121cb4d8 100644 --- a/tests/tcg/aarch64/Makefile.target +++ b/tests/tcg/aarch64/Makefile.target @@ -86,7 +86,11 @@ run-gdbstub-sve-ioctls: sve-ioctls EXTRA_RUNS += run-gdbstub-sysregs run-gdbstub-sve-ioctls endif +endif +ifneq ($(DOCKER_IMAGE)$(CROSS_CC_HAS_SVE2),) +AARCH64_TESTS += test-826 +test-826: CFLAGS+=-march=armv8.1-a+sve2 endif TESTS += $(AARCH64_TESTS) diff --git a/tests/tcg/aarch64/test-826.c b/tests/tcg/aarch64/test-826.c new file mode 100644 index 0000000000..f59740a8c5 --- /dev/null +++ b/tests/tcg/aarch64/test-826.c @@ -0,0 +1,50 @@ +#include +#include +#include +#include +#include +#include + +static void *expected; + +void sigsegv(int sig, siginfo_t *info, void *vuc) +{ + ucontext_t *uc = vuc; + + assert(info->si_addr == expected); + uc->uc_mcontext.pc += 4; +} + +int main() +{ + struct sigaction sa = { + .sa_sigaction = sigsegv, + .sa_flags = SA_SIGINFO + }; + + void *page; + long ofs; + + if (sigaction(SIGSEGV, &sa, NULL) < 0) { + perror("sigaction"); + return EXIT_FAILURE; + } + + page = mmap(0, getpagesize(), PROT_NONE, MAP_PRIVATE | MAP_ANON, -1, 0); + if (page == MAP_FAILED) { + perror("mmap"); + return EXIT_FAILURE; + } + + ofs = 0x124; + expected = page + ofs; + + asm("ptrue p0.d, vl1\n\t" + "dup z0.d, %0\n\t" + "ldnt1h {z1.d}, p0/z, [z0.d, %1]\n\t" + "dup z1.d, %1\n\t" + "ldnt1h {z0.d}, p0/z, [z1.d, %0]" + : : "r"(page), "r"(ofs) : "v0", "v1"); + + return EXIT_SUCCESS; +} diff --git a/tests/tcg/configure.sh b/tests/tcg/configure.sh index ed4b5ccb1f..84f928f7f8 100755 --- a/tests/tcg/configure.sh +++ b/tests/tcg/configure.sh @@ -299,6 +299,10 @@ for target in $target_list; do -march=armv8.1-a+sve -o $TMPE $TMPC; then echo "CROSS_CC_HAS_SVE=y" >> $config_target_mak fi + if do_compiler "$target_compiler" $target_compiler_cflags \ + -march=armv8.1-a+sve2 -o $TMPE $TMPC; then + echo "CROSS_CC_HAS_SVE2=y" >> $config_target_mak + fi if do_compiler "$target_compiler" $target_compiler_cflags \ -march=armv8.3-a -o $TMPE $TMPC; then echo "CROSS_CC_HAS_ARMV8_3=y" >> $config_target_mak