diff options
Diffstat (limited to 'target/arm')
| -rw-r--r-- | target/arm/hvf/hvf.c | 16 | ||||
| -rw-r--r-- | target/arm/internals.h | 9 | ||||
| -rw-r--r-- | target/arm/meson.build | 2 | ||||
| -rw-r--r-- | target/arm/tcg-stubs.c | 22 | ||||
| -rw-r--r-- | target/arm/tcg/meson.build | 1 | ||||
| -rw-r--r-- | target/arm/tcg/vfp_helper.c (renamed from target/arm/vfp_helper.c) | 189 | ||||
| -rw-r--r-- | target/arm/vfp_fpscr.c | 155 |
7 files changed, 221 insertions, 173 deletions
diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c index 0afd96018e..2439af63a0 100644 --- a/target/arm/hvf/hvf.c +++ b/target/arm/hvf/hvf.c @@ -899,6 +899,18 @@ static bool hvf_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf) clamp_id_aa64mmfr0_parange_to_ipa_size(&host_isar.id_aa64mmfr0); + /* + * Disable SME, which is not properly handled by QEMU hvf yet. + * To allow this through we would need to: + * - make sure that the SME state is correctly handled in the + * get_registers/put_registers functions + * - get the SME-specific CPU properties to work with accelerators + * other than TCG + * - fix any assumptions we made that SME implies SVE (since + * on the M4 there is SME but not SVE) + */ + host_isar.id_aa64pfr1 &= ~R_ID_AA64PFR1_SME_MASK; + ahcf->isar = host_isar; /* @@ -1971,6 +1983,7 @@ int hvf_vcpu_exec(CPUState *cpu) bool isv = syndrome & ARM_EL_ISV; bool iswrite = (syndrome >> 6) & 1; bool s1ptw = (syndrome >> 7) & 1; + bool sse = (syndrome >> 21) & 1; uint32_t sas = (syndrome >> 22) & 3; uint32_t len = 1 << sas; uint32_t srt = (syndrome >> 16) & 0x1f; @@ -1998,6 +2011,9 @@ int hvf_vcpu_exec(CPUState *cpu) address_space_read(&address_space_memory, hvf_exit->exception.physical_address, MEMTXATTRS_UNSPECIFIED, &val, len); + if (sse) { + val = sextract64(val, 0, len * 8); + } hvf_set_reg(cpu, srt, val); } diff --git a/target/arm/internals.h b/target/arm/internals.h index b318734145..a6ff228f9f 100644 --- a/target/arm/internals.h +++ b/target/arm/internals.h @@ -1833,5 +1833,14 @@ int alle1_tlbmask(CPUARMState *env); void arm_set_default_fp_behaviours(float_status *s); /* Set the float_status behaviour to match Arm FPCR.AH=1 behaviour */ void arm_set_ah_fp_behaviours(float_status *s); +/* Read the float_status info and return the appropriate FPSR value */ +uint32_t vfp_get_fpsr_from_host(CPUARMState *env); +/* Clear the exception status flags from all float_status fields */ +void vfp_clear_float_status_exc_flags(CPUARMState *env); +/* + * Update float_status fields to handle the bits of the FPCR + * specified by mask changing to the values in val. + */ +void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask); #endif diff --git a/target/arm/meson.build b/target/arm/meson.build index 2e10464dbb..3065081d24 100644 --- a/target/arm/meson.build +++ b/target/arm/meson.build @@ -4,7 +4,7 @@ arm_ss.add(files( 'debug_helper.c', 'gdbstub.c', 'helper.c', - 'vfp_helper.c', + 'vfp_fpscr.c', )) arm_ss.add(zlib) diff --git a/target/arm/tcg-stubs.c b/target/arm/tcg-stubs.c index f3f45d54f2..93a15cad61 100644 --- a/target/arm/tcg-stubs.c +++ b/target/arm/tcg-stubs.c @@ -30,3 +30,25 @@ void assert_hflags_rebuild_correctly(CPUARMState *env) void define_tlb_insn_regs(ARMCPU *cpu) { } + +/* With KVM, we never use float_status, so these can be no-ops */ +void arm_set_default_fp_behaviours(float_status *s) +{ +} + +void arm_set_ah_fp_behaviours(float_status *s) +{ +} + +uint32_t vfp_get_fpsr_from_host(CPUARMState *env) +{ + return 0; +} + +void vfp_clear_float_status_exc_flags(CPUARMState *env) +{ +} + +void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) +{ +} diff --git a/target/arm/tcg/meson.build b/target/arm/tcg/meson.build index 1f9077c372..dd12ccedb1 100644 --- a/target/arm/tcg/meson.build +++ b/target/arm/tcg/meson.build @@ -41,6 +41,7 @@ arm_ss.add(files( 'vec_helper.c', 'tlb-insns.c', 'arith_helper.c', + 'vfp_helper.c', )) arm_ss.add(when: 'TARGET_AARCH64', if_true: files( diff --git a/target/arm/vfp_helper.c b/target/arm/tcg/vfp_helper.c index 5d424477a2..b32e2f4e27 100644 --- a/target/arm/vfp_helper.c +++ b/target/arm/tcg/vfp_helper.c @@ -23,13 +23,7 @@ #include "internals.h" #include "cpu-features.h" #include "fpu/softfloat.h" -#ifdef CONFIG_TCG #include "qemu/log.h" -#endif - -/* VFP support. We follow the convention used for VFP instructions: - Single precision routines have a "s" suffix, double precision a - "d" suffix. */ /* * Set the float_status behaviour to match the Arm defaults: @@ -75,8 +69,6 @@ void arm_set_ah_fp_behaviours(float_status *s) set_float_default_nan_pattern(0b11000000, s); } -#ifdef CONFIG_TCG - /* Convert host exception flags to vfp form. */ static inline uint32_t vfp_exceptbits_from_host(int host_bits, bool ah) { @@ -113,7 +105,7 @@ static inline uint32_t vfp_exceptbits_from_host(int host_bits, bool ah) return target_bits; } -static uint32_t vfp_get_fpsr_from_host(CPUARMState *env) +uint32_t vfp_get_fpsr_from_host(CPUARMState *env) { uint32_t a32_flags = 0, a64_flags = 0; @@ -148,7 +140,7 @@ static uint32_t vfp_get_fpsr_from_host(CPUARMState *env) vfp_exceptbits_from_host(a32_flags, false); } -static void vfp_clear_float_status_exc_flags(CPUARMState *env) +void vfp_clear_float_status_exc_flags(CPUARMState *env) { /* * Clear out all the exception-flag information in the float_status @@ -176,7 +168,7 @@ static void vfp_sync_and_clear_float_status_exc_flags(CPUARMState *env) vfp_clear_float_status_exc_flags(env); } -static void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) +void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) { uint64_t changed = env->vfp.fpcr; @@ -261,166 +253,11 @@ static void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) } } -#else - -static uint32_t vfp_get_fpsr_from_host(CPUARMState *env) -{ - return 0; -} - -static void vfp_clear_float_status_exc_flags(CPUARMState *env) -{ -} - -static void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) -{ -} - -#endif - -uint32_t vfp_get_fpcr(CPUARMState *env) -{ - uint32_t fpcr = env->vfp.fpcr - | (env->vfp.vec_len << 16) - | (env->vfp.vec_stride << 20); - - /* - * M-profile LTPSIZE is the same bits [18:16] as A-profile Len; whichever - * of the two is not applicable to this CPU will always be zero. - */ - fpcr |= env->v7m.ltpsize << 16; - - return fpcr; -} - -uint32_t vfp_get_fpsr(CPUARMState *env) -{ - uint32_t fpsr = env->vfp.fpsr; - uint32_t i; - - fpsr |= vfp_get_fpsr_from_host(env); - - i = env->vfp.qc[0] | env->vfp.qc[1] | env->vfp.qc[2] | env->vfp.qc[3]; - fpsr |= i ? FPSR_QC : 0; - return fpsr; -} - -uint32_t HELPER(vfp_get_fpscr)(CPUARMState *env) -{ - return (vfp_get_fpcr(env) & FPSCR_FPCR_MASK) | - (vfp_get_fpsr(env) & FPSCR_FPSR_MASK); -} - -uint32_t vfp_get_fpscr(CPUARMState *env) -{ - return HELPER(vfp_get_fpscr)(env); -} - -void vfp_set_fpsr(CPUARMState *env, uint32_t val) -{ - ARMCPU *cpu = env_archcpu(env); - - if (arm_feature(env, ARM_FEATURE_NEON) || - cpu_isar_feature(aa32_mve, cpu)) { - /* - * The bit we set within vfp.qc[] is arbitrary; the array as a - * whole being zero/non-zero is what counts. - */ - env->vfp.qc[0] = val & FPSR_QC; - env->vfp.qc[1] = 0; - env->vfp.qc[2] = 0; - env->vfp.qc[3] = 0; - } - - /* - * NZCV lives only in env->vfp.fpsr. The cumulative exception flags - * IOC|DZC|OFC|UFC|IXC|IDC also live in env->vfp.fpsr, with possible - * extra pending exception information that hasn't yet been folded in - * living in the float_status values (for TCG). - * Since this FPSR write gives us the up to date values of the exception - * flags, we want to store into vfp.fpsr the NZCV and CEXC bits, zeroing - * anything else. We also need to clear out the float_status exception - * information so that the next vfp_get_fpsr does not fold in stale data. - */ - val &= FPSR_NZCV_MASK | FPSR_CEXC_MASK; - env->vfp.fpsr = val; - vfp_clear_float_status_exc_flags(env); -} - -static void vfp_set_fpcr_masked(CPUARMState *env, uint32_t val, uint32_t mask) -{ - /* - * We only set FPCR bits defined by mask, and leave the others alone. - * We assume the mask is sensible (e.g. doesn't try to set only - * part of a field) - */ - ARMCPU *cpu = env_archcpu(env); - - /* When ARMv8.2-FP16 is not supported, FZ16 is RES0. */ - if (!cpu_isar_feature(any_fp16, cpu)) { - val &= ~FPCR_FZ16; - } - if (!cpu_isar_feature(aa64_afp, cpu)) { - val &= ~(FPCR_FIZ | FPCR_AH | FPCR_NEP); - } - - if (!cpu_isar_feature(aa64_ebf16, cpu)) { - val &= ~FPCR_EBF; - } - - vfp_set_fpcr_to_host(env, val, mask); - - if (mask & (FPCR_LEN_MASK | FPCR_STRIDE_MASK)) { - if (!arm_feature(env, ARM_FEATURE_M)) { - /* - * Short-vector length and stride; on M-profile these bits - * are used for different purposes. - * We can't make this conditional be "if MVFR0.FPShVec != 0", - * because in v7A no-short-vector-support cores still had to - * allow Stride/Len to be written with the only effect that - * some insns are required to UNDEF if the guest sets them. - */ - env->vfp.vec_len = extract32(val, 16, 3); - env->vfp.vec_stride = extract32(val, 20, 2); - } else if (cpu_isar_feature(aa32_mve, cpu)) { - env->v7m.ltpsize = extract32(val, FPCR_LTPSIZE_SHIFT, - FPCR_LTPSIZE_LENGTH); - } - } - - /* - * We don't implement trapped exception handling, so the - * trap enable bits, IDE|IXE|UFE|OFE|DZE|IOE are all RAZ/WI (not RES0!) - * - * The FPCR bits we keep in vfp.fpcr are AHP, DN, FZ, RMode, EBF, FZ16, - * FIZ, AH, and NEP. - * Len, Stride and LTPSIZE we just handled. Store those bits - * there, and zero any of the other FPCR bits and the RES0 and RAZ/WI - * bits. - */ - val &= FPCR_AHP | FPCR_DN | FPCR_FZ | FPCR_RMODE_MASK | FPCR_FZ16 | - FPCR_EBF | FPCR_FIZ | FPCR_AH | FPCR_NEP; - env->vfp.fpcr &= ~mask; - env->vfp.fpcr |= val; -} - -void vfp_set_fpcr(CPUARMState *env, uint32_t val) -{ - vfp_set_fpcr_masked(env, val, MAKE_64BIT_MASK(0, 32)); -} - -void HELPER(vfp_set_fpscr)(CPUARMState *env, uint32_t val) -{ - vfp_set_fpcr_masked(env, val, FPSCR_FPCR_MASK); - vfp_set_fpsr(env, val & FPSCR_FPSR_MASK); -} - -void vfp_set_fpscr(CPUARMState *env, uint32_t val) -{ - HELPER(vfp_set_fpscr)(env, val); -} - -#ifdef CONFIG_TCG +/* + * VFP support. We follow the convention used for VFP instructions: + * Single precision routines have a "s" suffix, double precision a + * "d" suffix. + */ #define VFP_HELPER(name, p) HELPER(glue(glue(vfp_,name),p)) @@ -1520,4 +1357,12 @@ void HELPER(check_hcr_el2_trap)(CPUARMState *env, uint32_t rt, uint32_t reg) raise_exception(env, EXCP_HYP_TRAP, syndrome, 2); } -#endif +uint32_t HELPER(vfp_get_fpscr)(CPUARMState *env) +{ + return vfp_get_fpscr(env); +} + +void HELPER(vfp_set_fpscr)(CPUARMState *env, uint32_t val) +{ + vfp_set_fpscr(env, val); +} diff --git a/target/arm/vfp_fpscr.c b/target/arm/vfp_fpscr.c new file mode 100644 index 0000000000..92ea60ebbf --- /dev/null +++ b/target/arm/vfp_fpscr.c @@ -0,0 +1,155 @@ +/* + * ARM VFP floating-point: handling of FPSCR/FPCR/FPSR + * + * Copyright (c) 2003 Fabrice Bellard + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, see <http://www.gnu.org/licenses/>. + */ + +#include "qemu/osdep.h" +#include "cpu.h" +#include "internals.h" +#include "cpu-features.h" + +uint32_t vfp_get_fpcr(CPUARMState *env) +{ + uint32_t fpcr = env->vfp.fpcr + | (env->vfp.vec_len << 16) + | (env->vfp.vec_stride << 20); + + /* + * M-profile LTPSIZE is the same bits [18:16] as A-profile Len; whichever + * of the two is not applicable to this CPU will always be zero. + */ + fpcr |= env->v7m.ltpsize << 16; + + return fpcr; +} + +uint32_t vfp_get_fpsr(CPUARMState *env) +{ + uint32_t fpsr = env->vfp.fpsr; + uint32_t i; + + fpsr |= vfp_get_fpsr_from_host(env); + + i = env->vfp.qc[0] | env->vfp.qc[1] | env->vfp.qc[2] | env->vfp.qc[3]; + fpsr |= i ? FPSR_QC : 0; + return fpsr; +} + +uint32_t vfp_get_fpscr(CPUARMState *env) +{ + return (vfp_get_fpcr(env) & FPSCR_FPCR_MASK) | + (vfp_get_fpsr(env) & FPSCR_FPSR_MASK); +} + +void vfp_set_fpsr(CPUARMState *env, uint32_t val) +{ + ARMCPU *cpu = env_archcpu(env); + + if (arm_feature(env, ARM_FEATURE_NEON) || + cpu_isar_feature(aa32_mve, cpu)) { + /* + * The bit we set within vfp.qc[] is arbitrary; the array as a + * whole being zero/non-zero is what counts. + */ + env->vfp.qc[0] = val & FPSR_QC; + env->vfp.qc[1] = 0; + env->vfp.qc[2] = 0; + env->vfp.qc[3] = 0; + } + + /* + * NZCV lives only in env->vfp.fpsr. The cumulative exception flags + * IOC|DZC|OFC|UFC|IXC|IDC also live in env->vfp.fpsr, with possible + * extra pending exception information that hasn't yet been folded in + * living in the float_status values (for TCG). + * Since this FPSR write gives us the up to date values of the exception + * flags, we want to store into vfp.fpsr the NZCV and CEXC bits, zeroing + * anything else. We also need to clear out the float_status exception + * information so that the next vfp_get_fpsr does not fold in stale data. + */ + val &= FPSR_NZCV_MASK | FPSR_CEXC_MASK; + env->vfp.fpsr = val; + vfp_clear_float_status_exc_flags(env); +} + +static void vfp_set_fpcr_masked(CPUARMState *env, uint32_t val, uint32_t mask) +{ + /* + * We only set FPCR bits defined by mask, and leave the others alone. + * We assume the mask is sensible (e.g. doesn't try to set only + * part of a field) + */ + ARMCPU *cpu = env_archcpu(env); + + /* When ARMv8.2-FP16 is not supported, FZ16 is RES0. */ + if (!cpu_isar_feature(any_fp16, cpu)) { + val &= ~FPCR_FZ16; + } + if (!cpu_isar_feature(aa64_afp, cpu)) { + val &= ~(FPCR_FIZ | FPCR_AH | FPCR_NEP); + } + + if (!cpu_isar_feature(aa64_ebf16, cpu)) { + val &= ~FPCR_EBF; + } + + vfp_set_fpcr_to_host(env, val, mask); + + if (mask & (FPCR_LEN_MASK | FPCR_STRIDE_MASK)) { + if (!arm_feature(env, ARM_FEATURE_M)) { + /* + * Short-vector length and stride; on M-profile these bits + * are used for different purposes. + * We can't make this conditional be "if MVFR0.FPShVec != 0", + * because in v7A no-short-vector-support cores still had to + * allow Stride/Len to be written with the only effect that + * some insns are required to UNDEF if the guest sets them. + */ + env->vfp.vec_len = extract32(val, 16, 3); + env->vfp.vec_stride = extract32(val, 20, 2); + } else if (cpu_isar_feature(aa32_mve, cpu)) { + env->v7m.ltpsize = extract32(val, FPCR_LTPSIZE_SHIFT, + FPCR_LTPSIZE_LENGTH); + } + } + + /* + * We don't implement trapped exception handling, so the + * trap enable bits, IDE|IXE|UFE|OFE|DZE|IOE are all RAZ/WI (not RES0!) + * + * The FPCR bits we keep in vfp.fpcr are AHP, DN, FZ, RMode, EBF, FZ16, + * FIZ, AH, and NEP. + * Len, Stride and LTPSIZE we just handled. Store those bits + * there, and zero any of the other FPCR bits and the RES0 and RAZ/WI + * bits. + */ + val &= FPCR_AHP | FPCR_DN | FPCR_FZ | FPCR_RMODE_MASK | FPCR_FZ16 | + FPCR_EBF | FPCR_FIZ | FPCR_AH | FPCR_NEP; + env->vfp.fpcr &= ~mask; + env->vfp.fpcr |= val; +} + +void vfp_set_fpcr(CPUARMState *env, uint32_t val) +{ + vfp_set_fpcr_masked(env, val, MAKE_64BIT_MASK(0, 32)); +} + +void vfp_set_fpscr(CPUARMState *env, uint32_t val) +{ + vfp_set_fpcr_masked(env, val, FPSCR_FPCR_MASK); + vfp_set_fpsr(env, val & FPSCR_FPSR_MASK); +} |