* [PATCH v2] Set the correct env->fpip for x86 float instructions
@ 2021-04-30 2:19 Ziqiao Kong
2021-04-30 18:19 ` Richard Henderson
0 siblings, 1 reply; 2+ messages in thread
From: Ziqiao Kong @ 2021-04-30 2:19 UTC (permalink / raw)
To: qemu-devel; +Cc: pbonzini, richard.henderson, ehabkost, Ziqiao Kong
Thanks the review for v1 from Richard Henderson!
Changes since v1:
- Don't update FCS, FIP, FDS and FDP for x87 control instruction.
- Also write FCS, FDS and FDP for FSTENV.
- Clear FCS, FIP, FDS and FDP for FXSAVE as intel manual says.
Note:
During my test, I find that the implementation between some intel cpus
differs on updating FDS and FDP while the AMD Ryzen always update the
two registers correctly. Not sure wthether it's a bug or not.
Ziqiao
Signed-off-by: Ziqiao Kong <ziqiaokong@gmail.com>
---
target/i386/cpu.h | 4 +++
target/i386/tcg/fpu_helper.c | 50 ++++++++++++++++++++++--------------
target/i386/tcg/translate.c | 45 +++++++++++++++++++++++++++++++-
3 files changed, 79 insertions(+), 20 deletions(-)
diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index 570f916878..ba43ceb4ad 100644
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -705,6 +705,8 @@ typedef uint64_t FeatureWordArray[FEATURE_WORDS];
#define CPUID_7_0_EBX_INVPCID (1U << 10)
/* Restricted Transactional Memory */
#define CPUID_7_0_EBX_RTM (1U << 11)
+/* Deprecates FPU CS and FPU DS values */
+#define CPUID_7_0_EBX_FCS_FDS (1U << 13)
/* Memory Protection Extension */
#define CPUID_7_0_EBX_MPX (1U << 14)
/* AVX-512 Foundation */
@@ -1440,7 +1442,9 @@ typedef struct CPUX86State {
FPReg fpregs[8];
/* KVM-only so far */
uint16_t fpop;
+ uint16_t fpcs;
uint64_t fpip;
+ uint16_t fpds;
uint64_t fpdp;
/* emulator internal variables */
diff --git a/target/i386/tcg/fpu_helper.c b/target/i386/tcg/fpu_helper.c
index 60ed93520a..97cf68542b 100644
--- a/target/i386/tcg/fpu_helper.c
+++ b/target/i386/tcg/fpu_helper.c
@@ -766,6 +766,10 @@ void helper_fninit(CPUX86State *env)
{
env->fpus = 0;
env->fpstt = 0;
+ env->fpcs = 0;
+ env->fpip = 0;
+ env->fpds = 0;
+ env->fpdp = 0;
cpu_set_fpuc(env, 0x37f);
env->fptags[0] = 1;
env->fptags[1] = 1;
@@ -2368,6 +2372,7 @@ static void do_fstenv(CPUX86State *env, target_ulong ptr, int data32,
{
int fpus, fptag, exp, i;
uint64_t mant;
+ uint16_t fpcs, fpds;
CPU_LDoubleU tmp;
fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
@@ -2390,24 +2395,41 @@ static void do_fstenv(CPUX86State *env, target_ulong ptr, int data32,
}
}
}
+
+ /*
+ * If CR0.PE = 1, each instruction saves FCS and FDS into memory. If
+ * CPUID.(EAX=07H,ECX=0H):EBX[bit 13] = 1, the processor deprecates
+ * FCS and FDS; it saves each as 0000H.
+ */
+ if (!(env->features[FEAT_7_0_EBX] & CPUID_7_0_EBX_FCS_FDS)
+ && (env->cr[0] & CR0_PE_MASK)) {
+ fpcs = env->fpcs;
+ fpds = env->fpds;
+ } else {
+ fpcs = 0;
+ fpds = 0;
+ }
+
if (data32) {
/* 32 bit */
cpu_stl_data_ra(env, ptr, env->fpuc, retaddr);
cpu_stl_data_ra(env, ptr + 4, fpus, retaddr);
cpu_stl_data_ra(env, ptr + 8, fptag, retaddr);
- cpu_stl_data_ra(env, ptr + 12, 0, retaddr); /* fpip */
- cpu_stl_data_ra(env, ptr + 16, 0, retaddr); /* fpcs */
- cpu_stl_data_ra(env, ptr + 20, 0, retaddr); /* fpoo */
- cpu_stl_data_ra(env, ptr + 24, 0, retaddr); /* fpos */
+ cpu_stl_data_ra(env, ptr + 12, env->fpip, retaddr); /* fpip */
+ cpu_stw_data_ra(env, ptr + 16, fpcs, retaddr); /* fpcs */
+ cpu_stw_data_ra(env, ptr + 18, 0, retaddr);
+ cpu_stl_data_ra(env, ptr + 20, env->fpdp, retaddr); /* fpdp */
+ cpu_stw_data_ra(env, ptr + 24, fpds, retaddr); /* fpds */
+ cpu_stw_data_ra(env, ptr + 26, 0, retaddr);
} else {
/* 16 bit */
cpu_stw_data_ra(env, ptr, env->fpuc, retaddr);
cpu_stw_data_ra(env, ptr + 2, fpus, retaddr);
cpu_stw_data_ra(env, ptr + 4, fptag, retaddr);
- cpu_stw_data_ra(env, ptr + 6, 0, retaddr);
- cpu_stw_data_ra(env, ptr + 8, 0, retaddr);
- cpu_stw_data_ra(env, ptr + 10, 0, retaddr);
- cpu_stw_data_ra(env, ptr + 12, 0, retaddr);
+ cpu_stw_data_ra(env, ptr + 6, env->fpip, retaddr);
+ cpu_stw_data_ra(env, ptr + 8, fpcs, retaddr);
+ cpu_stw_data_ra(env, ptr + 10, env->fpdp, retaddr);
+ cpu_stw_data_ra(env, ptr + 12, fpds, retaddr);
}
}
@@ -2473,17 +2495,7 @@ void helper_fsave(CPUX86State *env, target_ulong ptr, int data32)
}
/* fninit */
- env->fpus = 0;
- env->fpstt = 0;
- cpu_set_fpuc(env, 0x37f);
- env->fptags[0] = 1;
- env->fptags[1] = 1;
- env->fptags[2] = 1;
- env->fptags[3] = 1;
- env->fptags[4] = 1;
- env->fptags[5] = 1;
- env->fptags[6] = 1;
- env->fptags[7] = 1;
+ helper_fninit(env);
}
void helper_frstor(CPUX86State *env, target_ulong ptr, int data32)
diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c
index 880bc45561..c26d343ab8 100644
--- a/target/i386/tcg/translate.c
+++ b/target/i386/tcg/translate.c
@@ -4486,6 +4486,11 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
target_ulong next_eip, tval;
int rex_w, rex_r;
target_ulong pc_start = s->base.pc_next;
+ /* For FCS, FIP, FDS and FDP. */
+ AddressParts last_addr;
+ TCGv ea;
+ bool update_fdp;
+ bool update_fip;
s->pc_start = s->pc = pc_start;
s->override = -1;
@@ -4506,6 +4511,9 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
rex_w = -1;
rex_r = 0;
+ update_fip = true;
+ update_fdp = false;
+
next_byte:
b = x86_ldub_code(env, s);
/* Collect prefixes. */
@@ -5850,7 +5858,11 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
op = ((b & 7) << 3) | ((modrm >> 3) & 7);
if (mod != 3) {
/* memory op */
- gen_lea_modrm(env, s, modrm);
+ update_fdp = true;
+ last_addr = gen_lea_modrm_0(env, s, modrm);
+ ea = gen_lea_modrm_1(s, last_addr);
+ gen_lea_v_seg(s, s->aflag, ea, last_addr.def_seg, s->override);
+
switch(op) {
case 0x00 ... 0x07: /* fxxxs */
case 0x10 ... 0x17: /* fixxxl */
@@ -5976,19 +5988,23 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
break;
case 0x0c: /* fldenv mem */
gen_helper_fldenv(cpu_env, s->A0, tcg_const_i32(dflag - 1));
+ update_fip = update_fdp = false;
break;
case 0x0d: /* fldcw mem */
tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
s->mem_index, MO_LEUW);
gen_helper_fldcw(cpu_env, s->tmp2_i32);
+ update_fip = update_fdp = false;
break;
case 0x0e: /* fnstenv mem */
gen_helper_fstenv(cpu_env, s->A0, tcg_const_i32(dflag - 1));
+ update_fip = update_fdp = false;
break;
case 0x0f: /* fnstcw mem */
gen_helper_fnstcw(s->tmp2_i32, cpu_env);
tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
s->mem_index, MO_LEUW);
+ update_fip = update_fdp = false;
break;
case 0x1d: /* fldt mem */
gen_helper_fldt_ST0(cpu_env, s->A0);
@@ -5999,14 +6015,17 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
break;
case 0x2c: /* frstor mem */
gen_helper_frstor(cpu_env, s->A0, tcg_const_i32(dflag - 1));
+ update_fip = update_fdp = false;
break;
case 0x2e: /* fnsave mem */
gen_helper_fsave(cpu_env, s->A0, tcg_const_i32(dflag - 1));
+ update_fip = update_fdp = false;
break;
case 0x2f: /* fnstsw mem */
gen_helper_fnstsw(s->tmp2_i32, cpu_env);
tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
s->mem_index, MO_LEUW);
+ update_fip = update_fdp = false;
break;
case 0x3c: /* fbld */
gen_helper_fbld_ST0(cpu_env, s->A0);
@@ -6047,6 +6066,7 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
case 0: /* fnop */
/* check exceptions (FreeBSD FPU probe) */
gen_helper_fwait(cpu_env);
+ update_fip = update_fdp = false;
break;
default:
goto unknown_op;
@@ -6214,9 +6234,11 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
break;
case 2: /* fclex */
gen_helper_fclex(cpu_env);
+ update_fip = update_fdp = false;
break;
case 3: /* fninit */
gen_helper_fninit(cpu_env);
+ update_fip = update_fdp = false;
break;
case 4: /* fsetpm (287 only, just do nop here) */
break;
@@ -6337,6 +6359,27 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
goto unknown_op;
}
}
+
+
+ if (update_fip) {
+ tcg_gen_movi_tl(s->T0, pc_start - s->cs_base);
+ tcg_gen_st_tl(s->T0, cpu_env, offsetof(CPUX86State, fpip));
+
+ tcg_gen_mov_tl(s->T0, cpu_seg_base[R_CS]);
+ tcg_gen_st16_tl(s->T0, cpu_env, offsetof(CPUX86State, fpcs));
+ }
+
+ if (update_fdp) {
+ if (s->override < 0) {
+ tcg_gen_mov_tl(s->A0, cpu_seg_base[last_addr.def_seg]);
+ } else {
+ tcg_gen_mov_tl(s->A0, cpu_seg_base[s->override]);
+ }
+ tcg_gen_st16_tl(s->A0, cpu_env, offsetof(CPUX86State, fpds));
+
+ ea = gen_lea_modrm_1(s, last_addr);
+ tcg_gen_st_tl(ea, cpu_env, offsetof(CPUX86State, fpdp));
+ }
break;
/************************/
/* string ops */
--
2.25.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH v2] Set the correct env->fpip for x86 float instructions
2021-04-30 2:19 [PATCH v2] Set the correct env->fpip for x86 float instructions Ziqiao Kong
@ 2021-04-30 18:19 ` Richard Henderson
0 siblings, 0 replies; 2+ messages in thread
From: Richard Henderson @ 2021-04-30 18:19 UTC (permalink / raw)
To: Ziqiao Kong, qemu-devel; +Cc: pbonzini, ehabkost
On 4/29/21 7:19 PM, Ziqiao Kong wrote:
> @@ -1440,7 +1442,9 @@ typedef struct CPUX86State {
> FPReg fpregs[8];
> /* KVM-only so far */
> uint16_t fpop;
> + uint16_t fpcs;
> uint64_t fpip;
> + uint16_t fpds;
> uint64_t fpdp;
Let's put all uint16_t together, just after fpop, to avoid extra padding
between the uint64_t.
> + cpu_stw_data_ra(env, ptr + 16, fpcs, retaddr); /* fpcs */
> + cpu_stw_data_ra(env, ptr + 18, 0, retaddr);
...
> + cpu_stw_data_ra(env, ptr + 24, fpds, retaddr); /* fpds */
> + cpu_stw_data_ra(env, ptr + 26, 0, retaddr);
Better to continue to use stl here, to zero-extend the field.
> @@ -5850,7 +5858,11 @@ static target_ulong disas_insn(DisasContext *s, CPUState *cpu)
> op = ((b & 7) << 3) | ((modrm >> 3) & 7);
> if (mod != 3) {
> /* memory op */
> - gen_lea_modrm(env, s, modrm);
> + update_fdp = true;
You should just move the declaration of update_fdp and last_addr into this
block. Then the store to fdp+fds goes at the end of this block.
> + last_addr = gen_lea_modrm_0(env, s, modrm);
> + ea = gen_lea_modrm_1(s, last_addr);
> + gen_lea_v_seg(s, s->aflag, ea, last_addr.def_seg, s->override);
I think you should copy ea into a temporary here.
> + tcg_gen_mov_tl(s->T0, cpu_seg_base[R_CS]);
> + tcg_gen_st16_tl(s->T0, cpu_env, offsetof(CPUX86State, fpcs));
cpu_seg_base is segment base, not the segment selector.
I believe that you want to copy offsetof(CPUX86State, segs[seg_reg].selector)
into this field.
> + ea = gen_lea_modrm_1(s, last_addr);
> + tcg_gen_st_tl(ea, cpu_env, offsetof(CPUX86State, fpdp));
Use the temporary saved above instead of re-computing.
r~
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-04-30 18:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-30 2:19 [PATCH v2] Set the correct env->fpip for x86 float instructions Ziqiao Kong
2021-04-30 18:19 ` Richard Henderson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).