Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 02/03/2016 09:06, Hervé Poussineau wrote: >> > > I just reconfirmed that > d6a2914984c89fa0a3125b9842e0cbf68de79a3d~1 + > 88c73d16ad1b6c22a2ab082064d0d521f756296a works, > while > d6a2914984c89fa0a3125b9842e0cbf68de79a3d + > 88c73d16ad1b6c22a2ab082064d0d521f756296a bugchecks. > > a5af12871fd4601c44f08d9e49131e9ca13ef102 that you tested is after > d6a2914984c89fa0a3125b9842e0cbf68de79a3d which broke Windows 9x, that's > why you're seeing it broken. > > It also has been a long time that Windows 9x doesn't work with > -enable-kvm. Maybe even always... This was tricky because the faulting instruction is some kind of generated thunk. In the good QEMU it is like: 0x8030a4da: mov$0x16f,%ax 0x8030a4dd: mov%ax,%es 0x8030a4df: movzwl %cx,%ecx 0x8030a4e3: addr32 mov %es:-0x400aa28c(%ecx),%edx 0x8030a4ec: ljmpl $0x167,$0xbff71903 In the bad one, the address is %es:0(%ecx) and everything goes south from there. Therefore I copied the old code, changed to generate the effective address in a new temp cpu_A1, then I added a helper that gets the two effective addresses and asserts if they mismatch. Sure enough it fired on an addr16 mov %gs:(%bx),%eax The issue is that the /* ADDSEG will only be false in 16-bit mode for LEA. */ comment is false when in 32-bit mode but with an addr16 prefix. I still have to forward port it, and test on a newer commit, but the attached patch can boot Windows 98. Paolo diff --git a/target-i386/translate.c b/target-i386/translate.c index 282f4a1..0a2d091 100644 --- a/target-i386/translate.c +++ b/target-i386/translate.c @@ -61,6 +61,7 @@ /* global register indexes */ static TCGv_ptr cpu_env; static TCGv cpu_A0; +static TCGv cpu_A1; static TCGv cpu_cc_dst, cpu_cc_src, cpu_cc_src2, cpu_cc_srcT; static TCGv_i32 cpu_cc_op; static TCGv cpu_regs[CPU_NB_REGS]; @@ -419,6 +420,18 @@ static inline void gen_op_add_reg_T0(TCGMemOp size, int reg) gen_op_mov_reg_v(size, reg, cpu_tmp0); } +static inline void gen_op_addl_A1_seg(DisasContext *s, int reg) +{ +tcg_gen_ld_tl(cpu_tmp0, cpu_env, offsetof(CPUX86State, segs[reg].base)); +if (CODE64(s)) { +tcg_gen_ext32u_tl(cpu_A1, cpu_A1); +tcg_gen_add_tl(cpu_A1, cpu_A1, cpu_tmp0); +} else { +tcg_gen_add_tl(cpu_A1, cpu_A1, cpu_tmp0); +tcg_gen_ext32u_tl(cpu_A1, cpu_A1); +} +} + static inline void gen_op_addl_A0_seg(DisasContext *s, int reg) { tcg_gen_ld_tl(cpu_tmp0, cpu_env, offsetof(CPUX86State, segs[reg].base)); @@ -485,13 +498,13 @@ static void gen_lea_v_seg(DisasContext *s, TCGv a0, int def_seg, int ovr_seg) break; case MO_16: /* 16 bit address */ -if (ovr_seg < 0) { -ovr_seg = def_seg; -} tcg_gen_ext16u_tl(cpu_A0, a0); -/* ADDSEG will only be false in 16-bit mode for LEA. */ -if (!s->addseg) { -return; +if (ovr_seg < 0) { +if (s->addseg) { +ovr_seg = def_seg; +} else { +return; +} } a0 = cpu_A0; break; @@ -1838,12 +1851,198 @@ static void gen_shifti(DisasContext *s1, int op, TCGMemOp ot, int d, int c) } } +static void gen_lea_modrm_old(CPUX86State *env, DisasContext *s, int modrm) +{ +target_long disp; +int havesib; +int base; +int index; +int scale; +int mod, rm, code, override, must_add_seg; +TCGv sum; + +override = s->override; +must_add_seg = s->addseg; +if (override >= 0) +must_add_seg = 1; +mod = (modrm >> 6) & 3; +rm = modrm & 7; + +switch (s->aflag) { +case MO_64: +case MO_32: +havesib = 0; +base = rm; +index = -1; +scale = 0; + +if (base == 4) { +havesib = 1; +code = cpu_ldub_code(env, s->pc++); +scale = (code >> 6) & 3; +index = ((code >> 3) & 7) | REX_X(s); +if (index == 4) { +index = -1; /* no index */ +} +base = (code & 7); +} +base |= REX_B(s); + +switch (mod) { +case 0: +if ((base & 7) == 5) { +base = -1; +disp = (int32_t)cpu_ldl_code(env, s->pc); +s->pc += 4; +if (CODE64(s) && !havesib) { +disp += s->pc + s->rip_offset; +} +} else { +disp = 0; +} +break; +case 1: +disp = (int8_t)cpu_ldub_code(env, s->pc++); +break; +default: +case 2: +disp = (int32_t)cpu_ldl_code(env, s->pc); +s->pc += 4; +break; +} + +/* For correct popl handling with esp. */ +if (base == R_ESP && s->popl_esp_hack) { +disp += s->popl_esp_hack; +} + +/* Compute the address, with a minimu
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 02/03/2016 09:06, Hervé Poussineau wrote: > I just reconfirmed that > d6a2914984c89fa0a3125b9842e0cbf68de79a3d~1 + > 88c73d16ad1b6c22a2ab082064d0d521f756296a works, > while > d6a2914984c89fa0a3125b9842e0cbf68de79a3d + > 88c73d16ad1b6c22a2ab082064d0d521f756296a bugchecks. > > a5af12871fd4601c44f08d9e49131e9ca13ef102 that you tested is after > d6a2914984c89fa0a3125b9842e0cbf68de79a3d which broke Windows 9x, that's > why you're seeing it broken. > > It also has been a long time that Windows 9x doesn't work with > -enable-kvm. Maybe even always... That's a different problem, Windows 9x doesn't like too fast clocks (similar to the Borland Pascal divide overflow found in several DOS games). TCG slows it down enough that it doesn't complain. Paolo
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
Le 02/03/2016 05:05, Richard Henderson a écrit : On 03/01/2016 12:03 PM, Hervé Poussineau wrote: Windows 95 still doesn't work, even with your two patches applied. The message is: "A fatal exception 0E has occurred at 0137:FD512607. The current application will be terminated. * Press any key to terminate the current application. * Press CTRL-ALT-DEL to restart your computer. You will lose any unsaved information in all applications. Press any key to continue." I get this same message (with 98) before all of the recent patches, i.e. testing a5af12871fd4601c44f08d9e49131e9ca13ef102. Of course, while I don't get this message from -enable-kvm, I'm still not able to boot 98 even with kvm. Which doesn't fill me with happy feelings about the state of the system, even though it's a fresh install. I just reconfirmed that d6a2914984c89fa0a3125b9842e0cbf68de79a3d~1 + 88c73d16ad1b6c22a2ab082064d0d521f756296a works, while d6a2914984c89fa0a3125b9842e0cbf68de79a3d + 88c73d16ad1b6c22a2ab082064d0d521f756296a bugchecks. a5af12871fd4601c44f08d9e49131e9ca13ef102 that you tested is after d6a2914984c89fa0a3125b9842e0cbf68de79a3d which broke Windows 9x, that's why you're seeing it broken. It also has been a long time that Windows 9x doesn't work with -enable-kvm. Maybe even always... Hervé
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 03/01/2016 12:03 PM, Hervé Poussineau wrote: Windows 95 still doesn't work, even with your two patches applied. The message is: "A fatal exception 0E has occurred at 0137:FD512607. The current application will be terminated. * Press any key to terminate the current application. * Press CTRL-ALT-DEL to restart your computer. You will lose any unsaved information in all applications. Press any key to continue." I get this same message (with 98) before all of the recent patches, i.e. testing a5af12871fd4601c44f08d9e49131e9ca13ef102. Of course, while I don't get this message from -enable-kvm, I'm still not able to boot 98 even with kvm. Which doesn't fill me with happy feelings about the state of the system, even though it's a fresh install. r~
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 01/03/2016 21:03, Hervé Poussineau wrote: > I also tested Windows 98. The first part setup correctly work, but > doesn't work just after the reboot (for the first boot). > The message is very similar to Windows 95. Indeed I stopped it after it got to the graphical part. I'll test again tomorrow, though perhaps Richard can beat me to it. Paolo
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
Le 01/03/2016 16:12, Paolo Bonzini a écrit : On 28/02/2016 22:49, Hervé Poussineau wrote: 3) MS-DOS 6 freezes when loading himem.sys since commit: commit 1906b2af7c2345037d9b2fdf484b457b5acd09d1 Author: Richard Henderson Date: Thu Jul 2 13:59:21 2015 +0100 target-i386: Rearrange processing of 0F 01 Rather than nesting tests of OP, MOD, and RM, decode them all at once with a switch. Fixes incorrect decoding of AMD Pacifica extensions (aka vmrun et al) via op==2 path. Signed-off-by: Richard Henderson I'm starting QEMU with -cpu 486. It works on master if I add -enable-kvm Please test the other patch I've just sent. I have not looked at 2, but that patch seems to fix the Windows 98 setup CD for me besides fixing the problem with this commit. If it's not enough for your Windows 95 testcase, please try getting a trace with "-d in_asm,op_opt,int". Windows 95 still doesn't work, even with your two patches applied. The message is: "A fatal exception 0E has occurred at 0137:FD512607. The current application will be terminated. * Press any key to terminate the current application. * Press CTRL-ALT-DEL to restart your computer. You will lose any unsaved information in all applications. Press any key to continue." I can provide debug log by email if required. I also tested Windows 98. The first part setup correctly work, but doesn't work just after the reboot (for the first boot). The message is very similar to Windows 95. Hervé
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 28/02/2016 22:49, Hervé Poussineau wrote: > > 3) MS-DOS 6 freezes when loading himem.sys since commit: > commit 1906b2af7c2345037d9b2fdf484b457b5acd09d1 > Author: Richard Henderson > Date: Thu Jul 2 13:59:21 2015 +0100 > > target-i386: Rearrange processing of 0F 01 > > Rather than nesting tests of OP, MOD, and RM, decode them > all at once with a switch. Fixes incorrect decoding of > AMD Pacifica extensions (aka vmrun et al) via op==2 path. > > Signed-off-by: Richard Henderson > > I'm starting QEMU with -cpu 486. > It works on master if I add -enable-kvm Please test the other patch I've just sent. I have not looked at 2, but that patch seems to fix the Windows 98 setup CD for me besides fixing the problem with this commit. If it's not enough for your Windows 95 testcase, please try getting a trace with "-d in_asm,op_opt,int". Paolo
Re: [Qemu-devel] [QEMU] Windows XP / Windows 95 / MS-DOS 6 regressions
On 28/02/2016 22:49, Hervé Poussineau wrote: > > > I currently see some regressions on Microsoft operating systems. > > 1) Windows XP bugchecks since commit: > commit 7f0b7141b4c7deab51efd8ee1e83eab2d9b7a9ea > Author: Richard Henderson > Date: Mon Jul 6 17:29:59 2015 +0100 > > target-i386: Perform set/reset_inhibit_irq inline > > With helpers that can be reused for other things. > > Signed-off-by: Richard Henderson > > I'm starting QEMU with -cpu pentium2. > Attached patch can be applied on master to work-around the problem. > Another work-around is to start with -enable-kvm. Ok, so let's go with the first one... The patch is incorrect because it looks at s->tb->flags. I'm posting a fix (mostly a revert) soon. Paolo