cpu: completely remove PM fpu state

Let back-ends to deal with save/restore.

src/base/emu-i386/cpu.c

@@ -93,7 +93,7 @@ static unsigned int TRs[2] =
 #endif
 
 /* fpu_state needs to be paragraph aligned for fxrstor/fxsave */
-fpregset_t vm86_fpu_state;
+struct emu_fpstate vm86_fpu_state __attribute__((aligned(16)));
 fenv_t dosemu_fenv;
 static void fpu_reset(void);
 
@@ -242,13 +242,13 @@ static void fpu_init(void)
 static void fpu_reset(void)
 {
 #ifdef __x86_64__
-  vm86_fpu_state->cwd = 0x0040;
-  vm86_fpu_state->swd = 0;
-  vm86_fpu_state->ftw = 0x5555;       //bochs
+  vm86_fpu_state.cwd = 0x0040;
+  vm86_fpu_state.swd = 0;
+  vm86_fpu_state.ftw = 0x5555;       //bochs
 #else
-  vm86_fpu_state->cw = 0x40;
-  vm86_fpu_state->sw = 0;
-  vm86_fpu_state->tag = 0x5555;       // bochs
+  vm86_fpu_state.cw = 0x40;
+  vm86_fpu_state.sw = 0;
+  vm86_fpu_state.tag = 0x5555;       // bochs
 #endif
 }
 
@@ -263,9 +263,9 @@ static void fpu_io_write(ioport_t port, Bit8u val)
   case 0xf0:
     /* not sure about this */
 #ifdef __x86_64__
-    vm86_fpu_state->swd &= ~0x8000;
+    vm86_fpu_state.swd &= ~0x8000;
 #else
-    vm86_fpu_state->sw &= ~0x8000;
+    vm86_fpu_state.sw &= ~0x8000;
 #endif
     break;
   case 0xf1:
@@ -300,8 +300,7 @@ void cpu_setup(void)
   io_dev.handler_name = "Math Coprocessor";
   port_register_handler(io_dev, 0);
 
-  vm86_fpu_state = aligned_alloc(16, sizeof(*vm86_fpu_state));
-  savefpstate(*vm86_fpu_state);
+  savefpstate(vm86_fpu_state);
 #ifdef FE_NOMASK_ENV
   feenableexcept(FE_DIVBYZERO | FE_OVERFLOW);
 #endif

src/base/emu-i386/do_vm86.c

@@ -446,7 +446,7 @@ static int true_vm86(union vm86_union *x)
     int ret;
     uint32_t old_flags = REG(eflags);
 
-    loadfpstate(*vm86_fpu_state);
+    loadfpstate(vm86_fpu_state);
 again:
 #if 0
     ret = vm86(&x->vm86ps);
@@ -464,7 +464,7 @@ static int true_vm86(union vm86_union *x)
      * TODO: check kernel version */
     REG(eflags) |= (old_flags & VIP);
 
-    savefpstate(*vm86_fpu_state);
+    savefpstate(vm86_fpu_state);
     /* there is no real need to save and restore the FPU state of the
        emulator itself: savefpstate (fnsave) also resets the current FPU
        state using fninit; fesetenv then restores trapping of division by

src/base/emu-i386/kvm.c

@@ -1075,6 +1075,7 @@ int kvm_dpmi(cpuctx_t *scp)
       }
 
       if (_trapno == 0x10) {
+#if 0
         struct kvm_fpu fpu;
         ioctl(vcpufd, KVM_GET_FPU, &fpu);
 #ifdef __x86_64__
@@ -1095,9 +1096,10 @@ int kvm_dpmi(cpuctx_t *scp)
         __fpstate->cssel = _cs;
         __fpstate->dataoff = fpu.last_dp;
         __fpstate->datasel = _ds;
+#endif
+        print_exception_info(scp);
 #endif
         dbug_printf("coprocessor exception, calling IRQ13\n");
-//        print_exception_info(scp);
         pic_untrigger(13);
         pic_request(13);
         ret = DPMI_RET_DOSEMU;

src/base/emu-i386/simx86/codegen-x86.h

@@ -39,8 +39,6 @@
 #include "trees.h"
 #include "vgaemu.h"
 
-#define HOST_ARCH_X86
-
 #define TAILSIZE	7
 #define TAILFIX		1
 

src/base/emu-i386/simx86/cpu-emu.c

@@ -523,6 +523,8 @@ static void Reg2Cpu (int mode)
   SetSegReal(SREG(gs),Ofs_GS);
   trans_addr     = LONG_CS + TheCPU.eip;
 
+  /* FPU state is loaded later on demand for JIT, not used for simulator */
+  TheCPU.fpstate = &vm86_fpu_state;
   if (debug_level('e')>1) {
 	if (debug_level('e')==9) e_printf("Reg2Cpu< vm86=%08x dpm=%08x emu=%08x\n%s\n",
 		REG(eflags),get_FLAGS(TheCPU.eflags),TheCPU.eflags,
@@ -567,6 +569,14 @@ void Cpu2Reg (void)
     clear_IF();
   REG(eflags) |= EFLAGS_IF;
 
+  if (TheCPU.fpstate == NULL) {
+    if (!CONFIG_CPUSIM)
+      savefpstate(vm86_fpu_state);
+#ifdef HOST_ARCH_X86
+    fesetenv(&dosemu_fenv);
+#endif
+  }
+
   if (debug_level('e')>1) e_printf("Cpu2Reg< vm86=%08x dpm=%08x emu=%08x\n",
 	REG(eflags),get_FLAGS(TheCPU.eflags),TheCPU.eflags);
 }
@@ -600,9 +610,7 @@ static void Scp2Cpu (cpuctx_t *scp)
   TheCPU.cr2 = _cr2;
   TheCPU.df_increments = (TheCPU.eflags&DF)?0xfcfeff:0x040201;
 
-  /* __fpstate is loaded later on demand for JIT, not used for simulator */
-  TheCPU.fpstate = &TheCPU._fpstate;
-  TheCPU._fpstate = *__fpstate;
+  TheCPU.fpstate = &vm86_fpu_state;
 }
 
 /*
@@ -642,17 +650,17 @@ static void Cpu2Scp (cpuctx_t *scp, int trapno)
    */
   if (!TheCPU.err) _err = 0;		//???
   if (TheCPU.fpstate == NULL) {
-    if (!CONFIG_CPUSIM) {
-      savefpstate(TheCPU._fpstate);
-      *get_fpstate(scp) = TheCPU._fpstate;
-    }
+    if (!CONFIG_CPUSIM)
+      savefpstate(vm86_fpu_state);
     /* there is no real need to save and restore the FPU state of the
        emulator itself: savefpstate (fnsave) also resets the current FPU
        state using fninit; fesetenv then restores trapping of division by
        zero and overflow which is good enough for calling FPU-using
        routines.
     */
+#ifdef HOST_ARCH_X86
     fesetenv(&dosemu_fenv);
+#endif
   }
 
   /* push running flags - same as eflags, RF is cosmetic */
@@ -966,8 +974,6 @@ int e_vm86(void)
   e_sigpa_count = 0;
   mode = ADDR16 | DATA16 | MREALA;
   TheCPU.StackMask = 0x0000ffff;
-  /* FPU state is loaded later on demand for JIT, not used for simulator */
-  TheCPU.fpstate = vm86_fpu_state;
 #ifdef SKIP_VM86_TRACE
   demusav=debug_level('e'); if (debug_level('e')) set_debug_level('e', 1);
 #endif
@@ -1053,11 +1059,6 @@ int e_vm86(void)
   while (retval < 0);
   /* ------ OUTER LOOP -- exit to user level ---------------------- */
 
-  if (TheCPU.fpstate == NULL) {
-    if (!CONFIG_CPUSIM) savefpstate(*vm86_fpu_state);
-    fesetenv(&dosemu_fenv);
-  }
-
   if (debug_level('e')>1)
     e_printf("EMU86: retval=%s\n", retdescs[retval&7]);
 

src/base/emu-i386/simx86/interp.c

@@ -2725,6 +2725,7 @@ intop3b:		{ int op = ArOpsFR[D_MO(opc)];
 				}
 			}
 			b &= 7;
+#ifdef HOST_ARCH_X86
 			if (TheCPU.fpstate) {
 				/* For simulator, only need to mask all
 				   exceptions; for JIT, load emulated FPU state
@@ -2735,6 +2736,7 @@ intop3b:		{ int op = ArOpsFR[D_MO(opc)];
 					loadfpstate(*TheCPU.fpstate);
 				TheCPU.fpstate = NULL;
 			}
+#endif
 			if (sim) {
 			    if (Fp87_op(exop,b)) { TheCPU.err = -96; return P0; }
 			}

src/base/emu-i386/simx86/syncpu.h

@@ -147,18 +147,17 @@ typedef struct {
 	unsigned short TR_SEL;
 	DTR  TR;
 
-	/* if not NULL, points to emulated FPU state
-	   if NULL, emulator uses FPU instructions, so flags that
-	   dosemu needs to restore its own FPU environment. */
-	fpregset_t fpstate;
-
 	void (*stub_read_8)(void);
 	void (*stub_read_16)(void);
 	void (*stub_read_32)(void);
 
 	/* should be moved to TSS once implemented */
 	struct revectored_struct int_revectored;
-	struct _libc_fpstate _fpstate __attribute__((aligned(16)));
+
+	/* if not NULL, points to emulated FPU state
+	   if NULL, emulator uses FPU instructions, so flags that
+	   dosemu needs to restore its own FPU environment. */
+	emu_fpregset_t fpstate;
 } SynCPU;
 
 union _SynCPU {

src/base/init/config.c

@@ -728,7 +728,7 @@ static void read_cpu_info(void)
 #endif
 #ifdef __i386__
         if (cpuflags && (strstr(cpuflags, "fxsr")) &&
-	    sizeof(*vm86_fpu_state) == (112+512)) {
+	    sizeof(vm86_fpu_state) == (112+512)) {
           config.cpufxsr = 1;
 	  if (cpuflags && strstr(cpuflags, "sse"))
 	    config.cpusse = 1;

src/dosext/dpmi/dnative.c

@@ -85,8 +85,9 @@ static void copy_to_dpmi(sigcontext_t *scp, cpuctx_t *s)
   _C(trapno);
   _C(err);
   _C(cr2);
-  if (scp->fpstate)
-    memcpy(scp->fpstate, get_fpstate(s), sizeof(struct _fpstate));
+  static_assert(sizeof(struct _fpstate) == sizeof(vm86_fpu_state),
+		"size mismatch");
+  scp->fpstate = (struct _fpstate *)&vm86_fpu_state;
 }
 
 static void copy_to_emu(cpuctx_t *d, sigcontext_t *scp)
@@ -111,8 +112,11 @@ static void copy_to_emu(cpuctx_t *d, sigcontext_t *scp)
   _D(trapno);
   _D(err);
   _D(cr2);
-  if (scp->fpstate)
-    memcpy(get_fpstate(d), _scp_fpstate, sizeof(*get_fpstate(d)));
+  if (scp->fpstate && scp->fpstate != (struct _fpstate *)&vm86_fpu_state) {
+    static_assert(sizeof(struct _fpstate) == sizeof(vm86_fpu_state),
+		"size mismatch");
+    memcpy(&vm86_fpu_state, scp->fpstate, sizeof(struct _fpstate));
+  }
 }
 
 #if WANT_SIGRETURN_WA

src/dosext/dpmi/dpmi.c

@@ -3997,11 +3997,6 @@ void dpmi_init(void)
   _es	= ES;
   _fs	= 0;
   _gs	= 0;
-#ifdef __linux__
-  /* fpu_state needs to be paragraph aligned for fxrstor/fxsave */
-//  __fpstate = aligned_alloc(16, sizeof(*__fpstate));
-  *__fpstate = *vm86_fpu_state;
-#endif
   NOCARRY;
   rm_to_pm_regs(&DPMI_CLIENT.stack_frame, ~0);
   /* on cpu-emu we work with IOPL=3 to avoid doom work-around */
@@ -4611,7 +4606,6 @@ static void do_dpmi_hlt(cpuctx_t *scp, uint8_t *lina, void *sp)
 	    DPMI_CLIENT.in_dpmi_pm_stack);
 
 	  pm_to_rm_regs(scp, ~0);
-	  old_ctx.fpregs = scp->fpregs;
 	  restore_pm_regs(&old_ctx);
 	  curscp = scp;
 	  scp = &old_ctx;

src/include/cpu.h

@@ -61,22 +61,20 @@ union word {
   struct { Bit8u l, h; } b;
 } MAY_ALIAS;
 
-#ifndef __linux__
 #define __ctx(fld) fld
 #ifdef __x86_64__
 /* taken from glibc, don't blame me :) */
-typedef long long int greg_t;
-struct _libc_fpxreg
+struct __fpxreg
 {
   unsigned short int __ctx(significand)[4];
   unsigned short int __ctx(exponent);
   unsigned short int __glibc_reserved1[3];
 };
-struct _libc_xmmreg
+struct __xmmreg
 {
   __uint32_t	__ctx(element)[4];
 };
-struct _libc_fpstate
+struct emu_fpstate
 {
   /* 64-bit FXSAVE format.  */
   uint16_t		__ctx(cwd);
@@ -87,20 +85,17 @@ struct _libc_fpstate
   uint64_t		__ctx(rdp);
   uint32_t		__ctx(mxcsr);
   uint32_t		__ctx(mxcr_mask);
-  struct _libc_fpxreg	_st[8];
-  struct _libc_xmmreg	_xmm[16];
+  struct __fpxreg	_st[8];
+  struct __xmmreg	_xmm[16];
   uint32_t		__glibc_reserved1[24];
 };
-/* Structure to describe FPU registers.  */
-typedef struct _libc_fpstate *fpregset_t;
 #else
-typedef int greg_t;
-struct _libc_fpreg
+struct __fpreg
 {
   unsigned short int __ctx(significand)[4];
   unsigned short int __ctx(exponent);
 };
-struct _libc_fpstate
+struct emu_fpstate
 {
   unsigned long int __ctx(cw);
   unsigned long int __ctx(sw);
@@ -109,11 +104,12 @@ struct _libc_fpstate
   unsigned long int __ctx(cssel);
   unsigned long int __ctx(dataoff);
   unsigned long int __ctx(datasel);
-  struct _libc_fpreg _st[8];
+  struct __fpreg _st[8];
   unsigned long int __ctx(status);
 };
 #endif
-#endif
+/* Structure to describe FPU registers.  */
+typedef struct emu_fpstate *emu_fpregset_t;
 
 union g_reg {
   greg_t reg;
@@ -245,7 +241,7 @@ static inline dosaddr_t FAR2ADDR(far_t ptr) {
 
 #define peek(seg, off)	(READ_WORD(SEGOFF2LINEAR(seg, off)))
 
-extern fpregset_t vm86_fpu_state;
+extern struct emu_fpstate vm86_fpu_state;
 extern fenv_t dosemu_fenv;
 
 /*
@@ -490,10 +486,9 @@ struct pm_regs {
 	unsigned trapno;
 	unsigned err;
 	unsigned long cr2;
-	struct _libc_fpstate fpregs;
 };
 typedef struct pm_regs cpuctx_t;
-#define REGS_SIZE offsetof(struct pm_regs, fpregs)
+#define REGS_SIZE offsetof(struct pm_regs, trapno)
 
 #define _es     (scp->es)
 #define _ds     (scp->ds)
@@ -518,7 +513,6 @@ typedef struct pm_regs cpuctx_t;
 #define get_trapno(s) ((s)->trapno)
 #define get_err(s)    ((s)->err)
 #define get_cr2(s)    ((s)->cr2)
-#define get_fpstate(s) (&(s)->fpregs)
 #define _edi    get_edi(scp)
 #define _esi    get_esi(scp)
 #define _ebp    get_ebp(scp)
@@ -547,7 +541,6 @@ typedef struct pm_regs cpuctx_t;
 #define _eflags_ (scp->eflags)
 #define _cr2    (scp->cr2)
 #define _trapno (scp->trapno)
-#define __fpstate (&scp->fpregs)
 /* compatibility */
 #define _rdi    _edi
 #define _rsi    _esi