pop ss, mov ss, x and sti disable irqs for the next instruction - began dispatch optimization by adding new x86 cpu 'hidden' flags

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@372 c046a42c-6fe2-441c-8c8c-71466251a162

8 files changed, 109 insertions, 58 deletions

diff --git a/cpu-exec.c b/cpu-exec.c
index 39bb933f51..7899a90862 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -186,7 +186,8 @@ int cpu_exec(CPUState *env1)
 #if defined(TARGET_I386)
                     /* if hardware interrupt pending, we execute it */
                     if ((interrupt_request & CPU_INTERRUPT_HARD) &&
-                        (env->eflags & IF_MASK)) {
+                        (env->eflags & IF_MASK) && 
+                        !(env->hflags & HF_INHIBIT_IRQ_MASK)) {
                         int intno;
                         intno = cpu_x86_get_pic_interrupt(env);
                         if (loglevel) {
@@ -233,21 +234,20 @@ int cpu_exec(CPUState *env1)
 #endif
                 }
 #endif
-                /* we compute the CPU state. We assume it will not
-                   change during the whole generated block. */
+                /* we record a subset of the CPU state. It will
+                   always be the same before a given translated block
+                   is executed. */
 #if defined(TARGET_I386)
                 flags = (env->segs[R_CS].flags & DESC_B_MASK)
-                    >> (DESC_B_SHIFT - GEN_FLAG_CODE32_SHIFT);
+                    >> (DESC_B_SHIFT - HF_CS32_SHIFT);
                 flags |= (env->segs[R_SS].flags & DESC_B_MASK)
-                    >> (DESC_B_SHIFT - GEN_FLAG_SS32_SHIFT);
+                    >> (DESC_B_SHIFT - HF_SS32_SHIFT);
                 flags |= (((unsigned long)env->segs[R_DS].base | 
                            (unsigned long)env->segs[R_ES].base |
                            (unsigned long)env->segs[R_SS].base) != 0) << 
-                    GEN_FLAG_ADDSEG_SHIFT;
-                flags |= env->cpl << GEN_FLAG_CPL_SHIFT;
-                flags |= env->soft_mmu << GEN_FLAG_SOFT_MMU_SHIFT;
-                flags |= (env->eflags & VM_MASK) >> (17 - GEN_FLAG_VM_SHIFT);
-                flags |= (env->eflags & (IOPL_MASK | TF_MASK));
+                    HF_ADDSEG_SHIFT;
+                flags |= env->hflags;
+                flags |= (env->eflags & (IOPL_MASK | TF_MASK | VM_MASK));
                 cs_base = env->segs[R_CS].base;
                 pc = cs_base + env->eip;
 #elif defined(TARGET_ARM)
@@ -337,8 +337,8 @@ int cpu_exec(CPUState *env1)
                 /* reset soft MMU for next block (it can currently
                    only be set by a memory fault) */
 #if defined(TARGET_I386) && !defined(CONFIG_SOFTMMU)
-                if (env->soft_mmu) {
-                    env->soft_mmu = 0;
+                if (env->hflags & HF_SOFTMMU_MASK) {
+                    env->hflags &= ~HF_SOFTMMU_MASK;
                     /* do not allow linking to another block */
                     T0 = 0;
                 }
@@ -499,7 +499,7 @@ static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
         raise_exception_err(EXCP0E_PAGE, env->error_code);
     } else {
         /* activate soft MMU for this block */
-        env->soft_mmu = 1;
+        env->hflags |= HF_SOFTMMU_MASK;
         sigprocmask(SIG_SETMASK, old_set, NULL);
         cpu_loop_exit();
     }
diff --git a/cpu-i386.h b/cpu-i386.h
index a60e959150..ad3a6b801b 100644
--- a/cpu-i386.h
+++ b/cpu-i386.h
@@ -73,6 +73,10 @@
 #define CC_S    0x0080
 #define CC_O    0x0800
 
+#define TF_SHIFT   8
+#define IOPL_SHIFT 12
+#define VM_SHIFT   17
+
 #define TF_MASK 		0x00000100
 #define IF_MASK 		0x00000200
 #define DF_MASK 		0x00000400
@@ -85,6 +89,29 @@
 #define VIP_MASK                0x00100000
 #define ID_MASK                 0x00200000
 
+/* hidden flags - used internally by qemu to represent additionnal cpu
+   states. Only the CPL and INHIBIT_IRQ are not redundant. We avoid
+   using the IOPL_MASK, TF_MASK and VM_MASK bit position to ease oring
+   with eflags. */
+/* current cpl */
+#define HF_CPL_SHIFT         0
+/* true if soft mmu is being used */
+#define HF_SOFTMMU_SHIFT     2
+/* true if hardware interrupts must be disabled for next instruction */
+#define HF_INHIBIT_IRQ_SHIFT 3
+/* 16 or 32 segments */
+#define HF_CS32_SHIFT        4
+#define HF_SS32_SHIFT        5
+/* zero base for DS, ES and SS */
+#define HF_ADDSEG_SHIFT      6
+
+#define HF_CPL_MASK          (3 << HF_CPL_SHIFT)
+#define HF_SOFTMMU_MASK      (1 << HF_SOFTMMU_SHIFT)
+#define HF_INHIBIT_IRQ_MASK  (1 << HF_INHIBIT_IRQ_SHIFT)
+#define HF_CS32_MASK         (1 << HF_CS32_SHIFT)
+#define HF_SS32_MASK         (1 << HF_CS32_SHIFT)
+#define HF_ADDSEG_MASK       (1 << HF_ADDSEG_SHIFT)
+
 #define CR0_PE_MASK  (1 << 0)
 #define CR0_TS_MASK  (1 << 3)
 #define CR0_WP_MASK  (1 << 16)
@@ -226,6 +253,7 @@ typedef struct CPUX86State {
     uint32_t cc_dst;
     uint32_t cc_op;
     int32_t df; /* D flag : 1 if D = 0, -1 if D = 1 */
+    uint32_t hflags; /* hidden flags, see HF_xxx constants */
 
     /* FPU state */
     unsigned int fpstt; /* top of stack index */
@@ -249,8 +277,6 @@ typedef struct CPUX86State {
     SegmentCache tr;
     SegmentCache gdt; /* only base and limit are used */
     SegmentCache idt; /* only base and limit are used */
-    int cpl;          /* current cpl */
-    int soft_mmu;     /* TRUE if soft mmu is being used */
     
     /* sysenter registers */
     uint32_t sysenter_cs;
@@ -303,7 +329,11 @@ void cpu_x86_load_seg(CPUX86State *s, int seg_reg, int selector);
 /* wrapper, just in case memory mappings must be changed */
 static inline void cpu_x86_set_cpl(CPUX86State *s, int cpl)
 {
-    s->cpl = cpl;
+#if HF_CPL_MASK == 3
+    s->hflags = (s->hflags & ~HF_CPL_MASK) | cpl;
+#else
+#error HF_CPL_MASK is hardcoded
+#endif
 }
 
 /* simulate fsave/frstor */
diff --git a/exec.h b/exec.h
index 026daaabe8..b63a66f512 100644
--- a/exec.h
+++ b/exec.h
@@ -61,16 +61,6 @@ extern uint8_t gen_opc_instr_start[OPC_BUF_SIZE];
 
 #if defined(TARGET_I386)
 
-#define GEN_FLAG_CODE32_SHIFT    0
-#define GEN_FLAG_ADDSEG_SHIFT    1
-#define GEN_FLAG_SS32_SHIFT      2
-#define GEN_FLAG_VM_SHIFT        3
-#define GEN_FLAG_ST_SHIFT        4
-#define GEN_FLAG_TF_SHIFT        8 /* same position as eflags */
-#define GEN_FLAG_CPL_SHIFT       9
-#define GEN_FLAG_SOFT_MMU_SHIFT 11
-#define GEN_FLAG_IOPL_SHIFT     12 /* same position as eflags */
-
 void optimize_flags_init(void);
 
 #endif
diff --git a/helper-i386.c b/helper-i386.c
index fdbb885b12..0d128ea448 100644
--- a/helper-i386.c
+++ b/helper-i386.c
@@ -189,7 +189,7 @@ static void do_interrupt_protected(int intno, int is_int, int error_code,
 {
     SegmentCache *dt;
     uint8_t *ptr, *ssp;
-    int type, dpl, selector, ss_dpl;
+    int type, dpl, selector, ss_dpl, cpl;
     int has_error_code, new_stack, shift;
     uint32_t e1, e2, offset, ss, esp, ss_e1, ss_e2, push_size;
     uint32_t old_cs, old_ss, old_esp, old_eip;
@@ -216,8 +216,9 @@ static void do_interrupt_protected(int intno, int is_int, int error_code,
         break;
     }
     dpl = (e2 >> DESC_DPL_SHIFT) & 3;
+    cpl = env->hflags & HF_CPL_MASK;
     /* check privledge if software int */
-    if (is_int && dpl < env->cpl)
+    if (is_int && dpl < cpl)
         raise_exception_err(EXCP0D_GPF, intno * 8 + 2);
     /* check valid bit */
     if (!(e2 & DESC_P_MASK))
@@ -232,11 +233,11 @@ static void do_interrupt_protected(int intno, int is_int, int error_code,
     if (!(e2 & DESC_S_MASK) || !(e2 & (DESC_CS_MASK)))
         raise_exception_err(EXCP0D_GPF, selector & 0xfffc);
     dpl = (e2 >> DESC_DPL_SHIFT) & 3;
-    if (dpl > env->cpl)
+    if (dpl > cpl)
         raise_exception_err(EXCP0D_GPF, selector & 0xfffc);
     if (!(e2 & DESC_P_MASK))
         raise_exception_err(EXCP0B_NOSEG, selector & 0xfffc);
-    if (!(e2 & DESC_C_MASK) && dpl < env->cpl) {
+    if (!(e2 & DESC_C_MASK) && dpl < cpl) {
         /* to inner priviledge */
         get_ss_esp_from_tss(&ss, &esp, dpl);
         if ((ss & 0xfffc) == 0)
@@ -255,7 +256,7 @@ static void do_interrupt_protected(int intno, int is_int, int error_code,
         if (!(ss_e2 & DESC_P_MASK))
             raise_exception_err(EXCP0A_TSS, ss & 0xfffc);
         new_stack = 1;
-    } else if ((e2 & DESC_C_MASK) || dpl == env->cpl) {
+    } else if ((e2 & DESC_C_MASK) || dpl == cpl) {
         /* to same priviledge */
         new_stack = 0;
     } else {
@@ -402,7 +403,7 @@ void do_interrupt_user(int intno, int is_int, int error_code,
 {
     SegmentCache *dt;
     uint8_t *ptr;
-    int dpl;
+    int dpl, cpl;
     uint32_t e2;
 
     dt = &env->idt;
@@ -410,8 +411,9 @@ void do_interrupt_user(int intno, int is_int, int error_code,
     e2 = ldl(ptr + 4);
     
     dpl = (e2 >> DESC_DPL_SHIFT) & 3;
+    cpl = env->hflags & HF_CPL_MASK;
     /* check privledge if software int */
-    if (is_int && dpl < env->cpl)
+    if (is_int && dpl < cpl)
         raise_exception_err(EXCP0D_GPF, intno * 8 + 2);
 
     /* Since we emulate only user space, we cannot do more than
@@ -742,7 +744,7 @@ void helper_ljmp_protected_T0_T1(void)
         raise_exception_err(EXCP0D_GPF, 0);
     if (load_segment(&e1, &e2, new_cs) != 0)
         raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
-    cpl = env->cpl;
+    cpl = env->hflags & HF_CPL_MASK;
     if (e2 & DESC_S_MASK) {
         if (!(e2 & DESC_CS_MASK))
             raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
@@ -826,7 +828,7 @@ void helper_lcall_protected_T0_T1(int shift, int next_eip)
         raise_exception_err(EXCP0D_GPF, 0);
     if (load_segment(&e1, &e2, new_cs) != 0)
         raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
-    cpl = env->cpl;
+    cpl = env->hflags & HF_CPL_MASK;
     if (e2 & DESC_S_MASK) {
         if (!(e2 & DESC_CS_MASK))
             raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
@@ -1079,7 +1081,7 @@ static inline void helper_ret_protected(int shift, int is_iret, int addend)
     if (!(e2 & DESC_S_MASK) ||
         !(e2 & DESC_CS_MASK))
         raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
-    cpl = env->cpl;
+    cpl = env->hflags & HF_CPL_MASK;
     rpl = new_cs & 3; 
     if (rpl < cpl)
         raise_exception_err(EXCP0D_GPF, new_cs & 0xfffc);
diff --git a/helper2-i386.c b/helper2-i386.c
index 7a6fc13b52..f006c72a28 100644
--- a/helper2-i386.c
+++ b/helper2-i386.c
@@ -52,7 +52,7 @@ CPUX86State *cpu_x86_init(void)
 
     tlb_flush(env);
 #ifdef CONFIG_SOFTMMU
-    env->soft_mmu = 1;
+    env->hflags |= HF_SOFTMMU_MASK;
 #endif
     /* init various static tables */
     if (!inited) {
@@ -228,7 +228,7 @@ int cpu_x86_handle_mmu_fault(CPUX86State *env, uint32_t addr, int is_write)
     int cpl, error_code, is_dirty, is_user, prot, page_size, ret;
     unsigned long pd;
     
-    cpl = env->cpl;
+    cpl = env->hflags & HF_CPL_MASK;
     is_user = (cpl == 3);
     
 #ifdef DEBUG_MMU
@@ -325,7 +325,7 @@ int cpu_x86_handle_mmu_fault(CPUX86State *env, uint32_t addr, int is_write)
     }
     
  do_mapping:
-    if (env->soft_mmu) {
+    if (env->hflags & HF_SOFTMMU_MASK) {
         unsigned long paddr, vaddr, address, addend, page_offset;
         int index;
 
@@ -359,7 +359,7 @@ int cpu_x86_handle_mmu_fault(CPUX86State *env, uint32_t addr, int is_write)
     if ((pd & 0xfff) != 0) {
         /* IO access: no mapping is done as it will be handled by the
            soft MMU */
-        if (!env->soft_mmu)
+        if (!(env->hflags & HF_SOFTMMU_MASK))
             ret = 2;
     } else {
         void *map_addr;
diff --git a/op-i386.c b/op-i386.c
index 833b1f9d38..7b0c3917ac 100644
--- a/op-i386.c
+++ b/op-i386.c
@@ -457,6 +457,16 @@ void OPPROTO op_sti(void)
     env->eflags |= IF_MASK;
 }
 
+void OPPROTO op_set_inhibit_irq(void)
+{
+    env->hflags |= HF_INHIBIT_IRQ_MASK;
+}
+
+void OPPROTO op_reset_inhibit_irq(void)
+{
+    env->hflags &= ~HF_INHIBIT_IRQ_MASK;
+}
+
 #if 0
 /* vm86plus instructions */
 void OPPROTO op_cli_vm(void)
diff --git a/softmmu_template.h b/softmmu_template.h
index c36e25ae20..765e913a66 100644
--- a/softmmu_template.h
+++ b/softmmu_template.h
@@ -90,7 +90,7 @@ DATA_TYPE REGPARM(1) glue(glue(__ld, SUFFIX), _mmu)(unsigned long addr)
     
     /* test if there is match for unaligned or IO access */
     /* XXX: could done more in memory macro in a non portable way */
-    is_user = (env->cpl == 3);
+    is_user = ((env->hflags & HF_CPL_MASK) == 3);
     index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
  redo:
     tlb_addr = env->tlb_read[is_user][index].address;
@@ -126,7 +126,7 @@ static DATA_TYPE glue(slow_ld, SUFFIX)(unsigned long addr, void *retaddr)
     int is_user, index, shift;
     unsigned long physaddr, tlb_addr, addr1, addr2;
 
-    is_user = (env->cpl == 3);
+    is_user = ((env->hflags & HF_CPL_MASK) == 3);
     index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
  redo:
     tlb_addr = env->tlb_read[is_user][index].address;
@@ -169,7 +169,7 @@ void REGPARM(2) glue(glue(__st, SUFFIX), _mmu)(unsigned long addr, DATA_TYPE val
     void *retaddr;
     int is_user, index;
     
-    is_user = (env->cpl == 3);
+    is_user = ((env->hflags & HF_CPL_MASK) == 3);
     index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
  redo:
     tlb_addr = env->tlb_write[is_user][index].address;
@@ -203,7 +203,7 @@ static void glue(slow_st, SUFFIX)(unsigned long addr, DATA_TYPE val,
     unsigned long physaddr, tlb_addr;
     int is_user, index, i;
 
-    is_user = (env->cpl == 3);
+    is_user = ((env->hflags & HF_CPL_MASK) == 3);
     index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
  redo:
     tlb_addr = env->tlb_write[is_user][index].address;
diff --git a/translate-i386.c b/translate-i386.c
index e527c901d2..0e72acf199 100644
--- a/translate-i386.c
+++ b/translate-i386.c
@@ -1552,7 +1552,9 @@ static void gen_movl_seg_T0(DisasContext *s, int seg_reg, unsigned int cur_eip)
     else
         gen_op_movl_seg_T0_vm(offsetof(CPUX86State,segs[seg_reg]));
     /* abort translation because the register may have a non zero base
-       or because ss32 may change */
+       or because ss32 may change. For R_SS, translation must always
+       stop as a special handling must be done to disable hardware
+       interrupts for the next instruction */
     if (seg_reg == R_SS || (!s->addseg && seg_reg < R_FS))
         s->is_jmp = 2; 
 }
@@ -2356,10 +2358,14 @@ long disas_insn(DisasContext *s, uint8_t *pc_start)
     case 0x07: /* pop es */
     case 0x17: /* pop ss */
     case 0x1f: /* pop ds */
+        reg = b >> 3;
         gen_pop_T0(s);
-        gen_movl_seg_T0(s, b >> 3, pc_start - s->cs_base);
+        gen_movl_seg_T0(s, reg, pc_start - s->cs_base);
         gen_pop_update(s);
-        /* XXX: if reg == SS, inhibit interrupts/trace */
+        if (reg == R_SS) {
+            /* if reg == SS, inhibit interrupts/trace */
+            gen_op_set_inhibit_irq();
+        }
         break;
     case 0x1a1: /* pop fs */
     case 0x1a9: /* pop gs */
@@ -2418,7 +2424,10 @@ long disas_insn(DisasContext *s, uint8_t *pc_start)
             goto illegal_op;
         gen_ldst_modrm(s, modrm, OT_WORD, OR_TMP0, 0);
         gen_movl_seg_T0(s, reg, pc_start - s->cs_base);
-        /* XXX: if reg == SS, inhibit interrupts/trace */
+        if (reg == R_SS) {
+            /* if reg == SS, inhibit interrupts/trace */
+            gen_op_set_inhibit_irq();
+        }
         break;
     case 0x8c: /* mov Gv, seg */
         modrm = ldub(s->pc++);
@@ -3704,6 +3713,8 @@ long disas_insn(DisasContext *s, uint8_t *pc_start)
         if (!s->vm86) {
             if (s->cpl <= s->iopl) {
                 gen_op_sti();
+                /* interruptions are enabled only the first insn after sti */
+                gen_op_set_inhibit_irq();
                 s->is_jmp = 2; /* give a chance to handle pending irqs */
             } else {
                 gen_exception(s, EXCP0D_GPF, pc_start - s->cs_base);
@@ -3711,12 +3722,13 @@ long disas_insn(DisasContext *s, uint8_t *pc_start)
         } else {
             if (s->iopl == 3) {
                 gen_op_sti();
+                /* interruptions are enabled only the first insn after sti */
+                gen_op_set_inhibit_irq();
                 s->is_jmp = 2; /* give a chance to handle pending irqs */
             } else {
                 gen_exception(s, EXCP0D_GPF, pc_start - s->cs_base);
             }
         }
-        /* XXX: interruptions are enabled only the first insn after sti */
         break;
     case 0x62: /* bound */
         ot = dflag ? OT_LONG : OT_WORD;
@@ -4380,21 +4392,21 @@ static inline int gen_intermediate_code_internal(CPUState *env,
     flags = tb->flags;
        
     dc->pe = env->cr[0] & CR0_PE_MASK;
-    dc->code32 = (flags >> GEN_FLAG_CODE32_SHIFT) & 1;
-    dc->ss32 = (flags >> GEN_FLAG_SS32_SHIFT) & 1;
-    dc->addseg = (flags >> GEN_FLAG_ADDSEG_SHIFT) & 1;
-    dc->f_st = (flags >> GEN_FLAG_ST_SHIFT) & 7;
-    dc->vm86 = (flags >> GEN_FLAG_VM_SHIFT) & 1;
-    dc->cpl = (flags >> GEN_FLAG_CPL_SHIFT) & 3;
-    dc->iopl = (flags >> GEN_FLAG_IOPL_SHIFT) & 3;
-    dc->tf = (flags >> GEN_FLAG_TF_SHIFT) & 1;
+    dc->code32 = (flags >> HF_CS32_SHIFT) & 1;
+    dc->ss32 = (flags >> HF_SS32_SHIFT) & 1;
+    dc->addseg = (flags >> HF_ADDSEG_SHIFT) & 1;
+    dc->f_st = 0;
+    dc->vm86 = (flags >> VM_SHIFT) & 1;
+    dc->cpl = (flags >> HF_CPL_SHIFT) & 3;
+    dc->iopl = (flags >> IOPL_SHIFT) & 3;
+    dc->tf = (flags >> TF_SHIFT) & 1;
     dc->cc_op = CC_OP_DYNAMIC;
     dc->cs_base = cs_base;
     dc->tb = tb;
     dc->popl_esp_hack = 0;
     /* select memory access functions */
     dc->mem_index = 0;
-    if ((flags >> GEN_FLAG_SOFT_MMU_SHIFT) & 1) {
+    if (flags & HF_SOFTMMU_MASK) {
         if (dc->cpl == 3)
             dc->mem_index = 6;
         else
@@ -4408,6 +4420,13 @@ static inline int gen_intermediate_code_internal(CPUState *env,
     dc->is_jmp = DISAS_NEXT;
     pc_ptr = pc_start;
     lj = -1;
+
+    /* if irq were inhibited for the next instruction, we can disable
+       them here as it is simpler (otherwise jumps would have to
+       handled as special case) */
+    if (flags & HF_INHIBIT_IRQ_MASK) {
+        gen_op_reset_inhibit_irq();
+    }
     do {
         if (env->nb_breakpoints > 0) {
             for(j = 0; j < env->nb_breakpoints; j++) {