converted more helpers to TCG - fixed some SVM issues

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@4459 c046a42c-6fe2-441c-8c8c-71466251a162

target-i386/TODO

@@ -3,12 +3,13 @@ Correctness issues:
 - rework eflags optimization (will be a consequence of TCG port)
 - SVM: rework the implementation: simplify code, move most intercept
   tests as dynamic, correct segment access, verify exception safety,
-  remove most of the added CPU state.
+  cpu save/restore, SMM save/restore. 
 - arpl eflags computation is invalid
 - x86_64: fxsave/fxrestore intel/amd differences
 - x86_64: lcall/ljmp intel/amd differences ?
 - x86_64: cmpxchgl intel/amd differences ?
-- x86_64: cmovl bug intel/amd differences ?
+- x86_64: cmovl intel/amd differences ?
+- cmpxchg16b + cmpxchg8b cpuid test
 - x86: monitor invalid 
 - better code fetch (different exception handling + CS.limit support)
 - user/kernel PUSHL/POPL in helper.c
@@ -19,10 +20,18 @@ Correctness issues:
 - full support of segment limit/rights 
 - full x87 exception support
 - improve x87 bit exactness (use bochs code ?)
+- DRx register support
+- CR0.AC emulation
+- SSE alignment checks
+- fix SSE min/max with nans
 
 Optimizations/Features:
 
 - finish TCG port
+- add SVM nested paging support
+- add VMX support
+- add AVX support
+- add SSE5 support
 - evaluate x87 stack pointer statically
 - find a way to avoid translating several time the same TB if CR0.TS
   is set or not.

target-i386/exec.h

@@ -105,16 +105,6 @@ typedef struct CCTable {
 
 extern CCTable cc_table[];
 
-void helper_load_seg(int seg_reg, int selector);
-void helper_ljmp_protected_T0_T1(int next_eip);
-void helper_lcall_real_T0_T1(int shift, int next_eip);
-void helper_lcall_protected_T0_T1(int shift, int next_eip);
-void helper_iret_real(int shift);
-void helper_iret_protected(int shift, int next_eip);
-void helper_lret_protected(int shift, int addend);
-void helper_movl_crN_T0(int reg);
-void helper_movl_drN_T0(int reg);
-void helper_invlpg(target_ulong addr);
 void cpu_x86_update_cr0(CPUX86State *env, uint32_t new_cr0);
 void cpu_x86_update_cr3(CPUX86State *env, target_ulong new_cr3);
 void cpu_x86_update_cr4(CPUX86State *env, uint32_t new_cr4);
@@ -141,17 +131,10 @@ void OPPROTO op_movl_T0_eflags(void);
 
 #include "helper.h"
 
-void helper_mulq_EAX_T0(void);
-void helper_imulq_EAX_T0(void);
-void helper_imulq_T0_T1(void);
-void helper_cmpxchg8b(void);
-
-void check_iob_T0(void);
-void check_iow_T0(void);
-void check_iol_T0(void);
-void check_iob_DX(void);
-void check_iow_DX(void);
-void check_iol_DX(void);
+static inline void svm_check_intercept(uint32_t type)
+{
+    helper_svm_check_intercept_param(type, 0);
+}
 
 #if !defined(CONFIG_USER_ONLY)
 
@@ -363,7 +346,6 @@ extern const CPU86_LDouble f15rk[7];
 void fpu_raise_exception(void);
 void restore_native_fp_state(CPUState *env);
 void save_native_fp_state(CPUState *env);
-void vmexit(uint64_t exit_code, uint64_t exit_info_1);
 
 extern const uint8_t parity_table[256];
 extern const uint8_t rclw_table[32];

target-i386/helper.h

@@ -1,5 +1,7 @@
 #define TCG_HELPER_PROTO
 
+void helper_lock(void);
+void helper_unlock(void);
 void helper_divb_AL(target_ulong t0);
 void helper_idivb_AL(target_ulong t0);
 void helper_divw_AX(target_ulong t0);
@@ -7,6 +9,9 @@ void helper_idivw_AX(target_ulong t0);
 void helper_divl_EAX(target_ulong t0);
 void helper_idivl_EAX(target_ulong t0);
 #ifdef TARGET_X86_64
+void helper_mulq_EAX_T0(target_ulong t0);
+void helper_imulq_EAX_T0(target_ulong t0);
+target_ulong helper_imulq_T0_T1(target_ulong t0, target_ulong t1);
 void helper_divq_EAX(target_ulong t0);
 void helper_idivq_EAX(target_ulong t0);
 #endif
@@ -18,26 +23,34 @@ void helper_aas(void);
 void helper_daa(void);
 void helper_das(void);
 
-void helper_lsl(uint32_t selector);
-void helper_lar(uint32_t selector);
+uint32_t helper_lsl(uint32_t selector);
+uint32_t helper_lar(uint32_t selector);
 void helper_verr(uint32_t selector);
 void helper_verw(uint32_t selector);
 void helper_lldt(int selector);
 void helper_ltr(int selector);
 void helper_load_seg(int seg_reg, int selector);
-void helper_ljmp_protected_T0_T1(int next_eip);
-void helper_lcall_real_T0_T1(int shift, int next_eip);
-void helper_lcall_protected_T0_T1(int shift, int next_eip);
+void helper_ljmp_protected(int new_cs, target_ulong new_eip,
+                           int next_eip_addend);
+void helper_lcall_real(int new_cs, target_ulong new_eip1,
+                       int shift, int next_eip);
+void helper_lcall_protected(int new_cs, target_ulong new_eip, 
+                            int shift, int next_eip_addend);
 void helper_iret_real(int shift);
 void helper_iret_protected(int shift, int next_eip);
 void helper_lret_protected(int shift, int addend);
-void helper_movl_crN_T0(int reg);
-void helper_movl_drN_T0(int reg);
+void helper_movl_crN_T0(int reg, target_ulong t0);
+void helper_lmsw(target_ulong t0);
+void helper_clts(void);
+#if !defined(CONFIG_USER_ONLY)
+target_ulong helper_movtl_T0_cr8(void);
+#endif
+void helper_movl_drN_T0(int reg, target_ulong t0);
 void helper_invlpg(target_ulong addr);
 
-void helper_enter_level(int level, int data32);
+void helper_enter_level(int level, int data32, target_ulong t1);
 #ifdef TARGET_X86_64
-void helper_enter64_level(int level, int data64);
+void helper_enter64_level(int level, int data64, target_ulong t1);
 #endif
 void helper_sysenter(void);
 void helper_sysexit(void);
@@ -55,9 +68,10 @@ void helper_cli(void);
 void helper_sti(void);
 void helper_set_inhibit_irq(void);
 void helper_reset_inhibit_irq(void);
-void helper_boundw(void);
-void helper_boundl(void);
+void helper_boundw(target_ulong a0, int v);
+void helper_boundl(target_ulong a0, int v);
 void helper_rsm(void);
+void helper_cmpxchg8b(target_ulong a0);
 void helper_single_step(void);
 void helper_cpuid(void);
 void helper_rdtsc(void);
@@ -65,6 +79,20 @@ void helper_rdpmc(void);
 void helper_rdmsr(void);
 void helper_wrmsr(void);
 
+void helper_check_iob(uint32_t t0);
+void helper_check_iow(uint32_t t0);
+void helper_check_iol(uint32_t t0);
+void helper_outb(uint32_t port, uint32_t data);
+target_ulong helper_inb(uint32_t port);
+void helper_outw(uint32_t port, uint32_t data);
+target_ulong helper_inw(uint32_t port);
+void helper_outl(uint32_t port, uint32_t data);
+target_ulong helper_inl(uint32_t port);
+
+void helper_svm_check_intercept_param(uint32_t type, uint64_t param);
+void helper_vmexit(uint32_t exit_code, uint64_t exit_info_1);
+void helper_svm_check_io(uint32_t port, uint32_t param, 
+                         uint32_t next_eip_addend);
 void helper_vmrun(void);
 void helper_vmmcall(void);
 void helper_vmload(void);

target-i386/op.c

@@ -276,17 +276,17 @@ void OPPROTO op_imull_T0_T1(void)
 #ifdef TARGET_X86_64
 void OPPROTO op_mulq_EAX_T0(void)
 {
-    helper_mulq_EAX_T0();
+    helper_mulq_EAX_T0(T0);
 }
 
 void OPPROTO op_imulq_EAX_T0(void)
 {
-    helper_imulq_EAX_T0();
+    helper_imulq_EAX_T0(T0);
 }
 
 void OPPROTO op_imulq_T0_T1(void)
 {
-    helper_imulq_T0_T1();
+    T0 = helper_imulq_T0_T1(T0, T1);
 }
 #endif
 
@@ -351,7 +351,7 @@ void OPPROTO op_into(void)
 
 void OPPROTO op_cmpxchg8b(void)
 {
-    helper_cmpxchg8b();
+    helper_cmpxchg8b(A0);
 }
 
 /* multiple size ops */
@@ -522,12 +522,6 @@ void OPPROTO op_das(void)
 
 /* segment handling */
 
-/* never use it with R_CS */
-void OPPROTO op_movl_seg_T0(void)
-{
-    helper_load_seg(PARAM1, T0);
-}
-
 /* faster VM86 version */
 void OPPROTO op_movl_seg_T0_vm(void)
 {
@@ -548,12 +542,20 @@ void OPPROTO op_movl_T0_seg(void)
 
 void OPPROTO op_lsl(void)
 {
-    helper_lsl(T0);
+    uint32_t val;
+    val = helper_lsl(T0);
+    if (CC_SRC & CC_Z)
+        T1 = val;
+    FORCE_RET();
 }
 
 void OPPROTO op_lar(void)
 {
-    helper_lar(T0);
+    uint32_t val;
+    val = helper_lar(T0);
+    if (CC_SRC & CC_Z)
+        T1 = val;
+    FORCE_RET();
 }
 
 void OPPROTO op_verr(void)
@@ -585,104 +587,6 @@ void OPPROTO op_arpl_update(void)
     CC_SRC = (eflags & ~CC_Z) | T1;
 }
 
-/* T0: segment, T1:eip */
-void OPPROTO op_ljmp_protected_T0_T1(void)
-{
-    helper_ljmp_protected_T0_T1(PARAM1);
-}
-
-void OPPROTO op_lcall_real_T0_T1(void)
-{
-    helper_lcall_real_T0_T1(PARAM1, PARAM2);
-}
-
-void OPPROTO op_lcall_protected_T0_T1(void)
-{
-    helper_lcall_protected_T0_T1(PARAM1, PARAM2);
-}
-
-void OPPROTO op_iret_real(void)
-{
-    helper_iret_real(PARAM1);
-}
-
-void OPPROTO op_iret_protected(void)
-{
-    helper_iret_protected(PARAM1, PARAM2);
-}
-
-void OPPROTO op_lret_protected(void)
-{
-    helper_lret_protected(PARAM1, PARAM2);
-}
-
-/* CR registers access. */
-void OPPROTO op_movl_crN_T0(void)
-{
-    helper_movl_crN_T0(PARAM1);
-}
-
-/* These pseudo-opcodes check for SVM intercepts. */
-void OPPROTO op_svm_check_intercept(void)
-{
-    A0 = PARAM1 & PARAM2;
-    svm_check_intercept(PARAMQ1);
-}
-
-void OPPROTO op_svm_check_intercept_param(void)
-{
-    A0 = PARAM1 & PARAM2;
-    svm_check_intercept_param(PARAMQ1, T1);
-}
-
-void OPPROTO op_svm_vmexit(void)
-{
-    A0 = PARAM1 & PARAM2;
-    vmexit(PARAMQ1, T1);
-}
-
-void OPPROTO op_geneflags(void)
-{
-    CC_SRC = cc_table[CC_OP].compute_all();
-}
-
-/* This pseudo-opcode checks for IO intercepts. */
-#if !defined(CONFIG_USER_ONLY)
-void OPPROTO op_svm_check_intercept_io(void)
-{
-    A0 = PARAM1 & PARAM2;
-    /* PARAMQ1 = TYPE (0 = OUT, 1 = IN; 4 = STRING; 8 = REP)
-       T0      = PORT
-       T1      = next eip */
-    stq_phys(env->vm_vmcb + offsetof(struct vmcb, control.exit_info_2), T1);
-    /* ASIZE does not appear on real hw */
-    svm_check_intercept_param(SVM_EXIT_IOIO,
-                              (PARAMQ1 & ~SVM_IOIO_ASIZE_MASK) |
-                              ((T0 & 0xffff) << 16));
-}
-#endif
-
-#if !defined(CONFIG_USER_ONLY)
-void OPPROTO op_movtl_T0_cr8(void)
-{
-    T0 = cpu_get_apic_tpr(env);
-}
-#endif
-
-/* DR registers access */
-void OPPROTO op_movl_drN_T0(void)
-{
-    helper_movl_drN_T0(PARAM1);
-}
-
-void OPPROTO op_lmsw_T0(void)
-{
-    /* only 4 lower bits of CR0 are modified. PE cannot be set to zero
-       if already set to one. */
-    T0 = (env->cr[0] & ~0xe) | (T0 & 0xf);
-    helper_movl_crN_T0(0);
-}
-
 void OPPROTO op_movl_T0_env(void)
 {
     T0 = *(uint32_t *)((char *)env + PARAM1);
@@ -718,12 +622,6 @@ void OPPROTO op_movtl_env_T1(void)
     *(target_ulong *)((char *)env + PARAM1) = T1;
 }
 
-void OPPROTO op_clts(void)
-{
-    env->cr[0] &= ~CR0_TS_MASK;
-    env->hflags &= ~HF_TS_MASK;
-}
-
 /* flags handling */
 
 void OPPROTO op_jmp_label(void)
@@ -1028,17 +926,6 @@ void OPPROTO op_fcomi_dummy(void)
     T0 = 0;
 }
 
-/* threading support */
-void OPPROTO op_lock(void)
-{
-    cpu_lock();
-}
-
-void OPPROTO op_unlock(void)
-{
-    cpu_unlock();
-}
-
 /* SSE support */
 void OPPROTO op_com_dummy(void)
 {

target-i386/ops_sse.h

@@ -471,12 +471,12 @@ void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s)
 #endif
 }
 
-void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s)
+void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0)
 {
     int i;
     for(i = 0; i < (8 << SHIFT); i++) {
         if (s->B(i) & 0x80)
-            stb(A0 + i, d->B(i));
+            stb(a0 + i, d->B(i));
     }
     FORCE_RET();
 }

target-i386/ops_sse_header.h

@@ -104,7 +104,7 @@ void glue(helper_pmuludq, SUFFIX) (Reg *d, Reg *s);
 void glue(helper_pmaddwd, SUFFIX) (Reg *d, Reg *s);
 
 void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s);
-void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s);
+void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0);
 void glue(helper_movl_mm_T0, SUFFIX) (Reg *d, uint32_t val);
 #ifdef TARGET_X86_64
 void glue(helper_movq_mm_T0, SUFFIX) (Reg *d, uint64_t val);

target-i386/ops_template.h

@@ -554,39 +554,6 @@ void OPPROTO glue(op_movl_T0_Dshift, SUFFIX)(void)
     T0 = DF << SHIFT;
 }
 
-/* port I/O */
-#if DATA_BITS <= 32
-void OPPROTO glue(glue(op_out, SUFFIX), _T0_T1)(void)
-{
-    glue(cpu_out, SUFFIX)(env, T0, T1 & DATA_MASK);
-}
-
-void OPPROTO glue(glue(op_in, SUFFIX), _T0_T1)(void)
-{
-    T1 = glue(cpu_in, SUFFIX)(env, T0);
-}
-
-void OPPROTO glue(glue(op_in, SUFFIX), _DX_T0)(void)
-{
-    T0 = glue(cpu_in, SUFFIX)(env, EDX & 0xffff);
-}
-
-void OPPROTO glue(glue(op_out, SUFFIX), _DX_T0)(void)
-{
-    glue(cpu_out, SUFFIX)(env, EDX & 0xffff, T0);
-}
-
-void OPPROTO glue(glue(op_check_io, SUFFIX), _T0)(void)
-{
-    glue(glue(check_io, SUFFIX), _T0)();
-}
-
-void OPPROTO glue(glue(op_check_io, SUFFIX), _DX)(void)
-{
-    glue(glue(check_io, SUFFIX), _DX)();
-}
-#endif
-
 #undef DATA_BITS
 #undef SHIFT_MASK
 #undef SHIFT1_MASK

target-i386/svm.h

@@ -71,8 +71,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 	uint32_t int_vector;
 	uint32_t int_state;
 	uint8_t reserved_3[4];
-	uint32_t exit_code;
-	uint32_t exit_code_hi;
+	uint64_t exit_code;
 	uint64_t exit_info_1;
 	uint64_t exit_info_2;
 	uint32_t exit_int_info;
@@ -323,14 +322,6 @@ struct __attribute__ ((__packed__)) vmcb {
 
 /* function references */
 
-void helper_stgi(void);
-void vmexit(uint64_t exit_code, uint64_t exit_info_1);
-int svm_check_intercept_param(uint32_t type, uint64_t param);
-static inline int svm_check_intercept(unsigned int type) {
-    return svm_check_intercept_param(type, 0);
-}
-
-
 #define INTERCEPTED(mask) (env->intercept & mask)
 #define INTERCEPTEDw(var, mask) (env->intercept ## var & mask)
 #define INTERCEPTEDl(var, mask) (env->intercept ## var & mask)