X-Git-Url: http://wien.tomnetworks.com/gitweb/?a=blobdiff_plain;f=mono%2Fmini%2Ftramp-amd64.c;h=7611b48d46250eddfdea141450036dbe6c82e752;hb=f233a1c472f36dd1d84e3d01de04eca9958262ea;hp=c13da0b1137c9066ff2581e02a18202e676eb0c8;hpb=6f5a0b555436eba699f9dd2659471044097b951b;p=mono.git diff --git a/mono/mini/tramp-amd64.c b/mono/mini/tramp-amd64.c index c13da0b1137..7611b48d462 100644 --- a/mono/mini/tramp-amd64.c +++ b/mono/mini/tramp-amd64.c @@ -18,9 +18,7 @@ #include #include -#ifdef HAVE_VALGRIND_MEMCHECK_H -#include -#endif +#include #include "mini.h" #include "mini-amd64.h" @@ -49,9 +47,7 @@ mono_arch_get_unbox_trampoline (MonoGenericSharingContext *gsctx, MonoMethod *m, this_reg = mono_arch_get_this_arg_reg (mono_method_signature (m), gsctx, NULL); - mono_domain_lock (domain); - start = code = mono_code_manager_reserve (domain->code_mp, 20); - mono_domain_unlock (domain); + start = code = mono_domain_code_reserve (domain, 20); amd64_alu_reg_imm (code, X86_ADD, this_reg, sizeof (MonoObject)); /* FIXME: Optimize this */ @@ -64,6 +60,67 @@ mono_arch_get_unbox_trampoline (MonoGenericSharingContext *gsctx, MonoMethod *m, return start; } +/* + * mono_arch_get_static_rgctx_trampoline: + * + * Create a trampoline which sets RGCTX_REG to MRGCTX, then jumps to ADDR. + */ +gpointer +mono_arch_get_static_rgctx_trampoline (MonoMethod *m, MonoMethodRuntimeGenericContext *mrgctx, gpointer addr) +{ + guint8 *code, *start; + int buf_len; + + MonoDomain *domain = mono_domain_get (); + +#ifdef MONO_ARCH_NOMAP32BIT + buf_len = 32; +#else + /* AOTed code could still have a non-32 bit address */ + if ((((guint64)addr) >> 32) == 0) + buf_len = 16; + else + buf_len = 30; +#endif + + start = code = mono_domain_code_reserve (domain, buf_len); + + amd64_mov_reg_imm (code, MONO_ARCH_RGCTX_REG, mrgctx); + amd64_jump_code (code, addr); + g_assert ((code - start) < buf_len); + + mono_arch_flush_icache (start, code - start); + + return start; +} + +gpointer +mono_arch_get_llvm_imt_trampoline (MonoDomain *domain, MonoMethod *m, int vt_offset) +{ + guint8 *code, *start; + int buf_len; + int this_reg; + + buf_len = 32; + + start = code = mono_domain_code_reserve (domain, buf_len); + + this_reg = mono_arch_get_this_arg_reg (mono_method_signature (m), NULL, NULL); + + /* Set imt arg */ + amd64_mov_reg_imm (code, MONO_ARCH_IMT_REG, m); + /* Load vtable address */ + amd64_mov_reg_membase (code, AMD64_RAX, this_reg, 0, 8); + amd64_jump_membase (code, AMD64_RAX, vt_offset); + amd64_ret (code); + + g_assert ((code - start) < buf_len); + + mono_arch_flush_icache (start, code - start); + + return start; +} + /* * mono_arch_patch_callsite: * @@ -83,12 +140,11 @@ mono_arch_patch_callsite (guint8 *method_start, guint8 *orig_code, guint8 *addr) if (code [-5] != 0xe8) { if (can_write) { InterlockedExchangePointer ((gpointer*)(orig_code - 11), addr); -#ifdef HAVE_VALGRIND_MEMCHECK_H VALGRIND_DISCARD_TRANSLATIONS (orig_code - 11, sizeof (gpointer)); -#endif } } else { if ((((guint64)(addr)) >> 32) != 0) { +#ifdef MONO_ARCH_NOMAP32BIT /* Print some diagnostics */ MonoJitInfo *ji = mono_jit_info_table_find (mono_domain_get (), (char*)orig_code); if (ji) @@ -98,13 +154,24 @@ mono_arch_patch_callsite (guint8 *method_start, guint8 *orig_code, guint8 *addr) if (ji) fprintf (stderr, "Callee: %s\n", mono_method_full_name (ji->method, TRUE)); g_assert_not_reached (); +#else + /* + * This might happen when calling AOTed code. Create a thunk. + */ + guint8 *thunk_start, *thunk_code; + + thunk_start = thunk_code = mono_domain_code_reserve (mono_domain_get (), 32); + amd64_jump_membase (thunk_code, AMD64_RIP, 0); + *(guint64*)thunk_code = (guint64)addr; + addr = thunk_start; + g_assert ((((guint64)(addr)) >> 32) == 0); + mono_arch_flush_icache (thunk_start, thunk_code - thunk_start); +#endif } g_assert ((((guint64)(orig_code)) >> 32) == 0); if (can_write) { InterlockedExchange ((gint32*)(orig_code - 4), ((gint64)addr - (gint64)orig_code)); -#ifdef HAVE_VALGRIND_MEMCHECK_H VALGRIND_DISCARD_TRANSLATIONS (orig_code - 5, 4); -#endif } } } @@ -113,15 +180,13 @@ mono_arch_patch_callsite (guint8 *method_start, guint8 *orig_code, guint8 *addr) gpointer *got_entry = (gpointer*)((guint8*)orig_code + (*(guint32*)(orig_code - 4))); if (can_write) { InterlockedExchangePointer (got_entry, addr); -#ifdef HAVE_VALGRIND_MEMCHECK_H VALGRIND_DISCARD_TRANSLATIONS (orig_code - 5, sizeof (gpointer)); -#endif } } } void -mono_arch_patch_plt_entry (guint8 *code, guint8 *addr) +mono_arch_patch_plt_entry (guint8 *code, gpointer *got, mgreg_t *regs, guint8 *addr) { gint32 disp; gpointer *plt_jump_table_entry; @@ -138,10 +203,19 @@ mono_arch_patch_plt_entry (guint8 *code, guint8 *addr) } void -mono_arch_nullify_class_init_trampoline (guint8 *code, gssize *regs) +mono_arch_nullify_class_init_trampoline (guint8 *code, mgreg_t *regs) { guint8 buf [16]; - gboolean can_write = mono_breakpoint_clean_code (NULL, code, 7, buf, sizeof (buf)); + MonoJitInfo *ji = NULL; + gboolean can_write; + + if (mono_use_llvm) { + /* code - 7 might be before the start of the method */ + /* FIXME: Avoid this expensive call somehow */ + ji = mono_jit_info_table_find (mono_domain_get (), (char*)code); + } + + can_write = mono_breakpoint_clean_code (ji ? ji->code_start : NULL, code, 7, buf, sizeof (buf)); if (!can_write) return; @@ -153,61 +227,57 @@ mono_arch_nullify_class_init_trampoline (guint8 *code, gssize *regs) * really careful about the ordering of the cases. Longer sequences * come first. */ - if ((code [-4] == 0x41) && (code [-3] == 0xff) && (code [-2] == 0x15)) { + if ((buf [0] == 0x41) && (buf [1] == 0xff) && (buf [2] == 0x15)) { gpointer *vtable_slot; /* call *(%rip) */ - vtable_slot = mono_arch_get_vcall_slot_addr (code + 3, (gpointer*)regs); + vtable_slot = mono_get_vcall_slot_addr (code + 3, regs); g_assert (vtable_slot); *vtable_slot = nullified_class_init_trampoline; - } else if (code [-2] == 0xe8) { + } else if (buf [2] == 0xe8) { /* call */ - guint8 *buf = code - 2; - + //guint8 *buf = code - 2; + + /* + * It would be better to replace the call with nops, but that doesn't seem + * to work on SMP machines even when the whole call is inside a cache line. + * Patching the call address seems to work. + */ + /* buf [0] = 0x66; buf [1] = 0x66; buf [2] = 0x90; buf [3] = 0x66; buf [4] = 0x90; - } else if ((code [0] == 0x41) && (code [1] == 0xff)) { - /* call */ - /* happens on machines without MAP_32BIT like freebsd */ - /* amd64_set_reg_template is 10 bytes long */ - guint8* buf = code - 10; - - /* FIXME: Make this thread safe */ - /* Padding code suggested by the AMD64 Opt Manual */ - buf [0] = 0x66; - buf [1] = 0x66; - buf [2] = 0x66; - buf [3] = 0x90; - buf [4] = 0x66; - buf [5] = 0x66; - buf [6] = 0x66; - buf [7] = 0x90; - buf [8] = 0x66; - buf [9] = 0x66; - buf [10] = 0x90; - buf [11] = 0x66; - buf [12] = 0x90; - } else if (code [0] == 0x90 || code [0] == 0xeb || code [0] == 0x66) { + */ + + mono_arch_patch_callsite (code - 2, code - 2 + 5, nullified_class_init_trampoline); + } else if ((buf [5] == 0xff) && x86_modrm_mod (buf [6]) == 3 && x86_modrm_reg (buf [6]) == 2) { + /* call * */ + /* Generated by the LLVM JIT or on platforms without MAP_32BIT set */ + guint8* buf = code; + + /* FIXME: Not thread safe */ + buf [1] = 0x90; + buf [2] = 0x90; + } else if (buf [4] == 0x90 || buf [5] == 0xeb || buf [6] == 0x66) { /* Already changed by another thread */ ; } else { - printf ("Invalid trampoline sequence: %x %x %x %x %x %x %x\n", code [0], code [1], code [2], code [3], - code [4], code [5], code [6]); + printf ("Invalid trampoline sequence: %x %x %x %x %x %x %x\n", buf [0], buf [1], buf [2], buf [3], + buf [4], buf [5], buf [6]); g_assert_not_reached (); } } void -mono_arch_nullify_plt_entry (guint8 *code) +mono_arch_nullify_plt_entry (guint8 *code, mgreg_t *regs) { if (mono_aot_only && !nullified_class_init_trampoline) nullified_class_init_trampoline = mono_aot_get_named_code ("nullified_class_init_trampoline"); - mono_arch_patch_plt_entry (code, nullified_class_init_trampoline); + mono_arch_patch_plt_entry (code, NULL, regs, nullified_class_init_trampoline); } guchar* @@ -215,28 +285,39 @@ mono_arch_create_trampoline_code (MonoTrampolineType tramp_type) { MonoJumpInfo *ji; guint32 code_size; + guchar *code; + GSList *unwind_ops, *l; + + code = mono_arch_create_trampoline_code_full (tramp_type, &code_size, &ji, &unwind_ops, FALSE); + + mono_save_trampoline_xdebug_info ("", code, code_size, unwind_ops); + + for (l = unwind_ops; l; l = l->next) + g_free (l->data); + g_slist_free (unwind_ops); - return mono_arch_create_trampoline_code_full (tramp_type, &code_size, &ji, FALSE); + return code; } guchar* -mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *code_size, MonoJumpInfo **ji, gboolean aot) +mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *code_size, MonoJumpInfo **ji, GSList **out_unwind_ops, gboolean aot) { guint8 *buf, *code, *tramp, *br [2], *r11_save_code, *after_r11_save_code; int i, lmf_offset, offset, res_offset, arg_offset, rax_offset, tramp_offset, saved_regs_offset; - int saved_fpregs_offset, rbp_offset, framesize, orig_rsp_to_rbp_offset; + int saved_fpregs_offset, rbp_offset, framesize, orig_rsp_to_rbp_offset, cfa_offset; gboolean has_caller; + GSList *unwind_ops = NULL; if (tramp_type == MONO_TRAMPOLINE_JUMP) has_caller = FALSE; else has_caller = TRUE; - code = buf = mono_global_codeman_reserve (524); + code = buf = mono_global_codeman_reserve (538); *ji = NULL; - framesize = 524 + sizeof (MonoLMF); + framesize = 538 + sizeof (MonoLMF); framesize = (framesize + (MONO_ARCH_FRAME_ALIGNMENT - 1)) & ~ (MONO_ARCH_FRAME_ALIGNMENT - 1); orig_rsp_to_rbp_offset = 0; @@ -245,16 +326,30 @@ mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *c code += 5; after_r11_save_code = code; + // CFA = sp + 16 (the trampoline address is on the stack) + cfa_offset = 16; + mono_add_unwind_op_def_cfa (unwind_ops, code, buf, AMD64_RSP, 16); + // IP saved at CFA - 8 + mono_add_unwind_op_offset (unwind_ops, code, buf, AMD64_RIP, -8); + /* Pop the return address off the stack */ amd64_pop_reg (code, AMD64_R11); orig_rsp_to_rbp_offset += 8; + cfa_offset -= 8; + mono_add_unwind_op_def_cfa_offset (unwind_ops, code, buf, cfa_offset); + /* * Allocate a new stack frame */ amd64_push_reg (code, AMD64_RBP); + cfa_offset += 8; + mono_add_unwind_op_def_cfa_offset (unwind_ops, code, buf, cfa_offset); + mono_add_unwind_op_offset (unwind_ops, code, buf, AMD64_RBP, - cfa_offset); + orig_rsp_to_rbp_offset -= 8; amd64_mov_reg_reg (code, AMD64_RBP, AMD64_RSP, 8); + mono_add_unwind_op_def_cfa_reg (unwind_ops, code, buf, AMD64_RBP); amd64_alu_reg_imm (code, X86_SUB, AMD64_RSP, framesize); offset = 0; @@ -366,7 +461,7 @@ mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *c amd64_mov_membase_imm (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, method), 0, 8); } /* Save callee saved regs */ -#ifdef PLATFORM_WIN32 +#ifdef TARGET_WIN32 amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rdi), AMD64_RDI, 8); amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rsi), AMD64_RSI, 8); #endif @@ -472,7 +567,7 @@ mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *c amd64_jump_membase (code, AMD64_RSP, rax_offset - 0x8); } - g_assert ((code - buf) <= 524); + g_assert ((code - buf) <= 538); mono_arch_flush_icache (buf, code - buf); @@ -485,6 +580,8 @@ mono_arch_create_trampoline_code_full (MonoTrampolineType tramp_type, guint32 *c nullified_class_init_trampoline = mono_arch_get_nullified_class_init_trampoline (&code_len); } + *out_unwind_ops = unwind_ops; + return buf; } @@ -516,9 +613,7 @@ mono_arch_create_specific_trampoline (gpointer arg1, MonoTrampolineType tramp_ty else size = 5 + 1 + 8; - mono_domain_lock (domain); - code = buf = mono_code_manager_reserve_align (domain->code_mp, size, 1); - mono_domain_unlock (domain); + code = buf = mono_domain_code_reserve_align (domain, size, 1); amd64_call_code (code, tramp); /* The trampoline code will obtain the argument from the instruction stream */ @@ -567,7 +662,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline_full (guint32 slot, guint32 *code_s mrgctx = MONO_RGCTX_SLOT_IS_MRGCTX (slot); index = MONO_RGCTX_SLOT_INDEX (slot); if (mrgctx) - index += sizeof (MonoMethodRuntimeGenericContext) / sizeof (gpointer); + index += MONO_SIZEOF_METHOD_RUNTIME_GENERIC_CONTEXT / sizeof (gpointer); for (depth = 0; ; ++depth) { int size = mono_class_rgctx_get_array_size (depth, mrgctx); @@ -598,7 +693,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline_full (guint32 slot, guint32 *code_s for (i = 0; i < depth; ++i) { /* load ptr to next array */ if (mrgctx && i == 0) - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, sizeof (MonoMethodRuntimeGenericContext), 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, MONO_SIZEOF_METHOD_RUNTIME_GENERIC_CONTEXT, 8); else amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, 0, 8); /* is the ptr null? */ @@ -648,6 +743,15 @@ mono_arch_create_rgctx_lazy_fetch_trampoline_full (guint32 slot, guint32 *code_s gpointer mono_arch_create_generic_class_init_trampoline (void) +{ + guint32 code_size; + MonoJumpInfo *ji; + + return mono_arch_create_generic_class_init_trampoline_full (&code_size, &ji, FALSE); +} + +gpointer +mono_arch_create_generic_class_init_trampoline_full (guint32 *code_size, MonoJumpInfo **ji, gboolean aot) { guint8 *tramp; guint8 *code, *buf; @@ -656,6 +760,8 @@ mono_arch_create_generic_class_init_trampoline (void) guint8 *jump; int tramp_size; + *ji = NULL; + tramp_size = 64; code = buf = mono_global_codeman_reserve (tramp_size); @@ -671,15 +777,23 @@ mono_arch_create_generic_class_init_trampoline (void) x86_patch (jump, code); - tramp = mono_arch_create_specific_trampoline (NULL, MONO_TRAMPOLINE_GENERIC_CLASS_INIT, mono_get_root_domain (), NULL); + if (aot) { + *ji = mono_patch_info_list_prepend (*ji, code - buf, MONO_PATCH_INFO_JIT_ICALL_ADDR, "specific_trampoline_generic_class_init"); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RIP, 0, 8); + amd64_jump_reg (code, AMD64_R11); + } else { + tramp = mono_arch_create_specific_trampoline (NULL, MONO_TRAMPOLINE_GENERIC_CLASS_INIT, mono_get_root_domain (), NULL); - /* jump to the actual trampoline */ - amd64_jump_code (code, tramp); + /* jump to the actual trampoline */ + amd64_jump_code (code, tramp); + } mono_arch_flush_icache (buf, code - buf); g_assert (code - buf <= tramp_size); + *code_size = code - buf; + return buf; } @@ -697,12 +811,15 @@ mono_arch_create_monitor_enter_trampoline (void) gpointer mono_arch_create_monitor_enter_trampoline_full (guint32 *code_size, MonoJumpInfo **ji, gboolean aot) { + guint8 *tramp; guint8 *code, *buf; guint8 *jump_obj_null, *jump_sync_null, *jump_cmpxchg_failed, *jump_other_owner, *jump_tid; int tramp_size; int owner_offset, nest_offset, dummy; + *ji = NULL; + g_assert (MONO_ARCH_MONITOR_OBJECT_REG == AMD64_RDI); mono_monitor_threads_sync_members_offset (&owner_offset, &nest_offset, &dummy); @@ -718,82 +835,82 @@ mono_arch_create_monitor_enter_trampoline_full (guint32 *code_size, MonoJumpInfo if (mono_thread_get_tls_offset () != -1) { /* MonoObject* obj is in RDI */ /* is obj null? */ - amd64_test_reg_reg (buf, AMD64_RDI, AMD64_RDI); + amd64_test_reg_reg (code, AMD64_RDI, AMD64_RDI); /* if yes, jump to actual trampoline */ - jump_obj_null = buf; - amd64_branch8 (buf, X86_CC_Z, -1, 1); + jump_obj_null = code; + amd64_branch8 (code, X86_CC_Z, -1, 1); /* load obj->synchronization to RCX */ - amd64_mov_reg_membase (buf, AMD64_RCX, AMD64_RDI, G_STRUCT_OFFSET (MonoObject, synchronisation), 8); + amd64_mov_reg_membase (code, AMD64_RCX, AMD64_RDI, G_STRUCT_OFFSET (MonoObject, synchronisation), 8); /* is synchronization null? */ - amd64_test_reg_reg (buf, AMD64_RCX, AMD64_RCX); + amd64_test_reg_reg (code, AMD64_RCX, AMD64_RCX); /* if yes, jump to actual trampoline */ - jump_sync_null = buf; - amd64_branch8 (buf, X86_CC_Z, -1, 1); + jump_sync_null = code; + amd64_branch8 (code, X86_CC_Z, -1, 1); - /* load MonoThread* into RDX */ - buf = mono_amd64_emit_tls_get (buf, AMD64_RDX, mono_thread_get_tls_offset ()); + /* load MonoInternalThread* into RDX */ + code = mono_amd64_emit_tls_get (code, AMD64_RDX, mono_thread_get_tls_offset ()); /* load TID into RDX */ - amd64_mov_reg_membase (buf, AMD64_RDX, AMD64_RDX, G_STRUCT_OFFSET (MonoThread, tid), 8); + amd64_mov_reg_membase (code, AMD64_RDX, AMD64_RDX, G_STRUCT_OFFSET (MonoInternalThread, tid), 8); /* is synchronization->owner null? */ - amd64_alu_membase_imm_size (buf, X86_CMP, AMD64_RCX, owner_offset, 0, 8); + amd64_alu_membase_imm_size (code, X86_CMP, AMD64_RCX, owner_offset, 0, 8); /* if not, jump to next case */ - jump_tid = buf; - amd64_branch8 (buf, X86_CC_NZ, -1, 1); + jump_tid = code; + amd64_branch8 (code, X86_CC_NZ, -1, 1); /* if yes, try a compare-exchange with the TID */ /* zero RAX */ - amd64_alu_reg_reg (buf, X86_XOR, AMD64_RAX, AMD64_RAX); + amd64_alu_reg_reg (code, X86_XOR, AMD64_RAX, AMD64_RAX); /* compare and exchange */ - amd64_prefix (buf, X86_LOCK_PREFIX); - amd64_cmpxchg_membase_reg_size (buf, AMD64_RCX, owner_offset, AMD64_RDX, 8); + amd64_prefix (code, X86_LOCK_PREFIX); + amd64_cmpxchg_membase_reg_size (code, AMD64_RCX, owner_offset, AMD64_RDX, 8); /* if not successful, jump to actual trampoline */ - jump_cmpxchg_failed = buf; - amd64_branch8 (buf, X86_CC_NZ, -1, 1); + jump_cmpxchg_failed = code; + amd64_branch8 (code, X86_CC_NZ, -1, 1); /* if successful, return */ - amd64_ret (buf); + amd64_ret (code); /* next case: synchronization->owner is not null */ - x86_patch (jump_tid, buf); + x86_patch (jump_tid, code); /* is synchronization->owner == TID? */ - amd64_alu_membase_reg_size (buf, X86_CMP, AMD64_RCX, owner_offset, AMD64_RDX, 8); + amd64_alu_membase_reg_size (code, X86_CMP, AMD64_RCX, owner_offset, AMD64_RDX, 8); /* if not, jump to actual trampoline */ - jump_other_owner = buf; - amd64_branch8 (buf, X86_CC_NZ, -1, 1); + jump_other_owner = code; + amd64_branch8 (code, X86_CC_NZ, -1, 1); /* if yes, increment nest */ - amd64_inc_membase_size (buf, AMD64_RCX, nest_offset, 4); + amd64_inc_membase_size (code, AMD64_RCX, nest_offset, 4); /* return */ - amd64_ret (buf); + amd64_ret (code); - x86_patch (jump_obj_null, buf); - x86_patch (jump_sync_null, buf); - x86_patch (jump_cmpxchg_failed, buf); - x86_patch (jump_other_owner, buf); + x86_patch (jump_obj_null, code); + x86_patch (jump_sync_null, code); + x86_patch (jump_cmpxchg_failed, code); + x86_patch (jump_other_owner, code); } /* jump to the actual trampoline */ #if MONO_AMD64_ARG_REG1 != AMD64_RDI - amd64_mov_reg_reg (buf, MONO_AMD64_ARG_REG1, AMD64_RDI); + amd64_mov_reg_reg (code, MONO_AMD64_ARG_REG1, AMD64_RDI); #endif if (aot) { - *ji = mono_patch_info_list_prepend (*ji, buf - code, MONO_PATCH_INFO_JIT_ICALL_ADDR, "specific_trampoline_monitor_enter"); - amd64_mov_reg_membase (buf, AMD64_R11, AMD64_RIP, 0, 8); - amd64_jump_reg (buf, AMD64_R11); + *ji = mono_patch_info_list_prepend (*ji, code - buf, MONO_PATCH_INFO_JIT_ICALL_ADDR, "specific_trampoline_monitor_enter"); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RIP, 0, 8); + amd64_jump_reg (code, AMD64_R11); } else { tramp = mono_arch_create_specific_trampoline (NULL, MONO_TRAMPOLINE_MONITOR_ENTER, mono_get_root_domain (), NULL); /* jump to the actual trampoline */ - amd64_jump_code (buf, tramp); + amd64_jump_code (code, tramp); } - mono_arch_flush_icache (buf, buf - code); - g_assert (buf - code <= tramp_size); + mono_arch_flush_icache (code, code - buf); + g_assert (code - buf <= tramp_size); - *code_size = buf - code; + *code_size = code - buf; - return code; + return buf; } gpointer @@ -815,6 +932,8 @@ mono_arch_create_monitor_exit_trampoline_full (guint32 *code_size, MonoJumpInfo int tramp_size; int owner_offset, nest_offset, entry_count_offset; + *ji = NULL; + g_assert (MONO_ARCH_MONITOR_OBJECT_REG == AMD64_RDI); mono_monitor_threads_sync_members_offset (&owner_offset, &nest_offset, &entry_count_offset); @@ -832,81 +951,81 @@ mono_arch_create_monitor_exit_trampoline_full (guint32 *code_size, MonoJumpInfo if (mono_thread_get_tls_offset () != -1) { /* MonoObject* obj is in RDI */ /* is obj null? */ - amd64_test_reg_reg (buf, AMD64_RDI, AMD64_RDI); + amd64_test_reg_reg (code, AMD64_RDI, AMD64_RDI); /* if yes, jump to actual trampoline */ - jump_obj_null = buf; - amd64_branch8 (buf, X86_CC_Z, -1, 1); + jump_obj_null = code; + amd64_branch8 (code, X86_CC_Z, -1, 1); /* load obj->synchronization to RCX */ - amd64_mov_reg_membase (buf, AMD64_RCX, AMD64_RDI, G_STRUCT_OFFSET (MonoObject, synchronisation), 8); + amd64_mov_reg_membase (code, AMD64_RCX, AMD64_RDI, G_STRUCT_OFFSET (MonoObject, synchronisation), 8); /* is synchronization null? */ - amd64_test_reg_reg (buf, AMD64_RCX, AMD64_RCX); + amd64_test_reg_reg (code, AMD64_RCX, AMD64_RCX); /* if not, jump to next case */ - jump_next = buf; - amd64_branch8 (buf, X86_CC_NZ, -1, 1); + jump_next = code; + amd64_branch8 (code, X86_CC_NZ, -1, 1); /* if yes, just return */ - amd64_ret (buf); + amd64_ret (code); /* next case: synchronization is not null */ - x86_patch (jump_next, buf); - /* load MonoThread* into RDX */ - buf = mono_amd64_emit_tls_get (buf, AMD64_RDX, mono_thread_get_tls_offset ()); + x86_patch (jump_next, code); + /* load MonoInternalThread* into RDX */ + code = mono_amd64_emit_tls_get (code, AMD64_RDX, mono_thread_get_tls_offset ()); /* load TID into RDX */ - amd64_mov_reg_membase (buf, AMD64_RDX, AMD64_RDX, G_STRUCT_OFFSET (MonoThread, tid), 8); + amd64_mov_reg_membase (code, AMD64_RDX, AMD64_RDX, G_STRUCT_OFFSET (MonoInternalThread, tid), 8); /* is synchronization->owner == TID */ - amd64_alu_membase_reg_size (buf, X86_CMP, AMD64_RCX, owner_offset, AMD64_RDX, 8); + amd64_alu_membase_reg_size (code, X86_CMP, AMD64_RCX, owner_offset, AMD64_RDX, 8); /* if yes, jump to next case */ - jump_next = buf; - amd64_branch8 (buf, X86_CC_Z, -1, 1); + jump_next = code; + amd64_branch8 (code, X86_CC_Z, -1, 1); /* if not, just return */ - amd64_ret (buf); + amd64_ret (code); /* next case: synchronization->owner == TID */ - x86_patch (jump_next, buf); + x86_patch (jump_next, code); /* is synchronization->nest == 1 */ - amd64_alu_membase_imm_size (buf, X86_CMP, AMD64_RCX, nest_offset, 1, 4); + amd64_alu_membase_imm_size (code, X86_CMP, AMD64_RCX, nest_offset, 1, 4); /* if not, jump to next case */ - jump_next = buf; - amd64_branch8 (buf, X86_CC_NZ, -1, 1); + jump_next = code; + amd64_branch8 (code, X86_CC_NZ, -1, 1); /* if yes, is synchronization->entry_count zero? */ - amd64_alu_membase_imm_size (buf, X86_CMP, AMD64_RCX, entry_count_offset, 0, 4); + amd64_alu_membase_imm_size (code, X86_CMP, AMD64_RCX, entry_count_offset, 0, 4); /* if not, jump to actual trampoline */ - jump_have_waiters = buf; - amd64_branch8 (buf, X86_CC_NZ, -1 , 1); + jump_have_waiters = code; + amd64_branch8 (code, X86_CC_NZ, -1 , 1); /* if yes, set synchronization->owner to null and return */ - amd64_mov_membase_imm (buf, AMD64_RCX, owner_offset, 0, 8); - amd64_ret (buf); + amd64_mov_membase_imm (code, AMD64_RCX, owner_offset, 0, 8); + amd64_ret (code); /* next case: synchronization->nest is not 1 */ - x86_patch (jump_next, buf); + x86_patch (jump_next, code); /* decrease synchronization->nest and return */ - amd64_dec_membase_size (buf, AMD64_RCX, nest_offset, 4); - amd64_ret (buf); + amd64_dec_membase_size (code, AMD64_RCX, nest_offset, 4); + amd64_ret (code); - x86_patch (jump_obj_null, buf); - x86_patch (jump_have_waiters, buf); + x86_patch (jump_obj_null, code); + x86_patch (jump_have_waiters, code); } /* jump to the actual trampoline */ #if MONO_AMD64_ARG_REG1 != AMD64_RDI - amd64_mov_reg_reg (buf, MONO_AMD64_ARG_REG1, AMD64_RDI); + amd64_mov_reg_reg (code, MONO_AMD64_ARG_REG1, AMD64_RDI); #endif if (aot) { - *ji = mono_patch_info_list_prepend (*ji, buf - code, MONO_PATCH_INFO_JIT_ICALL_ADDR, "specific_trampoline_monitor_exit"); - amd64_mov_reg_membase (buf, AMD64_R11, AMD64_RIP, 0, 8); - amd64_jump_reg (buf, AMD64_R11); + *ji = mono_patch_info_list_prepend (*ji, code - buf, MONO_PATCH_INFO_JIT_ICALL_ADDR, "specific_trampoline_monitor_exit"); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RIP, 0, 8); + amd64_jump_reg (code, AMD64_R11); } else { tramp = mono_arch_create_specific_trampoline (NULL, MONO_TRAMPOLINE_MONITOR_EXIT, mono_get_root_domain (), NULL); - amd64_jump_code (buf, tramp); + amd64_jump_code (code, tramp); } - mono_arch_flush_icache (buf, buf - code); - g_assert (buf - code <= tramp_size); + mono_arch_flush_icache (code, code - buf); + g_assert (code - buf <= tramp_size); - *code_size = buf - code; + *code_size = code - buf; - return code; + return buf; } #endif