X-Git-Url: http://wien.tomnetworks.com/gitweb/?a=blobdiff_plain;f=mono%2Fmini%2Ftramp-amd64.c;h=f576af9496cc3fbed04f3b5b283032ce5c03a5e3;hb=9342e7d1d4e5d0d3248de33d05b2c2b993314bd6;hp=cd42591d8769c0b6bfc61d79b41d996410ffc3ab;hpb=c9658fe44f0a22c0b1d3452466f129a0b7985d11;p=mono.git diff --git a/mono/mini/tramp-amd64.c b/mono/mini/tramp-amd64.c index cd42591d876..f576af9496c 100644 --- a/mono/mini/tramp-amd64.c +++ b/mono/mini/tramp-amd64.c @@ -25,6 +25,11 @@ #include "mini.h" #include "mini-amd64.h" +#if defined(__native_client_codegen__) && defined(__native_client__) +#include +#include +#endif + #define IS_REX(inst) (((inst) >= 0x40) && ((inst) <= 0x4f)) static guint8* nullified_class_init_trampoline; @@ -56,6 +61,8 @@ mono_arch_get_unbox_trampoline (MonoMethod *m, gpointer addr) amd64_jump_reg (code, AMD64_RAX); g_assert ((code - start) < 20); + nacl_domain_code_validate (domain, &start, 20, &code); + mono_arch_flush_icache (start, code - start); return start; @@ -90,6 +97,7 @@ mono_arch_get_static_rgctx_trampoline (MonoMethod *m, MonoMethodRuntimeGenericCo amd64_jump_code (code, addr); g_assert ((code - start) < buf_len); + nacl_domain_code_validate (domain, &start, buf_len, &code); mono_arch_flush_icache (start, code - start); return start; @@ -117,6 +125,8 @@ mono_arch_get_llvm_imt_trampoline (MonoDomain *domain, MonoMethod *m, int vt_off g_assert ((code - start) < buf_len); + nacl_domain_code_validate (domain, &start, buf_len, &code); + mono_arch_flush_icache (start, code - start); return start; @@ -131,12 +141,14 @@ mono_arch_get_llvm_imt_trampoline (MonoDomain *domain, MonoMethod *m, int vt_off void mono_arch_patch_callsite (guint8 *method_start, guint8 *orig_code, guint8 *addr) { +#if defined(__default_codegen__) guint8 *code; guint8 buf [16]; gboolean can_write = mono_breakpoint_clean_code (method_start, orig_code, 14, buf, sizeof (buf)); code = buf + 14; + /* mov 64-bit imm into r11 (followed by call reg?) or direct call*/ if (((code [-13] == 0x49) && (code [-12] == 0xbb)) || (code [-5] == 0xe8)) { if (code [-5] != 0xe8) { if (can_write) { @@ -184,6 +196,38 @@ mono_arch_patch_callsite (guint8 *method_start, guint8 *orig_code, guint8 *addr) VALGRIND_DISCARD_TRANSLATIONS (orig_code - 5, sizeof (gpointer)); } } +#elif defined(__native_client__) + /* These are essentially the same 2 cases as above, modified for NaCl*/ + + /* Target must be bundle-aligned */ + g_assert (((guint32)addr & kNaClAlignmentMask) == 0); + /* Return target must be bundle-aligned */ + g_assert (((guint32)orig_code & kNaClAlignmentMask) == 0); + + if (orig_code[-5] == 0xe8) { + /* Direct call */ + int ret; + gint32 offset = (gint32)addr - (gint32)orig_code; + guint8 buf[sizeof(gint32)]; + *((gint32*)(buf)) = offset; + ret = nacl_dyncode_modify (orig_code - sizeof(gint32), buf, sizeof(gint32)); + g_assert (ret == 0); + } + + else if (is_nacl_call_reg_sequence (orig_code - 10) && orig_code[-16] == 0x41 && orig_code[-15] == 0xbb) { + int ret; + guint8 buf[sizeof(gint32)]; + *((gint32 *)(buf)) = addr; + /* orig_code[-14] is the start of the immediate. */ + ret = nacl_dyncode_modify (orig_code - 14, buf, sizeof(gint32)); + g_assert (ret == 0); + } + else { + g_assert_not_reached (); + } + + return; +#endif } void @@ -192,6 +236,7 @@ mono_arch_patch_plt_entry (guint8 *code, gpointer *got, mgreg_t *regs, guint8 *a gint32 disp; gpointer *plt_jump_table_entry; +#if defined(__default_codegen__) /* A PLT entry: jmp *(%rip) */ g_assert (code [0] == 0xff); g_assert (code [1] == 0x25); @@ -199,6 +244,24 @@ mono_arch_patch_plt_entry (guint8 *code, gpointer *got, mgreg_t *regs, guint8 *a disp = *(gint32*)(code + 2); plt_jump_table_entry = (gpointer*)(code + 6 + disp); +#elif defined(__native_client_codegen__) + /* A PLT entry: */ + /* mov (%rip), %r11d */ + /* nacljmp *%r11 */ + + /* Verify the 'mov' */ + g_assert (code [0] == 0x45); + g_assert (code [1] == 0x8b); + g_assert (code [2] == 0x1d); + + disp = *(gint32*)(code + 3); + + /* 7 = 3 (mov opcode) + 4 (disp) */ + /* This needs to resolve to the target of the RIP-relative offset */ + plt_jump_table_entry = (gpointer*)(code + 7 + disp); + +#endif /* __native_client_codegen__ */ + InterlockedExchangePointer (plt_jump_table_entry, addr); } @@ -321,22 +384,25 @@ guchar* mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInfo **info, 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; - int buf_len, saved_regs_offset; + 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, cfa_offset; gboolean has_caller; GSList *unwind_ops = NULL; MonoJumpInfo *ji = NULL; + const guint kMaxCodeSize = NACL_SIZE (548, 548*2); + +#if defined(__native_client_codegen__) + const guint kNaClTrampOffset = 17; +#endif if (tramp_type == MONO_TRAMPOLINE_JUMP) has_caller = FALSE; else has_caller = TRUE; - buf_len = 548; - code = buf = mono_global_codeman_reserve (buf_len); + code = buf = mono_global_codeman_reserve (kMaxCodeSize); - framesize = 538 + sizeof (MonoLMF); + framesize = kMaxCodeSize + sizeof (MonoLMF); framesize = (framesize + (MONO_ARCH_FRAME_ALIGNMENT - 1)) & ~ (MONO_ARCH_FRAME_ALIGNMENT - 1); orig_rsp_to_rbp_offset = 0; @@ -353,72 +419,76 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf /* Pop the return address off the stack */ amd64_pop_reg (code, AMD64_R11); - orig_rsp_to_rbp_offset += 8; + orig_rsp_to_rbp_offset += sizeof(mgreg_t); - cfa_offset -= 8; + cfa_offset -= sizeof(mgreg_t); 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; + cfa_offset += sizeof(mgreg_t); 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); + orig_rsp_to_rbp_offset -= sizeof(mgreg_t); + amd64_mov_reg_reg (code, AMD64_RBP, AMD64_RSP, sizeof(mgreg_t)); 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; rbp_offset = - offset; - offset += 8; + offset += sizeof(mgreg_t); rax_offset = - offset; - offset += 8; + offset += sizeof(mgreg_t); tramp_offset = - offset; - offset += 8; + offset += sizeof(gpointer); arg_offset = - offset; /* Compute the trampoline address from the return address */ if (aot) { +#if defined(__default_codegen__) /* 7 = length of call *(rip) */ amd64_alu_reg_imm (code, X86_SUB, AMD64_R11, 7); +#elif defined(__native_client_codegen__) + amd64_alu_reg_imm (code, X86_SUB, AMD64_R11, kNaClTrampOffset); +#endif } else { /* 5 = length of amd64_call_membase () */ amd64_alu_reg_imm (code, X86_SUB, AMD64_R11, 5); } - amd64_mov_membase_reg (code, AMD64_RBP, tramp_offset, AMD64_R11, 8); + amd64_mov_membase_reg (code, AMD64_RBP, tramp_offset, AMD64_R11, sizeof(gpointer)); - offset += 8; + offset += sizeof(mgreg_t); res_offset = - offset; /* Save all registers */ - offset += AMD64_NREG * 8; + offset += AMD64_NREG * sizeof(mgreg_t); saved_regs_offset = - offset; for (i = 0; i < AMD64_NREG; ++i) { if (i == AMD64_RBP) { /* RAX is already saved */ - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RBP, rbp_offset, 8); - amd64_mov_membase_reg (code, AMD64_RBP, saved_regs_offset + (i * 8), AMD64_RAX, 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RBP, rbp_offset, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, saved_regs_offset + (i * sizeof(mgreg_t)), AMD64_RAX, sizeof(mgreg_t)); } else if (i != AMD64_R11) { - amd64_mov_membase_reg (code, AMD64_RBP, saved_regs_offset + (i * 8), i, 8); + amd64_mov_membase_reg (code, AMD64_RBP, saved_regs_offset + (i * sizeof(mgreg_t)), i, sizeof(mgreg_t)); } else { /* We have to save R11 right at the start of the trampoline code because it's used as a scratch register */ - amd64_mov_membase_reg (r11_save_code, AMD64_RSP, saved_regs_offset + orig_rsp_to_rbp_offset + (i * 8), i, 8); + amd64_mov_membase_reg (r11_save_code, AMD64_RSP, saved_regs_offset + orig_rsp_to_rbp_offset + (i * sizeof(mgreg_t)), i, sizeof(mgreg_t)); g_assert (r11_save_code == after_r11_save_code); } } - offset += 8 * 8; + offset += 8 * sizeof(mgreg_t); saved_fpregs_offset = - offset; for (i = 0; i < 8; ++i) - amd64_movsd_membase_reg (code, AMD64_RBP, saved_fpregs_offset + (i * 8), i); + amd64_movsd_membase_reg (code, AMD64_RBP, saved_fpregs_offset + (i * sizeof(mgreg_t)), i); if (tramp_type != MONO_TRAMPOLINE_GENERIC_CLASS_INIT && tramp_type != MONO_TRAMPOLINE_MONITOR_ENTER && @@ -426,14 +496,21 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf /* Obtain the trampoline argument which is encoded in the instruction stream */ if (aot) { /* Load the GOT offset */ - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, tramp_offset, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, tramp_offset, sizeof(gpointer)); +#if defined(__default_codegen__) amd64_mov_reg_membase (code, AMD64_RAX, AMD64_R11, 7, 4); +#elif defined(__native_client_codegen__) + /* The arg is hidden in a "push imm32" instruction, */ + /* add one to skip the opcode. */ + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_R11, kNaClTrampOffset+1, 4); +#endif /* Compute the address of the GOT slot */ - amd64_alu_reg_reg_size (code, X86_ADD, AMD64_R11, AMD64_RAX, 8); + amd64_alu_reg_reg_size (code, X86_ADD, AMD64_R11, AMD64_RAX, sizeof(gpointer)); /* Load the value */ - amd64_mov_reg_membase (code, AMD64_R11, AMD64_R11, 0, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_R11, 0, sizeof(gpointer)); } else { - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, tramp_offset, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, tramp_offset, sizeof(gpointer)); +#if defined(__default_codegen__) amd64_mov_reg_membase (code, AMD64_RAX, AMD64_R11, 5, 1); amd64_widen_reg (code, AMD64_RAX, AMD64_RAX, TRUE, FALSE); amd64_alu_reg_imm_size (code, X86_CMP, AMD64_RAX, 4, 1); @@ -447,11 +524,15 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf mono_amd64_patch (br [0], code); amd64_mov_reg_membase (code, AMD64_R11, AMD64_R11, 6, 8); mono_amd64_patch (br [1], code); +#elif defined(__native_client_codegen__) + /* All args are 32-bit pointers in NaCl */ + amd64_mov_reg_membase (code, AMD64_R11, AMD64_R11, 6, 4); +#endif } - amd64_mov_membase_reg (code, AMD64_RBP, arg_offset, AMD64_R11, 8); + amd64_mov_membase_reg (code, AMD64_RBP, arg_offset, AMD64_R11, sizeof(gpointer)); } else { - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, saved_regs_offset + (MONO_AMD64_ARG_REG1 * 8), 8); - amd64_mov_membase_reg (code, AMD64_RBP, arg_offset, AMD64_R11, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, saved_regs_offset + (MONO_AMD64_ARG_REG1 * sizeof(mgreg_t)), sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, arg_offset, AMD64_R11, sizeof(gpointer)); } /* Save LMF begin */ @@ -461,34 +542,34 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf /* Save ip */ if (has_caller) - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, 8, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, 8, sizeof(gpointer)); else amd64_mov_reg_imm (code, AMD64_R11, 0); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rip), AMD64_R11, 8); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rip), AMD64_R11, sizeof(mgreg_t)); /* Save fp */ - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RSP, framesize, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rbp), AMD64_R11, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RSP, framesize, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rbp), AMD64_R11, sizeof(mgreg_t)); /* Save sp */ - amd64_mov_reg_reg (code, AMD64_R11, AMD64_RSP, 8); + amd64_mov_reg_reg (code, AMD64_R11, AMD64_RSP, sizeof(mgreg_t)); amd64_alu_reg_imm (code, X86_ADD, AMD64_R11, framesize + 16); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rsp), AMD64_R11, 8); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rsp), AMD64_R11, sizeof(mgreg_t)); /* Save method */ if (tramp_type == MONO_TRAMPOLINE_JIT || tramp_type == MONO_TRAMPOLINE_JUMP) { - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, arg_offset, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, method), AMD64_R11, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, arg_offset, sizeof(gpointer)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, method), AMD64_R11, sizeof(gpointer)); } else { - amd64_mov_membase_imm (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, method), 0, 8); + amd64_mov_membase_imm (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, method), 0, sizeof(gpointer)); } /* Save callee saved regs */ #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); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rdi), AMD64_RDI, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rsi), AMD64_RSI, sizeof(mgreg_t)); #endif - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rbx), AMD64_RBX, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r12), AMD64_R12, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r13), AMD64_R13, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r14), AMD64_R14, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r15), AMD64_R15, 8); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, rbx), AMD64_RBX, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r12), AMD64_R12, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r13), AMD64_R13, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r14), AMD64_R14, sizeof(mgreg_t)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, r15), AMD64_R15, sizeof(mgreg_t)); if (aot) { code = mono_arch_emit_load_aotconst (buf, code, &ji, MONO_PATCH_INFO_JIT_ICALL_ADDR, "mono_get_lmf_addr"); @@ -498,15 +579,15 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf amd64_call_reg (code, AMD64_R11); /* Save lmf_addr */ - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, lmf_addr), AMD64_RAX, 8); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, lmf_addr), AMD64_RAX, sizeof(gpointer)); /* Save previous_lmf */ /* Set the lowest bit to 1 to signal that this LMF has the ip field set */ - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RAX, 0, 8); - amd64_alu_reg_imm_size (code, X86_ADD, AMD64_R11, 1, 8); - amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, previous_lmf), AMD64_R11, 8); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RAX, 0, sizeof(gpointer)); + amd64_alu_reg_imm_size (code, X86_ADD, AMD64_R11, 1, sizeof(gpointer)); + amd64_mov_membase_reg (code, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, previous_lmf), AMD64_R11, sizeof(gpointer)); /* Set new lmf */ amd64_lea_membase (code, AMD64_R11, AMD64_RBP, lmf_offset); - amd64_mov_membase_reg (code, AMD64_RAX, 0, AMD64_R11, 8); + amd64_mov_membase_reg (code, AMD64_RAX, 0, AMD64_R11, sizeof(gpointer)); /* Save LMF end */ @@ -515,15 +596,15 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf /* Arg2 is the address of the calling code */ if (has_caller) - amd64_mov_reg_membase (code, AMD64_ARG_REG2, AMD64_RBP, 8, 8); + amd64_mov_reg_membase (code, AMD64_ARG_REG2, AMD64_RBP, 8, sizeof(gpointer)); else amd64_mov_reg_imm (code, AMD64_ARG_REG2, 0); /* Arg3 is the method/vtable ptr */ - amd64_mov_reg_membase (code, AMD64_ARG_REG3, AMD64_RBP, arg_offset, 8); + amd64_mov_reg_membase (code, AMD64_ARG_REG3, AMD64_RBP, arg_offset, sizeof(gpointer)); /* Arg4 is the trampoline address */ - amd64_mov_reg_membase (code, AMD64_ARG_REG4, AMD64_RBP, tramp_offset, 8); + amd64_mov_reg_membase (code, AMD64_ARG_REG4, AMD64_RBP, tramp_offset, sizeof(gpointer)); if (aot) { char *icall_name = g_strdup_printf ("trampoline_func_%d", tramp_type); @@ -539,7 +620,7 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf /* * Have to call the _force_ variant, since there could be a protected wrapper on the top of the stack. */ - amd64_mov_membase_reg (code, AMD64_RBP, res_offset, AMD64_RAX, 8); + amd64_mov_membase_reg (code, AMD64_RBP, res_offset, AMD64_RAX, sizeof(mgreg_t)); if (aot) { code = mono_arch_emit_load_aotconst (buf, code, &ji, MONO_PATCH_INFO_JIT_ICALL_ADDR, "mono_thread_force_interruption_checkpoint"); } else { @@ -547,43 +628,45 @@ mono_arch_create_generic_trampoline (MonoTrampolineType tramp_type, MonoTrampInf } amd64_call_reg (code, AMD64_R11); - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RBP, res_offset, 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RBP, res_offset, sizeof(mgreg_t)); /* Restore LMF */ - amd64_mov_reg_membase (code, AMD64_RCX, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, previous_lmf), 8); - amd64_alu_reg_imm_size (code, X86_SUB, AMD64_RCX, 1, 8); - amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, lmf_addr), 8); - amd64_mov_membase_reg (code, AMD64_R11, 0, AMD64_RCX, 8); + amd64_mov_reg_membase (code, AMD64_RCX, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, previous_lmf), sizeof(gpointer)); + amd64_alu_reg_imm_size (code, X86_SUB, AMD64_RCX, 1, sizeof(gpointer)); + amd64_mov_reg_membase (code, AMD64_R11, AMD64_RBP, lmf_offset + G_STRUCT_OFFSET (MonoLMF, lmf_addr), sizeof(gpointer)); + amd64_mov_membase_reg (code, AMD64_R11, 0, AMD64_RCX, sizeof(gpointer)); /* * Save rax to the stack, after the leave instruction, this will become part of * the red zone. */ - amd64_mov_membase_reg (code, AMD64_RBP, rax_offset, AMD64_RAX, 8); + amd64_mov_membase_reg (code, AMD64_RBP, rax_offset, AMD64_RAX, sizeof(mgreg_t)); /* Restore argument registers, r10 (imt method/rgxtx) and rax (needed for direct calls to C vararg functions). */ for (i = 0; i < AMD64_NREG; ++i) if (AMD64_IS_ARGUMENT_REG (i) || i == AMD64_R10 || i == AMD64_RAX) - amd64_mov_reg_membase (code, i, AMD64_RBP, saved_regs_offset + (i * 8), 8); + amd64_mov_reg_membase (code, i, AMD64_RBP, saved_regs_offset + (i * sizeof(mgreg_t)), sizeof(mgreg_t)); for (i = 0; i < 8; ++i) - amd64_movsd_reg_membase (code, i, AMD64_RBP, saved_fpregs_offset + (i * 8)); + amd64_movsd_reg_membase (code, i, AMD64_RBP, saved_fpregs_offset + (i * sizeof(mgreg_t))); /* Restore stack */ amd64_leave (code); if (MONO_TRAMPOLINE_TYPE_MUST_RETURN (tramp_type)) { /* Load result */ - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RSP, rax_offset - 0x8, 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RSP, rax_offset - sizeof(mgreg_t), sizeof(mgreg_t)); amd64_ret (code); } else { /* call the compiled method using the saved rax */ - amd64_jump_membase (code, AMD64_RSP, rax_offset - 0x8); + amd64_jump_membase (code, AMD64_RSP, rax_offset - sizeof(mgreg_t)); } - g_assert ((code - buf) <= buf_len); + g_assert ((code - buf) <= kMaxCodeSize); + + nacl_global_codeman_validate (&buf, kMaxCodeSize, &code); mono_arch_flush_icache (buf, code - buf); @@ -606,6 +689,8 @@ mono_arch_get_nullified_class_init_trampoline (MonoTrampInfo **info) code = buf = mono_global_codeman_reserve (16); amd64_ret (code); + nacl_global_codeman_validate(&buf, 16, &code); + mono_arch_flush_icache (buf, code - buf); if (info) @@ -625,15 +710,25 @@ mono_arch_create_specific_trampoline (gpointer arg1, MonoTrampolineType tramp_ty tramp = mono_get_trampoline_code (tramp_type); +#if defined(__default_codegen__) if ((((guint64)arg1) >> 32) == 0) size = 5 + 1 + 4; else size = 5 + 1 + 8; code = buf = mono_domain_code_reserve_align (domain, size, 1); +#elif defined(__native_client_codegen__) + size = 5 + 1 + 4; + /* Aligning the call site below could */ + /* add up to kNaClAlignment-1 bytes */ + size += (kNaClAlignment-1); + buf = mono_domain_code_reserve_align (domain, size, kNaClAlignment); + code = buf; +#endif amd64_call_code (code, tramp); /* The trampoline code will obtain the argument from the instruction stream */ +#if defined(__default_codegen__) if ((((guint64)arg1) >> 32) == 0) { *code = 0x4; *(guint32*)(code + 1) = (gint64)arg1; @@ -643,12 +738,20 @@ mono_arch_create_specific_trampoline (gpointer arg1, MonoTrampolineType tramp_ty *(guint64*)(code + 1) = (gint64)arg1; code += 9; } +#elif defined(__native_client_codegen__) + /* For NaCl, all tramp args are 32-bit because they're pointers */ + *code = 0x68; /* push imm32 */ + *(guint32*)(code + 1) = (gint32)arg1; + code += 5; +#endif g_assert ((code - buf) <= size); if (code_len) *code_len = size; + nacl_domain_code_validate(domain, &buf, size, &code); + mono_arch_flush_icache (buf, size); return buf; @@ -679,7 +782,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info index -= size - 1; } - tramp_size = 64 + 8 * depth; + tramp_size = NACL_SIZE (64 + 8 * depth, 128 + 8 * depth); code = buf = mono_global_codeman_reserve (tramp_size); @@ -692,7 +795,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info amd64_mov_reg_reg (code, AMD64_RAX, AMD64_ARG_REG1, 8); } else { /* load rgctx ptr from vtable */ - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_ARG_REG1, G_STRUCT_OFFSET (MonoVTable, runtime_generic_context), 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_ARG_REG1, G_STRUCT_OFFSET (MonoVTable, runtime_generic_context), sizeof(gpointer)); /* is the rgctx ptr null? */ amd64_test_reg_reg (code, AMD64_RAX, AMD64_RAX); /* if yes, jump to actual trampoline */ @@ -703,9 +806,9 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info for (i = 0; i < depth; ++i) { /* load ptr to next array */ if (mrgctx && i == 0) - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, MONO_SIZEOF_METHOD_RUNTIME_GENERIC_CONTEXT, 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, MONO_SIZEOF_METHOD_RUNTIME_GENERIC_CONTEXT, sizeof(gpointer)); else - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, 0, 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, 0, sizeof(gpointer)); /* is the ptr null? */ amd64_test_reg_reg (code, AMD64_RAX, AMD64_RAX); /* if yes, jump to actual trampoline */ @@ -714,7 +817,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info } /* fetch slot */ - amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, sizeof (gpointer) * (index + 1), 8); + amd64_mov_reg_membase (code, AMD64_RAX, AMD64_RAX, sizeof (gpointer) * (index + 1), sizeof(gpointer)); /* is the slot null? */ amd64_test_reg_reg (code, AMD64_RAX, AMD64_RAX); /* if yes, jump to actual trampoline */ @@ -724,12 +827,12 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info amd64_ret (code); for (i = mrgctx ? 1 : 0; i <= depth + 1; ++i) - x86_patch (rgctx_null_jumps [i], code); + mono_amd64_patch (rgctx_null_jumps [i], code); g_free (rgctx_null_jumps); /* move the rgctx pointer to the VTABLE register */ - amd64_mov_reg_reg (code, MONO_ARCH_VTABLE_REG, AMD64_ARG_REG1, 8); + amd64_mov_reg_reg (code, MONO_ARCH_VTABLE_REG, AMD64_ARG_REG1, sizeof(gpointer)); if (aot) { code = mono_arch_emit_load_aotconst (buf, code, &ji, MONO_PATCH_INFO_JIT_ICALL_ADDR, g_strdup_printf ("specific_trampoline_lazy_fetch_%u", slot)); @@ -741,6 +844,7 @@ mono_arch_create_rgctx_lazy_fetch_trampoline (guint32 slot, MonoTrampInfo **info amd64_jump_code (code, tramp); } + nacl_global_codeman_validate (&buf, tramp_size, &code); mono_arch_flush_icache (buf, code - buf); g_assert (code - buf <= tramp_size); @@ -788,6 +892,8 @@ mono_arch_create_generic_class_init_trampoline (MonoTrampInfo **info, gboolean a amd64_jump_code (code, tramp); } + nacl_global_codeman_validate (&buf, tramp_size, &code); + mono_arch_flush_icache (buf, code - buf); g_assert (code - buf <= tramp_size); @@ -911,6 +1017,8 @@ mono_arch_create_monitor_enter_trampoline (MonoTrampInfo **info, gboolean aot) amd64_jump_code (code, tramp); } + nacl_global_codeman_validate (&buf, tramp_size, &code); + mono_arch_flush_icache (code, code - buf); g_assert (code - buf <= tramp_size); @@ -1027,6 +1135,8 @@ mono_arch_create_monitor_exit_trampoline (MonoTrampInfo **info, gboolean aot) amd64_jump_code (code, tramp); } + nacl_global_codeman_validate (&buf, tramp_size, &code); + mono_arch_flush_icache (code, code - buf); g_assert (code - buf <= tramp_size); @@ -1118,5 +1228,11 @@ mono_arch_get_call_target (guint8 *code) guint32 mono_arch_get_plt_info_offset (guint8 *plt_entry, mgreg_t *regs, guint8 *code) { +#if defined(__native_client__) || defined(__native_client_codegen__) + /* 18 = 3 (mov opcode) + 4 (disp) + 10 (nacljmp) + 1 (push opcode) */ + /* See aot-compiler.c arch_emit_plt_entry for details. */ + return *(guint32*)(plt_entry + 18); +#else return *(guint32*)(plt_entry + 6); +#endif }