From 44ded3d04821bec57407cc26a8b4db620da2be04 Mon Sep 17 00:00:00 2001 From: Emilio G. Cota Date: Fri, 23 Jun 2017 20:04:43 -0400 Subject: tcg: take tb_ctx out of TCGContext Groundwork for supporting multiple TCG contexts. Reviewed-by: Richard Henderson Reviewed-by: Alex BennĂ©e Signed-off-by: Emilio G. Cota Signed-off-by: Richard Henderson --- linux-user/main.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'linux-user') diff --git a/linux-user/main.c b/linux-user/main.c index dde04c769a..0a2a0d75b3 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -129,7 +129,7 @@ int cpu_get_pic_interrupt(CPUX86State *env) void fork_start(void) { cpu_list_lock(); - qemu_mutex_lock(&tcg_ctx.tb_ctx.tb_lock); + qemu_mutex_lock(&tb_ctx.tb_lock); mmap_fork_start(); } @@ -145,11 +145,11 @@ void fork_end(int child) QTAILQ_REMOVE(&cpus, cpu, node); } } - qemu_mutex_init(&tcg_ctx.tb_ctx.tb_lock); + qemu_mutex_init(&tb_ctx.tb_lock); qemu_init_cpu_list(); gdbserver_fork(thread_cpu); } else { - qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock); + qemu_mutex_unlock(&tb_ctx.tb_lock); cpu_list_unlock(); } } -- cgit v1.2.3-55-g7522 From b1311c4acf503dc9c1a310cc40b64f05b08833dc Mon Sep 17 00:00:00 2001 From: Emilio G. Cota Date: Wed, 12 Jul 2017 17:15:52 -0400 Subject: tcg: define tcg_init_ctx and make tcg_ctx a pointer Groundwork for supporting multiple TCG contexts. The core of this patch is this change to tcg/tcg.h: > -extern TCGContext tcg_ctx; > +extern TCGContext tcg_init_ctx; > +extern TCGContext *tcg_ctx; Note that for now we set *tcg_ctx to whatever TCGContext is passed to tcg_context_init -- in this case &tcg_init_ctx. Reviewed-by: Richard Henderson Signed-off-by: Emilio G. Cota Signed-off-by: Richard Henderson --- accel/tcg/tcg-runtime.c | 2 +- accel/tcg/translate-all.c | 109 +++++++++++++++++++++--------------------- bsd-user/main.c | 2 +- include/exec/gen-icount.h | 10 ++-- linux-user/main.c | 2 +- target/alpha/translate.c | 2 +- target/arm/translate.c | 2 +- target/cris/translate.c | 2 +- target/cris/translate_v10.c | 2 +- target/hppa/translate.c | 2 +- target/i386/translate.c | 2 +- target/lm32/translate.c | 2 +- target/m68k/translate.c | 2 +- target/microblaze/translate.c | 2 +- target/mips/translate.c | 2 +- target/moxie/translate.c | 2 +- target/nios2/translate.c | 2 +- target/openrisc/translate.c | 2 +- target/ppc/translate.c | 2 +- target/s390x/translate.c | 2 +- target/sh4/translate.c | 2 +- target/sparc/translate.c | 2 +- target/tilegx/translate.c | 2 +- target/tricore/translate.c | 2 +- target/unicore32/translate.c | 2 +- target/xtensa/translate.c | 2 +- tcg/tcg-op.c | 46 +++++++++--------- tcg/tcg.c | 22 +++++---- tcg/tcg.h | 21 ++++---- 29 files changed, 130 insertions(+), 126 deletions(-) (limited to 'linux-user') diff --git a/accel/tcg/tcg-runtime.c b/accel/tcg/tcg-runtime.c index 25f0cabfed..4172ffda82 100644 --- a/accel/tcg/tcg-runtime.c +++ b/accel/tcg/tcg-runtime.c @@ -153,7 +153,7 @@ void *HELPER(lookup_tb_ptr)(CPUArchState *env) tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, curr_cflags()); if (tb == NULL) { - return tcg_ctx.code_gen_epilogue; + return tcg_ctx->code_gen_epilogue; } qemu_log_mask_and_addr(CPU_LOG_EXEC, pc, "Chain %p [%d: " TARGET_FMT_lx "] %s\n", diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c index b238b724a8..7cd9ad5f9c 100644 --- a/accel/tcg/translate-all.c +++ b/accel/tcg/translate-all.c @@ -153,7 +153,8 @@ static int v_l2_levels; static void *l1_map[V_L1_MAX_SIZE]; /* code generation context */ -TCGContext tcg_ctx; +TCGContext tcg_init_ctx; +TCGContext *tcg_ctx; TBContext tb_ctx; bool parallel_cpus; @@ -209,7 +210,7 @@ static TranslationBlock *tb_find_pc(uintptr_t tc_ptr); void cpu_gen_init(void) { - tcg_context_init(&tcg_ctx); + tcg_context_init(&tcg_init_ctx); } /* Encode VAL as a signed leb128 sequence at P. @@ -267,7 +268,7 @@ static target_long decode_sleb128(uint8_t **pp) static int encode_search(TranslationBlock *tb, uint8_t *block) { - uint8_t *highwater = tcg_ctx.code_gen_highwater; + uint8_t *highwater = tcg_ctx->code_gen_highwater; uint8_t *p = block; int i, j, n; @@ -278,12 +279,12 @@ static int encode_search(TranslationBlock *tb, uint8_t *block) if (i == 0) { prev = (j == 0 ? tb->pc : 0); } else { - prev = tcg_ctx.gen_insn_data[i - 1][j]; + prev = tcg_ctx->gen_insn_data[i - 1][j]; } - p = encode_sleb128(p, tcg_ctx.gen_insn_data[i][j] - prev); + p = encode_sleb128(p, tcg_ctx->gen_insn_data[i][j] - prev); } - prev = (i == 0 ? 0 : tcg_ctx.gen_insn_end_off[i - 1]); - p = encode_sleb128(p, tcg_ctx.gen_insn_end_off[i] - prev); + prev = (i == 0 ? 0 : tcg_ctx->gen_insn_end_off[i - 1]); + p = encode_sleb128(p, tcg_ctx->gen_insn_end_off[i] - prev); /* Test for (pending) buffer overflow. The assumption is that any one row beginning below the high water mark cannot overrun @@ -343,8 +344,8 @@ static int cpu_restore_state_from_tb(CPUState *cpu, TranslationBlock *tb, restore_state_to_opc(env, tb, data); #ifdef CONFIG_PROFILER - tcg_ctx.restore_time += profile_getclock() - ti; - tcg_ctx.restore_count++; + tcg_ctx->restore_time += profile_getclock() - ti; + tcg_ctx->restore_count++; #endif return 0; } @@ -590,7 +591,7 @@ static inline void *split_cross_256mb(void *buf1, size_t size1) buf1 = buf2; } - tcg_ctx.code_gen_buffer_size = size1; + tcg_ctx->code_gen_buffer_size = size1; return buf1; } #endif @@ -653,16 +654,16 @@ static inline void *alloc_code_gen_buffer(void) size = full_size - qemu_real_host_page_size; /* Honor a command-line option limiting the size of the buffer. */ - if (size > tcg_ctx.code_gen_buffer_size) { - size = (((uintptr_t)buf + tcg_ctx.code_gen_buffer_size) + if (size > tcg_ctx->code_gen_buffer_size) { + size = (((uintptr_t)buf + tcg_ctx->code_gen_buffer_size) & qemu_real_host_page_mask) - (uintptr_t)buf; } - tcg_ctx.code_gen_buffer_size = size; + tcg_ctx->code_gen_buffer_size = size; #ifdef __mips__ if (cross_256mb(buf, size)) { buf = split_cross_256mb(buf, size); - size = tcg_ctx.code_gen_buffer_size; + size = tcg_ctx->code_gen_buffer_size; } #endif @@ -675,7 +676,7 @@ static inline void *alloc_code_gen_buffer(void) #elif defined(_WIN32) static inline void *alloc_code_gen_buffer(void) { - size_t size = tcg_ctx.code_gen_buffer_size; + size_t size = tcg_ctx->code_gen_buffer_size; void *buf1, *buf2; /* Perform the allocation in two steps, so that the guard page @@ -694,7 +695,7 @@ static inline void *alloc_code_gen_buffer(void) { int flags = MAP_PRIVATE | MAP_ANONYMOUS; uintptr_t start = 0; - size_t size = tcg_ctx.code_gen_buffer_size; + size_t size = tcg_ctx->code_gen_buffer_size; void *buf; /* Constrain the position of the buffer based on the host cpu. @@ -711,7 +712,7 @@ static inline void *alloc_code_gen_buffer(void) flags |= MAP_32BIT; /* Cannot expect to map more than 800MB in low memory. */ if (size > 800u * 1024 * 1024) { - tcg_ctx.code_gen_buffer_size = size = 800u * 1024 * 1024; + tcg_ctx->code_gen_buffer_size = size = 800u * 1024 * 1024; } # elif defined(__sparc__) start = 0x40000000ul; @@ -751,7 +752,7 @@ static inline void *alloc_code_gen_buffer(void) default: /* Split the original buffer. Free the smaller half. */ buf2 = split_cross_256mb(buf, size); - size2 = tcg_ctx.code_gen_buffer_size; + size2 = tcg_ctx->code_gen_buffer_size; if (buf == buf2) { munmap(buf + size2 + qemu_real_host_page_size, size - size2); } else { @@ -819,9 +820,9 @@ static gint tb_tc_cmp(gconstpointer ap, gconstpointer bp) static inline void code_gen_alloc(size_t tb_size) { - tcg_ctx.code_gen_buffer_size = size_code_gen_buffer(tb_size); - tcg_ctx.code_gen_buffer = alloc_code_gen_buffer(); - if (tcg_ctx.code_gen_buffer == NULL) { + tcg_ctx->code_gen_buffer_size = size_code_gen_buffer(tb_size); + tcg_ctx->code_gen_buffer = alloc_code_gen_buffer(); + if (tcg_ctx->code_gen_buffer == NULL) { fprintf(stderr, "Could not allocate dynamic translator buffer\n"); exit(1); } @@ -849,7 +850,7 @@ void tcg_exec_init(unsigned long tb_size) #if defined(CONFIG_SOFTMMU) /* There's no guest base to take into account, so go ahead and initialize the prologue now. */ - tcg_prologue_init(&tcg_ctx); + tcg_prologue_init(tcg_ctx); #endif } @@ -865,7 +866,7 @@ static TranslationBlock *tb_alloc(target_ulong pc) assert_tb_locked(); - tb = tcg_tb_alloc(&tcg_ctx); + tb = tcg_tb_alloc(tcg_ctx); if (unlikely(tb == NULL)) { return NULL; } @@ -949,11 +950,11 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count) g_tree_foreach(tb_ctx.tb_tree, tb_host_size_iter, &host_size); printf("qemu: flush code_size=%td nb_tbs=%zu avg_tb_size=%zu\n", - tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer, nb_tbs, + tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer, nb_tbs, nb_tbs > 0 ? host_size / nb_tbs : 0); } - if ((unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer) - > tcg_ctx.code_gen_buffer_size) { + if ((unsigned long)(tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer) + > tcg_ctx->code_gen_buffer_size) { cpu_abort(cpu, "Internal error: code buffer overflow\n"); } @@ -968,7 +969,7 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count) qht_reset_size(&tb_ctx.htable, CODE_GEN_HTABLE_SIZE); page_flush_tb(); - tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer; + tcg_ctx->code_gen_ptr = tcg_ctx->code_gen_buffer; /* XXX: flush processor icache at this point if cache flush is expensive */ atomic_mb_set(&tb_ctx.tb_flush_count, tb_ctx.tb_flush_count + 1); @@ -1316,44 +1317,44 @@ TranslationBlock *tb_gen_code(CPUState *cpu, cpu_loop_exit(cpu); } - gen_code_buf = tcg_ctx.code_gen_ptr; + gen_code_buf = tcg_ctx->code_gen_ptr; tb->tc.ptr = gen_code_buf; tb->pc = pc; tb->cs_base = cs_base; tb->flags = flags; tb->cflags = cflags; tb->trace_vcpu_dstate = *cpu->trace_dstate; - tcg_ctx.tb_cflags = cflags; + tcg_ctx->tb_cflags = cflags; #ifdef CONFIG_PROFILER - tcg_ctx.tb_count1++; /* includes aborted translations because of + tcg_ctx->tb_count1++; /* includes aborted translations because of exceptions */ ti = profile_getclock(); #endif - tcg_func_start(&tcg_ctx); + tcg_func_start(tcg_ctx); - tcg_ctx.cpu = ENV_GET_CPU(env); + tcg_ctx->cpu = ENV_GET_CPU(env); gen_intermediate_code(cpu, tb); - tcg_ctx.cpu = NULL; + tcg_ctx->cpu = NULL; trace_translate_block(tb, tb->pc, tb->tc.ptr); /* generate machine code */ tb->jmp_reset_offset[0] = TB_JMP_RESET_OFFSET_INVALID; tb->jmp_reset_offset[1] = TB_JMP_RESET_OFFSET_INVALID; - tcg_ctx.tb_jmp_reset_offset = tb->jmp_reset_offset; + tcg_ctx->tb_jmp_reset_offset = tb->jmp_reset_offset; if (TCG_TARGET_HAS_direct_jump) { - tcg_ctx.tb_jmp_insn_offset = tb->jmp_target_arg; - tcg_ctx.tb_jmp_target_addr = NULL; + tcg_ctx->tb_jmp_insn_offset = tb->jmp_target_arg; + tcg_ctx->tb_jmp_target_addr = NULL; } else { - tcg_ctx.tb_jmp_insn_offset = NULL; - tcg_ctx.tb_jmp_target_addr = tb->jmp_target_arg; + tcg_ctx->tb_jmp_insn_offset = NULL; + tcg_ctx->tb_jmp_target_addr = tb->jmp_target_arg; } #ifdef CONFIG_PROFILER - tcg_ctx.tb_count++; - tcg_ctx.interm_time += profile_getclock() - ti; + tcg_ctx->tb_count++; + tcg_ctx->interm_time += profile_getclock() - ti; ti = profile_getclock(); #endif @@ -1362,7 +1363,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu, the tcg optimization currently hidden inside tcg_gen_code. All that should be required is to flush the TBs, allocate a new TB, re-initialize it per above, and re-do the actual code generation. */ - gen_code_size = tcg_gen_code(&tcg_ctx, tb); + gen_code_size = tcg_gen_code(tcg_ctx, tb); if (unlikely(gen_code_size < 0)) { goto buffer_overflow; } @@ -1373,10 +1374,10 @@ TranslationBlock *tb_gen_code(CPUState *cpu, tb->tc.size = gen_code_size; #ifdef CONFIG_PROFILER - tcg_ctx.code_time += profile_getclock() - ti; - tcg_ctx.code_in_len += tb->size; - tcg_ctx.code_out_len += gen_code_size; - tcg_ctx.search_out_len += search_size; + tcg_ctx->code_time += profile_getclock() - ti; + tcg_ctx->code_in_len += tb->size; + tcg_ctx->code_out_len += gen_code_size; + tcg_ctx->search_out_len += search_size; #endif #ifdef DEBUG_DISAS @@ -1384,8 +1385,8 @@ TranslationBlock *tb_gen_code(CPUState *cpu, qemu_log_in_addr_range(tb->pc)) { qemu_log_lock(); qemu_log("OUT: [size=%d]\n", gen_code_size); - if (tcg_ctx.data_gen_ptr) { - size_t code_size = tcg_ctx.data_gen_ptr - tb->tc.ptr; + if (tcg_ctx->data_gen_ptr) { + size_t code_size = tcg_ctx->data_gen_ptr - tb->tc.ptr; size_t data_size = gen_code_size - code_size; size_t i; @@ -1394,12 +1395,12 @@ TranslationBlock *tb_gen_code(CPUState *cpu, for (i = 0; i < data_size; i += sizeof(tcg_target_ulong)) { if (sizeof(tcg_target_ulong) == 8) { qemu_log("0x%08" PRIxPTR ": .quad 0x%016" PRIx64 "\n", - (uintptr_t)tcg_ctx.data_gen_ptr + i, - *(uint64_t *)(tcg_ctx.data_gen_ptr + i)); + (uintptr_t)tcg_ctx->data_gen_ptr + i, + *(uint64_t *)(tcg_ctx->data_gen_ptr + i)); } else { qemu_log("0x%08" PRIxPTR ": .long 0x%08x\n", - (uintptr_t)tcg_ctx.data_gen_ptr + i, - *(uint32_t *)(tcg_ctx.data_gen_ptr + i)); + (uintptr_t)tcg_ctx->data_gen_ptr + i, + *(uint32_t *)(tcg_ctx->data_gen_ptr + i)); } } } else { @@ -1411,7 +1412,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu, } #endif - tcg_ctx.code_gen_ptr = (void *) + tcg_ctx->code_gen_ptr = (void *) ROUND_UP((uintptr_t)gen_code_buf + gen_code_size + search_size, CODE_GEN_ALIGN); @@ -1940,8 +1941,8 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf) * For avg host size we use the precise numbers from tb_tree_stats though. */ cpu_fprintf(f, "gen code size %td/%zd\n", - tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer, - tcg_ctx.code_gen_highwater - tcg_ctx.code_gen_buffer); + tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer, + tcg_ctx->code_gen_highwater - tcg_ctx->code_gen_buffer); cpu_fprintf(f, "TB count %zu\n", nb_tbs); cpu_fprintf(f, "TB avg target size %zu max=%zu bytes\n", nb_tbs ? tst.target_size / nb_tbs : 0, diff --git a/bsd-user/main.c b/bsd-user/main.c index 836daac15c..392c0ed5fb 100644 --- a/bsd-user/main.c +++ b/bsd-user/main.c @@ -977,7 +977,7 @@ int main(int argc, char **argv) /* Now that we've loaded the binary, GUEST_BASE is fixed. Delay generating the prologue until now so that the prologue can take the real value of GUEST_BASE into account. */ - tcg_prologue_init(&tcg_ctx); + tcg_prologue_init(tcg_ctx); /* build Task State */ memset(ts, 0, sizeof(TaskState)); diff --git a/include/exec/gen-icount.h b/include/exec/gen-icount.h index 48b566c1c9..c58b0b2585 100644 --- a/include/exec/gen-icount.h +++ b/include/exec/gen-icount.h @@ -19,7 +19,7 @@ static inline void gen_tb_start(TranslationBlock *tb) count = tcg_temp_new_i32(); } - tcg_gen_ld_i32(count, tcg_ctx.tcg_env, + tcg_gen_ld_i32(count, tcg_ctx->tcg_env, -ENV_OFFSET + offsetof(CPUState, icount_decr.u32)); if (tb_cflags(tb) & CF_USE_ICOUNT) { @@ -37,7 +37,7 @@ static inline void gen_tb_start(TranslationBlock *tb) tcg_gen_brcondi_i32(TCG_COND_LT, count, 0, exitreq_label); if (tb_cflags(tb) & CF_USE_ICOUNT) { - tcg_gen_st16_i32(count, tcg_ctx.tcg_env, + tcg_gen_st16_i32(count, tcg_ctx->tcg_env, -ENV_OFFSET + offsetof(CPUState, icount_decr.u16.low)); } @@ -56,13 +56,13 @@ static inline void gen_tb_end(TranslationBlock *tb, int num_insns) tcg_gen_exit_tb((uintptr_t)tb + TB_EXIT_REQUESTED); /* Terminate the linked list. */ - tcg_ctx.gen_op_buf[tcg_ctx.gen_op_buf[0].prev].next = 0; + tcg_ctx->gen_op_buf[tcg_ctx->gen_op_buf[0].prev].next = 0; } static inline void gen_io_start(void) { TCGv_i32 tmp = tcg_const_i32(1); - tcg_gen_st_i32(tmp, tcg_ctx.tcg_env, + tcg_gen_st_i32(tmp, tcg_ctx->tcg_env, -ENV_OFFSET + offsetof(CPUState, can_do_io)); tcg_temp_free_i32(tmp); } @@ -70,7 +70,7 @@ static inline void gen_io_start(void) static inline void gen_io_end(void) { TCGv_i32 tmp = tcg_const_i32(0); - tcg_gen_st_i32(tmp, tcg_ctx.tcg_env, + tcg_gen_st_i32(tmp, tcg_ctx->tcg_env, -ENV_OFFSET + offsetof(CPUState, can_do_io)); tcg_temp_free_i32(tmp); } diff --git a/linux-user/main.c b/linux-user/main.c index 0a2a0d75b3..8814906409 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -4476,7 +4476,7 @@ int main(int argc, char **argv, char **envp) /* Now that we've loaded the binary, GUEST_BASE is fixed. Delay generating the prologue until now so that the prologue can take the real value of GUEST_BASE into account. */ - tcg_prologue_init(&tcg_ctx); + tcg_prologue_init(tcg_ctx); #if defined(TARGET_I386) env->cr[0] = CR0_PG_MASK | CR0_WP_MASK | CR0_PE_MASK; diff --git a/target/alpha/translate.c b/target/alpha/translate.c index 53b8c036e2..f6247bf38d 100644 --- a/target/alpha/translate.c +++ b/target/alpha/translate.c @@ -127,7 +127,7 @@ void alpha_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < 31; i++) { cpu_std_ir[i] = tcg_global_mem_new_i64(cpu_env, diff --git a/target/arm/translate.c b/target/arm/translate.c index 397cc7afea..7873c03ae8 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -82,7 +82,7 @@ void arm_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < 16; i++) { cpu_R[i] = tcg_global_mem_new_i32(cpu_env, diff --git a/target/cris/translate.c b/target/cris/translate.c index 6774acc7af..6687b838d5 100644 --- a/target/cris/translate.c +++ b/target/cris/translate.c @@ -3369,7 +3369,7 @@ void cris_initialize_tcg(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cc_x = tcg_global_mem_new(cpu_env, offsetof(CPUCRISState, cc_x), "cc_x"); cc_src = tcg_global_mem_new(cpu_env, diff --git a/target/cris/translate_v10.c b/target/cris/translate_v10.c index 4a0b485d8e..5d489203f4 100644 --- a/target/cris/translate_v10.c +++ b/target/cris/translate_v10.c @@ -1273,7 +1273,7 @@ void cris_initialize_crisv10_tcg(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cc_x = tcg_global_mem_new(cpu_env, offsetof(CPUCRISState, cc_x), "cc_x"); cc_src = tcg_global_mem_new(cpu_env, diff --git a/target/hppa/translate.c b/target/hppa/translate.c index 08b2c73291..9059812d4e 100644 --- a/target/hppa/translate.c +++ b/target/hppa/translate.c @@ -127,7 +127,7 @@ void hppa_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; TCGV_UNUSED(cpu_gr[0]); for (i = 1; i < 32; i++) { diff --git a/target/i386/translate.c b/target/i386/translate.c index 70ba0b2d5a..649004393d 100644 --- a/target/i386/translate.c +++ b/target/i386/translate.c @@ -8368,7 +8368,7 @@ void tcg_x86_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_cc_op = tcg_global_mem_new_i32(cpu_env, offsetof(CPUX86State, cc_op), "cc_op"); cpu_cc_dst = tcg_global_mem_new(cpu_env, offsetof(CPUX86State, cc_dst), diff --git a/target/lm32/translate.c b/target/lm32/translate.c index d4a2e00165..6707967a2c 100644 --- a/target/lm32/translate.c +++ b/target/lm32/translate.c @@ -1209,7 +1209,7 @@ void lm32_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < ARRAY_SIZE(cpu_R); i++) { cpu_R[i] = tcg_global_mem_new(cpu_env, diff --git a/target/m68k/translate.c b/target/m68k/translate.c index d751faed7c..f6e902f2b6 100644 --- a/target/m68k/translate.c +++ b/target/m68k/translate.c @@ -70,7 +70,7 @@ void m68k_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; #define DEFO32(name, offset) \ QREG_##name = tcg_global_mem_new_i32(cpu_env, \ diff --git a/target/microblaze/translate.c b/target/microblaze/translate.c index c70a2d6644..22f8d6230b 100644 --- a/target/microblaze/translate.c +++ b/target/microblaze/translate.c @@ -1856,7 +1856,7 @@ void mb_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; env_debug = tcg_global_mem_new(cpu_env, offsetof(CPUMBState, debug), diff --git a/target/mips/translate.c b/target/mips/translate.c index aadffbec39..7dfa94ab26 100644 --- a/target/mips/translate.c +++ b/target/mips/translate.c @@ -20455,7 +20455,7 @@ void mips_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; TCGV_UNUSED(cpu_gpr[0]); for (i = 1; i < 32; i++) diff --git a/target/moxie/translate.c b/target/moxie/translate.c index 3f1e609028..59c70b5cef 100644 --- a/target/moxie/translate.c +++ b/target/moxie/translate.c @@ -102,7 +102,7 @@ void moxie_translate_init(void) }; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_pc = tcg_global_mem_new_i32(cpu_env, offsetof(CPUMoxieState, pc), "$pc"); for (i = 0; i < 16; i++) diff --git a/target/nios2/translate.c b/target/nios2/translate.c index d33e365892..b91fd206fb 100644 --- a/target/nios2/translate.c +++ b/target/nios2/translate.c @@ -948,7 +948,7 @@ void nios2_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < NUM_CORE_REGS; i++) { cpu_R[i] = tcg_global_mem_new(cpu_env, diff --git a/target/openrisc/translate.c b/target/openrisc/translate.c index 666d050650..b031f2db97 100644 --- a/target/openrisc/translate.c +++ b/target/openrisc/translate.c @@ -81,7 +81,7 @@ void openrisc_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_sr = tcg_global_mem_new(cpu_env, offsetof(CPUOpenRISCState, sr), "sr"); cpu_dflag = tcg_global_mem_new_i32(cpu_env, diff --git a/target/ppc/translate.c b/target/ppc/translate.c index ac5b8ea9a5..0ad84a75e4 100644 --- a/target/ppc/translate.c +++ b/target/ppc/translate.c @@ -86,7 +86,7 @@ void ppc_translate_init(void) size_t cpu_reg_names_size; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; p = cpu_reg_names; cpu_reg_names_size = sizeof(cpu_reg_names); diff --git a/target/s390x/translate.c b/target/s390x/translate.c index 241b708502..2bf6f48089 100644 --- a/target/s390x/translate.c +++ b/target/s390x/translate.c @@ -113,7 +113,7 @@ void s390x_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; psw_addr = tcg_global_mem_new_i64(cpu_env, offsetof(CPUS390XState, psw.addr), "psw_addr"); diff --git a/target/sh4/translate.c b/target/sh4/translate.c index f918bae978..c13be851ba 100644 --- a/target/sh4/translate.c +++ b/target/sh4/translate.c @@ -100,7 +100,7 @@ void sh4_translate_init(void) }; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < 24; i++) { cpu_gregs[i] = tcg_global_mem_new_i32(cpu_env, diff --git a/target/sparc/translate.c b/target/sparc/translate.c index 9dc41869a4..afef77976b 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -5912,7 +5912,7 @@ void sparc_tcg_init(void) unsigned int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_regwptr = tcg_global_mem_new_ptr(cpu_env, offsetof(CPUSPARCState, regwptr), diff --git a/target/tilegx/translate.c b/target/tilegx/translate.c index 5cd84f6b25..a744c38bb7 100644 --- a/target/tilegx/translate.c +++ b/target/tilegx/translate.c @@ -2446,7 +2446,7 @@ void tilegx_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_pc = tcg_global_mem_new_i64(cpu_env, offsetof(CPUTLGState, pc), "pc"); for (i = 0; i < TILEGX_R_COUNT; i++) { cpu_regs[i] = tcg_global_mem_new_i64(cpu_env, diff --git a/target/tricore/translate.c b/target/tricore/translate.c index 042c0e69bc..590cbbee8b 100644 --- a/target/tricore/translate.c +++ b/target/tricore/translate.c @@ -8882,7 +8882,7 @@ void tricore_tcg_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; /* reg init */ for (i = 0 ; i < 16 ; i++) { cpu_gpr_a[i] = tcg_global_mem_new(cpu_env, diff --git a/target/unicore32/translate.c b/target/unicore32/translate.c index d717de0335..070653e2d1 100644 --- a/target/unicore32/translate.c +++ b/target/unicore32/translate.c @@ -75,7 +75,7 @@ void uc32_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; for (i = 0; i < 32; i++) { cpu_R[i] = tcg_global_mem_new_i32(cpu_env, diff --git a/target/xtensa/translate.c b/target/xtensa/translate.c index f62319eddd..ab96b77d88 100644 --- a/target/xtensa/translate.c +++ b/target/xtensa/translate.c @@ -222,7 +222,7 @@ void xtensa_translate_init(void) int i; cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env"); - tcg_ctx.tcg_env = cpu_env; + tcg_ctx->tcg_env = cpu_env; cpu_pc = tcg_global_mem_new_i32(cpu_env, offsetof(CPUXtensaState, pc), "pc"); diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c index 8c7668de60..ba603281d3 100644 --- a/tcg/tcg-op.c +++ b/tcg/tcg-op.c @@ -48,7 +48,7 @@ extern TCGv_i32 TCGV_HIGH_link_error(TCGv_i64); static inline TCGOp *tcg_emit_op(TCGOpcode opc) { - TCGContext *ctx = &tcg_ctx; + TCGContext *ctx = tcg_ctx; int oi = ctx->gen_next_op_idx; int ni = oi + 1; int pi = oi - 1; @@ -121,7 +121,7 @@ void tcg_gen_op6(TCGOpcode opc, TCGArg a1, TCGArg a2, TCGArg a3, void tcg_gen_mb(TCGBar mb_type) { - if (tcg_ctx.tb_cflags & CF_PARALLEL) { + if (tcg_ctx->tb_cflags & CF_PARALLEL) { tcg_gen_op1(INDEX_op_mb, mb_type); } } @@ -2552,8 +2552,8 @@ void tcg_gen_goto_tb(unsigned idx) tcg_debug_assert(idx <= 1); #ifdef CONFIG_DEBUG_TCG /* Verify that we havn't seen this numbered exit before. */ - tcg_debug_assert((tcg_ctx.goto_tb_issue_mask & (1 << idx)) == 0); - tcg_ctx.goto_tb_issue_mask |= 1 << idx; + tcg_debug_assert((tcg_ctx->goto_tb_issue_mask & (1 << idx)) == 0); + tcg_ctx->goto_tb_issue_mask |= 1 << idx; #endif tcg_gen_op1i(INDEX_op_goto_tb, idx); } @@ -2562,7 +2562,7 @@ void tcg_gen_lookup_and_goto_ptr(void) { if (TCG_TARGET_HAS_goto_ptr && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) { TCGv_ptr ptr = tcg_temp_new_ptr(); - gen_helper_lookup_tb_ptr(ptr, tcg_ctx.tcg_env); + gen_helper_lookup_tb_ptr(ptr, tcg_ctx->tcg_env); tcg_gen_op1i(INDEX_op_goto_ptr, tcgv_ptr_arg(ptr)); tcg_temp_free_ptr(ptr); } else { @@ -2648,7 +2648,7 @@ void tcg_gen_qemu_ld_i32(TCGv_i32 val, TCGv addr, TCGArg idx, TCGMemOp memop) { tcg_gen_req_mo(TCG_MO_LD_LD | TCG_MO_ST_LD); memop = tcg_canonicalize_memop(memop, 0, 0); - trace_guest_mem_before_tcg(tcg_ctx.cpu, tcg_ctx.tcg_env, + trace_guest_mem_before_tcg(tcg_ctx->cpu, tcg_ctx->tcg_env, addr, trace_mem_get_info(memop, 0)); gen_ldst_i32(INDEX_op_qemu_ld_i32, val, addr, memop, idx); } @@ -2657,7 +2657,7 @@ void tcg_gen_qemu_st_i32(TCGv_i32 val, TCGv addr, TCGArg idx, TCGMemOp memop) { tcg_gen_req_mo(TCG_MO_LD_ST | TCG_MO_ST_ST); memop = tcg_canonicalize_memop(memop, 0, 1); - trace_guest_mem_before_tcg(tcg_ctx.cpu, tcg_ctx.tcg_env, + trace_guest_mem_before_tcg(tcg_ctx->cpu, tcg_ctx->tcg_env, addr, trace_mem_get_info(memop, 1)); gen_ldst_i32(INDEX_op_qemu_st_i32, val, addr, memop, idx); } @@ -2676,7 +2676,7 @@ void tcg_gen_qemu_ld_i64(TCGv_i64 val, TCGv addr, TCGArg idx, TCGMemOp memop) } memop = tcg_canonicalize_memop(memop, 1, 0); - trace_guest_mem_before_tcg(tcg_ctx.cpu, tcg_ctx.tcg_env, + trace_guest_mem_before_tcg(tcg_ctx->cpu, tcg_ctx->tcg_env, addr, trace_mem_get_info(memop, 0)); gen_ldst_i64(INDEX_op_qemu_ld_i64, val, addr, memop, idx); } @@ -2690,7 +2690,7 @@ void tcg_gen_qemu_st_i64(TCGv_i64 val, TCGv addr, TCGArg idx, TCGMemOp memop) } memop = tcg_canonicalize_memop(memop, 1, 1); - trace_guest_mem_before_tcg(tcg_ctx.cpu, tcg_ctx.tcg_env, + trace_guest_mem_before_tcg(tcg_ctx->cpu, tcg_ctx->tcg_env, addr, trace_mem_get_info(memop, 1)); gen_ldst_i64(INDEX_op_qemu_st_i64, val, addr, memop, idx); } @@ -2780,7 +2780,7 @@ void tcg_gen_atomic_cmpxchg_i32(TCGv_i32 retv, TCGv addr, TCGv_i32 cmpv, { memop = tcg_canonicalize_memop(memop, 0, 0); - if (!(tcg_ctx.tb_cflags & CF_PARALLEL)) { + if (!(tcg_ctx->tb_cflags & CF_PARALLEL)) { TCGv_i32 t1 = tcg_temp_new_i32(); TCGv_i32 t2 = tcg_temp_new_i32(); @@ -2806,11 +2806,11 @@ void tcg_gen_atomic_cmpxchg_i32(TCGv_i32 retv, TCGv addr, TCGv_i32 cmpv, #ifdef CONFIG_SOFTMMU { TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN, idx)); - gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv, oi); + gen(retv, tcg_ctx->tcg_env, addr, cmpv, newv, oi); tcg_temp_free_i32(oi); } #else - gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv); + gen(retv, tcg_ctx->tcg_env, addr, cmpv, newv); #endif if (memop & MO_SIGN) { @@ -2824,7 +2824,7 @@ void tcg_gen_atomic_cmpxchg_i64(TCGv_i64 retv, TCGv addr, TCGv_i64 cmpv, { memop = tcg_canonicalize_memop(memop, 1, 0); - if (!(tcg_ctx.tb_cflags & CF_PARALLEL)) { + if (!(tcg_ctx->tb_cflags & CF_PARALLEL)) { TCGv_i64 t1 = tcg_temp_new_i64(); TCGv_i64 t2 = tcg_temp_new_i64(); @@ -2851,14 +2851,14 @@ void tcg_gen_atomic_cmpxchg_i64(TCGv_i64 retv, TCGv addr, TCGv_i64 cmpv, #ifdef CONFIG_SOFTMMU { TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop, idx)); - gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv, oi); + gen(retv, tcg_ctx->tcg_env, addr, cmpv, newv, oi); tcg_temp_free_i32(oi); } #else - gen(retv, tcg_ctx.tcg_env, addr, cmpv, newv); + gen(retv, tcg_ctx->tcg_env, addr, cmpv, newv); #endif #else - gen_helper_exit_atomic(tcg_ctx.tcg_env); + gen_helper_exit_atomic(tcg_ctx->tcg_env); /* Produce a result, so that we have a well-formed opcode stream with respect to uses of the result in the (dead) code following. */ tcg_gen_movi_i64(retv, 0); @@ -2914,11 +2914,11 @@ static void do_atomic_op_i32(TCGv_i32 ret, TCGv addr, TCGv_i32 val, #ifdef CONFIG_SOFTMMU { TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN, idx)); - gen(ret, tcg_ctx.tcg_env, addr, val, oi); + gen(ret, tcg_ctx->tcg_env, addr, val, oi); tcg_temp_free_i32(oi); } #else - gen(ret, tcg_ctx.tcg_env, addr, val); + gen(ret, tcg_ctx->tcg_env, addr, val); #endif if (memop & MO_SIGN) { @@ -2959,14 +2959,14 @@ static void do_atomic_op_i64(TCGv_i64 ret, TCGv addr, TCGv_i64 val, #ifdef CONFIG_SOFTMMU { TCGv_i32 oi = tcg_const_i32(make_memop_idx(memop & ~MO_SIGN, idx)); - gen(ret, tcg_ctx.tcg_env, addr, val, oi); + gen(ret, tcg_ctx->tcg_env, addr, val, oi); tcg_temp_free_i32(oi); } #else - gen(ret, tcg_ctx.tcg_env, addr, val); + gen(ret, tcg_ctx->tcg_env, addr, val); #endif #else - gen_helper_exit_atomic(tcg_ctx.tcg_env); + gen_helper_exit_atomic(tcg_ctx->tcg_env); /* Produce a result, so that we have a well-formed opcode stream with respect to uses of the result in the (dead) code following. */ tcg_gen_movi_i64(ret, 0); @@ -3001,7 +3001,7 @@ static void * const table_##NAME[16] = { \ void tcg_gen_atomic_##NAME##_i32 \ (TCGv_i32 ret, TCGv addr, TCGv_i32 val, TCGArg idx, TCGMemOp memop) \ { \ - if (tcg_ctx.tb_cflags & CF_PARALLEL) { \ + if (tcg_ctx->tb_cflags & CF_PARALLEL) { \ do_atomic_op_i32(ret, addr, val, idx, memop, table_##NAME); \ } else { \ do_nonatomic_op_i32(ret, addr, val, idx, memop, NEW, \ @@ -3011,7 +3011,7 @@ void tcg_gen_atomic_##NAME##_i32 \ void tcg_gen_atomic_##NAME##_i64 \ (TCGv_i64 ret, TCGv addr, TCGv_i64 val, TCGArg idx, TCGMemOp memop) \ { \ - if (tcg_ctx.tb_cflags & CF_PARALLEL) { \ + if (tcg_ctx->tb_cflags & CF_PARALLEL) { \ do_atomic_op_i64(ret, addr, val, idx, memop, table_##NAME); \ } else { \ do_nonatomic_op_i64(ret, addr, val, idx, memop, NEW, \ diff --git a/tcg/tcg.c b/tcg/tcg.c index 3a73912827..62f418ac8a 100644 --- a/tcg/tcg.c +++ b/tcg/tcg.c @@ -243,7 +243,7 @@ static void tcg_out_label(TCGContext *s, TCGLabel *l, tcg_insn_unit *ptr) TCGLabel *gen_new_label(void) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; TCGLabel *l = tcg_malloc(sizeof(TCGLabel)); *l = (TCGLabel){ @@ -382,6 +382,8 @@ void tcg_context_init(TCGContext *s) for (; i < ARRAY_SIZE(tcg_target_reg_alloc_order); ++i) { indirect_reg_alloc_order[i] = tcg_target_reg_alloc_order[i]; } + + tcg_ctx = s; } /* @@ -522,7 +524,7 @@ void tcg_set_frame(TCGContext *s, TCGReg reg, intptr_t start, intptr_t size) TCGv_i32 tcg_global_reg_new_i32(TCGReg reg, const char *name) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; TCGTemp *t; if (tcg_regset_test_reg(s->reserved_regs, reg)) { @@ -534,7 +536,7 @@ TCGv_i32 tcg_global_reg_new_i32(TCGReg reg, const char *name) TCGv_i64 tcg_global_reg_new_i64(TCGReg reg, const char *name) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; TCGTemp *t; if (tcg_regset_test_reg(s->reserved_regs, reg)) { @@ -547,7 +549,7 @@ TCGv_i64 tcg_global_reg_new_i64(TCGReg reg, const char *name) TCGTemp *tcg_global_mem_new_internal(TCGType type, TCGv_ptr base, intptr_t offset, const char *name) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; TCGTemp *base_ts = tcgv_ptr_temp(base); TCGTemp *ts = tcg_global_alloc(s); int indirect_reg = 0, bigendian = 0; @@ -602,7 +604,7 @@ TCGTemp *tcg_global_mem_new_internal(TCGType type, TCGv_ptr base, static TCGTemp *tcg_temp_new_internal(TCGType type, int temp_local) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; TCGTemp *ts; int idx, k; @@ -659,7 +661,7 @@ TCGv_i64 tcg_temp_new_internal_i64(int temp_local) static void tcg_temp_free_internal(TCGTemp *ts) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; int k, idx; #if defined(CONFIG_DEBUG_TCG) @@ -723,13 +725,13 @@ TCGv_i64 tcg_const_local_i64(int64_t val) #if defined(CONFIG_DEBUG_TCG) void tcg_clear_temp_count(void) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; s->temps_in_use = 0; } int tcg_check_temp_count(void) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; if (s->temps_in_use) { /* Clear the count so that we don't give another * warning immediately next time around. @@ -969,7 +971,7 @@ bool tcg_op_supported(TCGOpcode op) and endian swap in tcg_reg_alloc_call(). */ void tcg_gen_callN(void *func, TCGTemp *ret, int nargs, TCGTemp **args) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; int i, real_args, nb_rets, pi; unsigned sizemask, flags; TCGHelperInfo *info; @@ -2908,7 +2910,7 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb) #ifdef CONFIG_PROFILER void tcg_dump_info(FILE *f, fprintf_function cpu_fprintf) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; int64_t tb_count = s->tb_count; int64_t tb_div_count = tb_count ? tb_count : 1; int64_t tot = s->interm_time + s->code_time; diff --git a/tcg/tcg.h b/tcg/tcg.h index 76324c9ad6..17fd146557 100644 --- a/tcg/tcg.h +++ b/tcg/tcg.h @@ -688,12 +688,13 @@ struct TCGContext { target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS]; }; -extern TCGContext tcg_ctx; +extern TCGContext tcg_init_ctx; +extern TCGContext *tcg_ctx; static inline size_t temp_idx(TCGTemp *ts) { - ptrdiff_t n = ts - tcg_ctx.temps; - tcg_debug_assert(n >= 0 && n < tcg_ctx.nb_temps); + ptrdiff_t n = ts - tcg_ctx->temps; + tcg_debug_assert(n >= 0 && n < tcg_ctx->nb_temps); return n; } @@ -713,7 +714,7 @@ static inline TCGTemp *arg_temp(TCGArg a) static inline TCGTemp *tcgv_i32_temp(TCGv_i32 v) { uintptr_t o = (uintptr_t)v; - TCGTemp *t = (void *)&tcg_ctx + o; + TCGTemp *t = (void *)tcg_ctx + o; tcg_debug_assert(offsetof(TCGContext, temps[temp_idx(t)]) == o); return t; } @@ -746,7 +747,7 @@ static inline TCGArg tcgv_ptr_arg(TCGv_ptr v) static inline TCGv_i32 temp_tcgv_i32(TCGTemp *t) { (void)temp_idx(t); /* trigger embedded assert */ - return (TCGv_i32)((void *)t - (void *)&tcg_ctx); + return (TCGv_i32)((void *)t - (void *)tcg_ctx); } static inline TCGv_i64 temp_tcgv_i64(TCGTemp *t) @@ -773,13 +774,13 @@ static inline TCGv_i32 TCGV_HIGH(TCGv_i64 t) static inline void tcg_set_insn_param(int op_idx, int arg, TCGArg v) { - tcg_ctx.gen_op_buf[op_idx].args[arg] = v; + tcg_ctx->gen_op_buf[op_idx].args[arg] = v; } /* The number of opcodes emitted so far. */ static inline int tcg_op_buf_count(void) { - return tcg_ctx.gen_next_op_idx; + return tcg_ctx->gen_next_op_idx; } /* Test for whether to terminate the TB for using too many opcodes. */ @@ -798,7 +799,7 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s); /* Called with tb_lock held. */ static inline void *tcg_malloc(int size) { - TCGContext *s = &tcg_ctx; + TCGContext *s = tcg_ctx; uint8_t *ptr, *ptr_end; /* ??? This is a weak placeholder for minimum malloc alignment. */ @@ -807,7 +808,7 @@ static inline void *tcg_malloc(int size) ptr = s->pool_cur; ptr_end = ptr + size; if (unlikely(ptr_end > s->pool_end)) { - return tcg_malloc_internal(&tcg_ctx, size); + return tcg_malloc_internal(tcg_ctx, size); } else { s->pool_cur = ptr_end; return ptr; @@ -1147,7 +1148,7 @@ static inline unsigned get_mmuidx(TCGMemOpIdx oi) uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr); #else # define tcg_qemu_tb_exec(env, tb_ptr) \ - ((uintptr_t (*)(void *, void *))tcg_ctx.code_gen_prologue)(env, tb_ptr) + ((uintptr_t (*)(void *, void *))tcg_ctx->code_gen_prologue)(env, tb_ptr) #endif void tcg_register_jit(void *buf, size_t buf_size); -- cgit v1.2.3-55-g7522 From e8feb96fcc6c16eab8923332e86ff4ef0e2ac276 Mon Sep 17 00:00:00 2001 From: Emilio G. Cota Date: Fri, 7 Jul 2017 19:24:20 -0400 Subject: tcg: introduce regions to split code_gen_buffer This is groundwork for supporting multiple TCG contexts. The naive solution here is to split code_gen_buffer statically among the TCG threads; this however results in poor utilization if translation needs are different across TCG threads. What we do here is to add an extra layer of indirection, assigning regions that act just like pages do in virtual memory allocation. (BTW if you are wondering about the chosen naming, I did not want to use blocks or pages because those are already heavily used in QEMU). We use a global lock to serialize allocations as well as statistics reporting (we now export the size of the used code_gen_buffer with tcg_code_size()). Note that for the allocator we could just use a counter and atomic_inc; however, that would complicate the gathering of tcg_code_size()-like stats. So given that the region operations are not a fast path, a lock seems the most reasonable choice. The effectiveness of this approach is clear after seeing some numbers. I used the bootup+shutdown of debian-arm with '-tb-size 80' as a benchmark. Note that I'm evaluating this after enabling per-thread TCG (which is done by a subsequent commit). * -smp 1, 1 region (entire buffer): qemu: flush code_size=83885014 nb_tbs=154739 avg_tb_size=357 qemu: flush code_size=83884902 nb_tbs=153136 avg_tb_size=363 qemu: flush code_size=83885014 nb_tbs=152777 avg_tb_size=364 qemu: flush code_size=83884950 nb_tbs=150057 avg_tb_size=373 qemu: flush code_size=83884998 nb_tbs=150234 avg_tb_size=373 qemu: flush code_size=83885014 nb_tbs=154009 avg_tb_size=360 qemu: flush code_size=83885014 nb_tbs=151007 avg_tb_size=370 qemu: flush code_size=83885014 nb_tbs=151816 avg_tb_size=367 That is, 8 flushes. * -smp 8, 32 regions (80/32 MB per region) [i.e. this patch]: qemu: flush code_size=76328008 nb_tbs=141040 avg_tb_size=356 qemu: flush code_size=75366534 nb_tbs=138000 avg_tb_size=361 qemu: flush code_size=76864546 nb_tbs=140653 avg_tb_size=361 qemu: flush code_size=76309084 nb_tbs=135945 avg_tb_size=375 qemu: flush code_size=74581856 nb_tbs=132909 avg_tb_size=375 qemu: flush code_size=73927256 nb_tbs=135616 avg_tb_size=360 qemu: flush code_size=78629426 nb_tbs=142896 avg_tb_size=365 qemu: flush code_size=76667052 nb_tbs=138508 avg_tb_size=368 Again, 8 flushes. Note how buffer utilization is not 100%, but it is close. Smaller region sizes would yield higher utilization, but we want region allocation to be rare (it acquires a lock), so we do not want to go too small. * -smp 8, static partitioning of 8 regions (10 MB per region): qemu: flush code_size=21936504 nb_tbs=40570 avg_tb_size=354 qemu: flush code_size=11472174 nb_tbs=20633 avg_tb_size=370 qemu: flush code_size=11603976 nb_tbs=21059 avg_tb_size=365 qemu: flush code_size=23254872 nb_tbs=41243 avg_tb_size=377 qemu: flush code_size=28289496 nb_tbs=52057 avg_tb_size=358 qemu: flush code_size=43605160 nb_tbs=78896 avg_tb_size=367 qemu: flush code_size=45166552 nb_tbs=82158 avg_tb_size=364 qemu: flush code_size=63289640 nb_tbs=116494 avg_tb_size=358 qemu: flush code_size=51389960 nb_tbs=93937 avg_tb_size=362 qemu: flush code_size=59665928 nb_tbs=107063 avg_tb_size=372 qemu: flush code_size=38380824 nb_tbs=68597 avg_tb_size=374 qemu: flush code_size=44884568 nb_tbs=79901 avg_tb_size=376 qemu: flush code_size=50782632 nb_tbs=90681 avg_tb_size=374 qemu: flush code_size=39848888 nb_tbs=71433 avg_tb_size=372 qemu: flush code_size=64708840 nb_tbs=119052 avg_tb_size=359 qemu: flush code_size=49830008 nb_tbs=90992 avg_tb_size=362 qemu: flush code_size=68372408 nb_tbs=123442 avg_tb_size=368 qemu: flush code_size=33555560 nb_tbs=59514 avg_tb_size=378 qemu: flush code_size=44748344 nb_tbs=80974 avg_tb_size=367 qemu: flush code_size=37104248 nb_tbs=67609 avg_tb_size=364 That is, 20 flushes. Note how a static partitioning approach uses the code buffer poorly, leading to many unnecessary flushes. Reviewed-by: Richard Henderson Signed-off-by: Emilio G. Cota Signed-off-by: Richard Henderson --- accel/tcg/translate-all.c | 63 +++++-------- bsd-user/main.c | 1 + cpus.c | 12 +++ linux-user/main.c | 1 + tcg/tcg.c | 222 +++++++++++++++++++++++++++++++++++++++++++++- tcg/tcg.h | 6 ++ 6 files changed, 260 insertions(+), 45 deletions(-) (limited to 'linux-user') diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c index 9061c0508c..f99bfd9309 100644 --- a/accel/tcg/translate-all.c +++ b/accel/tcg/translate-all.c @@ -606,15 +606,13 @@ static inline void *alloc_code_gen_buffer(void) { void *buf = static_code_gen_buffer; void *end = static_code_gen_buffer + sizeof(static_code_gen_buffer); - size_t full_size, size; + size_t size; /* page-align the beginning and end of the buffer */ buf = QEMU_ALIGN_PTR_UP(buf, qemu_real_host_page_size); end = QEMU_ALIGN_PTR_DOWN(end, qemu_real_host_page_size); - /* Reserve a guard page. */ - full_size = end - buf; - size = full_size - qemu_real_host_page_size; + size = end - buf; /* Honor a command-line option limiting the size of the buffer. */ if (size > tcg_ctx->code_gen_buffer_size) { @@ -633,9 +631,6 @@ static inline void *alloc_code_gen_buffer(void) if (qemu_mprotect_rwx(buf, size)) { abort(); } - if (qemu_mprotect_none(buf + size, qemu_real_host_page_size)) { - abort(); - } qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE); return buf; @@ -644,22 +639,16 @@ static inline void *alloc_code_gen_buffer(void) static inline void *alloc_code_gen_buffer(void) { size_t size = tcg_ctx->code_gen_buffer_size; - void *buf1, *buf2; - - /* Perform the allocation in two steps, so that the guard page - is reserved but uncommitted. */ - buf1 = VirtualAlloc(NULL, size + qemu_real_host_page_size, - MEM_RESERVE, PAGE_NOACCESS); - if (buf1 != NULL) { - buf2 = VirtualAlloc(buf1, size, MEM_COMMIT, PAGE_EXECUTE_READWRITE); - assert(buf1 == buf2); - } + void *buf; - return buf1; + buf = VirtualAlloc(NULL, size, MEM_RESERVE | MEM_COMMIT, + PAGE_EXECUTE_READWRITE); + return buf; } #else static inline void *alloc_code_gen_buffer(void) { + int prot = PROT_WRITE | PROT_READ | PROT_EXEC; int flags = MAP_PRIVATE | MAP_ANONYMOUS; uintptr_t start = 0; size_t size = tcg_ctx->code_gen_buffer_size; @@ -693,8 +682,7 @@ static inline void *alloc_code_gen_buffer(void) # endif # endif - buf = mmap((void *)start, size + qemu_real_host_page_size, - PROT_NONE, flags, -1, 0); + buf = mmap((void *)start, size, prot, flags, -1, 0); if (buf == MAP_FAILED) { return NULL; } @@ -704,24 +692,23 @@ static inline void *alloc_code_gen_buffer(void) /* Try again, with the original still mapped, to avoid re-acquiring that 256mb crossing. This time don't specify an address. */ size_t size2; - void *buf2 = mmap(NULL, size + qemu_real_host_page_size, - PROT_NONE, flags, -1, 0); + void *buf2 = mmap(NULL, size, prot, flags, -1, 0); switch ((int)(buf2 != MAP_FAILED)) { case 1: if (!cross_256mb(buf2, size)) { /* Success! Use the new buffer. */ - munmap(buf, size + qemu_real_host_page_size); + munmap(buf, size); break; } /* Failure. Work with what we had. */ - munmap(buf2, size + qemu_real_host_page_size); + munmap(buf2, size); /* fallthru */ default: /* Split the original buffer. Free the smaller half. */ buf2 = split_cross_256mb(buf, size); size2 = tcg_ctx->code_gen_buffer_size; if (buf == buf2) { - munmap(buf + size2 + qemu_real_host_page_size, size - size2); + munmap(buf + size2, size - size2); } else { munmap(buf, size - size2); } @@ -732,10 +719,6 @@ static inline void *alloc_code_gen_buffer(void) } #endif - /* Make the final buffer accessible. The guard page at the end - will remain inaccessible with PROT_NONE. */ - mprotect(buf, size, PROT_WRITE | PROT_READ | PROT_EXEC); - /* Request large pages for the buffer. */ qemu_madvise(buf, size, QEMU_MADV_HUGEPAGE); @@ -916,13 +899,8 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count) size_t host_size = 0; g_tree_foreach(tb_ctx.tb_tree, tb_host_size_iter, &host_size); - printf("qemu: flush code_size=%td nb_tbs=%zu avg_tb_size=%zu\n", - tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer, nb_tbs, - nb_tbs > 0 ? host_size / nb_tbs : 0); - } - if ((unsigned long)(tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer) - > tcg_ctx->code_gen_buffer_size) { - cpu_abort(cpu, "Internal error: code buffer overflow\n"); + printf("qemu: flush code_size=%zu nb_tbs=%zu avg_tb_size=%zu\n", + tcg_code_size(), nb_tbs, nb_tbs > 0 ? host_size / nb_tbs : 0); } CPU_FOREACH(cpu) { @@ -936,7 +914,7 @@ static void do_tb_flush(CPUState *cpu, run_on_cpu_data tb_flush_count) qht_reset_size(&tb_ctx.htable, CODE_GEN_HTABLE_SIZE); page_flush_tb(); - tcg_ctx->code_gen_ptr = tcg_ctx->code_gen_buffer; + tcg_region_reset_all(); /* XXX: flush processor icache at this point if cache flush is expensive */ atomic_mb_set(&tb_ctx.tb_flush_count, tb_ctx.tb_flush_count + 1); @@ -1274,9 +1252,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu, phys_pc = get_page_addr_code(env, pc); + buffer_overflow: tb = tb_alloc(pc); if (unlikely(!tb)) { - buffer_overflow: /* flush must be done */ tb_flush(cpu); mmap_unlock(); @@ -1380,9 +1358,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu, } #endif - tcg_ctx->code_gen_ptr = (void *) + atomic_set(&tcg_ctx->code_gen_ptr, (void *) ROUND_UP((uintptr_t)gen_code_buf + gen_code_size + search_size, - CODE_GEN_ALIGN); + CODE_GEN_ALIGN)); /* init jump list */ assert(((uintptr_t)tb & 3) == 0); @@ -1908,9 +1886,8 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf) * otherwise users might think "-tb-size" is not honoured. * For avg host size we use the precise numbers from tb_tree_stats though. */ - cpu_fprintf(f, "gen code size %td/%zd\n", - tcg_ctx->code_gen_ptr - tcg_ctx->code_gen_buffer, - tcg_ctx->code_gen_highwater - tcg_ctx->code_gen_buffer); + cpu_fprintf(f, "gen code size %zu/%zu\n", + tcg_code_size(), tcg_code_capacity()); cpu_fprintf(f, "TB count %zu\n", nb_tbs); cpu_fprintf(f, "TB avg target size %zu max=%zu bytes\n", nb_tbs ? tst.target_size / nb_tbs : 0, diff --git a/bsd-user/main.c b/bsd-user/main.c index 392c0ed5fb..f1b244b59b 100644 --- a/bsd-user/main.c +++ b/bsd-user/main.c @@ -978,6 +978,7 @@ int main(int argc, char **argv) generating the prologue until now so that the prologue can take the real value of GUEST_BASE into account. */ tcg_prologue_init(tcg_ctx); + tcg_region_init(); /* build Task State */ memset(ts, 0, sizeof(TaskState)); diff --git a/cpus.c b/cpus.c index c9a624003a..8e06257a74 100644 --- a/cpus.c +++ b/cpus.c @@ -1664,6 +1664,18 @@ static void qemu_tcg_init_vcpu(CPUState *cpu) char thread_name[VCPU_THREAD_NAME_SIZE]; static QemuCond *single_tcg_halt_cond; static QemuThread *single_tcg_cpu_thread; + static int tcg_region_inited; + + /* + * Initialize TCG regions--once. Now is a good time, because: + * (1) TCG's init context, prologue and target globals have been set up. + * (2) qemu_tcg_mttcg_enabled() works now (TCG init code runs before the + * -accel flag is processed, so the check doesn't work then). + */ + if (!tcg_region_inited) { + tcg_region_inited = 1; + tcg_region_init(); + } if (qemu_tcg_mttcg_enabled() || !single_tcg_cpu_thread) { cpu->thread = g_malloc0(sizeof(QemuThread)); diff --git a/linux-user/main.c b/linux-user/main.c index 8814906409..28353f1a75 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -4477,6 +4477,7 @@ int main(int argc, char **argv, char **envp) generating the prologue until now so that the prologue can take the real value of GUEST_BASE into account. */ tcg_prologue_init(tcg_ctx); + tcg_region_init(); #if defined(TARGET_I386) env->cr[0] = CR0_PG_MASK | CR0_WP_MASK | CR0_PE_MASK; diff --git a/tcg/tcg.c b/tcg/tcg.c index f1bbfe37ff..3de5f7cf97 100644 --- a/tcg/tcg.c +++ b/tcg/tcg.c @@ -121,6 +121,30 @@ static bool tcg_out_ldst_finalize(TCGContext *s); static TCGContext **tcg_ctxs; static unsigned int n_tcg_ctxs; +/* + * We divide code_gen_buffer into equally-sized "regions" that TCG threads + * dynamically allocate from as demand dictates. Given appropriate region + * sizing, this minimizes flushes even when some TCG threads generate a lot + * more code than others. + */ +struct tcg_region_state { + QemuMutex lock; + + /* fields set at init time */ + void *start; + void *start_aligned; + void *end; + size_t n; + size_t size; /* size of one region */ + size_t stride; /* .size + guard size */ + + /* fields protected by the lock */ + size_t current; /* current region index */ + size_t agg_size_full; /* aggregate size of full regions */ +}; + +static struct tcg_region_state region; + static TCGRegSet tcg_target_available_regs[2]; static TCGRegSet tcg_target_call_clobber_regs; @@ -258,6 +282,196 @@ TCGLabel *gen_new_label(void) #include "tcg-target.inc.c" +static void tcg_region_bounds(size_t curr_region, void **pstart, void **pend) +{ + void *start, *end; + + start = region.start_aligned + curr_region * region.stride; + end = start + region.size; + + if (curr_region == 0) { + start = region.start; + } + if (curr_region == region.n - 1) { + end = region.end; + } + + *pstart = start; + *pend = end; +} + +static void tcg_region_assign(TCGContext *s, size_t curr_region) +{ + void *start, *end; + + tcg_region_bounds(curr_region, &start, &end); + + s->code_gen_buffer = start; + s->code_gen_ptr = start; + s->code_gen_buffer_size = end - start; + s->code_gen_highwater = end - TCG_HIGHWATER; +} + +static bool tcg_region_alloc__locked(TCGContext *s) +{ + if (region.current == region.n) { + return true; + } + tcg_region_assign(s, region.current); + region.current++; + return false; +} + +/* + * Request a new region once the one in use has filled up. + * Returns true on error. + */ +static bool tcg_region_alloc(TCGContext *s) +{ + bool err; + /* read the region size now; alloc__locked will overwrite it on success */ + size_t size_full = s->code_gen_buffer_size; + + qemu_mutex_lock(®ion.lock); + err = tcg_region_alloc__locked(s); + if (!err) { + region.agg_size_full += size_full - TCG_HIGHWATER; + } + qemu_mutex_unlock(®ion.lock); + return err; +} + +/* + * Perform a context's first region allocation. + * This function does _not_ increment region.agg_size_full. + */ +static inline bool tcg_region_initial_alloc__locked(TCGContext *s) +{ + return tcg_region_alloc__locked(s); +} + +/* Call from a safe-work context */ +void tcg_region_reset_all(void) +{ + unsigned int i; + + qemu_mutex_lock(®ion.lock); + region.current = 0; + region.agg_size_full = 0; + + for (i = 0; i < n_tcg_ctxs; i++) { + bool err = tcg_region_initial_alloc__locked(tcg_ctxs[i]); + + g_assert(!err); + } + qemu_mutex_unlock(®ion.lock); +} + +/* + * Initializes region partitioning. + * + * Called at init time from the parent thread (i.e. the one calling + * tcg_context_init), after the target's TCG globals have been set. + */ +void tcg_region_init(void) +{ + void *buf = tcg_init_ctx.code_gen_buffer; + void *aligned; + size_t size = tcg_init_ctx.code_gen_buffer_size; + size_t page_size = qemu_real_host_page_size; + size_t region_size; + size_t n_regions; + size_t i; + + /* We do not yet support multiple TCG contexts, so use one region for now */ + n_regions = 1; + + /* The first region will be 'aligned - buf' bytes larger than the others */ + aligned = QEMU_ALIGN_PTR_UP(buf, page_size); + g_assert(aligned < tcg_init_ctx.code_gen_buffer + size); + /* + * Make region_size a multiple of page_size, using aligned as the start. + * As a result of this we might end up with a few extra pages at the end of + * the buffer; we will assign those to the last region. + */ + region_size = (size - (aligned - buf)) / n_regions; + region_size = QEMU_ALIGN_DOWN(region_size, page_size); + + /* A region must have at least 2 pages; one code, one guard */ + g_assert(region_size >= 2 * page_size); + + /* init the region struct */ + qemu_mutex_init(®ion.lock); + region.n = n_regions; + region.size = region_size - page_size; + region.stride = region_size; + region.start = buf; + region.start_aligned = aligned; + /* page-align the end, since its last page will be a guard page */ + region.end = QEMU_ALIGN_PTR_DOWN(buf + size, page_size); + /* account for that last guard page */ + region.end -= page_size; + + /* set guard pages */ + for (i = 0; i < region.n; i++) { + void *start, *end; + int rc; + + tcg_region_bounds(i, &start, &end); + rc = qemu_mprotect_none(end, page_size); + g_assert(!rc); + } + + /* We do not yet support multiple TCG contexts so allocate the region now */ + { + bool err = tcg_region_initial_alloc__locked(tcg_ctx); + + g_assert(!err); + } +} + +/* + * Returns the size (in bytes) of all translated code (i.e. from all regions) + * currently in the cache. + * See also: tcg_code_capacity() + * Do not confuse with tcg_current_code_size(); that one applies to a single + * TCG context. + */ +size_t tcg_code_size(void) +{ + unsigned int i; + size_t total; + + qemu_mutex_lock(®ion.lock); + total = region.agg_size_full; + for (i = 0; i < n_tcg_ctxs; i++) { + const TCGContext *s = tcg_ctxs[i]; + size_t size; + + size = atomic_read(&s->code_gen_ptr) - s->code_gen_buffer; + g_assert(size <= s->code_gen_buffer_size); + total += size; + } + qemu_mutex_unlock(®ion.lock); + return total; +} + +/* + * Returns the code capacity (in bytes) of the entire cache, i.e. including all + * regions. + * See also: tcg_code_size() + */ +size_t tcg_code_capacity(void) +{ + size_t guard_size, capacity; + + /* no need for synchronization; these variables are set at init time */ + guard_size = region.stride - region.size; + capacity = region.end + guard_size - region.start; + capacity -= region.n * (guard_size + TCG_HIGHWATER); + return capacity; +} + /* pool based memory allocation */ void *tcg_malloc_internal(TCGContext *s, int size) { @@ -401,13 +615,17 @@ TranslationBlock *tcg_tb_alloc(TCGContext *s) TranslationBlock *tb; void *next; + retry: tb = (void *)ROUND_UP((uintptr_t)s->code_gen_ptr, align); next = (void *)ROUND_UP((uintptr_t)(tb + 1), align); if (unlikely(next > s->code_gen_highwater)) { - return NULL; + if (tcg_region_alloc(s)) { + return NULL; + } + goto retry; } - s->code_gen_ptr = next; + atomic_set(&s->code_gen_ptr, next); s->data_gen_ptr = NULL; return tb; } diff --git a/tcg/tcg.h b/tcg/tcg.h index 4b9958a179..9f95648282 100644 --- a/tcg/tcg.h +++ b/tcg/tcg.h @@ -802,6 +802,12 @@ void *tcg_malloc_internal(TCGContext *s, int size); void tcg_pool_reset(TCGContext *s); TranslationBlock *tcg_tb_alloc(TCGContext *s); +void tcg_region_init(void); +void tcg_region_reset_all(void); + +size_t tcg_code_size(void); +size_t tcg_code_capacity(void); + /* Called with tb_lock held. */ static inline void *tcg_malloc(int size) { -- cgit v1.2.3-55-g7522 From 3468b59e18b179bc63c7ce934de912dfa9596122 Mon Sep 17 00:00:00 2001 From: Emilio G. Cota Date: Wed, 19 Jul 2017 18:57:58 -0400 Subject: tcg: enable multiple TCG contexts in softmmu This enables parallel TCG code generation. However, we do not take advantage of it yet since tb_lock is still held during tb_gen_code. In user-mode we use a single TCG context; see the documentation added to tcg_region_init for the rationale. Note that targets do not need any conversion: targets initialize a TCGContext (e.g. defining TCG globals), and after this initialization has finished, the context is cloned by the vCPU threads, each of them keeping a separate copy. TCG threads claim one entry in tcg_ctxs[] by atomically increasing n_tcg_ctxs. Do not be too annoyed by the subsequent atomic_read's of that variable and tcg_ctxs; they are there just to play nice with analysis tools such as thread sanitizer. Note that we do not allocate an array of contexts (we allocate an array of pointers instead) because when tcg_context_init is called, we do not know yet how many contexts we'll use since the bool behind qemu_tcg_mttcg_enabled() isn't set yet. Previous patches folded some TCG globals into TCGContext. The non-const globals remaining are only set at init time, i.e. before the TCG threads are spawned. Here is a list of these set-at-init-time globals under tcg/: Only written by tcg_context_init: - indirect_reg_alloc_order - tcg_op_defs Only written by tcg_target_init (called from tcg_context_init): - tcg_target_available_regs - tcg_target_call_clobber_regs - arm: arm_arch, use_idiv_instructions - i386: have_cmov, have_bmi1, have_bmi2, have_lzcnt, have_movbe, have_popcnt - mips: use_movnz_instructions, use_mips32_instructions, use_mips32r2_instructions, got_sigill (tcg_target_detect_isa) - ppc: have_isa_2_06, have_isa_3_00, tb_ret_addr - s390: tb_ret_addr, s390_facilities - sparc: qemu_ld_trampoline, qemu_st_trampoline (build_trampolines), use_vis3_instructions Only written by tcg_prologue_init: - 'struct jit_code_entry one_entry' - aarch64: tb_ret_addr - arm: tb_ret_addr - i386: tb_ret_addr, guest_base_flags - ia64: tb_ret_addr - mips: tb_ret_addr, bswap32_addr, bswap32u_addr, bswap64_addr Reviewed-by: Richard Henderson Signed-off-by: Emilio G. Cota Signed-off-by: Richard Henderson --- accel/tcg/translate-all.c | 2 +- cpus.c | 2 + linux-user/syscall.c | 1 + tcg/tcg.c | 146 +++++++++++++++++++++++++++++++++++++++++++--- tcg/tcg.h | 7 ++- 5 files changed, 145 insertions(+), 13 deletions(-) (limited to 'linux-user') diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c index f99bfd9309..5724149289 100644 --- a/accel/tcg/translate-all.c +++ b/accel/tcg/translate-all.c @@ -154,7 +154,7 @@ static void *l1_map[V_L1_MAX_SIZE]; /* code generation context */ TCGContext tcg_init_ctx; -TCGContext *tcg_ctx; +__thread TCGContext *tcg_ctx; TBContext tb_ctx; bool parallel_cpus; diff --git a/cpus.c b/cpus.c index 8e06257a74..114c29b6a0 100644 --- a/cpus.c +++ b/cpus.c @@ -1307,6 +1307,7 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) CPUState *cpu = arg; rcu_register_thread(); + tcg_register_thread(); qemu_mutex_lock_iothread(); qemu_thread_get_self(cpu->thread); @@ -1454,6 +1455,7 @@ static void *qemu_tcg_cpu_thread_fn(void *arg) g_assert(!use_icount); rcu_register_thread(); + tcg_register_thread(); qemu_mutex_lock_iothread(); qemu_thread_get_self(cpu->thread); diff --git a/linux-user/syscall.c b/linux-user/syscall.c index 9bf901fa11..d4497dec5d 100644 --- a/linux-user/syscall.c +++ b/linux-user/syscall.c @@ -6218,6 +6218,7 @@ static void *clone_func(void *arg) TaskState *ts; rcu_register_thread(); + tcg_register_thread(); env = info->env; cpu = ENV_GET_CPU(env); thread_cpu = cpu; diff --git a/tcg/tcg.c b/tcg/tcg.c index 3de5f7cf97..5574317736 100644 --- a/tcg/tcg.c +++ b/tcg/tcg.c @@ -58,6 +58,7 @@ #include "elf.h" #include "exec/log.h" +#include "sysemu/sysemu.h" /* Forward declarations for functions declared in tcg-target.inc.c and used here. */ @@ -353,25 +354,87 @@ static inline bool tcg_region_initial_alloc__locked(TCGContext *s) /* Call from a safe-work context */ void tcg_region_reset_all(void) { + unsigned int n_ctxs = atomic_read(&n_tcg_ctxs); unsigned int i; qemu_mutex_lock(®ion.lock); region.current = 0; region.agg_size_full = 0; - for (i = 0; i < n_tcg_ctxs; i++) { - bool err = tcg_region_initial_alloc__locked(tcg_ctxs[i]); + for (i = 0; i < n_ctxs; i++) { + TCGContext *s = atomic_read(&tcg_ctxs[i]); + bool err = tcg_region_initial_alloc__locked(s); g_assert(!err); } qemu_mutex_unlock(®ion.lock); } +#ifdef CONFIG_USER_ONLY +static size_t tcg_n_regions(void) +{ + return 1; +} +#else +/* + * It is likely that some vCPUs will translate more code than others, so we + * first try to set more regions than max_cpus, with those regions being of + * reasonable size. If that's not possible we make do by evenly dividing + * the code_gen_buffer among the vCPUs. + */ +static size_t tcg_n_regions(void) +{ + size_t i; + + /* Use a single region if all we have is one vCPU thread */ + if (max_cpus == 1 || !qemu_tcg_mttcg_enabled()) { + return 1; + } + + /* Try to have more regions than max_cpus, with each region being >= 2 MB */ + for (i = 8; i > 0; i--) { + size_t regions_per_thread = i; + size_t region_size; + + region_size = tcg_init_ctx.code_gen_buffer_size; + region_size /= max_cpus * regions_per_thread; + + if (region_size >= 2 * 1024u * 1024) { + return max_cpus * regions_per_thread; + } + } + /* If we can't, then just allocate one region per vCPU thread */ + return max_cpus; +} +#endif + /* * Initializes region partitioning. * * Called at init time from the parent thread (i.e. the one calling * tcg_context_init), after the target's TCG globals have been set. + * + * Region partitioning works by splitting code_gen_buffer into separate regions, + * and then assigning regions to TCG threads so that the threads can translate + * code in parallel without synchronization. + * + * In softmmu the number of TCG threads is bounded by max_cpus, so we use at + * least max_cpus regions in MTTCG. In !MTTCG we use a single region. + * Note that the TCG options from the command-line (i.e. -accel accel=tcg,[...]) + * must have been parsed before calling this function, since it calls + * qemu_tcg_mttcg_enabled(). + * + * In user-mode we use a single region. Having multiple regions in user-mode + * is not supported, because the number of vCPU threads (recall that each thread + * spawned by the guest corresponds to a vCPU thread) is only bounded by the + * OS, and usually this number is huge (tens of thousands is not uncommon). + * Thus, given this large bound on the number of vCPU threads and the fact + * that code_gen_buffer is allocated at compile-time, we cannot guarantee + * that the availability of at least one region per vCPU thread. + * + * However, this user-mode limitation is unlikely to be a significant problem + * in practice. Multi-threaded guests share most if not all of their translated + * code, which makes parallel code generation less appealing than in softmmu. */ void tcg_region_init(void) { @@ -383,8 +446,7 @@ void tcg_region_init(void) size_t n_regions; size_t i; - /* We do not yet support multiple TCG contexts, so use one region for now */ - n_regions = 1; + n_regions = tcg_n_regions(); /* The first region will be 'aligned - buf' bytes larger than the others */ aligned = QEMU_ALIGN_PTR_UP(buf, page_size); @@ -422,13 +484,66 @@ void tcg_region_init(void) g_assert(!rc); } - /* We do not yet support multiple TCG contexts so allocate the region now */ + /* In user-mode we support only one ctx, so do the initial allocation now */ +#ifdef CONFIG_USER_ONLY { bool err = tcg_region_initial_alloc__locked(tcg_ctx); g_assert(!err); } +#endif +} + +/* + * All TCG threads except the parent (i.e. the one that called tcg_context_init + * and registered the target's TCG globals) must register with this function + * before initiating translation. + * + * In user-mode we just point tcg_ctx to tcg_init_ctx. See the documentation + * of tcg_region_init() for the reasoning behind this. + * + * In softmmu each caller registers its context in tcg_ctxs[]. Note that in + * softmmu tcg_ctxs[] does not track tcg_ctx_init, since the initial context + * is not used anymore for translation once this function is called. + * + * Not tracking tcg_init_ctx in tcg_ctxs[] in softmmu keeps code that iterates + * over the array (e.g. tcg_code_size() the same for both softmmu and user-mode. + */ +#ifdef CONFIG_USER_ONLY +void tcg_register_thread(void) +{ + tcg_ctx = &tcg_init_ctx; +} +#else +void tcg_register_thread(void) +{ + TCGContext *s = g_malloc(sizeof(*s)); + unsigned int i, n; + bool err; + + *s = tcg_init_ctx; + + /* Relink mem_base. */ + for (i = 0, n = tcg_init_ctx.nb_globals; i < n; ++i) { + if (tcg_init_ctx.temps[i].mem_base) { + ptrdiff_t b = tcg_init_ctx.temps[i].mem_base - tcg_init_ctx.temps; + tcg_debug_assert(b >= 0 && b < n); + s->temps[i].mem_base = &s->temps[b]; + } + } + + /* Claim an entry in tcg_ctxs */ + n = atomic_fetch_inc(&n_tcg_ctxs); + g_assert(n < max_cpus); + atomic_set(&tcg_ctxs[n], s); + + tcg_ctx = s; + qemu_mutex_lock(®ion.lock); + err = tcg_region_initial_alloc__locked(tcg_ctx); + g_assert(!err); + qemu_mutex_unlock(®ion.lock); } +#endif /* !CONFIG_USER_ONLY */ /* * Returns the size (in bytes) of all translated code (i.e. from all regions) @@ -439,13 +554,14 @@ void tcg_region_init(void) */ size_t tcg_code_size(void) { + unsigned int n_ctxs = atomic_read(&n_tcg_ctxs); unsigned int i; size_t total; qemu_mutex_lock(®ion.lock); total = region.agg_size_full; - for (i = 0; i < n_tcg_ctxs; i++) { - const TCGContext *s = tcg_ctxs[i]; + for (i = 0; i < n_ctxs; i++) { + const TCGContext *s = atomic_read(&tcg_ctxs[i]); size_t size; size = atomic_read(&s->code_gen_ptr) - s->code_gen_buffer; @@ -601,8 +717,18 @@ void tcg_context_init(TCGContext *s) } tcg_ctx = s; + /* + * In user-mode we simply share the init context among threads, since we + * use a single region. See the documentation tcg_region_init() for the + * reasoning behind this. + * In softmmu we will have at most max_cpus TCG threads. + */ +#ifdef CONFIG_USER_ONLY tcg_ctxs = &tcg_ctx; n_tcg_ctxs = 1; +#else + tcg_ctxs = g_new(TCGContext *, max_cpus); +#endif } /* @@ -2951,10 +3077,12 @@ static void tcg_reg_alloc_call(TCGContext *s, TCGOp *op) static inline void tcg_profile_snapshot(TCGProfile *prof, bool counters, bool table) { + unsigned int n_ctxs = atomic_read(&n_tcg_ctxs); unsigned int i; - for (i = 0; i < n_tcg_ctxs; i++) { - const TCGProfile *orig = &tcg_ctxs[i]->prof; + for (i = 0; i < n_ctxs; i++) { + TCGContext *s = atomic_read(&tcg_ctxs[i]); + const TCGProfile *orig = &s->prof; if (counters) { PROF_ADD(prof, orig, tb_count1); diff --git a/tcg/tcg.h b/tcg/tcg.h index 9f95648282..3d022e448b 100644 --- a/tcg/tcg.h +++ b/tcg/tcg.h @@ -695,7 +695,7 @@ struct TCGContext { }; extern TCGContext tcg_init_ctx; -extern TCGContext *tcg_ctx; +extern __thread TCGContext *tcg_ctx; static inline size_t temp_idx(TCGTemp *ts) { @@ -797,7 +797,7 @@ static inline bool tcg_op_buf_full(void) /* pool based memory allocation */ -/* tb_lock must be held for tcg_malloc_internal. */ +/* user-mode: tb_lock must be held for tcg_malloc_internal. */ void *tcg_malloc_internal(TCGContext *s, int size); void tcg_pool_reset(TCGContext *s); TranslationBlock *tcg_tb_alloc(TCGContext *s); @@ -808,7 +808,7 @@ void tcg_region_reset_all(void); size_t tcg_code_size(void); size_t tcg_code_capacity(void); -/* Called with tb_lock held. */ +/* user-mode: Called with tb_lock held. */ static inline void *tcg_malloc(int size) { TCGContext *s = tcg_ctx; @@ -828,6 +828,7 @@ static inline void *tcg_malloc(int size) } void tcg_context_init(TCGContext *s); +void tcg_register_thread(void); void tcg_prologue_init(TCGContext *s); void tcg_func_start(TCGContext *s); -- cgit v1.2.3-55-g7522