diff options
Diffstat (limited to 'tcg')
| -rw-r--r-- | tcg/meson.build | 23 | ||||
| -rw-r--r-- | tcg/region.c | 27 | ||||
| -rw-r--r-- | tcg/tcg-internal.h | 2 | ||||
| -rw-r--r-- | tcg/tcg-op-ldst.c | 2 | ||||
| -rw-r--r-- | tcg/tcg.c | 17 |
5 files changed, 26 insertions, 45 deletions
diff --git a/tcg/meson.build b/tcg/meson.build index 69ebb49..7df378d 100644 --- a/tcg/meson.build +++ b/tcg/meson.build @@ -27,24 +27,5 @@ if host_os == 'linux' tcg_ss.add(files('perf.c')) endif -tcg_ss = tcg_ss.apply({}) - -libtcg_user = static_library('tcg_user', - tcg_ss.sources() + genh, - dependencies: tcg_ss.dependencies(), - c_args: '-DCONFIG_USER_ONLY', - build_by_default: false) - -tcg_user = declare_dependency(objects: libtcg_user.extract_all_objects(recursive: false), - dependencies: tcg_ss.dependencies()) -user_ss.add(tcg_user) - -libtcg_system = static_library('tcg_system', - tcg_ss.sources() + genh, - dependencies: tcg_ss.dependencies(), - c_args: '-DCONFIG_SOFTMMU', - build_by_default: false) - -tcg_system = declare_dependency(objects: libtcg_system.extract_all_objects(recursive: false), - dependencies: tcg_ss.dependencies()) -system_ss.add(tcg_system) +libuser_ss.add_all(tcg_ss) +libsystem_ss.add_all(tcg_ss) diff --git a/tcg/region.c b/tcg/region.c index 478ec05..7ea0b37 100644 --- a/tcg/region.c +++ b/tcg/region.c @@ -422,7 +422,7 @@ void tcg_region_reset_all(void) tcg_region_tree_reset_all(); } -static size_t tcg_n_regions(size_t tb_size, unsigned max_cpus) +static size_t tcg_n_regions(size_t tb_size, unsigned max_threads) { #ifdef CONFIG_USER_ONLY return 1; @@ -431,24 +431,25 @@ static size_t tcg_n_regions(size_t tb_size, unsigned max_cpus) /* * 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 + * so we first try to set more regions than threads, 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. + * + * Use a single region if all we have is one vCPU thread. */ - /* Use a single region if all we have is one vCPU thread */ - if (max_cpus == 1 || !qemu_tcg_mttcg_enabled()) { + if (max_threads == 1) { return 1; } /* - * Try to have more regions than max_cpus, with each region being >= 2 MB. + * Try to have more regions than threads, with each region being >= 2 MB. * If we can't, then just allocate one region per vCPU thread. */ n_regions = tb_size / (2 * MiB); - if (n_regions <= max_cpus) { - return max_cpus; + if (n_regions <= max_threads) { + return max_threads; } - return MIN(n_regions, max_cpus * 8); + return MIN(n_regions, max_threads * 8); #endif } @@ -731,11 +732,7 @@ static int alloc_code_gen_buffer(size_t size, int splitwx, Error **errp) * and then assigning regions to TCG threads so that the threads can translate * code in parallel without synchronization. * - * In system-mode 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 system-mode the number of TCG threads is bounded by max_threads, * * 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 @@ -749,7 +746,7 @@ static int alloc_code_gen_buffer(size_t size, int splitwx, Error **errp) * in practice. Multi-threaded guests share most if not all of their translated * code, which makes parallel code generation less appealing than in system-mode */ -void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus) +void tcg_region_init(size_t tb_size, int splitwx, unsigned max_threads) { const size_t page_size = qemu_real_host_page_size(); size_t region_size; @@ -787,7 +784,7 @@ void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus) * 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.n = tcg_n_regions(tb_size, max_cpus); + region.n = tcg_n_regions(tb_size, max_threads); region_size = tb_size / region.n; region_size = QEMU_ALIGN_DOWN(region_size, page_size); diff --git a/tcg/tcg-internal.h b/tcg/tcg-internal.h index a648ee7..ff85fb2 100644 --- a/tcg/tcg-internal.h +++ b/tcg/tcg-internal.h @@ -34,7 +34,7 @@ extern TCGContext **tcg_ctxs; extern unsigned int tcg_cur_ctxs; extern unsigned int tcg_max_ctxs; -void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus); +void tcg_region_init(size_t tb_size, int splitwx, unsigned max_threads); bool tcg_region_alloc(TCGContext *s); void tcg_region_initial_alloc(TCGContext *s); void tcg_region_prologue_set(TCGContext *s); diff --git a/tcg/tcg-op-ldst.c b/tcg/tcg-op-ldst.c index 73838e2..3b073b4 100644 --- a/tcg/tcg-op-ldst.c +++ b/tcg/tcg-op-ldst.c @@ -37,7 +37,7 @@ static void check_max_alignment(unsigned a_bits) { /* * The requested alignment cannot overlap the TLB flags. - * FIXME: Must keep the count up-to-date with "exec/cpu-all.h". + * FIXME: Must keep the count up-to-date with "exec/tlb-flags.h". */ if (tcg_use_softmmu) { tcg_debug_assert(a_bits + 5 <= tcg_ctx->page_bits); @@ -1499,7 +1499,7 @@ static void process_constraint_sets(void); static TCGTemp *tcg_global_reg_new_internal(TCGContext *s, TCGType type, TCGReg reg, const char *name); -static void tcg_context_init(unsigned max_cpus) +static void tcg_context_init(unsigned max_threads) { TCGContext *s = &tcg_init_ctx; int n, i; @@ -1538,15 +1538,15 @@ static void tcg_context_init(unsigned max_cpus) * 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 system-mode we will have at most max_cpus TCG threads. + * In system-mode we will have at most max_threads TCG threads. */ #ifdef CONFIG_USER_ONLY tcg_ctxs = &tcg_ctx; tcg_cur_ctxs = 1; tcg_max_ctxs = 1; #else - tcg_max_ctxs = max_cpus; - tcg_ctxs = g_new0(TCGContext *, max_cpus); + tcg_max_ctxs = max_threads; + tcg_ctxs = g_new0(TCGContext *, max_threads); #endif tcg_debug_assert(!tcg_regset_test_reg(s->reserved_regs, TCG_AREG0)); @@ -1554,10 +1554,10 @@ static void tcg_context_init(unsigned max_cpus) tcg_env = temp_tcgv_ptr(ts); } -void tcg_init(size_t tb_size, int splitwx, unsigned max_cpus) +void tcg_init(size_t tb_size, int splitwx, unsigned max_threads) { - tcg_context_init(max_cpus); - tcg_region_init(tb_size, splitwx, max_cpus); + tcg_context_init(max_threads); + tcg_region_init(tb_size, splitwx, max_threads); } /* @@ -4671,6 +4671,9 @@ static void temp_load(TCGContext *s, TCGTemp *ts, TCGRegSet desired_regs, ts->mem_coherent = 0; break; case TEMP_VAL_MEM: + if (!ts->mem_allocated) { + temp_allocate_frame(s, ts); + } reg = tcg_reg_alloc(s, desired_regs, allocated_regs, preferred_regs, ts->indirect_base); tcg_out_ld(s, ts->type, reg, ts->mem_base->reg, ts->mem_offset); |