diff --git a/tcg/region.c b/tcg/region.c index b143eaf69c..037a01e4ed 100644 --- a/tcg/region.c +++ b/tcg/region.c @@ -364,38 +364,33 @@ void tcg_region_reset_all(void) tcg_region_tree_reset_all(); } -static size_t tcg_n_regions(unsigned max_cpus) +static size_t tcg_n_regions(size_t tb_size, unsigned max_cpus) { #ifdef CONFIG_USER_ONLY return 1; #else + size_t n_regions; + /* * 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. */ - 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; - } + /* + * Try to have more regions than max_cpus, 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 we can't, then just allocate one region per vCPU thread */ - return max_cpus; + return MIN(n_regions, max_cpus * 8); #endif } @@ -833,7 +828,7 @@ void tcg_region_init(size_t tb_size, int splitwx, unsigned max_cpus) buf = tcg_init_ctx.code_gen_buffer; total_size = tcg_init_ctx.code_gen_buffer_size; page_size = qemu_real_host_page_size; - n_regions = tcg_n_regions(max_cpus); + n_regions = tcg_n_regions(total_size, max_cpus); /* The first region will be 'aligned - buf' bytes larger than the others */ aligned = QEMU_ALIGN_PTR_UP(buf, page_size);