@@ -17,5 +17,6 @@
void tcg_cpus_destroy(CPUState *cpu);
int tcg_cpus_exec(CPUState *cpu);
void tcg_handle_interrupt(CPUState *cpu, int mask);
+void tcg_set_cpus_cflags(uint32_t cflags_base);
#endif /* TCG_CPUS_H */
@@ -510,8 +510,6 @@ struct TranslationBlock {
uintptr_t jmp_dest[2];
};
-extern bool parallel_cpus;
-
/* Hide the qatomic_read to make code a little easier on the eyes */
static inline uint32_t tb_cflags(const TranslationBlock *tb)
{
@@ -521,8 +519,7 @@ static inline uint32_t tb_cflags(const TranslationBlock *tb)
/* current cflags for hashing/comparison */
static inline uint32_t curr_cflags(CPUState *cpu)
{
- uint32_t cflags = deposit32(0, CF_CLUSTER_SHIFT, 8, cpu->cluster_index);
- cflags |= parallel_cpus ? CF_PARALLEL : 0;
+ uint32_t cflags = cpu->cflags_base;
cflags |= icount_enabled() ? CF_USE_ICOUNT : 0;
return cflags;
}
@@ -282,6 +282,7 @@ struct qemu_work_item;
* to a cluster this will be UNASSIGNED_CLUSTER_INDEX; otherwise it will
* be the same as the cluster-id property of the CPU object's TYPE_CPU_CLUSTER
* QOM parent.
+ * @cflags_base: Precompute @cluster_index and #CF_PARALLEL for this cpu.
* @nr_cores: Number of cores within this CPU package.
* @nr_threads: Number of threads within this CPU.
* @running: #true if CPU is currently running (lockless).
@@ -412,6 +413,7 @@ struct CPUState {
/* TODO Move common fields from CPUArchState here. */
int cpu_index;
int cluster_index;
+ uint32_t cflags_base;
uint32_t halted;
uint32_t can_do_io;
int32_t exception_index;
@@ -267,8 +267,6 @@ void cpu_exec_step_atomic(CPUState *cpu)
mmap_unlock();
}
- /* Since we got here, we know that parallel_cpus must be true. */
- parallel_cpus = false;
cpu_exec_enter(cpu);
/* execute the generated code */
trace_exec_tb(tb, pc);
@@ -296,7 +294,6 @@ void cpu_exec_step_atomic(CPUState *cpu)
* the execution.
*/
g_assert(cpu_in_exclusive_context(cpu));
- parallel_cpus = true;
cpu->running = false;
end_exclusive();
}
@@ -114,8 +114,7 @@ void mttcg_start_vcpu_thread(CPUState *cpu)
char thread_name[VCPU_THREAD_NAME_SIZE];
g_assert(tcg_enabled());
-
- parallel_cpus = (current_machine->smp.max_cpus > 1);
+ tcg_set_cpus_cflags(current_machine->smp.max_cpus > 1 ? CF_PARALLEL : 0);
cpu->thread = g_malloc0(sizeof(QemuThread));
cpu->halt_cond = g_malloc0(sizeof(QemuCond));
@@ -269,7 +269,7 @@ void rr_start_vcpu_thread(CPUState *cpu)
static QemuThread *single_tcg_cpu_thread;
g_assert(tcg_enabled());
- parallel_cpus = false;
+ tcg_set_cpus_cflags(0);
if (!single_tcg_cpu_thread) {
cpu->thread = g_malloc0(sizeof(QemuThread));
@@ -41,6 +41,16 @@
/* common functionality among all TCG variants */
+void tcg_set_cpus_cflags(uint32_t cflags_base)
+{
+ CPUState *cpu;
+
+ CPU_FOREACH(cpu) {
+ cpu->cflags_base = deposit32(cflags_base, CF_CLUSTER_SHIFT, 8,
+ cpu->cluster_index);
+ }
+}
+
void tcg_cpus_destroy(CPUState *cpu)
{
cpu_thread_signal_destroyed(cpu);
@@ -224,7 +224,6 @@ static void *l1_map[V_L1_MAX_SIZE];
TCGContext tcg_init_ctx;
__thread TCGContext *tcg_ctx;
TBContext tb_ctx;
-bool parallel_cpus;
static void page_table_config_init(void)
{
@@ -205,6 +205,7 @@ CPUArchState *cpu_copy(CPUArchState *env)
/* Reset non arch specific state */
cpu_reset(new_cpu);
+ new_cpu->cflags_base = cpu->cflags_base;
memcpy(new_env, env, sizeof(CPUArchState));
/* Clone all break/watchpoints.
@@ -83,7 +83,7 @@ static abi_ulong get_sigframe(struct target_sigaction *ka,
}
/* Notice when we're in the middle of a gUSA region and reset.
- Note that this will only occur for !parallel_cpus, as we will
+ Note that this will only occur for !CF_PARALLEL, as we will
translate such sequences differently in a parallel context. */
static void unwind_gusa(CPUSH4State *regs)
{
@@ -6481,6 +6481,16 @@ static int do_fork(CPUArchState *env, unsigned int flags, abi_ulong newsp,
/* Grab a mutex so that thread setup appears atomic. */
pthread_mutex_lock(&clone_lock);
+ /*
+ * If this is our first additional thread, we need to ensure we
+ * generate code for parallel execution and flush old translations.
+ * Do this now so that the copy gets CF_PARALLEL too.
+ */
+ if (!(cpu->cflags_base & CF_PARALLEL)) {
+ cpu->cflags_base |= CF_PARALLEL;
+ tb_flush(cpu);
+ }
+
/* we create a new CPU instance. */
new_env = cpu_copy(env);
/* Init regs that differ from the parent. */
@@ -6521,14 +6531,6 @@ static int do_fork(CPUArchState *env, unsigned int flags, abi_ulong newsp,
sigprocmask(SIG_BLOCK, &sigmask, &info.sigmask);
cpu->random_seed = qemu_guest_random_seed_thread_part1();
- /* If this is our first additional thread, we need to ensure we
- * generate code for parallel execution and flush old translations.
- */
- if (!parallel_cpus) {
- parallel_cpus = true;
- tb_flush(cpu);
- }
-
ret = pthread_create(&info.thread, &attr, clone_func, &info);
/* TODO: Free new CPU state if thread creation failed. */
Precompute the initial tb->cflags value for a cpu from its cluster and the number of cpus live in the system. This avoids having to compute this constant data every time we look up a TB. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- Based-on: <20210224165811.11567-1-alex.bennee@linaro.org> ("Experimenting with tb-lookup tweaks") --- accel/tcg/tcg-accel-ops.h | 1 + include/exec/exec-all.h | 5 +---- include/hw/core/cpu.h | 2 ++ accel/tcg/cpu-exec.c | 3 --- accel/tcg/tcg-accel-ops-mttcg.c | 3 +-- accel/tcg/tcg-accel-ops-rr.c | 2 +- accel/tcg/tcg-accel-ops.c | 10 ++++++++++ accel/tcg/translate-all.c | 1 - linux-user/main.c | 1 + linux-user/sh4/signal.c | 2 +- linux-user/syscall.c | 18 ++++++++++-------- 11 files changed, 28 insertions(+), 20 deletions(-) -- 2.25.1