Re: [PATCH] cpus-common: stop using mb_set/mb_read

[Richard Henderson]

On 4/28/23 11:17, Paolo Bonzini wrote:

Use a store-release at the end of the work item, and a load-acquire when
waiting for the item to be completed.  This is the standard message passing
pattern and is both enough and clearer than mb_read/mb_set.

Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
---

Reviewed-by: Richard Henderson <richard.hender...@linaro.org>

r~

  cpus-common.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/cpus-common.c b/cpus-common.c
index b0047e456f93..a53716deb437 100644
--- a/cpus-common.c
+++ b/cpus-common.c
@@ -157,7 +157,7 @@ void do_run_on_cpu(CPUState *cpu, run_on_cpu_func func, 
run_on_cpu_data data,
      wi.exclusive = false;

queue_work_on_cpu(cpu, &wi);

-    while (!qatomic_mb_read(&wi.done)) {
+    while (!qatomic_load_acquire(&wi.done)) {
          CPUState *self_cpu = current_cpu;

qemu_cond_wait(&qemu_work_cond, mutex);

@@ -363,7 +363,7 @@ void process_queued_cpu_work(CPUState *cpu)
          if (wi->free) {
              g_free(wi);
          } else {
-            qatomic_mb_set(&wi->done, true);
+            qatomic_store_release(&wi->done, true);
          }
      }
      qemu_mutex_unlock(&cpu->work_mutex);