diff mbox series

[v4,07/13] accel-ops: Introduce create_vcpu_thread_precheck / postcheck handlers

Message ID 20220323171751.78612-8-philippe.mathieu.daude@gmail.com (mailing list archive)
State New, archived
Headers show
Series accel: Fix vCPU memory leaks | expand

Commit Message

Philippe Mathieu-Daudé March 23, 2022, 5:17 p.m. UTC
From: Philippe Mathieu-Daudé <f4bug@amsat.org>

Introduce precheck/postcheck handlers which will help to
refactor code common to the various create_vcpu_thread()
implementations.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
---
 include/sysemu/accel-ops.h | 4 ++++
 softmmu/cpus.c             | 8 +++++++-
 2 files changed, 11 insertions(+), 1 deletion(-)

Comments

Richard Henderson March 23, 2022, 10:26 p.m. UTC | #1
On 3/23/22 10:17, Philippe Mathieu-Daudé wrote:
> From: Philippe Mathieu-Daudé <f4bug@amsat.org>
> 
> Introduce precheck/postcheck handlers which will help to
> refactor code common to the various create_vcpu_thread()
> implementations.
> 
> Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> ---
>   include/sysemu/accel-ops.h | 4 ++++
>   softmmu/cpus.c             | 8 +++++++-
>   2 files changed, 11 insertions(+), 1 deletion(-)
> 
> diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
> index 6013c9444c..26b542d35c 100644
> --- a/include/sysemu/accel-ops.h
> +++ b/include/sysemu/accel-ops.h
> @@ -31,6 +31,10 @@ struct AccelOpsClass {
>       bool (*cpus_are_resettable)(void);
>   
>       void (*create_vcpu_thread)(CPUState *cpu); /* MANDATORY NON-NULL */
> +    /* If non-NULL, return whether common vCPU thread must be created */
> +    bool (*create_vcpu_thread_precheck)(CPUState *cpu);
> +    void (*create_vcpu_thread_postcheck)(CPUState *cpu);


I don't think this is the correct trade-off.
These new hooks are only used by rr, and at least in this patch set, incorrectly.

I think you should keep the single create_vcpu_thread hook, and if null, use your new 
common_vcpu_thread_create.  Leave rr to be the single accel setting the hook, and then 
it's easier not to break rr during the reorg as well.


r~
diff mbox series

Patch

diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
index 6013c9444c..26b542d35c 100644
--- a/include/sysemu/accel-ops.h
+++ b/include/sysemu/accel-ops.h
@@ -31,6 +31,10 @@  struct AccelOpsClass {
     bool (*cpus_are_resettable)(void);
 
     void (*create_vcpu_thread)(CPUState *cpu); /* MANDATORY NON-NULL */
+    /* If non-NULL, return whether common vCPU thread must be created */
+    bool (*create_vcpu_thread_precheck)(CPUState *cpu);
+    void (*create_vcpu_thread_postcheck)(CPUState *cpu);
+
     void (*kick_vcpu_thread)(CPUState *cpu);
     bool (*cpu_thread_is_idle)(CPUState *cpu);
 
diff --git a/softmmu/cpus.c b/softmmu/cpus.c
index 7b75bb66d5..857e2081ba 100644
--- a/softmmu/cpus.c
+++ b/softmmu/cpus.c
@@ -637,7 +637,13 @@  void qemu_init_vcpu(CPUState *cpu)
 
     /* accelerators all implement the AccelOpsClass */
     g_assert(cpus_accel != NULL && cpus_accel->create_vcpu_thread != NULL);
-    cpus_accel->create_vcpu_thread(cpu);
+    if (cpus_accel->create_vcpu_thread_precheck == NULL
+            || cpus_accel->create_vcpu_thread_precheck(cpu)) {
+        cpus_accel->create_vcpu_thread(cpu);
+    }
+    if (cpus_accel->create_vcpu_thread_postcheck) {
+        cpus_accel->create_vcpu_thread_postcheck(cpu);
+    }
 
     while (!cpu->created) {
         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);