diff mbox

[RFC,V7,16/19] translate-all: introduces tb_flush_safe.

Message ID 1439220437-23957-17-git-send-email-fred.konrad@greensocs.com
State New
Headers show

Commit Message

fred.konrad@greensocs.com Aug. 10, 2015, 3:27 p.m. UTC 
From: KONRAD Frederic <fred.konrad@greensocs.com>

tb_flush is not thread safe we definitely need to exit VCPUs to do that.
This introduces tb_flush_safe which just creates an async safe work which will
do a tb_flush later.

Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
 include/exec/exec-all.h |  1 +
 translate-all.c         | 15 +++++++++++++++
 2 files changed, 16 insertions(+)

Comments

Paolo Bonzini Aug. 10, 2015, 4:26 p.m. UTC | #1 
On 10/08/2015 17:27, fred.konrad@greensocs.com wrote:
> +
> +void tb_flush_safe(CPUState *cpu)
> +{
> +#if 0 /* !MTTCG */
> +    tb_flush(cpu);
> +#else
> +    async_run_safe_work_on_cpu(cpu, tb_flush_work, cpu);
> +#endif /* MTTCG */
> +}
> +

I think this can use first_cpu unconditionally; tb_flush only uses its
argument for an error message.

In fact, I think that by definition async_run_safe_work_on_cpu can use
any CPU for its work; it locks out everyone else, so it does not matter
which thread you're on.  So async_run_safe_work_on_cpu could drop the
@cpu argument (becoming async_run_safe_cpu_work) and use first_cpu
unconditionally.

Paolo
Paolo Bonzini Aug. 12, 2015, 2:09 p.m. UTC | #2 
On 10/08/2015 17:27, fred.konrad@greensocs.com wrote:
> From: KONRAD Frederic <fred.konrad@greensocs.com>
> 
> tb_flush is not thread safe we definitely need to exit VCPUs to do that.
> This introduces tb_flush_safe which just creates an async safe work which will
> do a tb_flush later.
> 
> Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>

You could also allocate a new code buffer and free the old one with
call_rcu.  This should simplify things a lot.

Paolo
fred.konrad@greensocs.com Aug. 12, 2015, 2:11 p.m. UTC | #3 
On 12/08/2015 16:09, Paolo Bonzini wrote:
>
> On 10/08/2015 17:27, fred.konrad@greensocs.com wrote:
>> From: KONRAD Frederic <fred.konrad@greensocs.com>
>>
>> tb_flush is not thread safe we definitely need to exit VCPUs to do that.
>> This introduces tb_flush_safe which just creates an async safe work which will
>> do a tb_flush later.
>>
>> Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
> You could also allocate a new code buffer and free the old one with
> call_rcu.  This should simplify things a lot.
>
> Paolo
Depending the size of the code buffer this might be a good idea. :).

Fred
Paolo Bonzini Aug. 12, 2015, 2:14 p.m. UTC | #4 
On 12/08/2015 16:11, Frederic Konrad wrote:
>> You could also allocate a new code buffer and free the old one with
>> call_rcu.  This should simplify things a lot.
>
> Depending the size of the code buffer this might be a good idea. :).

32 megabytes.

Paolo
diff mbox

Patch

diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index e9512df..246df68 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -200,6 +200,7 @@  struct TBContext {
 
 void tb_free(TranslationBlock *tb);
 void tb_flush(CPUState *cpu);
+void tb_flush_safe(CPUState *cpu);
 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr);
 
 #if defined(USE_DIRECT_JUMP)
diff --git a/translate-all.c b/translate-all.c
index fc5162a..7094bf0 100644
--- a/translate-all.c
+++ b/translate-all.c
@@ -818,6 +818,21 @@  static void page_flush_tb(void)
     }
 }
 
+static void tb_flush_work(void *opaque)
+{
+    CPUState *cpu = opaque;
+    tb_flush(cpu);
+}
+
+void tb_flush_safe(CPUState *cpu)
+{
+#if 0 /* !MTTCG */
+    tb_flush(cpu);
+#else
+    async_run_safe_work_on_cpu(cpu, tb_flush_work, cpu);
+#endif /* MTTCG */
+}
+
 /* flush all the translation blocks */
 /* XXX: tb_flush is currently not thread safe */
 void tb_flush(CPUState *cpu)