[Qemu-devel] [5636] Run timers from host alarm timer callback

[Qemu-devel] [5636] Run timers from host alarm timer callback

[Anthony Liguori]

Revision: 5636
          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5636
Author:   aliguori
Date:     2008-11-05 21:04:35 +0000 (Wed, 05 Nov 2008)

Log Message:
-----------
Run timers from host alarm timer callback

This further cleans up the main loop getting it a lot closer to what a main
loop should be.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Modified Paths:
--------------
    trunk/vl.c

Modified: trunk/vl.c
===================================================================
--- trunk/vl.c	2008-11-05 20:49:37 UTC (rev 5635)
+++ trunk/vl.c	2008-11-05 21:04:35 UTC (rev 5636)
@@ -1700,6 +1700,15 @@
         len = read(alarm_timer_rfd, buffer, sizeof(buffer));
     } while ((len == -1 && errno == EINTR) || len > 0);
 
+    /* vm time timers */
+    if (vm_running && likely(!(cur_cpu->singlestep_enabled & SSTEP_NOTIMER)))
+        qemu_run_timers(&active_timers[QEMU_TIMER_VIRTUAL],
+                        qemu_get_clock(vm_clock));
+
+    /* real time timers */
+    qemu_run_timers(&active_timers[QEMU_TIMER_REALTIME],
+                    qemu_get_clock(rt_clock));
+
     if (t->flags & ALARM_FLAG_EXPIRED) {
         alarm_timer->flags &= ~ALARM_FLAG_EXPIRED;
         qemu_rearm_alarm_timer(alarm_timer);
@@ -4551,16 +4560,6 @@
     }
 #endif
 
-    if (vm_running) {
-        if (likely(!(cur_cpu->singlestep_enabled & SSTEP_NOTIMER)))
-        qemu_run_timers(&active_timers[QEMU_TIMER_VIRTUAL],
-                        qemu_get_clock(vm_clock));
-    }
-
-    /* real time timers */
-    qemu_run_timers(&active_timers[QEMU_TIMER_REALTIME],
-                    qemu_get_clock(rt_clock));
-
     /* Check bottom-halves last in case any of the earlier events triggered
        them.  */
     qemu_bh_poll();

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Anthony Liguori]

Anthony Liguori wrote:
> Revision: 5636
>           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5636
> Author:   aliguori
> Date:     2008-11-05 21:04:35 +0000 (Wed, 05 Nov 2008)
>
> Log Message:
> -----------
> Run timers from host alarm timer callback
>
> This further cleans up the main loop getting it a lot closer to what a main
> loop should be.
>
> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
>   

I've tested this with a variety of clocks and guests but this could 
uncover more bugs like the previous 5s select sleep.  If things start 
slowing down, please try this changeset first and let me know.

Regards,

Anthony Liguori

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Luis Pureza]

Hi,

This commit caused a noticeable slowdown while using the instruction
counting mechanism. I'm testing with the sparc-test image and with
--icount auto.

Any ideas?

Thanks,

Luís Pureza

On Wed, Nov 5, 2008 at 9:08 PM, Anthony Liguori <anthony@codemonkey.ws> wrote:
> Anthony Liguori wrote:
>>
>> Revision: 5636
>>          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5636
>> Author:   aliguori
>> Date:     2008-11-05 21:04:35 +0000 (Wed, 05 Nov 2008)
>>
>> Log Message:
>> -----------
>> Run timers from host alarm timer callback
>>
>> This further cleans up the main loop getting it a lot closer to what a
>> main
>> loop should be.
>>
>> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
>>
>
> I've tested this with a variety of clocks and guests but this could uncover
> more bugs like the previous 5s select sleep.  If things start slowing down,
> please try this changeset first and let me know.
>
> Regards,
>
> Anthony Liguori
>
>
>
>

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Anthony Liguori]

Luis Pureza wrote:
> Hi,
>
> This commit caused a noticeable slowdown while using the instruction
> counting mechanism. I'm testing with the sparc-test image and with
> --icount auto.
>   

Yeah, I take it that all timers don't result in signals when using 
icount and that we rely on select() timeout instead.  Let me look a 
little more closely at the code and I'll come up with a fix.

Regards,

Anthony Liguori

> Any ideas?
>
> Thanks,
>
> Luís Pureza
>
> On Wed, Nov 5, 2008 at 9:08 PM, Anthony Liguori <anthony@codemonkey.ws> wrote:
>   
>> Anthony Liguori wrote:
>>     
>>> Revision: 5636
>>>          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5636
>>> Author:   aliguori
>>> Date:     2008-11-05 21:04:35 +0000 (Wed, 05 Nov 2008)
>>>
>>> Log Message:
>>> -----------
>>> Run timers from host alarm timer callback
>>>
>>> This further cleans up the main loop getting it a lot closer to what a
>>> main
>>> loop should be.
>>>
>>> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
>>>
>>>       
>> I've tested this with a variety of clocks and guests but this could uncover
>> more bugs like the previous 5s select sleep.  If things start slowing down,
>> please try this changeset first and let me know.
>>
>> Regards,
>>
>> Anthony Liguori
>>
>>
>>
>>
>>     
>
>
>   

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Paul Brook]

On Friday 07 November 2008, Anthony Liguori wrote:
> Luis Pureza wrote:
> > Hi,
> >
> > This commit caused a noticeable slowdown while using the instruction
> > counting mechanism. I'm testing with the sparc-test image and with
> > --icount auto.
>
> Yeah, I take it that all timers don't result in signals when using
> icount and that we rely on select() timeout instead.  Let me look a
> little more closely at the code and I'll come up with a fix.

In icount mode vm_clock timers don't use host timers at all.

Instead we calculate how long we want to wait before checking again (the 
EXCP_HALT code in vl.c:main_loop), and use that as the select timeout.

Paul

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Anthony Liguori]

Anthony Liguori wrote:
> Luis Pureza wrote:
>> Hi,
>>
>> This commit caused a noticeable slowdown while using the instruction
>> counting mechanism. I'm testing with the sparc-test image and with
>> --icount auto.
>>   
>
> Yeah, I take it that all timers don't result in signals when using 
> icount and that we rely on select() timeout instead.  Let me look a 
> little more closely at the code and I'll come up with a fix.

Did r5623 happen to fix this problem?

Regards,

Anthony Liguori

Re: [Qemu-devel] [5636] Run timers from host alarm timer callback

[Luis Pureza]

On Tue, Nov 11, 2008 at 9:00 PM, Anthony Liguori <anthony@codemonkey.ws> wrote:
> Anthony Liguori wrote:
> Did r5623 happen to fix this problem?

Sorry, no. Should it?

Luís Pureza