replay: check other timers for icount limit

[PATCH] replay: check other timers for icount limit

Pavel Dovgalyuk posted 1 patch 4 years, 4 months ago

Download mbox

Test asan failed

Test checkpatch failed

Test FreeBSD failed

Test docker-mingw@fedora failed

Test docker-clang@ubuntu failed

Test docker-quick@centos7 failed

Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/157675958855.14346.18049977447896411847.stgit@pasha-Precision-3630-Tower

Maintainers: Richard Henderson <rth@twiddle.net>, Paolo Bonzini <pbonzini@redhat.com>

cpus.c |    4 ++++
1 file changed, 4 insertions(+)

Expand all Fold all

[PATCH] replay: check other timers for icount limit

Posted by Pavel Dovgalyuk 4 years, 4 months ago

Record/replay can stall when there are no virtual devices that generate
events - it just uses all the time for vCPU thread. Therefore main loop
has to wait too much for the vCPU thread, because they are synchronized
in rr mode.
This patch does not let creating too long vCPU executions without
interrupting to main loop. It checks realtime timers that always exits
to control user input.

Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
---
 cpus.c |    4 ++++
 1 file changed, 4 insertions(+)

diff --git a/cpus.c b/cpus.c
index 63bda152f5..7dba4fa414 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1388,6 +1388,10 @@ static int64_t tcg_get_icount_limit(void)
          */
         deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL,
                                               QEMU_TIMER_ATTR_ALL);
+        /* Check realtime timers, because they help with input processing */
+        deadline = qemu_soonest_timeout(deadline,
+                qemu_clock_deadline_ns_all(QEMU_CLOCK_REALTIME,
+                                           QEMU_TIMER_ATTR_ALL));
 
         /* Maintain prior (possibly buggy) behaviour where if no deadline
          * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than

Re: [PATCH] replay: check other timers for icount limit

Posted by Paolo Bonzini 4 years, 4 months ago

On 19/12/19 13:46, Pavel Dovgalyuk wrote:
> Record/replay can stall when there are no virtual devices that generate
> events - it just uses all the time for vCPU thread. Therefore main loop
> has to wait too much for the vCPU thread, because they are synchronized
> in rr mode.
> This patch does not let creating too long vCPU executions without
> interrupting to main loop. It checks realtime timers that always exits
> to control user input.
> 
> Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
> ---
>  cpus.c |    4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/cpus.c b/cpus.c
> index 63bda152f5..7dba4fa414 100644
> --- a/cpus.c
> +++ b/cpus.c
> @@ -1388,6 +1388,10 @@ static int64_t tcg_get_icount_limit(void)
>           */
>          deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL,
>                                                QEMU_TIMER_ATTR_ALL);
> +        /* Check realtime timers, because they help with input processing */
> +        deadline = qemu_soonest_timeout(deadline,
> +                qemu_clock_deadline_ns_all(QEMU_CLOCK_REALTIME,
> +                                           QEMU_TIMER_ATTR_ALL));
>  
>          /* Maintain prior (possibly buggy) behaviour where if no deadline
>           * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
> 

Queued, thanks.

Paolo