Xen devel - Apr 2006 - [PATCH] avoid injection of missed timer interrupts while debugging a guest

Hi Keir, Ian,

Today if we stop the guest for debugging using the gdbserver and then
continue with the guest execution the PIT finds that the time has gone
ahead and it missed to inject so many timer ticks, and it tries to
inject all of those one by one. So after the gdb continue command the
guest just handles the timer interrupts for a long time before it can do
anything else.

    This patch changes this PIT behavior, if it finds that the guest is
being debugged then it does not try to inject missing timer ticks.

 

 Please apply or comment.

 

Signed-Off-By: Nitin A Kamble <nitin.a.kamble@intel.com>

 

Thanks & Regards,

Nitin

------------------------------------------------------------------------
-----------

Open Source Technology Center, Intel Corp

 




_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 19 Apr 2006, at 21:02, Kamble, Nitin A wrote:
> Hi Keir, Ian,
> Today if we stop the guest for debugging using the gdbserver and then 
> continue with the guest execution the PIT finds that the time has gone 
> ahead and it missed to inject so many timer ticks, and it tries to 
> inject all of those one by one. So after the gdb continue command the 
> guest just handles the timer interrupts for a long time before it can 
> do anything else.
>     This patch changes this PIT behavior, if it finds that the guest 
> is being debugged then it does not try to inject missing timer ticks.
How about not accounting missed ticks across DOMF_ctrl_pause''d periods?
This would mean that missing_ticks would still be accounted for a 
DOMF_debugger guest that is not currently being debugged, and will also 
nicely handle any other circumstances in which a guest might get paused 
by control software for any reason. I suppose it''ll be a bigger patch 
though.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Keir,
  Gdbserver does use the xc_ptrace(PTRACE_ATTACH), and the PTRACE_ATTACH uses
xc_domain_pause() to pause the running guest for debugging.
  This is the same interface called by "xm pause". 
One more thing, the PIT missed timer ticks injection code is invoked later after
the domain is un-paused, so at the time of that invocation the domain is no more
paused.
  I think what you are saying can be achieved by keeping a flag that will tell
the PIT code that the domain was paused & unpaused and so if you have missed
ticks do not try to inject it.
  Did I understand you correctly, and do you think the
"domain_was_paused" flag is the right solution?

Thanks & Regards,
Nitin
-----------------------------------------------------------------------------------
Open Source Technology Center, Intel Corp
>-----Original Message-----
>From: Keir Fraser [mailto:Keir.Fraser@cl.cam.ac.uk]
>Sent: Wednesday, April 19, 2006 1:18 PM
>To: Kamble, Nitin A
>Cc: xen-devel@lists.xensource.com; Ian Pratt
>Subject: Re: [PATCH] avoid injection of missed timer interrupts while
>debugging a guest
>
>
>On 19 Apr 2006, at 21:02, Kamble, Nitin A wrote:
>
>> Hi Keir, Ian,
>> Today if we stop the guest for debugging using the gdbserver and then
>> continue with the guest execution the PIT finds that the time has gone
>> ahead and it missed to inject so many timer ticks, and it tries to
>> inject all of those one by one. So after the gdb continue command the
>> guest just handles the timer interrupts for a long time before it can
>> do anything else.
>>     This patch changes this PIT behavior, if it finds that the guest
>> is being debugged then it does not try to inject missing timer ticks.
>
>How about not accounting missed ticks across DOMF_ctrl_pause''d
periods?
>This would mean that missing_ticks would still be accounted for a
>DOMF_debugger guest that is not currently being debugged, and will also
>nicely handle any other circumstances in which a guest might get paused
>by control software for any reason. I suppose it''ll be a bigger
patch
>though.
>
>  -- Keir
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 19 Apr 2006, at 22:28, Kamble, Nitin A wrote:
>   Gdbserver does use the xc_ptrace(PTRACE_ATTACH), and the 
> PTRACE_ATTACH uses xc_domain_pause() to pause the running guest for 
> debugging.
>   This is the same interface called by "xm pause".
> One more thing, the PIT missed timer ticks injection code is invoked 
> later after the domain is un-paused, so at the time of that invocation 
> the domain is no more paused.
>   I think what you are saying can be achieved by keeping a flag that 
> will tell the PIT code that the domain was paused & unpaused and so if 
> you have missed ticks do not try to inject it.
>   Did I understand you correctly, and do you think the 
> "domain_was_paused" flag is the right solution?
Yes, you''ve got it. It is rather a pain as you note -- you essentially 
want a notifier on pause so the hvm code can maintain this flag that 
you can check-and-clear the next time you try to account missed ticks. 
On the other hand, a check for debugger mode is a bit gross and bizarre 
in the guts of HVM timer handling code. :-) I would at least like to 
see a patch that tries the more general approach and then we can pick 
between them.

There is also the question of whether or not we do generally want to 
account missed ticks if a domain is paused for some reason. But I''m 
pretty sure the answer is ''no''.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Keir,
  Here is the patch for avoiding missing timer ticks injection for the
paused domain.

Signed-Off-by: Nitin A Kamble <nitin.a.kamble@intel.com>

Thanks & Regards,
Nitin
------------------------------------------------------------------------
-----------
Open Source Technology Center, Intel Corp
>-----Original Message-----
>From: Keir Fraser [mailto:Keir.Fraser@cl.cam.ac.uk]
>Sent: Wednesday, April 19, 2006 2:31 PM
>To: Kamble, Nitin A
>Cc: xen-devel@lists.xensource.com; Ian Pratt
>Subject: Re: [PATCH] avoid injection of missed timer interrupts while
>debugging a guest
>
>
>On 19 Apr 2006, at 22:28, Kamble, Nitin A wrote:
>
>>   Gdbserver does use the xc_ptrace(PTRACE_ATTACH), and the
>> PTRACE_ATTACH uses xc_domain_pause() to pause the running guest for
>> debugging.
>>   This is the same interface called by "xm pause".
>> One more thing, the PIT missed timer ticks injection code is invoked
>> later after the domain is un-paused, so at the time of that
invocation
>> the domain is no more paused.
>>   I think what you are saying can be achieved by keeping a flag that
>> will tell the PIT code that the domain was paused & unpaused and so
if
>> you have missed ticks do not try to inject it.
>>   Did I understand you correctly, and do you think the
>> "domain_was_paused" flag is the right solution?
>
>Yes, you''ve got it. It is rather a pain as you note -- you
essentially
>want a notifier on pause so the hvm code can maintain this flag that
>you can check-and-clear the next time you try to account missed ticks.
>On the other hand, a check for debugger mode is a bit gross and bizarre
>in the guts of HVM timer handling code. :-) I would at least like to
>see a patch that tries the more general approach and then we can pick
>between them.
>
>There is also the question of whether or not we do generally want to
>account missed ticks if a domain is paused for some reason. But I''m
>pretty sure the answer is ''no''.
>
>  -- Keir

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel