gettimeofday with interrupts disabled

gettimeofday with interrupts disabled

[Aziz Eker]

Hi,

I want to get some measurements about the time scheduler_tick()
function takes. To do that, I am finding time difference between the
start and end of the function using 'gettimeofday_us'. However, the
scheduler_tick() function is called with interrupts disabled. Does
'gettimeofday' continue to run even when interrupts are disabled? If
not, is there anyway to count the RTC in systemtap?

 Thanks,

Aziz

Re: gettimeofday with interrupts disabled

[Roland McGrath]

There are other parts of the kernel that call do_gettimeofday with
interrupts disabled and locks held, for whatever that's worth.

RE: gettimeofday with interrupts disabled

[Stone, Joshua I]

On Monday, May 08, 2006 3:12 PM, James Dickens wrote:
> On 5/8/06, Stone, Joshua I <joshua.i.stone@intel.com> wrote:
>> On Sunday, May 07, 2006 6:24 PM, Aziz Eker wrote:
>>> I want to get some measurements about the time scheduler_tick()
>>> function takes. To do that, I am finding time difference between the
>>> start and end of the function using 'gettimeofday_us'. However, the
>>> scheduler_tick() function is called with interrupts disabled. Does
>>> 'gettimeofday' continue to run even when interrupts are disabled? If
>>> not, is there anyway to count the RTC in systemtap?
>> 
>> Calling gettimeofday is only a problem for probes in code sections
>> that hold a write lock on xtime_lock.  See bug #2525 for more
>> details: http://sources.redhat.com/bugzilla/show_bug.cgi?id=2525
>> 
> what if its an smp machine, and and someone else is calling
> gettimeofday?  seems like a race to me..

A race for what?  'do_gettimeofday' only takes a read lock on
xtime_lock, and multiple read locks are perfectly fine.

Even if the read locks were exclusive, it still would not be a deadlock.
One thread would be left waiting for the other to release, but since
your scenario is running on independent CPUs, the lock will eventually
be released.


Josh

Re: gettimeofday with interrupts disabled

[James Dickens]

On 5/8/06, Stone, Joshua I <joshua.i.stone@intel.com> wrote:
> On Sunday, May 07, 2006 6:24 PM, Aziz Eker wrote:
> > I want to get some measurements about the time scheduler_tick()
> > function takes. To do that, I am finding time difference between the
> > start and end of the function using 'gettimeofday_us'. However, the
> > scheduler_tick() function is called with interrupts disabled. Does
> > 'gettimeofday' continue to run even when interrupts are disabled? If
> > not, is there anyway to count the RTC in systemtap?
>
> Calling gettimeofday is only a problem for probes in code sections that
> hold a write lock on xtime_lock.  See bug #2525 for more details:
> http://sources.redhat.com/bugzilla/show_bug.cgi?id=2525
>
what if its an smp machine, and and someone else is calling
gettimeofday?  seems like a race to me..

James Dickens
uadmin.blogspot.com

>
> Josh
>

RE: gettimeofday with interrupts disabled

[Stone, Joshua I]

On Sunday, May 07, 2006 6:24 PM, Aziz Eker wrote:
> I want to get some measurements about the time scheduler_tick()
> function takes. To do that, I am finding time difference between the
> start and end of the function using 'gettimeofday_us'. However, the
> scheduler_tick() function is called with interrupts disabled. Does
> 'gettimeofday' continue to run even when interrupts are disabled? If
> not, is there anyway to count the RTC in systemtap?

Calling gettimeofday is only a problem for probes in code sections that
hold a write lock on xtime_lock.  See bug #2525 for more details:
http://sources.redhat.com/bugzilla/show_bug.cgi?id=2525


Josh