Re: [PATCH 13/23] KVM: PPC: Book3S HV: Accumulate timing information for real-mode code

[Alexander Graf <agraf@suse.de>]

On 20.03.15 12:25, Paul Mackerras wrote:
> On Fri, Mar 20, 2015 at 12:15:15PM +0100, Alexander Graf wrote:
>>
>>
>> On 20.03.15 10:39, Paul Mackerras wrote:
>>> This reads the timebase at various points in the real-mode guest
>>> entry/exit code and uses that to accumulate total, minimum and
>>> maximum time spent in those parts of the code.  Currently these
>>> times are accumulated per vcpu in 5 parts of the code:
>>>
>>> * rm_entry - time taken from the start of kvmppc_hv_entry() until
>>>   just before entering the guest.
>>> * rm_intr - time from when we take a hypervisor interrupt in the
>>>   guest until we either re-enter the guest or decide to exit to the
>>>   host.  This includes time spent handling hcalls in real mode.
>>> * rm_exit - time from when we decide to exit the guest until the
>>>   return from kvmppc_hv_entry().
>>> * guest - time spend in the guest
>>> * cede - time spent napping in real mode due to an H_CEDE hcall
>>>   while other threads in the same vcore are active.
>>>
>>> These times are exposed in debugfs in a directory per vcpu that
>>> contains a file called "timings".  This file contains one line for
>>> each of the 5 timings above, with the name followed by a colon and
>>> 4 numbers, which are the count (number of times the code has been
>>> executed), the total time, the minimum time, and the maximum time,
>>> all in nanoseconds.
>>>
>>> Signed-off-by: Paul Mackerras <paulus@samba.org>
>>
>> Have you measure the additional overhead this brings?
> 
> I haven't - in fact I did this patch so I could measure the overhead
> or improvement from other changes I did, but it doesn't measure its
> own overhead, of course.  I guess I need a workload that does a
> defined number of guest entries and exits and measure how fast it runs
> with and without the patch (maybe something like H_SET_MODE in a
> loop).  I'll figure something out and post the results.  

Yeah, just measure the number of exits you can handle for a simple
hcall. If there is measurable overhead, it's probably a good idea to
move the statistics gathering into #ifdef paths for DEBUGFS or maybe
even a separate EXIT_TIMING config option as we have it for booke.


Alex