On 07/09/2013 04:49:32 PM, Alexander Graf wrote:
On 09.07.2013, at 20:29, Scott Wood wrote:
> On 07/09/2013 12:46:32 PM, Alexander Graf wrote:
>> On 07/09/2013 07:16 PM, Scott Wood wrote:
>>> On 07/08/2013 01:45:58 PM, Alexander Graf wrote:
>>>> On 03.07.2013, at 15:30, Mihai Caraman wrote:
>>>> > Some guests are making use of return from machine check
instruction
>>>> > to do crazy things even though the 64-bit kernel doesn't
handle yet
>>>> > this interrupt. Emulate MCSRR0/1 SPR and rfmci instruction
accordingly.
>>>> >
>>>> > Signed-off-by: Mihai Caraman <mihai.caraman@xxxxxxxxxxxxx>
>>>> > ---
>>>> > arch/powerpc/include/asm/kvm_host.h | 1 +
>>>> > arch/powerpc/kvm/booke_emulate.c | 25
+++++++++++++++++++++++++
>>>> > arch/powerpc/kvm/timing.c | 1 +
>>>> > 3 files changed, 27 insertions(+), 0 deletions(-)
>>>> >
>>>> > diff --git a/arch/powerpc/include/asm/kvm_host.h
b/arch/powerpc/include/asm/kvm_host.h
>>>> > index af326cd..0466789 100644
>>>> > --- a/arch/powerpc/include/asm/kvm_host.h
>>>> > +++ b/arch/powerpc/include/asm/kvm_host.h
>>>> > @@ -148,6 +148,7 @@ enum kvm_exit_types {
>>>> > EMULATED_TLBWE_EXITS,
>>>> > EMULATED_RFI_EXITS,
>>>> > EMULATED_RFCI_EXITS,
>>>> > + EMULATED_RFMCI_EXITS,
>>>> I would quite frankly prefer to see us abandon the whole exit
timing framework in the kernel and instead use trace points. Then we
don't have to maintain all of this randomly exercised code.
>>> Would this map well to tracepoints? We're not trying to track
discrete events, so much as accumulated time spent in different areas.
>> I think so. We'd just have to emit tracepoints as soon as we enter
handle_exit and in prepare_to_enter. Then a user space program should
have everything it needs to create statistics out of that. It would
certainly simplify the entry/exit path.
>
> I was hoping that wasn't going to be your answer. :-)
>
> Such a change would introduce a new dependency, more complexity,
and the possibility for bad totals to result from a ring buffer
filling faster than userspace can drain it.
Well, at least it would allow for optional tracing :). Today you have
to change a compile flag to enable / disable timing stats.
>
> I also don't see how it would simplify entry/exit, since we'd still
need to take timestamps in the same places, in order to record a
final event that says how long a particular event took.
Not sure I understand. What the timing stats do is that they measure
the time between [exit ... entry], right? We'd do the same thing,
just all in C code. That means we would become slightly less
accurate, but gain dynamic enabling of the traces and get rid of all
the timing stat asm code.
Compile-time enabling bothers me less than a loss of accuracy (not just
a small loss by moving into C code, but a potential for a large loss if
we overflow the buffer) and a dependency on a userspace tool (both in
terms of the tool needing to be written, and in the hassle of ensuring
that it's present in the root filesystem of whatever system I'm
testing). And the whole mechanism will be more complicated.
Lots of debug options are enabled at build time; why must this be
different?
-Scott
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
