Message ID | 1402991675-24905-6-git-send-email-mpe@ellerman.id.au |
---|---|
State | New |
Headers | show |
On Tue, 2014-06-17 at 10:27 +0200, Alexander Graf wrote: > On 17.06.14 09:54, Michael Ellerman wrote: > > Add support for powerpc platforms. We use uname -m, which allows us to > > detect ppc, ppc64 and ppc64le/el. > > > > Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> > > Could you please add support for PR KVM tracepoints along the way? There > we do know the exit reason for every single guest <-> host transition. I > would like to move to a similar model with HV in the future, so we can > hopefully just reuse this by then. So I think what you're saying is you want it to somehow support using 'kvm_exit' for PR and 'kvm_userspace_exit' for HV? Or actually use 'kvm_exit' if it exists and fall back to 'kvm_userspace_exit', so that if HV starts providing 'kvm_exit' the script will pick that up without further changes. cheers
On 18.06.14 02:50, Michael Ellerman wrote: > On Tue, 2014-06-17 at 10:27 +0200, Alexander Graf wrote: >> On 17.06.14 09:54, Michael Ellerman wrote: >>> Add support for powerpc platforms. We use uname -m, which allows us to >>> detect ppc, ppc64 and ppc64le/el. >>> >>> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> >> Could you please add support for PR KVM tracepoints along the way? There >> we do know the exit reason for every single guest <-> host transition. I >> would like to move to a similar model with HV in the future, so we can >> hopefully just reuse this by then. > So I think what you're saying is you want it to somehow support using > 'kvm_exit' for PR and 'kvm_userspace_exit' for HV? "kvm_userspace_exit" is implemented on both HV and PR. "kvm_exit" is PR only, but I'm hoping we can get it working in HV as well. > Or actually use 'kvm_exit' if it exists and fall back to 'kvm_userspace_exit', > so that if HV starts providing 'kvm_exit' the script will pick that up without > further changes. They are completely different things. "kvm_userspace_exit" tells us which exits we take from KVM -> QEMU. "kvm_exit" tells us which exits we take from guest -> KVM. In fact, IIRC x86 also implements kvm_userspace_exit - or at least something very similar to it. It's a completely separate category. Maybe it should be a command line switch to distinguish between the count types? Paolo, do you have any strong opinion here? Alex
On Wed, 2014-06-18 at 02:59 +0200, Alexander Graf wrote: > On 18.06.14 02:50, Michael Ellerman wrote: > > On Tue, 2014-06-17 at 10:27 +0200, Alexander Graf wrote: > >> On 17.06.14 09:54, Michael Ellerman wrote: > >>> Add support for powerpc platforms. We use uname -m, which allows us to > >>> detect ppc, ppc64 and ppc64le/el. > >>> > >>> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> > >> Could you please add support for PR KVM tracepoints along the way? There > >> we do know the exit reason for every single guest <-> host transition. I > >> would like to move to a similar model with HV in the future, so we can > >> hopefully just reuse this by then. > > So I think what you're saying is you want it to somehow support using > > 'kvm_exit' for PR and 'kvm_userspace_exit' for HV? > > "kvm_userspace_exit" is implemented on both HV and PR. "kvm_exit" is PR > only, but I'm hoping we can get it working in HV as well. > > > Or actually use 'kvm_exit' if it exists and fall back to 'kvm_userspace_exit', > > so that if HV starts providing 'kvm_exit' the script will pick that up without > > further changes. > > They are completely different things. "kvm_userspace_exit" tells us > which exits we take from KVM -> QEMU. "kvm_exit" tells us which exits we > take from guest -> KVM. > > In fact, IIRC x86 also implements kvm_userspace_exit - or at least > something very similar to it. It's a completely separate category. Right. Everyone implements kvm_userspace_exit, it's in virt/kvm/kvm_main.c > Maybe it should be a command line switch to distinguish between the count types? Or just we always read the kvm_userspace_exit counts, and if we find kvm_exit we expose that as well - with an arch specific set of reasons. cheers
On 18.06.14 03:37, Michael Ellerman wrote: > On Wed, 2014-06-18 at 02:59 +0200, Alexander Graf wrote: >> On 18.06.14 02:50, Michael Ellerman wrote: >>> On Tue, 2014-06-17 at 10:27 +0200, Alexander Graf wrote: >>>> On 17.06.14 09:54, Michael Ellerman wrote: >>>>> Add support for powerpc platforms. We use uname -m, which allows us to >>>>> detect ppc, ppc64 and ppc64le/el. >>>>> >>>>> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> >>>> Could you please add support for PR KVM tracepoints along the way? There >>>> we do know the exit reason for every single guest <-> host transition. I >>>> would like to move to a similar model with HV in the future, so we can >>>> hopefully just reuse this by then. >>> So I think what you're saying is you want it to somehow support using >>> 'kvm_exit' for PR and 'kvm_userspace_exit' for HV? >> "kvm_userspace_exit" is implemented on both HV and PR. "kvm_exit" is PR >> only, but I'm hoping we can get it working in HV as well. >> >>> Or actually use 'kvm_exit' if it exists and fall back to 'kvm_userspace_exit', >>> so that if HV starts providing 'kvm_exit' the script will pick that up without >>> further changes. >> They are completely different things. "kvm_userspace_exit" tells us >> which exits we take from KVM -> QEMU. "kvm_exit" tells us which exits we >> take from guest -> KVM. >> >> In fact, IIRC x86 also implements kvm_userspace_exit - or at least >> something very similar to it. It's a completely separate category. > Right. Everyone implements kvm_userspace_exit, it's in virt/kvm/kvm_main.c > >> Maybe it should be a command line switch to distinguish between the count types? > Or just we always read the kvm_userspace_exit counts, and if we find kvm_exit > we expose that as well - with an arch specific set of reasons. That would work too. I'm not sure how interesting the exit count of kvm_userspace_exit really is, but it certainly works for me to have it available as well. Alex
Thanks, applied the series at last. Paolo On 17/06/2014 09:54, Michael Ellerman wrote: > Add support for powerpc platforms. We use uname -m, which allows us to > detect ppc, ppc64 and ppc64le/el. > > Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> > --- > scripts/kvm/kvm_stat | 18 ++++++++++++++++-- > 1 file changed, 16 insertions(+), 2 deletions(-) > > diff --git a/scripts/kvm/kvm_stat b/scripts/kvm/kvm_stat > index 2e0f5ed..db867fe 100755 > --- a/scripts/kvm/kvm_stat > +++ b/scripts/kvm/kvm_stat > @@ -12,7 +12,7 @@ > # the COPYING file in the top-level directory. > > import curses > -import sys, os, time, optparse > +import sys, os, time, optparse, ctypes > > class DebugfsProvider(object): > def __init__(self): > @@ -199,7 +199,21 @@ def s390_init(): > 'sc_perf_evt_open' : 331, > }) > > +def ppc_init(): > + globals().update({ > + 'sc_perf_evt_open' : 319, > + 'ioctl_numbers' : { > + 'SET_FILTER' : 0x80002406 | (ctypes.sizeof(ctypes.c_char_p) << 16), > + 'ENABLE' : 0x20002400, > + 'DISABLE' : 0x20002401, > + } > + }) > + > def detect_platform(): > + if os.uname()[4].startswith('ppc'): > + ppc_init() > + return > + > for line in file('/proc/cpuinfo').readlines(): > if line.startswith('flags'): > for flag in line.split(): > @@ -220,7 +234,7 @@ def invert(d): > for f in filters: > filters[f] = (filters[f][0], invert(filters[f][1])) > > -import ctypes, struct, array > +import struct, array > > libc = ctypes.CDLL('libc.so.6') > syscall = libc.syscall >
On Fri, 2014-10-31 at 16:36 +0100, Paolo Bonzini wrote:
> Thanks, applied the series at last.
Thanks.
cheers
diff --git a/scripts/kvm/kvm_stat b/scripts/kvm/kvm_stat index 2e0f5ed..db867fe 100755 --- a/scripts/kvm/kvm_stat +++ b/scripts/kvm/kvm_stat @@ -12,7 +12,7 @@ # the COPYING file in the top-level directory. import curses -import sys, os, time, optparse +import sys, os, time, optparse, ctypes class DebugfsProvider(object): def __init__(self): @@ -199,7 +199,21 @@ def s390_init(): 'sc_perf_evt_open' : 331, }) +def ppc_init(): + globals().update({ + 'sc_perf_evt_open' : 319, + 'ioctl_numbers' : { + 'SET_FILTER' : 0x80002406 | (ctypes.sizeof(ctypes.c_char_p) << 16), + 'ENABLE' : 0x20002400, + 'DISABLE' : 0x20002401, + } + }) + def detect_platform(): + if os.uname()[4].startswith('ppc'): + ppc_init() + return + for line in file('/proc/cpuinfo').readlines(): if line.startswith('flags'): for flag in line.split(): @@ -220,7 +234,7 @@ def invert(d): for f in filters: filters[f] = (filters[f][0], invert(filters[f][1])) -import ctypes, struct, array +import struct, array libc = ctypes.CDLL('libc.so.6') syscall = libc.syscall
Add support for powerpc platforms. We use uname -m, which allows us to detect ppc, ppc64 and ppc64le/el. Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> --- scripts/kvm/kvm_stat | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-)