| Message ID | 1355400716-21359-2-git-send-email-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Dec 13, 2012 at 01:11:56PM +0100, Paolo Bonzini wrote: > This patch adds two scenarios to the vmexit test. One is a very simple > PIO case that is handled in the kernel (reading from ELCR). The other > is an unmapped PIO that is handled in userspace. The difference > between the two is roughly the cost of a userspace exit; the existing > inl_from_pmtimer test is not precise enough, because the device model > has a pretty high cost. > > With the current code in KVM, the difference between inl_from_kernel > and vmcall is the cost of emulation. This is arguably a bug, because > inl_from_kernel need not go through the whole emulation stuff. > You do not what to have separate code path for something that doesn't even matter. out does not go through the emulator since it is used for virtio signaling (and very simple to do too). > Example: > > vmcall 3898 > inl_from_pmtimer 24615 > inl_from_qemu 20574 > inl_from_kernel 7237 > > So the cost of exiting to userspace is 13000 cycles on this machine, > and the cost of emulation is 3300 cycles. > > Suggested-by: Avi Kivity <avi.kivity@gmail.com> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > x86/vmexit.c | 19 +++++++++++++++++++ > 1 file changed, 19 insertions(+) > > diff --git a/x86/vmexit.c b/x86/vmexit.c > index ad8ab55..545a8bc 100644 > --- a/x86/vmexit.c > +++ b/x86/vmexit.c > @@ -4,6 +4,13 @@ > #include "processor.h" > #include "atomic.h" > > +static unsigned int inb(unsigned short port) > +{ > + unsigned int val; > + asm volatile("xorl %0, %0; inb %w1, %b0" : "=a"(val) : "Nd"(port)); > + return val; > +} > + > static unsigned int inl(unsigned short port) > { > unsigned int val; > @@ -82,6 +89,16 @@ static void inl_pmtimer(void) > inl(0xb008); > } > > +static void inl_nop_qemu(void) > +{ > + inl(0x1234); > +} > + > +static void inl_nop_kernel(void) > +{ > + inb(0x4d0); > +} > + > static void ple_round_robin(void) > { > struct counter { > @@ -116,6 +133,8 @@ static struct test { > { mov_to_cr8, "mov_to_cr8" , .parallel = 1, }, > #endif > { inl_pmtimer, "inl_from_pmtimer", .parallel = 1, }, > + { inl_nop_qemu, "inl_from_qemu", .parallel = 1 }, > + { inl_nop_kernel, "inl_from_kernel", .parallel = 1 }, > { ipi, "ipi", is_smp, .parallel = 0, }, > { ipi_halt, "ipi+halt", is_smp, .parallel = 0, }, > { ple_round_robin, "ple-round-robin", .parallel = 1 }, > -- > 1.8.0.2 > > -- > To unsubscribe from this list: send the line "unsubscribe kvm" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- Gleb. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/x86/vmexit.c b/x86/vmexit.c index ad8ab55..545a8bc 100644 --- a/x86/vmexit.c +++ b/x86/vmexit.c @@ -4,6 +4,13 @@ #include "processor.h" #include "atomic.h" +static unsigned int inb(unsigned short port) +{ + unsigned int val; + asm volatile("xorl %0, %0; inb %w1, %b0" : "=a"(val) : "Nd"(port)); + return val; +} + static unsigned int inl(unsigned short port) { unsigned int val; @@ -82,6 +89,16 @@ static void inl_pmtimer(void) inl(0xb008); } +static void inl_nop_qemu(void) +{ + inl(0x1234); +} + +static void inl_nop_kernel(void) +{ + inb(0x4d0); +} + static void ple_round_robin(void) { struct counter { @@ -116,6 +133,8 @@ static struct test { { mov_to_cr8, "mov_to_cr8" , .parallel = 1, }, #endif { inl_pmtimer, "inl_from_pmtimer", .parallel = 1, }, + { inl_nop_qemu, "inl_from_qemu", .parallel = 1 }, + { inl_nop_kernel, "inl_from_kernel", .parallel = 1 }, { ipi, "ipi", is_smp, .parallel = 0, }, { ipi_halt, "ipi+halt", is_smp, .parallel = 0, }, { ple_round_robin, "ple-round-robin", .parallel = 1 },
This patch adds two scenarios to the vmexit test. One is a very simple PIO case that is handled in the kernel (reading from ELCR). The other is an unmapped PIO that is handled in userspace. The difference between the two is roughly the cost of a userspace exit; the existing inl_from_pmtimer test is not precise enough, because the device model has a pretty high cost. With the current code in KVM, the difference between inl_from_kernel and vmcall is the cost of emulation. This is arguably a bug, because inl_from_kernel need not go through the whole emulation stuff. Example: vmcall 3898 inl_from_pmtimer 24615 inl_from_qemu 20574 inl_from_kernel 7237 So the cost of exiting to userspace is 13000 cycles on this machine, and the cost of emulation is 3300 cycles. Suggested-by: Avi Kivity <avi.kivity@gmail.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- x86/vmexit.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)