RFC: Linux: disable APERF/MPERF feature in PV kernels

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Quoted 14 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Seems reasonable to me. Do all those RDMSR traps have a measurable performance effect?

Also, is there a symbolic constant for that bit?

J

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

measurement.
The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Awaiting your comments.

Regards,
Andre.

P.S. Of course this doesn't fix pure userland software like cpupower, but I would consider this in the user's responsibility to

not use these tools in Dom0, but instead use xenpm.

Signature · 10 lines

Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany

commit e802e47d85314b4541288e4a19d057e2ea885a28
Author: Andre Przywara <[email protected]>
Date:   Tue May 22 15:13:07 2012 +0200

    filter APERFMPERF feature in Xen to avoid kernel internal usage
    
    Signed-off-by: Andre Przywara <[email protected]>

arch/x86/xen/enlighten.c +5 -0

index 95dccce..71252d5 100644
@@ -240,6 +240,11 @@ static void xen_cpuid(unsigned int *ax, unsigned int *bx,
 		*dx = cpuid_leaf5_edx_val;
 		return;
 
+	case 6:
+		/* Disabling APERFMPERF for kernel usage */
+		maskecx = ~(1U << 0);
+		break;
+
 	case 0xb:
 		/* Suppress extended topology stuff */
 		maskebx = 0;

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Quoted 14 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Seems reasonable to me. Do all those RDMSR traps have a measurable performance effect?

Also, is there a symbolic constant for that bit?

Quoted 10 lines · Konrad Rzeszutek Wilk

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

arch/x86/kernel/cpu/sched.c contains code to read out and compute APERF/MPERF registers. I added a Xen debug-key to dump a usage counter added in traps.c and thus could prove that it is actually the kernel that accesses these registers.
As far as I understood this the idea is to learn about boosting and down-clocking (P-states) to get a fairer view on the actual computing time a process consumed.

Quoted 18 lines

Quoted 17 lines

measurement.
The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Awaiting your comments.

Regards,
Andre.

P.S. Of course this doesn't fix pure userland software like cpupower, but I would consider this in the user's responsibility to

Quoted 2 lines

Which would not work anymore as the cpufreq support is disabled when it boots under Xen.

Do you mean with "anymore" in a future kernel? I tested this on 3.4.0 and cpupower monitor worked fine. Right, cpufreq is not enabled, but cpupower uses the /dev/cpu/<n>/msr device file to directly read the MSRs. So I get this output if run on an idle Dom0:

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

Quoted 11 lines

measurement.
The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Awaiting your comments.

Quoted 5 lines

Regards,
Andre.

P.S. Of course this doesn't fix pure userland software like cpupower, but I would consider this in the user's responsibility to

Which would not work anymore as the cpufreq support is disabled when it boots under Xen.

Quoted 36 lines

not use these tools in Dom0, but instead use xenpm.

--
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany

commit e802e47d85314b4541288e4a19d057e2ea885a28

Author: Andre Przywara<[email protected]>
Date:   Tue May 22 15:13:07 2012 +0200

filter APERFMPERF feature in Xen to avoid kernel internal usage

Signed-off-by Andre Przywara<[email protected]>

arch/x86/xen/enlighten.c +5 -0

index 95dccce..71252d5 100644
@@ -240,6 +240,11 @@ static void xen_cpuid(unsigned int *ax, unsigned int *bx,
  		*dx = cpuid_leaf5_edx_val;
  		return;
 
+	case 6:
+		/* Disabling APERFMPERF for kernel usage */
+		maskecx = ~(1U<<  0);
+		break;
+
  	case 0xb:
  		/* Suppress extended topology stuff */
  		maskebx = 0;

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Quoted 12 lines · Andre Przywara

Quoted 10 lines · Konrad Rzeszutek Wilk

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

Quoted 7 lines

arch/x86/kernel/cpu/sched.c contains code to read out and compute APERF/MPERF registers. I added a Xen debug-key to dump a usage counter added in traps.c and thus could prove that it is actually the kernel that accesses these registers.
As far as I understood this the idea is to learn about boosting and down-clocking (P-states) to get a fairer view on the actual computing time a process consumed.

Looks like its looking for this:

X86_FEATURE_APERFMPERF

Perhaps masking that should do it? Something along this in enlighten.c:

          cpuid_leaf1_edx_mask =
                  ~((1<<  X86_FEATURE_MCE)  |  /* disable MCE */
                    (1<<  X86_FEATURE_MCA)  |  /* disable MCA */
                    (1<<  X86_FEATURE_MTRR) |  /* disable MTRR */
                    (1<<  X86_FEATURE_ACC));   /* thermal monitoring

would be more appropiate?

Or is that attribute on a different leaf?

Quoted 21 lines

Quoted 20 lines

Quoted 10 lines

measurement.
The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Quoted 7 lines

Awaiting your comments.

Regards,
Andre.

P.S. Of course this doesn't fix pure userland software like cpupower, but I would consider this in the user's responsibility to

Which would not work anymore as the cpufreq support is disabled when it boots under Xen.

Quoted 4 lines

Do you mean with "anymore" in a future kernel? I tested this on 3.4.0 and cpupower monitor worked fine. Right, cpufreq is not enabled, but cpupower uses the /dev/cpu/<n>/msr device file to directly read the MSRs. So I get this output if run on an idle Dom0:

Ahh. Neat. Will have to play with that.

Quoted 1 line

Quoted 1 line · Andre Przywara

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Awaiting your comments.

Quoted 1 line

Quoted 1 line · Malcolm Crossley

Quoted 18 lines · Jeremy Fitzhardinge

Quoted 14 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Seems reasonable to me. Do all those RDMSR traps have a measurable performance effect?

Quoted 8 lines

Also, is there a symbolic constant for that bit?

J

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Hi,

I've attached a patch which masks the matching CPUID leaves in the Xen pv_cpuid function.
Should the logic in pv_cpuid be changed to only pass through explictly allowed CPUID leaves rather than masking
them using case statements?

Malcolm

Quoted 14 lines

Quoted 1 line

Quoted 1 line · Andre Przywara

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Quoted 1 line

Awaiting your comments.

First of all I'm of the opinion that this indeed should not be masked in the hypervisor - there's no reason to disallow the guest to read these registers (but we should of course deny writes as long as Xen is controlling P-states, which we do).

Next I'd like to note that in our kernels we simply don't build arch/x86/kernel/cpu/sched.o. Together with CPU_FREQ being suppressed, there's no consumer of the feature flag in our kernels.

So I would think that your suggested change is appropriate, but I'm adding Konrad to Cc as these days he's the one to pick this up.

Jan

Quoted 1 line

Quoted 1 line · Andre Przywara

Quoted 4 lines · Jan Beulich

Next I'd like to note that in our kernels we simply don't build arch/x86/kernel/cpu/sched.o. Together with CPU_FREQ being suppressed, there's no consumer of the feature flag in our kernels.

With "our kernels" you mean OpenSuSE/SLES kernels? I quickly checked upstream as well as the repos on kernel.opensuse.org. In all of them sched.o is unconditionally included in the Makefile.
So is there a build patch to exclude this file for builds of distro Xen kernels?

Quoted 7 lines

Quoted 1 line

Quoted 1 line · Andre Przywara

Quoted 4 lines · Jan Beulich

Next I'd like to note that in our kernels we simply don't build arch/x86/kernel/cpu/sched.o. Together with CPU_FREQ being suppressed, there's no consumer of the feature flag in our kernels.

Quoted 5 lines

With "our kernels" you mean OpenSuSE/SLES kernels? I quickly checked upstream as well as the repos on kernel.opensuse.org. In all of them sched.o is unconditionally included in the Makefile.
So is there a build patch to exclude this file for builds of distro Xen kernels?

Did you perhaps overlook

disabled-obj-$(CONFIG_XEN) := hypervisor.o mshyperv.o perfctr-watchdog.o \
perf_event.o sched.o vmware.o

in that very Makefile?

Quoted 1 line

Quoted 1 line · Andre Przywara

But this works only for SuSE since you build different kernels for Xen and without Xen, right?
Starting with 3.x I now have only a single kernel image which I use for both native and as Dom0.

So the patch to disable APERFMPERF is still useful, at least for upstream, right?

Quoted 14 lines

Quoted 1 line

Quoted 1 line · Andre Przywara

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the measurement. The attached patch explicitly disables this CPU capability inside the Linux kernel, I couldn't measure any APERF/MPERF reads anymore with the patch applied.
I am not sure if the PVOPS code is the right place to fix this, we could as well do it in the HV's xen/arch/x86/traps.c:pv_cpuid(). Also when the Dom0 VCPUs are pinned, we could allow this, but I am not sure if it's worth to do so.

Quoted 1 line

Awaiting your comments.

First of all I'm of the opinion that this indeed should not be masked in the hypervisor - there's no reason to disallow the guest to read these registers (but we should of course deny writes as long as Xen is controlling P-states, which we do).

Next I'd like to note that in our kernels we simply don't build arch/x86/kernel/cpu/sched.o. Together with CPU_FREQ being suppressed, there's no consumer of the feature flag in our kernels.

So I would think that your suggested change is appropriate, but I'm adding Konrad to Cc as these days he's the one to pick this up.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Quoted 1 line

Quoted 1 line · Andrew Cooper

Quoted 4 lines · Jan Beulich

First of all I'm of the opinion that this indeed should not be masked in the hypervisor - there's no reason to disallow the guest to read these registers (but we should of course deny writes as long as Xen is controlling P-states, which we do).

I am sorry but I am going to have to disagree with you on this point.

We should not be advertising this feature to any guest at all if we can't provide an implementation which works as native expects. Else we are failing in our job of virtualisation.

There is 'dom0_vcpus_pin'[1] which identity pins dom0 vcpus, and prevents update of the affinity masks, and appears to conditionally allow access to certain MSRs. I think it would be fine to expose this feature iff dom0s vcpus are pinned in this fashion. That way, the measurement should succeed, even if dom0 only has read access to the MSRs.

Quoted 11 lines

Quoted 1 line

Quoted 1 line · Andrew Cooper

Quoted 4 lines · Jan Beulich

First of all I'm of the opinion that this indeed should not be masked in the hypervisor - there's no reason to disallow the guest to read these registers (but we should of course deny writes as long as Xen is controlling P-states, which we do).

I am sorry but I am going to have to disagree with you on this point.

We should not be advertising this feature to any guest at all if we can't provide an implementation which works as native expects. Else we are failing in our job of virtualisation.

That's perhaps a matter of the position you take - for HVM, I would agree with yours, but there's many more aspects (not the least related to accessing other MSRs) that we fail to "properly" virtualize for PV guests - my position is that it is the nature of PV that guest kernels have to be aware of being virtualized (and hence stay away from doing certain things unless [they think] they know what they're doing).

Quoted 5 lines

There is 'dom0_vcpus_pin'[1] which identity pins dom0 vcpus, and prevents update of the affinity masks, and appears to conditionally allow access to certain MSRs. I think it would be fine to expose this feature iff dom0s vcpus are pinned in this fashion. That way, the measurement should succeed, even if dom0 only has read access to the MSRs.

Restricting it to this case would be too restrictive - it really makes sense at any time where the vCPU's affinity has exactly one bit set (or to be precise, the intersection of it and the set of online pCPU-s).

Jan

Quoted 33 lines · Konrad Rzeszutek Wilk

Quoted 12 lines · Andre Przywara

Quoted 10 lines · Konrad Rzeszutek Wilk

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

Quoted 7 lines

arch/x86/kernel/cpu/sched.c contains code to read out and compute APERF/MPERF registers. I added a Xen debug-key to dump a usage counter added in traps.c and thus could prove that it is actually the kernel that accesses these registers.
As far as I understood this the idea is to learn about boosting and down-clocking (P-states) to get a fairer view on the actual computing time a process consumed.

Looks like its looking for this:

X86_FEATURE_APERFMPERF

Perhaps masking that should do it? Something along this in enlighten.c:

         cpuid_leaf1_edx_mask =
                 ~((1<<  X86_FEATURE_MCE)  |  /* disable MCE */
                   (1<<  X86_FEATURE_MCA)  |  /* disable MCA */
                   (1<<  X86_FEATURE_MTRR) |  /* disable MTRR */
                   (1<<  X86_FEATURE_ACC));   /* thermal monitoring

Quoted 3 lines

would be more appropiate?

Or is that attribute on a different leaf?

Right, it is bit 0 on level 6. That's why I couldn't use any of the predefined masks and I didn't feel like inventing a new one just for this single bit.
We could as well explicitly use clear_cpu_cap somewhere, but I didn't find any code place in the Xen tree already doing this, instead it looks like it belongs to where I put it (we handle leaf 5 in a special way already here)

Quoted 11 lines

Quoted 10 lines

Quoted 5 lines

Quoted 2 lines

P.S. Of course this doesn't fix pure userland software like cpupower, but I would consider this in the user's responsibility to

Which would not work anymore as the cpufreq support is disabled when it boots under Xen.

Do you mean with "anymore" in a future kernel? I tested this on 3.4.0 and cpupower monitor worked fine. Right, cpufreq is not enabled, but cpupower uses the /dev/cpu/<n>/msr device file to directly read the MSRs. So I get this output if run on an idle Dom0:

Quoted 1 line

Ahh. Neat. Will have to play with that.

Bad news is we cannot forbid cpupower querying the feature directly using the CPUID instruction in PV guests. Only we could patch it to use /proc/cpuinfo readout instead, as this reflects the kernel view of available features. With my patch aperfmperf is no longer there.

Quoted 24 lines · Jan Beulich

Quoted 11 lines

Quoted 1 line

Quoted 1 line · Andrew Cooper

Quoted 4 lines · Jan Beulich

First of all I'm of the opinion that this indeed should not be masked in the hypervisor - there's no reason to disallow the guest to read these registers (but we should of course deny writes as long as Xen is controlling P-states, which we do).

I am sorry but I am going to have to disagree with you on this point.

We should not be advertising this feature to any guest at all if we can't provide an implementation which works as native expects. Else we are failing in our job of virtualisation.

That's perhaps a matter of the position you take - for HVM, I would agree with yours, but there's many more aspects (not the least related to accessing other MSRs) that we fail to "properly" virtualize for PV guests - my position is that it is the nature of PV that guest kernels have to be aware of being virtualized (and hence stay away from doing certain things unless [they think] they know what they're doing).

Quoted 5 lines

There is 'dom0_vcpus_pin'[1] which identity pins dom0 vcpus, and prevents update of the affinity masks, and appears to conditionally allow access to certain MSRs. I think it would be fine to expose this feature iff dom0s vcpus are pinned in this fashion. That way, the measurement should succeed, even if dom0 only has read access to the MSRs.

Quoted 7 lines

Restricting it to this case would be too restrictive - it really makes sense at any time where the vCPU's affinity has exactly one bit set (or to be precise, the intersection of it and the set of online pCPU-s).

Jan

That is unfortunately too lax. You also need to be able to guarantee that the affinity mask is not updated (and vcpu rescheduled) while in the middle of a measurement. Xen cant sensibly work out if or when a guest is taking a measurement, nor can dom0. So the only safe solution I can see is for Xen to prevent the affinity masks from ever being updated. With more thought, this would also preclude migration of a guest to another host.

Quoted 38 lines · Andre Przywara

Quoted 33 lines · Konrad Rzeszutek Wilk

Quoted 12 lines · Andre Przywara

Quoted 10 lines · Konrad Rzeszutek Wilk

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

Quoted 7 lines

arch/x86/kernel/cpu/sched.c contains code to read out and compute APERF/MPERF registers. I added a Xen debug-key to dump a usage counter added in traps.c and thus could prove that it is actually the kernel that accesses these registers.
As far as I understood this the idea is to learn about boosting and down-clocking (P-states) to get a fairer view on the actual computing time a process consumed.

Looks like its looking for this:

X86_FEATURE_APERFMPERF

Perhaps masking that should do it? Something along this in enlighten.c:

          cpuid_leaf1_edx_mask =
                  ~((1<<   X86_FEATURE_MCE)  |  /* disable MCE */
                    (1<<   X86_FEATURE_MCA)  |  /* disable MCA */
                    (1<<   X86_FEATURE_MTRR) |  /* disable MTRR */
                    (1<<   X86_FEATURE_ACC));   /* thermal monitoring

Quoted 3 lines

would be more appropiate?

Or is that attribute on a different leaf?

Quoted 7 lines

Right, it is bit 0 on level 6. That's why I couldn't use any of the predefined masks and I didn't feel like inventing a new one just for this single bit.
We could as well explicitly use clear_cpu_cap somewhere, but I didn't find any code place in the Xen tree already doing this, instead it looks like it belongs to where I put it (we handle leaf 5 in a special way already here)

OK, can you resend the patch please, looking similar to what you sent earlier, but do use a #define if possible (you can have the #define in that file) and an comment explaining why this is neccessary - and point to the Linux source code that uses this.

Looks like a patch to cpupower should be cooked up too?

Quoted 47 lines · Konrad Rzeszutek Wilk

Quoted 38 lines · Andre Przywara

Quoted 33 lines · Konrad Rzeszutek Wilk

Quoted 12 lines · Andre Przywara

Quoted 10 lines · Konrad Rzeszutek Wilk

Quoted 7 lines · Andre Przywara

Hi,

while testing some APERF/MPERF semantics I discovered that this feature is enabled in Xen Dom0, but is not reliable.
The Linux kernel's scheduler uses this feature if it sees the CPUID bit, leading to costly RDMSR traps (a few 100,000s during a kernel compile) and bogus values due to VCPU migration during the

Can you point me to the Linux scheduler code that does this? Thanks.

Quoted 7 lines

arch/x86/kernel/cpu/sched.c contains code to read out and compute APERF/MPERF registers. I added a Xen debug-key to dump a usage counter added in traps.c and thus could prove that it is actually the kernel that accesses these registers.
As far as I understood this the idea is to learn about boosting and down-clocking (P-states) to get a fairer view on the actual computing time a process consumed.

Looks like its looking for this:

X86_FEATURE_APERFMPERF

Perhaps masking that should do it? Something along this in enlighten.c:

cpuid_leaf1_edx_mask =
~((1<< X86_FEATURE_MCE) | /* disable MCE */
(1<< X86_FEATURE_MCA) | /* disable MCA */
(1<< X86_FEATURE_MTRR) | /* disable MTRR */
(1<< X86_FEATURE_ACC)); /* thermal monitoring

Quoted 3 lines

would be more appropiate?

Or is that attribute on a different leaf?

Quoted 7 lines

Right, it is bit 0 on level 6. That's why I couldn't use any of the predefined masks and I didn't feel like inventing a new one just for this single bit.
We could as well explicitly use clear_cpu_cap somewhere, but I didn't find any code place in the Xen tree already doing this, instead it looks like it belongs to where I put it (we handle leaf 5 in a special way already here)

Quoted 4 lines

OK, can you resend the patch please, looking similar to what you sent earlier, but do use a #define if possible (you can have the #define in that file) and an comment explaining why this is neccessary - and point to the Linux source code that uses this.

Well, I was about to do this and wanted to see if this has any performance impact - only to discover that 3.4 does not trigger it anymore. After some debugging it turns out the guy reading APERF/MPERF was not arch/x86/kernel/cpu/sched.c, but drivers/cpufreq/mperf.c. So with disabling cpufreq the only real user is gone already. So the patch is kind of pointless as it on 3.4 with cpufreq already disabled. Remains to be investigated why sched.c is not called (I added a usage counter, it stays at zero).

To avoid future mis-uses of APERF/MPERF by the kernel, I'd like to add the patch anyway. I will send it again when I have a clearer picture of this.

.. snip..

Quoted 1 line

Looks like a patch to cpupower should be cooked up too?

I will contact the author.