When commit c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") switched the initialization of cpuc->guest_switch_msrs to use compound literals, it screwed up the boolean logic:
+ u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable; ... - arr[0].guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask; - arr[0].guest &= ~(cpuc->pebs_enabled & x86_pmu.pebs_capable); + .guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),
Before the patch, the value of arr[0].guest would have been intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask. The intent is to always treat PEBS events as host-only because, while the guest runs, there is no way to tell the processor about the virtual address where to put PEBS records intended for the host.
Unfortunately, the new expression can be expanded to
(intel_ctrl & ~cpuc->intel_ctrl_host_mask) | (intel_ctrl & ~pebs_mask)
which makes no sense; it includes any bit that isn't *both* marked as exclude_guest and using PEBS. So, reinstate the old logic. Another way to write it could be "intel_ctrl & ~(cpuc->intel_ctrl_host_mask | pebs_mask)", presumably the intention of the author of the faulty. However, I personally find the repeated application of A AND NOT B to be a bit more readable.
This shows up as guest failures when running concurrent long-running perf workloads on the host, and was reported to happen with rcutorture. All guests on a given host would die simultaneously with something like an instruction fault or a segmentation violation.
Reported-by: Paul E. McKenney paulmck@kernel.org Analyzed-by: Sean Christopherson seanjc@google.com Tested-by: Paul E. McKenney paulmck@kernel.org Cc: stable@vger.kernel.org Fixes: c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") Signed-off-by: Paolo Bonzini pbonzini@redhat.com --- arch/x86/events/intel/core.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c index ce1c777227b4..0f2786d4e405 100644 --- a/arch/x86/events/intel/core.c +++ b/arch/x86/events/intel/core.c @@ -4051,12 +4051,17 @@ static struct perf_guest_switch_msr *intel_guest_get_msrs(int *nr, void *data) u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable; int global_ctrl, pebs_enable;
+ /* + * In addition to obeying exclude_guest/exclude_host, remove bits being + * used for PEBS when running a guest, because PEBS writes to virtual + * addresses (not physical addresses). + */ *nr = 0; global_ctrl = (*nr)++; arr[global_ctrl] = (struct perf_guest_switch_msr){ .msr = MSR_CORE_PERF_GLOBAL_CTRL, .host = intel_ctrl & ~cpuc->intel_ctrl_guest_mask, - .guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask), + .guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask, };
if (!x86_pmu.pebs)
On 2024-01-04 10:39 a.m., Paolo Bonzini wrote:
When commit c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") switched the initialization of cpuc->guest_switch_msrs to use compound literals, it screwed up the boolean logic:
- u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable;
...
- arr[0].guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask;
- arr[0].guest &= ~(cpuc->pebs_enabled & x86_pmu.pebs_capable);
.guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),Before the patch, the value of arr[0].guest would have been intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask. The intent is to always treat PEBS events as host-only because, while the guest runs, there is no way to tell the processor about the virtual address where to put PEBS records intended for the host.
Unfortunately, the new expression can be expanded to
(intel_ctrl & ~cpuc->intel_ctrl_host_mask) | (intel_ctrl & ~pebs_mask)
which makes no sense; it includes any bit that isn't *both* marked as exclude_guest and using PEBS. So, reinstate the old logic.
I think the old logic will completely disable the PEBS in guest capability. Because the counter which is assigned to a guest PEBS event will also be set in the pebs_mask. The old logic disable the counter in GLOBAL_CTRL in guest. Nothing will be counted.
Like once proposed a fix in the intel_guest_get_msrs(). https://lore.kernel.org/lkml/20231129095055.88060-1-likexu@tencent.com/ It should work for the issue.
Ideally, we should prevent the host PEBS from profiling a guest via rejecting the event creation in the perf. But I couldn't find a good way to distinguish host-created PEBS and guest-created PEBS. So Like's proposal should be a good alternative so far.
Thanks, Kan
Another way to write it could be "intel_ctrl & ~(cpuc->intel_ctrl_host_mask | pebs_mask)", presumably the intention of the author of the faulty. However, I personally find the repeated application of A AND NOT B to be a bit more readable.
This shows up as guest failures when running concurrent long-running perf workloads on the host, and was reported to happen with rcutorture. All guests on a given host would die simultaneously with something like an instruction fault or a segmentation violation.
Reported-by: Paul E. McKenney paulmck@kernel.org Analyzed-by: Sean Christopherson seanjc@google.com Tested-by: Paul E. McKenney paulmck@kernel.org Cc: stable@vger.kernel.org Fixes: c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") Signed-off-by: Paolo Bonzini pbonzini@redhat.com
arch/x86/events/intel/core.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c index ce1c777227b4..0f2786d4e405 100644 --- a/arch/x86/events/intel/core.c +++ b/arch/x86/events/intel/core.c @@ -4051,12 +4051,17 @@ static struct perf_guest_switch_msr *intel_guest_get_msrs(int *nr, void *data) u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable; int global_ctrl, pebs_enable;
- /*
* In addition to obeying exclude_guest/exclude_host, remove bits being* used for PEBS when running a guest, because PEBS writes to virtual* addresses (not physical addresses).*/
.guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),
.guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask,if (!x86_pmu.pebs)
On Thu, Jan 04, 2024, Liang, Kan wrote:
On 2024-01-04 10:39 a.m., Paolo Bonzini wrote:
When commit c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") switched the initialization of cpuc->guest_switch_msrs to use compound literals, it screwed up the boolean logic:
- u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable;
...
- arr[0].guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask;
- arr[0].guest &= ~(cpuc->pebs_enabled & x86_pmu.pebs_capable);
.guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),Before the patch, the value of arr[0].guest would have been intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask. The intent is to always treat PEBS events as host-only because, while the guest runs, there is no way to tell the processor about the virtual address where to put PEBS records intended for the host.
Unfortunately, the new expression can be expanded to
(intel_ctrl & ~cpuc->intel_ctrl_host_mask) | (intel_ctrl & ~pebs_mask)
which makes no sense; it includes any bit that isn't *both* marked as exclude_guest and using PEBS. So, reinstate the old logic.
I think the old logic will completely disable the PEBS in guest capability. Because the counter which is assigned to a guest PEBS event will also be set in the pebs_mask. The old logic disable the counter in GLOBAL_CTRL in guest. Nothing will be counted.
Like once proposed a fix in the intel_guest_get_msrs(). https://lore.kernel.org/lkml/20231129095055.88060-1-likexu@tencent.com/ It should work for the issue.
No, that patch only affects the path where hardware supports enabling PEBS in the the guest, i.e. intel_guest_get_msrs() will bail before getting to that code due to the lack of x86_pmu.pebs_ept support, which IIUC is all pre-Icelake Intel CPUs.
if (!kvm_pmu || !x86_pmu.pebs_ept) return arr;
On 2024-01-04 1:22 p.m., Sean Christopherson wrote:
On Thu, Jan 04, 2024, Liang, Kan wrote:
On 2024-01-04 10:39 a.m., Paolo Bonzini wrote:
When commit c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS") switched the initialization of cpuc->guest_switch_msrs to use compound literals, it screwed up the boolean logic:
- u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable;
...
- arr[0].guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask;
- arr[0].guest &= ~(cpuc->pebs_enabled & x86_pmu.pebs_capable);
.guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),Before the patch, the value of arr[0].guest would have been intel_ctrl & ~cpuc->intel_ctrl_host_mask & ~pebs_mask. The intent is to always treat PEBS events as host-only because, while the guest runs, there is no way to tell the processor about the virtual address where to put PEBS records intended for the host.
Unfortunately, the new expression can be expanded to
(intel_ctrl & ~cpuc->intel_ctrl_host_mask) | (intel_ctrl & ~pebs_mask)
which makes no sense; it includes any bit that isn't *both* marked as exclude_guest and using PEBS. So, reinstate the old logic.
I think the old logic will completely disable the PEBS in guest capability. Because the counter which is assigned to a guest PEBS event will also be set in the pebs_mask. The old logic disable the counter in GLOBAL_CTRL in guest. Nothing will be counted.
Like once proposed a fix in the intel_guest_get_msrs(). https://lore.kernel.org/lkml/20231129095055.88060-1-likexu@tencent.com/ It should work for the issue.
No, that patch only affects the path where hardware supports enabling PEBS in the the guest, i.e. intel_guest_get_msrs() will bail before getting to that code due to the lack of x86_pmu.pebs_ept support, which IIUC is all pre-Icelake Intel CPUs.
if (!kvm_pmu || !x86_pmu.pebs_ept) return arr;
True, we have to disable all PEBS counters for pre-ICL as well.
I think what I missed is that the disable here is temporary. The arr[global_ctrl].guest will be updated later for the x86_pmu.pebs_ept platform, so the guest PEBS event should still work.
The patch looks good to me.
Reviewed-by: Kan Liang kan.liang@linux.intel.com
Thanks, Kan
linux-stable-mirror@lists.linaro.org
-
Liang, Kan -
Paolo Bonzini -
Sean Christopherson