| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20201109021254.79755-15-like.xu@linux.intel.com>
Date: Mon, 9 Nov 2020 10:12:51 +0800
From: Like Xu <like.xu@...ux.intel.com>
To: Peter Zijlstra <peterz@...radead.org>,
Paolo Bonzini <pbonzini@...hat.com>, kvm@...r.kernel.org
Cc: Sean Christopherson <sean.j.christopherson@...el.com>,
Vitaly Kuznetsov <vkuznets@...hat.com>,
Wanpeng Li <wanpengli@...cent.com>,
Jim Mattson <jmattson@...gle.com>,
Joerg Roedel <joro@...tes.org>,
Kan Liang <kan.liang@...ux.intel.com>, luwei.kang@...el.com,
Thomas Gleixner <tglx@...utronix.de>, wei.w.wang@...el.com,
Tony Luck <tony.luck@...el.com>,
Stephane Eranian <eranian@...gle.com>,
Mark Gross <mgross@...ux.intel.com>,
Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>,
linux-kernel@...r.kernel.org
Subject: [PATCH RFC v2 14/17] KVM: vmx/pmu: Limit pebs_interrupt_threshold in the guest DS area
If the host counter X is scheduled to the guest PEBS counter Y,
the guest ds pebs_interrupt_threshold field in guest DS area would
be changed to only ONE record before vm-entry which helps KVM
more easily and accurately handle the cross-mapping emulation
when the PEBS overflow PMI is generated.
In most cases, the guest counters would not be scheduled in a cross-mapped
way which means there is no need to change guest DS
pebs_interrupt_threshold and the applicable_counters fields in the guest
PEBS records are naturally correct. PEBS facility writes multiple PEBS
records into guest DS w/o interception and the performance is good.
AFAIK, we don't expect that changing the pebs_interrupt_threshold value
from the KVM side will break any guest PEBS drivers.
Signed-off-by: Like Xu <like.xu@...ux.intel.com>
---
arch/x86/include/asm/kvm_host.h | 1 +
arch/x86/kvm/pmu.c | 17 +++-----
arch/x86/kvm/pmu.h | 11 +++++
arch/x86/kvm/vmx/pmu_intel.c | 71 +++++++++++++++++++++++++++++++++
arch/x86/kvm/x86.c | 1 +
5 files changed, 90 insertions(+), 11 deletions(-)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index bffb384485da..77b529b8c16a 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -456,6 +456,7 @@ struct kvm_pmu {
u64 pebs_data_cfg_mask;
bool counter_cross_mapped;
+ bool need_rewrite_ds_pebs_interrupt_threshold;
/*
* The gate to release perf_events not marked in
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index f87be3c2140e..7c8e3ca5b7ad 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -471,17 +471,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
kvm_pmu_refresh(vcpu);
}
-static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
-{
- struct kvm_pmu *pmu = pmc_to_pmu(pmc);
-
- if (pmc_is_fixed(pmc))
- return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
- pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
-
- return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
-}
-
/* Release perf_events for vPMCs that have been unused for a full time slice. */
void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
{
@@ -576,4 +565,10 @@ void kvm_pmu_counter_cross_mapped_check(struct kvm_vcpu *vcpu)
break;
}
}
+
+ if (!pmu->counter_cross_mapped)
+ return;
+
+ if (pmu->need_rewrite_ds_pebs_interrupt_threshold)
+ kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
}
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index b1e52e33f08c..6cdc9fd03195 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -147,6 +147,17 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)
return sample_period;
}
+static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
+{
+ struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+
+ if (pmc_is_fixed(pmc))
+ return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
+ pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
+
+ return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
+}
+
void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index 2917105e584e..346b1104e674 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -211,6 +211,23 @@ static struct kvm_pmc *intel_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)
return pmc;
}
+static void intel_pmu_pebs_setup(struct kvm_pmu *pmu)
+{
+ struct kvm_pmc *pmc = NULL;
+ int bit;
+
+ pmu->need_rewrite_ds_pebs_interrupt_threshold = false;
+
+ for_each_set_bit(bit, (unsigned long *)&pmu->pebs_enable, X86_PMC_IDX_MAX) {
+ pmc = kvm_x86_ops.pmu_ops->pmc_idx_to_pmc(pmu, bit);
+
+ if (pmc && pmc_speculative_in_use(pmc)) {
+ pmu->need_rewrite_ds_pebs_interrupt_threshold = true;
+ break;
+ }
+ }
+}
+
static int intel_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
{
struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
@@ -287,6 +304,8 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
return 0;
if (kvm_valid_perf_global_ctrl(pmu, data)) {
global_ctrl_changed(pmu, data);
+ if (pmu->global_ctrl & pmu->pebs_enable)
+ intel_pmu_pebs_setup(pmu);
return 0;
}
break;
@@ -491,12 +510,64 @@ static void intel_pmu_reset(struct kvm_vcpu *vcpu)
pmu->global_ovf_ctrl = 0;
}
+static int rewrite_ds_pebs_interrupt_threshold(struct kvm_vcpu *vcpu)
+{
+ struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
+ struct debug_store *ds = NULL;
+ u64 new_threshold, offset;
+ gpa_t gpa;
+ int srcu_idx, ret = -ENOMEM;
+
+ ds = kmalloc(sizeof(struct debug_store), GFP_KERNEL);
+ if (!ds)
+ goto out;
+
+ ret = -EFAULT;
+ srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
+ gpa = kvm_mmu_gva_to_gpa_system(vcpu, pmu->ds_area, NULL);
+ if (gpa == UNMAPPED_GVA)
+ goto unlock_out;
+
+ if (kvm_read_guest(vcpu->kvm, gpa, ds, sizeof(struct debug_store)))
+ goto unlock_out;
+
+ /* Adding sizeof(struct pebs_basic) offset is enough to generate PMI. */
+ new_threshold = ds->pebs_buffer_base + sizeof(struct pebs_basic);
+ offset = offsetof(struct debug_store, pebs_interrupt_threshold);
+ gpa = kvm_mmu_gva_to_gpa_system(vcpu, pmu->ds_area + offset, NULL);
+ if (gpa == UNMAPPED_GVA)
+ goto unlock_out;
+
+ if (kvm_write_guest(vcpu->kvm, gpa, &new_threshold, sizeof(u64)))
+ goto unlock_out;
+
+ ret = 0;
+
+unlock_out:
+ srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
+
+out:
+ kfree(ds);
+ return ret;
+}
+
void intel_pmu_handle_event(struct kvm_vcpu *vcpu)
{
struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
+ int ret;
if (!(pmu->global_ctrl & pmu->pebs_enable))
return;
+
+ if (pmu->counter_cross_mapped && pmu->need_rewrite_ds_pebs_interrupt_threshold) {
+ ret = rewrite_ds_pebs_interrupt_threshold(vcpu);
+ pmu->need_rewrite_ds_pebs_interrupt_threshold = false;
+ }
+
+ if (ret == -ENOMEM)
+ pr_debug_ratelimited("%s: Fail to emulate guest PEBS due to OOM.", __func__);
+ else if (ret == -EFAULT)
+ pr_debug_ratelimited("%s: Fail to emulate guest PEBS due to GPA fault.", __func__);
}
struct kvm_pmu_ops intel_pmu_ops = {
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 88a544e6379f..8db0811c1dd3 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -5856,6 +5856,7 @@ gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
{
return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
}
+EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_system);
static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
struct kvm_vcpu *vcpu, u32 access,
--
2.21.3
Powered by blists - more mailing lists