From: Zhang, Yanmin <[email protected]>
Based on Ingo's idea, I implement a para virt interface for perf to support
statistics collection in guest os. That means we could run tool perf in guest
os directly.
Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
design suggestions. I also want to thank Yangsheng and LinMing for their generous
help.
The design is:
1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
2) Create a host perf_event per guest perf_event;
3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
kernel if a guest event overflows.
4) Guest kernel goes through all enabled event on current cpu and output data when they
overflows.
5) No change in user space.
Below is an example.
#perf top
--------------------------------------------------------------------------------------------------------------------------
PerfTop: 7954 irqs/sec kernel:79.5% exact: 0.0% [1000Hz cycles], (all, 8 CPUs)
--------------------------------------------------------------------------------------------------------------------------
samples pcnt function DSO
_______ _____ ________________________ _________________________________________________________
5315.00 4.9% copy_user_generic_string /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
3342.00 3.1% add_preempt_count /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
3338.00 3.1% sub_preempt_count /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
2454.00 2.3% pvclock_clocksource_read /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
2434.00 2.3% tcp_sendmsg /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
2090.00 1.9% child_run /bm/tmp/benchmarks/run_bmtbench/dbench/dbench-3.03/tbench
2081.00 1.9% debug_smp_processor_id /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
2003.00 1.9% __GI_strstr /lib64/libc-2.11.so
1999.00 1.9% __strchr_sse2 /lib64/libc-2.11.so
1983.00 1.8% tcp_ack /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
1800.00 1.7% tcp_transmit_skb /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
1727.00 1.6% schedule /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
1706.00 1.6% __libc_recv /lib64/libc-2.11.so
1702.00 1.6% __GI_memchr /lib64/libc-2.11.so
1580.00 1.5% tcp_recvmsg /lib/modules/2.6.35-rc1-tip-guestperf/build/vmlinux
The patch is against tip/master tree of June 1st.
Signed-off-by: Zhang Yanmin <[email protected]>
---
diff -Nraup linux-2.6_tip0601/arch/x86/include/asm/kvm_host.h linux-2.6_tip0601perfkvm/arch/x86/include/asm/kvm_host.h
--- linux-2.6_tip0601/arch/x86/include/asm/kvm_host.h 2010-06-02 10:01:52.147999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/include/asm/kvm_host.h 2010-06-06 15:46:31.874999850 +0800
@@ -561,6 +561,9 @@ int emulator_write_phys(struct kvm_vcpu
const void *val, int bytes);
int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
gpa_t addr, unsigned long *ret);
+int kvm_pv_perf_op(struct kvm_vcpu *vcpu, int op_code, unsigned long a1,
+ unsigned long a2, unsigned long *result);
+
u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
extern bool tdp_enabled;
diff -Nraup linux-2.6_tip0601/arch/x86/include/asm/kvm_para.h linux-2.6_tip0601perfkvm/arch/x86/include/asm/kvm_para.h
--- linux-2.6_tip0601/arch/x86/include/asm/kvm_para.h 2010-06-02 10:01:52.126999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/include/asm/kvm_para.h 2010-06-06 15:46:31.874999850 +0800
@@ -33,7 +33,14 @@
#define MSR_KVM_WALL_CLOCK_NEW 0x4b564d00
#define MSR_KVM_SYSTEM_TIME_NEW 0x4b564d01
-#define KVM_MAX_MMU_OP_BATCH 32
+/* Operations for KVM_PERF_OP */
+#define KVM_PERF_OP_OPEN 1
+#define KVM_PERF_OP_CLOSE 2
+#define KVM_PERF_OP_ENABLE 3
+#define KVM_PERF_OP_DISABLE 4
+#define KVM_PERF_OP_START 5
+#define KVM_PERF_OP_STOP 6
+#define KVM_PERF_OP_READ 7
/* Operations for KVM_HC_MMU_OP */
#define KVM_MMU_OP_WRITE_PTE 1
@@ -64,6 +71,12 @@ struct kvm_mmu_op_release_pt {
#ifdef __KERNEL__
#include <asm/processor.h>
+struct kvmperf_event {
+ unsigned int event_offset;
+ struct page *event_page;
+ struct page *event_page2;
+};
+
extern void kvmclock_init(void);
diff -Nraup linux-2.6_tip0601/arch/x86/Kconfig linux-2.6_tip0601perfkvm/arch/x86/Kconfig
--- linux-2.6_tip0601/arch/x86/Kconfig 2010-06-02 10:01:52.364999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/Kconfig 2010-06-06 15:46:31.875999850 +0800
@@ -552,6 +552,14 @@ config KVM_GUEST
This option enables various optimizations for running under the KVM
hypervisor.
+config KVM_PERF
+ bool "KVM Guest perf support"
+ select PARAVIRT
+ select PERF_EVENT
+ ---help---
+ This option enables various optimizations for running perf in
+ guest os under the KVM hypervisor.
+
source "arch/x86/lguest/Kconfig"
config PARAVIRT
diff -Nraup linux-2.6_tip0601/arch/x86/kernel/cpu/perf_event.c linux-2.6_tip0601perfkvm/arch/x86/kernel/cpu/perf_event.c
--- linux-2.6_tip0601/arch/x86/kernel/cpu/perf_event.c 2010-06-02 10:01:53.252999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kernel/cpu/perf_event.c 2010-06-06 15:46:31.875999850 +0800
@@ -25,6 +25,7 @@
#include <linux/highmem.h>
#include <linux/cpu.h>
#include <linux/bitops.h>
+#include <linux/kvm_para.h>
#include <asm/apic.h>
#include <asm/stacktrace.h>
@@ -582,10 +583,20 @@ static void x86_pmu_disable_all(void)
}
}
+#ifdef CONFIG_KVM_PERF
+extern int kvm_hw_perf_enable(void);
+extern int kvm_hw_perf_disable(void);
+#endif
+
void hw_perf_disable(void)
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+#ifdef CONFIG_KVM_PERF
+ if (!kvm_hw_perf_disable())
+ return;
+#endif
+
if (!x86_pmu_initialized())
return;
@@ -809,6 +820,11 @@ void hw_perf_enable(void)
struct hw_perf_event *hwc;
int i, added = cpuc->n_added;
+#ifdef CONFIG_KVM_PERF
+ if (!kvm_hw_perf_enable())
+ return;
+#endif
+
if (!x86_pmu_initialized())
return;
@@ -1254,6 +1270,7 @@ x86_get_event_constraints(struct cpu_hw_
#include "perf_event_intel_lbr.c"
#include "perf_event_intel_ds.c"
#include "perf_event_intel.c"
+#include "perf_event_kvm.c"
static int __cpuinit
x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
@@ -1307,6 +1324,11 @@ void __init init_hw_perf_events(void)
pr_info("Performance Events: ");
+#ifdef CONFIG_KVM_PERF
+ if (!kvm_init_hw_perf_events())
+ return;
+#endif
+
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_INTEL:
err = intel_pmu_init();
@@ -1535,6 +1557,13 @@ const struct pmu *hw_perf_event_init(str
const struct pmu *tmp;
int err;
+#ifdef CONFIG_KVM_PERF
+ if (kvm_para_available()) {
+ tmp = kvm_hw_perf_event_init(event);
+ return tmp;
+ }
+#endif
+
err = __hw_perf_event_init(event);
if (!err) {
/*
diff -Nraup linux-2.6_tip0601/arch/x86/kernel/cpu/perf_event_kvm.c linux-2.6_tip0601perfkvm/arch/x86/kernel/cpu/perf_event_kvm.c
--- linux-2.6_tip0601/arch/x86/kernel/cpu/perf_event_kvm.c 1970-01-01 08:00:00.000000000 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kernel/cpu/perf_event_kvm.c 2010-06-06 15:46:31.876999850 +0800
@@ -0,0 +1,367 @@
+/*
+ * Performance events
+ *
+ * Copyright (C) 2010 Intel Corporation, Zhang Yanmin <[email protected]>
+ *
+ * For licencing details see kernel-base/COPYING
+ */
+
+#ifdef CONFIG_KVM_PERF
+
+#ifdef CONFIG_X86_LOCAL_APIC
+
+static bool kvm_reserve_pmc_hardware(void)
+{
+ if (nmi_watchdog == NMI_LOCAL_APIC)
+ disable_lapic_nmi_watchdog();
+
+ return true;
+}
+
+static void kvm_release_pmc_hardware(void)
+{
+ if (nmi_watchdog == NMI_LOCAL_APIC)
+ enable_lapic_nmi_watchdog();
+}
+
+#else
+
+static bool kvm_reserve_pmc_hardware(void) { return true; }
+static void kvm_release_pmc_hardware(void) {}
+
+#endif
+
+static void kvm_hw_perf_event_destroy(struct perf_event *event)
+{
+ unsigned long addr;
+
+ addr = __pa(event);
+ kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_CLOSE,
+ addr, (unsigned long) event->shadow);
+
+ if (atomic_dec_and_mutex_lock(&active_events, &pmc_reserve_mutex)) {
+ kvm_release_pmc_hardware();
+ mutex_unlock(&pmc_reserve_mutex);
+ }
+}
+
+static int
+check_event_overflow(struct perf_event *event, struct pt_regs *regs)
+{
+ struct perf_sample_data data;
+ struct hw_perf_event *hwc = &event->hw;
+ s32 overflows;
+ int i;
+ int handled = 0;
+
+again:
+ overflows = atomic_read(&hwc->overflows);
+ if (atomic_cmpxchg(&hwc->overflows, overflows, 0) !=
+ overflows)
+ goto again;
+
+ for (i = 0; i < overflows; i++) {
+ perf_sample_data_init(&data, 0);
+
+ data.period = event->hw.last_period;
+
+ if (event->overflow_handler)
+ event->overflow_handler(event, 1, &data, regs);
+ else
+
+ perf_event_output(event, 1, &data, regs);
+
+ handled++;
+ }
+
+ return handled;
+}
+
+static int
+kvm_check_event_overflow(struct pt_regs *regs)
+{
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+ struct perf_event *event;
+ int i, max_count;
+ int handled = 0;
+
+ max_count = X86_PMC_IDX_MAX;
+ for (i = 0; i < max_count; i++) {
+ event = cpuc->event_list[i];
+ if (event)
+ handled += check_event_overflow(event, regs);
+ }
+ return handled;
+}
+
+static DEFINE_PER_CPU(int, kvm_nmi_entered);
+
+static int kvm_x86_pmu_handle_irq(struct pt_regs *regs)
+{
+ int handled = 0;
+
+ if (percpu_read(kvm_nmi_entered))
+ return 0;
+
+ percpu_write(kvm_nmi_entered, 1);
+
+ handled = kvm_check_event_overflow(regs);
+ if (handled)
+ inc_irq_stat(apic_perf_irqs);
+
+ percpu_write(kvm_nmi_entered, 0);
+
+ return handled;
+}
+
+static int __kprobes
+kvm_perf_event_nmi_handler(struct notifier_block *self,
+ unsigned long cmd, void *__args)
+{
+ struct die_args *args = __args;
+ struct pt_regs *regs;
+
+ if (!atomic_read(&active_events))
+ return NOTIFY_DONE;
+
+ switch (cmd) {
+ case DIE_NMI:
+ case DIE_NMI_IPI:
+ break;
+
+ default:
+ return NOTIFY_DONE;
+ }
+
+ regs = args->regs;
+
+ kvm_x86_pmu_handle_irq(regs);
+
+ return NOTIFY_STOP;
+}
+
+static __read_mostly struct notifier_block kvm_perf_event_nmi_notifier = {
+ .notifier_call = kvm_perf_event_nmi_handler,
+ .next = NULL,
+ .priority = 1
+};
+
+static int kvm_add_event(struct perf_event *event)
+{
+ int i, max_count;
+ unsigned long flags;
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+ int ret = -1;
+
+ local_irq_save(flags);
+ max_count = X86_PMC_IDX_MAX;
+
+ if (cpuc->n_events >= max_count) {
+ local_irq_restore(flags);
+ return -ENOSPC;
+ }
+ for (i = 0; i < max_count; i++) {
+ if (cpuc->event_list[i] == NULL) {
+ cpuc->event_list[i] = event;
+ cpuc->n_events++;
+ ret = 0;
+ break;
+ }
+ }
+ local_irq_restore(flags);
+ return ret;
+}
+
+static int kvm_del_event(struct perf_event *event)
+{
+ int i, max_count;
+ unsigned long flags;
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+ int ret = -1;
+
+ local_irq_save(flags);
+ max_count = X86_PMC_IDX_MAX;
+ for (i = 0; i < max_count; i++) {
+ if (cpuc->event_list[i] == event) {
+ cpuc->event_list[i] = NULL;
+ cpuc->n_events--;
+ ret = 0;
+ break;
+ }
+ }
+ local_irq_restore(flags);
+ return ret;
+}
+
+static int kvm_pmu_enable(struct perf_event *event)
+{
+ int ret;
+ unsigned long addr = __pa(event);
+
+ if (kvm_add_event(event))
+ return -1;
+
+ ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_ENABLE,
+ addr, (unsigned long) event->shadow);
+ return ret;
+}
+
+static void kvm_pmu_disable(struct perf_event *event)
+{
+ unsigned long addr = __pa(event);
+ kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_DISABLE,
+ addr, (unsigned long) event->shadow);
+ kvm_del_event(event);
+}
+
+static int kvm_pmu_start(struct perf_event *event)
+{
+ int ret;
+ unsigned long addr = __pa(event);
+
+ if (kvm_add_event(event))
+ return -1;
+
+ ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_START,
+ addr, (unsigned long) event->shadow);
+ return ret;
+}
+
+static void kvm_pmu_stop(struct perf_event *event)
+{
+ unsigned long addr = __pa(event);
+ kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_STOP,
+ addr, (unsigned long) event->shadow);
+ kvm_del_event(event);
+}
+
+static void kvm_pmu_read(struct perf_event *event)
+{
+ unsigned long addr;
+
+ addr = __pa(event);
+
+ kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_READ,
+ addr, (unsigned long) event->shadow);
+ return;
+}
+
+static void kvm_pmu_unthrottle(struct perf_event *event)
+{
+ return;
+}
+
+static const struct pmu kvm_pmu = {
+ .enable = kvm_pmu_enable,
+ .disable = kvm_pmu_disable,
+ .start = kvm_pmu_start,
+ .stop = kvm_pmu_stop,
+ .read = kvm_pmu_read,
+ .unthrottle = kvm_pmu_unthrottle,
+};
+
+static int kvm_default_x86_handle_irq(struct pt_regs *regs)
+{
+ return 1;
+}
+
+int __init kvm_init_hw_perf_events(void)
+{
+ if (!kvm_para_available())
+ return -1;
+
+ x86_pmu.handle_irq = kvm_default_x86_handle_irq;
+
+ pr_cont("KVM PARA PMU driver.\n");
+ register_die_notifier(&kvm_perf_event_nmi_notifier);
+
+ return 0;
+}
+
+static int __kvm_hw_perf_event_init(struct perf_event *event)
+{
+ struct hw_perf_event *hwc = &event->hw;
+ int err;
+ unsigned long result;
+ unsigned long addr;
+
+ err = 0;
+ if (!atomic_inc_not_zero(&active_events)) {
+ mutex_lock(&pmc_reserve_mutex);
+ if (atomic_read(&active_events) == 0) {
+ if (!kvm_reserve_pmc_hardware())
+ err = -EBUSY;
+ }
+ if (!err)
+ atomic_inc(&active_events);
+ mutex_unlock(&pmc_reserve_mutex);
+ if (err)
+ return err;
+ }
+
+ event->destroy = kvm_hw_perf_event_destroy;
+
+ hwc->idx = -1;
+ hwc->last_cpu = -1;
+ hwc->last_tag = ~0ULL;
+
+ addr = __pa(event);
+ result = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, addr, 0);
+
+ if (result)
+ event->shadow = (void *) result;
+ else
+ err = -1;
+
+ return err;
+}
+
+const struct pmu *kvm_hw_perf_event_init(struct perf_event *event)
+{
+ int err;
+
+ err = __kvm_hw_perf_event_init(event);
+ if (err)
+ return ERR_PTR(err);
+
+ return &kvm_pmu;
+}
+
+int kvm_hw_perf_enable(void)
+{
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+
+ if (!kvm_para_available())
+ return -1;
+
+ if (cpuc->enabled)
+ return 0;
+
+ if (cpuc->n_added)
+ cpuc->n_added = 0;
+
+ cpuc->enabled = 1;
+ barrier();
+
+ return 0;
+}
+
+int kvm_hw_perf_disable(void)
+{
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+
+ if (!kvm_para_available())
+ return -1;
+
+ if (!cpuc->enabled)
+ return 0;
+
+ cpuc->n_added = 0;
+ cpuc->enabled = 0;
+ barrier();
+
+ return 0;
+}
+
+#endif
+
diff -Nraup linux-2.6_tip0601/arch/x86/kvm/kvmperf_event.c linux-2.6_tip0601perfkvm/arch/x86/kvm/kvmperf_event.c
--- linux-2.6_tip0601/arch/x86/kvm/kvmperf_event.c 1970-01-01 08:00:00.000000000 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kvm/kvmperf_event.c 2010-06-06 16:36:32.714999849 +0800
@@ -0,0 +1,276 @@
+/*
+ * Performance events x86 kvm para architecture code
+ *
+ * Copyright (C) 2010 Intel Inc.
+ * Zhang Yanmin <[email protected]>
+ *
+ * For licencing details see kernel-base/COPYING
+ */
+
+#include <linux/perf_event.h>
+#include <linux/capability.h>
+#include <linux/notifier.h>
+#include <linux/hardirq.h>
+#include <linux/kprobes.h>
+#include <linux/module.h>
+#include <linux/kdebug.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/cpu.h>
+#include <linux/kvm.h>
+#include <linux/kvm_host.h>
+#include <linux/file.h>
+#include <linux/syscalls.h>
+#include <linux/init.h>
+
+#include <asm/apic.h>
+#include <asm/stacktrace.h>
+#include <asm/nmi.h>
+#include <asm/compat.h>
+
+#include "x86.h"
+
+static void kvm_sync_event_to_guest(struct perf_event *event, int overflows)
+{
+ struct hw_perf_event *hwc = &event->hw;
+ struct kvmperf_event *kvmevent;
+ int offset, len, data_len, copied, page_offset;
+ struct page *event_page;
+ void *shared_kaddr;
+
+ kvmevent = event->shadow;
+ offset = kvmevent->event_offset;
+
+ /* Copy perf_event->count firstly */
+ offset += offsetof(struct perf_event, count);
+ if (offset < PAGE_SIZE) {
+ event_page = kvmevent->event_page;
+ page_offset = offset;
+ } else {
+ event_page = kvmevent->event_page2;
+ page_offset = offset - PAGE_SIZE;
+ }
+ shared_kaddr = kmap_atomic(event_page, KM_USER0);
+ *((atomic64_t *)(shared_kaddr + page_offset)) = event->count;
+
+ offset = kvmevent->event_offset;
+ offset += offsetof(struct perf_event, hw);
+ if (offset < PAGE_SIZE) {
+ if (event_page == kvmevent->event_page2) {
+ kunmap_atomic(shared_kaddr, KM_USER0);
+ event_page = kvmevent->event_page;
+ shared_kaddr = kmap_atomic(event_page, KM_USER0);
+ }
+ page_offset = offset;
+ } else {
+ if (event_page == kvmevent->event_page) {
+ kunmap_atomic(shared_kaddr, KM_USER0);
+ event_page = kvmevent->event_page2;
+ shared_kaddr = kmap_atomic(event_page, KM_USER0);
+ }
+ page_offset = offset - PAGE_SIZE;
+ }
+
+ if (overflows)
+ atomic_add(overflows, (atomic_t *)(shared_kaddr + page_offset));
+
+ kunmap_atomic(shared_kaddr, KM_USER0);
+#if 0
+ offset += offsetof(struct hw_perf_event, prev_count);
+ data_len = sizeof(struct hw_perf_event) -
+ offsetof(struct hw_perf_event, prev_count);
+ if (event_page == kvmevent->event_page2) {
+ page_offset += offsetof(struct hw_perf_event, prev_count);
+ memcpy(shared_kaddr + page_offset,
+ &hwc->prev_count, data_len);
+ kunmap_atomic(shared_kaddr, KM_USER0);
+
+ return;
+ }
+
+ copied = 0;
+ if (offset < PAGE_SIZE) {
+ len = PAGE_SIZE - offset;
+ if (len > data_len)
+ len = data_len;
+ memcpy(shared_kaddr + offset,
+ &hwc->prev_count, data_len);
+ copied = len;
+ page_offset = 0;
+ } else
+ page_offset = offset - PAGE_SIZE;
+
+ kunmap_atomic(shared_kaddr, KM_USER0);
+ len = data_len - copied;
+ if (len) {
+ /* Copy across pages */
+ shared_kaddr = kmap_atomic(kvmevent->event_page2, KM_USER0);
+ memcpy(shared_kaddr + page_offset,
+ ((void *)&hwc->prev_count) + copied, len);
+ kunmap_atomic(shared_kaddr, KM_USER0);
+ }
+#endif
+}
+
+static void kvm_perf_event_overflow(struct perf_event *event, int nmi,
+ struct perf_sample_data *data,
+ struct pt_regs *regs)
+{
+ BUG_ON(event->shadow == NULL);
+ kvm_sync_event_to_guest(event, 1);
+
+ kvm_notify_event_overflow();
+}
+
+static struct perf_event *
+kvm_pv_perf_op_open(struct kvm_vcpu *vcpu, gpa_t addr)
+{
+ int ret;
+ struct perf_event *event;
+ struct perf_event *host_event = NULL;
+ struct kvmperf_event *shadow = NULL;
+
+ event = kzalloc(sizeof(*event), GFP_KERNEL);
+ if (!event)
+ goto out;
+
+ shadow = kzalloc(sizeof(*shadow), GFP_KERNEL);
+ if (!shadow)
+ goto out;
+
+ shadow->event_page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
+ shadow->event_offset = addr & ~PAGE_MASK;
+ if (shadow->event_offset + sizeof(struct perf_event) > PAGE_SIZE) {
+ shadow->event_page2 = gfn_to_page(vcpu->kvm,
+ (addr >> PAGE_SHIFT) + 1);
+ }
+
+ ret = kvm_read_guest(vcpu->kvm, addr, event, sizeof(*event));
+ if (ret)
+ goto out;
+
+ /*
+ * By default, we disable the host event. Later on, guets os
+ * triggers a perf_event_attach to enable it
+ */
+ event->attr.disabled = 1;
+ event->attr.inherit = 0;
+ event->attr.enable_on_exec = 0;
+ /*
+ * We don't support exclude mode of user and kernel for guest os,
+ * which mean we always collect both user and kernel for guest os
+ */
+ event->attr.exclude_user = 0;
+ event->attr.exclude_kernel = 0;
+ /* We always create a cpu context host perf event */
+
+ host_event = perf_event_create_kernel_counter(&event->attr, -1,
+ current->pid, kvm_perf_event_overflow);
+
+ if (IS_ERR(host_event)) {
+ host_event = NULL;
+ goto out;
+ }
+ host_event->shadow = shadow;
+
+out:
+ if (!host_event)
+ kfree(shadow);
+ kfree(event);
+
+ return host_event;
+}
+
+static int kvm_pv_perf_op_close(struct kvm_vcpu *vcpu,
+ struct perf_event *host_event)
+{
+ struct kvmperf_event *shadow = host_event->shadow;
+
+ perf_event_release_kernel(host_event);
+ put_page(shadow->event_page);
+ if (shadow->event_page2)
+ put_page(shadow->event_page2);
+ kfree(shadow);
+ return 0;
+}
+
+static int kvm_pv_perf_op_enable(struct perf_event *host_event)
+{
+ perf_event_attach(host_event);
+ return 0;
+}
+
+static int kvm_pv_perf_op_disable(struct perf_event *host_event)
+{
+ perf_event_detach(host_event);
+ return 0;
+}
+
+static int kvm_pv_perf_op_start(struct perf_event *host_event)
+{
+ perf_event_attach(host_event);
+ return 0;
+}
+
+static int kvm_pv_perf_op_stop(struct perf_event *host_event)
+{
+ perf_event_detach(host_event);
+ return 0;
+}
+
+static int kvm_pv_perf_op_read(struct perf_event *host_event)
+{
+ u64 enabled, running;
+ if (host_event->state == PERF_EVENT_STATE_ACTIVE)
+ perf_event_read_value(host_event, &enabled, &running);
+ kvm_sync_event_to_guest(host_event, 0);
+ return 0;
+}
+
+int kvm_pv_perf_op(struct kvm_vcpu *vcpu, int op_code, unsigned long a1,
+ unsigned long a2, unsigned long *result)
+{
+ unsigned long ret;
+ struct perf_event *host_event;
+ gpa_t addr;
+
+ addr = (gpa_t)(a1);
+
+ switch(op_code) {
+ case KVM_PERF_OP_OPEN:
+ ret = (unsigned long) kvm_pv_perf_op_open(vcpu, addr);
+ break;
+ case KVM_PERF_OP_CLOSE:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_close(vcpu, host_event);
+ break;
+ case KVM_PERF_OP_ENABLE:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_enable(host_event);
+ break;
+ case KVM_PERF_OP_DISABLE:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_disable(host_event);
+ break;
+ case KVM_PERF_OP_START:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_start(host_event);
+ break;
+ case KVM_PERF_OP_STOP:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_stop(host_event);
+ break;
+ case KVM_PERF_OP_READ:
+ host_event = (struct perf_event *) a2;
+ ret = kvm_pv_perf_op_read(host_event);
+ break;
+ default:
+ ret = -KVM_ENOSYS;
+ }
+
+ *result = ret;
+ return 0;
+}
+
diff -Nraup linux-2.6_tip0601/arch/x86/kvm/Makefile linux-2.6_tip0601perfkvm/arch/x86/kvm/Makefile
--- linux-2.6_tip0601/arch/x86/kvm/Makefile 2010-06-02 10:01:52.563999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kvm/Makefile 2010-06-06 15:46:31.876999850 +0800
@@ -11,7 +11,7 @@ kvm-y += $(addprefix ../../../virt/kvm
kvm-$(CONFIG_IOMMU_API) += $(addprefix ../../../virt/kvm/, iommu.o)
kvm-y += x86.o mmu.o emulate.o i8259.o irq.o lapic.o \
- i8254.o timer.o
+ i8254.o timer.o kvmperf_event.o
kvm-intel-y += vmx.o
kvm-amd-y += svm.o
diff -Nraup linux-2.6_tip0601/arch/x86/kvm/x86.c linux-2.6_tip0601perfkvm/arch/x86/kvm/x86.c
--- linux-2.6_tip0601/arch/x86/kvm/x86.c 2010-06-02 10:01:52.572999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kvm/x86.c 2010-06-06 16:33:28.977999849 +0800
@@ -6,12 +6,14 @@
* Copyright (C) 2006 Qumranet, Inc.
* Copyright (C) 2008 Qumranet, Inc.
* Copyright IBM Corporation, 2008
+ * Copyright Intel Corporation, 2010
*
* Authors:
* Avi Kivity <[email protected]>
* Yaniv Kamay <[email protected]>
* Amit Shah <[email protected]>
* Ben-Ami Yassour <[email protected]>
+ * Yanmin Zhang <[email protected]>
*
* This work is licensed under the terms of the GNU GPL, version 2. See
* the COPYING file in the top-level directory.
@@ -4052,6 +4054,16 @@ static unsigned long kvm_get_guest_ip(vo
return ip;
}
+int kvm_notify_event_overflow(void)
+{
+ if (percpu_read(current_vcpu)) {
+ kvm_inject_nmi(percpu_read(current_vcpu));
+ return 0;
+ }
+
+ return -1;
+}
+
static struct perf_guest_info_callbacks kvm_guest_cbs = {
.is_in_guest = kvm_is_in_guest,
.is_user_mode = kvm_is_user_mode,
@@ -4245,6 +4257,9 @@ int kvm_emulate_hypercall(struct kvm_vcp
case KVM_HC_MMU_OP:
r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
break;
+ case KVM_PERF_OP:
+ r = kvm_pv_perf_op(vcpu, a0, a1, a2, &ret);
+ break;
default:
ret = -KVM_ENOSYS;
break;
diff -Nraup linux-2.6_tip0601/arch/x86/kvm/x86.h linux-2.6_tip0601perfkvm/arch/x86/kvm/x86.h
--- linux-2.6_tip0601/arch/x86/kvm/x86.h 2010-06-02 10:01:52.552999849 +0800
+++ linux-2.6_tip0601perfkvm/arch/x86/kvm/x86.h 2010-06-06 16:30:45.918999850 +0800
@@ -74,5 +74,6 @@ static inline struct kvm_mem_aliases *kv
void kvm_before_handle_nmi(struct kvm_vcpu *vcpu);
void kvm_after_handle_nmi(struct kvm_vcpu *vcpu);
+int kvm_notify_event_overflow(void);
#endif
diff -Nraup linux-2.6_tip0601/include/linux/kvm_para.h linux-2.6_tip0601perfkvm/include/linux/kvm_para.h
--- linux-2.6_tip0601/include/linux/kvm_para.h 2010-06-02 10:02:08.780999849 +0800
+++ linux-2.6_tip0601perfkvm/include/linux/kvm_para.h 2010-06-06 15:46:31.877999850 +0800
@@ -17,6 +17,7 @@
#define KVM_HC_VAPIC_POLL_IRQ 1
#define KVM_HC_MMU_OP 2
+#define KVM_PERF_OP 3
/*
* hypercalls use architecture specific
diff -Nraup linux-2.6_tip0601/include/linux/perf_event.h linux-2.6_tip0601perfkvm/include/linux/perf_event.h
--- linux-2.6_tip0601/include/linux/perf_event.h 2010-06-02 10:02:08.055999849 +0800
+++ linux-2.6_tip0601perfkvm/include/linux/perf_event.h 2010-06-06 16:31:38.542999849 +0800
@@ -534,7 +534,20 @@ struct hw_perf_event {
/* breakpoint */
struct arch_hw_breakpoint info;
#endif
+#ifdef CONFIG_KVM_PERF
+ /*
+ * host will increase overflows of guest event variable,
+ * guest kernel checks it and output overflow data
+ */
+ atomic_t overflows;
+#endif
};
+
+ /*
+ * CAREFULL: prev_count should be the first member after the
+ * union. With KVM paravirt support, host side perf_event sync
+ * function just assumes that.
+ */
atomic64_t prev_count;
u64 sample_period;
u64 last_period;
@@ -731,6 +744,14 @@ struct perf_event {
perf_overflow_handler_t overflow_handler;
+ /*
+ * pointer to kvm guest/host perf_event peers:
+ * 1) If in host, shadow points to an area of guest event
+ * page mapping information;
+ * 2) If in guest, shadow points to it's host peer event;
+ */
+ void *shadow;
+
#ifdef CONFIG_EVENT_TRACING
struct ftrace_event_call *tp_event;
struct event_filter *filter;
@@ -849,6 +870,10 @@ perf_event_create_kernel_counter(struct
perf_overflow_handler_t callback);
extern u64 perf_event_read_value(struct perf_event *event,
u64 *enabled, u64 *running);
+extern void perf_event_output(struct perf_event *event, int nmi,
+ struct perf_sample_data *data, struct pt_regs *regs);
+void perf_event_attach(struct perf_event *event);
+void perf_event_detach(struct perf_event *event);
struct perf_sample_data {
u64 type;
@@ -1026,6 +1051,14 @@ perf_event_task_sched_in(struct task_str
static inline void
perf_event_task_sched_out(struct task_struct *task,
struct task_struct *next) { }
+
+static inline void
+perf_event_output(struct perf_event *event, int nmi,
+ struct perf_sample_data *data, struct pt_regs *regs) { }
+
+static inline void perf_event_attach(struct perf_event *event) { }
+static inline void perf_event_detach(struct perf_event *event) { }
+
static inline void
perf_event_task_tick(struct task_struct *task) { }
static inline int perf_event_init_task(struct task_struct *child) { return 0; }
diff -Nraup linux-2.6_tip0601/kernel/perf_event.c linux-2.6_tip0601perfkvm/kernel/perf_event.c
--- linux-2.6_tip0601/kernel/perf_event.c 2010-06-02 10:03:06.809999849 +0800
+++ linux-2.6_tip0601perfkvm/kernel/perf_event.c 2010-06-06 15:57:08.878999851 +0800
@@ -34,6 +34,7 @@
#include <linux/hw_breakpoint.h>
#include <asm/irq_regs.h>
+#include <asm/kvm_para.h>
/*
* Each CPU has a list of per CPU events:
@@ -754,6 +755,7 @@ static int group_can_go_on(struct perf_e
*/
if (event->attr.exclusive && cpuctx->active_oncpu)
return 0;
+
/*
* Otherwise, try to add it if all previous groups were able
* to go on.
@@ -1617,6 +1619,7 @@ void perf_event_task_tick(struct task_st
struct perf_cpu_context *cpuctx;
struct perf_event_context *ctx;
int rotate = 0;
+ int adjust_freq = 1;
if (!atomic_read(&nr_events))
return;
@@ -1630,9 +1633,16 @@ void perf_event_task_tick(struct task_st
if (ctx && ctx->nr_events && ctx->nr_events != ctx->nr_active)
rotate = 1;
- perf_ctx_adjust_freq(&cpuctx->ctx);
- if (ctx)
- perf_ctx_adjust_freq(ctx);
+#ifdef CONFIG_KVM_PERF
+ if (kvm_para_available())
+ adjust_freq = 0;
+#endif
+
+ if (adjust_freq) {
+ perf_ctx_adjust_freq(&cpuctx->ctx);
+ if (ctx)
+ perf_ctx_adjust_freq(ctx);
+ }
if (!rotate)
return;
@@ -3431,7 +3441,7 @@ void perf_prepare_sample(struct perf_eve
}
}
-static void perf_event_output(struct perf_event *event, int nmi,
+void perf_event_output(struct perf_event *event, int nmi,
struct perf_sample_data *data,
struct pt_regs *regs)
{
@@ -5242,6 +5252,47 @@ perf_event_create_kernel_counter(struct
}
EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter);
+void perf_event_attach(struct perf_event *event)
+{
+ struct perf_event_context *old_ctx, *new_ctx;
+
+ old_ctx = event->ctx;
+ new_ctx = find_get_context(current->pid, -1);
+ if (old_ctx != new_ctx) {
+ if (old_ctx) {
+ /* Delete from old ctx before joining new ctx */
+ mutex_lock(&old_ctx->mutex);
+ raw_spin_lock(&old_ctx->lock);
+ list_del_event(event, old_ctx);
+ raw_spin_unlock(&old_ctx->lock);
+ mutex_unlock(&old_ctx->mutex);
+ put_ctx(old_ctx);
+ }
+
+ mutex_lock(&new_ctx->mutex);
+ raw_spin_lock(&new_ctx->lock);
+ list_add_event(event, new_ctx);
+ event->ctx = new_ctx;
+ raw_spin_unlock(&new_ctx->lock);
+ mutex_unlock(&new_ctx->mutex);
+ } else
+ put_ctx(new_ctx);
+
+ perf_event_enable(event);
+}
+EXPORT_SYMBOL_GPL(perf_event_attach);
+
+void perf_event_detach(struct perf_event *event)
+{
+ /*
+ * Just disable the event and don't del it from
+ * ctx->event_list in case there is a race condition
+ * with perf_event_read_value
+ */
+ perf_event_disable(event);
+}
+EXPORT_SYMBOL_GPL(perf_event_detach);
+
/*
* inherit a event from parent task to child task:
*/
On 06/09/2010 06:30 AM, Zhang, Yanmin wrote:
> From: Zhang, Yanmin<[email protected]>
>
> Based on Ingo's idea, I implement a para virt interface for perf to support
> statistics collection in guest os. That means we could run tool perf in guest
> os directly.
>
> Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
> design suggestions. I also want to thank Yangsheng and LinMing for their generous
> help.
>
> The design is:
>
> 1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
> 2) Create a host perf_event per guest perf_event;
> 3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
> when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
> kernel if a guest event overflows.
> 4) Guest kernel goes through all enabled event on current cpu and output data when they
> overflows.
> 5) No change in user space.
>
One thing that's missing is documentation of the guest/host ABI. It
will be a requirement for inclusion, but it will also be a great help
for review, so please provide it ASAP.
Please also split the guest and host parts into separate patches.
>
> -#define KVM_MAX_MMU_OP_BATCH 32
>
Keep that please.
> +/* Operations for KVM_PERF_OP */
> +#define KVM_PERF_OP_OPEN 1
> +#define KVM_PERF_OP_CLOSE 2
> +#define KVM_PERF_OP_ENABLE 3
> +#define KVM_PERF_OP_DISABLE 4
> +#define KVM_PERF_OP_START 5
> +#define KVM_PERF_OP_STOP 6
> +#define KVM_PERF_OP_READ 7
>
Where is the hypercall number for the perf hypercall?
> +static bool kvm_reserve_pmc_hardware(void)
> +{
> + if (nmi_watchdog == NMI_LOCAL_APIC)
> + disable_lapic_nmi_watchdog();
> +
> + return true;
> +}
> +
> +static void kvm_release_pmc_hardware(void)
> +{
> + if (nmi_watchdog == NMI_LOCAL_APIC)
> + enable_lapic_nmi_watchdog();
> +}
>
Disabling the watchdog is unfortunate. Why is it necessary?
> +
> +static int kvm_pmu_enable(struct perf_event *event)
> +{
> + int ret;
> + unsigned long addr = __pa(event);
> +
> + if (kvm_add_event(event))
> + return -1;
> +
> + ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_ENABLE,
> + addr, (unsigned long) event->shadow);
>
This is suspicious. Passing virtual addresses to the host is always
problematic. Or event->shadow only used as an opaque cookie?
> +int __init kvm_init_hw_perf_events(void)
> +{
> + if (!kvm_para_available())
> + return -1;
> +
> + x86_pmu.handle_irq = kvm_default_x86_handle_irq;
> +
> + pr_cont("KVM PARA PMU driver.\n");
> + register_die_notifier(&kvm_perf_event_nmi_notifier);
> +
> + return 0;
> +}
>
Need to detect the kvm pmu via its own cpuid bit.
> +
> +static int __kvm_hw_perf_event_init(struct perf_event *event)
> +{
> + struct hw_perf_event *hwc =&event->hw;
> + int err;
> + unsigned long result;
> + unsigned long addr;
> +
> + err = 0;
> + if (!atomic_inc_not_zero(&active_events)) {
> + mutex_lock(&pmc_reserve_mutex);
> + if (atomic_read(&active_events) == 0) {
> + if (!kvm_reserve_pmc_hardware())
> + err = -EBUSY;
> + }
> + if (!err)
> + atomic_inc(&active_events);
> + mutex_unlock(&pmc_reserve_mutex);
> + if (err)
> + return err;
> + }
> +
> + event->destroy = kvm_hw_perf_event_destroy;
> +
> + hwc->idx = -1;
> + hwc->last_cpu = -1;
> + hwc->last_tag = ~0ULL;
> +
> + addr = __pa(event);
> + result = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, addr, 0);
> +
> + if (result)
> + event->shadow = (void *) result;
> + else
> + err = -1;
>
Ok, so event->shadow is never dereferenced. In that case better not
make it a pointer at all, keep it unsigned long.
Note that you can run 32-bit guests on 64-bit hosts, so the cookie
better not exceed 32 bits.
> +
> +static void kvm_sync_event_to_guest(struct perf_event *event, int overflows)
> +{
> + struct hw_perf_event *hwc =&event->hw;
> + struct kvmperf_event *kvmevent;
> + int offset, len, data_len, copied, page_offset;
> + struct page *event_page;
> + void *shared_kaddr;
> +
> + kvmevent = event->shadow;
> + offset = kvmevent->event_offset;
> +
> + /* Copy perf_event->count firstly */
> + offset += offsetof(struct perf_event, count);
> + if (offset< PAGE_SIZE) {
> + event_page = kvmevent->event_page;
> + page_offset = offset;
> + } else {
> + event_page = kvmevent->event_page2;
> + page_offset = offset - PAGE_SIZE;
> + }
> + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> + *((atomic64_t *)(shared_kaddr + page_offset)) = event->count;
>
This copy will not be atomic on 32-bit hosts.
> +
> + offset = kvmevent->event_offset;
> + offset += offsetof(struct perf_event, hw);
> + if (offset< PAGE_SIZE) {
> + if (event_page == kvmevent->event_page2) {
> + kunmap_atomic(shared_kaddr, KM_USER0);
> + event_page = kvmevent->event_page;
> + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> + }
> + page_offset = offset;
> + } else {
> + if (event_page == kvmevent->event_page) {
> + kunmap_atomic(shared_kaddr, KM_USER0);
> + event_page = kvmevent->event_page2;
> + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> + }
> + page_offset = offset - PAGE_SIZE;
> + }
> +
> + if (overflows)
> + atomic_add(overflows, (atomic_t *)(shared_kaddr + page_offset));
> +
> + kunmap_atomic(shared_kaddr, KM_USER0);
>
Where is the actual event copied? I'm afraid it's hard to understand
the code.
> +#if 0
> + offset += offsetof(struct hw_perf_event, prev_count);
> + data_len = sizeof(struct hw_perf_event) -
> + offsetof(struct hw_perf_event, prev_count);
> + if (event_page == kvmevent->event_page2) {
> + page_offset += offsetof(struct hw_perf_event, prev_count);
> + memcpy(shared_kaddr + page_offset,
> + &hwc->prev_count, data_len);
> + kunmap_atomic(shared_kaddr, KM_USER0);
> +
> + return;
> + }
> +
> + copied = 0;
> + if (offset< PAGE_SIZE) {
> + len = PAGE_SIZE - offset;
> + if (len> data_len)
> + len = data_len;
> + memcpy(shared_kaddr + offset,
> + &hwc->prev_count, data_len);
> + copied = len;
> + page_offset = 0;
> + } else
> + page_offset = offset - PAGE_SIZE;
> +
> + kunmap_atomic(shared_kaddr, KM_USER0);
> + len = data_len - copied;
> + if (len) {
> + /* Copy across pages */
> + shared_kaddr = kmap_atomic(kvmevent->event_page2, KM_USER0);
> + memcpy(shared_kaddr + page_offset,
> + ((void *)&hwc->prev_count) + copied, len);
> + kunmap_atomic(shared_kaddr, KM_USER0);
> + }
> +#endif
>
Maybe here, but the #if 0 doesn't help.
> +}
> +
>
> +static struct perf_event *
> +kvm_pv_perf_op_open(struct kvm_vcpu *vcpu, gpa_t addr)
> +{
> + int ret;
> + struct perf_event *event;
> + struct perf_event *host_event = NULL;
> + struct kvmperf_event *shadow = NULL;
> +
> + event = kzalloc(sizeof(*event), GFP_KERNEL);
> + if (!event)
> + goto out;
> +
> + shadow = kzalloc(sizeof(*shadow), GFP_KERNEL);
> + if (!shadow)
> + goto out;
> +
> + shadow->event_page = gfn_to_page(vcpu->kvm, addr>> PAGE_SHIFT);
> + shadow->event_offset = addr& ~PAGE_MASK;
>
Might truncate on 32-bit hosts. PAGE_MASK is 32-bit while addr is 64 bit.
> + if (shadow->event_offset + sizeof(struct perf_event)> PAGE_SIZE) {
> + shadow->event_page2 = gfn_to_page(vcpu->kvm,
> + (addr>> PAGE_SHIFT) + 1);
> + }
>
My be simpler to make event_pages an array instead of two independent
variables.
> +
> + ret = kvm_read_guest(vcpu->kvm, addr, event, sizeof(*event));
> + if (ret)
> + goto out;
>
I assume this is to check existence? It doesn't work well with memory
hotplug. In general it's preferred to use
kvm_read_guest()/kvm_write_guest() instead of gfn_to_page() during
initialization to prevent pinning and allow for hotplug.
> +
> + /*
> + * By default, we disable the host event. Later on, guets os
> + * triggers a perf_event_attach to enable it
> + */
> + event->attr.disabled = 1;
> + event->attr.inherit = 0;
> + event->attr.enable_on_exec = 0;
> + /*
> + * We don't support exclude mode of user and kernel for guest os,
> + * which mean we always collect both user and kernel for guest os
> + */
> + event->attr.exclude_user = 0;
> + event->attr.exclude_kernel = 0;
> + /* We always create a cpu context host perf event */
> +
> + host_event = perf_event_create_kernel_counter(&event->attr, -1,
> + current->pid, kvm_perf_event_overflow);
> +
> + if (IS_ERR(host_event)) {
> + host_event = NULL;
> + goto out;
> + }
> + host_event->shadow = shadow;
> +
> +out:
> + if (!host_event)
> + kfree(shadow);
> + kfree(event);
> +
> + return host_event;
> +}
> +
>
> +
> +int kvm_pv_perf_op(struct kvm_vcpu *vcpu, int op_code, unsigned long a1,
> + unsigned long a2, unsigned long *result)
> +{
> + unsigned long ret;
> + struct perf_event *host_event;
> + gpa_t addr;
> +
> + addr = (gpa_t)(a1);
>
A 32-bit guest has a 64-bit gpa_t but 32-bit ulongs, so gpas need to be
passed in two hypervall arguments.
> +
> + switch(op_code) {
> + case KVM_PERF_OP_OPEN:
> + ret = (unsigned long) kvm_pv_perf_op_open(vcpu, addr);
> + break;
> + case KVM_PERF_OP_CLOSE:
> + host_event = (struct perf_event *) a2;
>
First, you get truncation in a 32-bit guest. Second, you must validate
the argument! The guest kernel can easily subvert the host by passing a
bogus host_event.
It may be better to have the guest create an id to the host, and the
host can simply look up the id in a table:
if (a2 >= KVM_PMU_MAX_EVENTS)
bail out;
host_event = vcpu->pmu.host_events[a2];
> @@ -4052,6 +4054,16 @@ static unsigned long kvm_get_guest_ip(vo
> return ip;
> }
>
> +int kvm_notify_event_overflow(void)
> +{
> + if (percpu_read(current_vcpu)) {
> + kvm_inject_nmi(percpu_read(current_vcpu));
> + return 0;
> + }
> +
> + return -1;
> +}
>
This should really go through the APIC PERF LVT. No interface
currently, but we are working on it.
This way the guest can configure the perf interrupt to be NMI, INTR, or
anything it likes.
--
error compiling committee.c: too many arguments to function
On 06/09/2010 06:30 AM, Zhang, Yanmin wrote:
> From: Zhang, Yanmin<[email protected]>
>
> Based on Ingo's idea, I implement a para virt interface for perf to support
> statistics collection in guest os. That means we could run tool perf in guest
> os directly.
>
> Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
> design suggestions. I also want to thank Yangsheng and LinMing for their generous
> help.
>
> The design is:
>
> 1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
> 2) Create a host perf_event per guest perf_event;
> 3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
> when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
> kernel if a guest event overflows.
> 4) Guest kernel goes through all enabled event on current cpu and output data when they
> overflows.
> 5) No change in user space.
>
Other issues:
- save/restore support for live migration
- some way to limit the number of open handles (comes automatically with
the table approach I suggested earlier)
--
error compiling committee.c: too many arguments to function
On Wed, 2010-06-09 at 11:33 +0300, Avi Kivity wrote:
> On 06/09/2010 06:30 AM, Zhang, Yanmin wrote:
> > From: Zhang, Yanmin<[email protected]>
> >
> > Based on Ingo's idea, I implement a para virt interface for perf to support
> > statistics collection in guest os. That means we could run tool perf in guest
> > os directly.
> >
> > Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
> > design suggestions. I also want to thank Yangsheng and LinMing for their generous
> > help.
> >
> > The design is:
> >
> > 1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
> > 2) Create a host perf_event per guest perf_event;
> > 3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
> > when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
> > kernel if a guest event overflows.
> > 4) Guest kernel goes through all enabled event on current cpu and output data when they
> > overflows.
> > 5) No change in user space.
> >
>
Thanks for your kind comments.
> One thing that's missing is documentation of the guest/host ABI. It
> will be a requirement for inclusion, but it will also be a great help
> for review, so please provide it ASAP.
I will add such document. It will includes:
1) Data struct perf_event definition. Guest os and host os have to share the same
data structure as host kernel will sync data (counte/overflows and others if needed)
changes back to guest os.
2) A list of all hypercalls
3) Guest need have NMI handler which checks all guest events.
>
> Please also split the guest and host parts into separate patches.
I will do so.
>
> >
> > -#define KVM_MAX_MMU_OP_BATCH 32
> >
>
> Keep that please.
I will do so.
>
> > +/* Operations for KVM_PERF_OP */
> > +#define KVM_PERF_OP_OPEN 1
> > +#define KVM_PERF_OP_CLOSE 2
> > +#define KVM_PERF_OP_ENABLE 3
> > +#define KVM_PERF_OP_DISABLE 4
> > +#define KVM_PERF_OP_START 5
> > +#define KVM_PERF_OP_STOP 6
> > +#define KVM_PERF_OP_READ 7
> >
>
> Where is the hypercall number for the perf hypercall?
I defines it in file kvm_para.h like KVM_HC_MMU_OP.
>
> > +static bool kvm_reserve_pmc_hardware(void)
> > +{
> > + if (nmi_watchdog == NMI_LOCAL_APIC)
> > + disable_lapic_nmi_watchdog();
> > +
> > + return true;
> > +}
> > +
> > +static void kvm_release_pmc_hardware(void)
> > +{
> > + if (nmi_watchdog == NMI_LOCAL_APIC)
> > + enable_lapic_nmi_watchdog();
> > +}
> >
>
> Disabling the watchdog is unfortunate. Why is it necessary?
perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
set up in case they might have impact.
>
> > +
> > +static int kvm_pmu_enable(struct perf_event *event)
> > +{
> > + int ret;
> > + unsigned long addr = __pa(event);
> > +
> > + if (kvm_add_event(event))
> > + return -1;
> > +
> > + ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_ENABLE,
> > + addr, (unsigned long) event->shadow);
> >
>
> This is suspicious. Passing virtual addresses to the host is always
> problematic. Or event->shadow only used as an opaque cookie?
I use perf_event->shadow at both host and guest side.
1) At host side, perf_event->shadow points to an area including the page
mapping information about guest perf_event. Host kernel uses it to sync data
changes back to guest;
2) At guest side, perf_event->shadow save the virtual address of host
perf_event at host side. At guest side,
kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, ...) return the virtual address.
Guest os shouldn't use it but using it to pass it back to host kernel in following
hypercalls. It might be a security issue for host kernel. Originally, I planed guest
os not to use perf_event->shadow. Host kernel maintains a per-vcpu event-related
list whose key is addr of guest perf_event. But considering the vcpu thread migration
on logical cpu, such list needs lock and implementation becomes a little complicated.
I will double-check the list method as the security issue is there.
>
> > +int __init kvm_init_hw_perf_events(void)
> > +{
> > + if (!kvm_para_available())
> > + return -1;
> > +
> > + x86_pmu.handle_irq = kvm_default_x86_handle_irq;
> > +
> > + pr_cont("KVM PARA PMU driver.\n");
> > + register_die_notifier(&kvm_perf_event_nmi_notifier);
> > +
> > + return 0;
> > +}
> >
>
> Need to detect the kvm pmu via its own cpuid bit.
Ok. I will add a feature, KVM_FEATURE_PARA_PERF, something like
bit KVM_FEATURE_CLOCKSOURCE.
>
> > +
> > +static int __kvm_hw_perf_event_init(struct perf_event *event)
> > +{
> > + struct hw_perf_event *hwc =&event->hw;
> > + int err;
> > + unsigned long result;
> > + unsigned long addr;
> > +
> > + err = 0;
> > + if (!atomic_inc_not_zero(&active_events)) {
> > + mutex_lock(&pmc_reserve_mutex);
> > + if (atomic_read(&active_events) == 0) {
> > + if (!kvm_reserve_pmc_hardware())
> > + err = -EBUSY;
> > + }
> > + if (!err)
> > + atomic_inc(&active_events);
> > + mutex_unlock(&pmc_reserve_mutex);
> > + if (err)
> > + return err;
> > + }
> > +
> > + event->destroy = kvm_hw_perf_event_destroy;
> > +
> > + hwc->idx = -1;
> > + hwc->last_cpu = -1;
> > + hwc->last_tag = ~0ULL;
> > +
> > + addr = __pa(event);
> > + result = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, addr, 0);
> > +
> > + if (result)
> > + event->shadow = (void *) result;
> > + else
> > + err = -1;
> >
>
> Ok, so event->shadow is never dereferenced. In that case better not
> make it a pointer at all, keep it unsigned long.
Host kernel also uses it.
>
> Note that you can run 32-bit guests on 64-bit hosts, so the cookie
> better not exceed 32 bits.
I forgot 32 bits. I need double-check it. It seems I have to implement a list
at host side and don't use it at guest side any more.
>
> > +
> > +static void kvm_sync_event_to_guest(struct perf_event *event, int overflows)
> > +{
> > + struct hw_perf_event *hwc =&event->hw;
> > + struct kvmperf_event *kvmevent;
> > + int offset, len, data_len, copied, page_offset;
> > + struct page *event_page;
> > + void *shared_kaddr;
> > +
> > + kvmevent = event->shadow;
> > + offset = kvmevent->event_offset;
> > +
> > + /* Copy perf_event->count firstly */
> > + offset += offsetof(struct perf_event, count);
> > + if (offset< PAGE_SIZE) {
> > + event_page = kvmevent->event_page;
> > + page_offset = offset;
> > + } else {
> > + event_page = kvmevent->event_page2;
> > + page_offset = offset - PAGE_SIZE;
> > + }
> > + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> > + *((atomic64_t *)(shared_kaddr + page_offset)) = event->count;
> >
>
> This copy will not be atomic on 32-bit hosts.
Right. But it shouldn't be a problem as vcpu thread stops when vmexit to
host to process events. In addition, only current cpu in guest accesses
perf_events linked to current cpu.
>
> > +
> > + offset = kvmevent->event_offset;
> > + offset += offsetof(struct perf_event, hw);
> > + if (offset< PAGE_SIZE) {
> > + if (event_page == kvmevent->event_page2) {
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> > + event_page = kvmevent->event_page;
> > + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> > + }
> > + page_offset = offset;
> > + } else {
> > + if (event_page == kvmevent->event_page) {
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> > + event_page = kvmevent->event_page2;
> > + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> > + }
> > + page_offset = offset - PAGE_SIZE;
> > + }
> > +
> > + if (overflows)
> > + atomic_add(overflows, (atomic_t *)(shared_kaddr + page_offset));
> > +
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> >
>
> Where is the actual event copied? I'm afraid it's hard to understand
> the code.
Sorry, I should have more statements around the codes.
When an event overflows, host kernel sync perf_event->count to guest os
perf_event->count, and increases the perf_event->hw.overflows.
+struct kvmperf_event {
+ unsigned int event_offset;
+ struct page *event_page;
+ struct page *event_page2;
+};
Actually, at host side, perf_event->shadow points to a kvmperf_event.
kvmperf_event->event_page points the physical page of guest perf_event. event_page2
points to the 2nd page if perf_event data structure layouts across 2 pages.
event_offset marks the offset start in the 1st page.
>
> > +#if 0
> > + offset += offsetof(struct hw_perf_event, prev_count);
> > + data_len = sizeof(struct hw_perf_event) -
> > + offsetof(struct hw_perf_event, prev_count);
> > + if (event_page == kvmevent->event_page2) {
> > + page_offset += offsetof(struct hw_perf_event, prev_count);
> > + memcpy(shared_kaddr + page_offset,
> > + &hwc->prev_count, data_len);
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> > +
> > + return;
> > + }
> > +
> > + copied = 0;
> > + if (offset< PAGE_SIZE) {
> > + len = PAGE_SIZE - offset;
> > + if (len> data_len)
> > + len = data_len;
> > + memcpy(shared_kaddr + offset,
> > + &hwc->prev_count, data_len);
> > + copied = len;
> > + page_offset = 0;
> > + } else
> > + page_offset = offset - PAGE_SIZE;
> > +
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> > + len = data_len - copied;
> > + if (len) {
> > + /* Copy across pages */
> > + shared_kaddr = kmap_atomic(kvmevent->event_page2, KM_USER0);
> > + memcpy(shared_kaddr + page_offset,
> > + ((void *)&hwc->prev_count) + copied, len);
> > + kunmap_atomic(shared_kaddr, KM_USER0);
> > + }
> > +#endif
> >
>
> Maybe here, but the #if 0 doesn't help.
The codes in '#if 0' is trying to copy more data back to guest os
perf_event. I comment them out as they are not really used by guest os.
>
> > +}
> > +
> >
> > +static struct perf_event *
> > +kvm_pv_perf_op_open(struct kvm_vcpu *vcpu, gpa_t addr)
> > +{
> > + int ret;
> > + struct perf_event *event;
> > + struct perf_event *host_event = NULL;
> > + struct kvmperf_event *shadow = NULL;
> > +
> > + event = kzalloc(sizeof(*event), GFP_KERNEL);
> > + if (!event)
> > + goto out;
> > +
> > + shadow = kzalloc(sizeof(*shadow), GFP_KERNEL);
> > + if (!shadow)
> > + goto out;
> > +
> > + shadow->event_page = gfn_to_page(vcpu->kvm, addr>> PAGE_SHIFT);
> > + shadow->event_offset = addr& ~PAGE_MASK;
> >
>
> Might truncate on 32-bit hosts. PAGE_MASK is 32-bit while addr is 64 bit.
Above codes just run at host side. Is it possible that host kernel is 32 bit and
guest kernel is 64bits? I really need separate the patch to host and guest parts
as one patch misleads you guys.
>
> > + if (shadow->event_offset + sizeof(struct perf_event)> PAGE_SIZE) {
> > + shadow->event_page2 = gfn_to_page(vcpu->kvm,
> > + (addr>> PAGE_SHIFT) + 1);
> > + }
> >
>
> My be simpler to make event_pages an array instead of two independent
> variables.
Ok, I will do so.
>
> > +
> > + ret = kvm_read_guest(vcpu->kvm, addr, event, sizeof(*event));
> > + if (ret)
> > + goto out;
> >
>
> I assume this is to check existence?
Here calling kvm_read_guest is to get a copy of guest perf_event as it has
perf_event.attr. Host need the attr to create host perf_event.
> It doesn't work well with memory
> hotplug. In general it's preferred to use
> kvm_read_guest()/kvm_write_guest() instead of gfn_to_page() during
> initialization to prevent pinning and allow for hotplug.
That's an issue. But host kernel couldn't go to sleep when processing event
overflows under NMI context.
>
> > +
> > + /*
> > + * By default, we disable the host event. Later on, guets os
> > + * triggers a perf_event_attach to enable it
> > + */
> > + event->attr.disabled = 1;
> > + event->attr.inherit = 0;
> > + event->attr.enable_on_exec = 0;
> > + /*
> > + * We don't support exclude mode of user and kernel for guest os,
> > + * which mean we always collect both user and kernel for guest os
> > + */
> > + event->attr.exclude_user = 0;
> > + event->attr.exclude_kernel = 0;
> > + /* We always create a cpu context host perf event */
> > +
> > + host_event = perf_event_create_kernel_counter(&event->attr, -1,
> > + current->pid, kvm_perf_event_overflow);
> > +
> > + if (IS_ERR(host_event)) {
> > + host_event = NULL;
> > + goto out;
> > + }
> > + host_event->shadow = shadow;
> > +
> > +out:
> > + if (!host_event)
> > + kfree(shadow);
> > + kfree(event);
> > +
> > + return host_event;
> > +}
> > +
> >
> > +
> > +int kvm_pv_perf_op(struct kvm_vcpu *vcpu, int op_code, unsigned long a1,
> > + unsigned long a2, unsigned long *result)
> > +{
> > + unsigned long ret;
> > + struct perf_event *host_event;
> > + gpa_t addr;
> > +
> > + addr = (gpa_t)(a1);
> >
>
> A 32-bit guest has a 64-bit gpa_t but 32-bit ulongs, so gpas need to be
> passed in two hypervall arguments.
Ok. Originally I pass 2 parameters like hc_gpa. I will redo it.
>
> > +
> > + switch(op_code) {
> > + case KVM_PERF_OP_OPEN:
> > + ret = (unsigned long) kvm_pv_perf_op_open(vcpu, addr);
> > + break;
> > + case KVM_PERF_OP_CLOSE:
> > + host_event = (struct perf_event *) a2;
> >
>
> First, you get truncation in a 32-bit guest. Second, you must validate
> the argument! The guest kernel can easily subvert the host by passing a
> bogus host_event.
You are right. I will implement the list method at host side.
>
> It may be better to have the guest create an id to the host, and the
> host can simply look up the id in a table:
Perhaps the address of guest perf_event is the best id.
>
> if (a2 >= KVM_PMU_MAX_EVENTS)
> bail out;
> host_event = vcpu->pmu.host_events[a2];
>
> > @@ -4052,6 +4054,16 @@ static unsigned long kvm_get_guest_ip(vo
> > return ip;
> > }
> >
> > +int kvm_notify_event_overflow(void)
> > +{
> > + if (percpu_read(current_vcpu)) {
> > + kvm_inject_nmi(percpu_read(current_vcpu));
> > + return 0;
> > + }
> > +
> > + return -1;
> > +}
> >
>
> This should really go through the APIC PERF LVT. No interface
> currently, but we are working on it.
>
> This way the guest can configure the perf interrupt to be NMI, INTR, or
> anything it likes.
>
Thanks for the pointer.
Yanmin
On Wed, 2010-06-09 at 11:59 +0300, Avi Kivity wrote:
> On 06/09/2010 06:30 AM, Zhang, Yanmin wrote:
> > From: Zhang, Yanmin<[email protected]>
> >
> > Based on Ingo's idea, I implement a para virt interface for perf to support
> > statistics collection in guest os. That means we could run tool perf in guest
> > os directly.
> >
> > Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
> > design suggestions. I also want to thank Yangsheng and LinMing for their generous
> > help.
> >
> > The design is:
> >
> > 1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
> > 2) Create a host perf_event per guest perf_event;
> > 3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
> > when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
> > kernel if a guest event overflows.
> > 4) Guest kernel goes through all enabled event on current cpu and output data when they
> > overflows.
> > 5) No change in user space.
> >
>
> Other issues:
>
> - save/restore support for live migration
Well, it's a little hard to process perf_event under live migration case.
I will check it.
> - some way to limit the number of open handles (comes automatically with
> the table approach I suggested earlier)
Current perf doesn't restrict perf_event number. Kernel does a rotation to collect
statistics of all perf_events. My patch just follows this style.
The table method might be not good, because below scenario:
guest perf_event might be a per-task event at guest side. When the guest application task is
migrated to another cpu, the perf_event peer at host side should also be migrated to the new vcpu
thread. With table method, we need do some rearrangement on the table when event migration happens.
Here migration I mention is not guest live migration.
I will double-check it.
Thanks,
Yanmin
On 06/09/2010 12:21 PM, Zhang, Yanmin wrote:
>
>> One thing that's missing is documentation of the guest/host ABI. It
>> will be a requirement for inclusion, but it will also be a great help
>> for review, so please provide it ASAP.
>>
> I will add such document. It will includes:
> 1) Data struct perf_event definition. Guest os and host os have to share the same
> data structure as host kernel will sync data (counte/overflows and others if needed)
> changes back to guest os.
> 2) A list of all hypercalls
> 3) Guest need have NMI handler which checks all guest events.
>
Thanks. It may be easier to have just the documentation for the first
few iterations so we can converge on a good interface, then update the
code to match the interface.
>> Disabling the watchdog is unfortunate. Why is it necessary?
>>
> perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
> set up in case they might have impact.
>
Ok. Is that the case for the hardware pmus as well? If so it might be
done in common code.
>>> +
>>> +static int kvm_pmu_enable(struct perf_event *event)
>>> +{
>>> + int ret;
>>> + unsigned long addr = __pa(event);
>>> +
>>> + if (kvm_add_event(event))
>>> + return -1;
>>> +
>>> + ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_ENABLE,
>>> + addr, (unsigned long) event->shadow);
>>>
>>>
>> This is suspicious. Passing virtual addresses to the host is always
>> problematic. Or event->shadow only used as an opaque cookie?
>>
> I use perf_event->shadow at both host and guest side.
> 1) At host side, perf_event->shadow points to an area including the page
> mapping information about guest perf_event. Host kernel uses it to sync data
> changes back to guest;
> 2) At guest side, perf_event->shadow save the virtual address of host
> perf_event at host side. At guest side,
> kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, ...) return the virtual address.
> Guest os shouldn't use it but using it to pass it back to host kernel in following
> hypercalls. It might be a security issue for host kernel. Originally, I planed guest
> os not to use perf_event->shadow. Host kernel maintains a per-vcpu event-related
> list whose key is addr of guest perf_event. But considering the vcpu thread migration
> on logical cpu, such list needs lock and implementation becomes a little complicated.
>
> I will double-check the list method as the security issue is there.
>
Besides the other concern, you cannot live migrate a host virtual
address, since it changes from host to host. It's better to use a
guest-chosen bounded small integer.
>> Need to detect the kvm pmu via its own cpuid bit.
>>
> Ok. I will add a feature, KVM_FEATURE_PARA_PERF, something like
> bit KVM_FEATURE_CLOCKSOURCE.
>
Don't forget Documentation/kvm/cpuid.txt.
>> Ok, so event->shadow is never dereferenced. In that case better not
>> make it a pointer at all, keep it unsigned long.
>>
> Host kernel also uses it.
>
I see. So put it in a union. Or perhaps not even in a union - what if
a kvm guest is also acting as a kvm host?
>
>>
>>> +
>>> +static void kvm_sync_event_to_guest(struct perf_event *event, int overflows)
>>> +{
>>> + struct hw_perf_event *hwc =&event->hw;
>>> + struct kvmperf_event *kvmevent;
>>> + int offset, len, data_len, copied, page_offset;
>>> + struct page *event_page;
>>> + void *shared_kaddr;
>>> +
>>> + kvmevent = event->shadow;
>>> + offset = kvmevent->event_offset;
>>> +
>>> + /* Copy perf_event->count firstly */
>>> + offset += offsetof(struct perf_event, count);
>>> + if (offset< PAGE_SIZE) {
>>> + event_page = kvmevent->event_page;
>>> + page_offset = offset;
>>> + } else {
>>> + event_page = kvmevent->event_page2;
>>> + page_offset = offset - PAGE_SIZE;
>>> + }
>>> + shared_kaddr = kmap_atomic(event_page, KM_USER0);
>>> + *((atomic64_t *)(shared_kaddr + page_offset)) = event->count;
>>>
>>>
>> This copy will not be atomic on 32-bit hosts.
>>
> Right. But it shouldn't be a problem as vcpu thread stops when vmexit to
> host to process events. In addition, only current cpu in guest accesses
> perf_events linked to current cpu.
>
Ok. These restrictions should be documented.
>>
>>> +}
>>> +
>>>
>>> +static struct perf_event *
>>> +kvm_pv_perf_op_open(struct kvm_vcpu *vcpu, gpa_t addr)
>>> +{
>>> + int ret;
>>> + struct perf_event *event;
>>> + struct perf_event *host_event = NULL;
>>> + struct kvmperf_event *shadow = NULL;
>>> +
>>> + event = kzalloc(sizeof(*event), GFP_KERNEL);
>>> + if (!event)
>>> + goto out;
>>> +
>>> + shadow = kzalloc(sizeof(*shadow), GFP_KERNEL);
>>> + if (!shadow)
>>> + goto out;
>>> +
>>> + shadow->event_page = gfn_to_page(vcpu->kvm, addr>> PAGE_SHIFT);
>>> + shadow->event_offset = addr& ~PAGE_MASK;
>>>
>>>
>> Might truncate on 32-bit hosts. PAGE_MASK is 32-bit while addr is 64 bit.
>>
> Above codes just run at host side. Is it possible that host kernel is 32 bit and
> guest kernel is 64bits?
No, guest bitness always <= host bitness. But gpa_t is 64-bit even on
32-bit host/guest, so you can't use PAGE_MASK on that.
>>> +
>>> + ret = kvm_read_guest(vcpu->kvm, addr, event, sizeof(*event));
>>> + if (ret)
>>> + goto out;
>>>
>>>
>> I assume this is to check existence?
>>
> Here calling kvm_read_guest is to get a copy of guest perf_event as it has
> perf_event.attr. Host need the attr to create host perf_event.
>
>
>> It doesn't work well with memory
>> hotplug. In general it's preferred to use
>> kvm_read_guest()/kvm_write_guest() instead of gfn_to_page() during
>> initialization to prevent pinning and allow for hotplug.
>>
> That's an issue. But host kernel couldn't go to sleep when processing event
> overflows under NMI context.
>
You can set a bit in vcpu->requests and schedule the copying there.
vcpu->requests is always checked before guest entry.
>
>> It may be better to have the guest create an id to the host, and the
>> host can simply look up the id in a table:
>>
> Perhaps the address of guest perf_event is the best id.
>
That will need to be looked up in a hash table. A small id is best
because it can be easily looked up by both guest and host.
--
error compiling committee.c: too many arguments to function
On 06/09/2010 12:30 PM, Zhang, Yanmin wrote:
> On Wed, 2010-06-09 at 11:59 +0300, Avi Kivity wrote:
>
>> On 06/09/2010 06:30 AM, Zhang, Yanmin wrote:
>>
>>> From: Zhang, Yanmin<[email protected]>
>>>
>>> Based on Ingo's idea, I implement a para virt interface for perf to support
>>> statistics collection in guest os. That means we could run tool perf in guest
>>> os directly.
>>>
>>> Great thanks to Peter Zijlstra. He is really the architect and gave me architecture
>>> design suggestions. I also want to thank Yangsheng and LinMing for their generous
>>> help.
>>>
>>> The design is:
>>>
>>> 1) Add a kvm_pmu whose callbacks mostly just calls hypercall to vmexit to host kernel;
>>> 2) Create a host perf_event per guest perf_event;
>>> 3) Host kernel syncs perf_event count/overflows data changes to guest perf_event
>>> when processing perf_event overflows after NMI arrives. Host kernel inject NMI to guest
>>> kernel if a guest event overflows.
>>> 4) Guest kernel goes through all enabled event on current cpu and output data when they
>>> overflows.
>>> 5) No change in user space.
>>>
>>>
>> Other issues:
>>
>> - save/restore support for live migration
>>
> Well, it's a little hard to process perf_event under live migration case.
> I will check it.
>
It's probably the biggest benefit of paravirt PMU over non-paravirt PMU,
and live migration is one of the most important features of
virtualization. So we really need to get this working.
>> - some way to limit the number of open handles (comes automatically with
>> the table approach I suggested earlier)
>>
> Current perf doesn't restrict perf_event number. Kernel does a rotation to collect
> statistics of all perf_events.
We must have some restriction, since we consume host resources for each
perf_event.
> My patch just follows this style.
> The table method might be not good, because below scenario:
> guest perf_event might be a per-task event at guest side. When the guest application task is
> migrated to another cpu, the perf_event peer at host side should also be migrated to the new vcpu
> thread. With table method, we need do some rearrangement on the table when event migration happens.
> Here migration I mention is not guest live migration.
>
Yes. But the code for that already exists, no? Real hardware has
limited resources so perf multiplexes unlimited user perf_events on
limited hardware perf_events. The same can happen here, perhaps with a
larger limit.
--
error compiling committee.c: too many arguments to function
On Wed, 2010-06-09 at 12:41 +0300, Avi Kivity wrote:
>
> >> Disabling the watchdog is unfortunate. Why is it necessary?
> >>
> > perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
> > set up in case they might have impact.
> >
>
> Ok. Is that the case for the hardware pmus as well? If so it might be
> done in common code.
The x86 hardware pmu implementation disables the lapic watchdog too, but
recent kernels come with a watchdog implementation on top of perf, the
old lapic one will be depricated.
* Peter Zijlstra <[email protected]> wrote:
> On Wed, 2010-06-09 at 12:41 +0300, Avi Kivity wrote:
> >
> > >> Disabling the watchdog is unfortunate. Why is it necessary?
> > >>
> > > perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
> > > set up in case they might have impact.
> > >
> >
> > Ok. Is that the case for the hardware pmus as well? If so it might be
> > done in common code.
>
> The x86 hardware pmu implementation disables the lapic watchdog too, but
> recent kernels come with a watchdog implementation on top of perf, the old
> lapic one will be depricated.
Note, that code is in -tip, queued for v2.6.36.
Ingo
On 06/09/2010 01:08 PM, Peter Zijlstra wrote:
> On Wed, 2010-06-09 at 12:41 +0300, Avi Kivity wrote:
>
>>
>>>> Disabling the watchdog is unfortunate. Why is it necessary?
>>>>
>>>>
>>> perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
>>> set up in case they might have impact.
>>>
>>>
>> Ok. Is that the case for the hardware pmus as well? If so it might be
>> done in common code.
>>
> The x86 hardware pmu implementation disables the lapic watchdog too, but
> recent kernels come with a watchdog implementation on top of perf, the
> old lapic one will be depricated.
>
So this should indeed be in common code, and removed (for all pmus) when
the new watchdog is merged in 2.6.36.
--
error compiling committee.c: too many arguments to function
On Wed, 2010-06-09 at 12:41 +0300, Avi Kivity wrote:
> On 06/09/2010 12:21 PM, Zhang, Yanmin wrote:
> >
> >> One thing that's missing is documentation of the guest/host ABI. It
> >> will be a requirement for inclusion, but it will also be a great help
> >> for review, so please provide it ASAP.
> >>
> > I will add such document. It will includes:
> > 1) Data struct perf_event definition. Guest os and host os have to share the same
> > data structure as host kernel will sync data (counte/overflows and others if needed)
> > changes back to guest os.
> > 2) A list of all hypercalls
> > 3) Guest need have NMI handler which checks all guest events.
> >
>
> Thanks. It may be easier to have just the documentation for the first
> few iterations so we can converge on a good interface, then update the
> code to match the interface.
I thought over it last night. Your input is important. I need define a clear ABI.
At guest side, I plan to use perf_event->shadow to point to another data area, such like:
struct guest_perf_counter {
u64 count;
atomic_t overflows;
};
So host side just copy data to this area, then guest copy them to its own
perf_event.
The ABI becomes more simple than before. Function kvm_sync_event_to_guest also becomes
clearer. The ABI mostly includes the definition of struct perf_event_attr, guest_perf_counter,
and hypercalls.
> >> Disabling the watchdog is unfortunate. Why is it necessary?
> >>
> > perf always uses NMI, so we disable the nmi_watchdog when a perf_event is
> > set up in case they might have impact.
> >
>
> Ok. Is that the case for the hardware pmus as well? If so it might be
> done in common code.
>
> >>> +
> >>> +static int kvm_pmu_enable(struct perf_event *event)
> >>> +{
> >>> + int ret;
> >>> + unsigned long addr = __pa(event);
> >>> +
> >>> + if (kvm_add_event(event))
> >>> + return -1;
> >>> +
> >>> + ret = kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_ENABLE,
> >>> + addr, (unsigned long) event->shadow);
> >>>
> >>>
> >> This is suspicious. Passing virtual addresses to the host is always
> >> problematic. Or event->shadow only used as an opaque cookie?
> >>
> > I use perf_event->shadow at both host and guest side.
> > 1) At host side, perf_event->shadow points to an area including the page
> > mapping information about guest perf_event. Host kernel uses it to sync data
> > changes back to guest;
> > 2) At guest side, perf_event->shadow save the virtual address of host
> > perf_event at host side. At guest side,
> > kvm_hypercall3(KVM_PERF_OP, KVM_PERF_OP_OPEN, ...) return the virtual address.
> > Guest os shouldn't use it but using it to pass it back to host kernel in following
> > hypercalls. It might be a security issue for host kernel. Originally, I planed guest
> > os not to use perf_event->shadow. Host kernel maintains a per-vcpu event-related
> > list whose key is addr of guest perf_event. But considering the vcpu thread migration
> > on logical cpu, such list needs lock and implementation becomes a little complicated.
> >
> > I will double-check the list method as the security issue is there.
> >
>
> Besides the other concern, you cannot live migrate a host virtual
> address, since it changes from host to host. It's better to use a
> guest-chosen bounded small integer.
Ok. Perhaps a single u32 per guest os instance is enough. So I will change the shadow
pointing to a structure like below in guest kernel:
struct guest_perf_counter {
u64 count;
atomic_t overflows;
};
struct guest_perf_shadow {
u32 id;
struct guest_perf_counter sync_data;
};
atomic_t guest_perf_id; /*Global id counter per guest os*/
>
> >> Need to detect the kvm pmu via its own cpuid bit.
> >>
> > Ok. I will add a feature, KVM_FEATURE_PARA_PERF, something like
> > bit KVM_FEATURE_CLOCKSOURCE.
> >
>
> Don't forget Documentation/kvm/cpuid.txt.
Thanks for your kind reminder.
> >> Ok, so event->shadow is never dereferenced. In that case better not
> >> make it a pointer at all, keep it unsigned long.
> >>
> > Host kernel also uses it.
> >
>
> I see. So put it in a union. Or perhaps not even in a union - what if
> a kvm guest is also acting as a kvm host?
My patch has consideration on it. I compiled kernel with host and guest support
at the same time. The accessing to perf_event->shadow is really under specific
scenarios, or they are just in specific functions. These functions are called
just bu host kernel , or just by guest kernel.
>
>
> >
> >>
> >>> +
> >>> +static void kvm_sync_event_to_guest(struct perf_event *event, int overflows)
> >>> +{
> >>> + struct hw_perf_event *hwc =&event->hw;
> >>> + struct kvmperf_event *kvmevent;
> >>> + int offset, len, data_len, copied, page_offset;
> >>> + struct page *event_page;
> >>> + void *shared_kaddr;
> >>> +
> >>> + kvmevent = event->shadow;
> >>> + offset = kvmevent->event_offset;
> >>> +
> >>> + /* Copy perf_event->count firstly */
> >>> + offset += offsetof(struct perf_event, count);
> >>> + if (offset< PAGE_SIZE) {
> >>> + event_page = kvmevent->event_page;
> >>> + page_offset = offset;
> >>> + } else {
> >>> + event_page = kvmevent->event_page2;
> >>> + page_offset = offset - PAGE_SIZE;
> >>> + }
> >>> + shared_kaddr = kmap_atomic(event_page, KM_USER0);
> >>> + *((atomic64_t *)(shared_kaddr + page_offset)) = event->count;
> >>>
> >>>
> >> This copy will not be atomic on 32-bit hosts.
> >>
> > Right. But it shouldn't be a problem as vcpu thread stops when vmexit to
> > host to process events. In addition, only current cpu in guest accesses
> > perf_events linked to current cpu.
> >
>
> Ok. These restrictions should be documented.
Perhaps I need write them down as code comments in the patch.
>
> >>
> >>> +}
> >>> +
> >>>
> >>> +static struct perf_event *
> >>> +kvm_pv_perf_op_open(struct kvm_vcpu *vcpu, gpa_t addr)
> >>> +{
> >>> + int ret;
> >>> + struct perf_event *event;
> >>> + struct perf_event *host_event = NULL;
> >>> + struct kvmperf_event *shadow = NULL;
> >>> +
> >>> + event = kzalloc(sizeof(*event), GFP_KERNEL);
> >>> + if (!event)
> >>> + goto out;
> >>> +
> >>> + shadow = kzalloc(sizeof(*shadow), GFP_KERNEL);
> >>> + if (!shadow)
> >>> + goto out;
> >>> +
> >>> + shadow->event_page = gfn_to_page(vcpu->kvm, addr>> PAGE_SHIFT);
> >>> + shadow->event_offset = addr& ~PAGE_MASK;
> >>>
> >>>
> >> Might truncate on 32-bit hosts. PAGE_MASK is 32-bit while addr is 64 bit.
> >>
> > Above codes just run at host side. Is it possible that host kernel is 32 bit and
> > guest kernel is 64bits?
>
> No, guest bitness always <= host bitness. But gpa_t is 64-bit even on
> 32-bit host/guest, so you can't use PAGE_MASK on that.
I will check it.
>
> >>> +
> >>> + ret = kvm_read_guest(vcpu->kvm, addr, event, sizeof(*event));
> >>> + if (ret)
> >>> + goto out;
> >>>
> >>>
> >> I assume this is to check existence?
> >>
> > Here calling kvm_read_guest is to get a copy of guest perf_event as it has
> > perf_event.attr. Host need the attr to create host perf_event.
> >
> >
> >> It doesn't work well with memory
> >> hotplug. In general it's preferred to use
> >> kvm_read_guest()/kvm_write_guest() instead of gfn_to_page() during
> >> initialization to prevent pinning and allow for hotplug.
> >>
> > That's an issue. But host kernel couldn't go to sleep when processing event
> > overflows under NMI context.
> >
>
> You can set a bit in vcpu->requests and schedule the copying there.
> vcpu->requests is always checked before guest entry.
That becomes a little complicated as we need record overflowed events in vcpu.
Let me check it again.
>
>
> >
> >> It may be better to have the guest create an id to the host, and the
> >> host can simply look up the id in a table:
> >>
> > Perhaps the address of guest perf_event is the best id.
> >
>
> That will need to be looked up in a hash table. A small id is best
> because it can be easily looked up by both guest and host.
Yes. I will use a u32 or atomic_t.
Thanks for your patience!
Yanmin
On 06/10/2010 05:21 AM, Zhang, Yanmin wrote:
>
>> I see. So put it in a union. Or perhaps not even in a union - what if
>> a kvm guest is also acting as a kvm host?
>>
> My patch has consideration on it. I compiled kernel with host and guest support
> at the same time. The accessing to perf_event->shadow is really under specific
> scenarios, or they are just in specific functions. These functions are called
> just bu host kernel , or just by guest kernel.
>
But a kernel can be both guest and host at the same time. Currently
this only works on AMD, but there was some work to bring it to Intel as
well.
--
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.
On Thu, 2010-06-10 at 06:06 +0300, Avi Kivity wrote:
> On 06/10/2010 05:21 AM, Zhang, Yanmin wrote:
> >
> >> I see. So put it in a union. Or perhaps not even in a union - what if
> >> a kvm guest is also acting as a kvm host?
> >>
> > My patch has consideration on it. I compiled kernel with host and guest support
> > at the same time. The accessing to perf_event->shadow is really under specific
> > scenarios, or they are just in specific functions. These functions are called
> > just bu host kernel , or just by guest kernel.
> >
>
> But a kernel can be both guest and host at the same time. Currently
> this only works on AMD, but there was some work to bring it to Intel as
> well.
Oh, this is a fancy VT feature. My patch doesn't support this mode.
On Thu, 2010-06-10 at 10:21 +0800, Zhang, Yanmin wrote:
> . The ABI mostly includes the definition of struct perf_event_attr,
> guest_perf_counter,
> and hypercalls.
Note that perf_event_attr isn't guaranteed to be stable between kernels,
it can grow.
On Thu, 2010-06-10 at 11:50 +0200, Peter Zijlstra wrote:
> On Thu, 2010-06-10 at 10:21 +0800, Zhang, Yanmin wrote:
> > . The ABI mostly includes the definition of struct perf_event_attr,
> > guest_perf_counter,
> > and hypercalls.
>
> Note that perf_event_attr isn't guaranteed to be stable between kernels,
> it can grow.
Thanks. I need create a small new attr structure to save some key data members.