diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2016-05-27 13:41:54 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2016-05-27 13:41:54 -0700 |
commit | e28e909c36bb5d6319953822d84df00fce7cbd18 (patch) | |
tree | a4aca971908a7a604c6fdd9a95360728f9f721b3 /tools | |
parent | dc03c0f9d12d85286d5e3623aa96d5c2a271b8e6 (diff) | |
parent | fabc712866435660f7fa1070e1fabe29eba5bc4c (diff) | |
download | op-kernel-dev-e28e909c36bb5d6319953822d84df00fce7cbd18.zip op-kernel-dev-e28e909c36bb5d6319953822d84df00fce7cbd18.tar.gz |
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull second batch of KVM updates from Radim Krčmář:
"General:
- move kvm_stat tool from QEMU repo into tools/kvm/kvm_stat (kvm_stat
had nothing to do with QEMU in the first place -- the tool only
interprets debugfs)
- expose per-vm statistics in debugfs and support them in kvm_stat
(KVM always collected per-vm statistics, but they were summarised
into global statistics)
x86:
- fix dynamic APICv (VMX was improperly configured and a guest could
access host's APIC MSRs, CVE-2016-4440)
- minor fixes
ARM changes from Christoffer Dall:
- new vgic reimplementation of our horribly broken legacy vgic
implementation. The two implementations will live side-by-side
(with the new being the configured default) for one kernel release
and then we'll remove the legacy one.
- fix for a non-critical issue with virtual abort injection to guests"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (70 commits)
tools: kvm_stat: Add comments
tools: kvm_stat: Introduce pid monitoring
KVM: Create debugfs dir and stat files for each VM
MAINTAINERS: Add kvm tools
tools: kvm_stat: Powerpc related fixes
tools: Add kvm_stat man page
tools: Add kvm_stat vm monitor script
kvm:vmx: more complete state update on APICv on/off
KVM: SVM: Add more SVM_EXIT_REASONS
KVM: Unify traced vector format
svm: bitwise vs logical op typo
KVM: arm/arm64: vgic-new: Synchronize changes to active state
KVM: arm/arm64: vgic-new: enable build
KVM: arm/arm64: vgic-new: implement mapped IRQ handling
KVM: arm/arm64: vgic-new: Wire up irqfd injection
KVM: arm/arm64: vgic-new: Add vgic_v2/v3_enable
KVM: arm/arm64: vgic-new: vgic_init: implement map_resources
KVM: arm/arm64: vgic-new: vgic_init: implement vgic_init
KVM: arm/arm64: vgic-new: vgic_init: implement vgic_create
KVM: arm/arm64: vgic-new: vgic_init: implement kvm_vgic_hyp_init
...
Diffstat (limited to 'tools')
-rw-r--r-- | tools/Makefile | 6 | ||||
-rw-r--r-- | tools/kvm/kvm_stat/Makefile | 41 | ||||
-rwxr-xr-x | tools/kvm/kvm_stat/kvm_stat | 1127 | ||||
-rw-r--r-- | tools/kvm/kvm_stat/kvm_stat.txt | 63 |
4 files changed, 1236 insertions, 1 deletions
diff --git a/tools/Makefile b/tools/Makefile index 6bf68fe..f10b64d8 100644 --- a/tools/Makefile +++ b/tools/Makefile @@ -16,6 +16,7 @@ help: @echo ' gpio - GPIO tools' @echo ' hv - tools used when in Hyper-V clients' @echo ' iio - IIO tools' + @echo ' kvm_stat - top-like utility for displaying kvm statistics' @echo ' lguest - a minimal 32-bit x86 hypervisor' @echo ' net - misc networking tools' @echo ' perf - Linux performance measurement and analysis tool' @@ -110,10 +111,13 @@ tmon_install: freefall_install: $(call descend,laptop/$(@:_install=),install) +kvm_stat_install: + $(call descend,kvm/$(@:_install=),install) + install: acpi_install cgroup_install cpupower_install hv_install firewire_install lguest_install \ perf_install selftests_install turbostat_install usb_install \ virtio_install vm_install net_install x86_energy_perf_policy_install \ - tmon_install freefall_install objtool_install + tmon_install freefall_install objtool_install kvm_stat_install acpi_clean: $(call descend,power/acpi,clean) diff --git a/tools/kvm/kvm_stat/Makefile b/tools/kvm/kvm_stat/Makefile new file mode 100644 index 0000000..5b1cba5 --- /dev/null +++ b/tools/kvm/kvm_stat/Makefile @@ -0,0 +1,41 @@ +include ../../scripts/Makefile.include +include ../../scripts/utilities.mak +BINDIR=usr/bin +MANDIR=usr/share/man +MAN1DIR=$(MANDIR)/man1 + +MAN1=kvm_stat.1 + +A2X=a2x +a2x_path := $(call get-executable,$(A2X)) + +all: man + +ifneq ($(findstring $(MAKEFLAGS),s),s) + ifneq ($(V),1) + QUIET_A2X = @echo ' A2X '$@; + endif +endif + +%.1: %.txt +ifeq ($(a2x_path),) + $(error "You need to install asciidoc for man pages") +else + $(QUIET_A2X)$(A2X) --doctype manpage --format manpage $< +endif + +clean: + rm -f $(MAN1) + +man: $(MAN1) + +install-man: man + install -d -m 755 $(INSTALL_ROOT)/$(MAN1DIR) + install -m 644 kvm_stat.1 $(INSTALL_ROOT)/$(MAN1DIR) + +install-tools: + install -d -m 755 $(INSTALL_ROOT)/$(BINDIR) + install -m 755 -p "kvm_stat" "$(INSTALL_ROOT)/$(BINDIR)/$(TARGET)" + +install: install-tools install-man +.PHONY: all clean man install-tools install-man install diff --git a/tools/kvm/kvm_stat/kvm_stat b/tools/kvm/kvm_stat/kvm_stat new file mode 100755 index 0000000..581278c --- /dev/null +++ b/tools/kvm/kvm_stat/kvm_stat @@ -0,0 +1,1127 @@ +#!/usr/bin/python +# +# top-like utility for displaying kvm statistics +# +# Copyright 2006-2008 Qumranet Technologies +# Copyright 2008-2011 Red Hat, Inc. +# +# Authors: +# Avi Kivity <avi@redhat.com> +# +# This work is licensed under the terms of the GNU GPL, version 2. See +# the COPYING file in the top-level directory. +"""The kvm_stat module outputs statistics about running KVM VMs + +Three different ways of output formatting are available: +- as a top-like text ui +- in a key -> value format +- in an all keys, all values format + +The data is sampled from the KVM's debugfs entries and its perf events. +""" + +import curses +import sys +import os +import time +import optparse +import ctypes +import fcntl +import resource +import struct +import re +from collections import defaultdict +from time import sleep + +VMX_EXIT_REASONS = { + 'EXCEPTION_NMI': 0, + 'EXTERNAL_INTERRUPT': 1, + 'TRIPLE_FAULT': 2, + 'PENDING_INTERRUPT': 7, + 'NMI_WINDOW': 8, + 'TASK_SWITCH': 9, + 'CPUID': 10, + 'HLT': 12, + 'INVLPG': 14, + 'RDPMC': 15, + 'RDTSC': 16, + 'VMCALL': 18, + 'VMCLEAR': 19, + 'VMLAUNCH': 20, + 'VMPTRLD': 21, + 'VMPTRST': 22, + 'VMREAD': 23, + 'VMRESUME': 24, + 'VMWRITE': 25, + 'VMOFF': 26, + 'VMON': 27, + 'CR_ACCESS': 28, + 'DR_ACCESS': 29, + 'IO_INSTRUCTION': 30, + 'MSR_READ': 31, + 'MSR_WRITE': 32, + 'INVALID_STATE': 33, + 'MWAIT_INSTRUCTION': 36, + 'MONITOR_INSTRUCTION': 39, + 'PAUSE_INSTRUCTION': 40, + 'MCE_DURING_VMENTRY': 41, + 'TPR_BELOW_THRESHOLD': 43, + 'APIC_ACCESS': 44, + 'EPT_VIOLATION': 48, + 'EPT_MISCONFIG': 49, + 'WBINVD': 54, + 'XSETBV': 55, + 'APIC_WRITE': 56, + 'INVPCID': 58, +} + +SVM_EXIT_REASONS = { + 'READ_CR0': 0x000, + 'READ_CR3': 0x003, + 'READ_CR4': 0x004, + 'READ_CR8': 0x008, + 'WRITE_CR0': 0x010, + 'WRITE_CR3': 0x013, + 'WRITE_CR4': 0x014, + 'WRITE_CR8': 0x018, + 'READ_DR0': 0x020, + 'READ_DR1': 0x021, + 'READ_DR2': 0x022, + 'READ_DR3': 0x023, + 'READ_DR4': 0x024, + 'READ_DR5': 0x025, + 'READ_DR6': 0x026, + 'READ_DR7': 0x027, + 'WRITE_DR0': 0x030, + 'WRITE_DR1': 0x031, + 'WRITE_DR2': 0x032, + 'WRITE_DR3': 0x033, + 'WRITE_DR4': 0x034, + 'WRITE_DR5': 0x035, + 'WRITE_DR6': 0x036, + 'WRITE_DR7': 0x037, + 'EXCP_BASE': 0x040, + 'INTR': 0x060, + 'NMI': 0x061, + 'SMI': 0x062, + 'INIT': 0x063, + 'VINTR': 0x064, + 'CR0_SEL_WRITE': 0x065, + 'IDTR_READ': 0x066, + 'GDTR_READ': 0x067, + 'LDTR_READ': 0x068, + 'TR_READ': 0x069, + 'IDTR_WRITE': 0x06a, + 'GDTR_WRITE': 0x06b, + 'LDTR_WRITE': 0x06c, + 'TR_WRITE': 0x06d, + 'RDTSC': 0x06e, + 'RDPMC': 0x06f, + 'PUSHF': 0x070, + 'POPF': 0x071, + 'CPUID': 0x072, + 'RSM': 0x073, + 'IRET': 0x074, + 'SWINT': 0x075, + 'INVD': 0x076, + 'PAUSE': 0x077, + 'HLT': 0x078, + 'INVLPG': 0x079, + 'INVLPGA': 0x07a, + 'IOIO': 0x07b, + 'MSR': 0x07c, + 'TASK_SWITCH': 0x07d, + 'FERR_FREEZE': 0x07e, + 'SHUTDOWN': 0x07f, + 'VMRUN': 0x080, + 'VMMCALL': 0x081, + 'VMLOAD': 0x082, + 'VMSAVE': 0x083, + 'STGI': 0x084, + 'CLGI': 0x085, + 'SKINIT': 0x086, + 'RDTSCP': 0x087, + 'ICEBP': 0x088, + 'WBINVD': 0x089, + 'MONITOR': 0x08a, + 'MWAIT': 0x08b, + 'MWAIT_COND': 0x08c, + 'XSETBV': 0x08d, + 'NPF': 0x400, +} + +# EC definition of HSR (from arch/arm64/include/asm/kvm_arm.h) +AARCH64_EXIT_REASONS = { + 'UNKNOWN': 0x00, + 'WFI': 0x01, + 'CP15_32': 0x03, + 'CP15_64': 0x04, + 'CP14_MR': 0x05, + 'CP14_LS': 0x06, + 'FP_ASIMD': 0x07, + 'CP10_ID': 0x08, + 'CP14_64': 0x0C, + 'ILL_ISS': 0x0E, + 'SVC32': 0x11, + 'HVC32': 0x12, + 'SMC32': 0x13, + 'SVC64': 0x15, + 'HVC64': 0x16, + 'SMC64': 0x17, + 'SYS64': 0x18, + 'IABT': 0x20, + 'IABT_HYP': 0x21, + 'PC_ALIGN': 0x22, + 'DABT': 0x24, + 'DABT_HYP': 0x25, + 'SP_ALIGN': 0x26, + 'FP_EXC32': 0x28, + 'FP_EXC64': 0x2C, + 'SERROR': 0x2F, + 'BREAKPT': 0x30, + 'BREAKPT_HYP': 0x31, + 'SOFTSTP': 0x32, + 'SOFTSTP_HYP': 0x33, + 'WATCHPT': 0x34, + 'WATCHPT_HYP': 0x35, + 'BKPT32': 0x38, + 'VECTOR32': 0x3A, + 'BRK64': 0x3C, +} + +# From include/uapi/linux/kvm.h, KVM_EXIT_xxx +USERSPACE_EXIT_REASONS = { + 'UNKNOWN': 0, + 'EXCEPTION': 1, + 'IO': 2, + 'HYPERCALL': 3, + 'DEBUG': 4, + 'HLT': 5, + 'MMIO': 6, + 'IRQ_WINDOW_OPEN': 7, + 'SHUTDOWN': 8, + 'FAIL_ENTRY': 9, + 'INTR': 10, + 'SET_TPR': 11, + 'TPR_ACCESS': 12, + 'S390_SIEIC': 13, + 'S390_RESET': 14, + 'DCR': 15, + 'NMI': 16, + 'INTERNAL_ERROR': 17, + 'OSI': 18, + 'PAPR_HCALL': 19, + 'S390_UCONTROL': 20, + 'WATCHDOG': 21, + 'S390_TSCH': 22, + 'EPR': 23, + 'SYSTEM_EVENT': 24, +} + +IOCTL_NUMBERS = { + 'SET_FILTER': 0x40082406, + 'ENABLE': 0x00002400, + 'DISABLE': 0x00002401, + 'RESET': 0x00002403, +} + +class Arch(object): + """Encapsulates global architecture specific data. + + Contains the performance event open syscall and ioctl numbers, as + well as the VM exit reasons for the architecture it runs on. + + """ + @staticmethod + def get_arch(): + machine = os.uname()[4] + + if machine.startswith('ppc'): + return ArchPPC() + elif machine.startswith('aarch64'): + return ArchA64() + elif machine.startswith('s390'): + return ArchS390() + else: + # X86_64 + for line in open('/proc/cpuinfo'): + if not line.startswith('flags'): + continue + + flags = line.split() + if 'vmx' in flags: + return ArchX86(VMX_EXIT_REASONS) + if 'svm' in flags: + return ArchX86(SVM_EXIT_REASONS) + return + +class ArchX86(Arch): + def __init__(self, exit_reasons): + self.sc_perf_evt_open = 298 + self.ioctl_numbers = IOCTL_NUMBERS + self.exit_reasons = exit_reasons + +class ArchPPC(Arch): + def __init__(self): + self.sc_perf_evt_open = 319 + self.ioctl_numbers = IOCTL_NUMBERS + self.ioctl_numbers['ENABLE'] = 0x20002400 + self.ioctl_numbers['DISABLE'] = 0x20002401 + self.ioctl_numbers['RESET'] = 0x20002403 + + # PPC comes in 32 and 64 bit and some generated ioctl + # numbers depend on the wordsize. + char_ptr_size = ctypes.sizeof(ctypes.c_char_p) + self.ioctl_numbers['SET_FILTER'] = 0x80002406 | char_ptr_size << 16 + self.exit_reasons = {} + +class ArchA64(Arch): + def __init__(self): + self.sc_perf_evt_open = 241 + self.ioctl_numbers = IOCTL_NUMBERS + self.exit_reasons = AARCH64_EXIT_REASONS + +class ArchS390(Arch): + def __init__(self): + self.sc_perf_evt_open = 331 + self.ioctl_numbers = IOCTL_NUMBERS + self.exit_reasons = None + +ARCH = Arch.get_arch() + + +def walkdir(path): + """Returns os.walk() data for specified directory. + + As it is only a wrapper it returns the same 3-tuple of (dirpath, + dirnames, filenames). + """ + return next(os.walk(path)) + + +def parse_int_list(list_string): + """Returns an int list from a string of comma separated integers and + integer ranges.""" + integers = [] + members = list_string.split(',') + + for member in members: + if '-' not in member: + integers.append(int(member)) + else: + int_range = member.split('-') + integers.extend(range(int(int_range[0]), + int(int_range[1]) + 1)) + + return integers + + +def get_online_cpus(): + """Returns a list of cpu id integers.""" + with open('/sys/devices/system/cpu/online') as cpu_list: + cpu_string = cpu_list.readline() + return parse_int_list(cpu_string) + + +def get_filters(): + """Returns a dict of trace events, their filter ids and + the values that can be filtered. + + Trace events can be filtered for special values by setting a + filter string via an ioctl. The string normally has the format + identifier==value. For each filter a new event will be created, to + be able to distinguish the events. + + """ + filters = {} + filters['kvm_userspace_exit'] = ('reason', USERSPACE_EXIT_REASONS) + if ARCH.exit_reasons: + filters['kvm_exit'] = ('exit_reason', ARCH.exit_reasons) + return filters + +libc = ctypes.CDLL('libc.so.6', use_errno=True) +syscall = libc.syscall + +class perf_event_attr(ctypes.Structure): + """Struct that holds the necessary data to set up a trace event. + + For an extensive explanation see perf_event_open(2) and + include/uapi/linux/perf_event.h, struct perf_event_attr + + All fields that are not initialized in the constructor are 0. + + """ + _fields_ = [('type', ctypes.c_uint32), + ('size', ctypes.c_uint32), + ('config', ctypes.c_uint64), + ('sample_freq', ctypes.c_uint64), + ('sample_type', ctypes.c_uint64), + ('read_format', ctypes.c_uint64), + ('flags', ctypes.c_uint64), + ('wakeup_events', ctypes.c_uint32), + ('bp_type', ctypes.c_uint32), + ('bp_addr', ctypes.c_uint64), + ('bp_len', ctypes.c_uint64), + ] + + def __init__(self): + super(self.__class__, self).__init__() + self.type = PERF_TYPE_TRACEPOINT + self.size = ctypes.sizeof(self) + self.read_format = PERF_FORMAT_GROUP + +def perf_event_open(attr, pid, cpu, group_fd, flags): + """Wrapper for the sys_perf_evt_open() syscall. + + Used to set up performance events, returns a file descriptor or -1 + on error. + + Attributes are: + - syscall number + - struct perf_event_attr * + - pid or -1 to monitor all pids + - cpu number or -1 to monitor all cpus + - The file descriptor of the group leader or -1 to create a group. + - flags + + """ + return syscall(ARCH.sc_perf_evt_open, ctypes.pointer(attr), + ctypes.c_int(pid), ctypes.c_int(cpu), + ctypes.c_int(group_fd), ctypes.c_long(flags)) + +PERF_TYPE_TRACEPOINT = 2 +PERF_FORMAT_GROUP = 1 << 3 + +PATH_DEBUGFS_TRACING = '/sys/kernel/debug/tracing' +PATH_DEBUGFS_KVM = '/sys/kernel/debug/kvm' + +class Group(object): + """Represents a perf event group.""" + + def __init__(self): + self.events = [] + + def add_event(self, event): + self.events.append(event) + + def read(self): + """Returns a dict with 'event name: value' for all events in the + group. + + Values are read by reading from the file descriptor of the + event that is the group leader. See perf_event_open(2) for + details. + + Read format for the used event configuration is: + struct read_format { + u64 nr; /* The number of events */ + struct { + u64 value; /* The value of the event */ + } values[nr]; + }; + + """ + length = 8 * (1 + len(self.events)) + read_format = 'xxxxxxxx' + 'Q' * len(self.events) + return dict(zip([event.name for event in self.events], + struct.unpack(read_format, + os.read(self.events[0].fd, length)))) + +class Event(object): + """Represents a performance event and manages its life cycle.""" + def __init__(self, name, group, trace_cpu, trace_pid, trace_point, + trace_filter, trace_set='kvm'): + self.name = name + self.fd = None + self.setup_event(group, trace_cpu, trace_pid, trace_point, + trace_filter, trace_set) + + def __del__(self): + """Closes the event's file descriptor. + + As no python file object was created for the file descriptor, + python will not reference count the descriptor and will not + close it itself automatically, so we do it. + + """ + if self.fd: + os.close(self.fd) + + def setup_event_attribute(self, trace_set, trace_point): + """Returns an initialized ctype perf_event_attr struct.""" + + id_path = os.path.join(PATH_DEBUGFS_TRACING, 'events', trace_set, + trace_point, 'id') + + event_attr = perf_event_attr() + event_attr.config = int(open(id_path).read()) + return event_attr + + def setup_event(self, group, trace_cpu, trace_pid, trace_point, + trace_filter, trace_set): + """Sets up the perf event in Linux. + + Issues the syscall to register the event in the kernel and + then sets the optional filter. + + """ + + event_attr = self.setup_event_attribute(trace_set, trace_point) + + # First event will be group leader. + group_leader = -1 + + # All others have to pass the leader's descriptor instead. + if group.events: + group_leader = group.events[0].fd + + fd = perf_event_open(event_attr, trace_pid, + trace_cpu, group_leader, 0) + if fd == -1: + err = ctypes.get_errno() + raise OSError(err, os.strerror(err), + 'while calling sys_perf_event_open().') + + if trace_filter: + fcntl.ioctl(fd, ARCH.ioctl_numbers['SET_FILTER'], + trace_filter) + + self.fd = fd + + def enable(self): + """Enables the trace event in the kernel. + + Enabling the group leader makes reading counters from it and the + events under it possible. + + """ + fcntl.ioctl(self.fd, ARCH.ioctl_numbers['ENABLE'], 0) + + def disable(self): + """Disables the trace event in the kernel. + + Disabling the group leader makes reading all counters under it + impossible. + + """ + fcntl.ioctl(self.fd, ARCH.ioctl_numbers['DISABLE'], 0) + + def reset(self): + """Resets the count of the trace event in the kernel.""" + fcntl.ioctl(self.fd, ARCH.ioctl_numbers['RESET'], 0) + +class TracepointProvider(object): + """Data provider for the stats class. + + Manages the events/groups from which it acquires its data. + + """ + def __init__(self): + self.group_leaders = [] + self.filters = get_filters() + self._fields = self.get_available_fields() + self._pid = 0 + + def get_available_fields(self): + """Returns a list of available event's of format 'event name(filter + name)'. + + All available events have directories under + /sys/kernel/debug/tracing/events/ which export information + about the specific event. Therefore, listing the dirs gives us + a list of all available events. + + Some events like the vm exit reasons can be filtered for + specific values. To take account for that, the routine below + creates special fields with the following format: + event name(filter name) + + """ + path = os.path.join(PATH_DEBUGFS_TRACING, 'events', 'kvm') + fields = walkdir(path)[1] + extra = [] + for field in fields: + if field in self.filters: + filter_name_, filter_dicts = self.filters[field] + for name in filter_dicts: + extra.append(field + '(' + name + ')') + fields += extra + return fields + + def setup_traces(self): + """Creates all event and group objects needed to be able to retrieve + data.""" + if self._pid > 0: + # Fetch list of all threads of the monitored pid, as qemu + # starts a thread for each vcpu. + path = os.path.join('/proc', str(self._pid), 'task') + groupids = walkdir(path)[1] + else: + groupids = get_online_cpus() + + # The constant is needed as a buffer for python libs, std + # streams and other files that the script opens. + newlim = len(groupids) * len(self._fields) + 50 + try: + softlim_, hardlim = resource.getrlimit(resource.RLIMIT_NOFILE) + + if hardlim < newlim: + # Now we need CAP_SYS_RESOURCE, to increase the hard limit. + resource.setrlimit(resource.RLIMIT_NOFILE, (newlim, newlim)) + else: + # Raising the soft limit is sufficient. + resource.setrlimit(resource.RLIMIT_NOFILE, (newlim, hardlim)) + + except ValueError: + sys.exit("NOFILE rlimit could not be raised to {0}".format(newlim)) + + for groupid in groupids: + group = Group() + for name in self._fields: + tracepoint = name + tracefilter = None + match = re.match(r'(.*)\((.*)\)', name) + if match: + tracepoint, sub = match.groups() + tracefilter = ('%s==%d\0' % + (self.filters[tracepoint][0], + self.filters[tracepoint][1][sub])) + + # From perf_event_open(2): + # pid > 0 and cpu == -1 + # This measures the specified process/thread on any CPU. + # + # pid == -1 and cpu >= 0 + # This measures all processes/threads on the specified CPU. + trace_cpu = groupid if self._pid == 0 else -1 + trace_pid = int(groupid) if self._pid != 0 else -1 + + group.add_event(Event(name=name, + group=group, + trace_cpu=trace_cpu, + trace_pid=trace_pid, + trace_point=tracepoint, + trace_filter=tracefilter)) + + self.group_leaders.append(group) + + def available_fields(self): + return self.get_available_fields() + + @property + def fields(self): + return self._fields + + @fields.setter + def fields(self, fields): + """Enables/disables the (un)wanted events""" + self._fields = fields + for group in self.group_leaders: + for index, event in enumerate(group.events): + if event.name in fields: + event.reset() + event.enable() + else: + # Do not disable the group leader. + # It would disable all of its events. + if index != 0: + event.disable() + + @property + def pid(self): + return self._pid + + @pid.setter + def pid(self, pid): + """Changes the monitored pid by setting new traces.""" + self._pid = pid + # The garbage collector will get rid of all Event/Group + # objects and open files after removing the references. + self.group_leaders = [] + self.setup_traces() + self.fields = self._fields + + def read(self): + """Returns 'event name: current value' for all enabled events.""" + ret = defaultdict(int) + for group in self.group_leaders: + for name, val in group.read().iteritems(): + if name in self._fields: + ret[name] += val + return ret + +class DebugfsProvider(object): + """Provides data from the files that KVM creates in the kvm debugfs + folder.""" + def __init__(self): + self._fields = self.get_available_fields() + self._pid = 0 + self.do_read = True + + def get_available_fields(self): + """"Returns a list of available fields. + + The fields are all available KVM debugfs files + + """ + return walkdir(PATH_DEBUGFS_KVM)[2] + + @property + def fields(self): + return self._fields + + @fields.setter + def fields(self, fields): + self._fields = fields + + @property + def pid(self): + return self._pid + + @pid.setter + def pid(self, pid): + if pid != 0: + self._pid = pid + + vms = walkdir(PATH_DEBUGFS_KVM)[1] + if len(vms) == 0: + self.do_read = False + + self.paths = filter(lambda x: "{}-".format(pid) in x, vms) + + else: + self.paths = [''] + self.do_read = True + + def read(self): + """Returns a dict with format:'file name / field -> current value'.""" + results = {} + + # If no debugfs filtering support is available, then don't read. + if not self.do_read: + return results + + for path in self.paths: + for field in self._fields: + results[field] = results.get(field, 0) \ + + self.read_field(field, path) + + return results + + def read_field(self, field, path): + """Returns the value of a single field from a specific VM.""" + try: + return int(open(os.path.join(PATH_DEBUGFS_KVM, + path, + field)) + .read()) + except IOError: + return 0 + +class Stats(object): + """Manages the data providers and the data they provide. + + It is used to set filters on the provider's data and collect all + provider data. + + """ + def __init__(self, providers, pid, fields=None): + self.providers = providers + self._pid_filter = pid + self._fields_filter = fields + self.values = {} + self.update_provider_pid() + self.update_provider_filters() + + def update_provider_filters(self): + """Propagates fields filters to providers.""" + def wanted(key): + if not self._fields_filter: + return True + return re.match(self._fields_filter, key) is not None + + # As we reset the counters when updating the fields we can + # also clear the cache of old values. + self.values = {} + for provider in self.providers: + provider_fields = [key for key in provider.get_available_fields() + if wanted(key)] + provider.fields = provider_fields + + def update_provider_pid(self): + """Propagates pid filters to providers.""" + for provider in self.providers: + provider.pid = self._pid_filter + + @property + def fields_filter(self): + return self._fields_filter + + @fields_filter.setter + def fields_filter(self, fields_filter): + self._fields_filter = fields_filter + self.update_provider_filters() + + @property + def pid_filter(self): + return self._pid_filter + + @pid_filter.setter + def pid_filter(self, pid): + self._pid_filter = pid + self.values = {} + self.update_provider_pid() + + def get(self): + """Returns a dict with field -> (value, delta to last value) of all + provider data.""" + for provider in self.providers: + new = provider.read() + for key in provider.fields: + oldval = self.values.get(key, (0, 0)) + newval = new.get(key, 0) + newdelta = None + if oldval is not None: + newdelta = newval - oldval[0] + self.values[key] = (newval, newdelta) + return self.values + +LABEL_WIDTH = 40 +NUMBER_WIDTH = 10 + +class Tui(object): + """Instruments curses to draw a nice text ui.""" + def __init__(self, stats): + self.stats = stats + self.screen = None + self.drilldown = False + self.update_drilldown() + + def __enter__(self): + """Initialises curses for later use. Based on curses.wrapper + implementation from the Python standard library.""" + self.screen = curses.initscr() + curses.noecho() + curses.cbreak() + + # The try/catch works around a minor bit of + # over-conscientiousness in the curses module, the error + # return from C start_color() is ignorable. + try: + curses.start_color() + except: + pass + + curses.use_default_colors() + return self + + def __exit__(self, *exception): + """Resets the terminal to its normal state. Based on curses.wrappre + implementation from the Python standard library.""" + if self.screen: + self.screen.keypad(0) + curses.echo() + curses.nocbreak() + curses.endwin() + + def update_drilldown(self): + """Sets or removes a filter that only allows fields without braces.""" + if not self.stats.fields_filter: + self.stats.fields_filter = r'^[^\(]*$' + + elif self.stats.fields_filter == r'^[^\(]*$': + self.stats.fields_filter = None + + def update_pid(self, pid): + """Propagates pid selection to stats object.""" + self.stats.pid_filter = pid + + def refresh(self, sleeptime): + """Refreshes on-screen data.""" + self.screen.erase() + if self.stats.pid_filter > 0: + self.screen.addstr(0, 0, 'kvm statistics - pid {0}' + .format(self.stats.pid_filter), + curses.A_BOLD) + else: + self.screen.addstr(0, 0, 'kvm statistics - summary', curses.A_BOLD) + self.screen.addstr(2, 1, 'Event') + self.screen.addstr(2, 1 + LABEL_WIDTH + NUMBER_WIDTH - + len('Total'), 'Total') + self.screen.addstr(2, 1 + LABEL_WIDTH + NUMBER_WIDTH + 8 - + len('Current'), 'Current') + row = 3 + stats = self.stats.get() + def sortkey(x): + if stats[x][1]: + return (-stats[x][1], -stats[x][0]) + else: + return (0, -stats[x][0]) + for key in sorted(stats.keys(), key=sortkey): + + if row >= self.screen.getmaxyx()[0]: + break + values = stats[key] + if not values[0] and not values[1]: + break + col = 1 + self.screen.addstr(row, col, key) + col += LABEL_WIDTH + self.screen.addstr(row, col, '%10d' % (values[0],)) + col += NUMBER_WIDTH + if values[1] is not None: + self.screen.addstr(row, col, '%8d' % (values[1] / sleeptime,)) + row += 1 + self.screen.refresh() + + def show_filter_selection(self): + """Draws filter selection mask. + + Asks for a valid regex and sets the fields filter accordingly. + + """ + while True: + self.screen.erase() + self.screen.addstr(0, 0, + "Show statistics for events matching a regex.", + curses.A_BOLD) + self.screen.addstr(2, 0, + "Current regex: {0}" + .format(self.stats.fields_filter)) + self.screen.addstr(3, 0, "New regex: ") + curses.echo() + regex = self.screen.getstr() + curses.noecho() + if len(regex) == 0: + return + try: + re.compile(regex) + self.stats.fields_filter = regex + return + except re.error: + continue + + def show_vm_selection(self): + """Draws PID selection mask. + + Asks for a pid until a valid pid or 0 has been entered. + + """ + while True: + self.screen.erase() + self.screen.addstr(0, 0, + 'Show statistics for specific pid.', + curses.A_BOLD) + self.screen.addstr(1, 0, + 'This might limit the shown data to the trace ' + 'statistics.') + + curses.echo() + self.screen.addstr(3, 0, "Pid [0 or pid]: ") + pid = self.screen.getstr() + curses.noecho() + + try: + pid = int(pid) + + if pid == 0: + self.update_pid(pid) + break + else: + if not os.path.isdir(os.path.join('/proc/', str(pid))): + continue + else: + self.update_pid(pid) + break + + except ValueError: + continue + + def show_stats(self): + """Refreshes the screen and processes user input.""" + sleeptime = 0.25 + while True: + self.refresh(sleeptime) + curses.halfdelay(int(sleeptime * 10)) + sleeptime = 3 + try: + char = self.screen.getkey() + if char == 'x': + self.drilldown = not self.drilldown + self.update_drilldown() + if char == 'q': + break + if char == 'f': + self.show_filter_selection() + if char == 'p': + self.show_vm_selection() + except KeyboardInterrupt: + break + except curses.error: + continue + +def batch(stats): + """Prints statistics in a key, value format.""" + s = stats.get() + time.sleep(1) + s = stats.get() + for key in sorted(s.keys()): + values = s[key] + print '%-42s%10d%10d' % (key, values[0], values[1]) + +def log(stats): + """Prints statistics as reiterating key block, multiple value blocks.""" + keys = sorted(stats.get().iterkeys()) + def banner(): + for k in keys: + print '%s' % k, + print + def statline(): + s = stats.get() + for k in keys: + print ' %9d' % s[k][1], + print + line = 0 + banner_repeat = 20 + while True: + time.sleep(1) + if line % banner_repeat == 0: + banner() + statline() + line += 1 + +def get_options(): + """Returns processed program arguments.""" + description_text = """ +This script displays various statistics about VMs running under KVM. +The statistics are gathered from the KVM debugfs entries and / or the +currently available perf traces. + +The monitoring takes additional cpu cycles and might affect the VM's +performance. + +Requirements: +- Access to: + /sys/kernel/debug/kvm + /sys/kernel/debug/trace/events/* + /proc/pid/task +- /proc/sys/kernel/perf_event_paranoid < 1 if user has no + CAP_SYS_ADMIN and perf events are used. +- CAP_SYS_RESOURCE if the hard limit is not high enough to allow + the large number of files that are possibly opened. +""" + + class PlainHelpFormatter(optparse.IndentedHelpFormatter): + def format_description(self, description): + if description: + return description + "\n" + else: + return "" + + optparser = optparse.OptionParser(description=description_text, + formatter=PlainHelpFormatter()) + optparser.add_option('-1', '--once', '--batch', + action='store_true', + default=False, + dest='once', + help='run in batch mode for one second', + ) + optparser.add_option('-l', '--log', + action='store_true', + default=False, + dest='log', + help='run in logging mode (like vmstat)', + ) + optparser.add_option('-t', '--tracepoints', + action='store_true', + default=False, + dest='tracepoints', + help='retrieve statistics from tracepoints', + ) + optparser.add_option('-d', '--debugfs', + action='store_true', + default=False, + dest='debugfs', + help='retrieve statistics from debugfs', + ) + optparser.add_option('-f', '--fields', + action='store', + default=None, + dest='fields', + help='fields to display (regex)', + ) + optparser.add_option('-p', '--pid', + action='store', + default=0, + type=int, + dest='pid', + help='restrict statistics to pid', + ) + (options, _) = optparser.parse_args(sys.argv) + return options + +def get_providers(options): + """Returns a list of data providers depending on the passed options.""" + providers = [] + + if options.tracepoints: + providers.append(TracepointProvider()) + if options.debugfs: + providers.append(DebugfsProvider()) + if len(providers) == 0: + providers.append(TracepointProvider()) + + return providers + +def check_access(options): + """Exits if the current user can't access all needed directories.""" + if not os.path.exists('/sys/kernel/debug'): + sys.stderr.write('Please enable CONFIG_DEBUG_FS in your kernel.') + sys.exit(1) + + if not os.path.exists(PATH_DEBUGFS_KVM): + sys.stderr.write("Please make sure, that debugfs is mounted and " + "readable by the current user:\n" + "('mount -t debugfs debugfs /sys/kernel/debug')\n" + "Also ensure, that the kvm modules are loaded.\n") + sys.exit(1) + + if not os.path.exists(PATH_DEBUGFS_TRACING) and (options.tracepoints + or not options.debugfs): + sys.stderr.write("Please enable CONFIG_TRACING in your kernel " + "when using the option -t (default).\n" + "If it is enabled, make {0} readable by the " + "current user.\n" + .format(PATH_DEBUGFS_TRACING)) + if options.tracepoints: + sys.exit(1) + + sys.stderr.write("Falling back to debugfs statistics!\n") + options.debugfs = True + sleep(5) + + return options + +def main(): + options = get_options() + options = check_access(options) + + if (options.pid > 0 and + not os.path.isdir(os.path.join('/proc/', + str(options.pid)))): + sys.stderr.write('Did you use a (unsupported) tid instead of a pid?\n') + sys.exit('Specified pid does not exist.') + + providers = get_providers(options) + stats = Stats(providers, options.pid, fields=options.fields) + + if options.log: + log(stats) + elif not options.once: + with Tui(stats) as tui: + tui.show_stats() + else: + batch(stats) + +if __name__ == "__main__": + main() diff --git a/tools/kvm/kvm_stat/kvm_stat.txt b/tools/kvm/kvm_stat/kvm_stat.txt new file mode 100644 index 0000000..b92a153 --- /dev/null +++ b/tools/kvm/kvm_stat/kvm_stat.txt @@ -0,0 +1,63 @@ +kvm_stat(1) +=========== + +NAME +---- +kvm_stat - Report KVM kernel module event counters + +SYNOPSIS +-------- +[verse] +'kvm_stat' [OPTION]... + +DESCRIPTION +----------- +kvm_stat prints counts of KVM kernel module trace events. These events signify +state transitions such as guest mode entry and exit. + +This tool is useful for observing guest behavior from the host perspective. +Often conclusions about performance or buggy behavior can be drawn from the +output. + +The set of KVM kernel module trace events may be specific to the kernel version +or architecture. It is best to check the KVM kernel module source code for the +meaning of events. + +OPTIONS +------- +-1:: +--once:: +--batch:: + run in batch mode for one second + +-l:: +--log:: + run in logging mode (like vmstat) + +-t:: +--tracepoints:: + retrieve statistics from tracepoints + +-d:: +--debugfs:: + retrieve statistics from debugfs + +-p<pid>:: +--pid=<pid>:: + limit statistics to one virtual machine (pid) + +-f<fields>:: +--fields=<fields>:: + fields to display (regex) + +-h:: +--help:: + show help message + +SEE ALSO +-------- +'perf'(1), 'trace-cmd'(1) + +AUTHOR +------ +Stefan Hajnoczi <stefanha@redhat.com> |