KVM: stats: Add fd-based API to read binary stats data

This commit defines the API for userspace and prepare the common functionalities to support per VM/VCPU binary stats data readings. The KVM stats now is only accessible by debugfs, which has some shortcomings this change series are supposed to fix: 1. The current debugfs stats solution in KVM could be disabled when kernel Lockdown mode is enabled, which is a potential rick for production. 2. The current debugfs stats solution in KVM is organized as "one stats per file", it is good for debugging, but not efficient for production. 3. The stats read/clear in current debugfs solution in KVM are protected by the global kvm_lock. Besides that, there are some other benefits with this change: 1. All KVM VM/VCPU stats can be read out in a bulk by one copy to userspace. 2. A schema is used to describe KVM statistics. From userspace's perspective, the KVM statistics are self-describing. 3. With the fd-based solution, a separate telemetry would be able to read KVM stats in a less privileged environment. 4. After the initial setup by reading in stats descriptors, a telemetry only needs to read the stats data itself, no more parsing or setup is needed.

Jing Zhang committed 5 years ago

Commit cb082bf1 parent 0193cc99 files · +307 −7Browse files

Unified · Split

9 files

arch/arm64/kvm/Makefile +1 −1
arch/mips/kvm/Makefile +1 −1
arch/powerpc/kvm/Makefile +1 −1
arch/s390/kvm/Makefile +2 −1
arch/x86/kvm/Makefile +1 −1
include/linux/kvm_host.h +80 −2
include/linux/kvm_types.h +2 −0
include/uapi/linux/kvm.h +73 −0
virt/kvm/binary_stats.c +146 −0

arch/arm64/kvm/Makefile+1−1

@@ -11,7 +11,7 @@ obj-$(CONFIG_KVM) += kvm.o · 10 unmodified lines
11	11	obj-$(CONFIG_KVM) += hyp/
12	12
13	13	kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \
14		− $(KVM)/vfio.o $(KVM)/irqchip.o \
	14	+ $(KVM)/vfio.o $(KVM)/irqchip.o $(KVM)/binary_stats.o \
15	15	arm.o mmu.o mmio.o psci.o perf.o hypercalls.o pvtime.o \
16	16	inject_fault.o va_layout.o handle_exit.o \
17	17	guest.o debug.o reset.o sys_regs.o \

arch/mips/kvm/Makefile+1−1

@@ -2,7 +2,7 @@ · 1 unmodified line
2	2	# Makefile for KVM support for MIPS
3	3	#
4	4
5		−common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o)
	5	+common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o binary_stats.o)
6	6
7	7	EXTRA_CFLAGS += -Ivirt/kvm -Iarch/mips/kvm
8	8

arch/powerpc/kvm/Makefile+1−1

@@ -6,7 +6,7 @@ · 5 unmodified lines
6	6	ccflags-y := -Ivirt/kvm -Iarch/powerpc/kvm
7	7	KVM := ../../../virt/kvm
8	8
9		−common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o
	9	+common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/binary_stats.o
10	10	common-objs-$(CONFIG_KVM_VFIO) += $(KVM)/vfio.o
11	11	common-objs-$(CONFIG_KVM_MMIO) += $(KVM)/coalesced_mmio.o
12	12

arch/s390/kvm/Makefile+2−1

@@ -4,7 +4,8 @@ · 3 unmodified lines
4	4	# Copyright IBM Corp. 2008
5	5
6	6	KVM := ../../../virt/kvm
7		−common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/async_pf.o $(KVM)/irqchip.o $(KVM)/vfio.o
	7	+common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/async_pf.o \
	8	+ $(KVM)/irqchip.o $(KVM)/vfio.o $(KVM)/binary_stats.o
8	9
9	10	ccflags-y := -Ivirt/kvm -Iarch/s390/kvm
10	11

arch/x86/kvm/Makefile+1−1

@@ -11,7 +11,7 @@ KVM := ../../../virt/kvm · 10 unmodified lines
11	11
12	12	kvm-y += $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o \
13	13	$(KVM)/eventfd.o $(KVM)/irqchip.o $(KVM)/vfio.o \
14		− $(KVM)/dirty_ring.o
	14	+ $(KVM)/dirty_ring.o $(KVM)/binary_stats.o
15	15	kvm-$(CONFIG_KVM_ASYNC_PF) += $(KVM)/async_pf.o
16	16
17	17	kvm-y += x86.o emulate.o i8259.o irq.o lapic.o \

include/linux/kvm_host.h+80−2

@@ -1272,16 +1272,94 @@ struct kvm_stats_debugfs_item { · 1271 unmodified lines
1272	1272	int mode;
1273	1273	};
1274	1274
	1275	+struct _kvm_stats_desc {
	1276	+ struct kvm_stats_desc desc;
	1277	+ char name[KVM_STATS_NAME_SIZE];
	1278	+};
	1279	+
1275	1280	#define KVM_DBGFS_GET_MODE(dbgfs_item) \
1276	1281	((dbgfs_item)->mode ? (dbgfs_item)->mode : 0644)
1277	1282
1278		−#define VM_STAT(n, x, ...) \
	1283	+#define VM_STAT(n, x, ...) \
1279	1284	{ n, offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__ }
1280		−#define VCPU_STAT(n, x, ...) \
	1285	+#define VCPU_STAT(n, x, ...) \
1281	1286	{ n, offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__ }
1282	1287
	1288	+#define STATS_DESC_COMMON(type, unit, base, exp) \
	1289	+ .flags = type \| unit \| base \| \
	1290	+ BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) \| \
	1291	+ BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) \| \
	1292	+ BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
	1293	+ .exponent = exp, \
	1294	+ .size = 1
	1295	+
	1296	+#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp) \
	1297	+ { \
	1298	+ { \
	1299	+ STATS_DESC_COMMON(type, unit, base, exp), \
	1300	+ .offset = offsetof(struct kvm_vm_stat, generic.stat) \
	1301	+ }, \
	1302	+ .name = #stat, \
	1303	+ }
	1304	+#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp) \
	1305	+ { \
	1306	+ { \
	1307	+ STATS_DESC_COMMON(type, unit, base, exp), \
	1308	+ .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
	1309	+ }, \
	1310	+ .name = #stat, \
	1311	+ }
	1312	+#define VM_STATS_DESC(stat, type, unit, base, exp) \
	1313	+ { \
	1314	+ { \
	1315	+ STATS_DESC_COMMON(type, unit, base, exp), \
	1316	+ .offset = offsetof(struct kvm_vm_stat, stat) \
	1317	+ }, \
	1318	+ .name = #stat, \
	1319	+ }
	1320	+#define VCPU_STATS_DESC(stat, type, unit, base, exp) \
	1321	+ { \
	1322	+ { \
	1323	+ STATS_DESC_COMMON(type, unit, base, exp), \
	1324	+ .offset = offsetof(struct kvm_vcpu_stat, stat) \
	1325	+ }, \
	1326	+ .name = #stat, \
	1327	+ }
	1328	+/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
	1329	+#define STATS_DESC(SCOPE, stat, type, unit, base, exp) \
	1330	+ SCOPE##_STATS_DESC(stat, type, unit, base, exp)
	1331	+
	1332	+#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
	1333	+ STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, unit, base, exponent)
	1334	+#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
	1335	+ STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, unit, base, exponent)
	1336	+#define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
	1337	+ STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, unit, base, exponent)
	1338	+
	1339	+/* Cumulative counter, read/write */
	1340	+#define STATS_DESC_COUNTER(SCOPE, name) \
	1341	+ STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
	1342	+ KVM_STATS_BASE_POW10, 0)
	1343	+/* Instantaneous counter, read only */
	1344	+#define STATS_DESC_ICOUNTER(SCOPE, name) \
	1345	+ STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
	1346	+ KVM_STATS_BASE_POW10, 0)
	1347	+/* Peak counter, read/write */
	1348	+#define STATS_DESC_PCOUNTER(SCOPE, name) \
	1349	+ STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
	1350	+ KVM_STATS_BASE_POW10, 0)
	1351	+
	1352	+/* Cumulative time in nanosecond */
	1353	+#define STATS_DESC_TIME_NSEC(SCOPE, name) \
	1354	+ STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
	1355	+ KVM_STATS_BASE_POW10, -9)
	1356	+
1283	1357	extern struct kvm_stats_debugfs_item debugfs_entries[];
1284	1358	extern struct dentry *kvm_debugfs_dir;
	1359	+ssize_t kvm_stats_read(char id, const struct kvm_stats_header header,
	1360	+ const struct _kvm_stats_desc *desc,
	1361	+ void *stats, size_t size_stats,
	1362	+ char __user user_buffer, size_t size, loff_t offset);
1285	1363
1286	1364	#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1287	1365	static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)

include/linux/kvm_types.h+2−0

@@ -89,4 +89,6 @@ struct kvm_vcpu_stat_generic { · 88 unmodified lines
89	89	u64 halt_poll_fail_ns;
90	90	};
91	91
	92	+#define KVM_STATS_NAME_SIZE 48
	93	+
92	94	#endif /* __KVM_TYPES_H__ */

include/uapi/linux/kvm.h+73−0

@@ -1087,6 +1087,7 @@ struct kvm_ppc_resize_hpt { · 1086 unmodified lines
1087	1087	#define KVM_CAP_SREGS2 200
1088	1088	#define KVM_CAP_EXIT_HYPERCALL 201
1089	1089	#define KVM_CAP_PPC_RPT_INVALIDATE 202
	1090	+#define KVM_CAP_BINARY_STATS_FD 203
1090	1091
1091	1092	#ifdef KVM_CAP_IRQ_ROUTING
1092	1093
@@ -1906,4 +1907,76 @@ struct kvm_dirty_gfn { · 813 unmodified lines
1906	1907	#define KVM_BUS_LOCK_DETECTION_OFF (1 << 0)
1907	1908	#define KVM_BUS_LOCK_DETECTION_EXIT (1 << 1)
1908	1909
	1910	+/**
	1911	+ * struct kvm_stats_header - Header of per vm/vcpu binary statistics data.
	1912	+ * @flags: Some extra information for header, always 0 for now.
	1913	+ * @name_size: The size in bytes of the memory which contains statistics
	1914	+ * name string including trailing '\0'. The memory is allocated
	1915	+ * at the send of statistics descriptor.
	1916	+ * @num_desc: The number of statistics the vm or vcpu has.
	1917	+ * @id_offset: The offset of the vm/vcpu stats' id string in the file pointed
	1918	+ * by vm/vcpu stats fd.
	1919	+ * @desc_offset: The offset of the vm/vcpu stats' descriptor block in the file
	1920	+ * pointd by vm/vcpu stats fd.
	1921	+ * @data_offset: The offset of the vm/vcpu stats' data block in the file
	1922	+ * pointed by vm/vcpu stats fd.
	1923	+ *
	1924	+ * This is the header userspace needs to read from stats fd before any other
	1925	+ * readings. It is used by userspace to discover all the information about the
	1926	+ * vm/vcpu's binary statistics.
	1927	+ * Userspace reads this header from the start of the vm/vcpu's stats fd.
	1928	+ */
	1929	+struct kvm_stats_header {
	1930	+ __u32 flags;
	1931	+ __u32 name_size;
	1932	+ __u32 num_desc;
	1933	+ __u32 id_offset;
	1934	+ __u32 desc_offset;
	1935	+ __u32 data_offset;
	1936	+};
	1937	+
	1938	+#define KVM_STATS_TYPE_SHIFT 0
	1939	+#define KVM_STATS_TYPE_MASK (0xF << KVM_STATS_TYPE_SHIFT)
	1940	+#define KVM_STATS_TYPE_CUMULATIVE (0x0 << KVM_STATS_TYPE_SHIFT)
	1941	+#define KVM_STATS_TYPE_INSTANT (0x1 << KVM_STATS_TYPE_SHIFT)
	1942	+#define KVM_STATS_TYPE_PEAK (0x2 << KVM_STATS_TYPE_SHIFT)
	1943	+#define KVM_STATS_TYPE_MAX KVM_STATS_TYPE_PEAK
	1944	+
	1945	+#define KVM_STATS_UNIT_SHIFT 4
	1946	+#define KVM_STATS_UNIT_MASK (0xF << KVM_STATS_UNIT_SHIFT)
	1947	+#define KVM_STATS_UNIT_NONE (0x0 << KVM_STATS_UNIT_SHIFT)
	1948	+#define KVM_STATS_UNIT_BYTES (0x1 << KVM_STATS_UNIT_SHIFT)
	1949	+#define KVM_STATS_UNIT_SECONDS (0x2 << KVM_STATS_UNIT_SHIFT)
	1950	+#define KVM_STATS_UNIT_CYCLES (0x3 << KVM_STATS_UNIT_SHIFT)
	1951	+#define KVM_STATS_UNIT_MAX KVM_STATS_UNIT_CYCLES
	1952	+
	1953	+#define KVM_STATS_BASE_SHIFT 8
	1954	+#define KVM_STATS_BASE_MASK (0xF << KVM_STATS_BASE_SHIFT)
	1955	+#define KVM_STATS_BASE_POW10 (0x0 << KVM_STATS_BASE_SHIFT)
	1956	+#define KVM_STATS_BASE_POW2 (0x1 << KVM_STATS_BASE_SHIFT)
	1957	+#define KVM_STATS_BASE_MAX KVM_STATS_BASE_POW2
	1958	+
	1959	+/**
	1960	+ * struct kvm_stats_desc - Descriptor of a KVM statistics.
	1961	+ * @flags: Annotations of the stats, like type, unit, etc.
	1962	+ * @exponent: Used together with @flags to determine the unit.
	1963	+ * @size: The number of data items for this stats.
	1964	+ * Every data item is of type __u64.
	1965	+ * @offset: The offset of the stats to the start of stat structure in
	1966	+ * struture kvm or kvm_vcpu.
	1967	+ * @unused: Unused field for future usage. Always 0 for now.
	1968	+ * @name: The name string for the stats. Its size is indicated by the
	1969	+ * &kvm_stats_header->name_size.
	1970	+ */
	1971	+struct kvm_stats_desc {
	1972	+ __u32 flags;
	1973	+ __s16 exponent;
	1974	+ __u16 size;
	1975	+ __u32 offset;
	1976	+ __u32 unused;
	1977	+ char name[];
	1978	+};
	1979	+
	1980	+#define KVM_GET_STATS_FD _IO(KVMIO, 0xce)
	1981	+
1909	1982	#endif /* __LINUX_KVM_H */

virt/kvm/binary_stats.c+146−0

@@ -0,0 +1,146 @@
	1	+// SPDX-License-Identifier: GPL-2.0-only
	2	+/*
	3	+ * KVM binary statistics interface implementation
	4	+ *
	5	+ * Copyright 2021 Google LLC
	6	+ */
	7	+
	8	+#include <linux/kvm_host.h>
	9	+#include <linux/kvm.h>
	10	+#include <linux/errno.h>
	11	+#include <linux/uaccess.h>
	12	+
	13	+/**
	14	+ * kvm_stats_read() - Common function to read from the binary statistics
	15	+ * file descriptor.
	16	+ *
	17	+ * @id: identification string of the stats
	18	+ * @header: stats header for a vm or a vcpu
	19	+ * @desc: start address of an array of stats descriptors for a vm or a vcpu
	20	+ * @stats: start address of stats data block for a vm or a vcpu
	21	+ * @size_stats: the size of stats data block pointed by @stats
	22	+ * @user_buffer: start address of userspace buffer
	23	+ * @size: requested read size from userspace
	24	+ * @offset: the start position from which the content will be read for the
	25	+ * corresponding vm or vcp file descriptor
	26	+ *
	27	+ * The file content of a vm/vcpu file descriptor is now defined as below:
	28	+ * +-------------+
	29	+ * \| Header \|
	30	+ * +-------------+
	31	+ * \| id string \|
	32	+ * +-------------+
	33	+ * \| Descriptors \|
	34	+ * +-------------+
	35	+ * \| Stats Data \|
	36	+ * +-------------+
	37	+ * Although this function allows userspace to read any amount of data (as long
	38	+ * as in the limit) from any position, the typical usage would follow below
	39	+ * steps:
	40	+ * 1. Read header from offset 0. Get the offset of descriptors and stats data
	41	+ * and some other necessary information. This is a one-time work for the
	42	+ * lifecycle of the corresponding vm/vcpu stats fd.
	43	+ * 2. Read id string from its offset. This is a one-time work for the lifecycle
	44	+ * of the corresponding vm/vcpu stats fd.
	45	+ * 3. Read descriptors from its offset and discover all the stats by parsing
	46	+ * descriptors. This is a one-time work for the lifecycle of the
	47	+ * corresponding vm/vcpu stats fd.
	48	+ * 4. Periodically read stats data from its offset using pread.
	49	+ *
	50	+ * Return: the number of bytes that has been successfully read
	51	+ */
	52	+ssize_t kvm_stats_read(char id, const struct kvm_stats_header header,
	53	+ const struct _kvm_stats_desc *desc,
	54	+ void *stats, size_t size_stats,
	55	+ char __user user_buffer, size_t size, loff_t offset)
	56	+{
	57	+ ssize_t len;
	58	+ ssize_t copylen;
	59	+ ssize_t remain = size;
	60	+ size_t size_desc;
	61	+ size_t size_header;
	62	+ void *src;
	63	+ loff_t pos = *offset;
	64	+ char __user *dest = user_buffer;
	65	+
	66	+ size_header = sizeof(*header);
	67	+ size_desc = header->num_desc * sizeof(*desc);
	68	+
	69	+ len = KVM_STATS_NAME_SIZE + size_header + size_desc + size_stats - pos;
	70	+ len = min(len, remain);
	71	+ if (len <= 0)
	72	+ return 0;
	73	+ remain = len;
	74	+
	75	+ /*
	76	+ * Copy kvm stats header.
	77	+ * The header is the first block of content userspace usually read out.
	78	+ * The pos is 0 and the copylen and remain would be the size of header.
	79	+ * The copy of the header would be skipped if offset is larger than the
	80	+ * size of header. That usually happens when userspace reads stats
	81	+ * descriptors and stats data.
	82	+ */
	83	+ copylen = size_header - pos;
	84	+ copylen = min(copylen, remain);
	85	+ if (copylen > 0) {
	86	+ src = (void *)header + pos;
	87	+ if (copy_to_user(dest, src, copylen))
	88	+ return -EFAULT;
	89	+ remain -= copylen;
	90	+ pos += copylen;
	91	+ dest += copylen;
	92	+ }
	93	+
	94	+ /*
	95	+ * Copy kvm stats header id string.
	96	+ * The id string is unique for every vm/vcpu, which is stored in kvm
	97	+ * and kvm_vcpu structure.
	98	+ * The id string is part of the stat header from the perspective of
	99	+ * userspace, it is usually read out together with previous constant
	100	+ * header part and could be skipped for later descriptors and stats
	101	+ * data readings.
	102	+ */
	103	+ copylen = header->id_offset + KVM_STATS_NAME_SIZE - pos;
	104	+ copylen = min(copylen, remain);
	105	+ if (copylen > 0) {
	106	+ src = id + pos - header->id_offset;
	107	+ if (copy_to_user(dest, src, copylen))
	108	+ return -EFAULT;
	109	+ remain -= copylen;
	110	+ pos += copylen;
	111	+ dest += copylen;
	112	+ }
	113	+
	114	+ /*
	115	+ * Copy kvm stats descriptors.
	116	+ * The descriptors copy would be skipped in the typical case that
	117	+ * userspace periodically read stats data, since the pos would be
	118	+ * greater than the end address of descriptors
	119	+ * (header->header.desc_offset + size_desc) causing copylen <= 0.
	120	+ */
	121	+ copylen = header->desc_offset + size_desc - pos;
	122	+ copylen = min(copylen, remain);
	123	+ if (copylen > 0) {
	124	+ src = (void *)desc + pos - header->desc_offset;
	125	+ if (copy_to_user(dest, src, copylen))
	126	+ return -EFAULT;
	127	+ remain -= copylen;
	128	+ pos += copylen;
	129	+ dest += copylen;
	130	+ }
	131	+
	132	+ /* Copy kvm stats values */
	133	+ copylen = header->data_offset + size_stats - pos;
	134	+ copylen = min(copylen, remain);
	135	+ if (copylen > 0) {
	136	+ src = stats + pos - header->data_offset;
	137	+ if (copy_to_user(dest, src, copylen))
	138	+ return -EFAULT;
	139	+ remain -= copylen;
	140	+ pos += copylen;
	141	+ dest += copylen;
	142	+ }
	143	+
	144	+ *offset = pos;
	145	+ return len;
	146	+}