diff mbox series

[v11,2/7] KVM: stats: Add fd-based API to read binary stats data

Message ID 20210618044819.3690166-3-jingzhangos@google.com (mailing list archive)
State New, archived
Headers show
Series KVM statistics data fd-based binary interface | expand

Commit Message

Jing Zhang June 18, 2021, 4:48 a.m. UTC
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.

Reviewed-by: David Matlack <dmatlack@google.com>
Reviewed-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com> #arm64
Signed-off-by: Jing Zhang <jingzhangos@google.com>
---
 arch/arm64/kvm/Makefile   |   2 +-
 arch/mips/kvm/Makefile    |   2 +-
 arch/powerpc/kvm/Makefile |   2 +-
 arch/s390/kvm/Makefile    |   3 +-
 arch/x86/kvm/Makefile     |   2 +-
 include/linux/kvm_host.h  | 145 ++++++++++++++++++++++++++++++++++++++
 include/uapi/linux/kvm.h  |  42 +++++++++++
 virt/kvm/binary_stats.c   | 130 ++++++++++++++++++++++++++++++++++
 8 files changed, 323 insertions(+), 5 deletions(-)
 create mode 100644 virt/kvm/binary_stats.c

Comments

Greg KH June 18, 2021, 6:57 a.m. UTC | #1
Minor comment nits:

On Fri, Jun 18, 2021 at 04:48:14AM +0000, Jing Zhang wrote:
> +/*
> + * Common vm/vcpu stats read function to userspace.

Should you use a real kernel-doc style here?  You almost are, might as
well do it "right" :)

> + * @id: identification string of the stats
> + * @header: stats header for a vm or a vcpu
> + * @desc: start address of an array of stats descriptors for a vm or a vcpu
> + * @stats: start address of stats data block for a vm or a vcpu
> + * @size_stats: the size of stats data block pointed by @stats
> + * @user_buffer: start address of userspace buffer
> + * @size: requested read size from userspace
> + * @offset: the start position from which the content will be read for the
> + *          corresponding vm or vcp file descriptor
> + *
> + * The file content of a vm/vcpu file descriptor is now defined as below:
> + * +-------------+
> + * |   Header    |
> + * +-------------+
> + * | Descriptors |
> + * +-------------+
> + * | Stats Data  |
> + * +-------------+

Where is the "header id string"?  In the header?

> + * Although this function allows userspace to read any amount of data (as long
> + * as in the limit) from any position, the typical usage would follow below
> + * steps:
> + * 1. Read header from offset 0. Get the offset of descriptors and stats data
> + *    and some other necessary information. This is a one-time work for the
> + *    lifecycle of the corresponding vm/vcpu stats fd.
> + * 2. Read descriptors from its offset and discover all the stats by parsing
> + *    descriptors. This is a one-time work for the lifecycle of the
> + *    corresponding vm/vcpu stats fd.
> + * 3. Periodically read stats data from its offset.

You forgot "2.5.  rewind fd pointer position", see below...

> + */
> +ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
> +		struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
> +		char __user *user_buffer, size_t size, loff_t *offset)
> +{
> +	ssize_t len;
> +	ssize_t copylen;
> +	ssize_t remain = size;
> +	size_t size_desc;
> +	size_t size_header;
> +	void *src;
> +	loff_t pos = *offset;
> +	char __user *dest = user_buffer;
> +
> +	size_header = sizeof(*header);
> +	size_desc = header->count * sizeof(*desc);
> +
> +	len = KVM_STATS_ID_MAXLEN + size_header + size_desc + size_stats - pos;
> +	len = min(len, remain);
> +	if (len <= 0)
> +		return 0;
> +	remain = len;
> +
> +	/* Copy kvm stats header.
> +	 * The header is the first block of content userspace usually read out.
> +	 * The pos is 0 and the copylen and remain would be the size of header.
> +	 * The copy of the header would be skipped if offset is larger than the
> +	 * size of header. That usually happens when userspace reads stats
> +	 * descriptors and stats data.
> +	 */

Looks like this is the networking "style" of multi-line comments, not
the rest of the kernel.  You might want to fix this up to be the normal
style which would be:

	/*
	 * Copy kvm stats header.
	 * The header is the first block of content userspace usually read out.
	 * The pos is 0 and the copylen and remain would be the size of header.
	 * The copy of the header would be skipped if offset is larger than the
	 * size of header. That usually happens when userspace reads stats
	 * descriptors and stats data.
	 */

I do not know how picky the kvm maintainers are about this, that's up to
them :)


> +	copylen = size_header - pos;
> +	copylen = min(copylen, remain);
> +	if (copylen > 0) {
> +		src = (void *)header + pos;
> +		if (copy_to_user(dest, src, copylen))
> +			return -EFAULT;
> +		remain -= copylen;
> +		pos += copylen;
> +		dest += copylen;
> +	}
> +
> +	/* Copy kvm stats header id string.
> +	 * The id string is unique for every vm/vcpu, which is stored in kvm
> +	 * and kvm_vcpu structure.
> +	 */

This header too is skipped if necessary, so you should say that as well.


> +	copylen = size_header + KVM_STATS_ID_MAXLEN - pos;
> +	copylen = min(copylen, remain);
> +	if (copylen > 0) {
> +		src = id + pos - size_header;
> +		if (copy_to_user(dest, src, copylen))
> +			return -EFAULT;
> +		remain -= copylen;
> +		pos += copylen;
> +		dest += copylen;
> +	}
> +
> +	/* Copy kvm stats descriptors.
> +	 * The descriptors copy would be skipped in the typical case that
> +	 * userspace periodically read stats data, since the pos would be
> +	 * greater than the end address of descriptors
> +	 * (header->header.desc_offset + size_desc) causing copylen <= 0.
> +	 */

But you say that it is skipped here.

> +	copylen = header->desc_offset + size_desc - pos;
> +	copylen = min(copylen, remain);
> +	if (copylen > 0) {
> +		src = (void *)desc + pos - header->desc_offset;
> +		if (copy_to_user(dest, src, copylen))
> +			return -EFAULT;
> +		remain -= copylen;
> +		pos += copylen;
> +		dest += copylen;
> +	}
> +
> +	/* Copy kvm stats values */
> +	copylen = header->data_offset + size_stats - pos;
> +	copylen = min(copylen, remain);
> +	if (copylen > 0) {
> +		src = stats + pos - header->data_offset;

This lets you sync to the end of the header and read just the stats, but
does that mean that userspace keeps needing to "rewind" back to the end
of the header to read the stats again?

Or can it just keep reading off the end of the previous read?

It's not quite obvious here, and I mention that above in step "2.5", so
maybe I am wrong, which is fine, but then I'm confused :)


thanks,

greg k-h
Greg KH June 18, 2021, 7 a.m. UTC | #2
On Fri, Jun 18, 2021 at 04:48:14AM +0000, Jing Zhang wrote:
> 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.
> 
> Reviewed-by: David Matlack <dmatlack@google.com>
> Reviewed-by: Ricardo Koller <ricarkol@google.com>
> Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
> Reviewed-by: Fuad Tabba <tabba@google.com>
> Tested-by: Fuad Tabba <tabba@google.com> #arm64
> Signed-off-by: Jing Zhang <jingzhangos@google.com>
> ---
>  arch/arm64/kvm/Makefile   |   2 +-
>  arch/mips/kvm/Makefile    |   2 +-
>  arch/powerpc/kvm/Makefile |   2 +-
>  arch/s390/kvm/Makefile    |   3 +-
>  arch/x86/kvm/Makefile     |   2 +-
>  include/linux/kvm_host.h  | 145 ++++++++++++++++++++++++++++++++++++++
>  include/uapi/linux/kvm.h  |  42 +++++++++++
>  virt/kvm/binary_stats.c   | 130 ++++++++++++++++++++++++++++++++++
>  8 files changed, 323 insertions(+), 5 deletions(-)
>  create mode 100644 virt/kvm/binary_stats.c
> 
> diff --git a/arch/arm64/kvm/Makefile b/arch/arm64/kvm/Makefile
> index 589921392cb1..989bb5dad2c8 100644
> --- a/arch/arm64/kvm/Makefile
> +++ b/arch/arm64/kvm/Makefile
> @@ -11,7 +11,7 @@ obj-$(CONFIG_KVM) += kvm.o
>  obj-$(CONFIG_KVM) += hyp/
>  
>  kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \
> -	 $(KVM)/vfio.o $(KVM)/irqchip.o \
> +	 $(KVM)/vfio.o $(KVM)/irqchip.o $(KVM)/binary_stats.o \
>  	 arm.o mmu.o mmio.o psci.o perf.o hypercalls.o pvtime.o \
>  	 inject_fault.o va_layout.o handle_exit.o \
>  	 guest.o debug.o reset.o sys_regs.o \
> diff --git a/arch/mips/kvm/Makefile b/arch/mips/kvm/Makefile
> index 30cc060857c7..c67250a956b8 100644
> --- a/arch/mips/kvm/Makefile
> +++ b/arch/mips/kvm/Makefile
> @@ -2,7 +2,7 @@
>  # Makefile for KVM support for MIPS
>  #
>  
> -common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o)
> +common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o binary_stats.o)
>  
>  EXTRA_CFLAGS += -Ivirt/kvm -Iarch/mips/kvm
>  
> diff --git a/arch/powerpc/kvm/Makefile b/arch/powerpc/kvm/Makefile
> index 2bfeaa13befb..b347d043b932 100644
> --- a/arch/powerpc/kvm/Makefile
> +++ b/arch/powerpc/kvm/Makefile
> @@ -6,7 +6,7 @@
>  ccflags-y := -Ivirt/kvm -Iarch/powerpc/kvm
>  KVM := ../../../virt/kvm
>  
> -common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o
> +common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/binary_stats.o
>  common-objs-$(CONFIG_KVM_VFIO) += $(KVM)/vfio.o
>  common-objs-$(CONFIG_KVM_MMIO) += $(KVM)/coalesced_mmio.o
>  
> diff --git a/arch/s390/kvm/Makefile b/arch/s390/kvm/Makefile
> index 12decca22e7c..b3aaadc60ead 100644
> --- a/arch/s390/kvm/Makefile
> +++ b/arch/s390/kvm/Makefile
> @@ -4,7 +4,8 @@
>  # Copyright IBM Corp. 2008
>  
>  KVM := ../../../virt/kvm
> -common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o $(KVM)/irqchip.o $(KVM)/vfio.o
> +common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o \
> +	      $(KVM)/irqchip.o $(KVM)/vfio.o $(KVM)/binary_stats.o
>  
>  ccflags-y := -Ivirt/kvm -Iarch/s390/kvm
>  
> diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
> index 83331376b779..75dfd27b6e8a 100644
> --- a/arch/x86/kvm/Makefile
> +++ b/arch/x86/kvm/Makefile
> @@ -11,7 +11,7 @@ KVM := ../../../virt/kvm
>  
>  kvm-y			+= $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o \
>  				$(KVM)/eventfd.o $(KVM)/irqchip.o $(KVM)/vfio.o \
> -				$(KVM)/dirty_ring.o
> +				$(KVM)/dirty_ring.o $(KVM)/binary_stats.o
>  kvm-$(CONFIG_KVM_ASYNC_PF)	+= $(KVM)/async_pf.o
>  
>  kvm-y			+= x86.o emulate.o i8259.o irq.o lapic.o \
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index 5a31e0696360..2f0d12064ae7 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -1272,6 +1272,12 @@ struct kvm_stats_debugfs_item {
>  	int mode;
>  };
>  
> +#define KVM_STATS_NAME_LEN	48
> +struct _kvm_stats_desc {
> +	struct kvm_stats_desc desc;
> +	char name[KVM_STATS_NAME_LEN];
> +};
> +
>  #define KVM_DBGFS_GET_MODE(dbgfs_item)                                         \
>  	((dbgfs_item)->mode ? (dbgfs_item)->mode : 0644)
>  
> @@ -1285,8 +1291,147 @@ struct kvm_stats_debugfs_item {
>  	{ n, offsetof(struct kvm_vcpu, stat.generic.x),			       \
>  	  KVM_STAT_VCPU, ## __VA_ARGS__ }
>  
> +#define STATS_DESC_COMMON(type, unit, base, exp)			       \
> +	.flags = type | unit | base |					       \
> +	    BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |		       \
> +	    BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |		       \
> +	    BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),		       \
> +	.exponent = exp,						       \
> +	.size = 1
> +
> +#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp)		       \
> +	{								       \
> +		{							       \
> +			STATS_DESC_COMMON(type, unit, base, exp),	       \
> +			.offset = offsetof(struct kvm_vm_stat, generic.stat)   \
> +		},							       \
> +		.name = #stat,						       \
> +	}
> +#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp)		       \
> +	{								       \
> +		{							       \
> +			STATS_DESC_COMMON(type, unit, base, exp),	       \
> +			.offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
> +		},							       \
> +		.name = #stat,						       \
> +	}
> +#define VM_STATS_DESC(stat, type, unit, base, exp)			       \
> +	{								       \
> +		{							       \
> +			STATS_DESC_COMMON(type, unit, base, exp),	       \
> +			.offset = offsetof(struct kvm_vm_stat, stat)	       \
> +		},							       \
> +		.name = #stat,						       \
> +	}
> +#define VCPU_STATS_DESC(stat, type, unit, base, exp)			       \
> +	{								       \
> +		{							       \
> +			STATS_DESC_COMMON(type, unit, base, exp),	       \
> +			.offset = offsetof(struct kvm_vcpu_stat, stat)	       \
> +		},							       \
> +		.name = #stat,						       \
> +	}
> +/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
> +#define STATS_DESC(SCOPE, stat, type, unit, base, exp)			       \
> +	SCOPE##_STATS_DESC(stat, type, unit, base, exp)
> +
> +#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)	       \
> +	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,		       \
> +		      unit, base, exponent)
> +#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)		       \
> +	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, unit, base, exponent)  \
> +
> +/* Cumulative counter */
> +#define STATS_DESC_COUNTER(SCOPE, name)					       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
> +		KVM_STATS_BASE_POW10, 0)
> +/* Instantaneous counter */
> +#define STATS_DESC_ICOUNTER(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
> +		KVM_STATS_BASE_POW10, 0)
> +
> +/* Cumulative clock cycles */
> +#define STATS_DESC_CYCLE(SCOPE, name)					       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_CYCLES,	       \
> +		KVM_STATS_BASE_POW10, 0)
> +/* Instantaneous clock cycles */
> +#define STATS_DESC_ICYCLE(SCOPE, name)					       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_CYCLES,		       \
> +		KVM_STATS_BASE_POW10, 0)
> +
> +/* Cumulative memory size in Byte */
> +#define STATS_DESC_SIZE_BYTE(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
> +		KVM_STATS_BASE_POW2, 0)
> +/* Cumulative memory size in KiByte */
> +#define STATS_DESC_SIZE_KBYTE(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
> +		KVM_STATS_BASE_POW2, 10)
> +/* Cumulative memory size in MiByte */
> +#define STATS_DESC_SIZE_MBYTE(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
> +		KVM_STATS_BASE_POW2, 20)
> +/* Cumulative memory size in GiByte */
> +#define STATS_DESC_SIZE_GBYTE(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
> +		KVM_STATS_BASE_POW2, 30)
> +
> +/* Instantaneous memory size in Byte */
> +#define STATS_DESC_ISIZE_BYTE(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
> +		KVM_STATS_BASE_POW2, 0)
> +/* Instantaneous memory size in KiByte */
> +#define STATS_DESC_ISIZE_KBYTE(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
> +		KVM_STATS_BASE_POW2, 10)
> +/* Instantaneous memory size in MiByte */
> +#define STATS_DESC_ISIZE_MBYTE(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
> +		KVM_STATS_BASE_POW2, 20)
> +/* Instantaneous memory size in GiByte */
> +#define STATS_DESC_ISIZE_GBYTE(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
> +		KVM_STATS_BASE_POW2, 30)
> +
> +/* Cumulative time in second */
> +#define STATS_DESC_TIME_SEC(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
> +		KVM_STATS_BASE_POW10, 0)
> +/* Cumulative time in millisecond */
> +#define STATS_DESC_TIME_MSEC(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
> +		KVM_STATS_BASE_POW10, -3)
> +/* Cumulative time in microsecond */
> +#define STATS_DESC_TIME_USEC(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
> +		KVM_STATS_BASE_POW10, -6)
> +/* Cumulative time in nanosecond */
> +#define STATS_DESC_TIME_NSEC(SCOPE, name)				       \
> +	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
> +		KVM_STATS_BASE_POW10, -9)
> +
> +/* Instantaneous time in second */
> +#define STATS_DESC_ITIME_SEC(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
> +		KVM_STATS_BASE_POW10, 0)
> +/* Instantaneous time in millisecond */
> +#define STATS_DESC_ITIME_MSEC(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
> +		KVM_STATS_BASE_POW10, -3)
> +/* Instantaneous time in microsecond */
> +#define STATS_DESC_ITIME_USEC(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
> +		KVM_STATS_BASE_POW10, -6)
> +/* Instantaneous time in nanosecond */
> +#define STATS_DESC_ITIME_NSEC(SCOPE, name)				       \
> +	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
> +		KVM_STATS_BASE_POW10, -9)
> +
>  extern struct kvm_stats_debugfs_item debugfs_entries[];
>  extern struct dentry *kvm_debugfs_dir;
> +ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
> +		struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
> +		char __user *user_buffer, size_t size, loff_t *offset);
>  
>  #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
>  static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index 9febe1412f7a..ab73e905105c 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -1086,6 +1086,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_HYPERV_ENFORCE_CPUID 199
>  #define KVM_CAP_SREGS2 200
>  #define KVM_CAP_EXIT_HYPERCALL 201
> +#define KVM_CAP_BINARY_STATS_FD 202
>  
>  #ifdef KVM_CAP_IRQ_ROUTING
>  
> @@ -1905,4 +1906,45 @@ struct kvm_dirty_gfn {
>  #define KVM_BUS_LOCK_DETECTION_OFF             (1 << 0)
>  #define KVM_BUS_LOCK_DETECTION_EXIT            (1 << 1)
>  
> +#define KVM_STATS_ID_MAXLEN		64
> +
> +struct kvm_stats_header {
> +	__u32 name_size;
> +	__u32 count;
> +	__u32 desc_offset;
> +	__u32 data_offset;
> +	char id[];
> +};

You mentioned before that the size of this really is the size of the
structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?

If so, why not put that value explicitly in:
	char id[THE_REST_OF_THE_HEADER_SPACE];

As this is not a variable header size at all, and you can not change it
going forward, so the variable length array here feels disingenuous.



> +
> +#define KVM_STATS_TYPE_SHIFT		0
> +#define KVM_STATS_TYPE_MASK		(0xF << KVM_STATS_TYPE_SHIFT)
> +#define KVM_STATS_TYPE_CUMULATIVE	(0x0 << KVM_STATS_TYPE_SHIFT)
> +#define KVM_STATS_TYPE_INSTANT		(0x1 << KVM_STATS_TYPE_SHIFT)
> +#define KVM_STATS_TYPE_MAX		KVM_STATS_TYPE_INSTANT
> +
> +#define KVM_STATS_UNIT_SHIFT		4
> +#define KVM_STATS_UNIT_MASK		(0xF << KVM_STATS_UNIT_SHIFT)
> +#define KVM_STATS_UNIT_NONE		(0x0 << KVM_STATS_UNIT_SHIFT)
> +#define KVM_STATS_UNIT_BYTES		(0x1 << KVM_STATS_UNIT_SHIFT)
> +#define KVM_STATS_UNIT_SECONDS		(0x2 << KVM_STATS_UNIT_SHIFT)
> +#define KVM_STATS_UNIT_CYCLES		(0x3 << KVM_STATS_UNIT_SHIFT)
> +#define KVM_STATS_UNIT_MAX		KVM_STATS_UNIT_CYCLES
> +
> +#define KVM_STATS_BASE_SHIFT		8
> +#define KVM_STATS_BASE_MASK		(0xF << KVM_STATS_BASE_SHIFT)
> +#define KVM_STATS_BASE_POW10		(0x0 << KVM_STATS_BASE_SHIFT)
> +#define KVM_STATS_BASE_POW2		(0x1 << KVM_STATS_BASE_SHIFT)
> +#define KVM_STATS_BASE_MAX		KVM_STATS_BASE_POW2
> +
> +struct kvm_stats_desc {
> +	__u32 flags;
> +	__s16 exponent;
> +	__u16 size;
> +	__u32 offset;
> +	__u32 unused;
> +	char name[];
> +};

What is the max length of name?

Why aren't these structures defined here in kerneldoc so that we can
understand them better?  Putting them in a .rst file guarantees they
will get out of sync, and you can always directly import the kerneldoc
into the .rst file.

thanks,

greg k-h
Paolo Bonzini June 18, 2021, 8:02 a.m. UTC | #3
On 18/06/21 09:00, Greg KH wrote:
>> +struct kvm_stats_header {
>> +	__u32 name_size;
>> +	__u32 count;
>> +	__u32 desc_offset;
>> +	__u32 data_offset;
>> +	char id[];
>> +};
> 
> You mentioned before that the size of this really is the size of the
> structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
> 
> If so, why not put that value explicitly in:
> 	char id[THE_REST_OF_THE_HEADER_SPACE];
> 
> As this is not a variable header size at all, and you can not change it
> going forward, so the variable length array here feels disingenuous.

It can change; the header goes up to desc_offset.  Let's rename 
desc_offset to header_size.

>> +struct kvm_stats_desc {
>> +	__u32 flags;
>> +	__s16 exponent;
>> +	__u16 size;
>> +	__u32 offset;
>> +	__u32 unused;
>> +	char name[];
>> +};
> 
> What is the max length of name?

It's name_size in the header.

> Why aren't these structures defined here in kerneldoc so that we can
> understand them better?  Putting them in a .rst file guarantees they
> will get out of sync, and you can always directly import the kerneldoc
> into the .rst file.

This is a problem in general with Documentation/virt/kvm/api.rst.  The 
file is organized to match the kerneldoc structs to the ioctl that they 
are used for, and sometimes a ioctl includes different structs for each 
architecture.

It is probably possible to do it using :identifiers: and/or :doc:, but 
it would require running scripts/kernel-doc on the uAPI headers dozens 
of times.  That is quite expensive at 0.3s each run, but that's what you 
get with Perl (gcc -fsyntax-only is 20 times faster).

Paolo
Greg KH June 18, 2021, 8:23 a.m. UTC | #4
On Fri, Jun 18, 2021 at 10:02:57AM +0200, Paolo Bonzini wrote:
> On 18/06/21 09:00, Greg KH wrote:
> > > +struct kvm_stats_header {
> > > +	__u32 name_size;
> > > +	__u32 count;
> > > +	__u32 desc_offset;
> > > +	__u32 data_offset;
> > > +	char id[];
> > > +};
> > 
> > You mentioned before that the size of this really is the size of the
> > structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
> > 
> > If so, why not put that value explicitly in:
> > 	char id[THE_REST_OF_THE_HEADER_SPACE];
> > 
> > As this is not a variable header size at all, and you can not change it
> > going forward, so the variable length array here feels disingenuous.
> 
> It can change; the header goes up to desc_offset.  Let's rename desc_offset
> to header_size.

"Traditionally" the first field of a variable length structure like this
has the size.  So maybe this needs to be:

struct kvm_stats_header {
	__u32 header_size;
	__u32 name_size;
	__u32 num_desc;
	__u32 desc_offset;
	__u32 data_offset;
	char id[];
};

I just guessed at what "count" is as I do not remember at the moment,
but obviously "count" wasn't descriptive :)

Wait, what is "name_size" here for?

> > > +struct kvm_stats_desc {
> > > +	__u32 flags;
> > > +	__s16 exponent;
> > > +	__u16 size;
> > > +	__u32 offset;
> > > +	__u32 unused;
> > > +	char name[];
> > > +};
> > 
> > What is the max length of name?
> 
> It's name_size in the header.

So it's specified in the _previous_ header?  That feels wrong, shouldn't
this descriptor define what is in it?

I'm not trying to nit-pick here, I'm actually confused now.  Structures
that contain "headers" should have in those headers at least two things:
	- declare the size of themselves if they are variable length
	- declare offsets to other structures

Don't put a size in this header for the size of a later structure,
that's just extra complexity that is not needed.

Think of this as a stream of bytes across the wire like a hardware
descriptor.  We have loads of experience dealing with this with
protocols like USB and Greybus and PCI and the like.  Let's learn from
those experiences and not try to mess things up where we don't need to
:)


> 
> > Why aren't these structures defined here in kerneldoc so that we can
> > understand them better?  Putting them in a .rst file guarantees they
> > will get out of sync, and you can always directly import the kerneldoc
> > into the .rst file.
> 
> This is a problem in general with Documentation/virt/kvm/api.rst.  The file
> is organized to match the kerneldoc structs to the ioctl that they are used
> for, and sometimes a ioctl includes different structs for each architecture.
> 
> It is probably possible to do it using :identifiers: and/or :doc:, but it
> would require running scripts/kernel-doc on the uAPI headers dozens of
> times.  That is quite expensive at 0.3s each run, but that's what you get
> with Perl (gcc -fsyntax-only is 20 times faster).

Is that what v4l and drm do today?  That's still safer and more
"obvious" than trying to keep two different files in sync which, as I
well know, almost impossible to do well over the _years_ in which you
will have to maintain these files.

Let's make it easier for everyone, put it only in one place and if
people want to see the documentation, they can generate it (it's
auto-generated on kernel.org anyway), no need to worry about multiple
passes or not.

thanks,

greg k-h
Paolo Bonzini June 18, 2021, 8:28 a.m. UTC | #5
On 18/06/21 08:57, Greg KH wrote:
>> + * 2. Read descriptors from its offset and discover all the stats by parsing
>> + *    descriptors. This is a one-time work for the lifecycle of the
>> + *    corresponding vm/vcpu stats fd.
>> + * 3. Periodically read stats data from its offset.
> You forgot "2.5.  rewind fd pointer position", see below...

Or use pread (that's what the test does).  I'll do a copy-editing pass 
and be sure to mention that as well.

Paolo
Jing Zhang June 18, 2021, 12:40 p.m. UTC | #6
On Fri, Jun 18, 2021 at 1:57 AM Greg KH <gregkh@linuxfoundation.org> wrote:
>
> Minor comment nits:
>
> On Fri, Jun 18, 2021 at 04:48:14AM +0000, Jing Zhang wrote:
> > +/*
> > + * Common vm/vcpu stats read function to userspace.
>
> Should you use a real kernel-doc style here?  You almost are, might as
> well do it "right" :)
>
Will fix that.
> > + * @id: identification string of the stats
> > + * @header: stats header for a vm or a vcpu
> > + * @desc: start address of an array of stats descriptors for a vm or a vcpu
> > + * @stats: start address of stats data block for a vm or a vcpu
> > + * @size_stats: the size of stats data block pointed by @stats
> > + * @user_buffer: start address of userspace buffer
> > + * @size: requested read size from userspace
> > + * @offset: the start position from which the content will be read for the
> > + *          corresponding vm or vcp file descriptor
> > + *
> > + * The file content of a vm/vcpu file descriptor is now defined as below:
> > + * +-------------+
> > + * |   Header    |
> > + * +-------------+
> > + * | Descriptors |
> > + * +-------------+
> > + * | Stats Data  |
> > + * +-------------+
>
> Where is the "header id string"?  In the header?
>
Yes, the id string is in the header.
> > + * Although this function allows userspace to read any amount of data (as long
> > + * as in the limit) from any position, the typical usage would follow below
> > + * steps:
> > + * 1. Read header from offset 0. Get the offset of descriptors and stats data
> > + *    and some other necessary information. This is a one-time work for the
> > + *    lifecycle of the corresponding vm/vcpu stats fd.
> > + * 2. Read descriptors from its offset and discover all the stats by parsing
> > + *    descriptors. This is a one-time work for the lifecycle of the
> > + *    corresponding vm/vcpu stats fd.
> > + * 3. Periodically read stats data from its offset.
>
> You forgot "2.5.  rewind fd pointer position", see below...
>
Sure, will clarify that.
> > + */
> > +ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
> > +             struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
> > +             char __user *user_buffer, size_t size, loff_t *offset)
> > +{
> > +     ssize_t len;
> > +     ssize_t copylen;
> > +     ssize_t remain = size;
> > +     size_t size_desc;
> > +     size_t size_header;
> > +     void *src;
> > +     loff_t pos = *offset;
> > +     char __user *dest = user_buffer;
> > +
> > +     size_header = sizeof(*header);
> > +     size_desc = header->count * sizeof(*desc);
> > +
> > +     len = KVM_STATS_ID_MAXLEN + size_header + size_desc + size_stats - pos;
> > +     len = min(len, remain);
> > +     if (len <= 0)
> > +             return 0;
> > +     remain = len;
> > +
> > +     /* Copy kvm stats header.
> > +      * The header is the first block of content userspace usually read out.
> > +      * The pos is 0 and the copylen and remain would be the size of header.
> > +      * The copy of the header would be skipped if offset is larger than the
> > +      * size of header. That usually happens when userspace reads stats
> > +      * descriptors and stats data.
> > +      */
>
> Looks like this is the networking "style" of multi-line comments, not
> the rest of the kernel.  You might want to fix this up to be the normal
> style which would be:
>
>         /*
>          * Copy kvm stats header.
>          * The header is the first block of content userspace usually read out.
>          * The pos is 0 and the copylen and remain would be the size of header.
>          * The copy of the header would be skipped if offset is larger than the
>          * size of header. That usually happens when userspace reads stats
>          * descriptors and stats data.
>          */
>
> I do not know how picky the kvm maintainers are about this, that's up to
> them :)
>
>
Will fix it.
> > +     copylen = size_header - pos;
> > +     copylen = min(copylen, remain);
> > +     if (copylen > 0) {
> > +             src = (void *)header + pos;
> > +             if (copy_to_user(dest, src, copylen))
> > +                     return -EFAULT;
> > +             remain -= copylen;
> > +             pos += copylen;
> > +             dest += copylen;
> > +     }
> > +
> > +     /* Copy kvm stats header id string.
> > +      * The id string is unique for every vm/vcpu, which is stored in kvm
> > +      * and kvm_vcpu structure.
> > +      */
>
> This header too is skipped if necessary, so you should say that as well.
>
>
Sure, will clarify that.
> > +     copylen = size_header + KVM_STATS_ID_MAXLEN - pos;
> > +     copylen = min(copylen, remain);
> > +     if (copylen > 0) {
> > +             src = id + pos - size_header;
> > +             if (copy_to_user(dest, src, copylen))
> > +                     return -EFAULT;
> > +             remain -= copylen;
> > +             pos += copylen;
> > +             dest += copylen;
> > +     }
> > +
> > +     /* Copy kvm stats descriptors.
> > +      * The descriptors copy would be skipped in the typical case that
> > +      * userspace periodically read stats data, since the pos would be
> > +      * greater than the end address of descriptors
> > +      * (header->header.desc_offset + size_desc) causing copylen <= 0.
> > +      */
>
> But you say that it is skipped here.
>
> > +     copylen = header->desc_offset + size_desc - pos;
> > +     copylen = min(copylen, remain);
> > +     if (copylen > 0) {
> > +             src = (void *)desc + pos - header->desc_offset;
> > +             if (copy_to_user(dest, src, copylen))
> > +                     return -EFAULT;
> > +             remain -= copylen;
> > +             pos += copylen;
> > +             dest += copylen;
> > +     }
> > +
> > +     /* Copy kvm stats values */
> > +     copylen = header->data_offset + size_stats - pos;
> > +     copylen = min(copylen, remain);
> > +     if (copylen > 0) {
> > +             src = stats + pos - header->data_offset;
>
> This lets you sync to the end of the header and read just the stats, but
> does that mean that userspace keeps needing to "rewind" back to the end
> of the header to read the stats again?
>
> Or can it just keep reading off the end of the previous read?
>
> It's not quite obvious here, and I mention that above in step "2.5", so
> maybe I am wrong, which is fine, but then I'm confused :)
Userspace needs to rewind back to read the stats again or just use pread
as Paolo mentioned and that's used in the testcase.
>
>
> thanks,
>
> greg k-h
Jing Zhang June 18, 2021, 12:53 p.m. UTC | #7
On Fri, Jun 18, 2021 at 2:01 AM Greg KH <gregkh@linuxfoundation.org> wrote:
>
> On Fri, Jun 18, 2021 at 04:48:14AM +0000, Jing Zhang wrote:
> > 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.
> >
> > Reviewed-by: David Matlack <dmatlack@google.com>
> > Reviewed-by: Ricardo Koller <ricarkol@google.com>
> > Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
> > Reviewed-by: Fuad Tabba <tabba@google.com>
> > Tested-by: Fuad Tabba <tabba@google.com> #arm64
> > Signed-off-by: Jing Zhang <jingzhangos@google.com>
> > ---
> >  arch/arm64/kvm/Makefile   |   2 +-
> >  arch/mips/kvm/Makefile    |   2 +-
> >  arch/powerpc/kvm/Makefile |   2 +-
> >  arch/s390/kvm/Makefile    |   3 +-
> >  arch/x86/kvm/Makefile     |   2 +-
> >  include/linux/kvm_host.h  | 145 ++++++++++++++++++++++++++++++++++++++
> >  include/uapi/linux/kvm.h  |  42 +++++++++++
> >  virt/kvm/binary_stats.c   | 130 ++++++++++++++++++++++++++++++++++
> >  8 files changed, 323 insertions(+), 5 deletions(-)
> >  create mode 100644 virt/kvm/binary_stats.c
> >
> > diff --git a/arch/arm64/kvm/Makefile b/arch/arm64/kvm/Makefile
> > index 589921392cb1..989bb5dad2c8 100644
> > --- a/arch/arm64/kvm/Makefile
> > +++ b/arch/arm64/kvm/Makefile
> > @@ -11,7 +11,7 @@ obj-$(CONFIG_KVM) += kvm.o
> >  obj-$(CONFIG_KVM) += hyp/
> >
> >  kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \
> > -      $(KVM)/vfio.o $(KVM)/irqchip.o \
> > +      $(KVM)/vfio.o $(KVM)/irqchip.o $(KVM)/binary_stats.o \
> >        arm.o mmu.o mmio.o psci.o perf.o hypercalls.o pvtime.o \
> >        inject_fault.o va_layout.o handle_exit.o \
> >        guest.o debug.o reset.o sys_regs.o \
> > diff --git a/arch/mips/kvm/Makefile b/arch/mips/kvm/Makefile
> > index 30cc060857c7..c67250a956b8 100644
> > --- a/arch/mips/kvm/Makefile
> > +++ b/arch/mips/kvm/Makefile
> > @@ -2,7 +2,7 @@
> >  # Makefile for KVM support for MIPS
> >  #
> >
> > -common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o)
> > +common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o binary_stats.o)
> >
> >  EXTRA_CFLAGS += -Ivirt/kvm -Iarch/mips/kvm
> >
> > diff --git a/arch/powerpc/kvm/Makefile b/arch/powerpc/kvm/Makefile
> > index 2bfeaa13befb..b347d043b932 100644
> > --- a/arch/powerpc/kvm/Makefile
> > +++ b/arch/powerpc/kvm/Makefile
> > @@ -6,7 +6,7 @@
> >  ccflags-y := -Ivirt/kvm -Iarch/powerpc/kvm
> >  KVM := ../../../virt/kvm
> >
> > -common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o
> > +common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/binary_stats.o
> >  common-objs-$(CONFIG_KVM_VFIO) += $(KVM)/vfio.o
> >  common-objs-$(CONFIG_KVM_MMIO) += $(KVM)/coalesced_mmio.o
> >
> > diff --git a/arch/s390/kvm/Makefile b/arch/s390/kvm/Makefile
> > index 12decca22e7c..b3aaadc60ead 100644
> > --- a/arch/s390/kvm/Makefile
> > +++ b/arch/s390/kvm/Makefile
> > @@ -4,7 +4,8 @@
> >  # Copyright IBM Corp. 2008
> >
> >  KVM := ../../../virt/kvm
> > -common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o $(KVM)/irqchip.o $(KVM)/vfio.o
> > +common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o \
> > +           $(KVM)/irqchip.o $(KVM)/vfio.o $(KVM)/binary_stats.o
> >
> >  ccflags-y := -Ivirt/kvm -Iarch/s390/kvm
> >
> > diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
> > index 83331376b779..75dfd27b6e8a 100644
> > --- a/arch/x86/kvm/Makefile
> > +++ b/arch/x86/kvm/Makefile
> > @@ -11,7 +11,7 @@ KVM := ../../../virt/kvm
> >
> >  kvm-y                        += $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o \
> >                               $(KVM)/eventfd.o $(KVM)/irqchip.o $(KVM)/vfio.o \
> > -                             $(KVM)/dirty_ring.o
> > +                             $(KVM)/dirty_ring.o $(KVM)/binary_stats.o
> >  kvm-$(CONFIG_KVM_ASYNC_PF)   += $(KVM)/async_pf.o
> >
> >  kvm-y                        += x86.o emulate.o i8259.o irq.o lapic.o \
> > diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> > index 5a31e0696360..2f0d12064ae7 100644
> > --- a/include/linux/kvm_host.h
> > +++ b/include/linux/kvm_host.h
> > @@ -1272,6 +1272,12 @@ struct kvm_stats_debugfs_item {
> >       int mode;
> >  };
> >
> > +#define KVM_STATS_NAME_LEN   48
> > +struct _kvm_stats_desc {
> > +     struct kvm_stats_desc desc;
> > +     char name[KVM_STATS_NAME_LEN];
> > +};
> > +
> >  #define KVM_DBGFS_GET_MODE(dbgfs_item)                                         \
> >       ((dbgfs_item)->mode ? (dbgfs_item)->mode : 0644)
> >
> > @@ -1285,8 +1291,147 @@ struct kvm_stats_debugfs_item {
> >       { n, offsetof(struct kvm_vcpu, stat.generic.x),                        \
> >         KVM_STAT_VCPU, ## __VA_ARGS__ }
> >
> > +#define STATS_DESC_COMMON(type, unit, base, exp)                            \
> > +     .flags = type | unit | base |                                          \
> > +         BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |                   \
> > +         BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |                   \
> > +         BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),                    \
> > +     .exponent = exp,                                                       \
> > +     .size = 1
> > +
> > +#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp)                  \
> > +     {                                                                      \
> > +             {                                                              \
> > +                     STATS_DESC_COMMON(type, unit, base, exp),              \
> > +                     .offset = offsetof(struct kvm_vm_stat, generic.stat)   \
> > +             },                                                             \
> > +             .name = #stat,                                                 \
> > +     }
> > +#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp)                \
> > +     {                                                                      \
> > +             {                                                              \
> > +                     STATS_DESC_COMMON(type, unit, base, exp),              \
> > +                     .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
> > +             },                                                             \
> > +             .name = #stat,                                                 \
> > +     }
> > +#define VM_STATS_DESC(stat, type, unit, base, exp)                          \
> > +     {                                                                      \
> > +             {                                                              \
> > +                     STATS_DESC_COMMON(type, unit, base, exp),              \
> > +                     .offset = offsetof(struct kvm_vm_stat, stat)           \
> > +             },                                                             \
> > +             .name = #stat,                                                 \
> > +     }
> > +#define VCPU_STATS_DESC(stat, type, unit, base, exp)                        \
> > +     {                                                                      \
> > +             {                                                              \
> > +                     STATS_DESC_COMMON(type, unit, base, exp),              \
> > +                     .offset = offsetof(struct kvm_vcpu_stat, stat)         \
> > +             },                                                             \
> > +             .name = #stat,                                                 \
> > +     }
> > +/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
> > +#define STATS_DESC(SCOPE, stat, type, unit, base, exp)                              \
> > +     SCOPE##_STATS_DESC(stat, type, unit, base, exp)
> > +
> > +#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)            \
> > +     STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,                     \
> > +                   unit, base, exponent)
> > +#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)                       \
> > +     STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, unit, base, exponent)  \
> > +
> > +/* Cumulative counter */
> > +#define STATS_DESC_COUNTER(SCOPE, name)                                             \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,                \
> > +             KVM_STATS_BASE_POW10, 0)
> > +/* Instantaneous counter */
> > +#define STATS_DESC_ICOUNTER(SCOPE, name)                                    \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,                   \
> > +             KVM_STATS_BASE_POW10, 0)
> > +
> > +/* Cumulative clock cycles */
> > +#define STATS_DESC_CYCLE(SCOPE, name)                                               \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_CYCLES,              \
> > +             KVM_STATS_BASE_POW10, 0)
> > +/* Instantaneous clock cycles */
> > +#define STATS_DESC_ICYCLE(SCOPE, name)                                              \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_CYCLES,                 \
> > +             KVM_STATS_BASE_POW10, 0)
> > +
> > +/* Cumulative memory size in Byte */
> > +#define STATS_DESC_SIZE_BYTE(SCOPE, name)                                   \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,               \
> > +             KVM_STATS_BASE_POW2, 0)
> > +/* Cumulative memory size in KiByte */
> > +#define STATS_DESC_SIZE_KBYTE(SCOPE, name)                                  \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,               \
> > +             KVM_STATS_BASE_POW2, 10)
> > +/* Cumulative memory size in MiByte */
> > +#define STATS_DESC_SIZE_MBYTE(SCOPE, name)                                  \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,               \
> > +             KVM_STATS_BASE_POW2, 20)
> > +/* Cumulative memory size in GiByte */
> > +#define STATS_DESC_SIZE_GBYTE(SCOPE, name)                                  \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,               \
> > +             KVM_STATS_BASE_POW2, 30)
> > +
> > +/* Instantaneous memory size in Byte */
> > +#define STATS_DESC_ISIZE_BYTE(SCOPE, name)                                  \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,                  \
> > +             KVM_STATS_BASE_POW2, 0)
> > +/* Instantaneous memory size in KiByte */
> > +#define STATS_DESC_ISIZE_KBYTE(SCOPE, name)                                 \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,                  \
> > +             KVM_STATS_BASE_POW2, 10)
> > +/* Instantaneous memory size in MiByte */
> > +#define STATS_DESC_ISIZE_MBYTE(SCOPE, name)                                 \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,                  \
> > +             KVM_STATS_BASE_POW2, 20)
> > +/* Instantaneous memory size in GiByte */
> > +#define STATS_DESC_ISIZE_GBYTE(SCOPE, name)                                 \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,                  \
> > +             KVM_STATS_BASE_POW2, 30)
> > +
> > +/* Cumulative time in second */
> > +#define STATS_DESC_TIME_SEC(SCOPE, name)                                    \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,             \
> > +             KVM_STATS_BASE_POW10, 0)
> > +/* Cumulative time in millisecond */
> > +#define STATS_DESC_TIME_MSEC(SCOPE, name)                                   \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,             \
> > +             KVM_STATS_BASE_POW10, -3)
> > +/* Cumulative time in microsecond */
> > +#define STATS_DESC_TIME_USEC(SCOPE, name)                                   \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,             \
> > +             KVM_STATS_BASE_POW10, -6)
> > +/* Cumulative time in nanosecond */
> > +#define STATS_DESC_TIME_NSEC(SCOPE, name)                                   \
> > +     STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,             \
> > +             KVM_STATS_BASE_POW10, -9)
> > +
> > +/* Instantaneous time in second */
> > +#define STATS_DESC_ITIME_SEC(SCOPE, name)                                   \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,                \
> > +             KVM_STATS_BASE_POW10, 0)
> > +/* Instantaneous time in millisecond */
> > +#define STATS_DESC_ITIME_MSEC(SCOPE, name)                                  \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,                \
> > +             KVM_STATS_BASE_POW10, -3)
> > +/* Instantaneous time in microsecond */
> > +#define STATS_DESC_ITIME_USEC(SCOPE, name)                                  \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,                \
> > +             KVM_STATS_BASE_POW10, -6)
> > +/* Instantaneous time in nanosecond */
> > +#define STATS_DESC_ITIME_NSEC(SCOPE, name)                                  \
> > +     STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,                \
> > +             KVM_STATS_BASE_POW10, -9)
> > +
> >  extern struct kvm_stats_debugfs_item debugfs_entries[];
> >  extern struct dentry *kvm_debugfs_dir;
> > +ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
> > +             struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
> > +             char __user *user_buffer, size_t size, loff_t *offset);
> >
> >  #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
> >  static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
> > diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> > index 9febe1412f7a..ab73e905105c 100644
> > --- a/include/uapi/linux/kvm.h
> > +++ b/include/uapi/linux/kvm.h
> > @@ -1086,6 +1086,7 @@ struct kvm_ppc_resize_hpt {
> >  #define KVM_CAP_HYPERV_ENFORCE_CPUID 199
> >  #define KVM_CAP_SREGS2 200
> >  #define KVM_CAP_EXIT_HYPERCALL 201
> > +#define KVM_CAP_BINARY_STATS_FD 202
> >
> >  #ifdef KVM_CAP_IRQ_ROUTING
> >
> > @@ -1905,4 +1906,45 @@ struct kvm_dirty_gfn {
> >  #define KVM_BUS_LOCK_DETECTION_OFF             (1 << 0)
> >  #define KVM_BUS_LOCK_DETECTION_EXIT            (1 << 1)
> >
> > +#define KVM_STATS_ID_MAXLEN          64
> > +
> > +struct kvm_stats_header {
> > +     __u32 name_size;
> > +     __u32 count;
> > +     __u32 desc_offset;
> > +     __u32 data_offset;
> > +     char id[];
> > +};
>
> You mentioned before that the size of this really is the size of the
> structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
>
> If so, why not put that value explicitly in:
>         char id[THE_REST_OF_THE_HEADER_SPACE];
>
> As this is not a variable header size at all, and you can not change it
> going forward, so the variable length array here feels disingenuous.
The size of header is not supposed to change, it is sizeof(struct
kvm_stats_header) +
KVM_STATS_ID_MAXLEN. Userspace needs this when reading the header.
Sorry for the confusion, will make that clear.

The reason char id[] is used instead of char id[KVM_STATS_ID_MAXLEN] is because
the header is only defined (and can be const as you mentioned) for
every architecture, but
the id string is different for every VM and VCPU. The content of the
header is combined
on-the-fly in the kvm_stats_read function.
If we use id[KVM_STATS_ID_MAXLEN] here, then we will waste some memory in every
header definition.
Another option is to define another internal header structure without
id[] field, which can be
used to define the const part of header for every architecture. But
that will have two structures
for the same header, we need to make sure these two structures are the
same except the id[]
field.

So, Greg, Paolo, What do you think? Which solution should we choose
from 3 options?
>
>
>
> > +
> > +#define KVM_STATS_TYPE_SHIFT         0
> > +#define KVM_STATS_TYPE_MASK          (0xF << KVM_STATS_TYPE_SHIFT)
> > +#define KVM_STATS_TYPE_CUMULATIVE    (0x0 << KVM_STATS_TYPE_SHIFT)
> > +#define KVM_STATS_TYPE_INSTANT               (0x1 << KVM_STATS_TYPE_SHIFT)
> > +#define KVM_STATS_TYPE_MAX           KVM_STATS_TYPE_INSTANT
> > +
> > +#define KVM_STATS_UNIT_SHIFT         4
> > +#define KVM_STATS_UNIT_MASK          (0xF << KVM_STATS_UNIT_SHIFT)
> > +#define KVM_STATS_UNIT_NONE          (0x0 << KVM_STATS_UNIT_SHIFT)
> > +#define KVM_STATS_UNIT_BYTES         (0x1 << KVM_STATS_UNIT_SHIFT)
> > +#define KVM_STATS_UNIT_SECONDS               (0x2 << KVM_STATS_UNIT_SHIFT)
> > +#define KVM_STATS_UNIT_CYCLES                (0x3 << KVM_STATS_UNIT_SHIFT)
> > +#define KVM_STATS_UNIT_MAX           KVM_STATS_UNIT_CYCLES
> > +
> > +#define KVM_STATS_BASE_SHIFT         8
> > +#define KVM_STATS_BASE_MASK          (0xF << KVM_STATS_BASE_SHIFT)
> > +#define KVM_STATS_BASE_POW10         (0x0 << KVM_STATS_BASE_SHIFT)
> > +#define KVM_STATS_BASE_POW2          (0x1 << KVM_STATS_BASE_SHIFT)
> > +#define KVM_STATS_BASE_MAX           KVM_STATS_BASE_POW2
> > +
> > +struct kvm_stats_desc {
> > +     __u32 flags;
> > +     __s16 exponent;
> > +     __u16 size;
> > +     __u32 offset;
> > +     __u32 unused;
> > +     char name[];
> > +};
>
> What is the max length of name?
>
As Paolo mentioned, it is read from the header.
> Why aren't these structures defined here in kerneldoc so that we can
> understand them better?  Putting them in a .rst file guarantees they
> will get out of sync, and you can always directly import the kerneldoc
> into the .rst file.
>
> thanks,
>
> greg k-h
Jing Zhang June 18, 2021, 1:02 p.m. UTC | #8
On Fri, Jun 18, 2021 at 3:23 AM Greg KH <gregkh@linuxfoundation.org> wrote:
>
> On Fri, Jun 18, 2021 at 10:02:57AM +0200, Paolo Bonzini wrote:
> > On 18/06/21 09:00, Greg KH wrote:
> > > > +struct kvm_stats_header {
> > > > + __u32 name_size;
> > > > + __u32 count;
> > > > + __u32 desc_offset;
> > > > + __u32 data_offset;
> > > > + char id[];
> > > > +};
> > >
> > > You mentioned before that the size of this really is the size of the
> > > structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
> > >
> > > If so, why not put that value explicitly in:
> > >     char id[THE_REST_OF_THE_HEADER_SPACE];
> > >
> > > As this is not a variable header size at all, and you can not change it
> > > going forward, so the variable length array here feels disingenuous.
> >
> > It can change; the header goes up to desc_offset.  Let's rename desc_offset
> > to header_size.
>
> "Traditionally" the first field of a variable length structure like this
> has the size.  So maybe this needs to be:
>
> struct kvm_stats_header {
>         __u32 header_size;
>         __u32 name_size;
>         __u32 num_desc;
>         __u32 desc_offset;
>         __u32 data_offset;
>         char id[];
> };
>
> I just guessed at what "count" is as I do not remember at the moment,
> but obviously "count" wasn't descriptive :)
>
> Wait, what is "name_size" here for?
>
> > > > +struct kvm_stats_desc {
> > > > + __u32 flags;
> > > > + __s16 exponent;
> > > > + __u16 size;
> > > > + __u32 offset;
> > > > + __u32 unused;
> > > > + char name[];
> > > > +};
> > >
> > > What is the max length of name?
> >
> > It's name_size in the header.
>
> So it's specified in the _previous_ header?  That feels wrong, shouldn't
> this descriptor define what is in it?
>
> I'm not trying to nit-pick here, I'm actually confused now.  Structures
> that contain "headers" should have in those headers at least two things:
>         - declare the size of themselves if they are variable length
>         - declare offsets to other structures
>
> Don't put a size in this header for the size of a later structure,
> that's just extra complexity that is not needed.
>
> Think of this as a stream of bytes across the wire like a hardware
> descriptor.  We have loads of experience dealing with this with
> protocols like USB and Greybus and PCI and the like.  Let's learn from
> those experiences and not try to mess things up where we don't need to
> :)
>
>
The name_size in the header is necessary for userspace to discover the
length of a descriptor of a KVM statistics, since the size of name[] in every
descriptor could be different (increased) in future versions.
One thing worth mentioning is that, for every file content, there is only one
header, but many descriptors, that's why it makes sense to have the size of
name[] in the header instead of in the descriptor. We don't want every
descriptor contain the same duplicated information.
> >
> > > Why aren't these structures defined here in kerneldoc so that we can
> > > understand them better?  Putting them in a .rst file guarantees they
> > > will get out of sync, and you can always directly import the kerneldoc
> > > into the .rst file.
> >
> > This is a problem in general with Documentation/virt/kvm/api.rst.  The file
> > is organized to match the kerneldoc structs to the ioctl that they are used
> > for, and sometimes a ioctl includes different structs for each architecture.
> >
> > It is probably possible to do it using :identifiers: and/or :doc:, but it
> > would require running scripts/kernel-doc on the uAPI headers dozens of
> > times.  That is quite expensive at 0.3s each run, but that's what you get
> > with Perl (gcc -fsyntax-only is 20 times faster).
>
> Is that what v4l and drm do today?  That's still safer and more
> "obvious" than trying to keep two different files in sync which, as I
> well know, almost impossible to do well over the _years_ in which you
> will have to maintain these files.
>
> Let's make it easier for everyone, put it only in one place and if
> people want to see the documentation, they can generate it (it's
> auto-generated on kernel.org anyway), no need to worry about multiple
> passes or not.
>
> thanks,
>
> greg k-h
Paolo Bonzini June 18, 2021, 3:51 p.m. UTC | #9
On 18/06/21 10:23, Greg KH wrote:
> On Fri, Jun 18, 2021 at 10:02:57AM +0200, Paolo Bonzini wrote:
>> On 18/06/21 09:00, Greg KH wrote:
>>>> +struct kvm_stats_header {
>>>> +	__u32 name_size;
>>>> +	__u32 count;
>>>> +	__u32 desc_offset;
>>>> +	__u32 data_offset;
>>>> +	char id[];
>>>> +};
>>>
>>> You mentioned before that the size of this really is the size of the
>>> structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
>>>
>>> If so, why not put that value explicitly in:
>>> 	char id[THE_REST_OF_THE_HEADER_SPACE];
>>>
>>> As this is not a variable header size at all, and you can not change it
>>> going forward, so the variable length array here feels disingenuous.
>>
>> It can change; the header goes up to desc_offset.  Let's rename desc_offset
>> to header_size.
> 
> "Traditionally" the first field of a variable length structure like this
> has the size.  So maybe this needs to be:
> 
> struct kvm_stats_header {
> 	__u32 header_size;

Thinking more about it, I slightly prefer id_offset so that we can later 
give a meaning to any bytes after kvm_stats_header and before id_offset.

Adding four unused bytes (for now always zero) is also useful to future 
proof the struct a bit, thus:

struct kvm_stats_header {
	__u32 flags;
	__u32 name_size;
	__u32 num_desc;
	__u32 id_offset;
	__u32 desc_offset;
	__u32 data_offset;
}

(Indeed num_desc is better than count).

> Wait, what is "name_size" here for?

So that you know the full size of the descriptors is (name_size + 
sizeof(kvm_stats_desc) + name_size) * num_desc.  That's the memory you 
allocate and the size that you can then pass to a single pread system 
call starting from offset desc_offset.

There is certainly room for improvement in that the length of id[] and 
name[] can be unified to name_size.

>>>> +struct kvm_stats_desc {
>>>> +	__u32 flags;
>>>> +	__s16 exponent;
>>>> +	__u16 size;
>>>> +	__u32 offset;
>>>> +	__u32 unused;
>>>> +	char name[];
>>>> +};
>>>
>>> What is the max length of name?
>>
>> It's name_size in the header.
> 
> So it's specified in the _previous_ header?  That feels wrong, shouldn't
> this descriptor define what is in it?

Compared to e.g. PCI where you can do random-access reads from memory or 
configuration space, reading from a file has slightly different 
tradeoffs.  So designing a file format is slightly different compared to 
designing an in-memory format, or a wire protocol.

Paolo
Jing Zhang June 18, 2021, 5:57 p.m. UTC | #10
Hi Paolo,

On Fri, Jun 18, 2021 at 10:51 AM Paolo Bonzini <bonzini@gnu.org> wrote:
>
> On 18/06/21 10:23, Greg KH wrote:
> > On Fri, Jun 18, 2021 at 10:02:57AM +0200, Paolo Bonzini wrote:
> >> On 18/06/21 09:00, Greg KH wrote:
> >>>> +struct kvm_stats_header {
> >>>> +  __u32 name_size;
> >>>> +  __u32 count;
> >>>> +  __u32 desc_offset;
> >>>> +  __u32 data_offset;
> >>>> +  char id[];
> >>>> +};
> >>>
> >>> You mentioned before that the size of this really is the size of the
> >>> structure + KVM_STATS_ID_MAXLEN, right?  Or is it - KVM_STATS_ID_MAXLEN?
> >>>
> >>> If so, why not put that value explicitly in:
> >>>     char id[THE_REST_OF_THE_HEADER_SPACE];
> >>>
> >>> As this is not a variable header size at all, and you can not change it
> >>> going forward, so the variable length array here feels disingenuous.
> >>
> >> It can change; the header goes up to desc_offset.  Let's rename desc_offset
> >> to header_size.
> >
> > "Traditionally" the first field of a variable length structure like this
> > has the size.  So maybe this needs to be:
> >
> > struct kvm_stats_header {
> >       __u32 header_size;
>
> Thinking more about it, I slightly prefer id_offset so that we can later
> give a meaning to any bytes after kvm_stats_header and before id_offset.
>
> Adding four unused bytes (for now always zero) is also useful to future
> proof the struct a bit, thus:
>
> struct kvm_stats_header {
>         __u32 flags;
>         __u32 name_size;
>         __u32 num_desc;
>         __u32 id_offset;
>         __u32 desc_offset;
>         __u32 data_offset;
> }
>
> (Indeed num_desc is better than count).
>
> > Wait, what is "name_size" here for?
>
> So that you know the full size of the descriptors is (name_size +
> sizeof(kvm_stats_desc) + name_size) * num_desc.  That's the memory you
> allocate and the size that you can then pass to a single pread system
> call starting from offset desc_offset.
>
> There is certainly room for improvement in that the length of id[] and
> name[] can be unified to name_size.
>
Thanks for all these ideas, which indeed make it more clear and neat.
Will improve by this and post another version later.
> >>>> +struct kvm_stats_desc {
> >>>> +  __u32 flags;
> >>>> +  __s16 exponent;
> >>>> +  __u16 size;
> >>>> +  __u32 offset;
> >>>> +  __u32 unused;
> >>>> +  char name[];
> >>>> +};
> >>>
> >>> What is the max length of name?
> >>
> >> It's name_size in the header.
> >
> > So it's specified in the _previous_ header?  That feels wrong, shouldn't
> > this descriptor define what is in it?
>
> Compared to e.g. PCI where you can do random-access reads from memory or
> configuration space, reading from a file has slightly different
> tradeoffs.  So designing a file format is slightly different compared to
> designing an in-memory format, or a wire protocol.
>
> Paolo

Jing
diff mbox series

Patch

diff --git a/arch/arm64/kvm/Makefile b/arch/arm64/kvm/Makefile
index 589921392cb1..989bb5dad2c8 100644
--- a/arch/arm64/kvm/Makefile
+++ b/arch/arm64/kvm/Makefile
@@ -11,7 +11,7 @@  obj-$(CONFIG_KVM) += kvm.o
 obj-$(CONFIG_KVM) += hyp/
 
 kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \
-	 $(KVM)/vfio.o $(KVM)/irqchip.o \
+	 $(KVM)/vfio.o $(KVM)/irqchip.o $(KVM)/binary_stats.o \
 	 arm.o mmu.o mmio.o psci.o perf.o hypercalls.o pvtime.o \
 	 inject_fault.o va_layout.o handle_exit.o \
 	 guest.o debug.o reset.o sys_regs.o \
diff --git a/arch/mips/kvm/Makefile b/arch/mips/kvm/Makefile
index 30cc060857c7..c67250a956b8 100644
--- a/arch/mips/kvm/Makefile
+++ b/arch/mips/kvm/Makefile
@@ -2,7 +2,7 @@ 
 # Makefile for KVM support for MIPS
 #
 
-common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o)
+common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o eventfd.o binary_stats.o)
 
 EXTRA_CFLAGS += -Ivirt/kvm -Iarch/mips/kvm
 
diff --git a/arch/powerpc/kvm/Makefile b/arch/powerpc/kvm/Makefile
index 2bfeaa13befb..b347d043b932 100644
--- a/arch/powerpc/kvm/Makefile
+++ b/arch/powerpc/kvm/Makefile
@@ -6,7 +6,7 @@ 
 ccflags-y := -Ivirt/kvm -Iarch/powerpc/kvm
 KVM := ../../../virt/kvm
 
-common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o
+common-objs-y = $(KVM)/kvm_main.o $(KVM)/eventfd.o $(KVM)/binary_stats.o
 common-objs-$(CONFIG_KVM_VFIO) += $(KVM)/vfio.o
 common-objs-$(CONFIG_KVM_MMIO) += $(KVM)/coalesced_mmio.o
 
diff --git a/arch/s390/kvm/Makefile b/arch/s390/kvm/Makefile
index 12decca22e7c..b3aaadc60ead 100644
--- a/arch/s390/kvm/Makefile
+++ b/arch/s390/kvm/Makefile
@@ -4,7 +4,8 @@ 
 # Copyright IBM Corp. 2008
 
 KVM := ../../../virt/kvm
-common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o $(KVM)/irqchip.o $(KVM)/vfio.o
+common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o \
+	      $(KVM)/irqchip.o $(KVM)/vfio.o $(KVM)/binary_stats.o
 
 ccflags-y := -Ivirt/kvm -Iarch/s390/kvm
 
diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
index 83331376b779..75dfd27b6e8a 100644
--- a/arch/x86/kvm/Makefile
+++ b/arch/x86/kvm/Makefile
@@ -11,7 +11,7 @@  KVM := ../../../virt/kvm
 
 kvm-y			+= $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o \
 				$(KVM)/eventfd.o $(KVM)/irqchip.o $(KVM)/vfio.o \
-				$(KVM)/dirty_ring.o
+				$(KVM)/dirty_ring.o $(KVM)/binary_stats.o
 kvm-$(CONFIG_KVM_ASYNC_PF)	+= $(KVM)/async_pf.o
 
 kvm-y			+= x86.o emulate.o i8259.o irq.o lapic.o \
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 5a31e0696360..2f0d12064ae7 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -1272,6 +1272,12 @@  struct kvm_stats_debugfs_item {
 	int mode;
 };
 
+#define KVM_STATS_NAME_LEN	48
+struct _kvm_stats_desc {
+	struct kvm_stats_desc desc;
+	char name[KVM_STATS_NAME_LEN];
+};
+
 #define KVM_DBGFS_GET_MODE(dbgfs_item)                                         \
 	((dbgfs_item)->mode ? (dbgfs_item)->mode : 0644)
 
@@ -1285,8 +1291,147 @@  struct kvm_stats_debugfs_item {
 	{ n, offsetof(struct kvm_vcpu, stat.generic.x),			       \
 	  KVM_STAT_VCPU, ## __VA_ARGS__ }
 
+#define STATS_DESC_COMMON(type, unit, base, exp)			       \
+	.flags = type | unit | base |					       \
+	    BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |		       \
+	    BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |		       \
+	    BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),		       \
+	.exponent = exp,						       \
+	.size = 1
+
+#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp)		       \
+	{								       \
+		{							       \
+			STATS_DESC_COMMON(type, unit, base, exp),	       \
+			.offset = offsetof(struct kvm_vm_stat, generic.stat)   \
+		},							       \
+		.name = #stat,						       \
+	}
+#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp)		       \
+	{								       \
+		{							       \
+			STATS_DESC_COMMON(type, unit, base, exp),	       \
+			.offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
+		},							       \
+		.name = #stat,						       \
+	}
+#define VM_STATS_DESC(stat, type, unit, base, exp)			       \
+	{								       \
+		{							       \
+			STATS_DESC_COMMON(type, unit, base, exp),	       \
+			.offset = offsetof(struct kvm_vm_stat, stat)	       \
+		},							       \
+		.name = #stat,						       \
+	}
+#define VCPU_STATS_DESC(stat, type, unit, base, exp)			       \
+	{								       \
+		{							       \
+			STATS_DESC_COMMON(type, unit, base, exp),	       \
+			.offset = offsetof(struct kvm_vcpu_stat, stat)	       \
+		},							       \
+		.name = #stat,						       \
+	}
+/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
+#define STATS_DESC(SCOPE, stat, type, unit, base, exp)			       \
+	SCOPE##_STATS_DESC(stat, type, unit, base, exp)
+
+#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)	       \
+	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,		       \
+		      unit, base, exponent)
+#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)		       \
+	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, unit, base, exponent)  \
+
+/* Cumulative counter */
+#define STATS_DESC_COUNTER(SCOPE, name)					       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
+		KVM_STATS_BASE_POW10, 0)
+/* Instantaneous counter */
+#define STATS_DESC_ICOUNTER(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
+		KVM_STATS_BASE_POW10, 0)
+
+/* Cumulative clock cycles */
+#define STATS_DESC_CYCLE(SCOPE, name)					       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_CYCLES,	       \
+		KVM_STATS_BASE_POW10, 0)
+/* Instantaneous clock cycles */
+#define STATS_DESC_ICYCLE(SCOPE, name)					       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_CYCLES,		       \
+		KVM_STATS_BASE_POW10, 0)
+
+/* Cumulative memory size in Byte */
+#define STATS_DESC_SIZE_BYTE(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
+		KVM_STATS_BASE_POW2, 0)
+/* Cumulative memory size in KiByte */
+#define STATS_DESC_SIZE_KBYTE(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
+		KVM_STATS_BASE_POW2, 10)
+/* Cumulative memory size in MiByte */
+#define STATS_DESC_SIZE_MBYTE(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
+		KVM_STATS_BASE_POW2, 20)
+/* Cumulative memory size in GiByte */
+#define STATS_DESC_SIZE_GBYTE(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_BYTES,	       \
+		KVM_STATS_BASE_POW2, 30)
+
+/* Instantaneous memory size in Byte */
+#define STATS_DESC_ISIZE_BYTE(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
+		KVM_STATS_BASE_POW2, 0)
+/* Instantaneous memory size in KiByte */
+#define STATS_DESC_ISIZE_KBYTE(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
+		KVM_STATS_BASE_POW2, 10)
+/* Instantaneous memory size in MiByte */
+#define STATS_DESC_ISIZE_MBYTE(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
+		KVM_STATS_BASE_POW2, 20)
+/* Instantaneous memory size in GiByte */
+#define STATS_DESC_ISIZE_GBYTE(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BYTES,		       \
+		KVM_STATS_BASE_POW2, 30)
+
+/* Cumulative time in second */
+#define STATS_DESC_TIME_SEC(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
+		KVM_STATS_BASE_POW10, 0)
+/* Cumulative time in millisecond */
+#define STATS_DESC_TIME_MSEC(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
+		KVM_STATS_BASE_POW10, -3)
+/* Cumulative time in microsecond */
+#define STATS_DESC_TIME_USEC(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
+		KVM_STATS_BASE_POW10, -6)
+/* Cumulative time in nanosecond */
+#define STATS_DESC_TIME_NSEC(SCOPE, name)				       \
+	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
+		KVM_STATS_BASE_POW10, -9)
+
+/* Instantaneous time in second */
+#define STATS_DESC_ITIME_SEC(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
+		KVM_STATS_BASE_POW10, 0)
+/* Instantaneous time in millisecond */
+#define STATS_DESC_ITIME_MSEC(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
+		KVM_STATS_BASE_POW10, -3)
+/* Instantaneous time in microsecond */
+#define STATS_DESC_ITIME_USEC(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
+		KVM_STATS_BASE_POW10, -6)
+/* Instantaneous time in nanosecond */
+#define STATS_DESC_ITIME_NSEC(SCOPE, name)				       \
+	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_SECONDS,		       \
+		KVM_STATS_BASE_POW10, -9)
+
 extern struct kvm_stats_debugfs_item debugfs_entries[];
 extern struct dentry *kvm_debugfs_dir;
+ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
+		struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
+		char __user *user_buffer, size_t size, loff_t *offset);
 
 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index 9febe1412f7a..ab73e905105c 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -1086,6 +1086,7 @@  struct kvm_ppc_resize_hpt {
 #define KVM_CAP_HYPERV_ENFORCE_CPUID 199
 #define KVM_CAP_SREGS2 200
 #define KVM_CAP_EXIT_HYPERCALL 201
+#define KVM_CAP_BINARY_STATS_FD 202
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1905,4 +1906,45 @@  struct kvm_dirty_gfn {
 #define KVM_BUS_LOCK_DETECTION_OFF             (1 << 0)
 #define KVM_BUS_LOCK_DETECTION_EXIT            (1 << 1)
 
+#define KVM_STATS_ID_MAXLEN		64
+
+struct kvm_stats_header {
+	__u32 name_size;
+	__u32 count;
+	__u32 desc_offset;
+	__u32 data_offset;
+	char id[];
+};
+
+#define KVM_STATS_TYPE_SHIFT		0
+#define KVM_STATS_TYPE_MASK		(0xF << KVM_STATS_TYPE_SHIFT)
+#define KVM_STATS_TYPE_CUMULATIVE	(0x0 << KVM_STATS_TYPE_SHIFT)
+#define KVM_STATS_TYPE_INSTANT		(0x1 << KVM_STATS_TYPE_SHIFT)
+#define KVM_STATS_TYPE_MAX		KVM_STATS_TYPE_INSTANT
+
+#define KVM_STATS_UNIT_SHIFT		4
+#define KVM_STATS_UNIT_MASK		(0xF << KVM_STATS_UNIT_SHIFT)
+#define KVM_STATS_UNIT_NONE		(0x0 << KVM_STATS_UNIT_SHIFT)
+#define KVM_STATS_UNIT_BYTES		(0x1 << KVM_STATS_UNIT_SHIFT)
+#define KVM_STATS_UNIT_SECONDS		(0x2 << KVM_STATS_UNIT_SHIFT)
+#define KVM_STATS_UNIT_CYCLES		(0x3 << KVM_STATS_UNIT_SHIFT)
+#define KVM_STATS_UNIT_MAX		KVM_STATS_UNIT_CYCLES
+
+#define KVM_STATS_BASE_SHIFT		8
+#define KVM_STATS_BASE_MASK		(0xF << KVM_STATS_BASE_SHIFT)
+#define KVM_STATS_BASE_POW10		(0x0 << KVM_STATS_BASE_SHIFT)
+#define KVM_STATS_BASE_POW2		(0x1 << KVM_STATS_BASE_SHIFT)
+#define KVM_STATS_BASE_MAX		KVM_STATS_BASE_POW2
+
+struct kvm_stats_desc {
+	__u32 flags;
+	__s16 exponent;
+	__u16 size;
+	__u32 offset;
+	__u32 unused;
+	char name[];
+};
+
+#define KVM_GET_STATS_FD  _IO(KVMIO,  0xce)
+
 #endif /* __LINUX_KVM_H */
diff --git a/virt/kvm/binary_stats.c b/virt/kvm/binary_stats.c
new file mode 100644
index 000000000000..1517e91d295b
--- /dev/null
+++ b/virt/kvm/binary_stats.c
@@ -0,0 +1,130 @@ 
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * KVM binary statistics interface implementation
+ *
+ * Copyright 2021 Google LLC
+ */
+
+#include <linux/kvm_host.h>
+#include <linux/kvm.h>
+#include <linux/errno.h>
+#include <linux/uaccess.h>
+
+/*
+ * Common vm/vcpu stats read function to userspace.
+ * @id: identification string of the stats
+ * @header: stats header for a vm or a vcpu
+ * @desc: start address of an array of stats descriptors for a vm or a vcpu
+ * @stats: start address of stats data block for a vm or a vcpu
+ * @size_stats: the size of stats data block pointed by @stats
+ * @user_buffer: start address of userspace buffer
+ * @size: requested read size from userspace
+ * @offset: the start position from which the content will be read for the
+ *          corresponding vm or vcp file descriptor
+ *
+ * The file content of a vm/vcpu file descriptor is now defined as below:
+ * +-------------+
+ * |   Header    |
+ * +-------------+
+ * | Descriptors |
+ * +-------------+
+ * | Stats Data  |
+ * +-------------+
+ * Although this function allows userspace to read any amount of data (as long
+ * as in the limit) from any position, the typical usage would follow below
+ * steps:
+ * 1. Read header from offset 0. Get the offset of descriptors and stats data
+ *    and some other necessary information. This is a one-time work for the
+ *    lifecycle of the corresponding vm/vcpu stats fd.
+ * 2. Read descriptors from its offset and discover all the stats by parsing
+ *    descriptors. This is a one-time work for the lifecycle of the
+ *    corresponding vm/vcpu stats fd.
+ * 3. Periodically read stats data from its offset.
+ */
+ssize_t kvm_stats_read(char *id, struct kvm_stats_header *header,
+		struct _kvm_stats_desc *desc, void *stats, size_t size_stats,
+		char __user *user_buffer, size_t size, loff_t *offset)
+{
+	ssize_t len;
+	ssize_t copylen;
+	ssize_t remain = size;
+	size_t size_desc;
+	size_t size_header;
+	void *src;
+	loff_t pos = *offset;
+	char __user *dest = user_buffer;
+
+	size_header = sizeof(*header);
+	size_desc = header->count * sizeof(*desc);
+
+	len = KVM_STATS_ID_MAXLEN + size_header + size_desc + size_stats - pos;
+	len = min(len, remain);
+	if (len <= 0)
+		return 0;
+	remain = len;
+
+	/* Copy kvm stats header.
+	 * The header is the first block of content userspace usually read out.
+	 * The pos is 0 and the copylen and remain would be the size of header.
+	 * The copy of the header would be skipped if offset is larger than the
+	 * size of header. That usually happens when userspace reads stats
+	 * descriptors and stats data.
+	 */
+	copylen = size_header - pos;
+	copylen = min(copylen, remain);
+	if (copylen > 0) {
+		src = (void *)header + pos;
+		if (copy_to_user(dest, src, copylen))
+			return -EFAULT;
+		remain -= copylen;
+		pos += copylen;
+		dest += copylen;
+	}
+
+	/* Copy kvm stats header id string.
+	 * The id string is unique for every vm/vcpu, which is stored in kvm
+	 * and kvm_vcpu structure.
+	 */
+	copylen = size_header + KVM_STATS_ID_MAXLEN - pos;
+	copylen = min(copylen, remain);
+	if (copylen > 0) {
+		src = id + pos - size_header;
+		if (copy_to_user(dest, src, copylen))
+			return -EFAULT;
+		remain -= copylen;
+		pos += copylen;
+		dest += copylen;
+	}
+
+	/* Copy kvm stats descriptors.
+	 * The descriptors copy would be skipped in the typical case that
+	 * userspace periodically read stats data, since the pos would be
+	 * greater than the end address of descriptors
+	 * (header->header.desc_offset + size_desc) causing copylen <= 0.
+	 */
+	copylen = header->desc_offset + size_desc - pos;
+	copylen = min(copylen, remain);
+	if (copylen > 0) {
+		src = (void *)desc + pos - header->desc_offset;
+		if (copy_to_user(dest, src, copylen))
+			return -EFAULT;
+		remain -= copylen;
+		pos += copylen;
+		dest += copylen;
+	}
+
+	/* Copy kvm stats values */
+	copylen = header->data_offset + size_stats - pos;
+	copylen = min(copylen, remain);
+	if (copylen > 0) {
+		src = stats + pos - header->data_offset;
+		if (copy_to_user(dest, src, copylen))
+			return -EFAULT;
+		remain -= copylen;
+		pos += copylen;
+		dest += copylen;
+	}
+
+	*offset = pos;
+	return len;
+}