KVM virtual PTP driver (v4)

This patchset implements a virtual PTP driver which allows guest to sync its clock to the host clock with high precision
(error is < 1us on an idle guest).

Changelog from v3

Patch3:

v2: export kvmclock clocksource structure directly (Paolo)

Patch4:

v2: drop timekeeper spinlock, move back to drivers/ptp/ptp_chardev.c (Paolo)
ptp_clock_gettime: support drivers with crosstimestamp but not gettime64 callbacks (Paolo)

Patch5:

v4: remove gettime64 callback (Paolo)

Changelog from v2

Cross timestamping, emulation of PTP_SYS_OFFSET with cross timestamping (Paolo, Miroslav, Radim).

Changelog from v1

Patch1:
v2: unify nsec_base (Radim)

Patch2:
v2: improve documentation (Radim)
change hypercall name to KVM_HC_CLOCK_PAIRING (Radim) increase padding size

Patch3:
v2: check for kvmclock (Radim)
initialize global variables before device registration (Radim)

PHC0 0 3 377 7 -141ns[ +234ns] +/- 7ns

#* PHC0                          0   3   377     5   -197ns[ -403ns] +/-    3ns
#* PHC0                          0   3   377    11   +157ns[ +264ns] +/-    3ns
#* PHC0                          0   3   377     9    +92ns[ +304ns] +/-    2ns
#* PHC0                          0   3   377     7   -102ns[ -164ns] +/-    3ns
#* PHC0                          0   3   377     5   -157ns[ -269ns] +/-    2ns
#* PHC0                          0   3   377    10    -63ns[  -98ns] +/-    2ns
#* PHC0                          0   3   377     8   +351ns[ +399ns] +/-    2ns
#* PHC0                          0   3   377     6   +385ns[ +399ns] +/-    4ns
#* PHC0                          0   3   377     4   -133ns[ -110ns] +/-    6ns
#* PHC0                          0   3   377    10    -59ns[  -90ns] +/-    3ns
#* PHC0                          0   3   377     8    -90ns[ +381ns] +/-    9ns
#* PHC0                          0   3   377     6     +6ns[ +216ns] +/-    6ns
#* PHC0                          0   3   377     4   +166ns[ -666ns] +/-   11ns
#* PHC0                          0   3   377    10    -18ns[ +323ns] +/-   10ns
#* PHC0                          0   3   377     8    -12ns[ +121ns] +/-    5ns
#* PHC0                          0   3   377     5     +4ns[ +218ns] +/-    7ns
#* PHC0                          0   3   377     4   +162ns[ -683ns] +/-   11ns
#* PHC0                          0   3   377    10    -82ns[ +310ns] +/-   12ns
#* PHC0                          0   3   377     7     +5ns[ -320ns] +/-    9ns
#* PHC0                          0   3   377     5    -13ns[ +165ns] +/-    7ns
#* PHC0                          0   3   377     3     +6ns[ +105ns] +/-    2ns
#* PHC0                          0   3   377     9    -19ns[  -67ns] +/-    3ns
#* PHC0                          0   3   377     8    +89ns[ +181ns] +/-    3ns
#* PHC0                          0   3   377     6    +93ns[ +168ns] +/-    2ns
#* PHC0                          0   3   377     4   +100ns[ +154ns] +/-    2ns
#* PHC0                          0   3   377    10   -249ns[ +180ns] +/-    7ns
#* PHC0                          0   3   377     8   +158ns[ -582ns] +/-   22ns
#* PHC0                          0   3   377     6    +10ns[ +213ns] +/-    9ns
#* PHC0                          0   3   377     3    -35ns[ +258ns] +/-    5ns
#* PHC0                          0   3   377    10    +25ns[ -620ns] +/-    8ns
#* PHC0                          0   3   377     7    +13ns[ +153ns] +/-    2ns

Expose the realtime host clock and save the TSC value used for the clock calculation.

---
 arch/x86/kvm/x86.c |   36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

v2: unify nsec_base (Radim)

Index: kvm-ptpdriver/arch/x86/kvm/x86.c
===================================================================
@@ -1139,6 +1139,7 @@
 
 	u64		boot_ns;
 	u64		nsec_base;
+	u64		wall_time_sec;
 };
 
 static struct pvclock_gtod_data pvclock_gtod_data;

@@ -1162,6 +1163,8 @@
 	vdata->boot_ns			= boot_ns;
 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
 
+	vdata->wall_time_sec            = tk->xtime_sec;
+
 	write_seqcount_end(&vdata->seq);
 }
 #endif

@@ -1623,6 +1626,28 @@
 	return mode;
 }
 
+static int do_realtime(struct timespec *ts, cycle_t *cycle_now)
+{
+	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
+	unsigned long seq;
+	int mode;
+	u64 ns;
+
+	do {
+		seq = read_seqcount_begin(&gtod->seq);
+		mode = gtod->clock.vclock_mode;
+		ts->tv_sec = gtod->wall_time_sec;
+		ns = gtod->nsec_base;
+		ns += vgettsc(cycle_now);
+		ns >>= gtod->clock.shift;
+	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
+
+	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
+	ts->tv_nsec = ns;
+
+	return mode;
+}
+
 /* returns true if host is using tsc clocksource */
 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
 {

@@ -1632,6 +1657,17 @@
 
 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
 }
+
+/* returns true if host is using tsc clocksource */
+static bool kvm_get_walltime_and_clockread(struct timespec *ts,
+					   cycle_t *cycle_now)
+{
+	/* checked again under seqlock below */
+	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
+		return false;
+
+	return do_realtime(ts, cycle_now) == VCLOCK_TSC;
+}
 #endif
 
 /*

Add a hypercall to retrieve the host realtime clock and the TSC value used to calculate that clock read.

Used to implement clock synchronization between
host and guest.

---
 Documentation/virtual/kvm/hypercalls.txt |   33 ++++++++++++++++++++++++
 arch/x86/include/uapi/asm/kvm_para.h     |    9 ++++++
 arch/x86/kvm/x86.c                       |   41 +++++++++++++++++++++++++++++++
 include/uapi/linux/kvm_para.h            |    3 ++
 4 files changed, 86 insertions(+)

v2: improve documentation (Radim)
change hypercall name to KVM_HC_CLOCK_PAIRING (Radim) increase padding size

Index: kvm-ptpdriver/arch/x86/include/uapi/asm/kvm_para.h
===================================================================
@@ -50,6 +50,15 @@
 	__u32 pad[11];
 };
 
+#define KVM_CLOCK_PAIRING_WALLCLOCK 0
+struct kvm_clock_offset {
+	__s64 sec;
+	__s64 nsec;
+	__u64 tsc;
+	__u32 flags;
+	__u32 pad[9];
+};
+
 #define KVM_STEAL_ALIGNMENT_BITS 5
 #define KVM_STEAL_VALID_BITS ((-1ULL << (KVM_STEAL_ALIGNMENT_BITS + 1)))
 #define KVM_STEAL_RESERVED_MASK (((1 << KVM_STEAL_ALIGNMENT_BITS) - 1 ) << 1)

Index: kvm-ptpdriver/Documentation/virtual/kvm/hypercalls.txt
===================================================================
@@ -81,3 +81,36 @@
 same guest can wakeup the sleeping vcpu by issuing KVM_HC_KICK_CPU hypercall,
 specifying APIC ID (a1) of the vcpu to be woken up. An additional argument (a0)
 is used in the hypercall for future use.
+
+
+6. KVM_HC_CLOCK_PAIRING
+------------------------
+Architecture: x86
+Status: active
+Purpose: Hypercall used to synchronize host and guest clocks.
+Usage:
+
+a0: guest physical address where host copies
+"struct kvm_clock_offset" structure.
+
+a1: clock_type, ATM only KVM_CLOCK_PAIRING_WALLCLOCK (0)
+is supported (hosts CLOCK_REALTIME clock).
+
+		struct kvm_clock_offset {
+			__s64 sec;
+			__s64 nsec;
+			__u64 tsc;
+			__u32 flags;
+			__u32 pad[9];
+		};
+
+       Where:
+               * sec: seconds from clock_type clock.
+               * nsec: nanoseconds from clock_type clock.
+               * tsc: TSC value used to calculate sec/nsec pair
+                 (this hypercall only works when host uses TSC clocksource).
+               * flags: flags, unused (0) at the moment.
+
+Returns KVM_EOPNOTSUPP if the host does not use TSC clocksource,
+or if clock type is different than KVM_CLOCK_PAIRING_WALLCLOCK.
+

Index: kvm-ptpdriver/include/uapi/linux/kvm_para.h
===================================================================
@@ -14,6 +14,8 @@
 #define KVM_EFAULT		EFAULT
 #define KVM_E2BIG		E2BIG
 #define KVM_EPERM		EPERM
+#define KVM_EOPNOTSUPP		EOPNOTSUPP
+#define KVM_ENOMEM		ENOMEM
 
 #define KVM_HC_VAPIC_POLL_IRQ		1
 #define KVM_HC_MMU_OP			2

@@ -23,6 +25,7 @@
 #define KVM_HC_MIPS_GET_CLOCK_FREQ	6
 #define KVM_HC_MIPS_EXIT_VM		7
 #define KVM_HC_MIPS_CONSOLE_OUTPUT	8
+#define KVM_HC_CLOCK_PAIRING		9
 
 /*
  * hypercalls use architecture specific

To be used by KVM PTP driver.

---
 arch/x86/include/asm/kvmclock.h |    6 ++++++
 arch/x86/kernel/kvmclock.c      |    3 ++-
 2 files changed, 8 insertions(+), 1 deletion(-)

v2: export kvmclock clocksource structure directly (Paolo)

Index: kvm-ptpdriver/arch/x86/include/asm/kvmclock.h
===================================================================
@@ -0,0 +1,6 @@
+#ifndef _ASM_X86_KVM_CLOCK_H
+#define _ASM_X86_KVM_CLOCK_H
+
+extern struct clocksource kvm_clock;
+
+#endif /* _ASM_X86_KVM_CLOCK_H */

Index: kvm-ptpdriver/arch/x86/kernel/kvmclock.c
===================================================================
@@ -28,6 +28,7 @@
 
 #include <asm/x86_init.h>
 #include <asm/reboot.h>
+#include <asm/kvmclock.h>
 
 static int kvmclock __ro_after_init = 1;
 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;

@@ -174,7 +175,7 @@
 	return ret;
 }
 
-static struct clocksource kvm_clock = {
+struct clocksource kvm_clock = {
 	.name = "kvm-clock",
 	.read = kvm_clock_get_cycles,
 	.rating = 400,

Emulate PTP_SYS_OFFSET by using an arithmetic mean of the realtime samples from ->getcrosststamp callback.

Note: if the realtime clock changes during cross timestamp sampling, then the mean will return a value that does not exist. Given the nature of PTP_SYS_OFFSET users (time synchronization), that should not be a problem.

---
 drivers/ptp/ptp_chardev.c |   73 ++++++++++++++++++++++++++++++++++++++++++++++
 drivers/ptp/ptp_clock.c   |   10 ++++++
 2 files changed, 83 insertions(+)

v2: drop timekeeper spinlock, move back to drivers/ptp/ptp_chardev.c (Paolo)
ptp_clock_gettime: support drivers with crosstimestamp but not gettime64 callbacks

Index: kvm-ptpdriver/drivers/ptp/ptp_chardev.c
===================================================================
@@ -116,6 +116,74 @@
 	return 0;
 }
 
+static void mean_sys_realtime(struct system_device_crosststamp *xtstamp,
+			      struct system_device_crosststamp *n_xtstamp,
+			      struct ptp_clock_time *pct)
+{
+	ktime_t sys_realtime, n_sys_realtime;
+	struct timespec ts;
+	u64 ns;
+
+	/* estimate realtime with arithmetic mean of two crosststamp samples */
+	sys_realtime = xtstamp->sys_realtime;
+	n_sys_realtime = n_xtstamp->sys_realtime;
+	ns = ktime_divns(ktime_add(sys_realtime, n_sys_realtime), 2);
+	ts = ktime_to_timespec(ns_to_ktime(ns));
+	pct->sec = ts.tv_sec;
+	pct->nsec = ts.tv_nsec;
+}
+
+static int emulate_ptp_sys_offset(struct ptp_clock_info *info,
+				  struct ptp_sys_offset *sysoff,
+				  unsigned long arg)
+{
+	int i, err, len;
+	int n_samples;
+	struct system_device_crosststamp *xtstamps, *xtstamp, *n_xtstamp;
+	struct ptp_clock_time *pct;
+
+	n_samples = sysoff->n_samples + 2;
+
+	len = sizeof(struct system_device_crosststamp);
+	xtstamps = kzalloc(len * n_samples, GFP_KERNEL);
+	if (xtstamps == NULL)
+		return -ENOMEM;
+
+	xtstamp = xtstamps;
+	for (i = 0; i < n_samples; i++) {
+		err = info->getcrosststamp(info, xtstamp);
+		if (err) {
+			kfree(xtstamps);
+			return err;
+		}
+		xtstamp++;
+	}
+
+	pct = &sysoff->ts[0];
+	xtstamp = n_xtstamp = xtstamps;
+	n_xtstamp++;
+	for (i = 0; i < sysoff->n_samples; i++) {
+		struct timespec ts;
+
+		mean_sys_realtime(xtstamp, n_xtstamp, pct);
+		pct++;
+
+		ts = ktime_to_timespec(n_xtstamp->device);
+		pct->sec = ts.tv_sec;
+		pct->nsec = ts.tv_nsec;
+		pct++;
+		xtstamp++;
+		n_xtstamp++;
+	}
+
+	mean_sys_realtime(xtstamp, n_xtstamp, pct);
+	kfree(xtstamps);
+	if (copy_to_user((void __user *)arg, sysoff, sizeof(*sysoff)))
+		return -EFAULT;
+
+	return 0;
+}
+
 long ptp_ioctl(struct posix_clock *pc, unsigned int cmd, unsigned long arg)
 {
 	struct ptp_clock_caps caps;

@@ -219,6 +287,11 @@
 			err = -EINVAL;
 			break;
 		}
+
+		if (!ptp->info->gettime64 && ptp->info->getcrosststamp) {
+			err = emulate_ptp_sys_offset(ptp->info, sysoff, arg);
+			break;
+		}
 		pct = &sysoff->ts[0];
 		for (i = 0; i < sysoff->n_samples; i++) {
 			getnstimeofday64(&ts);

Index: kvm-ptpdriver/drivers/ptp/ptp_clock.c
===================================================================
@@ -118,6 +118,16 @@
 	struct timespec64 ts;
 	int err;
 
+	if (!ptp->info->gettime64 && ptp->info->getcrosststamp) {
+		struct system_device_crosststamp xtstamp;
+
+		err = ptp->info->getcrosststamp(ptp->info, &xtstamp);
+		if (err)
+			return err;
+		*tp = ktime_to_timespec(xtstamp.device);
+		return err;
+	}
+
 	err = ptp->info->gettime64(ptp->info, &ts);
 	if (!err)
 		*tp = timespec64_to_timespec(ts);

Add a driver with gettime method returning hosts realtime clock. This allows Chrony to synchronize host and guest clocks with high precision (see results below).

chronyc> sources
MS Name/IP address Stratum Poll Reach LastRx Last sample
===============================================================================
#* PHC0 0 3 377 4 +162ns[ -683ns] +/- 11ns

To configure Chronyd to use PHC refclock, add the
following line to its configuration file:

refclock PHC /dev/ptpX poll 3 dpoll -2 offset 0

Where /dev/ptpX is the kvmclock PTP clock.

---
 drivers/ptp/Kconfig   |   12 +++
 drivers/ptp/Makefile  |    1 
 drivers/ptp/ptp_kvm.c |  179 ++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 192 insertions(+)

v2: check for kvmclock (Radim)
initialize global variables before device registration (Radim)
v3: use cross timestamps callback (Paolo, Miroslav, Radim)
v4: remove gettime64 callback (Paolo)

Index: kvm-ptpdriver/drivers/ptp/Kconfig
===================================================================
@@ -90,4 +90,16 @@
 	  To compile this driver as a module, choose M here: the module
 	  will be called ptp_pch.
 
+config PTP_1588_CLOCK_KVM
+	tristate "KVM virtual PTP clock"
+	depends on PTP_1588_CLOCK
+	depends on KVM_GUEST
+	default y
+	help
+	  This driver adds support for using kvm infrastructure as a PTP
+	  clock. This clock is only useful if you are using KVM guests.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called ptp_kvm.
+
 endmenu

Index: kvm-ptpdriver/drivers/ptp/Makefile
===================================================================
@@ -6,3 +6,4 @@
 obj-$(CONFIG_PTP_1588_CLOCK)		+= ptp.o
 obj-$(CONFIG_PTP_1588_CLOCK_IXP46X)	+= ptp_ixp46x.o
 obj-$(CONFIG_PTP_1588_CLOCK_PCH)	+= ptp_pch.o
+obj-$(CONFIG_PTP_1588_CLOCK_KVM)	+= ptp_kvm.o

Add a driver with gettime method returning hosts realtime clock. This allows Chrony to synchronize host and guest clocks with high precision (see results below).

chronyc> sources
MS Name/IP address Stratum Poll Reach LastRx Last sample
===============================================================================
#* PHC0 0 3 377 4 +162ns[ -683ns] +/- 11ns

To configure Chronyd to use PHC refclock, add the
following line to its configuration file:

refclock PHC /dev/ptpX poll 3 dpoll -2 offset 0

Where /dev/ptpX is the kvmclock PTP clock.

---
 drivers/ptp/Kconfig   |   12 +++
 drivers/ptp/Makefile  |    1 
 drivers/ptp/ptp_kvm.c |  177 ++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 190 insertions(+)

v2: check for kvmclock (Radim)
initialize global variables before device registration (Radim)
v3: use cross timestamps callback (Paolo, Miroslav, Radim)
v4: remove gettime64 callback (Paolo)
v5: drop useless nsec assignment (Radim)

Index: kvm-ptpdriver/drivers/ptp/Kconfig
===================================================================
@@ -90,4 +90,16 @@
 	  To compile this driver as a module, choose M here: the module
 	  will be called ptp_pch.
 
+config PTP_1588_CLOCK_KVM
+	tristate "KVM virtual PTP clock"
+	depends on PTP_1588_CLOCK
+	depends on KVM_GUEST
+	default y
+	help
+	  This driver adds support for using kvm infrastructure as a PTP
+	  clock. This clock is only useful if you are using KVM guests.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called ptp_kvm.
+
 endmenu

Index: kvm-ptpdriver/drivers/ptp/Makefile
===================================================================
@@ -6,3 +6,4 @@
 obj-$(CONFIG_PTP_1588_CLOCK)		+= ptp.o
 obj-$(CONFIG_PTP_1588_CLOCK_IXP46X)	+= ptp_ixp46x.o
 obj-$(CONFIG_PTP_1588_CLOCK_PCH)	+= ptp_pch.o
+obj-$(CONFIG_PTP_1588_CLOCK_KVM)	+= ptp_kvm.o