[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH RFC 5/6] x86/time: implement PVCLOCK_TSC_STABLE_BIT

When using TSC as clocksource we will solely rely on TSC for updating
vcpu time infos (pvti). Right now, each vCPU takes the tsc_timestamp at
different instants meaning every EPOCH + delta. This delta is variable
depending on the time the CPU calibrates with CPU 0 (master), and will
likely be different and variable across vCPUS. This means that each VCPU
pvti won't account to its calibration error which could lead to time
going backwards, and allowing a situation where time read on VCPU B
immediately after A being smaller. While this doesn't happen a lot, I
was able to observe (for clocksource=tsc) around 50 times in an hour
having warps of < 100 ns.

This patch proposes relying on host TSC synchronization and passthrough
of the master tsc to the guest, when running on a TSC-safe platform. On
the rendezvous function we will retrieve the platform time in ns and the
last count read by the clocksource that was used to compute system time.
master will write both master_tsc_stamp and master_stime, and the other
vCPUS (slave) will use it to update their correspondent time infos.
This way we can guarantee that on a platform with a constant and
reliable TSC, that the time read on vcpu B right after A is bigger
independently of the VCPU calibration error. Since pvclock time infos
are monotonic as seen by any vCPU we set PVCLOCK_TSC_STABLE_BIT flag,
which then enables usage of VDSO on Linux. IIUC, this is similar to how
it's implemented on KVM.

Signed-off-by: Joao Martins <joao.m.martins@xxxxxxxxxx>
 xen/arch/x86/time.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 3f96ce6..e623891 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -910,6 +910,8 @@ static void __update_vcpu_system_time(struct vcpu *v, int 
     _u.tsc_timestamp = tsc_stamp;
     _u.system_time   = t->stime_local_stamp;
+    if ( clocksource_is_tsc )
+        _u.flags    |= PVCLOCK_TSC_STABLE_BIT;
     if ( is_hvm_domain(d) )
         _u.tsc_timestamp += v->arch.hvm_vcpu.cache_tsc_offset;
@@ -1370,9 +1372,12 @@ static void time_calibration_std_rendezvous(void *_r)
     if ( smp_processor_id() == 0 )
+        u64 last_counter;
         while ( atomic_read(&r->semaphore) != (total_cpus - 1) )
-        r->master_stime = read_platform_stime();
+        r->master_stime = read_platform_stime(&last_counter);
+        if ( clocksource_is_tsc )
+            r->master_tsc_stamp = last_counter;
         mb(); /* write r->master_stime /then/ signal */
@@ -1384,7 +1389,10 @@ static void time_calibration_std_rendezvous(void *_r)
         mb(); /* receive signal /then/ read r->master_stime */
-    c->local_tsc_stamp = rdtsc();
+    if ( clocksource_is_tsc )
+        c->local_tsc_stamp = r->master_tsc_stamp;
+    else
+        c->local_tsc_stamp = rdtsc();
     c->stime_local_stamp = get_s_time();
     c->stime_master_stamp = r->master_stime;

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.