hvm pmtimer: correct pmtimer accuracy

Several seconds of backward time drift per minute can be seen on a
RHEL6 HVM guest by switching the clocksource to 'acpi_pm' and then
running gettimeofday() in a loop. This is due to the accumulation
of small inaccuracies that are caused by shifting out the lower 32
bits when pmt_update_time() computes 'tmr_val'.

The patch makes sure that the lower 32 bits of the computed value
are not lost. They are saved in a new field 'not_accounted' in the
PMTState structure and are accounted the next time pmt_update_time()
is called.

From: Ulrich Obergfell <uobergfe@redhat.com>
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
xen-unstable changeset:   22174:632c02167f97
xen-unstable date:        Fri Sep 17 16:59:21 2010 +0100

diff --git a/xen/arch/x86/hvm/pmtimer.c b/xen/arch/x86/hvm/pmtimer.c
index 48fe26ac87b3514cf324de9f2af595cd7dc4009d..1b9ab7ba18fad02c5bd182e9e44d64e32cafbce7 100644 (file)
--- a/xen/arch/x86/hvm/pmtimer.c
+++ b/xen/arch/x86/hvm/pmtimer.c
@@ -83,14 +83,16 @@ void hvm_acpi_sleep_button(struct domain *d)
  * since the last time we did that. */
 static void pmt_update_time(PMTState *s)
 {
-    uint64_t curr_gtime;
+    uint64_t curr_gtime, tmp;
     uint32_t msb = s->pm.tmr_val & TMR_VAL_MSB;
     
     ASSERT(spin_is_locked(&s->lock));
 
     /* Update the timer */
     curr_gtime = hvm_get_guest_time(s->vcpu);
-    s->pm.tmr_val += ((curr_gtime - s->last_gtime) * s->scale) >> 32;
+    tmp = ((curr_gtime - s->last_gtime) * s->scale) + s->not_accounted;
+    s->not_accounted = (uint32_t)tmp;
+    s->pm.tmr_val += tmp >> 32;
     s->pm.tmr_val &= TMR_VAL_MASK;
     s->last_gtime = curr_gtime;
     
@@ -257,6 +259,7 @@ static int pmtimer_load(struct domain *d, hvm_domain_context_t *h)
 
     /* Calculate future counter values from now. */
     s->last_gtime = hvm_get_guest_time(s->vcpu);
+    s->not_accounted = 0;
 
     /* Set the SCI state from the registers */ 
     pmt_update_sci(s);
@@ -276,6 +279,7 @@ void pmtimer_init(struct vcpu *v)
     spin_lock_init(&s->lock);
 
     s->scale = ((uint64_t)FREQUENCE_PMTIMER << 32) / SYSTEM_TIME_HZ;
+    s->not_accounted = 0;
     s->vcpu = v;
 
     /* Intercept port I/O (need two handlers because PM1a_CNT is between

diff --git a/xen/include/asm-x86/hvm/vpt.h b/xen/include/asm-x86/hvm/vpt.h
index 6dcac7732860dc15bdc5925e0b1cba4d96f23475..65b0dffb98d2bd998c84adfcfeb9794d6f426ea8 100644 (file)
--- a/xen/include/asm-x86/hvm/vpt.h
+++ b/xen/include/asm-x86/hvm/vpt.h
@@ -117,6 +117,7 @@ typedef struct PMTState {
     struct hvm_hw_pmtimer pm;   /* 32bit timer value */
     struct vcpu *vcpu;          /* Keeps sync with this vcpu's guest-time */
     uint64_t last_gtime;        /* Last (guest) time we updated the timer */
+    uint32_t not_accounted;     /* time not accounted at last update */
     uint64_t scale;             /* Multiplier to get from tsc to timer ticks */
     struct timer timer;         /* To make sure we send SCIs */
     spinlock_t lock;