Merge with xen-ia64-unstable

xen/arch/x86/hvm/hpet.c

@@ -113,6 +113,8 @@ static inline int hpet_check_access_length(
 
 static inline uint64_t hpet_read_maincounter(HPETState *h)
 {
+    ASSERT(spin_is_locked(&h->lock));
+
     if ( hpet_enabled(h) )
         return guest_time_hpet(h->vcpu) + h->mc_offset;
     else 
@@ -131,6 +133,8 @@ static unsigned long hpet_read(
     if ( hpet_check_access_length(addr, length) != 0 )
         return ~0UL;
 
+    spin_lock(&h->lock);
+
     val = hpet_read64(h, addr & ~7);
     if ( (addr & ~7) == HPET_COUNTER )
         val = hpet_read_maincounter(h);
@@ -139,12 +143,15 @@ static unsigned long hpet_read(
     if ( length != 8 )
         result = (val >> ((addr & 7) * 8)) & ((1UL << (length * 8)) - 1);
 
+    spin_unlock(&h->lock);
+
     return result;
 }
 
 static void hpet_stop_timer(HPETState *h, unsigned int tn)
 {
     ASSERT(tn < HPET_TIMER_NUM);
+    ASSERT(spin_is_locked(&h->lock));
     stop_timer(&h->timers[tn]);
 }
 
@@ -157,7 +164,8 @@ static void hpet_set_timer(HPETState *h, unsigned int tn)
     uint64_t tn_cmp, cur_tick, diff;
 
     ASSERT(tn < HPET_TIMER_NUM);
-    
+    ASSERT(spin_is_locked(&h->lock));
+
     if ( !hpet_enabled(h) || !timer_enabled(h, tn) )
         return;
 
@@ -213,6 +221,8 @@ static void hpet_write(
     if ( hpet_check_access_length(addr, length) != 0 )
         return;
 
+    spin_lock(&h->lock);
+
     old_val = hpet_read64(h, addr & ~7);
     if ( (addr & ~7) == HPET_COUNTER )
         old_val = hpet_read_maincounter(h);
@@ -302,6 +312,8 @@ static void hpet_write(
         /* Ignore writes to unsupported and reserved registers. */
         break;
     }
+
+    spin_unlock(&h->lock);
 }
 
 static int hpet_range(struct vcpu *v, unsigned long addr)
@@ -321,6 +333,8 @@ static void hpet_route_interrupt(HPETState *h, unsigned int tn)
     unsigned int tn_int_route = timer_int_route(h, tn);
     struct domain *d = h->vcpu->domain;
 
+    ASSERT(spin_is_locked(&h->lock));
+
     if ( (tn <= 1) && (h->hpet.config & HPET_CFG_LEGACY) )
     {
         /* if LegacyReplacementRoute bit is set, HPET specification requires
@@ -352,8 +366,13 @@ static void hpet_timer_fn(void *opaque)
     HPETState *h = htfi->hs;
     unsigned int tn = htfi->tn;
 
+    spin_lock(&h->lock);
+
     if ( !hpet_enabled(h) || !timer_enabled(h, tn) )
+    {
+        spin_unlock(&h->lock);
         return;
+    }
 
     hpet_route_interrupt(h, tn);
 
@@ -374,6 +393,8 @@ static void hpet_timer_fn(void *opaque)
         set_timer(&h->timers[tn], 
                   NOW() + hpet_tick_to_ns(h, h->hpet.period[tn]));
     }
+
+    spin_unlock(&h->lock);
 }
 
 void hpet_migrate_timers(struct vcpu *v)
@@ -391,12 +412,19 @@ void hpet_migrate_timers(struct vcpu *v)
 static int hpet_save(struct domain *d, hvm_domain_context_t *h)
 {
     HPETState *hp = &d->arch.hvm_domain.pl_time.vhpet;
+    int rc;
+
+    spin_lock(&hp->lock);
 
     /* Write the proper value into the main counter */
     hp->hpet.mc64 = hp->mc_offset + guest_time_hpet(hp->vcpu);
 
     /* Save the HPET registers */
-    return hvm_save_entry(HPET, 0, h, &hp->hpet);
+    rc = hvm_save_entry(HPET, 0, h, &hp->hpet);
+
+    spin_unlock(&hp->lock);
+
+    return rc;
 }
 
 static int hpet_load(struct domain *d, hvm_domain_context_t *h)
@@ -404,9 +432,14 @@ static int hpet_load(struct domain *d, hvm_domain_context_t *h)
     HPETState *hp = &d->arch.hvm_domain.pl_time.vhpet;
     int i;
 
+    spin_lock(&hp->lock);
+
     /* Reload the HPET registers */
     if ( hvm_load_entry(HPET, h, &hp->hpet) )
+    {
+        spin_unlock(&hp->lock);
         return -EINVAL;
+    }
     
     /* Recalculate the offset between the main counter and guest time */
     hp->mc_offset = hp->hpet.mc64 - guest_time_hpet(hp->vcpu);
@@ -415,6 +448,8 @@ static int hpet_load(struct domain *d, hvm_domain_context_t *h)
     for ( i = 0; i < HPET_TIMER_NUM; i++ )
         hpet_set_timer(hp, i);
 
+    spin_unlock(&hp->lock);
+
     return 0;
 }
 
@@ -427,6 +462,8 @@ void hpet_init(struct vcpu *v)
 
     memset(h, 0, sizeof(HPETState));
 
+    spin_lock_init(&h->lock);
+
     h->vcpu = v;
     h->tsc_freq = ticks_per_sec(v);
 

xen/arch/x86/hvm/hvm.c

@@ -401,6 +401,7 @@ int hvm_vcpu_initialise(struct vcpu *v)
         get_ioreq(v)->vp_eport = v->arch.hvm_vcpu.xen_port;
     spin_unlock(&v->domain->arch.hvm_domain.ioreq.lock);
 
+    spin_lock_init(&v->arch.hvm_vcpu.tm_lock);
     INIT_LIST_HEAD(&v->arch.hvm_vcpu.tm_list);
 
     if ( v->vcpu_id == 0 )

xen/arch/x86/hvm/i8254.c

@@ -82,6 +82,8 @@ static int pit_get_count(PITState *pit, int channel)
     struct hvm_hw_pit_channel *c = &pit->hw.channels[channel];
     struct vcpu *v = vpit_vcpu(pit);
 
+    ASSERT(spin_is_locked(&pit->lock));
+
     d = muldiv64(hvm_get_guest_time(v) - pit->count_load_time[channel],
                  PIT_FREQ, ticks_per_sec(v));
 
@@ -111,6 +113,8 @@ static int pit_get_out(PITState *pit, int channel)
     int out;
     struct vcpu *v = vpit_vcpu(pit);
 
+    ASSERT(spin_is_locked(&pit->lock));
+
     d = muldiv64(hvm_get_guest_time(v) - pit->count_load_time[channel], 
                  PIT_FREQ, ticks_per_sec(v));
 
@@ -143,6 +147,8 @@ static void pit_set_gate(PITState *pit, int channel, int val)
     struct hvm_hw_pit_channel *s = &pit->hw.channels[channel];
     struct vcpu *v = vpit_vcpu(pit);
 
+    ASSERT(spin_is_locked(&pit->lock));
+
     switch ( s->mode )
     {
     default:
@@ -165,6 +171,7 @@ static void pit_set_gate(PITState *pit, int channel, int val)
 
 int pit_get_gate(PITState *pit, int channel)
 {
+    ASSERT(spin_is_locked(&pit->lock));
     return pit->hw.channels[channel].gate;
 }
 
@@ -181,10 +188,15 @@ static void pit_load_count(PITState *pit, int channel, int val)
     struct periodic_time *pt = &pit->pt[channel];
     struct vcpu *v = vpit_vcpu(pit);
 
+    ASSERT(spin_is_locked(&pit->lock));
+
     if ( val == 0 )
         val = 0x10000;
 
-    pit->count_load_time[channel] = hvm_get_guest_time(pt->vcpu);
+    if ( v == NULL )
+        rdtscll(pit->count_load_time[channel]);
+    else
+        pit->count_load_time[channel] = hvm_get_guest_time(v);
     s->count = val;
     period = DIV_ROUND((val * 1000000000ULL), PIT_FREQ);
 
@@ -209,23 +221,29 @@ static void pit_load_count(PITState *pit, int channel, int val)
     }
 }
 
-static void pit_latch_count(PITState *s, int channel)
+static void pit_latch_count(PITState *pit, int channel)
 {
-    struct hvm_hw_pit_channel *c = &s->hw.channels[channel];
+    struct hvm_hw_pit_channel *c = &pit->hw.channels[channel];
+
+    ASSERT(spin_is_locked(&pit->lock));
+
     if ( !c->count_latched )
     {
-        c->latched_count = pit_get_count(s, channel);
+        c->latched_count = pit_get_count(pit, channel);
         c->count_latched = c->rw_mode;
     }
 }
 
-static void pit_latch_status(PITState *s, int channel)
+static void pit_latch_status(PITState *pit, int channel)
 {
-    struct hvm_hw_pit_channel *c = &s->hw.channels[channel];
+    struct hvm_hw_pit_channel *c = &pit->hw.channels[channel];
+
+    ASSERT(spin_is_locked(&pit->lock));
+
     if ( !c->status_latched )
     {
         /* TODO: Return NULL COUNT (bit 6). */
-        c->status = ((pit_get_out(s, channel) << 7) |
+        c->status = ((pit_get_out(pit, channel) << 7) |
                      (c->rw_mode << 4) |
                      (c->mode << 1) |
                      c->bcd);
@@ -241,6 +259,8 @@ static void pit_ioport_write(struct PITState *pit, uint32_t addr, uint32_t val)
     val  &= 0xff;
     addr &= 3;
 
+    spin_lock(&pit->lock);
+
     if ( addr == 3 )
     {
         channel = val >> 6;
@@ -304,6 +324,8 @@ static void pit_ioport_write(struct PITState *pit, uint32_t addr, uint32_t val)
             break;
         }
     }
+
+    spin_unlock(&pit->lock);
 }
 
 static uint32_t pit_ioport_read(struct PITState *pit, uint32_t addr)
@@ -314,6 +336,8 @@ static uint32_t pit_ioport_read(struct PITState *pit, uint32_t addr)
     addr &= 3;
     s = &pit->hw.channels[addr];
 
+    spin_lock(&pit->lock);
+
     if ( s->status_latched )
     {
         s->status_latched = 0;
@@ -364,12 +388,16 @@ static uint32_t pit_ioport_read(struct PITState *pit, uint32_t addr)
         }
     }
 
+    spin_unlock(&pit->lock);
+
     return ret;
 }
 
 void pit_stop_channel0_irq(PITState *pit)
 {
+    spin_lock(&pit->lock);
     destroy_periodic_time(&pit->pt[0]);
+    spin_unlock(&pit->lock);
 }
 
 #ifdef HVM_DEBUG_SUSPEND
@@ -422,11 +450,18 @@ static void pit_info(PITState *pit)
 static int pit_save(struct domain *d, hvm_domain_context_t *h)
 {
     PITState *pit = domain_vpit(d);
+    int rc;
+
+    spin_lock(&pit->lock);
     
     pit_info(pit);
 
     /* Save the PIT hardware state */
-    return hvm_save_entry(PIT, 0, h, &pit->hw);
+    rc = hvm_save_entry(PIT, 0, h, &pit->hw);
+
+    spin_unlock(&pit->lock);
+
+    return rc;
 }
 
 static int pit_load(struct domain *d, hvm_domain_context_t *h)
@@ -434,9 +469,14 @@ static int pit_load(struct domain *d, hvm_domain_context_t *h)
     PITState *pit = domain_vpit(d);
     int i;
 
+    spin_lock(&pit->lock);
+
     /* Restore the PIT hardware state */
     if ( hvm_load_entry(PIT, h, &pit->hw) )
+    {
+        spin_unlock(&pit->lock);
         return 1;
+    }
     
     /* Recreate platform timers from hardware state.  There will be some 
      * time jitter here, but the wall-clock will have jumped massively, so 
@@ -448,6 +488,9 @@ static int pit_load(struct domain *d, hvm_domain_context_t *h)
     }
 
     pit_info(pit);
+
+    spin_unlock(&pit->lock);
+
     return 0;
 }
 
@@ -456,17 +499,15 @@ HVM_REGISTER_SAVE_RESTORE(PIT, pit_save, pit_load, 1, HVMSR_PER_DOM);
 void pit_init(struct vcpu *v, unsigned long cpu_khz)
 {
     PITState *pit = vcpu_vpit(v);
-    struct periodic_time *pt;
     struct hvm_hw_pit_channel *s;
     int i;
 
-    pt = &pit->pt[0];  
-    pt[0].vcpu = v;
-    pt[1].vcpu = v;
-    pt[2].vcpu = v;
+    spin_lock_init(&pit->lock);
+
+    /* Some sub-functions assert that they are called with the lock held. */
+    spin_lock(&pit->lock);
 
     register_portio_handler(v->domain, PIT_BASE, 4, handle_pit_io);
-    /* register the speaker port */
     register_portio_handler(v->domain, 0x61, 1, handle_speaker_io);
     ticks_per_sec(v) = cpu_khz * (int64_t)1000;
 
@@ -477,6 +518,8 @@ void pit_init(struct vcpu *v, unsigned long cpu_khz)
         s->gate = (i != 2);
         pit_load_count(pit, i, 0);
     }
+
+    spin_unlock(&pit->lock);
 }
 
 void pit_deinit(struct domain *d)
@@ -492,10 +535,10 @@ static int handle_pit_io(ioreq_t *p)
 
     if ( (p->size != 1) || p->data_is_ptr || (p->type != IOREQ_TYPE_PIO) )
     {
-        gdprintk(XENLOG_WARNING, "HVM_PIT bad access\n");
+        gdprintk(XENLOG_WARNING, "PIT bad access\n");
         return 1;
     }
-    
+
     if ( p->dir == IOREQ_WRITE )
     {
         pit_ioport_write(vpit, p->addr, p->data);
@@ -505,7 +548,7 @@ static int handle_pit_io(ioreq_t *p)
         if ( (p->addr & 3) != 3 )
             p->data = pit_ioport_read(vpit, p->addr);
         else
-            gdprintk(XENLOG_WARNING, "HVM_PIT: read A1:A0=3!\n");
+            gdprintk(XENLOG_WARNING, "PIT: read A1:A0=3!\n");
     }
 
     return 1;
@@ -533,15 +576,19 @@ static int handle_speaker_io(ioreq_t *p)
 
     if ( (p->size != 1) || p->data_is_ptr || (p->type != IOREQ_TYPE_PIO) )
     {
-        gdprintk(XENLOG_WARNING, "HVM_SPEAKER bad access\n");
+        gdprintk(XENLOG_WARNING, "PIT_SPEAKER bad access\n");
         return 1;
     }
 
+    spin_lock(&vpit->lock);
+
     if ( p->dir == IOREQ_WRITE )
         speaker_ioport_write(vpit, p->addr, p->data);
     else
         p->data = speaker_ioport_read(vpit, p->addr);
 
+    spin_unlock(&vpit->lock);
+
     return 1;
 }
 

xen/arch/x86/hvm/pmtimer.c

@@ -53,6 +53,8 @@
 /* Dispatch SCIs based on the PM1a_STS and PM1a_EN registers */
 static void pmt_update_sci(PMTState *s)
 {
+    ASSERT(spin_is_locked(&s->lock));
+
     if ( s->pm.pm1a_en & s->pm.pm1a_sts & SCI_MASK )
         hvm_isa_irq_assert(s->vcpu->domain, SCI_IRQ);
     else
@@ -66,6 +68,8 @@ static void pmt_update_time(PMTState *s)
     uint64_t curr_gtime;
     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;
@@ -89,6 +93,8 @@ static void pmt_timer_callback(void *opaque)
     uint32_t pmt_cycles_until_flip;
     uint64_t time_until_flip;
 
+    spin_lock(&s->lock);
+
     /* Recalculate the timer and make sure we get an SCI if we need one */
     pmt_update_time(s);
 
@@ -103,8 +109,9 @@ static void pmt_timer_callback(void *opaque)
 
     /* Wake up again near the next bit-flip */
     set_timer(&s->timer, NOW() + time_until_flip + MILLISECS(1));
-}
 
+    spin_unlock(&s->lock);
+}
 
 /* Handle port I/O to the PM1a_STS and PM1a_EN registers */
 static int handle_evt_io(ioreq_t *p)
@@ -114,7 +121,9 @@ static int handle_evt_io(ioreq_t *p)
     uint32_t addr, data, byte;
     int i;
 
-    if ( p->dir == 0 ) /* Write */
+    spin_lock(&s->lock);
+
+    if ( p->dir == IOREQ_WRITE )
     {
         /* Handle this I/O one byte at a time */
         for ( i = p->size, addr = p->addr, data = p->data;
@@ -122,7 +131,7 @@ static int handle_evt_io(ioreq_t *p)
               i--, addr++, data >>= 8 )
         {
             byte = data & 0xff;
-            switch(addr) 
+            switch ( addr )
             {
                 /* PM1a_STS register bits are write-to-clear */
             case PM1a_STS_ADDR:
@@ -149,7 +158,7 @@ static int handle_evt_io(ioreq_t *p)
         /* Fix up the SCI state to match the new register state */
         pmt_update_sci(s);
     }
-    else /* Read */
+    else /* p->dir == IOREQ_READ */
     {
         data = s->pm.pm1a_sts | (((uint32_t) s->pm.pm1a_en) << 16);
         data >>= 8 * (p->addr - PM1a_STS_ADDR);
@@ -157,6 +166,9 @@ static int handle_evt_io(ioreq_t *p)
         else if ( p->size == 2 ) data &= 0xffff;
         p->data = data;
     }
+
+    spin_unlock(&s->lock);
+
     return 1;
 }
 
@@ -167,29 +179,31 @@ static int handle_pmt_io(ioreq_t *p)
     struct vcpu *v = current;
     PMTState *s = &v->domain->arch.hvm_domain.pl_time.vpmt;
 
-    if (p->size != 4 ||
-        p->data_is_ptr ||
-        p->type != IOREQ_TYPE_PIO){
-        printk("HVM_PMT: wrong PM timer IO\n");
+    if ( (p->size != 4) || p->data_is_ptr || (p->type != IOREQ_TYPE_PIO) )
+    {
+        gdprintk(XENLOG_WARNING, "HVM_PMT bad access\n");
         return 1;
     }
     
-    if (p->dir == 0) { /* write */
-        /* PM_TMR_BLK is read-only */
-        return 1;
-    } else if (p->dir == 1) { /* read */
+    if ( p->dir == IOREQ_READ )
+    {
+        spin_lock(&s->lock);
         pmt_update_time(s);
         p->data = s->pm.tmr_val;
+        spin_unlock(&s->lock);
         return 1;
     }
+
     return 0;
 }
 
 static int pmtimer_save(struct domain *d, hvm_domain_context_t *h)
 {
     PMTState *s = &d->arch.hvm_domain.pl_time.vpmt;
-    uint32_t msb = s->pm.tmr_val & TMR_VAL_MSB;
-    uint32_t x;
+    uint32_t x, msb = s->pm.tmr_val & TMR_VAL_MSB;
+    int rc;
+
+    spin_lock(&s->lock);
 
     /* Update the counter to the guest's current time.  We always save
      * with the domain paused, so the saved time should be after the
@@ -202,22 +216,33 @@ static int pmtimer_save(struct domain *d, hvm_domain_context_t *h)
     /* No point in setting the SCI here because we'll already have saved the 
      * IRQ and *PIC state; we'll fix it up when we restore the domain */
 
-    return hvm_save_entry(PMTIMER, 0, h, &s->pm);
+    rc = hvm_save_entry(PMTIMER, 0, h, &s->pm);
+
+    spin_unlock(&s->lock);
+
+    return rc;
 }
 
 static int pmtimer_load(struct domain *d, hvm_domain_context_t *h)
 {
     PMTState *s = &d->arch.hvm_domain.pl_time.vpmt;
 
+    spin_lock(&s->lock);
+
     /* Reload the registers */
     if ( hvm_load_entry(PMTIMER, h, &s->pm) )
+    {
+        spin_unlock(&s->lock);
         return -EINVAL;
+    }
 
     /* Calculate future counter values from now. */
     s->last_gtime = hvm_get_guest_time(s->vcpu);
 
     /* Set the SCI state from the registers */ 
     pmt_update_sci(s);
+
+    spin_unlock(&s->lock);
     
     return 0;
 }
@@ -225,14 +250,11 @@ static int pmtimer_load(struct domain *d, hvm_domain_context_t *h)
 HVM_REGISTER_SAVE_RESTORE(PMTIMER, pmtimer_save, pmtimer_load, 
                           1, HVMSR_PER_DOM);
 
-
 void pmtimer_init(struct vcpu *v)
 {
     PMTState *s = &v->domain->arch.hvm_domain.pl_time.vpmt;
 
-    s->pm.tmr_val = 0;
-    s->pm.pm1a_sts = 0;
-    s->pm.pm1a_en = 0;
+    spin_lock_init(&s->lock);
 
     s->scale = ((uint64_t)FREQUENCE_PMTIMER << 32) / ticks_per_sec(v);
     s->vcpu = v;

xen/arch/x86/hvm/rtc.c

@@ -34,10 +34,12 @@
                                        arch.hvm_domain.pl_time.vrtc))
 #define vrtc_vcpu(rtc)   (vrtc_domain(rtc)->vcpu[0])
 
-void rtc_periodic_cb(struct vcpu *v, void *opaque)
+static void rtc_periodic_cb(struct vcpu *v, void *opaque)
 {
     RTCState *s = opaque;
+    spin_lock(&s->lock);
     s->hw.cmos_data[RTC_REG_C] |= 0xc0;
+    spin_unlock(&s->lock);
 }
 
 int is_rtc_periodic_irq(void *opaque)
@@ -55,6 +57,8 @@ static void rtc_timer_update(RTCState *s)
     int period_code, period;
     struct vcpu *v = vrtc_vcpu(s);
 
+    ASSERT(spin_is_locked(&s->lock));
+
     period_code = s->hw.cmos_data[RTC_REG_A] & RTC_RATE_SELECT;
     if ( (period_code != 0) && (s->hw.cmos_data[RTC_REG_B] & RTC_PIE) )
     {
@@ -78,14 +82,21 @@ static int rtc_ioport_write(void *opaque, uint32_t addr, uint32_t data)
 {
     RTCState *s = opaque;
 
+    spin_lock(&s->lock);
+
     if ( (addr & 1) == 0 )
     {
-        s->hw.cmos_index = data & 0x7f;
-        return (s->hw.cmos_index < RTC_CMOS_SIZE);
+        data &= 0x7f;
+        s->hw.cmos_index = data;
+        spin_unlock(&s->lock);
+        return (data < RTC_CMOS_SIZE);
     }
 
     if ( s->hw.cmos_index >= RTC_CMOS_SIZE )
+    {
+        spin_unlock(&s->lock);
         return 0;
+    }
 
     switch ( s->hw.cmos_index )
     {
@@ -134,6 +145,8 @@ static int rtc_ioport_write(void *opaque, uint32_t addr, uint32_t data)
         break;
     }
 
+    spin_unlock(&s->lock);
+
     return 1;
 }
 
@@ -158,6 +171,8 @@ static void rtc_set_time(RTCState *s)
     struct tm *tm = &s->current_tm;
     unsigned long before, after; /* XXX s_time_t */
       
+    ASSERT(spin_is_locked(&s->lock));
+
     before = mktime(tm->tm_year, tm->tm_mon, tm->tm_mday,
 		    tm->tm_hour, tm->tm_min, tm->tm_sec);
     
@@ -182,6 +197,8 @@ static void rtc_copy_date(RTCState *s)
     const struct tm *tm = &s->current_tm;
     struct domain *d = vrtc_domain(s);
 
+    ASSERT(spin_is_locked(&s->lock));
+
     if ( s->time_offset_seconds != d->time_offset_seconds )
     {
         s->current_tm = gmtime(get_localtime(d));
@@ -231,6 +248,8 @@ static void rtc_next_second(RTCState *s)
     int days_in_month;
     struct domain *d = vrtc_domain(s);
 
+    ASSERT(spin_is_locked(&s->lock));
+
     if ( s->time_offset_seconds != d->time_offset_seconds )
     {
         s->current_tm = gmtime(get_localtime(d));
@@ -279,6 +298,8 @@ static void rtc_update_second(void *opaque)
 {
     RTCState *s = opaque;
 
+    spin_lock(&s->lock);
+
     /* if the oscillator is not in normal operation, we do not update */
     if ( (s->hw.cmos_data[RTC_REG_A] & RTC_DIV_CTL) != RTC_REF_CLCK_32KHZ )
     {
@@ -295,6 +316,8 @@ static void rtc_update_second(void *opaque)
         /* Delay time before update cycle */
         set_timer(&s->second_timer2, s->next_second_time + 244000);
     }
+
+    spin_unlock(&s->lock);
 }
 
 static void rtc_update_second2(void *opaque)
@@ -302,6 +325,8 @@ static void rtc_update_second2(void *opaque)
     RTCState *s = opaque;
     struct domain *d = vrtc_domain(s);
 
+    spin_lock(&s->lock);
+
     if ( !(s->hw.cmos_data[RTC_REG_B] & RTC_SET) )
         rtc_copy_date(s);
 
@@ -337,16 +362,19 @@ static void rtc_update_second2(void *opaque)
 
     s->next_second_time += 1000000000ULL;
     set_timer(&s->second_timer, s->next_second_time);
+
+    spin_unlock(&s->lock);
 }
 
-static uint32_t rtc_ioport_read(void *opaque, uint32_t addr)
+static uint32_t rtc_ioport_read(RTCState *s, uint32_t addr)
 {
-    RTCState *s = opaque;
     int ret;
 
     if ( (addr & 1) == 0 )
         return 0xff;
 
+    spin_lock(&s->lock);
+
     switch ( s->hw.cmos_index )
     {
     case RTC_SECONDS:
@@ -371,6 +399,8 @@ static uint32_t rtc_ioport_read(void *opaque, uint32_t addr)
         break;
     }
 
+    spin_unlock(&s->lock);
+
     return ret;
 }
 
@@ -413,7 +443,11 @@ void rtc_migrate_timers(struct vcpu *v)
 static int rtc_save(struct domain *d, hvm_domain_context_t *h)
 {
     RTCState *s = domain_vrtc(d);
-    return hvm_save_entry(RTC, 0, h, &s->hw);
+    int rc;
+    spin_lock(&s->lock);
+    rc = hvm_save_entry(RTC, 0, h, &s->hw);
+    spin_unlock(&s->lock);
+    return rc;
 }
 
 /* Reload the hardware state from a saved domain */
@@ -421,9 +455,14 @@ static int rtc_load(struct domain *d, hvm_domain_context_t *h)
 {
     RTCState *s = domain_vrtc(d);
 
+    spin_lock(&s->lock);
+
     /* Restore the registers */
     if ( hvm_load_entry(RTC, h, &s->hw) != 0 )
+    {
+        spin_unlock(&s->lock);
         return -EINVAL;
+    }
 
     /* Reset the wall-clock time.  In normal running, this runs with host 
      * time, so let's keep doing that. */
@@ -436,6 +475,8 @@ static int rtc_load(struct domain *d, hvm_domain_context_t *h)
     /* Reset the periodic interrupt timer based on the registers */
     rtc_timer_update(s);
 
+    spin_unlock(&s->lock);
+
     return 0;
 }
 
@@ -446,13 +487,18 @@ void rtc_init(struct vcpu *v, int base)
 {
     RTCState *s = vcpu_vrtc(v);
 
+    spin_lock_init(&s->lock);
+
     s->hw.cmos_data[RTC_REG_A] = RTC_REF_CLCK_32KHZ | 6; /* ~1kHz */
     s->hw.cmos_data[RTC_REG_B] = RTC_24H;
     s->hw.cmos_data[RTC_REG_C] = 0;
     s->hw.cmos_data[RTC_REG_D] = RTC_VRT;
 
     s->current_tm = gmtime(get_localtime(v->domain));
+
+    spin_lock(&s->lock);
     rtc_copy_date(s);
+    spin_unlock(&s->lock);
 
     init_timer(&s->second_timer, rtc_update_second, s, v->processor);
     init_timer(&s->second_timer2, rtc_update_second2, s, v->processor);

xen/arch/x86/hvm/svm/vmcb.c

@@ -236,6 +236,16 @@ static int construct_vmcb(struct vcpu *v)
         vmcb->g_pat = 0x0007040600070406ULL; /* guest PAT */
         vmcb->h_cr3 = pagetable_get_paddr(v->domain->arch.phys_table);
         vmcb->cr4 = arch_svm->cpu_shadow_cr4 = 0;
+
+        /* No point in intercepting CR0/3/4 reads, because the hardware 
+         * will return the guest versions anyway. */
+        vmcb->cr_intercepts &= ~(CR_INTERCEPT_CR0_READ
+                                 |CR_INTERCEPT_CR3_READ
+                                 |CR_INTERCEPT_CR4_READ);
+
+        /* No point in intercepting INVLPG if we don't have shadow pagetables 
+         * that need to be fixed up. */
+        vmcb->general1_intercepts &= ~GENERAL1_INTERCEPT_INVLPG;
     }
     else
     {

xen/arch/x86/hvm/vpt.c

@@ -17,11 +17,31 @@
  * Place - Suite 330, Boston, MA 02111-1307 USA.
  *
  */
+
 #include <xen/time.h>
 #include <asm/hvm/support.h>
 #include <asm/hvm/vpt.h>
 #include <asm/event.h>
 
+static void pt_lock(struct periodic_time *pt)
+{
+    struct vcpu *v;
+
+    for ( ; ; )
+    {
+        v = pt->vcpu;
+        spin_lock(&v->arch.hvm_vcpu.tm_lock);
+        if ( likely(pt->vcpu == v) )
+            break;
+        spin_unlock(&v->arch.hvm_vcpu.tm_lock);
+    }
+}
+
+static void pt_unlock(struct periodic_time *pt)
+{
+    spin_unlock(&pt->vcpu->arch.hvm_vcpu.tm_lock);
+}
+
 static void missed_ticks(struct periodic_time *pt)
 {
     s_time_t missed_ticks;
@@ -52,10 +72,14 @@ void pt_freeze_time(struct vcpu *v)
     if ( test_bit(_VPF_blocked, &v->pause_flags) )
         return;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     v->arch.hvm_vcpu.guest_time = hvm_get_guest_time(v);
 
     list_for_each_entry ( pt, head, list )
         stop_timer(&pt->timer);
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
 }
 
 void pt_thaw_time(struct vcpu *v)
@@ -63,6 +87,8 @@ void pt_thaw_time(struct vcpu *v)
     struct list_head *head = &v->arch.hvm_vcpu.tm_list;
     struct periodic_time *pt;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     if ( v->arch.hvm_vcpu.guest_time )
     {
         hvm_set_guest_time(v, v->arch.hvm_vcpu.guest_time);
@@ -74,12 +100,16 @@ void pt_thaw_time(struct vcpu *v)
             set_timer(&pt->timer, pt->scheduled);
         }
     }
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
 }
 
 static void pt_timer_fn(void *data)
 {
     struct periodic_time *pt = data;
 
+    pt_lock(pt);
+
     pt->pending_intr_nr++;
     pt->scheduled += pt->period;
 
@@ -89,6 +119,8 @@ static void pt_timer_fn(void *data)
         set_timer(&pt->timer, pt->scheduled);
 
     vcpu_kick(pt->vcpu);
+
+    pt_unlock(pt);
 }
 
 void pt_update_irq(struct vcpu *v)
@@ -98,6 +130,8 @@ void pt_update_irq(struct vcpu *v)
     uint64_t max_lag = -1ULL;
     int irq = -1;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     list_for_each_entry ( pt, head, list )
     {
         if ( !is_isa_irq_masked(v, pt->irq) && pt->pending_intr_nr &&
@@ -108,6 +142,8 @@ void pt_update_irq(struct vcpu *v)
         }
     }
 
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
+
     if ( is_lvtt(v, irq) )
     {
         vlapic_set_irq(vcpu_vlapic(v), irq, 0);
@@ -119,7 +155,7 @@ void pt_update_irq(struct vcpu *v)
     }
 }
 
-struct periodic_time *is_pt_irq(struct vcpu *v, int vector, int type)
+static struct periodic_time *is_pt_irq(struct vcpu *v, int vector, int type)
 {
     struct list_head *head = &v->arch.hvm_vcpu.tm_list;
     struct periodic_time *pt;
@@ -152,19 +188,34 @@ struct periodic_time *is_pt_irq(struct vcpu *v, int vector, int type)
 
 void pt_intr_post(struct vcpu *v, int vector, int type)
 {
-    struct periodic_time *pt = is_pt_irq(v, vector, type);
+    struct periodic_time *pt;
+    time_cb *cb;
+    void *cb_priv;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
+    pt = is_pt_irq(v, vector, type);
     if ( pt == NULL )
+    {
+        spin_unlock(&v->arch.hvm_vcpu.tm_lock);
         return;
+    }
+
+    ASSERT(pt->vcpu == v);
 
     pt->pending_intr_nr--;
     pt->last_plt_gtime += pt->period_cycles;
 
-    if ( hvm_get_guest_time(pt->vcpu) < pt->last_plt_gtime )
-        hvm_set_guest_time(pt->vcpu, pt->last_plt_gtime);
+    if ( hvm_get_guest_time(v) < pt->last_plt_gtime )
+        hvm_set_guest_time(v, pt->last_plt_gtime);
 
-    if ( pt->cb != NULL )
-        pt->cb(pt->vcpu, pt->priv);
+    cb = pt->cb;
+    cb_priv = pt->priv;
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
+
+    if ( cb != NULL )
+        cb(v, cb_priv);
 }
 
 void pt_reset(struct vcpu *v)
@@ -172,6 +223,8 @@ void pt_reset(struct vcpu *v)
     struct list_head *head = &v->arch.hvm_vcpu.tm_list;
     struct periodic_time *pt;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     list_for_each_entry ( pt, head, list )
     {
         if ( pt->enabled )
@@ -182,6 +235,8 @@ void pt_reset(struct vcpu *v)
             set_timer(&pt->timer, pt->scheduled);
         }
     }
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
 }
 
 void pt_migrate(struct vcpu *v)
@@ -189,11 +244,15 @@ void pt_migrate(struct vcpu *v)
     struct list_head *head = &v->arch.hvm_vcpu.tm_list;
     struct periodic_time *pt;
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     list_for_each_entry ( pt, head, list )
     {
         if ( pt->enabled )
             migrate_timer(&pt->timer, v->processor);
     }
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
 }
 
 void create_periodic_time(
@@ -202,6 +261,8 @@ void create_periodic_time(
 {
     destroy_periodic_time(pt);
 
+    spin_lock(&v->arch.hvm_vcpu.tm_lock);
+
     init_timer(&pt->timer, pt_timer_fn, pt, v->processor);
     pt->enabled = 1;
     if ( period < 900000 ) /* < 0.9 ms */
@@ -223,6 +284,8 @@ void create_periodic_time(
 
     list_add(&pt->list, &v->arch.hvm_vcpu.tm_list);
     set_timer(&pt->timer, pt->scheduled);
+
+    spin_unlock(&v->arch.hvm_vcpu.tm_lock);
 }
 
 void destroy_periodic_time(struct periodic_time *pt)
@@ -230,8 +293,10 @@ void destroy_periodic_time(struct periodic_time *pt)
     if ( !pt->enabled )
         return;
 
+    pt_lock(pt);
     pt->enabled = 0;
     pt->pending_intr_nr = 0;
     list_del(&pt->list);
     kill_timer(&pt->timer);
+    pt_unlock(pt);
 }

xen/arch/x86/mm/hap/hap.c

@@ -55,14 +55,14 @@
 /* hap code to call when log_dirty is enable. return 0 if no problem found. */
 int hap_enable_log_dirty(struct domain *d)
 {
-    hap_lock(d);
     /* turn on PG_log_dirty bit in paging mode */
+    hap_lock(d);
     d->arch.paging.mode |= PG_log_dirty;
-    /* set l1e entries of P2M table to NOT_WRITABLE. */
-    p2m_set_flags_global(d, (_PAGE_PRESENT|_PAGE_USER));
-    flush_tlb_all_pge();
     hap_unlock(d);
 
+    /* set l1e entries of P2M table to NOT_WRITABLE. */
+    p2m_set_flags_global(d, (_PAGE_PRESENT|_PAGE_USER));
+    flush_tlb_mask(d->domain_dirty_cpumask);
     return 0;
 }
 
@@ -70,19 +70,20 @@ int hap_disable_log_dirty(struct domain *d)
 {
     hap_lock(d);
     d->arch.paging.mode &= ~PG_log_dirty;
-    /* set l1e entries of P2M table with normal mode */
-    p2m_set_flags_global(d, __PAGE_HYPERVISOR|_PAGE_USER);
     hap_unlock(d);
-    
-    return 1;
+
+    /* set l1e entries of P2M table with normal mode */
+    p2m_set_flags_global(d, __PAGE_HYPERVISOR|_PAGE_USER);    
+    return 0;
 }
 
 void hap_clean_dirty_bitmap(struct domain *d)
 {
     /* mark physical memory as NOT_WRITEABLE and flush the TLB */
     p2m_set_flags_global(d, (_PAGE_PRESENT|_PAGE_USER));
-    flush_tlb_all_pge();
+    flush_tlb_mask(d->domain_dirty_cpumask);
 }
+
 /************************************************/
 /*             HAP SUPPORT FUNCTIONS            */
 /************************************************/
@@ -268,6 +269,7 @@ void hap_install_xen_entries_in_l2h(struct vcpu *v, mfn_t sl2hmfn)
 {
     struct domain *d = v->domain;
     l2_pgentry_t *sl2e;
+    l3_pgentry_t *p2m;
 
     int i;
 
@@ -290,23 +292,18 @@ void hap_install_xen_entries_in_l2h(struct vcpu *v, mfn_t sl2hmfn)
         sl2e[l2_table_offset(LINEAR_PT_VIRT_START) + i] =
             l2e_empty();
 
-    if ( paging_mode_translate(d) )
+    /* Install the domain-specific p2m table */
+    ASSERT(pagetable_get_pfn(d->arch.phys_table) != 0);
+    p2m = hap_map_domain_page(pagetable_get_mfn(d->arch.phys_table));
+    for ( i = 0; i < MACHPHYS_MBYTES>>1; i++ )
     {
-        /* Install the domain-specific p2m table */
-        l3_pgentry_t *p2m;
-        ASSERT(pagetable_get_pfn(d->arch.phys_table) != 0);
-        p2m = hap_map_domain_page(pagetable_get_mfn(d->arch.phys_table));
-        for ( i = 0; i < MACHPHYS_MBYTES>>1; i++ )
-        {
-            sl2e[l2_table_offset(RO_MPT_VIRT_START) + i] =
-                (l3e_get_flags(p2m[i]) & _PAGE_PRESENT)
-                ? l2e_from_pfn(mfn_x(_mfn(l3e_get_pfn(p2m[i]))),
-                                      __PAGE_HYPERVISOR)
-                : l2e_empty();
-        }
-        hap_unmap_domain_page(p2m);
+        sl2e[l2_table_offset(RO_MPT_VIRT_START) + i] =
+            (l3e_get_flags(p2m[i]) & _PAGE_PRESENT)
+            ? l2e_from_pfn(mfn_x(_mfn(l3e_get_pfn(p2m[i]))),
+                           __PAGE_HYPERVISOR)
+            : l2e_empty();
     }
-
+    hap_unmap_domain_page(p2m);
     hap_unmap_domain_page(sl2e);
 }
 #endif
@@ -565,61 +562,37 @@ void hap_vcpu_init(struct vcpu *v)
 /************************************************/
 /*          HAP PAGING MODE FUNCTIONS           */
 /************************************************/
-/* In theory, hap should not intercept guest page fault. This function can 
- * be recycled to handle host/nested page fault, if needed.
+/* 
+ * HAP guests can handle page faults (in the guest page tables) without
+ * needing any action from Xen, so we should not be intercepting them.
  */
 int hap_page_fault(struct vcpu *v, unsigned long va, 
                    struct cpu_user_regs *regs)
 {
-    HERE_I_AM;
+    HAP_ERROR("Intercepted a guest #PF (%u:%u) with HAP enabled.\n",
+              v->domain->domain_id, v->vcpu_id);
     domain_crash(v->domain);
     return 0;
 }
 
-/* called when guest issues a invlpg request. 
- * Return 1 if need to issue page invalidation on CPU; Return 0 if does not
- * need to do so.
+/* 
+ * HAP guests can handle invlpg without needing any action from Xen, so
+ * should not be intercepting it. 
  */
 int hap_invlpg(struct vcpu *v, unsigned long va)
 {
-    HERE_I_AM;
+    HAP_ERROR("Intercepted a guest INVLPG (%u:%u) with HAP enabled.\n",
+              v->domain->domain_id, v->vcpu_id);
+    domain_crash(v->domain);
     return 0;
 }
 
+/*
+ * HAP guests do not need to take any action on CR3 writes (they are still
+ * intercepted, so that Xen's copy of the guest's CR3 can be kept in sync.)
+ */
 void hap_update_cr3(struct vcpu *v, int do_locking)
 {
-    struct domain *d = v->domain;
-    mfn_t gmfn;
-
-    HERE_I_AM;
-    /* Don't do anything on an uninitialised vcpu */
-    if ( !is_hvm_domain(d) && !v->is_initialised )
-    {
-        ASSERT(v->arch.cr3 == 0);
-        return;
-    }
-
-    if ( do_locking )
-        hap_lock(v->domain);
-    
-    ASSERT(hap_locked_by_me(v->domain));
-    ASSERT(v->arch.paging.mode);
-    
-    gmfn = pagetable_get_mfn(v->arch.guest_table);
-
-    make_cr3(v, pagetable_get_pfn(v->arch.monitor_table));
-    
-    hvm_update_guest_cr3(v, pagetable_get_paddr(v->arch.monitor_table));
-
-    HAP_PRINTK("d=%u v=%u guest_table=%05lx, monitor_table = %05lx\n", 
-               d->domain_id, v->vcpu_id, 
-               (unsigned long)pagetable_get_pfn(v->arch.guest_table),
-               (unsigned long)pagetable_get_pfn(v->arch.monitor_table));
-
-    flush_tlb_mask(d->domain_dirty_cpumask);
-
-    if ( do_locking )
-        hap_unlock(v->domain);
 }
 
 void hap_update_paging_modes(struct vcpu *v)
@@ -647,7 +620,7 @@ void hap_update_paging_modes(struct vcpu *v)
         v->arch.paging.mode = &hap_paging_real_mode;
     }
 
-    v->arch.paging.translate_enabled = !!hvm_paging_enabled(v);    
+    v->arch.paging.translate_enabled = !!hvm_paging_enabled(v);
 
     if ( pagetable_is_null(v->arch.monitor_table) ) {
         mfn_t mmfn = hap_make_monitor_table(v);
@@ -655,7 +628,6 @@ void hap_update_paging_modes(struct vcpu *v)
         make_cr3(v, mfn_x(mmfn));
     }
 
-    flush_tlb_mask(d->domain_dirty_cpumask);
     hap_unlock(d);
 }
 
@@ -702,29 +674,18 @@ void
 hap_write_p2m_entry(struct vcpu *v, unsigned long gfn, l1_pgentry_t *p,
                     l1_pgentry_t new, unsigned int level)
 {
-    int do_locking;
-
-    /* This function can be called from two directions (P2M and log dirty). We
-     *  need to make sure this lock has been held or not.
-     */
-    do_locking = !hap_locked_by_me(v->domain);
-
-    if ( do_locking )
-        hap_lock(v->domain);
+    hap_lock(v->domain);
 
     safe_write_pte(p, new);
 #if CONFIG_PAGING_LEVELS == 3
     /* install P2M in monitor table for PAE Xen */
-    if ( level == 3 ) {
+    if ( level == 3 ) 
 	/* We have written to the p2m l3: need to sync the per-vcpu
          * copies of it in the monitor tables */
 	p2m_install_entry_in_monitors(v->domain, (l3_pgentry_t *)p);
-	
-    }
 #endif
     
-    if ( do_locking )
-        hap_unlock(v->domain);
+    hap_unlock(v->domain);
 }
 
 /* Entry points into this mode of the hap code. */

xen/arch/x86/mm/p2m.c

@@ -32,9 +32,13 @@
 #define P2M_AUDIT     0
 #define P2M_DEBUGGING 1
 
-/* The P2M lock.  This protects all updates to the p2m table.
+/*
+ * The P2M lock.  This protects all updates to the p2m table.
  * Updates are expected to be safe against concurrent reads, 
- * which do *not* require the lock */
+ * which do *not* require the lock.
+ *
+ * Locking discipline: always acquire this lock before the shadow or HAP one
+ */
 
 #define p2m_lock_init(_d)                            \
     do {                                             \

xen/common/kexec.c

@@ -19,6 +19,7 @@
 #include <asm/atomic.h>
 #include <xen/spinlock.h>
 #include <xen/version.h>
+#include <xen/console.h>
 #include <public/elfnote.h>
 
 #ifndef COMPAT
@@ -110,6 +111,8 @@ void kexec_crash(void)
     if ( !test_bit(KEXEC_IMAGE_CRASH_BASE + pos, &kexec_flags) )
         return;
 
+    console_start_sync();
+
     one_cpu_only();
     kexec_crash_save_cpu();
     machine_crash_shutdown();

xen/common/timer.c

@@ -183,7 +183,7 @@ static inline void timer_lock(struct timer *timer)
 
 static inline void timer_unlock(struct timer *timer)
 {
-        spin_unlock(&per_cpu(timers, timer->cpu).lock);
+    spin_unlock(&per_cpu(timers, timer->cpu).lock);
 }
 
 #define timer_unlock_irq(t) \

xen/include/asm-x86/hvm/vcpu.h

@@ -35,6 +35,9 @@ struct hvm_vcpu {
     struct vlapic       vlapic;
     s64                 cache_tsc_offset;
     u64                 guest_time;
+
+    /* Lock and list for virtual platform timers. */
+    spinlock_t          tm_lock;
     struct list_head    tm_list;
 
     /* For AP startup */

xen/include/asm-x86/hvm/vpt.h

@@ -31,7 +31,6 @@
 #include <asm/hvm/vpic.h>
 #include <public/hvm/save.h>
 
-
 struct HPETState;
 struct HPET_timer_fn_info {
     struct HPETState       *hs;
@@ -45,6 +44,7 @@ typedef struct HPETState {
     uint64_t mc_offset;
     struct timer timers[HPET_TIMER_NUM];
     struct HPET_timer_fn_info timer_fn_info[HPET_TIMER_NUM]; 
+    spinlock_t lock;
 } HPETState;
 
 
@@ -80,6 +80,7 @@ typedef struct PITState {
     int64_t count_load_time[3];
     /* irq handling */
     struct periodic_time pt[3];
+    spinlock_t lock;
 } PITState;
 
 typedef struct RTCState {
@@ -93,6 +94,7 @@ typedef struct RTCState {
     struct timer second_timer2;
     struct periodic_time pt;
     int32_t time_offset_seconds;
+    spinlock_t lock;
 } RTCState;
 
 #define FREQUENCE_PMTIMER  3579545  /* Timer should run at 3.579545 MHz */
@@ -102,6 +104,7 @@ typedef struct PMTState {
     uint64_t last_gtime;        /* Last (guest) time we updated the timer */
     uint64_t scale;             /* Multiplier to get from tsc to timer ticks */
     struct timer timer;         /* To make sure we send SCIs */
+    spinlock_t lock;
 } PMTState;
 
 struct pl_time {    /* platform time */
@@ -116,7 +119,6 @@ struct pl_time {    /* platform time */
 void pt_freeze_time(struct vcpu *v);
 void pt_thaw_time(struct vcpu *v);
 void pt_update_irq(struct vcpu *v);
-struct periodic_time *is_pt_irq(struct vcpu *v, int vector, int type);
 void pt_intr_post(struct vcpu *v, int vector, int type);
 void pt_reset(struct vcpu *v);
 void pt_migrate(struct vcpu *v);