Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "Misc fixes:

   - fix hotplug bugs
   - fix irq live lock
   - fix various topology handling bugs
   - fix APIC ACK ordering
   - fix PV iopl handling
   - fix speling
   - fix/tweak memcpy_mcsafe() return value
   - fix fbcon bug
   - remove stray prototypes"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/msr: Remove unused native_read_tscp()
  x86/apic: Remove declaration of unused hw_nmi_is_cpu_stuck
  x86/oprofile/nmi: Add missing hotplug FROZEN handling
  x86/hpet: Use proper mask to modify hotplug action
  x86/apic/uv: Fix the hotplug notifier
  x86/apb/timer: Use proper mask to modify hotplug action
  x86/topology: Use total_cpus not nr_cpu_ids for logical packages
  x86/topology: Fix Intel HT disable
  x86/topology: Fix logical package mapping
  x86/irq: Cure live lock in fixup_irqs()
  x86/tsc: Prevent NULL pointer deref in calibrate_delay_is_known()
  x86/apic: Fix suspicious RCU usage in smp_trace_call_function_interrupt()
  x86/iopl: Fix iopl capability check on Xen PV
  x86/iopl/64: Properly context-switch IOPL on Xen PV
  selftests/x86: Add an iopl test
  x86/mm, x86/mce: Fix return type/value for memcpy_mcsafe()
  x86/video: Don't assume all FB devices are PCI devices
  arch/x86/irq: Purge useless handler declarations from hw_irq.h
  x86: Fix misspellings in comments

3 files changed, 74 insertions, 21 deletions

diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
index 531b9611c51d..d356987a04e9 100644
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -1611,7 +1611,7 @@ void __init enable_IR_x2apic(void)
 	legacy_pic->mask_all();
 	mask_ioapic_entries();
 
-	/* If irq_remapping_prepare() succeded, try to enable it */
+	/* If irq_remapping_prepare() succeeded, try to enable it */
 	if (ir_stat >= 0)
 		ir_stat = try_to_enable_IR();
 	/* ir_stat contains the remap mode or an error code */
diff --git a/arch/x86/kernel/apic/vector.c b/arch/x86/kernel/apic/vector.c
index 3b670df4ba7b..ad59d70bcb1a 100644
--- a/arch/x86/kernel/apic/vector.c
+++ b/arch/x86/kernel/apic/vector.c
@@ -213,6 +213,7 @@ update:
 	 */
 	cpumask_and(d->old_domain, d->old_domain, cpu_online_mask);
 	d->move_in_progress = !cpumask_empty(d->old_domain);
+	d->cfg.old_vector = d->move_in_progress ? d->cfg.vector : 0;
 	d->cfg.vector = vector;
 	cpumask_copy(d->domain, vector_cpumask);
 success:
@@ -655,46 +656,97 @@ void irq_complete_move(struct irq_cfg *cfg)
 }
 
 /*
- * Called with @desc->lock held and interrupts disabled.
+ * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
  */
 void irq_force_complete_move(struct irq_desc *desc)
 {
 	struct irq_data *irqdata = irq_desc_get_irq_data(desc);
 	struct apic_chip_data *data = apic_chip_data(irqdata);
 	struct irq_cfg *cfg = data ? &data->cfg : NULL;
+	unsigned int cpu;
 
 	if (!cfg)
 		return;
 
-	__irq_complete_move(cfg, cfg->vector);
-
 	/*
 	 * This is tricky. If the cleanup of @data->old_domain has not been
 	 * done yet, then the following setaffinity call will fail with
 	 * -EBUSY. This can leave the interrupt in a stale state.
 	 *
-	 * The cleanup cannot make progress because we hold @desc->lock. So in
-	 * case @data->old_domain is not yet cleaned up, we need to drop the
-	 * lock and acquire it again. @desc cannot go away, because the
-	 * hotplug code holds the sparse irq lock.
+	 * All CPUs are stuck in stop machine with interrupts disabled so
+	 * calling __irq_complete_move() would be completely pointless.
 	 */
 	raw_spin_lock(&vector_lock);
-	/* Clean out all offline cpus (including ourself) first. */
+	/*
+	 * Clean out all offline cpus (including the outgoing one) from the
+	 * old_domain mask.
+	 */
 	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
-	while (!cpumask_empty(data->old_domain)) {
+
+	/*
+	 * If move_in_progress is cleared and the old_domain mask is empty,
+	 * then there is nothing to cleanup. fixup_irqs() will take care of
+	 * the stale vectors on the outgoing cpu.
+	 */
+	if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
 		raw_spin_unlock(&vector_lock);
-		raw_spin_unlock(&desc->lock);
-		cpu_relax();
-		raw_spin_lock(&desc->lock);
+		return;
+	}
+
+	/*
+	 * 1) The interrupt is in move_in_progress state. That means that we
+	 *    have not seen an interrupt since the io_apic was reprogrammed to
+	 *    the new vector.
+	 *
+	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
+	 *    have not been processed yet.
+	 */
+	if (data->move_in_progress) {
 		/*
-		 * Reevaluate apic_chip_data. It might have been cleared after
-		 * we dropped @desc->lock.
+		 * In theory there is a race:
+		 *
+		 * set_ioapic(new_vector) <-- Interrupt is raised before update
+		 *			      is effective, i.e. it's raised on
+		 *			      the old vector.
+		 *
+		 * So if the target cpu cannot handle that interrupt before
+		 * the old vector is cleaned up, we get a spurious interrupt
+		 * and in the worst case the ioapic irq line becomes stale.
+		 *
+		 * But in case of cpu hotplug this should be a non issue
+		 * because if the affinity update happens right before all
+		 * cpus rendevouz in stop machine, there is no way that the
+		 * interrupt can be blocked on the target cpu because all cpus
+		 * loops first with interrupts enabled in stop machine, so the
+		 * old vector is not yet cleaned up when the interrupt fires.
+		 *
+		 * So the only way to run into this issue is if the delivery
+		 * of the interrupt on the apic/system bus would be delayed
+		 * beyond the point where the target cpu disables interrupts
+		 * in stop machine. I doubt that it can happen, but at least
+		 * there is a theroretical chance. Virtualization might be
+		 * able to expose this, but AFAICT the IOAPIC emulation is not
+		 * as stupid as the real hardware.
+		 *
+		 * Anyway, there is nothing we can do about that at this point
+		 * w/o refactoring the whole fixup_irq() business completely.
+		 * We print at least the irq number and the old vector number,
+		 * so we have the necessary information when a problem in that
+		 * area arises.
 		 */
-		data = apic_chip_data(irqdata);
-		if (!data)
-			return;
-		raw_spin_lock(&vector_lock);
+		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
+			irqdata->irq, cfg->old_vector);
 	}
+	/*
+	 * If old_domain is not empty, then other cpus still have the irq
+	 * descriptor set in their vector array. Clean it up.
+	 */
+	for_each_cpu(cpu, data->old_domain)
+		per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
+
+	/* Cleanup the left overs of the (half finished) move */
+	cpumask_clear(data->old_domain);
+	data->move_in_progress = 0;
 	raw_spin_unlock(&vector_lock);
 }
 #endif
diff --git a/arch/x86/kernel/apic/x2apic_uv_x.c b/arch/x86/kernel/apic/x2apic_uv_x.c
index 624db00583f4..8f4942e2bcbb 100644
--- a/arch/x86/kernel/apic/x2apic_uv_x.c
+++ b/arch/x86/kernel/apic/x2apic_uv_x.c
@@ -792,7 +792,8 @@ static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
 {
 	long cpu = (long)hcpu;
 
-	switch (action) {
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_DOWN_FAILED:
 	case CPU_ONLINE:
 		uv_heartbeat_enable(cpu);
 		break;
@@ -860,7 +861,7 @@ int uv_set_vga_state(struct pci_dev *pdev, bool decode,
  */
 void uv_cpu_init(void)
 {
-	/* CPU 0 initilization will be done via uv_system_init. */
+	/* CPU 0 initialization will be done via uv_system_init. */
 	if (!uv_blade_info)
 		return;