xen/events: defer eoi in case of excessive number of events

In case rogue guests are sending events at high frequency it might
happen that xen_evtchn_do_upcall() won't stop processing events in
dom0. As this is done in irq handling a crash might be the result.

In order to avoid that, delay further inter-domain events after some
time in xen_evtchn_do_upcall() by forcing eoi processing into a
worker on the same cpu, thus inhibiting new events coming in.

The time after which eoi processing is to be delayed is configurable
via a new module parameter "event_loop_timeout" which specifies the
maximum event loop time in jiffies (default: 2, the value was chosen
after some tests showing that a value of 2 was the lowest with an
only slight drop of dom0 network throughput while multiple guests
performed an event storm).

How long eoi processing will be delayed can be specified via another
parameter "event_eoi_delay" (again in jiffies, default 10, again the
value was chosen after testing with different delay values).

This is part of XSA-332.

Cc: stable@vger.kernel.org
Reported-by: Julien Grall <julien@xen.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Reviewed-by: Wei Liu <wl@xen.org>

1 files changed, 184 insertions, 5 deletions

diff --git a/drivers/xen/events/events_base.c b/drivers/xen/events/events_base.c
index 9cbfea5e9a08..cde096a6f11d 100644
--- a/drivers/xen/events/events_base.c
+++ b/drivers/xen/events/events_base.c
@@ -35,6 +35,8 @@
 #include <linux/pci.h>
 #include <linux/spinlock.h>
 #include <linux/cpuhotplug.h>
+#include <linux/atomic.h>
+#include <linux/ktime.h>
 
 #ifdef CONFIG_X86
 #include <asm/desc.h>
@@ -65,6 +67,15 @@
 
 #include "events_internal.h"
 
+#undef MODULE_PARAM_PREFIX
+#define MODULE_PARAM_PREFIX "xen."
+
+static uint __read_mostly event_loop_timeout = 2;
+module_param(event_loop_timeout, uint, 0644);
+
+static uint __read_mostly event_eoi_delay = 10;
+module_param(event_eoi_delay, uint, 0644);
+
 const struct evtchn_ops *evtchn_ops;
 
 /*
@@ -88,6 +99,7 @@ static DEFINE_RWLOCK(evtchn_rwlock);
  * irq_mapping_update_lock
  *   evtchn_rwlock
  *     IRQ-desc lock
+ *       percpu eoi_list_lock
  */
 
 static LIST_HEAD(xen_irq_list_head);
@@ -120,6 +132,8 @@ static struct irq_chip xen_pirq_chip;
 static void enable_dynirq(struct irq_data *data);
 static void disable_dynirq(struct irq_data *data);
 
+static DEFINE_PER_CPU(unsigned int, irq_epoch);
+
 static void clear_evtchn_to_irq_row(unsigned row)
 {
 	unsigned col;
@@ -399,17 +413,120 @@ void notify_remote_via_irq(int irq)
 }
 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 
+struct lateeoi_work {
+	struct delayed_work delayed;
+	spinlock_t eoi_list_lock;
+	struct list_head eoi_list;
+};
+
+static DEFINE_PER_CPU(struct lateeoi_work, lateeoi);
+
+static void lateeoi_list_del(struct irq_info *info)
+{
+	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
+	unsigned long flags;
+
+	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
+	list_del_init(&info->eoi_list);
+	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
+}
+
+static void lateeoi_list_add(struct irq_info *info)
+{
+	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
+	struct irq_info *elem;
+	u64 now = get_jiffies_64();
+	unsigned long delay;
+	unsigned long flags;
+
+	if (now < info->eoi_time)
+		delay = info->eoi_time - now;
+	else
+		delay = 1;
+
+	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
+
+	if (list_empty(&eoi->eoi_list)) {
+		list_add(&info->eoi_list, &eoi->eoi_list);
+		mod_delayed_work_on(info->eoi_cpu, system_wq,
+				    &eoi->delayed, delay);
+	} else {
+		list_for_each_entry_reverse(elem, &eoi->eoi_list, eoi_list) {
+			if (elem->eoi_time <= info->eoi_time)
+				break;
+		}
+		list_add(&info->eoi_list, &elem->eoi_list);
+	}
+
+	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
+}
+
 static void xen_irq_lateeoi_locked(struct irq_info *info)
 {
 	evtchn_port_t evtchn;
+	unsigned int cpu;
 
 	evtchn = info->evtchn;
-	if (!VALID_EVTCHN(evtchn))
+	if (!VALID_EVTCHN(evtchn) || !list_empty(&info->eoi_list))
 		return;
 
+	cpu = info->eoi_cpu;
+	if (info->eoi_time && info->irq_epoch == per_cpu(irq_epoch, cpu)) {
+		lateeoi_list_add(info);
+		return;
+	}
+
+	info->eoi_time = 0;
 	unmask_evtchn(evtchn);
 }
 
+static void xen_irq_lateeoi_worker(struct work_struct *work)
+{
+	struct lateeoi_work *eoi;
+	struct irq_info *info;
+	u64 now = get_jiffies_64();
+	unsigned long flags;
+
+	eoi = container_of(to_delayed_work(work), struct lateeoi_work, delayed);
+
+	read_lock_irqsave(&evtchn_rwlock, flags);
+
+	while (true) {
+		spin_lock(&eoi->eoi_list_lock);
+
+		info = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
+						eoi_list);
+
+		if (info == NULL || now < info->eoi_time) {
+			spin_unlock(&eoi->eoi_list_lock);
+			break;
+		}
+
+		list_del_init(&info->eoi_list);
+
+		spin_unlock(&eoi->eoi_list_lock);
+
+		info->eoi_time = 0;
+
+		xen_irq_lateeoi_locked(info);
+	}
+
+	if (info)
+		mod_delayed_work_on(info->eoi_cpu, system_wq,
+				    &eoi->delayed, info->eoi_time - now);
+
+	read_unlock_irqrestore(&evtchn_rwlock, flags);
+}
+
+static void xen_cpu_init_eoi(unsigned int cpu)
+{
+	struct lateeoi_work *eoi = &per_cpu(lateeoi, cpu);
+
+	INIT_DELAYED_WORK(&eoi->delayed, xen_irq_lateeoi_worker);
+	spin_lock_init(&eoi->eoi_list_lock);
+	INIT_LIST_HEAD(&eoi->eoi_list);
+}
+
 void xen_irq_lateeoi(unsigned int irq, unsigned int eoi_flags)
 {
 	struct irq_info *info;
@@ -429,6 +546,7 @@ EXPORT_SYMBOL_GPL(xen_irq_lateeoi);
 static void xen_irq_init(unsigned irq)
 {
 	struct irq_info *info;
+
 #ifdef CONFIG_SMP
 	/* By default all event channels notify CPU#0. */
 	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
@@ -443,6 +561,7 @@ static void xen_irq_init(unsigned irq)
 
 	set_info_for_irq(irq, info);
 
+	INIT_LIST_HEAD(&info->eoi_list);
 	list_add_tail(&info->list, &xen_irq_list_head);
 }
 
@@ -498,6 +617,9 @@ static void xen_free_irq(unsigned irq)
 
 	write_lock_irqsave(&evtchn_rwlock, flags);
 
+	if (!list_empty(&info->eoi_list))
+		lateeoi_list_del(info);
+
 	list_del(&info->list);
 
 	set_info_for_irq(irq, NULL);
@@ -1358,17 +1480,66 @@ void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
 	notify_remote_via_irq(irq);
 }
 
+struct evtchn_loop_ctrl {
+	ktime_t timeout;
+	unsigned count;
+	bool defer_eoi;
+};
+
+void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl)
+{
+	int irq;
+	struct irq_info *info;
+
+	irq = get_evtchn_to_irq(port);
+	if (irq == -1)
+		return;
+
+	/*
+	 * Check for timeout every 256 events.
+	 * We are setting the timeout value only after the first 256
+	 * events in order to not hurt the common case of few loop
+	 * iterations. The 256 is basically an arbitrary value.
+	 *
+	 * In case we are hitting the timeout we need to defer all further
+	 * EOIs in order to ensure to leave the event handling loop rather
+	 * sooner than later.
+	 */
+	if (!ctrl->defer_eoi && !(++ctrl->count & 0xff)) {
+		ktime_t kt = ktime_get();
+
+		if (!ctrl->timeout) {
+			kt = ktime_add_ms(kt,
+					  jiffies_to_msecs(event_loop_timeout));
+			ctrl->timeout = kt;
+		} else if (kt > ctrl->timeout) {
+			ctrl->defer_eoi = true;
+		}
+	}
+
+	info = info_for_irq(irq);
+
+	if (ctrl->defer_eoi) {
+		info->eoi_cpu = smp_processor_id();
+		info->irq_epoch = __this_cpu_read(irq_epoch);
+		info->eoi_time = get_jiffies_64() + event_eoi_delay;
+	}
+
+	generic_handle_irq(irq);
+}
+
 static void __xen_evtchn_do_upcall(void)
 {
 	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
 	int cpu = smp_processor_id();
+	struct evtchn_loop_ctrl ctrl = { 0 };
 
 	read_lock(&evtchn_rwlock);
 
 	do {
 		vcpu_info->evtchn_upcall_pending = 0;
 
-		xen_evtchn_handle_events(cpu);
+		xen_evtchn_handle_events(cpu, &ctrl);
 
 		BUG_ON(!irqs_disabled());
 
@@ -1377,6 +1548,13 @@ static void __xen_evtchn_do_upcall(void)
 	} while (vcpu_info->evtchn_upcall_pending);
 
 	read_unlock(&evtchn_rwlock);
+
+	/*
+	 * Increment irq_epoch only now to defer EOIs only for
+	 * xen_irq_lateeoi() invocations occurring from inside the loop
+	 * above.
+	 */
+	__this_cpu_inc(irq_epoch);
 }
 
 void xen_evtchn_do_upcall(struct pt_regs *regs)
@@ -1825,9 +2003,6 @@ void xen_setup_callback_vector(void) {}
 static inline void xen_alloc_callback_vector(void) {}
 #endif
 
-#undef MODULE_PARAM_PREFIX
-#define MODULE_PARAM_PREFIX "xen."
-
 static bool fifo_events = true;
 module_param(fifo_events, bool, 0);
 
@@ -1835,6 +2010,8 @@ static int xen_evtchn_cpu_prepare(unsigned int cpu)
 {
 	int ret = 0;
 
+	xen_cpu_init_eoi(cpu);
+
 	if (evtchn_ops->percpu_init)
 		ret = evtchn_ops->percpu_init(cpu);
 
@@ -1861,6 +2038,8 @@ void __init xen_init_IRQ(void)
 	if (ret < 0)
 		xen_evtchn_2l_init();
 
+	xen_cpu_init_eoi(smp_processor_id());
+
 	cpuhp_setup_state_nocalls(CPUHP_XEN_EVTCHN_PREPARE,
 				  "xen/evtchn:prepare",
 				  xen_evtchn_cpu_prepare, xen_evtchn_cpu_dead);