um: Implement time-travel=ext

This implements synchronized time-travel mode which - using a special
application on a unix socket - lets multiple machines take part in a
time-travelling simulation together.

The protocol for the unix domain socket is defined in the new file
include/uapi/linux/um_timetravel.h.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Richard Weinberger <richard@nod.at>

arch/um/drivers/virtio_uml.c

@@ -26,6 +26,7 @@
 #include <linux/virtio.h>
 #include <linux/virtio_config.h>
 #include <linux/virtio_ring.h>
+#include <linux/time-internal.h>
 #include <shared/as-layout.h>
 #include <irq_kern.h>
 #include <init.h>
@@ -64,6 +65,11 @@ struct virtio_uml_device {
 struct virtio_uml_vq_info {
 	int kick_fd, call_fd;
 	char name[32];
+#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
+	struct virtqueue *vq;
+	vq_callback_t *callback;
+	struct time_travel_event defer;
+#endif
 };
 
 extern unsigned long long physmem_size, highmem;
@@ -118,10 +124,27 @@ static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
 
 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
 			   int fd, struct vhost_user_msg *msg,
-			   size_t max_payload_size)
+			   size_t max_payload_size, bool wait)
 {
 	size_t size;
-	int rc = vhost_user_recv_header(fd, msg);
+	int rc;
+
+	/*
+	 * In virtio time-travel mode, we're handling all the vhost-user
+	 * FDs by polling them whenever appropriate. However, we may get
+	 * into a situation where we're sending out an interrupt message
+	 * to a device (e.g. a net device) and need to handle a simulation
+	 * time message while doing so, e.g. one that tells us to update
+	 * our idea of how long we can run without scheduling.
+	 *
+	 * Thus, we need to not just read() from the given fd, but need
+	 * to also handle messages for the simulation time - this function
+	 * does that for us while waiting for the given fd to be readable.
+	 */
+	if (wait)
+		time_travel_wait_readable(fd);
+
+	rc = vhost_user_recv_header(fd, msg);
 
 	if (rc == -ECONNRESET && vu_dev->registered) {
 		struct virtio_uml_platform_data *pdata;
@@ -143,7 +166,8 @@ static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
 				struct vhost_user_msg *msg,
 				size_t max_payload_size)
 {
-	int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg, max_payload_size);
+	int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg,
+				 max_payload_size, true);
 
 	if (rc)
 		return rc;
@@ -173,7 +197,8 @@ static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
 			       struct vhost_user_msg *msg,
 			       size_t max_payload_size)
 {
-	int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg, max_payload_size);
+	int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg,
+				 max_payload_size, false);
 
 	if (rc)
 		return rc;
@@ -700,6 +725,8 @@ static bool vu_notify(struct virtqueue *vq)
 	const uint64_t n = 1;
 	int rc;
 
+	time_travel_propagate_time();
+
 	if (info->kick_fd < 0) {
 		struct virtio_uml_device *vu_dev;
 
@@ -847,6 +874,23 @@ static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
 	return rc;
 }
 
+#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
+static void vu_defer_irq_handle(struct time_travel_event *d)
+{
+	struct virtio_uml_vq_info *info;
+
+	info = container_of(d, struct virtio_uml_vq_info, defer);
+	info->callback(info->vq);
+}
+
+static void vu_defer_irq_callback(struct virtqueue *vq)
+{
+	struct virtio_uml_vq_info *info = vq->priv;
+
+	time_travel_add_irq_event(&info->defer);
+}
+#endif
+
 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
 				     unsigned index, vq_callback_t *callback,
 				     const char *name, bool ctx)
@@ -866,6 +910,19 @@ static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
 	snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
 		 pdev->id, name);
 
+#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
+	/*
+	 * When we get an interrupt, we must bounce it through the simulation
+	 * calendar (the simtime device), except for the simtime device itself
+	 * since that's part of the simulation control.
+	 */
+	if (time_travel_mode == TT_MODE_EXTERNAL && callback) {
+		info->callback = callback;
+		callback = vu_defer_irq_callback;
+		time_travel_set_event_fn(&info->defer, vu_defer_irq_handle);
+	}
+#endif
+
 	vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
 				    ctx, vu_notify, callback, info->name);
 	if (!vq) {
@@ -874,6 +931,9 @@ static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
 	}
 	vq->priv = info;
 	num = virtqueue_get_vring_size(vq);
+#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
+	info->vq = vq;
+#endif
 
 	if (vu_dev->protocol_features &
 			BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS)) {

arch/um/include/linux/time-internal.h

@@ -15,6 +15,7 @@ enum time_travel_mode {
 	TT_MODE_OFF,
 	TT_MODE_BASIC,
 	TT_MODE_INFCPU,
+	TT_MODE_EXTERNAL,
 };
 
 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
@@ -35,6 +36,24 @@ time_travel_set_event_fn(struct time_travel_event *e,
 {
 	e->fn = fn;
 }
+
+void __time_travel_propagate_time(void);
+
+static inline void time_travel_propagate_time(void)
+{
+	if (time_travel_mode == TT_MODE_EXTERNAL)
+		__time_travel_propagate_time();
+}
+
+void __time_travel_wait_readable(int fd);
+
+static inline void time_travel_wait_readable(int fd)
+{
+	if (time_travel_mode == TT_MODE_EXTERNAL)
+		__time_travel_wait_readable(fd);
+}
+
+void time_travel_add_irq_event(struct time_travel_event *e);
 #else
 struct time_travel_event {
 };
@@ -47,5 +66,13 @@ static inline void time_travel_sleep(unsigned long long duration)
 
 /* this is a macro so the event/function need not exist */
 #define time_travel_set_event_fn(e, fn) do {} while (0)
+
+static inline void time_travel_propagate_time(void)
+{
+}
+
+static inline void time_travel_wait_readable(int fd)
+{
+}
 #endif /* CONFIG_UML_TIME_TRAVEL_SUPPORT */
 #endif /* __TIMER_INTERNAL_H__ */

arch/um/include/shared/os.h

@@ -181,6 +181,7 @@ extern int os_falloc_punch(int fd, unsigned long long offset, int count);
 extern int os_eventfd(unsigned int initval, int flags);
 extern int os_sendmsg_fds(int fd, const void *buf, unsigned int len,
 			  const int *fds, unsigned int fds_num);
+int os_poll(unsigned int n, const int *fds);
 
 /* start_up.c */
 extern void os_early_checks(void);

arch/um/kernel/skas/syscall.c

@@ -24,7 +24,8 @@ void handle_syscall(struct uml_pt_regs *r)
 	 * went to sleep, even if said userspace interacts with the kernel in
 	 * various ways.
 	 */
-	if (time_travel_mode == TT_MODE_INFCPU)
+	if (time_travel_mode == TT_MODE_INFCPU ||
+	    time_travel_mode == TT_MODE_EXTERNAL)
 		schedule();
 
 	/* Initialize the syscall number and default return value. */