Merge tag 'uml-for-linus-6.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/uml/linux

Pull UML updates from Richard Weinberger:

 - Removal of dead code (TT mode leftovers, etc)

 - Fixes for the network vector driver

 - Fixes for time-travel mode

* tag 'uml-for-linus-6.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/uml/linux:
  um: fix time-travel syscall scheduling hack
  um: Remove outdated asm/sysrq.h header
  um: Remove the declaration of user_thread function
  um: Remove the call to SUBARCH_EXECVE1 macro
  um: Remove unused mm_fd field from mm_id
  um: Remove unused fields from thread_struct
  um: Remove the redundant newpage check in update_pte_range
  um: Remove unused kpte_clear_flush macro
  um: Remove obsoleted declaration for execute_syscall_skas
  user_mode_linux_howto_v2: add VDE vector support in doc
  vector_user: add VDE support
  um: remove ARCH_NO_PREEMPT_DYNAMIC
  um: vector: Fix NAPI budget handling
  um: vector: Replace locks guarding queue depth with atomics
  um: remove variable stack array in os_rcv_fd_msg()

3 files changed, 195 insertions, 104 deletions

diff --git a/arch/um/drivers/vector_kern.c b/arch/um/drivers/vector_kern.c
index 2d473282ab51..c992da83268d 100644
--- a/arch/um/drivers/vector_kern.c
+++ b/arch/um/drivers/vector_kern.c
@@ -22,6 +22,7 @@
 #include <linux/interrupt.h>
 #include <linux/firmware.h>
 #include <linux/fs.h>
+#include <asm/atomic.h>
 #include <uapi/linux/filter.h>
 #include <init.h>
 #include <irq_kern.h>
@@ -102,18 +103,33 @@ static const struct {
 
 static void vector_reset_stats(struct vector_private *vp)
 {
+	/* We reuse the existing queue locks for stats */
+
+	/* RX stats are modified with RX head_lock held
+	 * in vector_poll.
+	 */
+
+	spin_lock(&vp->rx_queue->head_lock);
 	vp->estats.rx_queue_max = 0;
 	vp->estats.rx_queue_running_average = 0;
-	vp->estats.tx_queue_max = 0;
-	vp->estats.tx_queue_running_average = 0;
 	vp->estats.rx_encaps_errors = 0;
+	vp->estats.sg_ok = 0;
+	vp->estats.sg_linearized = 0;
+	spin_unlock(&vp->rx_queue->head_lock);
+
+	/* TX stats are modified with TX head_lock held
+	 * in vector_send.
+	 */
+
+	spin_lock(&vp->tx_queue->head_lock);
 	vp->estats.tx_timeout_count = 0;
 	vp->estats.tx_restart_queue = 0;
 	vp->estats.tx_kicks = 0;
 	vp->estats.tx_flow_control_xon = 0;
 	vp->estats.tx_flow_control_xoff = 0;
-	vp->estats.sg_ok = 0;
-	vp->estats.sg_linearized = 0;
+	vp->estats.tx_queue_max = 0;
+	vp->estats.tx_queue_running_average = 0;
+	spin_unlock(&vp->tx_queue->head_lock);
 }
 
 static int get_mtu(struct arglist *def)
@@ -232,12 +248,6 @@ static int get_transport_options(struct arglist *def)
 
 static char *drop_buffer;
 
-/* Array backed queues optimized for bulk enqueue/dequeue and
- * 1:N (small values of N) or 1:1 enqueuer/dequeuer ratios.
- * For more details and full design rationale see
- * http://foswiki.cambridgegreys.com/Main/EatYourTailAndEnjoyIt
- */
-
 
 /*
  * Advance the mmsg queue head by n = advance. Resets the queue to
@@ -247,27 +257,13 @@ static char *drop_buffer;
 
 static int vector_advancehead(struct vector_queue *qi, int advance)
 {
-	int queue_depth;
-
 	qi->head =
 		(qi->head + advance)
 			% qi->max_depth;
 
 
-	spin_lock(&qi->tail_lock);
-	qi->queue_depth -= advance;
-
-	/* we are at 0, use this to
-	 * reset head and tail so we can use max size vectors
-	 */
-
-	if (qi->queue_depth == 0) {
-		qi->head = 0;
-		qi->tail = 0;
-	}
-	queue_depth = qi->queue_depth;
-	spin_unlock(&qi->tail_lock);
-	return queue_depth;
+	atomic_sub(advance, &qi->queue_depth);
+	return atomic_read(&qi->queue_depth);
 }
 
 /*	Advance the queue tail by n = advance.
@@ -277,16 +273,11 @@ static int vector_advancehead(struct vector_queue *qi, int advance)
 
 static int vector_advancetail(struct vector_queue *qi, int advance)
 {
-	int queue_depth;
-
 	qi->tail =
 		(qi->tail + advance)
 			% qi->max_depth;
-	spin_lock(&qi->head_lock);
-	qi->queue_depth += advance;
-	queue_depth = qi->queue_depth;
-	spin_unlock(&qi->head_lock);
-	return queue_depth;
+	atomic_add(advance, &qi->queue_depth);
+	return atomic_read(&qi->queue_depth);
 }
 
 static int prep_msg(struct vector_private *vp,
@@ -339,9 +330,7 @@ static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
 	int iov_count;
 
 	spin_lock(&qi->tail_lock);
-	spin_lock(&qi->head_lock);
-	queue_depth = qi->queue_depth;
-	spin_unlock(&qi->head_lock);
+	queue_depth = atomic_read(&qi->queue_depth);
 
 	if (skb)
 		packet_len = skb->len;
@@ -360,6 +349,7 @@ static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
 		mmsg_vector->msg_hdr.msg_iovlen = iov_count;
 		mmsg_vector->msg_hdr.msg_name = vp->fds->remote_addr;
 		mmsg_vector->msg_hdr.msg_namelen = vp->fds->remote_addr_size;
+		wmb(); /* Make the packet visible to the NAPI poll thread */
 		queue_depth = vector_advancetail(qi, 1);
 	} else
 		goto drop;
@@ -398,7 +388,7 @@ static int consume_vector_skbs(struct vector_queue *qi, int count)
 }
 
 /*
- * Generic vector deque via sendmmsg with support for forming headers
+ * Generic vector dequeue via sendmmsg with support for forming headers
  * using transport specific callback. Allows GRE, L2TPv3, RAW and
  * other transports to use a common dequeue procedure in vector mode
  */
@@ -408,69 +398,64 @@ static int vector_send(struct vector_queue *qi)
 {
 	struct vector_private *vp = netdev_priv(qi->dev);
 	struct mmsghdr *send_from;
-	int result = 0, send_len, queue_depth = qi->max_depth;
+	int result = 0, send_len;
 
 	if (spin_trylock(&qi->head_lock)) {
-		if (spin_trylock(&qi->tail_lock)) {
-			/* update queue_depth to current value */
-			queue_depth = qi->queue_depth;
-			spin_unlock(&qi->tail_lock);
-			while (queue_depth > 0) {
-				/* Calculate the start of the vector */
-				send_len = queue_depth;
-				send_from = qi->mmsg_vector;
-				send_from += qi->head;
-				/* Adjust vector size if wraparound */
-				if (send_len + qi->head > qi->max_depth)
-					send_len = qi->max_depth - qi->head;
-				/* Try to TX as many packets as possible */
-				if (send_len > 0) {
-					result = uml_vector_sendmmsg(
-						 vp->fds->tx_fd,
-						 send_from,
-						 send_len,
-						 0
-					);
-					vp->in_write_poll =
-						(result != send_len);
-				}
-				/* For some of the sendmmsg error scenarios
-				 * we may end being unsure in the TX success
-				 * for all packets. It is safer to declare
-				 * them all TX-ed and blame the network.
-				 */
-				if (result < 0) {
-					if (net_ratelimit())
-						netdev_err(vp->dev, "sendmmsg err=%i\n",
-							result);
-					vp->in_error = true;
-					result = send_len;
-				}
-				if (result > 0) {
-					queue_depth =
-						consume_vector_skbs(qi, result);
-					/* This is equivalent to an TX IRQ.
-					 * Restart the upper layers to feed us
-					 * more packets.
-					 */
-					if (result > vp->estats.tx_queue_max)
-						vp->estats.tx_queue_max = result;
-					vp->estats.tx_queue_running_average =
-						(vp->estats.tx_queue_running_average + result) >> 1;
-				}
-				netif_wake_queue(qi->dev);
-				/* if TX is busy, break out of the send loop,
-				 *  poll write IRQ will reschedule xmit for us
+		/* update queue_depth to current value */
+		while (atomic_read(&qi->queue_depth) > 0) {
+			/* Calculate the start of the vector */
+			send_len = atomic_read(&qi->queue_depth);
+			send_from = qi->mmsg_vector;
+			send_from += qi->head;
+			/* Adjust vector size if wraparound */
+			if (send_len + qi->head > qi->max_depth)
+				send_len = qi->max_depth - qi->head;
+			/* Try to TX as many packets as possible */
+			if (send_len > 0) {
+				result = uml_vector_sendmmsg(
+					 vp->fds->tx_fd,
+					 send_from,
+					 send_len,
+					 0
+				);
+				vp->in_write_poll =
+					(result != send_len);
+			}
+			/* For some of the sendmmsg error scenarios
+			 * we may end being unsure in the TX success
+			 * for all packets. It is safer to declare
+			 * them all TX-ed and blame the network.
+			 */
+			if (result < 0) {
+				if (net_ratelimit())
+					netdev_err(vp->dev, "sendmmsg err=%i\n",
+						result);
+				vp->in_error = true;
+				result = send_len;
+			}
+			if (result > 0) {
+				consume_vector_skbs(qi, result);
+				/* This is equivalent to an TX IRQ.
+				 * Restart the upper layers to feed us
+				 * more packets.
 				 */
-				if (result != send_len) {
-					vp->estats.tx_restart_queue++;
-					break;
-				}
+				if (result > vp->estats.tx_queue_max)
+					vp->estats.tx_queue_max = result;
+				vp->estats.tx_queue_running_average =
+					(vp->estats.tx_queue_running_average + result) >> 1;
+			}
+			netif_wake_queue(qi->dev);
+			/* if TX is busy, break out of the send loop,
+			 *  poll write IRQ will reschedule xmit for us.
+			 */
+			if (result != send_len) {
+				vp->estats.tx_restart_queue++;
+				break;
 			}
 		}
 		spin_unlock(&qi->head_lock);
 	}
-	return queue_depth;
+	return atomic_read(&qi->queue_depth);
 }
 
 /* Queue destructor. Deliberately stateless so we can use
@@ -589,7 +574,7 @@ static struct vector_queue *create_queue(
 	}
 	spin_lock_init(&result->head_lock);
 	spin_lock_init(&result->tail_lock);
-	result->queue_depth = 0;
+	atomic_set(&result->queue_depth, 0);
 	result->head = 0;
 	result->tail = 0;
 	return result;
@@ -668,18 +653,27 @@ done:
 }
 
 
-/* Prepare queue for recvmmsg one-shot rx - fill with fresh sk_buffs*/
+/* Prepare queue for recvmmsg one-shot rx - fill with fresh sk_buffs */
 
 static void prep_queue_for_rx(struct vector_queue *qi)
 {
 	struct vector_private *vp = netdev_priv(qi->dev);
 	struct mmsghdr *mmsg_vector = qi->mmsg_vector;
 	void **skbuff_vector = qi->skbuff_vector;
-	int i;
+	int i, queue_depth;
+
+	queue_depth = atomic_read(&qi->queue_depth);
 
-	if (qi->queue_depth == 0)
+	if (queue_depth == 0)
 		return;
-	for (i = 0; i < qi->queue_depth; i++) {
+
+	/* RX is always emptied 100% during each cycle, so we do not
+	 * have to do the tail wraparound math for it.
+	 */
+
+	qi->head = qi->tail = 0;
+
+	for (i = 0; i < queue_depth; i++) {
 		/* it is OK if allocation fails - recvmmsg with NULL data in
 		 * iov argument still performs an RX, just drops the packet
 		 * This allows us stop faffing around with a "drop buffer"
@@ -689,7 +683,7 @@ static void prep_queue_for_rx(struct vector_queue *qi)
 		skbuff_vector++;
 		mmsg_vector++;
 	}
-	qi->queue_depth = 0;
+	atomic_set(&qi->queue_depth, 0);
 }
 
 static struct vector_device *find_device(int n)
@@ -972,7 +966,7 @@ static int vector_mmsg_rx(struct vector_private *vp, int budget)
 		budget = qi->max_depth;
 
 	packet_count = uml_vector_recvmmsg(
-		vp->fds->rx_fd, qi->mmsg_vector, qi->max_depth, 0);
+		vp->fds->rx_fd, qi->mmsg_vector, budget, 0);
 
 	if (packet_count < 0)
 		vp->in_error = true;
@@ -985,7 +979,7 @@ static int vector_mmsg_rx(struct vector_private *vp, int budget)
 	 * many do we need to prep the next time prep_queue_for_rx() is called.
 	 */
 
-	qi->queue_depth = packet_count;
+	atomic_add(packet_count, &qi->queue_depth);
 
 	for (i = 0; i < packet_count; i++) {
 		skb = (*skbuff_vector);
@@ -1172,6 +1166,7 @@ static int vector_poll(struct napi_struct *napi, int budget)
 
 	if ((vp->options & VECTOR_TX) != 0)
 		tx_enqueued = (vector_send(vp->tx_queue) > 0);
+	spin_lock(&vp->rx_queue->head_lock);
 	if ((vp->options & VECTOR_RX) > 0)
 		err = vector_mmsg_rx(vp, budget);
 	else {
@@ -1179,12 +1174,13 @@ static int vector_poll(struct napi_struct *napi, int budget)
 		if (err > 0)
 			err = 1;
 	}
+	spin_unlock(&vp->rx_queue->head_lock);
 	if (err > 0)
 		work_done += err;
 
 	if (tx_enqueued || err > 0)
 		napi_schedule(napi);
-	if (work_done < budget)
+	if (work_done <= budget)
 		napi_complete_done(napi, work_done);
 	return work_done;
 }
@@ -1225,7 +1221,7 @@ static int vector_net_open(struct net_device *dev)
 			vp->rx_header_size,
 			MAX_IOV_SIZE
 		);
-		vp->rx_queue->queue_depth = get_depth(vp->parsed);
+		atomic_set(&vp->rx_queue->queue_depth, get_depth(vp->parsed));
 	} else {
 		vp->header_rxbuffer = kmalloc(
 			vp->rx_header_size,
@@ -1467,7 +1463,17 @@ static void vector_get_ethtool_stats(struct net_device *dev,
 {
 	struct vector_private *vp = netdev_priv(dev);
 
+	/* Stats are modified in the dequeue portions of
+	 * rx/tx which are protected by the head locks
+	 * grabbing these locks here ensures they are up
+	 * to date.
+	 */
+
+	spin_lock(&vp->tx_queue->head_lock);
+	spin_lock(&vp->rx_queue->head_lock);
 	memcpy(tmp_stats, &vp->estats, sizeof(struct vector_estats));
+	spin_unlock(&vp->rx_queue->head_lock);
+	spin_unlock(&vp->tx_queue->head_lock);
 }
 
 static int vector_get_coalesce(struct net_device *netdev,
diff --git a/arch/um/drivers/vector_kern.h b/arch/um/drivers/vector_kern.h
index 806df551be0b..417834793658 100644
--- a/arch/um/drivers/vector_kern.h
+++ b/arch/um/drivers/vector_kern.h
@@ -14,6 +14,7 @@
 #include <linux/ctype.h>
 #include <linux/workqueue.h>
 #include <linux/interrupt.h>
+#include <asm/atomic.h>
 
 #include "vector_user.h"
 
@@ -44,7 +45,8 @@ struct vector_queue {
 	struct net_device *dev;
 	spinlock_t head_lock;
 	spinlock_t tail_lock;
-	int queue_depth, head, tail, max_depth, max_iov_frags;
+	atomic_t queue_depth;
+	int head, tail, max_depth, max_iov_frags;
 	short options;
 };
 
diff --git a/arch/um/drivers/vector_user.c b/arch/um/drivers/vector_user.c
index b16a5e5619d3..2ea67e6fd067 100644
--- a/arch/um/drivers/vector_user.c
+++ b/arch/um/drivers/vector_user.c
@@ -46,6 +46,9 @@
 #define TRANS_FD "fd"
 #define TRANS_FD_LEN strlen(TRANS_FD)
 
+#define TRANS_VDE "vde"
+#define TRANS_VDE_LEN strlen(TRANS_VDE)
+
 #define VNET_HDR_FAIL "could not enable vnet headers on fd %d"
 #define TUN_GET_F_FAIL "tapraw: TUNGETFEATURES failed: %s"
 #define L2TPV3_BIND_FAIL "l2tpv3_open : could not bind socket err=%i"
@@ -434,6 +437,84 @@ fd_cleanup:
 	return NULL;
 }
 
+/* enough char to store an int type */
+#define ENOUGH(type) ((CHAR_BIT * sizeof(type) - 1) / 3 + 2)
+#define ENOUGH_OCTAL(type) ((CHAR_BIT * sizeof(type) + 2) / 3)
+/* vde_plug --descr xx --port2 xx --mod2 xx --group2 xx seqpacket://NN vnl (NULL) */
+#define VDE_MAX_ARGC 12
+#define VDE_SEQPACKET_HEAD "seqpacket://"
+#define VDE_SEQPACKET_HEAD_LEN (sizeof(VDE_SEQPACKET_HEAD) - 1)
+#define VDE_DEFAULT_DESCRIPTION "UML"
+
+static struct vector_fds *user_init_vde_fds(struct arglist *ifspec)
+{
+	char seqpacketvnl[VDE_SEQPACKET_HEAD_LEN + ENOUGH(int) + 1];
+	char *argv[VDE_MAX_ARGC] = {"vde_plug"};
+	int argc = 1;
+	int rv;
+	int sv[2];
+	struct vector_fds *result = NULL;
+
+	char *vnl = uml_vector_fetch_arg(ifspec,"vnl");
+	char *descr = uml_vector_fetch_arg(ifspec,"descr");
+	char *port = uml_vector_fetch_arg(ifspec,"port");
+	char *mode = uml_vector_fetch_arg(ifspec,"mode");
+	char *group = uml_vector_fetch_arg(ifspec,"group");
+	if (descr == NULL) descr = VDE_DEFAULT_DESCRIPTION;
+
+	argv[argc++] = "--descr";
+	argv[argc++] = descr;
+	if (port != NULL) {
+		argv[argc++] = "--port2";
+		argv[argc++] = port;
+	}
+	if (mode != NULL) {
+		argv[argc++] = "--mod2";
+		argv[argc++] = mode;
+	}
+	if (group != NULL) {
+		argv[argc++] = "--group2";
+		argv[argc++] = group;
+	}
+	argv[argc++] = seqpacketvnl;
+	argv[argc++] = vnl;
+	argv[argc++] = NULL;
+
+	rv = socketpair(AF_UNIX, SOCK_SEQPACKET, 0, sv);
+	if (rv  < 0) {
+		printk(UM_KERN_ERR "vde: seqpacket socketpair err %d", -errno);
+		return NULL;
+	}
+	rv = os_set_exec_close(sv[0]);
+	if (rv  < 0) {
+		printk(UM_KERN_ERR "vde: seqpacket socketpair cloexec err %d", -errno);
+		goto vde_cleanup_sv;
+	}
+	snprintf(seqpacketvnl, sizeof(seqpacketvnl), VDE_SEQPACKET_HEAD "%d", sv[1]);
+
+	run_helper(NULL, NULL, argv);
+
+	close(sv[1]);
+
+	result = uml_kmalloc(sizeof(struct vector_fds), UM_GFP_KERNEL);
+	if (result == NULL) {
+		printk(UM_KERN_ERR "fd open: allocation failed");
+		goto vde_cleanup;
+	}
+
+	result->rx_fd = sv[0];
+	result->tx_fd = sv[0];
+	result->remote_addr_size = 0;
+	result->remote_addr = NULL;
+	return result;
+
+vde_cleanup_sv:
+	close(sv[1]);
+vde_cleanup:
+	close(sv[0]);
+	return NULL;
+}
+
 static struct vector_fds *user_init_raw_fds(struct arglist *ifspec)
 {
 	int rxfd = -1, txfd = -1;
@@ -673,6 +754,8 @@ struct vector_fds *uml_vector_user_open(
 		return user_init_unix_fds(parsed, ID_BESS);
 	if (strncmp(transport, TRANS_FD, TRANS_FD_LEN) == 0)
 		return user_init_fd_fds(parsed);
+	if (strncmp(transport, TRANS_VDE, TRANS_VDE_LEN) == 0)
+		return user_init_vde_fds(parsed);
 	return NULL;
 }