New upstream version 17.11.2upstream/17.11.2

Change-Id: Iccf80e68222e7fc45af045bfed265a962b5ccca6
Signed-off-by: Luca Boccassi <luca.boccassi@gmail.com>

2 files changed, 130 insertions, 20 deletions

diff --git a/examples/vhost/virtio_net.c b/examples/vhost/virtio_net.c
index 1ab57f52..31c3dd06 100644
--- a/examples/vhost/virtio_net.c
+++ b/examples/vhost/virtio_net.c
@@ -85,16 +85,20 @@ enqueue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 	    struct rte_mbuf *m, uint16_t desc_idx)
 {
 	uint32_t desc_avail, desc_offset;
+	uint64_t desc_chunck_len;
 	uint32_t mbuf_avail, mbuf_offset;
 	uint32_t cpy_len;
 	struct vring_desc *desc;
-	uint64_t desc_addr;
+	uint64_t desc_addr, desc_gaddr;
 	struct virtio_net_hdr virtio_hdr = {0, 0, 0, 0, 0, 0};
 	/* A counter to avoid desc dead loop chain */
 	uint16_t nr_desc = 1;
 
 	desc = &vr->desc[desc_idx];
-	desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+	desc_chunck_len = desc->len;
+	desc_gaddr = desc->addr;
+	desc_addr = rte_vhost_va_from_guest_pa(
+			dev->mem, desc_gaddr, &desc_chunck_len);
 	/*
 	 * Checking of 'desc_addr' placed outside of 'unlikely' macro to avoid
 	 * performance issue with some versions of gcc (4.8.4 and 5.3.0) which
@@ -106,9 +110,42 @@ enqueue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 	rte_prefetch0((void *)(uintptr_t)desc_addr);
 
 	/* write virtio-net header */
-	*(struct virtio_net_hdr *)(uintptr_t)desc_addr = virtio_hdr;
+	if (likely(desc_chunck_len >= dev->hdr_len)) {
+		*(struct virtio_net_hdr *)(uintptr_t)desc_addr = virtio_hdr;
+		desc_offset = dev->hdr_len;
+	} else {
+		uint64_t len;
+		uint64_t remain = dev->hdr_len;
+		uint64_t src = (uint64_t)(uintptr_t)&virtio_hdr, dst;
+		uint64_t guest_addr = desc_gaddr;
+
+		while (remain) {
+			len = remain;
+			dst = rte_vhost_va_from_guest_pa(dev->mem,
+					guest_addr, &len);
+			if (unlikely(!dst || !len))
+				return -1;
+
+			rte_memcpy((void *)(uintptr_t)dst,
+					(void *)(uintptr_t)src,
+					len);
+
+			remain -= len;
+			guest_addr += len;
+			dst += len;
+		}
+
+		desc_chunck_len = desc->len - dev->hdr_len;
+		desc_gaddr += dev->hdr_len;
+		desc_addr = rte_vhost_va_from_guest_pa(
+				dev->mem, desc_gaddr,
+				&desc_chunck_len);
+		if (unlikely(!desc_addr))
+			return -1;
+
+		desc_offset = 0;
+	}
 
-	desc_offset = dev->hdr_len;
 	desc_avail  = desc->len - dev->hdr_len;
 
 	mbuf_avail  = rte_pktmbuf_data_len(m);
@@ -133,15 +170,28 @@ enqueue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 				return -1;
 
 			desc = &vr->desc[desc->next];
-			desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+			desc_chunck_len = desc->len;
+			desc_gaddr = desc->addr;
+			desc_addr = rte_vhost_va_from_guest_pa(
+					dev->mem, desc_gaddr, &desc_chunck_len);
 			if (unlikely(!desc_addr))
 				return -1;
 
 			desc_offset = 0;
 			desc_avail  = desc->len;
+		} else if (unlikely(desc_chunck_len == 0)) {
+			desc_chunck_len = desc_avail;
+			desc_gaddr += desc_offset;
+			desc_addr = rte_vhost_va_from_guest_pa(dev->mem,
+					desc_gaddr,
+					&desc_chunck_len);
+			if (unlikely(!desc_addr))
+				return -1;
+
+			desc_offset = 0;
 		}
 
-		cpy_len = RTE_MIN(desc_avail, mbuf_avail);
+		cpy_len = RTE_MIN(desc_chunck_len, mbuf_avail);
 		rte_memcpy((void *)((uintptr_t)(desc_addr + desc_offset)),
 			rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
 			cpy_len);
@@ -150,6 +200,7 @@ enqueue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 		mbuf_offset += cpy_len;
 		desc_avail  -= cpy_len;
 		desc_offset += cpy_len;
+		desc_chunck_len -= cpy_len;
 	}
 
 	return 0;
@@ -223,8 +274,9 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 	    struct rte_mempool *mbuf_pool)
 {
 	struct vring_desc *desc;
-	uint64_t desc_addr;
+	uint64_t desc_addr, desc_gaddr;
 	uint32_t desc_avail, desc_offset;
+	uint64_t desc_chunck_len;
 	uint32_t mbuf_avail, mbuf_offset;
 	uint32_t cpy_len;
 	struct rte_mbuf *cur = m, *prev = m;
@@ -236,7 +288,10 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 			(desc->flags & VRING_DESC_F_INDIRECT))
 		return -1;
 
-	desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+	desc_chunck_len = desc->len;
+	desc_gaddr = desc->addr;
+	desc_addr = rte_vhost_va_from_guest_pa(
+			dev->mem, desc_gaddr, &desc_chunck_len);
 	if (unlikely(!desc_addr))
 		return -1;
 
@@ -250,7 +305,10 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 	 * header.
 	 */
 	desc = &vr->desc[desc->next];
-	desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+	desc_chunck_len = desc->len;
+	desc_gaddr = desc->addr;
+	desc_addr = rte_vhost_va_from_guest_pa(
+			dev->mem, desc_gaddr, &desc_chunck_len);
 	if (unlikely(!desc_addr))
 		return -1;
 	rte_prefetch0((void *)(uintptr_t)desc_addr);
@@ -262,7 +320,7 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 	mbuf_offset = 0;
 	mbuf_avail  = m->buf_len - RTE_PKTMBUF_HEADROOM;
 	while (1) {
-		cpy_len = RTE_MIN(desc_avail, mbuf_avail);
+		cpy_len = RTE_MIN(desc_chunck_len, mbuf_avail);
 		rte_memcpy(rte_pktmbuf_mtod_offset(cur, void *,
 						   mbuf_offset),
 			(void *)((uintptr_t)(desc_addr + desc_offset)),
@@ -272,6 +330,7 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 		mbuf_offset += cpy_len;
 		desc_avail  -= cpy_len;
 		desc_offset += cpy_len;
+		desc_chunck_len -= cpy_len;
 
 		/* This desc reaches to its end, get the next one */
 		if (desc_avail == 0) {
@@ -283,13 +342,26 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
 				return -1;
 			desc = &vr->desc[desc->next];
 
-			desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+			desc_chunck_len = desc->len;
+			desc_gaddr = desc->addr;
+			desc_addr = rte_vhost_va_from_guest_pa(
+					dev->mem, desc_gaddr, &desc_chunck_len);
 			if (unlikely(!desc_addr))
 				return -1;
 			rte_prefetch0((void *)(uintptr_t)desc_addr);
 
 			desc_offset = 0;
 			desc_avail  = desc->len;
+		} else if (unlikely(desc_chunck_len == 0)) {
+			desc_chunck_len = desc_avail;
+			desc_gaddr += desc_offset;
+			desc_addr = rte_vhost_va_from_guest_pa(dev->mem,
+					desc_gaddr,
+					&desc_chunck_len);
+			if (unlikely(!desc_addr))
+				return -1;
+
+			desc_offset = 0;
 		}
 
 		/*
diff --git a/examples/vhost_scsi/vhost_scsi.c b/examples/vhost_scsi/vhost_scsi.c
index b4f1f8d2..b40f9936 100644
--- a/examples/vhost_scsi/vhost_scsi.c
+++ b/examples/vhost_scsi/vhost_scsi.c
@@ -68,7 +68,7 @@ vhost_scsi_ctrlr_find(__rte_unused const char *ctrlr_name)
 	return g_vhost_ctrlr;
 }
 
-static uint64_t gpa_to_vva(int vid, uint64_t gpa)
+static uint64_t gpa_to_vva(int vid, uint64_t gpa, uint64_t *len)
 {
 	char path[PATH_MAX];
 	struct vhost_scsi_ctrlr *ctrlr;
@@ -88,7 +88,7 @@ static uint64_t gpa_to_vva(int vid, uint64_t gpa)
 
 	assert(ctrlr->mem != NULL);
 
-	return rte_vhost_gpa_to_vva(ctrlr->mem, gpa);
+	return rte_vhost_va_from_guest_pa(ctrlr->mem, gpa, len);
 }
 
 static struct vring_desc *
@@ -138,15 +138,29 @@ static void
 vhost_process_read_payload_chain(struct vhost_scsi_task *task)
 {
 	void *data;
+	uint64_t chunck_len;
 
 	task->iovs_cnt = 0;
+	chunck_len = task->desc->len;
 	task->resp = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
-						   task->desc->addr);
+						   task->desc->addr,
+						   &chunck_len);
+	if (!task->resp || chunck_len != task->desc->len) {
+		fprintf(stderr, "failed to translate desc address.\n");
+		return;
+	}
 
 	while (descriptor_has_next(task->desc)) {
 		task->desc = descriptor_get_next(task->vq->desc, task->desc);
+		chunck_len = task->desc->len;
 		data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
-						     task->desc->addr);
+						     task->desc->addr,
+							 &chunck_len);
+		if (!data || chunck_len != task->desc->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
+
 		task->iovs[task->iovs_cnt].iov_base = data;
 		task->iovs[task->iovs_cnt].iov_len = task->desc->len;
 		task->data_len += task->desc->len;
@@ -158,12 +172,20 @@ static void
 vhost_process_write_payload_chain(struct vhost_scsi_task *task)
 {
 	void *data;
+	uint64_t chunck_len;
 
 	task->iovs_cnt = 0;
 
 	do {
+		chunck_len = task->desc->len;
 		data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
-						     task->desc->addr);
+						     task->desc->addr,
+							 &chunck_len);
+		if (!data || chunck_len != task->desc->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
+
 		task->iovs[task->iovs_cnt].iov_base = data;
 		task->iovs[task->iovs_cnt].iov_len = task->desc->len;
 		task->data_len += task->desc->len;
@@ -171,8 +193,12 @@ vhost_process_write_payload_chain(struct vhost_scsi_task *task)
 		task->desc = descriptor_get_next(task->vq->desc, task->desc);
 	} while (descriptor_has_next(task->desc));
 
+	chunck_len = task->desc->len;
 	task->resp = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
-						   task->desc->addr);
+						   task->desc->addr,
+						   &chunck_len);
+	if (!task->resp || chunck_len != task->desc->len)
+		fprintf(stderr, "failed to translate desc address.\n");
 }
 
 static struct vhost_block_dev *
@@ -218,6 +244,7 @@ process_requestq(struct vhost_scsi_ctrlr *ctrlr, uint32_t q_idx)
 		int req_idx;
 		uint16_t last_idx;
 		struct vhost_scsi_task *task;
+		uint64_t chunck_len;
 
 		last_idx = scsi_vq->last_used_idx & (vq->size - 1);
 		req_idx = vq->avail->ring[last_idx];
@@ -235,16 +262,27 @@ process_requestq(struct vhost_scsi_ctrlr *ctrlr, uint32_t q_idx)
 		assert((task->desc->flags & VRING_DESC_F_INDIRECT) == 0);
 		scsi_vq->last_used_idx++;
 
+		chunck_len = task->desc->len;
 		task->req = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
-							  task->desc->addr);
+							  task->desc->addr,
+							  &chunck_len);
+		if (!task->req || chunck_len != task->desc->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
 
 		task->desc = descriptor_get_next(task->vq->desc, task->desc);
 		if (!descriptor_has_next(task->desc)) {
 			task->dxfer_dir = SCSI_DIR_NONE;
+			chunck_len = task->desc->len;
 			task->resp = (void *)(uintptr_t)
 					      gpa_to_vva(task->bdev->vid,
-							 task->desc->addr);
-
+							 task->desc->addr,
+							 &chunck_len);
+			if (!task->resp || chunck_len != task->desc->len) {
+				fprintf(stderr, "failed to translate desc address.\n");
+				return;
+			}
 		} else if (!descriptor_is_wr(task->desc)) {
 			task->dxfer_dir = SCSI_DIR_TO_DEV;
 			vhost_process_write_payload_chain(task);