[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [UNIKRAFT PATCH v4 07/12] plat/xen/drivers/net: Configure netfront rx queue



Hello Costin,

Please find the comment inline:

Thanks & Regards

Sharan

On 8/13/20 10:53 AM, Costin Lupu wrote:
Incoming packets are saved in buffers allocated with user-provided callbacks.
Whenever such packet is available, the driver is notified via event channels.
This patch introduces the configuration logic for rx queues and the handler
used in dealing with the notifications.

Signed-off-by: Costin Lupu <costin.lupu@xxxxxxxxx>
Signed-off-by: Razvan Cojocaru <razvan.cojocaru93@xxxxxxxxx>
---
  plat/xen/drivers/net/netfront.c | 212 ++++++++++++++++++++++++++++++--
  plat/xen/drivers/net/netfront.h |  29 +++++
  2 files changed, 232 insertions(+), 9 deletions(-)

diff --git a/plat/xen/drivers/net/netfront.c b/plat/xen/drivers/net/netfront.c
index 3c936743..742e05cf 100644
--- a/plat/xen/drivers/net/netfront.c
+++ b/plat/xen/drivers/net/netfront.c
@@ -52,6 +52,11 @@
  static struct uk_alloc *drv_allocator;
+static uint16_t xennet_rxidx(RING_IDX idx)
+{
+       return (uint16_t) (idx & (NET_RX_RING_SIZE - 1));
+}
+
  static void add_id_to_freelist(uint16_t id, uint16_t *freelist)
  {
        freelist[id + 1] = freelist[0];
@@ -67,6 +72,82 @@ static uint16_t get_id_from_freelist(uint16_t *freelist)
        return id;
  }
+static int netfront_rxq_enqueue(struct uk_netdev_rx_queue *rxq,
+               struct uk_netbuf *netbuf)
+{
+       RING_IDX req_prod;
+       uint16_t id;
+       netif_rx_request_t *rx_req;
+       struct netfront_dev *nfdev;
+       int notify;
+
+       /* buffer must be page aligned */
+       UK_ASSERT(((unsigned long) netbuf->data & ~PAGE_MASK) == 0);
+
+       if (RING_FULL(&rxq->ring)) {
+               uk_pr_debug("rx queue is full\n");
+               return -ENOSPC;
+       }
+
+       /* get request */
+       req_prod = rxq->ring.req_prod_pvt;
+       id = xennet_rxidx(req_prod);
+       rx_req = RING_GET_REQUEST(&rxq->ring, req_prod);
+       rx_req->id = id;
+
+       /* save buffer */
+       rxq->netbuf[id] = netbuf;
+       /* setup grant for buffer data */
+       nfdev = rxq->netfront_dev;
+       rxq->gref[id] = rx_req->gref =
+               gnttab_grant_access(nfdev->xendev->otherend_id,
+                       virt_to_mfn(netbuf->data), 0);
+       UK_ASSERT(rx_req->gref != GRANT_INVALID_REF);
+
+       wmb(); /* Ensure backend sees requests */
+       rxq->ring.req_prod_pvt = req_prod + 1;
+
+       RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rxq->ring, notify);
+       if (notify)
+               notify_remote_via_evtchn(rxq->evtchn);
+
+       return 0;
+}
+
+static int netfront_rx_fillup(struct uk_netdev_rx_queue *rxq, uint16_t nb_desc)
+{
+       struct uk_netbuf *netbuf[nb_desc];
+       int rc, status = 0;
+       uint16_t cnt;
+
+       cnt = rxq->alloc_rxpkts(rxq->alloc_rxpkts_argp, netbuf, nb_desc);
+
+       for (uint16_t i = 0; i < cnt; i++) {
+               rc = netfront_rxq_enqueue(rxq, netbuf[i]);
+               if (unlikely(rc < 0)) {
+                       uk_pr_err("Failed to add a buffer to rx queue %p: %d\n",
+                               rxq, rc);
+
+                       /*
+                        * Release netbufs that we are not going
+                        * to use anymore
+                        */
+                       for (uint16_t j = i; j < cnt; j++)
+                               uk_netbuf_free(netbuf[j]);
+
+                       status |= UK_NETDEV_STATUS_UNDERRUN;
+
+                       goto out;
+               }
+       }
+
+       if (unlikely(cnt < nb_desc))
+               status |= UK_NETDEV_STATUS_UNDERRUN;
+
+out:
+       return status;
+}
+
  static struct uk_netdev_tx_queue *netfront_txq_setup(struct uk_netdev *n,
                uint16_t queue_id,
                uint16_t nb_desc __unused,
@@ -103,15 +184,19 @@ static struct uk_netdev_tx_queue 
*netfront_txq_setup(struct uk_netdev *n,
        UK_ASSERT(txq->ring_ref != GRANT_INVALID_REF);
/* Setup event channel */
-       rc = evtchn_alloc_unbound(nfdev->xendev->otherend_id,
-                       NULL, NULL,
-                       &txq->evtchn);
-       if (rc) {
-               uk_pr_err("Error creating event channel: %d\n", rc);
-               gnttab_end_access(txq->ring_ref);
-               uk_pfree(drv_allocator, sring, 1);
-               return ERR2PTR(-rc);
-       }
+       if (nfdev->split_evtchn || !nfdev->rxqs[queue_id].initialized) {
+               rc = evtchn_alloc_unbound(nfdev->xendev->otherend_id,
+                               NULL, NULL,
+                               &txq->evtchn);
+               if (rc) {
+                       uk_pr_err("Error creating event channel: %d\n", rc);
+                       gnttab_end_access(txq->ring_ref);
+                       uk_pfree(drv_allocator, sring, 1);
+                       return ERR2PTR(rc);
+               }
+       } else
+               txq->evtchn = nfdev->rxqs[queue_id].evtchn;

In the next version maybe this can go to the previous patch.


+
        /* Events are always disabled for tx queue */
        mask_evtchn(txq->evtchn);
@@ -128,6 +213,86 @@ static struct uk_netdev_tx_queue *netfront_txq_setup(struct uk_netdev *n,
        return txq;
  }
+static void netfront_handler(evtchn_port_t port __unused,
+               struct __regs *regs __unused, void *arg)
+{
+       struct uk_netdev_rx_queue *rxq = arg;
+
+       /* Indicate to the network stack about an event */
+       uk_netdev_drv_rx_event(&rxq->netfront_dev->netdev, rxq->lqueue_id);
+}
+
+static struct uk_netdev_rx_queue *netfront_rxq_setup(struct uk_netdev *n,
+               uint16_t queue_id,
+               uint16_t nb_desc __unused,
+               struct uk_netdev_rxqueue_conf *conf)
+{
+       int rc;
+       struct netfront_dev *nfdev;
+       struct uk_netdev_rx_queue *rxq;
+       netif_rx_sring_t *sring;
+
+       UK_ASSERT(n != NULL);
+       UK_ASSERT(conf != NULL);
+
+       nfdev = to_netfront_dev(n);
+       if (queue_id >= nfdev->max_queue_pairs) {
+               uk_pr_err("Invalid queue identifier: %"__PRIu16"\n", queue_id);
+               return ERR2PTR(-EINVAL);
+       }
+
+       rxq = &nfdev->rxqs[queue_id];
+       UK_ASSERT(!rxq->initialized);
+       rxq->netfront_dev = nfdev;
+       rxq->lqueue_id = queue_id;
+
+       /* Setup shared ring */
+       sring = uk_palloc(drv_allocator, 1);
Allocator from the conf should be used here.
+       if (!sring)
+               return ERR2PTR(-ENOMEM);
+       memset(sring, 0, PAGE_SIZE);
+       SHARED_RING_INIT(sring);
+       FRONT_RING_INIT(&rxq->ring, sring, PAGE_SIZE);
+       rxq->ring_size = NET_RX_RING_SIZE;
+       rxq->ring_ref = gnttab_grant_access(nfdev->xendev->otherend_id,
+               virt_to_mfn(sring), 0);
+       UK_ASSERT(rxq->ring_ref != GRANT_INVALID_REF);
+
+       /* Setup event channel */
+       if (nfdev->split_evtchn || !nfdev->txqs[queue_id].initialized) {
+               rc = evtchn_alloc_unbound(nfdev->xendev->otherend_id,
+                               netfront_handler, rxq,
+                               &rxq->evtchn);
+               if (rc) {
+                       uk_pr_err("Error creating event channel: %d\n", rc);
+                       gnttab_end_access(rxq->ring_ref);
+                       uk_pfree(drv_allocator, sring, 1);
+                       return ERR2PTR(rc);
+               }
+       } else {
+               rxq->evtchn = nfdev->txqs[queue_id].evtchn;
+               /* overwriting event handler */
+               bind_evtchn(rxq->evtchn, netfront_handler, rxq);
+       }
+       /*
+        * By default, events are disabled and it is up to the user or
+        * network stack to explicitly enable them.
+        */
+       mask_evtchn(rxq->evtchn);
+       rxq->intr_enabled = 0;
+
+       rxq->alloc_rxpkts = conf->alloc_rxpkts;
+       rxq->alloc_rxpkts_argp = conf->alloc_rxpkts_argp;
+
+       /* Allocate receive buffers for this queue */
+       netfront_rx_fillup(rxq, rxq->ring_size);
+
+       rxq->initialized = true;
+       nfdev->rxqs_num++;
+
+       return rxq;
+}
+
  static int netfront_rxtx_alloc(struct netfront_dev *nfdev,
                const struct uk_netdev_conf *conf)
  {
@@ -196,6 +361,33 @@ exit:
        return rc;
  }
+static int netfront_rxq_info_get(struct uk_netdev *n,
+               uint16_t queue_id,
+               struct uk_netdev_queue_info *qinfo)
+{
+       struct netfront_dev *nfdev;
+       struct uk_netdev_rx_queue *rxq;
+       int rc = 0;
+
+       UK_ASSERT(n != NULL);
+       UK_ASSERT(qinfo != NULL);
+
+       nfdev = to_netfront_dev(n);
+       if (unlikely(queue_id >= nfdev->rxqs_num)) {
Same as the txq info get
+               uk_pr_err("Invalid queue id: %"__PRIu16"\n", queue_id);
+               rc = -EINVAL;
+               goto exit;
+       }
+       rxq = &nfdev->rxqs[queue_id];
+       qinfo->nb_min = rxq->ring_size;
+       qinfo->nb_max = rxq->ring_size;
+       qinfo->nb_align = PAGE_SIZE;
+       qinfo->nb_is_power_of_two = 1;
+
+exit:
+       return rc;
+}
+
  static int netfront_configure(struct uk_netdev *n,
                const struct uk_netdev_conf *conf)
  {
@@ -286,7 +478,9 @@ static unsigned int netfront_promisc_get(struct uk_netdev 
*n)
  static const struct uk_netdev_ops netfront_ops = {
        .configure = netfront_configure,
        .txq_configure = netfront_txq_setup,
+       .rxq_configure = netfront_rxq_setup,
        .txq_info_get = netfront_txq_info_get,
+       .rxq_info_get = netfront_rxq_info_get,
        .info_get = netfront_info_get,
        .einfo_get = netfront_einfo_get,
        .hwaddr_get = netfront_mac_get,
diff --git a/plat/xen/drivers/net/netfront.h b/plat/xen/drivers/net/netfront.h
index d3f603b3..c8364815 100644
--- a/plat/xen/drivers/net/netfront.h
+++ b/plat/xen/drivers/net/netfront.h
@@ -44,6 +44,7 @@
#define NET_TX_RING_SIZE __CONST_RING_SIZE(netif_tx, PAGE_SIZE)
+#define NET_RX_RING_SIZE __CONST_RING_SIZE(netif_rx, PAGE_SIZE)
/**
   * internal structure to represent the transmit queue.
@@ -77,6 +78,33 @@ struct uk_netdev_tx_queue {
   * internal structure to represent the receive queue.
   */
  struct uk_netdev_rx_queue {
+       /* The netfront device */
+       struct netfront_dev *netfront_dev;
+       /* The libuknet queue identifier */
+       uint16_t lqueue_id;
+       /* True if initialized */
+       bool initialized;
+
+       /* Shared ring size */
+       uint16_t ring_size;
+       /* Shared ring */
+       netif_rx_front_ring_t ring;
+       /* Shared ring grant ref */
+       grant_ref_t ring_ref;
+       /* Queue event channel */
+       evtchn_port_t evtchn;
+
+       /* The flag to interrupt on the transmit queue */
+       uint8_t intr_enabled;
+
+       /* User-provided receive buffer allocation function */
+       uk_netdev_alloc_rxpkts alloc_rxpkts;
+       void *alloc_rxpkts_argp;
+
+       /* Receive buffers for incoming packets */
+       struct uk_netbuf *netbuf[NET_RX_RING_SIZE];
+       /* Grants for receive buffers */
+       grant_ref_t gref[NET_RX_RING_SIZE];
  };
struct xs_econf {
@@ -93,6 +121,7 @@ struct netfront_dev {
/* List of the Rx/Tx queues */
        uint16_t txqs_num;
+       uint16_t rxqs_num;
        struct uk_netdev_tx_queue *txqs;
        struct uk_netdev_rx_queue *rxqs;
        /* Maximum number of queue pairs */



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.