#include <rte_memory.h>
#include <rte_flow.h>
#include <rte_kvargs.h>
+#include <rte_interrupts.h>
/* Generated configuration header. */
#include "mlx4_autoconf.h"
NULL,
};
+static int
+mlx4_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx);
+
+static int
+mlx4_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx);
+
+static int
+priv_rx_intr_vec_enable(struct priv *priv);
+
+static void
+priv_rx_intr_vec_disable(struct priv *priv);
+
/**
* Check if running as a secondary process.
*
}
if (rxqs_n == priv->rxqs_n)
return 0;
- if (!rte_is_power_of_2(rxqs_n)) {
+ if (!rte_is_power_of_2(rxqs_n) && !priv->isolated) {
unsigned n_active;
n_active = rte_align32pow2(rxqs_n + 1) >> 1;
/* For best performance, this function should not be inlined. */
static struct ibv_mr *mlx4_mp2mr(struct ibv_pd *, struct rte_mempool *)
- __attribute__((noinline));
+ __rte_noinline;
/**
* Register mempool as a memory region.
{
unsigned int i;
+ assert(!priv->isolated);
assert(mac_index < elemof(priv->mac));
if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
return;
rxq->if_cq,
¶ms));
}
- if (rxq->qp != NULL) {
+ if (rxq->qp != NULL && !rxq->priv->isolated) {
rxq_promiscuous_disable(rxq);
rxq_allmulticast_disable(rxq);
rxq_mac_addrs_del(rxq);
- claim_zero(ibv_destroy_qp(rxq->qp));
}
+ if (rxq->qp != NULL)
+ claim_zero(ibv_destroy_qp(rxq->qp));
if (rxq->cq != NULL)
claim_zero(ibv_destroy_cq(rxq->cq));
+ if (rxq->channel != NULL)
+ claim_zero(ibv_destroy_comp_channel(rxq->channel));
if (rxq->rd != NULL) {
struct ibv_exp_destroy_res_domain_attr attr = {
.comp_mask = 0,
NB_SEGS(rep) = 0x2a;
PORT(rep) = 0x2a;
rep->ol_flags = -1;
+ /*
+ * Clear special flags in mbuf to avoid
+ * crashing while freeing.
+ */
+ rep->ol_flags &=
+ ~(uint64_t)(IND_ATTACHED_MBUF |
+ CTRL_MBUF_FLAG);
#endif
assert(rep->buf_len == seg->buf_len);
/* Reconfigure sge to use rep instead of seg. */
return 0;
}
/* Remove attached flows if RSS is disabled (no parent queue). */
- if (!priv->rss) {
+ if (!priv->rss && !priv->isolated) {
rxq_allmulticast_disable(&tmpl);
rxq_promiscuous_disable(&tmpl);
rxq_mac_addrs_del(&tmpl);
return err;
};
/* Reconfigure flows. Do not care for errors. */
- if (!priv->rss) {
+ if (!priv->rss && !priv->isolated) {
rxq_mac_addrs_add(&tmpl);
if (priv->promisc)
rxq_promiscuous_enable(&tmpl);
(void *)dev, strerror(ret));
goto error;
}
+ if (dev->data->dev_conf.intr_conf.rxq) {
+ tmpl.channel = ibv_create_comp_channel(priv->ctx);
+ if (tmpl.channel == NULL) {
+ ret = ENOMEM;
+ ERROR("%p: Rx interrupt completion channel creation"
+ " failure: %s",
+ (void *)dev, strerror(ret));
+ goto error;
+ }
+ }
attr.cq = (struct ibv_exp_cq_init_attr){
.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
.res_domain = tmpl.rd,
};
- tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+ tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0,
+ &attr.cq);
if (tmpl.cq == NULL) {
ret = ENOMEM;
ERROR("%p: CQ creation failure: %s",
(void *)dev, strerror(ret));
goto error;
}
- if ((parent) || (!priv->rss)) {
+ if (!priv->isolated && (parent || !priv->rss)) {
/* Configure MAC and broadcast addresses. */
ret = rxq_mac_addrs_add(&tmpl);
if (ret) {
}
DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
priv->started = 1;
- if (priv->rss) {
+ if (priv->isolated) {
+ rxq = NULL;
+ r = 1;
+ } else if (priv->rss) {
rxq = &priv->rxq_parent;
r = 1;
} else {
(void *)dev);
goto err;
}
+ ret = priv_rx_intr_vec_enable(priv);
+ if (ret) {
+ ERROR("%p: Rx interrupt vector creation failed",
+ (void *)dev);
+ goto err;
+ }
ret = mlx4_priv_flow_start(priv);
if (ret) {
ERROR("%p: flow start failed: %s",
}
DEBUG("%p: detaching flows from all RX queues", (void *)dev);
priv->started = 0;
- if (priv->rss) {
+ if (priv->isolated) {
+ rxq = NULL;
+ r = 1;
+ } else if (priv->rss) {
rxq = &priv->rxq_parent;
r = 1;
} else {
assert(priv->ctx == NULL);
priv_dev_removal_interrupt_handler_uninstall(priv, dev);
priv_dev_link_interrupt_handler_uninstall(priv, dev);
+ priv_rx_intr_vec_disable(priv);
priv_unlock(priv);
memset(priv, 0, sizeof(*priv));
}
unsigned int max;
char ifname[IF_NAMESIZE];
- info->pci_dev = RTE_DEV_TO_PCI(dev->device);
+ info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
if (priv == NULL)
return;
if (mlx4_is_secondary())
return;
priv_lock(priv);
+ if (priv->isolated)
+ goto end;
DEBUG("%p: removing MAC address from index %" PRIu32,
(void *)dev, index);
/* Last array entry is reserved for broadcast. */
return -ENOTSUP;
(void)vmdq;
priv_lock(priv);
+ if (priv->isolated) {
+ DEBUG("%p: cannot add MAC address, "
+ "device is in isolated mode", (void *)dev);
+ re = EPERM;
+ goto end;
+ }
DEBUG("%p: adding MAC address at index %" PRIu32,
(void *)dev, index);
/* Last array entry is reserved for broadcast. */
if (mlx4_is_secondary())
return;
priv_lock(priv);
+ if (priv->isolated) {
+ DEBUG("%p: cannot enable promiscuous, "
+ "device is in isolated mode", (void *)dev);
+ priv_unlock(priv);
+ return;
+ }
if (priv->promisc) {
priv_unlock(priv);
return;
if (mlx4_is_secondary())
return;
priv_lock(priv);
- if (!priv->promisc) {
+ if (!priv->promisc || priv->isolated) {
priv_unlock(priv);
return;
}
if (mlx4_is_secondary())
return;
priv_lock(priv);
+ if (priv->isolated) {
+ DEBUG("%p: cannot enable allmulticast, "
+ "device is in isolated mode", (void *)dev);
+ priv_unlock(priv);
+ return;
+ }
if (priv->allmulti) {
priv_unlock(priv);
return;
if (mlx4_is_secondary())
return;
priv_lock(priv);
- if (!priv->allmulti) {
+ if (!priv->allmulti || priv->isolated) {
priv_unlock(priv);
return;
}
}
/* Reenable non-RSS queue attributes. No need to check
* for errors at this stage. */
- if (!priv->rss) {
+ if (!priv->rss && !priv->isolated) {
rxq_mac_addrs_add(rxq);
if (priv->promisc)
rxq_promiscuous_enable(rxq);
if (mlx4_is_secondary())
return -E_RTE_SECONDARY;
priv_lock(priv);
+ if (priv->isolated) {
+ DEBUG("%p: cannot set vlan filter, "
+ "device is in isolated mode", (void *)dev);
+ priv_unlock(priv);
+ return -EINVAL;
+ }
ret = vlan_filter_set(dev, vlan_id, on);
priv_unlock(priv);
assert(ret >= 0);
.destroy = mlx4_flow_destroy,
.flush = mlx4_flow_flush,
.query = NULL,
+ .isolate = mlx4_flow_isolate,
};
/**
.mac_addr_set = mlx4_mac_addr_set,
.mtu_set = mlx4_dev_set_mtu,
.filter_ctrl = mlx4_dev_filter_ctrl,
+ .rx_queue_intr_enable = mlx4_rx_intr_enable,
+ .rx_queue_intr_disable = mlx4_rx_intr_disable,
};
/**
/* Extract information. */
if (sscanf(line,
"PCI_SLOT_NAME="
- "%" SCNx16 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
+ "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
&pci_addr->domain,
&pci_addr->bus,
&pci_addr->devid,
{
struct ibv_async_event event;
int port_change = 0;
+ struct rte_eth_link *link = &dev->data->dev_link;
int ret = 0;
*events = 0;
event.event_type, event.element.port_num);
ibv_ack_async_event(&event);
}
-
- if (port_change ^ priv->pending_alarm) {
- struct rte_eth_link *link = &dev->data->dev_link;
-
- priv->pending_alarm = 0;
- mlx4_link_update(dev, 0);
- if (((link->link_speed == 0) && link->link_status) ||
- ((link->link_speed != 0) && !link->link_status)) {
+ if (!port_change)
+ return ret;
+ mlx4_link_update(dev, 0);
+ if (((link->link_speed == 0) && link->link_status) ||
+ ((link->link_speed != 0) && !link->link_status)) {
+ if (!priv->pending_alarm) {
/* Inconsistent status, check again later. */
priv->pending_alarm = 1;
rte_eal_alarm_set(MLX4_ALARM_TIMEOUT_US,
mlx4_dev_link_status_handler,
dev);
- } else {
- *events |= (1 << RTE_ETH_EVENT_INTR_LSC);
}
+ } else {
+ *events |= (1 << RTE_ETH_EVENT_INTR_LSC);
}
return ret;
}
priv_lock(priv);
assert(priv->pending_alarm == 1);
+ priv->pending_alarm = 0;
ret = priv_dev_status_handler(priv, dev, &events);
priv_unlock(priv);
if (ret > 0 && events & (1 << RTE_ETH_EVENT_INTR_LSC))
- _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
+ _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
+ NULL);
}
/**
i++) {
if (ev & (1 << i)) {
ev &= ~(1 << i);
- _rte_eth_dev_callback_process(dev, i, NULL);
+ _rte_eth_dev_callback_process(dev, i, NULL,
+ NULL);
ret--;
}
}
return 0;
}
+/**
+ * Allocate queue vector and fill epoll fd list for Rx interrupts.
+ *
+ * @param priv
+ * Pointer to private structure.
+ *
+ * @return
+ * 0 on success, negative on failure.
+ */
+static int
+priv_rx_intr_vec_enable(struct priv *priv)
+{
+ unsigned int i;
+ unsigned int rxqs_n = priv->rxqs_n;
+ unsigned int n = RTE_MIN(rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
+ unsigned int count = 0;
+ struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
+
+ if (!priv->dev->data->dev_conf.intr_conf.rxq)
+ return 0;
+ priv_rx_intr_vec_disable(priv);
+ intr_handle->intr_vec = malloc(sizeof(intr_handle->intr_vec[rxqs_n]));
+ if (intr_handle->intr_vec == NULL) {
+ ERROR("failed to allocate memory for interrupt vector,"
+ " Rx interrupts will not be supported");
+ return -ENOMEM;
+ }
+ intr_handle->type = RTE_INTR_HANDLE_EXT;
+ for (i = 0; i != n; ++i) {
+ struct rxq *rxq = (*priv->rxqs)[i];
+ int fd;
+ int flags;
+ int rc;
+
+ /* Skip queues that cannot request interrupts. */
+ if (!rxq || !rxq->channel) {
+ /* Use invalid intr_vec[] index to disable entry. */
+ intr_handle->intr_vec[i] =
+ RTE_INTR_VEC_RXTX_OFFSET +
+ RTE_MAX_RXTX_INTR_VEC_ID;
+ continue;
+ }
+ if (count >= RTE_MAX_RXTX_INTR_VEC_ID) {
+ ERROR("too many Rx queues for interrupt vector size"
+ " (%d), Rx interrupts cannot be enabled",
+ RTE_MAX_RXTX_INTR_VEC_ID);
+ priv_rx_intr_vec_disable(priv);
+ return -1;
+ }
+ fd = rxq->channel->fd;
+ flags = fcntl(fd, F_GETFL);
+ rc = fcntl(fd, F_SETFL, flags | O_NONBLOCK);
+ if (rc < 0) {
+ ERROR("failed to make Rx interrupt file descriptor"
+ " %d non-blocking for queue index %d", fd, i);
+ priv_rx_intr_vec_disable(priv);
+ return rc;
+ }
+ intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + count;
+ intr_handle->efds[count] = fd;
+ count++;
+ }
+ if (!count)
+ priv_rx_intr_vec_disable(priv);
+ else
+ intr_handle->nb_efd = count;
+ return 0;
+}
+
+/**
+ * Clean up Rx interrupts handler.
+ *
+ * @param priv
+ * Pointer to private structure.
+ */
+static void
+priv_rx_intr_vec_disable(struct priv *priv)
+{
+ struct rte_intr_handle *intr_handle = priv->dev->intr_handle;
+
+ rte_intr_free_epoll_fd(intr_handle);
+ free(intr_handle->intr_vec);
+ intr_handle->nb_efd = 0;
+ intr_handle->intr_vec = NULL;
+}
+
+/**
+ * DPDK callback for Rx queue interrupt enable.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ * @param idx
+ * Rx queue index.
+ *
+ * @return
+ * 0 on success, negative on failure.
+ */
+static int
+mlx4_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
+{
+ struct priv *priv = dev->data->dev_private;
+ struct rxq *rxq = (*priv->rxqs)[idx];
+ int ret;
+
+ if (!rxq || !rxq->channel)
+ ret = EINVAL;
+ else
+ ret = ibv_req_notify_cq(rxq->cq, 0);
+ if (ret)
+ WARN("unable to arm interrupt on rx queue %d", idx);
+ return -ret;
+}
+
+/**
+ * DPDK callback for Rx queue interrupt disable.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ * @param idx
+ * Rx queue index.
+ *
+ * @return
+ * 0 on success, negative on failure.
+ */
+static int
+mlx4_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
+{
+ struct priv *priv = dev->data->dev_private;
+ struct rxq *rxq = (*priv->rxqs)[idx];
+ struct ibv_cq *ev_cq;
+ void *ev_ctx;
+ int ret;
+
+ if (!rxq || !rxq->channel) {
+ ret = EINVAL;
+ } else {
+ ret = ibv_get_cq_event(rxq->cq->channel, &ev_cq, &ev_ctx);
+ if (ret || ev_cq != rxq->cq)
+ ret = EINVAL;
+ }
+ if (ret)
+ WARN("unable to disable interrupt on rx queue %d",
+ idx);
+ else
+ ibv_ack_cq_events(rxq->cq, 1);
+ return -ret;
+}
+
/**
* Verify and store value for device argument.
*
eth_dev->device->driver = &mlx4_driver.driver;
+ /*
+ * Copy and override interrupt handle to prevent it from
+ * being shared between all ethdev instances of a given PCI
+ * device. This is required to properly handle Rx interrupts
+ * on all ports.
+ */
+ priv->intr_handle_dev = *eth_dev->intr_handle;
+ eth_dev->intr_handle = &priv->intr_handle_dev;
+
priv->dev = eth_dev;
eth_dev->dev_ops = &mlx4_dev_ops;