* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
-/*
- * Known limitations:
- * - RSS hash key and options cannot be modified.
- */
-
/* System headers. */
#include <stddef.h>
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
-#include <limits.h>
#include <assert.h>
#include <net/if.h>
#include <dirent.h>
#include <rte_prefetch.h>
#include <rte_malloc.h>
#include <rte_spinlock.h>
-#include <rte_atomic.h>
#include <rte_log.h>
#include <rte_alarm.h>
#include <rte_memory.h>
#define WR_ID(o) (((wr_id_t *)&(o))->data)
-/* Transpose flags. Useful to convert IBV to DPDK flags. */
-#define TRANSPOSE(val, from, to) \
- (((from) >= (to)) ? \
- (((val) & (from)) / ((from) / (to))) : \
- (((val) & (from)) * ((to) / (from))))
-
/** Configuration structure for device arguments. */
struct mlx4_conf {
struct {
static int
rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
- unsigned int socket, int inactive,
- const struct rte_eth_rxconf *conf,
- struct rte_mempool *mp, int children_n,
- struct rxq *rxq_parent);
+ unsigned int socket, const struct rte_eth_rxconf *conf,
+ struct rte_mempool *mp);
static void
rxq_cleanup(struct rxq *rxq);
-/**
- * Create RSS parent queue.
- *
- * The new parent is inserted in front of the list in the private structure.
- *
- * @param priv
- * Pointer to private structure.
- * @param queues
- * Queues indices array, if NULL use all Rx queues.
- * @param children_n
- * The number of entries in queues[].
- *
- * @return
- * Pointer to a parent rxq structure, NULL on failure.
- */
-struct rxq *
-priv_parent_create(struct priv *priv,
- uint16_t queues[],
- uint16_t children_n)
-{
- int ret;
- uint16_t i;
- struct rxq *parent;
-
- parent = rte_zmalloc("parent queue",
- sizeof(*parent),
- RTE_CACHE_LINE_SIZE);
- if (!parent) {
- ERROR("cannot allocate memory for RSS parent queue");
- return NULL;
- }
- ret = rxq_setup(priv->dev, parent, 0, 0, 0,
- NULL, NULL, children_n, NULL);
- if (ret) {
- rte_free(parent);
- return NULL;
- }
- parent->rss.queues_n = children_n;
- if (queues) {
- for (i = 0; i < children_n; ++i)
- parent->rss.queues[i] = queues[i];
- } else {
- /* the default RSS ring case */
- assert(priv->rxqs_n == children_n);
- for (i = 0; i < priv->rxqs_n; ++i)
- parent->rss.queues[i] = i;
- }
- LIST_INSERT_HEAD(&priv->parents, parent, next);
- return parent;
-}
-
-/**
- * Clean up RX queue parent structure.
- *
- * @param parent
- * RX queue parent structure.
- */
-void
-rxq_parent_cleanup(struct rxq *parent)
-{
- LIST_REMOVE(parent, next);
- rxq_cleanup(parent);
- rte_free(parent);
-}
-
-/**
- * Clean up parent structures from the parent list.
- *
- * @param priv
- * Pointer to private structure.
- */
static void
-priv_parent_list_cleanup(struct priv *priv)
-{
- while (!LIST_EMPTY(&priv->parents))
- rxq_parent_cleanup(LIST_FIRST(&priv->parents));
-}
+priv_mac_addr_del(struct priv *priv);
/**
* Ethernet device configuration.
*
* Prepare the driver for a given number of TX and RX queues.
- * Allocate parent RSS queue when several RX queues are requested.
*
* @param dev
* Pointer to Ethernet device structure.
struct priv *priv = dev->data->dev_private;
unsigned int rxqs_n = dev->data->nb_rx_queues;
unsigned int txqs_n = dev->data->nb_tx_queues;
- unsigned int tmp;
priv->rxqs = (void *)dev->data->rx_queues;
priv->txqs = (void *)dev->data->tx_queues;
(void *)dev, priv->txqs_n, txqs_n);
priv->txqs_n = txqs_n;
}
- if (rxqs_n == priv->rxqs_n)
- return 0;
- if (!rte_is_power_of_2(rxqs_n) && !priv->isolated) {
- unsigned n_active;
-
- n_active = rte_align32pow2(rxqs_n + 1) >> 1;
- WARN("%p: number of RX queues must be a power"
- " of 2: %u queues among %u will be active",
- (void *)dev, n_active, rxqs_n);
- }
-
- INFO("%p: RX queues number update: %u -> %u",
- (void *)dev, priv->rxqs_n, rxqs_n);
- /* If RSS is enabled, disable it first. */
- if (priv->rss) {
- unsigned int i;
-
- /* Only if there are no remaining child RX queues. */
- for (i = 0; (i != priv->rxqs_n); ++i)
- if ((*priv->rxqs)[i] != NULL)
- return EINVAL;
- priv_parent_list_cleanup(priv);
- priv->rss = 0;
- priv->rxqs_n = 0;
- }
- if (rxqs_n <= 1) {
- /* Nothing else to do. */
+ if (rxqs_n != priv->rxqs_n) {
+ INFO("%p: Rx queues number update: %u -> %u",
+ (void *)dev, priv->rxqs_n, rxqs_n);
priv->rxqs_n = rxqs_n;
- return 0;
- }
- /* Allocate a new RSS parent queue if supported by hardware. */
- if (!priv->hw_rss) {
- ERROR("%p: only a single RX queue can be configured when"
- " hardware doesn't support RSS",
- (void *)dev);
- return EINVAL;
- }
- /* Fail if hardware doesn't support that many RSS queues. */
- if (rxqs_n >= priv->max_rss_tbl_sz) {
- ERROR("%p: only %u RX queues can be configured for RSS",
- (void *)dev, priv->max_rss_tbl_sz);
- return EINVAL;
}
- priv->rss = 1;
- tmp = priv->rxqs_n;
- priv->rxqs_n = rxqs_n;
- if (priv->isolated)
- return 0;
- if (priv_parent_create(priv, NULL, priv->rxqs_n))
- return 0;
- /* Failure, rollback. */
- priv->rss = 0;
- priv->rxqs_n = tmp;
- return ENOMEM;
+ return 0;
}
/**
unsigned int i;
struct txq_elt (*elts)[elts_n] =
rte_calloc_socket("TXQ", 1, sizeof(*elts), 0, txq->socket);
- linear_t (*elts_linear)[elts_n] =
- rte_calloc_socket("TXQ", 1, sizeof(*elts_linear), 0,
- txq->socket);
- struct ibv_mr *mr_linear = NULL;
int ret = 0;
- if ((elts == NULL) || (elts_linear == NULL)) {
+ if (elts == NULL) {
ERROR("%p: can't allocate packets array", (void *)txq);
ret = ENOMEM;
goto error;
}
- mr_linear =
- ibv_reg_mr(txq->priv->pd, elts_linear, sizeof(*elts_linear),
- IBV_ACCESS_LOCAL_WRITE);
- if (mr_linear == NULL) {
- ERROR("%p: unable to configure MR, ibv_reg_mr() failed",
- (void *)txq);
- ret = EINVAL;
- goto error;
- }
for (i = 0; (i != elts_n); ++i) {
struct txq_elt *elt = &(*elts)[i];
((MLX4_PMD_TX_PER_COMP_REQ < (elts_n / 4)) ?
MLX4_PMD_TX_PER_COMP_REQ : (elts_n / 4));
txq->elts_comp_cd = txq->elts_comp_cd_init;
- txq->elts_linear = elts_linear;
- txq->mr_linear = mr_linear;
assert(ret == 0);
return 0;
error:
- if (mr_linear != NULL)
- claim_zero(ibv_dereg_mr(mr_linear));
-
- rte_free(elts_linear);
rte_free(elts);
DEBUG("%p: failed, freed everything", (void *)txq);
unsigned int elts_head = txq->elts_head;
unsigned int elts_tail = txq->elts_tail;
struct txq_elt (*elts)[elts_n] = txq->elts;
- linear_t (*elts_linear)[elts_n] = txq->elts_linear;
- struct ibv_mr *mr_linear = txq->mr_linear;
DEBUG("%p: freeing WRs", (void *)txq);
txq->elts_n = 0;
txq->elts_comp_cd = 0;
txq->elts_comp_cd_init = 0;
txq->elts = NULL;
- txq->elts_linear = NULL;
- txq->mr_linear = NULL;
- if (mr_linear != NULL)
- claim_zero(ibv_dereg_mr(mr_linear));
-
- rte_free(elts_linear);
if (elts == NULL)
return;
while (elts_tail != elts_head) {
txq_mp2mr(txq, mp);
}
-/**
- * Copy scattered mbuf contents to a single linear buffer.
- *
- * @param[out] linear
- * Linear output buffer.
- * @param[in] buf
- * Scattered input buffer.
- *
- * @return
- * Number of bytes copied to the output buffer or 0 if not large enough.
- */
-static unsigned int
-linearize_mbuf(linear_t *linear, struct rte_mbuf *buf)
-{
- unsigned int size = 0;
- unsigned int offset;
-
- do {
- unsigned int len = DATA_LEN(buf);
-
- offset = size;
- size += len;
- if (unlikely(size > sizeof(*linear)))
- return 0;
- memcpy(&(*linear)[offset],
- rte_pktmbuf_mtod(buf, uint8_t *),
- len);
- buf = NEXT(buf);
- } while (buf != NULL);
- return size;
-}
-
-/**
- * Handle scattered buffers for mlx4_tx_burst().
- *
- * @param txq
- * TX queue structure.
- * @param segs
- * Number of segments in buf.
- * @param elt
- * TX queue element to fill.
- * @param[in] buf
- * Buffer to process.
- * @param elts_head
- * Index of the linear buffer to use if necessary (normally txq->elts_head).
- * @param[out] sges
- * Array filled with SGEs on success.
- *
- * @return
- * A structure containing the processed packet size in bytes and the
- * number of SGEs. Both fields are set to (unsigned int)-1 in case of
- * failure.
- */
-static struct tx_burst_sg_ret {
- unsigned int length;
- unsigned int num;
-}
-tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt,
- struct rte_mbuf *buf, unsigned int elts_head,
- struct ibv_sge (*sges)[MLX4_PMD_SGE_WR_N])
-{
- unsigned int sent_size = 0;
- unsigned int j;
- int linearize = 0;
-
- /* When there are too many segments, extra segments are
- * linearized in the last SGE. */
- if (unlikely(segs > elemof(*sges))) {
- segs = (elemof(*sges) - 1);
- linearize = 1;
- }
- /* Update element. */
- elt->buf = buf;
- /* Register segments as SGEs. */
- for (j = 0; (j != segs); ++j) {
- struct ibv_sge *sge = &(*sges)[j];
- uint32_t lkey;
-
- /* Retrieve Memory Region key for this memory pool. */
- lkey = txq_mp2mr(txq, txq_mb2mp(buf));
- if (unlikely(lkey == (uint32_t)-1)) {
- /* MR does not exist. */
- DEBUG("%p: unable to get MP <-> MR association",
- (void *)txq);
- /* Clean up TX element. */
- elt->buf = NULL;
- goto stop;
- }
- /* Update SGE. */
- sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
- if (txq->priv->vf)
- rte_prefetch0((volatile void *)
- (uintptr_t)sge->addr);
- sge->length = DATA_LEN(buf);
- sge->lkey = lkey;
- sent_size += sge->length;
- buf = NEXT(buf);
- }
- /* If buf is not NULL here and is not going to be linearized,
- * nb_segs is not valid. */
- assert(j == segs);
- assert((buf == NULL) || (linearize));
- /* Linearize extra segments. */
- if (linearize) {
- struct ibv_sge *sge = &(*sges)[segs];
- linear_t *linear = &(*txq->elts_linear)[elts_head];
- unsigned int size = linearize_mbuf(linear, buf);
-
- assert(segs == (elemof(*sges) - 1));
- if (size == 0) {
- /* Invalid packet. */
- DEBUG("%p: packet too large to be linearized.",
- (void *)txq);
- /* Clean up TX element. */
- elt->buf = NULL;
- goto stop;
- }
- /* If MLX4_PMD_SGE_WR_N is 1, free mbuf immediately. */
- if (elemof(*sges) == 1) {
- do {
- struct rte_mbuf *next = NEXT(buf);
-
- rte_pktmbuf_free_seg(buf);
- buf = next;
- } while (buf != NULL);
- elt->buf = NULL;
- }
- /* Update SGE. */
- sge->addr = (uintptr_t)&(*linear)[0];
- sge->length = size;
- sge->lkey = txq->mr_linear->lkey;
- sent_size += size;
- /* Include last segment. */
- segs++;
- }
- return (struct tx_burst_sg_ret){
- .length = sent_size,
- .num = segs,
- };
-stop:
- return (struct tx_burst_sg_ret){
- .length = -1,
- .num = -1,
- };
-}
-
/**
* DPDK callback for TX.
*
++elts_comp;
send_flags |= IBV_EXP_QP_BURST_SIGNALED;
}
- /* Should we enable HW CKSUM offload */
- if (buf->ol_flags &
- (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
- send_flags |= IBV_EXP_QP_BURST_IP_CSUM;
- /* HW does not support checksum offloads at arbitrary
- * offsets but automatically recognizes the packet
- * type. For inner L3/L4 checksums, only VXLAN (UDP)
- * tunnels are currently supported. */
- if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type))
- send_flags |= IBV_EXP_QP_BURST_TUNNEL;
- }
if (likely(segs == 1)) {
uintptr_t addr;
uint32_t length;
goto stop;
sent_size += length;
} else {
- struct ibv_sge sges[MLX4_PMD_SGE_WR_N];
- struct tx_burst_sg_ret ret;
-
- ret = tx_burst_sg(txq, segs, elt, buf, elts_head,
- &sges);
- if (ret.length == (unsigned int)-1)
- goto stop;
- RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
- /* Put SG list into send queue. */
- err = txq->if_qp->send_pending_sg_list
- (txq->qp,
- sges,
- ret.num,
- send_flags);
- if (unlikely(err))
- goto stop;
- sent_size += ret.length;
+ err = -1;
+ goto stop;
}
elts_head = elts_head_next;
/* Increment sent bytes counter. */
(void)conf; /* Thresholds configuration (ignored). */
if (priv == NULL)
return EINVAL;
- if ((desc == 0) || (desc % MLX4_PMD_SGE_WR_N)) {
- ERROR("%p: invalid number of TX descriptors (must be a"
- " multiple of %d)", (void *)dev, MLX4_PMD_SGE_WR_N);
+ if (desc == 0) {
+ ERROR("%p: invalid number of Tx descriptors", (void *)dev);
return EINVAL;
}
- desc /= MLX4_PMD_SGE_WR_N;
/* MRs will be registered in mp2mr[] later. */
attr.rd = (struct ibv_exp_res_domain_init_attr){
.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
priv->device_attr.max_qp_wr :
desc),
/* Max number of scatter/gather elements in a WR. */
- .max_send_sge = ((priv->device_attr.max_sge <
- MLX4_PMD_SGE_WR_N) ?
- priv->device_attr.max_sge :
- MLX4_PMD_SGE_WR_N),
+ .max_send_sge = 1,
.max_inline_data = MLX4_PMD_MAX_INLINE,
},
.qp_type = IBV_QPT_RAW_PACKET,
/* RX queues handling. */
-/**
- * Allocate RX queue elements with scattered packets support.
- *
- * @param rxq
- * Pointer to RX queue structure.
- * @param elts_n
- * Number of elements to allocate.
- * @param[in] pool
- * If not NULL, fetch buffers from this array instead of allocating them
- * with rte_pktmbuf_alloc().
- *
- * @return
- * 0 on success, errno value on failure.
- */
-static int
-rxq_alloc_elts_sp(struct rxq *rxq, unsigned int elts_n,
- struct rte_mbuf **pool)
-{
- unsigned int i;
- struct rxq_elt_sp (*elts)[elts_n] =
- rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
- rxq->socket);
- int ret = 0;
-
- if (elts == NULL) {
- ERROR("%p: can't allocate packets array", (void *)rxq);
- ret = ENOMEM;
- goto error;
- }
- /* For each WR (packet). */
- for (i = 0; (i != elts_n); ++i) {
- unsigned int j;
- struct rxq_elt_sp *elt = &(*elts)[i];
- struct ibv_recv_wr *wr = &elt->wr;
- struct ibv_sge (*sges)[(elemof(elt->sges))] = &elt->sges;
-
- /* These two arrays must have the same size. */
- assert(elemof(elt->sges) == elemof(elt->bufs));
- /* Configure WR. */
- wr->wr_id = i;
- wr->next = &(*elts)[(i + 1)].wr;
- wr->sg_list = &(*sges)[0];
- wr->num_sge = elemof(*sges);
- /* For each SGE (segment). */
- for (j = 0; (j != elemof(elt->bufs)); ++j) {
- struct ibv_sge *sge = &(*sges)[j];
- struct rte_mbuf *buf;
-
- if (pool != NULL) {
- buf = *(pool++);
- assert(buf != NULL);
- rte_pktmbuf_reset(buf);
- } else
- buf = rte_pktmbuf_alloc(rxq->mp);
- if (buf == NULL) {
- assert(pool == NULL);
- ERROR("%p: empty mbuf pool", (void *)rxq);
- ret = ENOMEM;
- goto error;
- }
- elt->bufs[j] = buf;
- /* Headroom is reserved by rte_pktmbuf_alloc(). */
- assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
- /* Buffer is supposed to be empty. */
- assert(rte_pktmbuf_data_len(buf) == 0);
- assert(rte_pktmbuf_pkt_len(buf) == 0);
- /* sge->addr must be able to store a pointer. */
- assert(sizeof(sge->addr) >= sizeof(uintptr_t));
- if (j == 0) {
- /* The first SGE keeps its headroom. */
- sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
- sge->length = (buf->buf_len -
- RTE_PKTMBUF_HEADROOM);
- } else {
- /* Subsequent SGEs lose theirs. */
- assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
- SET_DATA_OFF(buf, 0);
- sge->addr = (uintptr_t)buf->buf_addr;
- sge->length = buf->buf_len;
- }
- sge->lkey = rxq->mr->lkey;
- /* Redundant check for tailroom. */
- assert(sge->length == rte_pktmbuf_tailroom(buf));
- }
- }
- /* The last WR pointer must be NULL. */
- (*elts)[(i - 1)].wr.next = NULL;
- DEBUG("%p: allocated and configured %u WRs (%zu segments)",
- (void *)rxq, elts_n, (elts_n * elemof((*elts)[0].sges)));
- rxq->elts_n = elts_n;
- rxq->elts_head = 0;
- rxq->elts.sp = elts;
- assert(ret == 0);
- return 0;
-error:
- if (elts != NULL) {
- assert(pool == NULL);
- for (i = 0; (i != elemof(*elts)); ++i) {
- unsigned int j;
- struct rxq_elt_sp *elt = &(*elts)[i];
-
- for (j = 0; (j != elemof(elt->bufs)); ++j) {
- struct rte_mbuf *buf = elt->bufs[j];
-
- if (buf != NULL)
- rte_pktmbuf_free_seg(buf);
- }
- }
- rte_free(elts);
- }
- DEBUG("%p: failed, freed everything", (void *)rxq);
- assert(ret > 0);
- return ret;
-}
-
-/**
- * Free RX queue elements with scattered packets support.
- *
- * @param rxq
- * Pointer to RX queue structure.
- */
-static void
-rxq_free_elts_sp(struct rxq *rxq)
-{
- unsigned int i;
- unsigned int elts_n = rxq->elts_n;
- struct rxq_elt_sp (*elts)[elts_n] = rxq->elts.sp;
-
- DEBUG("%p: freeing WRs", (void *)rxq);
- rxq->elts_n = 0;
- rxq->elts.sp = NULL;
- if (elts == NULL)
- return;
- for (i = 0; (i != elemof(*elts)); ++i) {
- unsigned int j;
- struct rxq_elt_sp *elt = &(*elts)[i];
-
- for (j = 0; (j != elemof(elt->bufs)); ++j) {
- struct rte_mbuf *buf = elt->bufs[j];
-
- if (buf != NULL)
- rte_pktmbuf_free_seg(buf);
- }
- }
- rte_free(elts);
-}
-
/**
* Allocate RX queue elements.
*
* Pointer to RX queue structure.
* @param elts_n
* Number of elements to allocate.
- * @param[in] pool
- * If not NULL, fetch buffers from this array instead of allocating them
- * with rte_pktmbuf_alloc().
*
* @return
* 0 on success, errno value on failure.
*/
static int
-rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n, struct rte_mbuf **pool)
+rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n)
{
unsigned int i;
struct rxq_elt (*elts)[elts_n] =
struct rxq_elt *elt = &(*elts)[i];
struct ibv_recv_wr *wr = &elt->wr;
struct ibv_sge *sge = &(*elts)[i].sge;
- struct rte_mbuf *buf;
+ struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
- if (pool != NULL) {
- buf = *(pool++);
- assert(buf != NULL);
- rte_pktmbuf_reset(buf);
- } else
- buf = rte_pktmbuf_alloc(rxq->mp);
if (buf == NULL) {
- assert(pool == NULL);
ERROR("%p: empty mbuf pool", (void *)rxq);
ret = ENOMEM;
goto error;
(void *)rxq, elts_n);
rxq->elts_n = elts_n;
rxq->elts_head = 0;
- rxq->elts.no_sp = elts;
+ rxq->elts = elts;
assert(ret == 0);
return 0;
error:
if (elts != NULL) {
- assert(pool == NULL);
for (i = 0; (i != elemof(*elts)); ++i) {
struct rxq_elt *elt = &(*elts)[i];
struct rte_mbuf *buf;
{
unsigned int i;
unsigned int elts_n = rxq->elts_n;
- struct rxq_elt (*elts)[elts_n] = rxq->elts.no_sp;
+ struct rxq_elt (*elts)[elts_n] = rxq->elts;
DEBUG("%p: freeing WRs", (void *)rxq);
rxq->elts_n = 0;
- rxq->elts.no_sp = NULL;
+ rxq->elts = NULL;
if (elts == NULL)
return;
for (i = 0; (i != elemof(*elts)); ++i) {
}
/**
- * Delete flow steering rule.
+ * Unregister a MAC address.
*
- * @param rxq
- * Pointer to RX queue structure.
- * @param mac_index
- * MAC address index.
+ * @param priv
+ * Pointer to private structure.
*/
static void
-rxq_del_flow(struct rxq *rxq, unsigned int mac_index)
+priv_mac_addr_del(struct priv *priv)
{
#ifndef NDEBUG
- struct priv *priv = rxq->priv;
- const uint8_t (*mac)[ETHER_ADDR_LEN] =
- (const uint8_t (*)[ETHER_ADDR_LEN])
- priv->mac[mac_index].addr_bytes;
+ uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes;
#endif
- assert(rxq->mac_flow[mac_index] != NULL);
- DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
- (void *)rxq,
- (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
- mac_index);
- claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index]));
- rxq->mac_flow[mac_index] = NULL;
-}
-/**
- * Unregister a MAC address from a RX queue.
- *
- * @param rxq
- * Pointer to RX queue structure.
- * @param mac_index
- * MAC address index.
- */
-static void
-rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index)
-{
- assert(mac_index < elemof(rxq->priv->mac));
- if (!BITFIELD_ISSET(rxq->mac_configured, mac_index))
+ if (!priv->mac_flow)
return;
- rxq_del_flow(rxq, mac_index);
- BITFIELD_RESET(rxq->mac_configured, mac_index);
+ DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x",
+ (void *)priv,
+ (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]);
+ claim_zero(ibv_destroy_flow(priv->mac_flow));
+ priv->mac_flow = NULL;
}
/**
- * Unregister all MAC addresses from a RX queue.
+ * Register a MAC address.
*
- * @param rxq
- * Pointer to RX queue structure.
- */
-static void
-rxq_mac_addrs_del(struct rxq *rxq)
-{
- struct priv *priv = rxq->priv;
- unsigned int i;
-
- for (i = 0; (i != elemof(priv->mac)); ++i)
- rxq_mac_addr_del(rxq, i);
-}
-
-/**
- * Add single flow steering rule.
+ * The MAC address is registered in queue 0.
*
- * @param rxq
- * Pointer to RX queue structure.
- * @param mac_index
- * MAC address index to register.
+ * @param priv
+ * Pointer to private structure.
*
* @return
* 0 on success, errno value on failure.
*/
static int
-rxq_add_flow(struct rxq *rxq, unsigned int mac_index)
+priv_mac_addr_add(struct priv *priv)
{
+ uint8_t (*mac)[ETHER_ADDR_LEN] = &priv->mac.addr_bytes;
+ struct rxq *rxq;
struct ibv_flow *flow;
- struct priv *priv = rxq->priv;
- const uint8_t (*mac)[ETHER_ADDR_LEN] =
- (const uint8_t (*)[ETHER_ADDR_LEN])
- priv->mac[mac_index].addr_bytes;
+
+ /* If device isn't started, this is all we need to do. */
+ if (!priv->started)
+ return 0;
+ if (priv->isolated)
+ return 0;
+ if (*priv->rxqs && (*priv->rxqs)[0])
+ rxq = (*priv->rxqs)[0];
+ else
+ return 0;
/* Allocate flow specification on the stack. */
struct __attribute__((packed)) {
struct ibv_flow_attr *attr = &data.attr;
struct ibv_flow_spec_eth *spec = &data.spec;
- assert(mac_index < elemof(priv->mac));
+ if (priv->mac_flow)
+ priv_mac_addr_del(priv);
/*
* No padding must be inserted by the compiler between attr and spec.
* This layout is expected by libibverbs.
.dst_mac = "\xff\xff\xff\xff\xff\xff",
}
};
- DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
- (void *)rxq,
- (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
- mac_index);
+ DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x",
+ (void *)priv,
+ (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5]);
/* Create related flow. */
errno = 0;
flow = ibv_create_flow(rxq->qp, attr);
return errno;
return EINVAL;
}
- assert(rxq->mac_flow[mac_index] == NULL);
- rxq->mac_flow[mac_index] = flow;
+ assert(priv->mac_flow == NULL);
+ priv->mac_flow = flow;
return 0;
}
/**
- * Register a MAC address in a RX queue.
- *
- * @param rxq
- * Pointer to RX queue structure.
- * @param mac_index
- * MAC address index to register.
+ * Clean up a RX queue.
*
- * @return
- * 0 on success, errno value on failure.
- */
-static int
-rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index)
-{
- int ret;
-
- assert(mac_index < elemof(rxq->priv->mac));
- if (BITFIELD_ISSET(rxq->mac_configured, mac_index))
- rxq_mac_addr_del(rxq, mac_index);
- ret = rxq_add_flow(rxq, mac_index);
- if (ret)
- return ret;
- BITFIELD_SET(rxq->mac_configured, mac_index);
- return 0;
-}
-
-/**
- * Register all MAC addresses in a RX queue.
+ * Destroy objects, free allocated memory and reset the structure for reuse.
*
* @param rxq
* Pointer to RX queue structure.
- *
- * @return
- * 0 on success, errno value on failure.
- */
-static int
-rxq_mac_addrs_add(struct rxq *rxq)
-{
- struct priv *priv = rxq->priv;
- unsigned int i;
- int ret;
-
- for (i = 0; (i != elemof(priv->mac)); ++i) {
- if (!BITFIELD_ISSET(priv->mac_configured, i))
- continue;
- ret = rxq_mac_addr_add(rxq, i);
- if (!ret)
- continue;
- /* Failure, rollback. */
- while (i != 0)
- rxq_mac_addr_del(rxq, --i);
- assert(ret > 0);
- return ret;
- }
- return 0;
-}
-
-/**
- * Unregister a MAC address.
- *
- * In RSS mode, the MAC address is unregistered from the parent queue,
- * otherwise it is unregistered from each queue directly.
- *
- * @param priv
- * Pointer to private structure.
- * @param mac_index
- * MAC address index.
*/
static void
-priv_mac_addr_del(struct priv *priv, unsigned int mac_index)
+rxq_cleanup(struct rxq *rxq)
{
- unsigned int i;
-
- assert(!priv->isolated);
- assert(mac_index < elemof(priv->mac));
- if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
- return;
- if (priv->rss) {
- rxq_mac_addr_del(LIST_FIRST(&priv->parents), mac_index);
- goto end;
- }
- for (i = 0; (i != priv->dev->data->nb_rx_queues); ++i)
- rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
-end:
- BITFIELD_RESET(priv->mac_configured, mac_index);
-}
-
-/**
- * Register a MAC address.
- *
- * In RSS mode, the MAC address is registered in the parent queue,
- * otherwise it is registered in each queue directly.
- *
- * @param priv
- * Pointer to private structure.
- * @param mac_index
- * MAC address index to use.
- * @param mac
- * MAC address to register.
- *
- * @return
- * 0 on success, errno value on failure.
- */
-static int
-priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
- const uint8_t (*mac)[ETHER_ADDR_LEN])
-{
- unsigned int i;
- int ret;
-
- assert(mac_index < elemof(priv->mac));
- /* First, make sure this address isn't already configured. */
- for (i = 0; (i != elemof(priv->mac)); ++i) {
- /* Skip this index, it's going to be reconfigured. */
- if (i == mac_index)
- continue;
- if (!BITFIELD_ISSET(priv->mac_configured, i))
- continue;
- if (memcmp(priv->mac[i].addr_bytes, *mac, sizeof(*mac)))
- continue;
- /* Address already configured elsewhere, return with error. */
- return EADDRINUSE;
- }
- if (BITFIELD_ISSET(priv->mac_configured, mac_index))
- priv_mac_addr_del(priv, mac_index);
- priv->mac[mac_index] = (struct ether_addr){
- {
- (*mac)[0], (*mac)[1], (*mac)[2],
- (*mac)[3], (*mac)[4], (*mac)[5]
- }
- };
- /* If device isn't started, this is all we need to do. */
- if (!priv->started) {
-#ifndef NDEBUG
- /* Verify that all queues have this index disabled. */
- for (i = 0; (i != priv->rxqs_n); ++i) {
- if ((*priv->rxqs)[i] == NULL)
- continue;
- assert(!BITFIELD_ISSET
- ((*priv->rxqs)[i]->mac_configured, mac_index));
- }
-#endif
- goto end;
- }
- if (priv->rss) {
- ret = rxq_mac_addr_add(LIST_FIRST(&priv->parents), mac_index);
- if (ret)
- return ret;
- goto end;
- }
- for (i = 0; (i != priv->rxqs_n); ++i) {
- if ((*priv->rxqs)[i] == NULL)
- continue;
- ret = rxq_mac_addr_add((*priv->rxqs)[i], mac_index);
- if (!ret)
- continue;
- /* Failure, rollback. */
- while (i != 0)
- if ((*priv->rxqs)[(--i)] != NULL)
- rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
- return ret;
- }
-end:
- BITFIELD_SET(priv->mac_configured, mac_index);
- return 0;
-}
-
-/**
- * Clean up a RX queue.
- *
- * Destroy objects, free allocated memory and reset the structure for reuse.
- *
- * @param rxq
- * Pointer to RX queue structure.
- */
-static void
-rxq_cleanup(struct rxq *rxq)
-{
- struct ibv_exp_release_intf_params params;
+ struct ibv_exp_release_intf_params params;
DEBUG("cleaning up %p", (void *)rxq);
- if (rxq->sp)
- rxq_free_elts_sp(rxq);
- else
- rxq_free_elts(rxq);
+ rxq_free_elts(rxq);
if (rxq->if_qp != NULL) {
assert(rxq->priv != NULL);
assert(rxq->priv->ctx != NULL);
rxq->if_cq,
¶ms));
}
- if (rxq->qp != NULL && !rxq->priv->isolated) {
- rxq_mac_addrs_del(rxq);
- }
if (rxq->qp != NULL)
claim_zero(ibv_destroy_qp(rxq->qp));
if (rxq->cq != NULL)
memset(rxq, 0, sizeof(*rxq));
}
-/**
- * Translate RX completion flags to packet type.
- *
- * @param flags
- * RX completion flags returned by poll_length_flags().
- *
- * @note: fix mlx4_dev_supported_ptypes_get() if any change here.
- *
- * @return
- * Packet type for struct rte_mbuf.
- */
-static inline uint32_t
-rxq_cq_to_pkt_type(uint32_t flags)
-{
- uint32_t pkt_type;
-
- if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET)
- pkt_type =
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
- RTE_PTYPE_L3_IPV4_EXT_UNKNOWN) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
- RTE_PTYPE_L3_IPV6_EXT_UNKNOWN) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV4_PACKET,
- RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV6_PACKET,
- RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN);
- else
- pkt_type =
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV4_PACKET,
- RTE_PTYPE_L3_IPV4_EXT_UNKNOWN) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV6_PACKET,
- RTE_PTYPE_L3_IPV6_EXT_UNKNOWN);
- return pkt_type;
-}
-
-/**
- * Translate RX completion flags to offload flags.
- *
- * @param[in] rxq
- * Pointer to RX queue structure.
- * @param flags
- * RX completion flags returned by poll_length_flags().
- *
- * @return
- * Offload flags (ol_flags) for struct rte_mbuf.
- */
-static inline uint32_t
-rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags)
-{
- uint32_t ol_flags = 0;
-
- if (rxq->csum)
- ol_flags |=
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IP_CSUM_OK,
- PKT_RX_IP_CKSUM_GOOD) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK,
- PKT_RX_L4_CKSUM_GOOD);
- if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
- ol_flags |=
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK,
- PKT_RX_IP_CKSUM_GOOD) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK,
- PKT_RX_L4_CKSUM_GOOD);
- return ol_flags;
-}
-
-static uint16_t
-mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n);
-
-/**
- * DPDK callback for RX with scattered packets support.
- *
- * @param dpdk_rxq
- * Generic pointer to RX queue structure.
- * @param[out] pkts
- * Array to store received packets.
- * @param pkts_n
- * Maximum number of packets in array.
- *
- * @return
- * Number of packets successfully received (<= pkts_n).
- */
-static uint16_t
-mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
-{
- struct rxq *rxq = (struct rxq *)dpdk_rxq;
- struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
- const unsigned int elts_n = rxq->elts_n;
- unsigned int elts_head = rxq->elts_head;
- struct ibv_recv_wr head;
- struct ibv_recv_wr **next = &head.next;
- struct ibv_recv_wr *bad_wr;
- unsigned int i;
- unsigned int pkts_ret = 0;
- int ret;
-
- if (unlikely(!rxq->sp))
- return mlx4_rx_burst(dpdk_rxq, pkts, pkts_n);
- if (unlikely(elts == NULL)) /* See RTE_DEV_CMD_SET_MTU. */
- return 0;
- for (i = 0; (i != pkts_n); ++i) {
- struct rxq_elt_sp *elt = &(*elts)[elts_head];
- struct ibv_recv_wr *wr = &elt->wr;
- uint64_t wr_id = wr->wr_id;
- unsigned int len;
- unsigned int pkt_buf_len;
- struct rte_mbuf *pkt_buf = NULL; /* Buffer returned in pkts. */
- struct rte_mbuf **pkt_buf_next = &pkt_buf;
- unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM;
- unsigned int j = 0;
- uint32_t flags;
-
- /* Sanity checks. */
-#ifdef NDEBUG
- (void)wr_id;
-#endif
- assert(wr_id < rxq->elts_n);
- assert(wr->sg_list == elt->sges);
- assert(wr->num_sge == elemof(elt->sges));
- assert(elts_head < rxq->elts_n);
- assert(rxq->elts_head < rxq->elts_n);
- ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
- &flags);
- if (unlikely(ret < 0)) {
- struct ibv_wc wc;
- int wcs_n;
-
- DEBUG("rxq=%p, poll_length() failed (ret=%d)",
- (void *)rxq, ret);
- /* ibv_poll_cq() must be used in case of failure. */
- wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
- if (unlikely(wcs_n == 0))
- break;
- if (unlikely(wcs_n < 0)) {
- DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
- (void *)rxq, wcs_n);
- break;
- }
- assert(wcs_n == 1);
- if (unlikely(wc.status != IBV_WC_SUCCESS)) {
- /* Whatever, just repost the offending WR. */
- DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
- " completion status (%d): %s",
- (void *)rxq, wc.wr_id, wc.status,
- ibv_wc_status_str(wc.status));
- /* Increment dropped packets counter. */
- ++rxq->stats.idropped;
- /* Link completed WRs together for repost. */
- *next = wr;
- next = &wr->next;
- goto repost;
- }
- ret = wc.byte_len;
- }
- if (ret == 0)
- break;
- len = ret;
- pkt_buf_len = len;
- /* Link completed WRs together for repost. */
- *next = wr;
- next = &wr->next;
- /*
- * Replace spent segments with new ones, concatenate and
- * return them as pkt_buf.
- */
- while (1) {
- struct ibv_sge *sge = &elt->sges[j];
- struct rte_mbuf *seg = elt->bufs[j];
- struct rte_mbuf *rep;
- unsigned int seg_tailroom;
-
- /*
- * Fetch initial bytes of packet descriptor into a
- * cacheline while allocating rep.
- */
- rte_prefetch0(seg);
- rep = rte_mbuf_raw_alloc(rxq->mp);
- if (unlikely(rep == NULL)) {
- /*
- * Unable to allocate a replacement mbuf,
- * repost WR.
- */
- DEBUG("rxq=%p, wr_id=%" PRIu64 ":"
- " can't allocate a new mbuf",
- (void *)rxq, wr_id);
- if (pkt_buf != NULL) {
- *pkt_buf_next = NULL;
- rte_pktmbuf_free(pkt_buf);
- }
- /* Increase out of memory counters. */
- ++rxq->stats.rx_nombuf;
- ++rxq->priv->dev->data->rx_mbuf_alloc_failed;
- goto repost;
- }
-#ifndef NDEBUG
- /* Poison user-modifiable fields in rep. */
- NEXT(rep) = (void *)((uintptr_t)-1);
- SET_DATA_OFF(rep, 0xdead);
- DATA_LEN(rep) = 0xd00d;
- PKT_LEN(rep) = 0xdeadd00d;
- 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. */
- assert(sge->lkey == rxq->mr->lkey);
- sge->addr = ((uintptr_t)rep->buf_addr + seg_headroom);
- elt->bufs[j] = rep;
- ++j;
- /* Update pkt_buf if it's the first segment, or link
- * seg to the previous one and update pkt_buf_next. */
- *pkt_buf_next = seg;
- pkt_buf_next = &NEXT(seg);
- /* Update seg information. */
- seg_tailroom = (seg->buf_len - seg_headroom);
- assert(sge->length == seg_tailroom);
- SET_DATA_OFF(seg, seg_headroom);
- if (likely(len <= seg_tailroom)) {
- /* Last segment. */
- DATA_LEN(seg) = len;
- PKT_LEN(seg) = len;
- /* Sanity check. */
- assert(rte_pktmbuf_headroom(seg) ==
- seg_headroom);
- assert(rte_pktmbuf_tailroom(seg) ==
- (seg_tailroom - len));
- break;
- }
- DATA_LEN(seg) = seg_tailroom;
- PKT_LEN(seg) = seg_tailroom;
- /* Sanity check. */
- assert(rte_pktmbuf_headroom(seg) == seg_headroom);
- assert(rte_pktmbuf_tailroom(seg) == 0);
- /* Fix len and clear headroom for next segments. */
- len -= seg_tailroom;
- seg_headroom = 0;
- }
- /* Update head and tail segments. */
- *pkt_buf_next = NULL;
- assert(pkt_buf != NULL);
- assert(j != 0);
- NB_SEGS(pkt_buf) = j;
- PORT(pkt_buf) = rxq->port_id;
- PKT_LEN(pkt_buf) = pkt_buf_len;
- pkt_buf->packet_type = rxq_cq_to_pkt_type(flags);
- pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
-
- /* Return packet. */
- *(pkts++) = pkt_buf;
- ++pkts_ret;
- /* Increase bytes counter. */
- rxq->stats.ibytes += pkt_buf_len;
-repost:
- if (++elts_head >= elts_n)
- elts_head = 0;
- continue;
- }
- if (unlikely(i == 0))
- return 0;
- *next = NULL;
- /* Repost WRs. */
- ret = ibv_post_recv(rxq->qp, head.next, &bad_wr);
- if (unlikely(ret)) {
- /* Inability to repost WRs is fatal. */
- DEBUG("%p: ibv_post_recv(): failed for WR %p: %s",
- (void *)rxq->priv,
- (void *)bad_wr,
- strerror(ret));
- abort();
- }
- rxq->elts_head = elts_head;
- /* Increase packets counter. */
- rxq->stats.ipackets += pkts_ret;
- return pkts_ret;
-}
-
/**
* DPDK callback for RX.
*
- * The following function is the same as mlx4_rx_burst_sp(), except it doesn't
- * manage scattered packets. Improves performance when MRU is lower than the
- * size of the first segment.
+ * The following function doesn't manage scattered packets.
*
* @param dpdk_rxq
* Generic pointer to RX queue structure.
mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
struct rxq *rxq = (struct rxq *)dpdk_rxq;
- struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
+ struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts;
const unsigned int elts_n = rxq->elts_n;
unsigned int elts_head = rxq->elts_head;
struct ibv_sge sges[pkts_n];
unsigned int pkts_ret = 0;
int ret;
- if (unlikely(rxq->sp))
- return mlx4_rx_burst_sp(dpdk_rxq, pkts, pkts_n);
for (i = 0; (i != pkts_n); ++i) {
struct rxq_elt *elt = &(*elts)[elts_head];
struct ibv_recv_wr *wr = &elt->wr;
NEXT(seg) = NULL;
PKT_LEN(seg) = len;
DATA_LEN(seg) = len;
- seg->packet_type = rxq_cq_to_pkt_type(flags);
- seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
+ seg->packet_type = 0;
+ seg->ol_flags = 0;
/* Return packet. */
*(pkts++) = seg;
++pkts_ret;
/* Increase bytes counter. */
rxq->stats.ibytes += len;
-repost:
- if (++elts_head >= elts_n)
- elts_head = 0;
- continue;
- }
- if (unlikely(i == 0))
- return 0;
- /* Repost WRs. */
- ret = rxq->if_qp->recv_burst(rxq->qp, sges, i);
- if (unlikely(ret)) {
- /* Inability to repost WRs is fatal. */
- DEBUG("%p: recv_burst(): failed (ret=%d)",
- (void *)rxq->priv,
- ret);
- abort();
- }
- rxq->elts_head = elts_head;
- /* Increase packets counter. */
- rxq->stats.ipackets += pkts_ret;
- return pkts_ret;
-}
-
-/**
- * Allocate a Queue Pair.
- * Optionally setup inline receive if supported.
- *
- * @param priv
- * Pointer to private structure.
- * @param cq
- * Completion queue to associate with QP.
- * @param desc
- * Number of descriptors in QP (hint only).
- *
- * @return
- * QP pointer or NULL in case of error.
- */
-static struct ibv_qp *
-rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
- struct ibv_exp_res_domain *rd)
-{
- struct ibv_exp_qp_init_attr attr = {
- /* CQ to be associated with the send queue. */
- .send_cq = cq,
- /* CQ to be associated with the receive queue. */
- .recv_cq = cq,
- .cap = {
- /* Max number of outstanding WRs. */
- .max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
- priv->device_attr.max_qp_wr :
- desc),
- /* Max number of scatter/gather elements in a WR. */
- .max_recv_sge = ((priv->device_attr.max_sge <
- MLX4_PMD_SGE_WR_N) ?
- priv->device_attr.max_sge :
- MLX4_PMD_SGE_WR_N),
- },
- .qp_type = IBV_QPT_RAW_PACKET,
- .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
- IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
- .pd = priv->pd,
- .res_domain = rd,
- };
-
- attr.max_inl_recv = priv->inl_recv_size;
- attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
- return ibv_exp_create_qp(priv->ctx, &attr);
-}
-
-/**
- * Allocate a RSS Queue Pair.
- * Optionally setup inline receive if supported.
- *
- * @param priv
- * Pointer to private structure.
- * @param cq
- * Completion queue to associate with QP.
- * @param desc
- * Number of descriptors in QP (hint only).
- * @param children_n
- * If nonzero, a number of children for parent QP and zero for a child.
- * @param rxq_parent
- * Pointer for a parent in a child case, NULL otherwise.
- *
- * @return
- * QP pointer or NULL in case of error.
- */
-static struct ibv_qp *
-rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
- int children_n, struct ibv_exp_res_domain *rd,
- struct rxq *rxq_parent)
-{
- struct ibv_exp_qp_init_attr attr = {
- /* CQ to be associated with the send queue. */
- .send_cq = cq,
- /* CQ to be associated with the receive queue. */
- .recv_cq = cq,
- .cap = {
- /* Max number of outstanding WRs. */
- .max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
- priv->device_attr.max_qp_wr :
- desc),
- /* Max number of scatter/gather elements in a WR. */
- .max_recv_sge = ((priv->device_attr.max_sge <
- MLX4_PMD_SGE_WR_N) ?
- priv->device_attr.max_sge :
- MLX4_PMD_SGE_WR_N),
- },
- .qp_type = IBV_QPT_RAW_PACKET,
- .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
- IBV_EXP_QP_INIT_ATTR_RES_DOMAIN |
- IBV_EXP_QP_INIT_ATTR_QPG),
- .pd = priv->pd,
- .res_domain = rd,
- };
-
- attr.max_inl_recv = priv->inl_recv_size,
- attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV;
- if (children_n > 0) {
- attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
- /* TSS isn't necessary. */
- attr.qpg.parent_attrib.tss_child_count = 0;
- attr.qpg.parent_attrib.rss_child_count =
- rte_align32pow2(children_n + 1) >> 1;
- DEBUG("initializing parent RSS queue");
- } else {
- attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
- attr.qpg.qpg_parent = rxq_parent->qp;
- DEBUG("initializing child RSS queue");
- }
- return ibv_exp_create_qp(priv->ctx, &attr);
-}
-
-/**
- * Reconfigure a RX queue with new parameters.
- *
- * rxq_rehash() does not allocate mbufs, which, if not done from the right
- * thread (such as a control thread), may corrupt the pool.
- * In case of failure, the queue is left untouched.
- *
- * @param dev
- * Pointer to Ethernet device structure.
- * @param rxq
- * RX queue pointer.
- *
- * @return
- * 0 on success, errno value on failure.
- */
-static int
-rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
-{
- struct priv *priv = rxq->priv;
- struct rxq tmpl = *rxq;
- unsigned int mbuf_n;
- unsigned int desc_n;
- struct rte_mbuf **pool;
- unsigned int i, k;
- struct ibv_exp_qp_attr mod;
- struct ibv_recv_wr *bad_wr;
- unsigned int mb_len;
- int err;
-
- mb_len = rte_pktmbuf_data_room_size(rxq->mp);
- DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq);
- /* Number of descriptors and mbufs currently allocated. */
- desc_n = (tmpl.elts_n * (tmpl.sp ? MLX4_PMD_SGE_WR_N : 1));
- mbuf_n = desc_n;
- /* Toggle RX checksum offload if hardware supports it. */
- if (priv->hw_csum) {
- tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
- rxq->csum = tmpl.csum;
- }
- if (priv->hw_csum_l2tun) {
- tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
- rxq->csum_l2tun = tmpl.csum_l2tun;
- }
- /* Enable scattered packets support for this queue if necessary. */
- assert(mb_len >= RTE_PKTMBUF_HEADROOM);
- if (dev->data->dev_conf.rxmode.enable_scatter &&
- (dev->data->dev_conf.rxmode.max_rx_pkt_len >
- (mb_len - RTE_PKTMBUF_HEADROOM))) {
- tmpl.sp = 1;
- desc_n /= MLX4_PMD_SGE_WR_N;
- } else
- tmpl.sp = 0;
- DEBUG("%p: %s scattered packets support (%u WRs)",
- (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc_n);
- /* If scatter mode is the same as before, nothing to do. */
- if (tmpl.sp == rxq->sp) {
- DEBUG("%p: nothing to do", (void *)dev);
- return 0;
- }
- /* Remove attached flows if RSS is disabled (no parent queue). */
- if (!priv->rss && !priv->isolated) {
- rxq_mac_addrs_del(&tmpl);
- /* Update original queue in case of failure. */
- memcpy(rxq->mac_configured, tmpl.mac_configured,
- sizeof(rxq->mac_configured));
- memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
- }
- /* From now on, any failure will render the queue unusable.
- * Reinitialize QP. */
- if (!tmpl.qp)
- goto skip_init;
- mod = (struct ibv_exp_qp_attr){ .qp_state = IBV_QPS_RESET };
- err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
- if (err) {
- ERROR("%p: cannot reset QP: %s", (void *)dev, strerror(err));
- assert(err > 0);
- return err;
- }
- mod = (struct ibv_exp_qp_attr){
- /* Move the QP to this state. */
- .qp_state = IBV_QPS_INIT,
- /* Primary port number. */
- .port_num = priv->port
- };
- err = ibv_exp_modify_qp(tmpl.qp, &mod,
- (IBV_EXP_QP_STATE |
- IBV_EXP_QP_PORT));
- if (err) {
- ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
- (void *)dev, strerror(err));
- assert(err > 0);
- return err;
- };
-skip_init:
- err = ibv_resize_cq(tmpl.cq, desc_n);
- if (err) {
- ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err));
- assert(err > 0);
- return err;
- }
- /* Reconfigure flows. Do not care for errors. */
- if (!priv->rss && !priv->isolated) {
- rxq_mac_addrs_add(&tmpl);
- /* Update original queue in case of failure. */
- memcpy(rxq->mac_configured, tmpl.mac_configured,
- sizeof(rxq->mac_configured));
- memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
- }
- /* Allocate pool. */
- pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
- if (pool == NULL) {
- ERROR("%p: cannot allocate memory", (void *)dev);
- return ENOBUFS;
- }
- /* Snatch mbufs from original queue. */
- k = 0;
- if (rxq->sp) {
- struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
-
- for (i = 0; (i != elemof(*elts)); ++i) {
- struct rxq_elt_sp *elt = &(*elts)[i];
- unsigned int j;
-
- for (j = 0; (j != elemof(elt->bufs)); ++j) {
- assert(elt->bufs[j] != NULL);
- pool[k++] = elt->bufs[j];
- }
- }
- } else {
- struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
-
- for (i = 0; (i != elemof(*elts)); ++i) {
- struct rxq_elt *elt = &(*elts)[i];
- struct rte_mbuf *buf = (void *)
- ((uintptr_t)elt->sge.addr -
- WR_ID(elt->wr.wr_id).offset);
-
- assert(WR_ID(elt->wr.wr_id).id == i);
- pool[k++] = buf;
- }
- }
- assert(k == mbuf_n);
- tmpl.elts_n = 0;
- tmpl.elts.sp = NULL;
- assert((void *)&tmpl.elts.sp == (void *)&tmpl.elts.no_sp);
- err = ((tmpl.sp) ?
- rxq_alloc_elts_sp(&tmpl, desc_n, pool) :
- rxq_alloc_elts(&tmpl, desc_n, pool));
- if (err) {
- ERROR("%p: cannot reallocate WRs, aborting", (void *)dev);
- rte_free(pool);
- assert(err > 0);
- return err;
+repost:
+ if (++elts_head >= elts_n)
+ elts_head = 0;
+ continue;
}
- assert(tmpl.elts_n == desc_n);
- assert(tmpl.elts.sp != NULL);
- rte_free(pool);
- /* Clean up original data. */
- rxq->elts_n = 0;
- rte_free(rxq->elts.sp);
- rxq->elts.sp = NULL;
- if (!tmpl.qp)
- goto skip_rtr;
- /* Post WRs. */
- err = ibv_post_recv(tmpl.qp,
- (tmpl.sp ?
- &(*tmpl.elts.sp)[0].wr :
- &(*tmpl.elts.no_sp)[0].wr),
- &bad_wr);
- if (err) {
- ERROR("%p: ibv_post_recv() failed for WR %p: %s",
- (void *)dev,
- (void *)bad_wr,
- strerror(err));
- goto skip_rtr;
+ if (unlikely(i == 0))
+ return 0;
+ /* Repost WRs. */
+ ret = rxq->if_qp->recv_burst(rxq->qp, sges, i);
+ if (unlikely(ret)) {
+ /* Inability to repost WRs is fatal. */
+ DEBUG("%p: recv_burst(): failed (ret=%d)",
+ (void *)rxq->priv,
+ ret);
+ abort();
}
- mod = (struct ibv_exp_qp_attr){
- .qp_state = IBV_QPS_RTR
- };
- err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
- if (err)
- ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
- (void *)dev, strerror(err));
-skip_rtr:
- *rxq = tmpl;
- assert(err >= 0);
- return err;
+ rxq->elts_head = elts_head;
+ /* Increase packets counter. */
+ rxq->stats.ipackets += pkts_ret;
+ return pkts_ret;
}
/**
- * Create verbs QP resources associated with a rxq.
+ * Allocate a Queue Pair.
+ * Optionally setup inline receive if supported.
*
- * @param rxq
- * Pointer to RX queue structure.
+ * @param priv
+ * Pointer to private structure.
+ * @param cq
+ * Completion queue to associate with QP.
* @param desc
- * Number of descriptors to configure in queue.
- * @param inactive
- * If true, the queue is disabled because its index is higher or
- * equal to the real number of queues, which must be a power of 2.
- * @param children_n
- * The number of children in a parent case, zero for a child.
- * @param rxq_parent
- * The pointer to a parent RX structure for a child in RSS case,
- * NULL for parent.
+ * Number of descriptors in QP (hint only).
*
* @return
- * 0 on success, errno value on failure.
+ * QP pointer or NULL in case of error.
*/
-int
-rxq_create_qp(struct rxq *rxq,
- uint16_t desc,
- int inactive,
- int children_n,
- struct rxq *rxq_parent)
+static struct ibv_qp *
+rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+ struct ibv_exp_res_domain *rd)
{
- int ret;
- struct ibv_exp_qp_attr mod;
- struct ibv_exp_query_intf_params params;
- enum ibv_exp_query_intf_status status;
- struct ibv_recv_wr *bad_wr;
- int parent = (children_n > 0);
- struct priv *priv = rxq->priv;
-
- if (priv->rss && !inactive && (rxq_parent || parent))
- rxq->qp = rxq_setup_qp_rss(priv, rxq->cq, desc,
- children_n, rxq->rd,
- rxq_parent);
- else
- rxq->qp = rxq_setup_qp(priv, rxq->cq, desc, rxq->rd);
- if (rxq->qp == NULL) {
- ret = (errno ? errno : EINVAL);
- ERROR("QP creation failure: %s",
- strerror(ret));
- return ret;
- }
- mod = (struct ibv_exp_qp_attr){
- /* Move the QP to this state. */
- .qp_state = IBV_QPS_INIT,
- /* Primary port number. */
- .port_num = priv->port
- };
- ret = ibv_exp_modify_qp(rxq->qp, &mod,
- (IBV_EXP_QP_STATE |
- (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
- IBV_EXP_QP_PORT));
- if (ret) {
- ERROR("QP state to IBV_QPS_INIT failed: %s",
- strerror(ret));
- return ret;
- }
- if (!priv->isolated && (parent || !priv->rss)) {
- /* Configure MAC and broadcast addresses. */
- ret = rxq_mac_addrs_add(rxq);
- if (ret) {
- ERROR("QP flow attachment failed: %s",
- strerror(ret));
- return ret;
- }
- }
- if (!parent) {
- ret = ibv_post_recv(rxq->qp,
- (rxq->sp ?
- &(*rxq->elts.sp)[0].wr :
- &(*rxq->elts.no_sp)[0].wr),
- &bad_wr);
- if (ret) {
- ERROR("ibv_post_recv() failed for WR %p: %s",
- (void *)bad_wr,
- strerror(ret));
- return ret;
- }
- }
- mod = (struct ibv_exp_qp_attr){
- .qp_state = IBV_QPS_RTR
- };
- ret = ibv_exp_modify_qp(rxq->qp, &mod, IBV_EXP_QP_STATE);
- if (ret) {
- ERROR("QP state to IBV_QPS_RTR failed: %s",
- strerror(ret));
- return ret;
- }
- params = (struct ibv_exp_query_intf_params){
- .intf_scope = IBV_EXP_INTF_GLOBAL,
- .intf = IBV_EXP_INTF_QP_BURST,
- .obj = rxq->qp,
+ struct ibv_exp_qp_init_attr attr = {
+ /* CQ to be associated with the send queue. */
+ .send_cq = cq,
+ /* CQ to be associated with the receive queue. */
+ .recv_cq = cq,
+ .cap = {
+ /* Max number of outstanding WRs. */
+ .max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
+ priv->device_attr.max_qp_wr :
+ desc),
+ /* Max number of scatter/gather elements in a WR. */
+ .max_recv_sge = 1,
+ },
+ .qp_type = IBV_QPT_RAW_PACKET,
+ .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+ IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
+ .pd = priv->pd,
+ .res_domain = rd,
};
- rxq->if_qp = ibv_exp_query_intf(priv->ctx, ¶ms, &status);
- if (rxq->if_qp == NULL) {
- ERROR("QP interface family query failed with status %d",
- status);
- return errno;
- }
- return 0;
+
+ return ibv_exp_create_qp(priv->ctx, &attr);
}
/**
* Number of descriptors to configure in queue.
* @param socket
* NUMA socket on which memory must be allocated.
- * @param inactive
- * If true, the queue is disabled because its index is higher or
- * equal to the real number of queues, which must be a power of 2.
* @param[in] conf
* Thresholds parameters.
* @param mp
* Memory pool for buffer allocations.
- * @param children_n
- * The number of children in a parent case, zero for a child.
- * @param rxq_parent
- * The pointer to a parent RX structure (or NULL) in a child case,
- * NULL for parent.
*
* @return
* 0 on success, errno value on failure.
*/
static int
rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
- unsigned int socket, int inactive,
- const struct rte_eth_rxconf *conf,
- struct rte_mempool *mp, int children_n,
- struct rxq *rxq_parent)
+ unsigned int socket, const struct rte_eth_rxconf *conf,
+ struct rte_mempool *mp)
{
struct priv *priv = dev->data->dev_private;
struct rxq tmpl = {
.mp = mp,
.socket = socket
};
+ struct ibv_exp_qp_attr mod;
union {
struct ibv_exp_query_intf_params params;
struct ibv_exp_cq_init_attr cq;
struct ibv_exp_res_domain_init_attr rd;
} attr;
enum ibv_exp_query_intf_status status;
+ struct ibv_recv_wr *bad_wr;
unsigned int mb_len;
int ret = 0;
- int parent = (children_n > 0);
(void)conf; /* Thresholds configuration (ignored). */
- /*
- * If this is a parent queue, hardware must support RSS and
- * RSS must be enabled.
- */
- assert((!parent) || ((priv->hw_rss) && (priv->rss)));
- if (parent) {
- /* Even if unused, ibv_create_cq() requires at least one
- * descriptor. */
- desc = 1;
- goto skip_mr;
- }
mb_len = rte_pktmbuf_data_room_size(mp);
- if ((desc == 0) || (desc % MLX4_PMD_SGE_WR_N)) {
- ERROR("%p: invalid number of RX descriptors (must be a"
- " multiple of %d)", (void *)dev, MLX4_PMD_SGE_WR_N);
+ if (desc == 0) {
+ ERROR("%p: invalid number of Rx descriptors", (void *)dev);
return EINVAL;
}
- /* Toggle RX checksum offload if hardware supports it. */
- if (priv->hw_csum)
- tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
- if (priv->hw_csum_l2tun)
- tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
/* Enable scattered packets support for this queue if necessary. */
assert(mb_len >= RTE_PKTMBUF_HEADROOM);
if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
(mb_len - RTE_PKTMBUF_HEADROOM)) {
- tmpl.sp = 0;
+ ;
} else if (dev->data->dev_conf.rxmode.enable_scatter) {
- tmpl.sp = 1;
- desc /= MLX4_PMD_SGE_WR_N;
+ WARN("%p: scattered mode has been requested but is"
+ " not supported, this may lead to packet loss",
+ (void *)dev);
} else {
WARN("%p: the requested maximum Rx packet size (%u) is"
" larger than a single mbuf (%u) and scattered"
dev->data->dev_conf.rxmode.max_rx_pkt_len,
mb_len - RTE_PKTMBUF_HEADROOM);
}
- DEBUG("%p: %s scattered packets support (%u WRs)",
- (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc);
/* Use the entire RX mempool as the memory region. */
tmpl.mr = mlx4_mp2mr(priv->pd, mp);
if (tmpl.mr == NULL) {
(void *)dev, strerror(ret));
goto error;
}
-skip_mr:
attr.rd = (struct ibv_exp_res_domain_init_attr){
.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
IBV_EXP_RES_DOMAIN_MSG_MODEL),
priv->device_attr.max_qp_wr);
DEBUG("priv->device_attr.max_sge is %d",
priv->device_attr.max_sge);
- /* Allocate descriptors for RX queues, except for the RSS parent. */
- if (parent)
- goto skip_alloc;
- if (tmpl.sp)
- ret = rxq_alloc_elts_sp(&tmpl, desc, NULL);
- else
- ret = rxq_alloc_elts(&tmpl, desc, NULL);
+ tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
+ if (tmpl.qp == NULL) {
+ ret = (errno ? errno : EINVAL);
+ ERROR("%p: QP creation failure: %s",
+ (void *)dev, strerror(ret));
+ goto error;
+ }
+ mod = (struct ibv_exp_qp_attr){
+ /* Move the QP to this state. */
+ .qp_state = IBV_QPS_INIT,
+ /* Primary port number. */
+ .port_num = priv->port
+ };
+ ret = ibv_exp_modify_qp(tmpl.qp, &mod,
+ IBV_EXP_QP_STATE |
+ IBV_EXP_QP_PORT);
+ if (ret) {
+ ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+ (void *)dev, strerror(ret));
+ goto error;
+ }
+ ret = rxq_alloc_elts(&tmpl, desc);
if (ret) {
ERROR("%p: RXQ allocation failed: %s",
(void *)dev, strerror(ret));
- return ret;
+ goto error;
}
-skip_alloc:
- if (parent || rxq_parent || !priv->rss) {
- ret = rxq_create_qp(&tmpl, desc, inactive,
- children_n, rxq_parent);
- if (ret)
- goto error;
+ ret = ibv_post_recv(tmpl.qp, &(*tmpl.elts)[0].wr, &bad_wr);
+ if (ret) {
+ ERROR("%p: ibv_post_recv() failed for WR %p: %s",
+ (void *)dev,
+ (void *)bad_wr,
+ strerror(ret));
+ goto error;
+ }
+ mod = (struct ibv_exp_qp_attr){
+ .qp_state = IBV_QPS_RTR
+ };
+ ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+ if (ret) {
+ ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+ (void *)dev, strerror(ret));
+ goto error;
}
/* Save port ID. */
tmpl.port_id = dev->data->port_id;
};
tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
if (tmpl.if_cq == NULL) {
- ret = EINVAL;
ERROR("%p: CQ interface family query failed with status %d",
(void *)dev, status);
goto error;
}
+ attr.params = (struct ibv_exp_query_intf_params){
+ .intf_scope = IBV_EXP_INTF_GLOBAL,
+ .intf = IBV_EXP_INTF_QP_BURST,
+ .obj = tmpl.qp,
+ };
+ tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+ if (tmpl.if_qp == NULL) {
+ ERROR("%p: QP interface family query failed with status %d",
+ (void *)dev, status);
+ goto error;
+ }
/* Clean up rxq in case we're reinitializing it. */
DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq);
rxq_cleanup(rxq);
unsigned int socket, const struct rte_eth_rxconf *conf,
struct rte_mempool *mp)
{
- struct rxq *parent;
struct priv *priv = dev->data->dev_private;
struct rxq *rxq = (*priv->rxqs)[idx];
- int inactive = 0;
int ret;
priv_lock(priv);
return -EEXIST;
}
(*priv->rxqs)[idx] = NULL;
+ if (idx == 0)
+ priv_mac_addr_del(priv);
rxq_cleanup(rxq);
} else {
rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket);
return -ENOMEM;
}
}
- if (priv->rss && !priv->isolated) {
- /* The list consists of the single default one. */
- parent = LIST_FIRST(&priv->parents);
- if (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1)
- inactive = 1;
- } else {
- parent = NULL;
- }
- ret = rxq_setup(dev, rxq, desc, socket,
- inactive, conf, mp, 0, parent);
+ ret = rxq_setup(dev, rxq, desc, socket, conf, mp);
if (ret)
rte_free(rxq);
else {
(void *)dev, (void *)rxq);
(*priv->rxqs)[idx] = rxq;
/* Update receive callback. */
- if (rxq->sp)
- dev->rx_pkt_burst = mlx4_rx_burst_sp;
- else
- dev->rx_pkt_burst = mlx4_rx_burst;
+ dev->rx_pkt_burst = mlx4_rx_burst;
}
priv_unlock(priv);
return -ret;
DEBUG("%p: removing RX queue %p from list",
(void *)priv->dev, (void *)rxq);
(*priv->rxqs)[i] = NULL;
+ if (i == 0)
+ priv_mac_addr_del(priv);
break;
}
rxq_cleanup(rxq);
mlx4_dev_start(struct rte_eth_dev *dev)
{
struct priv *priv = dev->data->dev_private;
- unsigned int i = 0;
- unsigned int r;
- struct rxq *rxq;
int ret;
priv_lock(priv);
}
DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
priv->started = 1;
- if (priv->isolated) {
- rxq = NULL;
- r = 1;
- } else if (priv->rss) {
- rxq = LIST_FIRST(&priv->parents);
- r = 1;
- } else {
- rxq = (*priv->rxqs)[0];
- r = priv->rxqs_n;
- }
- /* Iterate only once when RSS is enabled. */
- do {
- /* Ignore nonexistent RX queues. */
- if (rxq == NULL)
- continue;
- ret = rxq_mac_addrs_add(rxq);
- if (!ret)
- continue;
- WARN("%p: QP flow attachment failed: %s",
- (void *)dev, strerror(ret));
+ ret = priv_mac_addr_add(priv);
+ if (ret)
goto err;
- } while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
ret = priv_dev_link_interrupt_handler_install(priv, dev);
if (ret) {
ERROR("%p: LSC handler install failed",
return 0;
err:
/* Rollback. */
- while (i != 0) {
- rxq = (*priv->rxqs)[i--];
- if (rxq != NULL) {
- rxq_mac_addrs_del(rxq);
- }
- }
+ priv_mac_addr_del(priv);
priv->started = 0;
priv_unlock(priv);
return -ret;
mlx4_dev_stop(struct rte_eth_dev *dev)
{
struct priv *priv = dev->data->dev_private;
- unsigned int i = 0;
- unsigned int r;
- struct rxq *rxq;
priv_lock(priv);
if (!priv->started) {
}
DEBUG("%p: detaching flows from all RX queues", (void *)dev);
priv->started = 0;
- if (priv->isolated) {
- rxq = NULL;
- r = 1;
- } else if (priv->rss) {
- rxq = LIST_FIRST(&priv->parents);
- r = 1;
- } else {
- rxq = (*priv->rxqs)[0];
- r = priv->rxqs_n;
- }
mlx4_priv_flow_stop(priv);
- /* Iterate only once when RSS is enabled. */
- do {
- /* Ignore nonexistent RX queues. */
- if (rxq == NULL)
- continue;
- rxq_mac_addrs_del(rxq);
- } while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+ priv_mac_addr_del(priv);
priv_unlock(priv);
}
DEBUG("%p: closing device \"%s\"",
(void *)dev,
((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+ priv_mac_addr_del(priv);
/* Prevent crashes when queues are still in use. This is unfortunately
* still required for DPDK 1.3 because some programs (such as testpmd)
* never release them before closing the device. */
priv->txqs_n = 0;
priv->txqs = NULL;
}
- if (priv->rss)
- priv_parent_list_cleanup(priv);
if (priv->pd != NULL) {
assert(priv->ctx != NULL);
claim_zero(ibv_dealloc_pd(priv->pd));
{
struct rte_eth_dev *dev = priv->dev;
int err;
- unsigned int i;
if (up) {
err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
if (err)
return err;
- for (i = 0; i < priv->rxqs_n; i++)
- if ((*priv->rxqs)[i]->sp)
- break;
- /* Check if an sp queue exists.
- * Note: Some old frames might be received.
- */
- if (i == priv->rxqs_n)
- dev->rx_pkt_burst = mlx4_rx_burst;
- else
- dev->rx_pkt_burst = mlx4_rx_burst_sp;
- dev->tx_pkt_burst = mlx4_tx_burst;
+ dev->rx_pkt_burst = mlx4_rx_burst;
} else {
err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
if (err)
info->max_rx_queues = max;
info->max_tx_queues = max;
/* Last array entry is reserved for broadcast. */
- info->max_mac_addrs = (elemof(priv->mac) - 1);
- info->rx_offload_capa =
- (priv->hw_csum ?
- (DEV_RX_OFFLOAD_IPV4_CKSUM |
- DEV_RX_OFFLOAD_UDP_CKSUM |
- DEV_RX_OFFLOAD_TCP_CKSUM) :
- 0);
- info->tx_offload_capa =
- (priv->hw_csum ?
- (DEV_TX_OFFLOAD_IPV4_CKSUM |
- DEV_TX_OFFLOAD_UDP_CKSUM |
- DEV_TX_OFFLOAD_TCP_CKSUM) :
- 0);
+ info->max_mac_addrs = 1;
+ info->rx_offload_capa = 0;
+ info->tx_offload_capa = 0;
if (priv_get_ifname(priv, &ifname) == 0)
info->if_index = if_nametoindex(ifname);
info->speed_capa =
priv_unlock(priv);
}
-static const uint32_t *
-mlx4_dev_supported_ptypes_get(struct rte_eth_dev *dev)
-{
- static const uint32_t ptypes[] = {
- /* refers to rxq_cq_to_pkt_type() */
- RTE_PTYPE_L3_IPV4,
- RTE_PTYPE_L3_IPV6,
- RTE_PTYPE_INNER_L3_IPV4,
- RTE_PTYPE_INNER_L3_IPV6,
- RTE_PTYPE_UNKNOWN
- };
-
- if (dev->rx_pkt_burst == mlx4_rx_burst ||
- dev->rx_pkt_burst == mlx4_rx_burst_sp)
- return ptypes;
- return NULL;
-}
-
/**
* DPDK callback to get device statistics.
*
priv_unlock(priv);
}
-/**
- * DPDK callback to remove a MAC address.
- *
- * @param dev
- * Pointer to Ethernet device structure.
- * @param index
- * MAC address index.
- */
-static void
-mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
-{
- struct priv *priv = dev->data->dev_private;
-
- 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. */
- if (index >= (elemof(priv->mac) - 1))
- goto end;
- priv_mac_addr_del(priv, index);
-end:
- priv_unlock(priv);
-}
-
-/**
- * DPDK callback to add a MAC address.
- *
- * @param dev
- * Pointer to Ethernet device structure.
- * @param mac_addr
- * MAC address to register.
- * @param index
- * MAC address index.
- * @param vmdq
- * VMDq pool index to associate address with (ignored).
- */
-static int
-mlx4_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
- uint32_t index, uint32_t vmdq)
-{
- struct priv *priv = dev->data->dev_private;
- int re;
-
- (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 (index >= (elemof(priv->mac) - 1)) {
- re = EINVAL;
- goto end;
- }
- re = priv_mac_addr_add(priv, index,
- (const uint8_t (*)[ETHER_ADDR_LEN])
- mac_addr->addr_bytes);
-end:
- priv_unlock(priv);
- return -re;
-}
-
-/**
- * DPDK callback to set the primary MAC address.
- *
- * @param dev
- * Pointer to Ethernet device structure.
- * @param mac_addr
- * MAC address to register.
- */
-static void
-mlx4_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
-{
- DEBUG("%p: setting primary MAC address", (void *)dev);
- mlx4_mac_addr_remove(dev, 0);
- mlx4_mac_addr_add(dev, mac_addr, 0, 0);
-}
-
/**
* DPDK callback to retrieve physical link information.
*
/**
* DPDK callback to change the MTU.
*
- * Setting the MTU affects hardware MRU (packets larger than the MTU cannot be
- * received). Use this as a hint to enable/disable scattered packets support
- * and improve performance when not needed.
- * Since failure is not an option, reconfiguring queues on the fly is not
- * recommended.
- *
* @param dev
* Pointer to Ethernet device structure.
* @param in_mtu
{
struct priv *priv = dev->data->dev_private;
int ret = 0;
- unsigned int i;
- uint16_t (*rx_func)(void *, struct rte_mbuf **, uint16_t) =
- mlx4_rx_burst;
priv_lock(priv);
/* Set kernel interface MTU first. */
} else
DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
priv->mtu = mtu;
- /* Temporarily replace RX handler with a fake one, assuming it has not
- * been copied elsewhere. */
- dev->rx_pkt_burst = removed_rx_burst;
- /* Make sure everyone has left mlx4_rx_burst() and uses
- * removed_rx_burst() instead. */
- rte_wmb();
- usleep(1000);
- /* Reconfigure each RX queue. */
- for (i = 0; (i != priv->rxqs_n); ++i) {
- struct rxq *rxq = (*priv->rxqs)[i];
- unsigned int max_frame_len;
-
- if (rxq == NULL)
- continue;
- /* Calculate new maximum frame length according to MTU. */
- max_frame_len = (priv->mtu + ETHER_HDR_LEN +
- (ETHER_MAX_VLAN_FRAME_LEN - ETHER_MAX_LEN));
- /* Provide new values to rxq_setup(). */
- dev->data->dev_conf.rxmode.jumbo_frame =
- (max_frame_len > ETHER_MAX_LEN);
- dev->data->dev_conf.rxmode.max_rx_pkt_len = max_frame_len;
- ret = rxq_rehash(dev, rxq);
- if (ret) {
- /* Force SP RX if that queue requires it and abort. */
- if (rxq->sp)
- rx_func = mlx4_rx_burst_sp;
- break;
- }
- /* Reenable non-RSS queue attributes. No need to check
- * for errors at this stage. */
- if (!priv->rss && !priv->isolated) {
- rxq_mac_addrs_add(rxq);
- }
- /* Scattered burst function takes priority. */
- if (rxq->sp)
- rx_func = mlx4_rx_burst_sp;
- }
- /* Burst functions can now be called again. */
- rte_wmb();
- dev->rx_pkt_burst = rx_func;
out:
priv_unlock(priv);
assert(ret >= 0);
.stats_get = mlx4_stats_get,
.stats_reset = mlx4_stats_reset,
.dev_infos_get = mlx4_dev_infos_get,
- .dev_supported_ptypes_get = mlx4_dev_supported_ptypes_get,
.rx_queue_setup = mlx4_rx_queue_setup,
.tx_queue_setup = mlx4_tx_queue_setup,
.rx_queue_release = mlx4_rx_queue_release,
.tx_queue_release = mlx4_tx_queue_release,
.flow_ctrl_get = mlx4_dev_get_flow_ctrl,
.flow_ctrl_set = mlx4_dev_set_flow_ctrl,
- .mac_addr_remove = mlx4_mac_addr_remove,
- .mac_addr_add = mlx4_mac_addr_add,
- .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,
return 0;
}
-/**
- * Retrieve integer value from environment variable.
- *
- * @param[in] name
- * Environment variable name.
- *
- * @return
- * Integer value, 0 if the variable is not set.
- */
-static int
-mlx4_getenv_int(const char *name)
-{
- const char *val = getenv(name);
-
- if (val == NULL)
- return 0;
- return atoi(val);
-}
-
static void
mlx4_dev_link_status_handler(void *);
static void
struct ibv_pd *pd = NULL;
struct priv *priv = NULL;
struct rte_eth_dev *eth_dev = NULL;
- struct ibv_exp_device_attr exp_device_attr;
struct ether_addr mac;
/* If port is not enabled, skip. */
if (!(conf.ports.enabled & (1 << i)))
continue;
- exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
- exp_device_attr.comp_mask |= IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ;
DEBUG("using port %u", port);
priv->port = port;
priv->pd = pd;
priv->mtu = ETHER_MTU;
- if (ibv_exp_query_device(ctx, &exp_device_attr)) {
- ERROR("ibv_exp_query_device() failed");
- err = ENODEV;
- goto port_error;
- }
- if ((exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_QPG) &&
- (exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_UD_RSS) &&
- (exp_device_attr.comp_mask &
- IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ) &&
- (exp_device_attr.max_rss_tbl_sz > 0)) {
- priv->hw_qpg = 1;
- priv->hw_rss = 1;
- priv->max_rss_tbl_sz = exp_device_attr.max_rss_tbl_sz;
- } else {
- priv->hw_qpg = 0;
- priv->hw_rss = 0;
- priv->max_rss_tbl_sz = 0;
- }
- priv->hw_tss = !!(exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_UD_TSS);
- DEBUG("device flags: %s%s%s",
- (priv->hw_qpg ? "IBV_DEVICE_QPG " : ""),
- (priv->hw_tss ? "IBV_DEVICE_TSS " : ""),
- (priv->hw_rss ? "IBV_DEVICE_RSS " : ""));
- if (priv->hw_rss)
- DEBUG("maximum RSS indirection table size: %u",
- exp_device_attr.max_rss_tbl_sz);
-
- priv->hw_csum =
- ((exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
- (exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
- DEBUG("checksum offloading is %ssupported",
- (priv->hw_csum ? "" : "not "));
-
- priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
- IBV_EXP_DEVICE_VXLAN_SUPPORT);
- DEBUG("L2 tunnel checksum offloads are %ssupported",
- (priv->hw_csum_l2tun ? "" : "not "));
-
- priv->inl_recv_size = mlx4_getenv_int("MLX4_INLINE_RECV_SIZE");
-
- if (priv->inl_recv_size) {
- exp_device_attr.comp_mask =
- IBV_EXP_DEVICE_ATTR_INLINE_RECV_SZ;
- if (ibv_exp_query_device(ctx, &exp_device_attr)) {
- INFO("Couldn't query device for inline-receive"
- " capabilities.");
- priv->inl_recv_size = 0;
- } else {
- if ((unsigned)exp_device_attr.inline_recv_sz <
- priv->inl_recv_size) {
- INFO("Max inline-receive (%d) <"
- " requested inline-receive (%u)",
- exp_device_attr.inline_recv_sz,
- priv->inl_recv_size);
- priv->inl_recv_size =
- exp_device_attr.inline_recv_sz;
- }
- }
- INFO("Set inline receive size to %u",
- priv->inl_recv_size);
- }
priv->vf = vf;
/* Configure the first MAC address by default. */
mac.addr_bytes[0], mac.addr_bytes[1],
mac.addr_bytes[2], mac.addr_bytes[3],
mac.addr_bytes[4], mac.addr_bytes[5]);
- /* Register MAC and broadcast addresses. */
- claim_zero(priv_mac_addr_add(priv, 0,
- (const uint8_t (*)[ETHER_ADDR_LEN])
- mac.addr_bytes));
- claim_zero(priv_mac_addr_add(priv, (elemof(priv->mac) - 1),
- &(const uint8_t [ETHER_ADDR_LEN])
- { "\xff\xff\xff\xff\xff\xff" }));
+ /* Register MAC address. */
+ priv->mac = mac;
+ if (priv_mac_addr_add(priv))
+ goto port_error;
#ifndef NDEBUG
{
char ifname[IF_NAMESIZE];
}
eth_dev->data->dev_private = priv;
- eth_dev->data->mac_addrs = priv->mac;
+ eth_dev->data->mac_addrs = &priv->mac;
eth_dev->device = &pci_dev->device;
rte_eth_copy_pci_info(eth_dev, pci_dev);