* [dpdk-dev] [PATCH v1 4/4] net/mlx4: relax Rx queue configuration order
2017-10-19 16:11 [dpdk-dev] [PATCH v1 0/4] net/mlx4: RSS fixes Adrien Mazarguil
` (2 preceding siblings ...)
2017-10-19 16:11 ` [dpdk-dev] [PATCH v1 3/4] net/mlx4: fix indirection table error rollback Adrien Mazarguil
@ 2017-10-19 16:11 ` Adrien Mazarguil
2017-10-23 19:18 ` [dpdk-dev] [PATCH v1 0/4] net/mlx4: RSS fixes Ferruh Yigit
4 siblings, 0 replies; 6+ messages in thread
From: Adrien Mazarguil @ 2017-10-19 16:11 UTC (permalink / raw)
To: Ferruh Yigit; +Cc: Nelio Laranjeiro, dev
Various hardware limitations apply to RSS indirection tables, one of them
being they must be an exact 1:1 mapping of the configured Rx queue indices.
While this restriction is enforced when creating RSS flow rules, it is not
the case when Rx queues themselves are created; underlying WQ numbers are
assigned in turn, not according to queue index.
Applications such as l3fwd-power that create Rx queues from highest to
lowest index (or any other non-sequential order) thus fail to get a working
RSS context.
This commit postpones WQ initialization to dev_start(), once all Rx queues
are configured in order to address this issue.
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
drivers/net/mlx4/mlx4.c | 23 +--
drivers/net/mlx4/mlx4_rxq.c | 357 +++++++++++++++++++++++++++++---------
drivers/net/mlx4/mlx4_rxtx.h | 5 +
3 files changed, 288 insertions(+), 97 deletions(-)
diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c
index a297b9a..9be875b 100644
--- a/drivers/net/mlx4/mlx4.c
+++ b/drivers/net/mlx4/mlx4.c
@@ -50,7 +50,6 @@
#pragma GCC diagnostic ignored "-Wpedantic"
#endif
#include <infiniband/verbs.h>
-#include <infiniband/mlx4dv.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-Wpedantic"
#endif
@@ -100,20 +99,8 @@ mlx4_dev_configure(struct rte_eth_dev *dev)
{
struct priv *priv = dev->data->dev_private;
struct rte_flow_error error;
- uint8_t log2_range = rte_log2_u32(dev->data->nb_rx_queues);
int ret;
- /* Prepare range for RSS contexts before creating the first WQ. */
- ret = mlx4dv_set_context_attr(priv->ctx,
- MLX4DV_SET_CTX_ATTR_LOG_WQS_RANGE_SZ,
- &log2_range);
- if (ret) {
- ERROR("cannot set up range size for RSS context to %u"
- " (for %u Rx queues), error: %s",
- 1 << log2_range, dev->data->nb_rx_queues, strerror(ret));
- rte_errno = ret;
- return -ret;
- }
/* Prepare internal flow rules. */
ret = mlx4_flow_sync(priv, &error);
if (ret) {
@@ -128,7 +115,8 @@ mlx4_dev_configure(struct rte_eth_dev *dev)
/**
* DPDK callback to start the device.
*
- * Simulate device start by attaching all configured flows.
+ * Simulate device start by initializing common RSS resources and attaching
+ * all configured flows.
*
* @param dev
* Pointer to Ethernet device structure.
@@ -147,6 +135,12 @@ mlx4_dev_start(struct rte_eth_dev *dev)
return 0;
DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
priv->started = 1;
+ ret = mlx4_rss_init(priv);
+ if (ret) {
+ ERROR("%p: cannot initialize RSS resources: %s",
+ (void *)dev, strerror(-ret));
+ goto err;
+ }
ret = mlx4_intr_install(priv);
if (ret) {
ERROR("%p: interrupt handler installation failed",
@@ -194,6 +188,7 @@ mlx4_dev_stop(struct rte_eth_dev *dev)
rte_wmb();
mlx4_flow_sync(priv, NULL);
mlx4_intr_uninstall(priv);
+ mlx4_rss_deinit(priv);
}
/**
diff --git a/drivers/net/mlx4/mlx4_rxq.c b/drivers/net/mlx4/mlx4_rxq.c
index ee29556..fb28290 100644
--- a/drivers/net/mlx4/mlx4_rxq.c
+++ b/drivers/net/mlx4/mlx4_rxq.c
@@ -46,6 +46,7 @@
#ifdef PEDANTIC
#pragma GCC diagnostic ignored "-Wpedantic"
#endif
+#include <infiniband/mlx4dv.h>
#include <infiniband/verbs.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-Wpedantic"
@@ -201,7 +202,7 @@ int mlx4_rss_attach(struct mlx4_rss *rss)
struct ibv_wq *ind_tbl[rss->queues];
struct priv *priv = rss->priv;
const char *msg;
- unsigned int i;
+ unsigned int i = 0;
int ret;
if (!rte_is_power_of_2(RTE_DIM(ind_tbl))) {
@@ -220,6 +221,12 @@ int mlx4_rss_attach(struct mlx4_rss *rss)
msg = "RSS target queue is not configured";
goto error;
}
+ ret = mlx4_rxq_attach(rxq);
+ if (ret) {
+ ret = -ret;
+ msg = "unable to attach RSS target queue";
+ goto error;
+ }
ind_tbl[i] = rxq->wq;
}
rss->ind = ibv_create_rwq_ind_table
@@ -286,6 +293,8 @@ int mlx4_rss_attach(struct mlx4_rss *rss)
claim_zero(ibv_destroy_rwq_ind_table(rss->ind));
rss->ind = NULL;
}
+ while (i--)
+ mlx4_rxq_detach(priv->dev->data->rx_queues[rss->queue_id[i]]);
ERROR("mlx4: %s", msg);
--rss->usecnt;
rte_errno = ret;
@@ -305,6 +314,9 @@ int mlx4_rss_attach(struct mlx4_rss *rss)
*/
void mlx4_rss_detach(struct mlx4_rss *rss)
{
+ struct priv *priv = rss->priv;
+ unsigned int i;
+
assert(rss->refcnt);
assert(rss->qp);
assert(rss->ind);
@@ -314,10 +326,164 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
rss->qp = NULL;
claim_zero(ibv_destroy_rwq_ind_table(rss->ind));
rss->ind = NULL;
+ for (i = 0; i != rss->queues; ++i)
+ mlx4_rxq_detach(priv->dev->data->rx_queues[rss->queue_id[i]]);
}
/**
- * Allocate Rx queue elements.
+ * Initialize common RSS context resources.
+ *
+ * Because ConnectX-3 hardware limitations require a fixed order in the
+ * indirection table, WQs must be allocated sequentially to be part of a
+ * common RSS context.
+ *
+ * Since a newly created WQ cannot be moved to a different context, this
+ * function allocates them all at once, one for each configured Rx queue,
+ * as well as all related resources (CQs and mbufs).
+ *
+ * This must therefore be done before creating any Rx flow rules relying on
+ * indirection tables.
+ *
+ * @param priv
+ * Pointer to private structure.
+ *
+ * @return
+ * 0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx4_rss_init(struct priv *priv)
+{
+ struct rte_eth_dev *dev = priv->dev;
+ uint8_t log2_range = rte_log2_u32(dev->data->nb_rx_queues);
+ uint32_t wq_num_prev = 0;
+ const char *msg;
+ unsigned int i;
+ int ret;
+
+ /* Prepare range for RSS contexts before creating the first WQ. */
+ ret = mlx4dv_set_context_attr(priv->ctx,
+ MLX4DV_SET_CTX_ATTR_LOG_WQS_RANGE_SZ,
+ &log2_range);
+ if (ret) {
+ ERROR("cannot set up range size for RSS context to %u"
+ " (for %u Rx queues), error: %s",
+ 1 << log2_range, dev->data->nb_rx_queues, strerror(ret));
+ rte_errno = ret;
+ return -ret;
+ }
+ for (i = 0; i != priv->dev->data->nb_rx_queues; ++i) {
+ struct rxq *rxq = priv->dev->data->rx_queues[i];
+ struct ibv_cq *cq;
+ struct ibv_wq *wq;
+ uint32_t wq_num;
+
+ /* Attach the configured Rx queues. */
+ if (rxq) {
+ assert(!rxq->usecnt);
+ ret = mlx4_rxq_attach(rxq);
+ if (!ret) {
+ wq_num = rxq->wq->wq_num;
+ goto wq_num_check;
+ }
+ ret = -ret;
+ msg = "unable to create Rx queue resources";
+ goto error;
+ }
+ /*
+ * WQs are temporarily allocated for unconfigured Rx queues
+ * to maintain proper index alignment in indirection table
+ * by skipping unused WQ numbers.
+ *
+ * The reason this works at all even though these WQs are
+ * immediately destroyed is that WQNs are allocated
+ * sequentially and are guaranteed to never be reused in the
+ * same context by the underlying implementation.
+ */
+ cq = ibv_create_cq(priv->ctx, 1, NULL, NULL, 0);
+ if (!cq) {
+ ret = ENOMEM;
+ msg = "placeholder CQ creation failure";
+ goto error;
+ }
+ wq = ibv_create_wq
+ (priv->ctx,
+ &(struct ibv_wq_init_attr){
+ .wq_type = IBV_WQT_RQ,
+ .max_wr = 1,
+ .max_sge = 1,
+ .pd = priv->pd,
+ .cq = cq,
+ });
+ if (wq) {
+ wq_num = wq->wq_num;
+ claim_zero(ibv_destroy_wq(wq));
+ }
+ claim_zero(ibv_destroy_cq(cq));
+ if (!wq) {
+ ret = ENOMEM;
+ msg = "placeholder WQ creation failure";
+ goto error;
+ }
+wq_num_check:
+ /*
+ * While guaranteed by the implementation, make sure WQ
+ * numbers are really sequential (as the saying goes,
+ * trust, but verify).
+ */
+ if (i && wq_num - wq_num_prev != 1) {
+ if (rxq)
+ mlx4_rxq_detach(rxq);
+ ret = ERANGE;
+ msg = "WQ numbers are not sequential";
+ goto error;
+ }
+ wq_num_prev = wq_num;
+ }
+ return 0;
+error:
+ ERROR("cannot initialize common RSS resources (queue %u): %s: %s",
+ i, msg, strerror(ret));
+ while (i--) {
+ struct rxq *rxq = priv->dev->data->rx_queues[i];
+
+ if (rxq)
+ mlx4_rxq_detach(rxq);
+ }
+ rte_errno = ret;
+ return -ret;
+}
+
+/**
+ * Release common RSS context resources.
+ *
+ * As the reverse of mlx4_rss_init(), this must be done after removing all
+ * flow rules relying on indirection tables.
+ *
+ * @param priv
+ * Pointer to private structure.
+ */
+void
+mlx4_rss_deinit(struct priv *priv)
+{
+ unsigned int i;
+
+ for (i = 0; i != priv->dev->data->nb_rx_queues; ++i) {
+ struct rxq *rxq = priv->dev->data->rx_queues[i];
+
+ if (rxq) {
+ assert(rxq->usecnt == 1);
+ mlx4_rxq_detach(rxq);
+ }
+ }
+}
+
+/**
+ * Attach a user to a Rx queue.
+ *
+ * Used when the resources of an Rx queue must be instantiated for it to
+ * become in a usable state.
+ *
+ * This function increments the usage count of the Rx queue.
*
* @param rxq
* Pointer to Rx queue structure.
@@ -325,17 +491,76 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
* @return
* 0 on success, negative errno value otherwise and rte_errno is set.
*/
-static int
-mlx4_rxq_alloc_elts(struct rxq *rxq)
+int
+mlx4_rxq_attach(struct rxq *rxq)
{
+ if (rxq->usecnt++) {
+ assert(rxq->cq);
+ assert(rxq->wq);
+ assert(rxq->wqes);
+ assert(rxq->rq_db);
+ return 0;
+ }
+
+ struct priv *priv = rxq->priv;
const uint32_t elts_n = 1 << rxq->elts_n;
const uint32_t sges_n = 1 << rxq->sges_n;
struct rte_mbuf *(*elts)[elts_n] = rxq->elts;
+ struct mlx4dv_obj mlxdv;
+ struct mlx4dv_rwq dv_rwq;
+ struct mlx4dv_cq dv_cq;
+ const char *msg;
+ struct ibv_cq *cq = NULL;
+ struct ibv_wq *wq = NULL;
+ volatile struct mlx4_wqe_data_seg (*wqes)[];
unsigned int i;
+ int ret;
assert(rte_is_power_of_2(elts_n));
+ cq = ibv_create_cq(priv->ctx, elts_n / sges_n, NULL, rxq->channel, 0);
+ if (!cq) {
+ ret = ENOMEM;
+ msg = "CQ creation failure";
+ goto error;
+ }
+ wq = ibv_create_wq
+ (priv->ctx,
+ &(struct ibv_wq_init_attr){
+ .wq_type = IBV_WQT_RQ,
+ .max_wr = elts_n / sges_n,
+ .max_sge = sges_n,
+ .pd = priv->pd,
+ .cq = cq,
+ });
+ if (!wq) {
+ ret = errno ? errno : EINVAL;
+ msg = "WQ creation failure";
+ goto error;
+ }
+ ret = ibv_modify_wq
+ (wq,
+ &(struct ibv_wq_attr){
+ .attr_mask = IBV_WQ_ATTR_STATE,
+ .wq_state = IBV_WQS_RDY,
+ });
+ if (ret) {
+ msg = "WQ state change to IBV_WQS_RDY failed";
+ goto error;
+ }
+ /* Retrieve device queue information. */
+ mlxdv.cq.in = cq;
+ mlxdv.cq.out = &dv_cq;
+ mlxdv.rwq.in = wq;
+ mlxdv.rwq.out = &dv_rwq;
+ ret = mlx4dv_init_obj(&mlxdv, MLX4DV_OBJ_RWQ | MLX4DV_OBJ_CQ);
+ if (ret) {
+ msg = "failed to obtain device information from WQ/CQ objects";
+ goto error;
+ }
+ wqes = (volatile struct mlx4_wqe_data_seg (*)[])
+ ((uintptr_t)dv_rwq.buf.buf + dv_rwq.rq.offset);
for (i = 0; i != RTE_DIM(*elts); ++i) {
- volatile struct mlx4_wqe_data_seg *scat = &(*rxq->wqes)[i];
+ volatile struct mlx4_wqe_data_seg *scat = &(*wqes)[i];
struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
if (buf == NULL) {
@@ -343,8 +568,9 @@ mlx4_rxq_alloc_elts(struct rxq *rxq)
rte_pktmbuf_free_seg((*elts)[i]);
(*elts)[i] = NULL;
}
- rte_errno = ENOMEM;
- return -rte_errno;
+ ret = ENOMEM;
+ msg = "cannot allocate mbuf";
+ goto error;
}
/* Headroom is reserved by rte_pktmbuf_alloc(). */
assert(buf->data_off == RTE_PKTMBUF_HEADROOM);
@@ -368,21 +594,55 @@ mlx4_rxq_alloc_elts(struct rxq *rxq)
}
DEBUG("%p: allocated and configured %u segments (max %u packets)",
(void *)rxq, elts_n, elts_n / sges_n);
+ rxq->cq = cq;
+ rxq->wq = wq;
+ rxq->wqes = wqes;
+ rxq->rq_db = dv_rwq.rdb;
+ rxq->mcq.buf = dv_cq.buf.buf;
+ rxq->mcq.cqe_cnt = dv_cq.cqe_cnt;
+ rxq->mcq.set_ci_db = dv_cq.set_ci_db;
+ rxq->mcq.cqe_64 = (dv_cq.cqe_size & 64) ? 1 : 0;
+ /* Update doorbell counter. */
+ rxq->rq_ci = elts_n / sges_n;
+ rte_wmb();
+ *rxq->rq_db = rte_cpu_to_be_32(rxq->rq_ci);
return 0;
+error:
+ if (wq)
+ claim_zero(ibv_destroy_wq(wq));
+ if (cq)
+ claim_zero(ibv_destroy_cq(cq));
+ rte_errno = ret;
+ ERROR("error while attaching Rx queue %p: %s: %s",
+ (void *)rxq, msg, strerror(ret));
+ return -ret;
}
/**
- * Free Rx queue elements.
+ * Detach a user from a Rx queue.
+ *
+ * This function decrements the usage count of the Rx queue and destroys
+ * usage resources after reaching 0.
*
* @param rxq
* Pointer to Rx queue structure.
*/
-static void
-mlx4_rxq_free_elts(struct rxq *rxq)
+void
+mlx4_rxq_detach(struct rxq *rxq)
{
unsigned int i;
struct rte_mbuf *(*elts)[1 << rxq->elts_n] = rxq->elts;
+ if (--rxq->usecnt)
+ return;
+ rxq->rq_ci = 0;
+ memset(&rxq->mcq, 0, sizeof(rxq->mcq));
+ rxq->rq_db = NULL;
+ rxq->wqes = NULL;
+ claim_zero(ibv_destroy_wq(rxq->wq));
+ rxq->wq = NULL;
+ claim_zero(ibv_destroy_cq(rxq->cq));
+ rxq->cq = NULL;
DEBUG("%p: freeing Rx queue elements", (void *)rxq);
for (i = 0; (i != RTE_DIM(*elts)); ++i) {
if (!(*elts)[i])
@@ -417,9 +677,6 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
struct rte_mempool *mp)
{
struct priv *priv = dev->data->dev_private;
- struct mlx4dv_obj mlxdv;
- struct mlx4dv_rwq dv_rwq;
- struct mlx4dv_cq dv_cq;
uint32_t mb_len = rte_pktmbuf_data_room_size(mp);
struct rte_mbuf *(*elts)[rte_align32pow2(desc)];
struct rxq *rxq;
@@ -560,73 +817,8 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
goto error;
}
}
- rxq->cq = ibv_create_cq(priv->ctx, desc >> rxq->sges_n, NULL,
- rxq->channel, 0);
- if (!rxq->cq) {
- rte_errno = ENOMEM;
- ERROR("%p: CQ creation failure: %s",
- (void *)dev, strerror(rte_errno));
- goto error;
- }
- rxq->wq = ibv_create_wq
- (priv->ctx,
- &(struct ibv_wq_init_attr){
- .wq_type = IBV_WQT_RQ,
- .max_wr = desc >> rxq->sges_n,
- .max_sge = 1 << rxq->sges_n,
- .pd = priv->pd,
- .cq = rxq->cq,
- });
- if (!rxq->wq) {
- rte_errno = errno ? errno : EINVAL;
- ERROR("%p: WQ creation failure: %s",
- (void *)dev, strerror(rte_errno));
- goto error;
- }
- ret = ibv_modify_wq
- (rxq->wq,
- &(struct ibv_wq_attr){
- .attr_mask = IBV_WQ_ATTR_STATE,
- .wq_state = IBV_WQS_RDY,
- });
- if (ret) {
- rte_errno = ret;
- ERROR("%p: WQ state to IBV_WPS_RDY failed: %s",
- (void *)dev, strerror(rte_errno));
- goto error;
- }
- /* Retrieve device queue information. */
- mlxdv.cq.in = rxq->cq;
- mlxdv.cq.out = &dv_cq;
- mlxdv.rwq.in = rxq->wq;
- mlxdv.rwq.out = &dv_rwq;
- ret = mlx4dv_init_obj(&mlxdv, MLX4DV_OBJ_RWQ | MLX4DV_OBJ_CQ);
- if (ret) {
- rte_errno = EINVAL;
- ERROR("%p: failed to obtain device information", (void *)dev);
- goto error;
- }
- rxq->wqes =
- (volatile struct mlx4_wqe_data_seg (*)[])
- ((uintptr_t)dv_rwq.buf.buf + dv_rwq.rq.offset);
- rxq->rq_db = dv_rwq.rdb;
- rxq->rq_ci = 0;
- rxq->mcq.buf = dv_cq.buf.buf;
- rxq->mcq.cqe_cnt = dv_cq.cqe_cnt;
- rxq->mcq.set_ci_db = dv_cq.set_ci_db;
- rxq->mcq.cqe_64 = (dv_cq.cqe_size & 64) ? 1 : 0;
- ret = mlx4_rxq_alloc_elts(rxq);
- if (ret) {
- ERROR("%p: RXQ allocation failed: %s",
- (void *)dev, strerror(rte_errno));
- goto error;
- }
DEBUG("%p: adding Rx queue %p to list", (void *)dev, (void *)rxq);
dev->data->rx_queues[idx] = rxq;
- /* Update doorbell counter. */
- rxq->rq_ci = desc >> rxq->sges_n;
- rte_wmb();
- *rxq->rq_db = rte_cpu_to_be_32(rxq->rq_ci);
return 0;
error:
dev->data->rx_queues[idx] = NULL;
@@ -660,11 +852,10 @@ mlx4_rx_queue_release(void *dpdk_rxq)
priv->dev->data->rx_queues[i] = NULL;
break;
}
- mlx4_rxq_free_elts(rxq);
- if (rxq->wq)
- claim_zero(ibv_destroy_wq(rxq->wq));
- if (rxq->cq)
- claim_zero(ibv_destroy_cq(rxq->cq));
+ assert(!rxq->cq);
+ assert(!rxq->wq);
+ assert(!rxq->wqes);
+ assert(!rxq->rq_db);
if (rxq->channel)
claim_zero(ibv_destroy_comp_channel(rxq->channel));
if (rxq->mr)
diff --git a/drivers/net/mlx4/mlx4_rxtx.h b/drivers/net/mlx4/mlx4_rxtx.h
index e10bbca..7d67748 100644
--- a/drivers/net/mlx4/mlx4_rxtx.h
+++ b/drivers/net/mlx4/mlx4_rxtx.h
@@ -83,6 +83,7 @@ struct rxq {
struct mlx4_cq mcq; /**< Info for directly manipulating the CQ. */
struct mlx4_rxq_stats stats; /**< Rx queue counters. */
unsigned int socket; /**< CPU socket ID for allocations. */
+ uint32_t usecnt; /**< Number of users relying on queue resources. */
uint8_t data[]; /**< Remaining queue resources. */
};
@@ -146,12 +147,16 @@ struct txq {
/* mlx4_rxq.c */
uint8_t mlx4_rss_hash_key_default[MLX4_RSS_HASH_KEY_SIZE];
+int mlx4_rss_init(struct priv *priv);
+void mlx4_rss_deinit(struct priv *priv);
struct mlx4_rss *mlx4_rss_get(struct priv *priv, uint64_t fields,
uint8_t key[MLX4_RSS_HASH_KEY_SIZE],
uint16_t queues, const uint16_t queue_id[]);
void mlx4_rss_put(struct mlx4_rss *rss);
int mlx4_rss_attach(struct mlx4_rss *rss);
void mlx4_rss_detach(struct mlx4_rss *rss);
+int mlx4_rxq_attach(struct rxq *rxq);
+void mlx4_rxq_detach(struct rxq *rxq);
int mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx,
uint16_t desc, unsigned int socket,
const struct rte_eth_rxconf *conf,
--
2.1.4
^ permalink raw reply [flat|nested] 6+ messages in thread