* [dpdk-dev] [PATCH v2] net/mlx4: fix no Rx interrupts
@ 2017-10-24 13:11 Moti Haimovsky
2017-10-24 14:18 ` Nélio Laranjeiro
0 siblings, 1 reply; 2+ messages in thread
From: Moti Haimovsky @ 2017-10-24 13:11 UTC (permalink / raw)
To: adrien.mazarguil; +Cc: dev, Moti Haimovsky
This commit addresses the issue of rx interrupts support with
the new Rx datapath introduced in DPDK version 17.11.
In order to generate an Rx interrupt an event queue is armed with the
consumer index of the Rx completion queue. Since version 17.11 this
index is handled by the PMD so it is now the responsibility of the
PMD to write this value when enabling Rx interrupts.
Fixes: 6681b845034c ("net/mlx4: add Rx bypassing Verbs")
Signed-off-by: Moti Haimovsky <motih@mellanox.com>
---
V2:
* Rebased on top of ff3397e9 ("net/mlx4: relax Rx queue configuration order")
---
drivers/net/mlx4/mlx4_intr.c | 38 +++++++++++++++++++++++++++++++++++++-
drivers/net/mlx4/mlx4_prm.h | 9 +++++++++
drivers/net/mlx4/mlx4_rxq.c | 6 +++++-
3 files changed, 51 insertions(+), 2 deletions(-)
diff --git a/drivers/net/mlx4/mlx4_intr.c b/drivers/net/mlx4/mlx4_intr.c
index 3806322..217c591 100644
--- a/drivers/net/mlx4/mlx4_intr.c
+++ b/drivers/net/mlx4/mlx4_intr.c
@@ -53,6 +53,7 @@
#include <rte_alarm.h>
#include <rte_errno.h>
#include <rte_ethdev.h>
+#include <rte_io.h>
#include <rte_interrupts.h>
#include "mlx4.h"
@@ -239,6 +240,40 @@
}
/**
+ * MLX4 CQ notification .
+ *
+ * @param rxq
+ * Pointer to receive queue structure.
+ * @param solicited
+ * Is request solicited or not.
+ *
+ * @return
+ * Always 0.
+ */
+static int
+mlx4_arm_cq(struct rxq *rxq, int solicited)
+{
+ struct mlx4_cq *cq = &rxq->mcq;
+ uint64_t doorbell;
+ uint32_t sn = cq->arm_sn & MLX4_CQ_DB_GEQ_N_MASK;
+ uint32_t ci = cq->cons_index & MLX4_CQ_DB_CI_MASK;
+ uint32_t cmd = solicited ? MLX4_CQ_DB_REQ_NOT_SOL : MLX4_CQ_DB_REQ_NOT;
+ void *cq_db_reg = (char *)cq->cq_uar + MLX4_CQ_DOORBELL;
+
+ *cq->arm_db = rte_cpu_to_be_32(sn << 28 | cmd | ci);
+ /*
+ * Make sure that the doorbell record in host memory is
+ * written before ringing the doorbell via PCI MMIO.
+ */
+ rte_wmb();
+ doorbell = sn << 28 | cmd | cq->cqn;
+ doorbell <<= 32;
+ doorbell |= ci;
+ rte_write64(rte_cpu_to_be_64(doorbell), cq_db_reg);
+ return 0;
+}
+
+/**
* Uninstall interrupt handler.
*
* @param priv
@@ -333,6 +368,7 @@
WARN("unable to disable interrupt on rx queue %d",
idx);
} else {
+ rxq->mcq.arm_sn++;
ibv_ack_cq_events(rxq->cq, 1);
}
return -ret;
@@ -358,7 +394,7 @@
if (!rxq || !rxq->channel)
ret = EINVAL;
else
- ret = ibv_req_notify_cq(rxq->cq, 0);
+ ret = mlx4_arm_cq(rxq, 0);
if (ret) {
rte_errno = ret;
WARN("unable to arm interrupt on rx queue %d", idx);
diff --git a/drivers/net/mlx4/mlx4_prm.h b/drivers/net/mlx4/mlx4_prm.h
index 3a77502..92f2fec 100644
--- a/drivers/net/mlx4/mlx4_prm.h
+++ b/drivers/net/mlx4/mlx4_prm.h
@@ -78,6 +78,11 @@ enum {
MLX4_CQE_L2_TUNNEL_IPOK = (int)(1u << 31),
};
+/* Completion queue events, numbers and masks. */
+#define MLX4_CQ_DB_GEQ_N_MASK 0x3
+#define MLX4_CQ_DB_CI_MASK 0xffffff
+#define MLX4_CQ_DOORBELL 0x20
+
/* Send queue information. */
struct mlx4_sq {
uint8_t *buf; /**< SQ buffer. */
@@ -100,6 +105,10 @@ struct mlx4_cq {
uint32_t cqe_64:1; /**< CQ entry size is 64 bytes. */
uint32_t cons_index; /**< Last queue entry that was handled. */
uint32_t *set_ci_db; /**< Pointer to the completion queue doorbell. */
+ uint32_t *arm_db; /**< Pointer to doorbell for arming Rx events. */
+ int arm_sn; /**< Rx event counter. */
+ void *cq_uar; /**< CQ user access region. */
+ uint32_t cqn; /**< CQ number. */
};
/**
diff --git a/drivers/net/mlx4/mlx4_rxq.c b/drivers/net/mlx4/mlx4_rxq.c
index 4c50077..24262cb 100644
--- a/drivers/net/mlx4/mlx4_rxq.c
+++ b/drivers/net/mlx4/mlx4_rxq.c
@@ -510,7 +510,7 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
struct rte_mbuf *(*elts)[elts_n] = rxq->elts;
struct mlx4dv_obj mlxdv;
struct mlx4dv_rwq dv_rwq;
- struct mlx4dv_cq dv_cq;
+ struct mlx4dv_cq dv_cq = { .comp_mask = MLX4DV_CQ_MASK_UAR, };
const char *msg;
struct ibv_cq *cq = NULL;
struct ibv_wq *wq = NULL;
@@ -604,6 +604,10 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
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;
+ rxq->mcq.arm_db = dv_cq.arm_db;
+ rxq->mcq.arm_sn = dv_cq.arm_sn;
+ rxq->mcq.cqn = dv_cq.cqn;
+ rxq->mcq.cq_uar = dv_cq.cq_uar;
/* Update doorbell counter. */
rxq->rq_ci = elts_n / sges_n;
rte_wmb();
--
1.8.3.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [dpdk-dev] [PATCH v2] net/mlx4: fix no Rx interrupts
2017-10-24 13:11 [dpdk-dev] [PATCH v2] net/mlx4: fix no Rx interrupts Moti Haimovsky
@ 2017-10-24 14:18 ` Nélio Laranjeiro
0 siblings, 0 replies; 2+ messages in thread
From: Nélio Laranjeiro @ 2017-10-24 14:18 UTC (permalink / raw)
To: Moti Haimovsky; +Cc: adrien.mazarguil, dev
Hi Moti,
On Tue, Oct 24, 2017 at 04:11:14PM +0300, Moti Haimovsky wrote:
> This commit addresses the issue of rx interrupts support with
> the new Rx datapath introduced in DPDK version 17.11.
> In order to generate an Rx interrupt an event queue is armed with the
> consumer index of the Rx completion queue. Since version 17.11 this
> index is handled by the PMD so it is now the responsibility of the
> PMD to write this value when enabling Rx interrupts.
>
> Fixes: 6681b845034c ("net/mlx4: add Rx bypassing Verbs")
>
> Signed-off-by: Moti Haimovsky <motih@mellanox.com>
> ---
> V2:
> * Rebased on top of ff3397e9 ("net/mlx4: relax Rx queue configuration order")
> ---
> drivers/net/mlx4/mlx4_intr.c | 38 +++++++++++++++++++++++++++++++++++++-
> drivers/net/mlx4/mlx4_prm.h | 9 +++++++++
> drivers/net/mlx4/mlx4_rxq.c | 6 +++++-
> 3 files changed, 51 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/mlx4/mlx4_intr.c b/drivers/net/mlx4/mlx4_intr.c
> index 3806322..217c591 100644
> --- a/drivers/net/mlx4/mlx4_intr.c
> +++ b/drivers/net/mlx4/mlx4_intr.c
> @@ -53,6 +53,7 @@
> #include <rte_alarm.h>
> #include <rte_errno.h>
> #include <rte_ethdev.h>
> +#include <rte_io.h>
> #include <rte_interrupts.h>
>
> #include "mlx4.h"
> @@ -239,6 +240,40 @@
> }
>
> /**
> + * MLX4 CQ notification .
> + *
> + * @param rxq
> + * Pointer to receive queue structure.
> + * @param solicited
> + * Is request solicited or not.
> + *
> + * @return
> + * Always 0.
Why not setting the return type to void in this case?
> + */
> +static int
> +mlx4_arm_cq(struct rxq *rxq, int solicited)
> +{
> + struct mlx4_cq *cq = &rxq->mcq;
> + uint64_t doorbell;
> + uint32_t sn = cq->arm_sn & MLX4_CQ_DB_GEQ_N_MASK;
> + uint32_t ci = cq->cons_index & MLX4_CQ_DB_CI_MASK;
> + uint32_t cmd = solicited ? MLX4_CQ_DB_REQ_NOT_SOL : MLX4_CQ_DB_REQ_NOT;
> + void *cq_db_reg = (char *)cq->cq_uar + MLX4_CQ_DOORBELL;
use uint8_t* instead of char*.
By the way, why not storing the address to the doorbell directly in the
structure instead of computing it each time it is necessary?
> + *cq->arm_db = rte_cpu_to_be_32(sn << 28 | cmd | ci);
> + /*
> + * Make sure that the doorbell record in host memory is
> + * written before ringing the doorbell via PCI MMIO.
> + */
> + rte_wmb();
> + doorbell = sn << 28 | cmd | cq->cqn;
> + doorbell <<= 32;
> + doorbell |= ci;
> + rte_write64(rte_cpu_to_be_64(doorbell), cq_db_reg);
> + return 0;
> +}
> +
> +/**
> * Uninstall interrupt handler.
> *
> * @param priv
> @@ -333,6 +368,7 @@
> WARN("unable to disable interrupt on rx queue %d",
> idx);
> } else {
> + rxq->mcq.arm_sn++;
> ibv_ack_cq_events(rxq->cq, 1);
> }
> return -ret;
> @@ -358,7 +394,7 @@
> if (!rxq || !rxq->channel)
> ret = EINVAL;
> else
> - ret = ibv_req_notify_cq(rxq->cq, 0);
> + ret = mlx4_arm_cq(rxq, 0);
> if (ret) {
> rte_errno = ret;
> WARN("unable to arm interrupt on rx queue %d", idx);
> diff --git a/drivers/net/mlx4/mlx4_prm.h b/drivers/net/mlx4/mlx4_prm.h
> index 3a77502..92f2fec 100644
> --- a/drivers/net/mlx4/mlx4_prm.h
> +++ b/drivers/net/mlx4/mlx4_prm.h
> @@ -78,6 +78,11 @@ enum {
> MLX4_CQE_L2_TUNNEL_IPOK = (int)(1u << 31),
> };
>
> +/* Completion queue events, numbers and masks. */
> +#define MLX4_CQ_DB_GEQ_N_MASK 0x3
> +#define MLX4_CQ_DB_CI_MASK 0xffffff
You could have replaced all the places where this mask is hard coded by
this new define for consistency.
> +#define MLX4_CQ_DOORBELL 0x20
> +
> /* Send queue information. */
> struct mlx4_sq {
> uint8_t *buf; /**< SQ buffer. */
> @@ -100,6 +105,10 @@ struct mlx4_cq {
> uint32_t cqe_64:1; /**< CQ entry size is 64 bytes. */
> uint32_t cons_index; /**< Last queue entry that was handled. */
> uint32_t *set_ci_db; /**< Pointer to the completion queue doorbell. */
> + uint32_t *arm_db; /**< Pointer to doorbell for arming Rx events. */
> + int arm_sn; /**< Rx event counter. */
> + void *cq_uar; /**< CQ user access region. */
> + uint32_t cqn; /**< CQ number. */
Try to re-arrange the variable order to avoid holes in the structure.
> };
>
> /**
> diff --git a/drivers/net/mlx4/mlx4_rxq.c b/drivers/net/mlx4/mlx4_rxq.c
> index 4c50077..24262cb 100644
> --- a/drivers/net/mlx4/mlx4_rxq.c
> +++ b/drivers/net/mlx4/mlx4_rxq.c
> @@ -510,7 +510,7 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
> struct rte_mbuf *(*elts)[elts_n] = rxq->elts;
> struct mlx4dv_obj mlxdv;
> struct mlx4dv_rwq dv_rwq;
> - struct mlx4dv_cq dv_cq;
> + struct mlx4dv_cq dv_cq = { .comp_mask = MLX4DV_CQ_MASK_UAR, };
> const char *msg;
> struct ibv_cq *cq = NULL;
> struct ibv_wq *wq = NULL;
> @@ -604,6 +604,10 @@ void mlx4_rss_detach(struct mlx4_rss *rss)
> 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;
> + rxq->mcq.arm_db = dv_cq.arm_db;
> + rxq->mcq.arm_sn = dv_cq.arm_sn;
> + rxq->mcq.cqn = dv_cq.cqn;
> + rxq->mcq.cq_uar = dv_cq.cq_uar;
> /* Update doorbell counter. */
> rxq->rq_ci = elts_n / sges_n;
> rte_wmb();
> --
> 1.8.3.1
Thanks,
--
Nélio Laranjeiro
6WIND
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2017-10-24 14:18 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-10-24 13:11 [dpdk-dev] [PATCH v2] net/mlx4: fix no Rx interrupts Moti Haimovsky
2017-10-24 14:18 ` Nélio Laranjeiro
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).