[dpdk-dev] [PATCH v2 06/32] net/sfc: support link status change interrupt

[Andrew Rybchenko <arybchenko@solarflare.com>]

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
Reviewed-by: Robert Stonehouse <rstonehouse@solarflare.com>
---
 doc/guides/nics/features/sfc_efx.ini |   1 +
 doc/guides/nics/sfc_efx.rst          |   4 +-
 drivers/net/sfc/sfc.c                |   4 +-
 drivers/net/sfc/sfc.h                |   4 +
 drivers/net/sfc/sfc_ethdev.c         |   1 +
 drivers/net/sfc/sfc_ev.c             |  30 +++++-
 drivers/net/sfc/sfc_intr.c           | 204 +++++++++++++++++++++++++++++++++++
 7 files changed, 242 insertions(+), 6 deletions(-)

diff --git a/doc/guides/nics/features/sfc_efx.ini b/doc/guides/nics/features/sfc_efx.ini
index 25472f8..693d35e 100644
--- a/doc/guides/nics/features/sfc_efx.ini
+++ b/doc/guides/nics/features/sfc_efx.ini
@@ -5,6 +5,7 @@
 ;
 [Features]
 Link status          = Y
+Link status event    = Y
 Flow control         = Y
 L3 checksum offload  = P
 L4 checksum offload  = P
diff --git a/doc/guides/nics/sfc_efx.rst b/doc/guides/nics/sfc_efx.rst
index 06b0c14..58f8242 100644
--- a/doc/guides/nics/sfc_efx.rst
+++ b/doc/guides/nics/sfc_efx.rst
@@ -44,7 +44,7 @@ SFC EFX PMD has support for:
 
 - Multiple transmit and receive queues
 
-- Link state information
+- Link state information including link status change interrupt
 
 - IPv4/IPv6 TCP/UDP transmit checksum offload
 
@@ -61,8 +61,6 @@ Non-supported Features
 
 The features not yet supported include:
 
-- Link status change interrupt
-
 - Receive queue interupts
 
 - Priority-based flow control
diff --git a/drivers/net/sfc/sfc.c b/drivers/net/sfc/sfc.c
index ef9e0d4..36044a0 100644
--- a/drivers/net/sfc/sfc.c
+++ b/drivers/net/sfc/sfc.c
@@ -116,7 +116,9 @@ sfc_check_conf(struct sfc_adapter *sa)
 		rc = EINVAL;
 	}
 
-	if (conf->intr_conf.lsc != 0) {
+	if ((conf->intr_conf.lsc != 0) &&
+	    (sa->intr.type != EFX_INTR_LINE) &&
+	    (sa->intr.type != EFX_INTR_MESSAGE)) {
 		sfc_err(sa, "Link status change interrupt not supported");
 		rc = EINVAL;
 	}
diff --git a/drivers/net/sfc/sfc.h b/drivers/net/sfc/sfc.h
index e9f5183..be33464 100644
--- a/drivers/net/sfc/sfc.h
+++ b/drivers/net/sfc/sfc.h
@@ -112,6 +112,8 @@ struct sfc_mcdi {
 
 struct sfc_intr {
 	efx_intr_type_t			type;
+	rte_intr_callback_fn		handler;
+	boolean_t			lsc_intr;
 };
 
 struct sfc_evq_info;
@@ -119,6 +121,8 @@ struct sfc_rxq_info;
 struct sfc_txq_info;
 
 struct sfc_port {
+	unsigned int			lsc_seq;
+
 	unsigned int			flow_ctrl;
 	boolean_t			flow_ctrl_autoneg;
 	size_t				pdu;
diff --git a/drivers/net/sfc/sfc_ethdev.c b/drivers/net/sfc/sfc_ethdev.c
index eff648b..8c46500 100644
--- a/drivers/net/sfc/sfc_ethdev.c
+++ b/drivers/net/sfc/sfc_ethdev.c
@@ -682,6 +682,7 @@ static struct eth_driver sfc_efx_pmd = {
 	.pci_drv = {
 		.id_table = pci_id_sfc_efx_map,
 		.drv_flags =
+			RTE_PCI_DRV_INTR_LSC |
 			RTE_PCI_DRV_NEED_MAPPING,
 		.probe = rte_eth_dev_pci_probe,
 		.remove = rte_eth_dev_pci_remove,
diff --git a/drivers/net/sfc/sfc_ev.c b/drivers/net/sfc/sfc_ev.c
index de42845..c788986 100644
--- a/drivers/net/sfc/sfc_ev.c
+++ b/drivers/net/sfc/sfc_ev.c
@@ -287,11 +287,25 @@ sfc_ev_link_change(void *arg, efx_link_mode_t link_mode)
 	struct sfc_adapter *sa = evq->sa;
 	struct rte_eth_link *dev_link = &sa->eth_dev->data->dev_link;
 	struct rte_eth_link new_link;
+	uint64_t new_link_u64;
+	uint64_t old_link_u64;
 
 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 
 	sfc_port_link_mode_to_info(link_mode, &new_link);
-	rte_atomic64_set((rte_atomic64_t *)dev_link, *(uint64_t *)&new_link);
+
+	new_link_u64 = *(uint64_t *)&new_link;
+	do {
+		old_link_u64 = rte_atomic64_read((rte_atomic64_t *)dev_link);
+		if (old_link_u64 == new_link_u64)
+			break;
+
+		if (rte_atomic64_cmpset((volatile uint64_t *)dev_link,
+					old_link_u64, new_link_u64)) {
+			evq->sa->port.lsc_seq++;
+			break;
+		}
+	} while (B_TRUE);
 
 	return B_FALSE;
 }
@@ -521,6 +535,12 @@ sfc_ev_start(struct sfc_adapter *sa)
 	if (rc != 0)
 		goto fail_mgmt_evq_start;
 
+	if (sa->intr.lsc_intr) {
+		rc = sfc_ev_qprime(sa->evq_info[sa->mgmt_evq_index].evq);
+		if (rc != 0)
+			goto fail_evq0_prime;
+	}
+
 	rte_spinlock_unlock(&sa->mgmt_evq_lock);
 
 	/*
@@ -538,6 +558,9 @@ sfc_ev_start(struct sfc_adapter *sa)
 
 	return 0;
 
+fail_evq0_prime:
+	sfc_ev_qstop(sa, 0);
+
 fail_mgmt_evq_start:
 	rte_spinlock_unlock(&sa->mgmt_evq_lock);
 	efx_ev_fini(sa->nic);
@@ -639,7 +662,10 @@ sfc_ev_qinit_info(struct sfc_adapter *sa, unsigned int sw_index)
 	SFC_ASSERT(rte_is_power_of_2(max_entries));
 
 	evq_info->max_entries = max_entries;
-	evq_info->flags = sa->evq_flags | EFX_EVQ_FLAGS_NOTIFY_DISABLED;
+	evq_info->flags = sa->evq_flags |
+		((sa->intr.lsc_intr && sw_index == sa->mgmt_evq_index) ?
+			EFX_EVQ_FLAGS_NOTIFY_INTERRUPT :
+			EFX_EVQ_FLAGS_NOTIFY_DISABLED);
 
 	return 0;
 }
diff --git a/drivers/net/sfc/sfc_intr.c b/drivers/net/sfc/sfc_intr.c
index 1b7dcdd..e0b1693 100644
--- a/drivers/net/sfc/sfc_intr.c
+++ b/drivers/net/sfc/sfc_intr.c
@@ -27,10 +27,130 @@
  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
+/*
+ * At the momemt of writing DPDK v16.07 has notion of two types of
+ * interrupts: LSC (link status change) and RXQ (receive indication).
+ * It allows to register interrupt callback for entire device which is
+ * not intended to be used for receive indication (i.e. link status
+ * change indication only). The handler has no information which HW
+ * interrupt has triggered it, so we don't know which event queue should
+ * be polled/reprimed (except qmask in the case of legacy line interrupt).
+ */
+
+#include <rte_common.h>
+#include <rte_interrupts.h>
+
 #include "efx.h"
 
 #include "sfc.h"
 #include "sfc_log.h"
+#include "sfc_ev.h"
+
+static void
+sfc_intr_handle_mgmt_evq(struct sfc_adapter *sa)
+{
+	struct sfc_evq *evq;
+
+	rte_spinlock_lock(&sa->mgmt_evq_lock);
+
+	evq = sa->evq_info[sa->mgmt_evq_index].evq;
+
+	if (evq->init_state != SFC_EVQ_STARTED) {
+		sfc_log_init(sa, "interrupt on stopped EVQ %u", evq->evq_index);
+	} else {
+		sfc_ev_qpoll(evq);
+
+		if (sfc_ev_qprime(evq) != 0)
+			sfc_err(sa, "cannot prime EVQ %u", evq->evq_index);
+	}
+
+	rte_spinlock_unlock(&sa->mgmt_evq_lock);
+}
+
+static void
+sfc_intr_line_handler(struct rte_intr_handle *intr_handle, void *cb_arg)
+{
+	struct sfc_adapter *sa = (struct sfc_adapter *)cb_arg;
+	efx_nic_t *enp = sa->nic;
+	boolean_t fatal;
+	uint32_t qmask;
+	unsigned int lsc_seq = sa->port.lsc_seq;
+
+	sfc_log_init(sa, "entry");
+
+	if (sa->state != SFC_ADAPTER_STARTED &&
+	    sa->state != SFC_ADAPTER_STARTING &&
+	    sa->state != SFC_ADAPTER_STOPPING) {
+		sfc_log_init(sa,
+			     "interrupt on stopped adapter, don't reenable");
+		goto exit;
+	}
+
+	efx_intr_status_line(enp, &fatal, &qmask);
+	if (fatal) {
+		(void)efx_intr_disable(enp);
+		(void)efx_intr_fatal(enp);
+		sfc_err(sa, "fatal, interrupts disabled");
+		goto exit;
+	}
+
+	if (qmask & (1 << sa->mgmt_evq_index))
+		sfc_intr_handle_mgmt_evq(sa);
+
+	if (rte_intr_enable(intr_handle) != 0)
+		sfc_err(sa, "cannot reenable interrupts");
+
+	sfc_log_init(sa, "done");
+
+exit:
+	if (lsc_seq != sa->port.lsc_seq) {
+		sfc_info(sa, "link status change event: link %s",
+			 sa->eth_dev->data->dev_link.link_status ?
+			 "UP" : "DOWN");
+		_rte_eth_dev_callback_process(sa->eth_dev,
+					      RTE_ETH_EVENT_INTR_LSC, NULL);
+	}
+}
+
+static void
+sfc_intr_message_handler(struct rte_intr_handle *intr_handle, void *cb_arg)
+{
+	struct sfc_adapter *sa = (struct sfc_adapter *)cb_arg;
+	efx_nic_t *enp = sa->nic;
+	boolean_t fatal;
+	unsigned int lsc_seq = sa->port.lsc_seq;
+
+	sfc_log_init(sa, "entry");
+
+	if (sa->state != SFC_ADAPTER_STARTED &&
+	    sa->state != SFC_ADAPTER_STARTING &&
+	    sa->state != SFC_ADAPTER_STOPPING) {
+		sfc_log_init(sa, "adapter not-started, don't reenable");
+		goto exit;
+	}
+
+	efx_intr_status_message(enp, sa->mgmt_evq_index, &fatal);
+	if (fatal) {
+		(void)efx_intr_disable(enp);
+		(void)efx_intr_fatal(enp);
+		sfc_err(sa, "fatal, interrupts disabled");
+		goto exit;
+	}
+
+	sfc_intr_handle_mgmt_evq(sa);
+
+	if (rte_intr_enable(intr_handle) != 0)
+		sfc_err(sa, "cannot reenable interrupts");
+
+	sfc_log_init(sa, "done");
+
+exit:
+	if (lsc_seq != sa->port.lsc_seq) {
+		sfc_info(sa, "link status change event");
+		_rte_eth_dev_callback_process(sa->eth_dev,
+					      RTE_ETH_EVENT_INTR_LSC, NULL);
+	}
+}
 
 int
 sfc_intr_start(struct sfc_adapter *sa)
@@ -54,11 +174,49 @@ sfc_intr_start(struct sfc_adapter *sa)
 
 	intr_handle = &sa->eth_dev->pci_dev->intr_handle;
 
+	if (intr->handler != NULL) {
+		sfc_log_init(sa, "rte_intr_callback_register");
+		rc = rte_intr_callback_register(intr_handle, intr->handler,
+						(void *)sa);
+		if (rc != 0) {
+			sfc_err(sa,
+				"cannot register interrupt handler (rc=%d)",
+				rc);
+			/*
+			 * Convert error code from negative returned by RTE API
+			 * to positive used in the driver.
+			 */
+			rc = -rc;
+			goto fail_rte_intr_cb_reg;
+		}
+
+		sfc_log_init(sa, "rte_intr_enable");
+		rc = rte_intr_enable(intr_handle);
+		if (rc != 0) {
+			sfc_err(sa, "cannot enable interrupts (rc=%d)", rc);
+			/*
+			 * Convert error code from negative returned by RTE API
+			 * to positive used in the driver.
+			 */
+			rc = -rc;
+			goto fail_rte_intr_enable;
+		}
+
+		sfc_log_init(sa, "efx_intr_enable");
+		efx_intr_enable(sa->nic);
+	}
+
 	sfc_log_init(sa, "done type=%u max_intr=%d nb_efd=%u vec=%p",
 		     intr_handle->type, intr_handle->max_intr,
 		     intr_handle->nb_efd, intr_handle->intr_vec);
 	return 0;
 
+fail_rte_intr_enable:
+	rte_intr_callback_unregister(intr_handle, intr->handler, (void *)sa);
+
+fail_rte_intr_cb_reg:
+	efx_intr_fini(sa->nic);
+
 fail_intr_init:
 	sfc_log_init(sa, "failed %d", rc);
 	return rc;
@@ -67,8 +225,29 @@ sfc_intr_start(struct sfc_adapter *sa)
 void
 sfc_intr_stop(struct sfc_adapter *sa)
 {
+	struct sfc_intr *intr = &sa->intr;
+
 	sfc_log_init(sa, "entry");
 
+	if (intr->handler != NULL) {
+		struct rte_intr_handle *intr_handle;
+		int rc;
+
+		efx_intr_disable(sa->nic);
+
+		intr_handle = &sa->eth_dev->pci_dev->intr_handle;
+		if (rte_intr_disable(intr_handle) != 0)
+			sfc_err(sa, "cannot disable interrupts");
+
+		while ((rc = rte_intr_callback_unregister(intr_handle,
+				intr->handler, (void *)sa)) == -EAGAIN)
+			;
+		if (rc != 1)
+			sfc_err(sa,
+				"cannot unregister interrupt handler %d",
+				rc);
+	}
+
 	efx_intr_fini(sa->nic);
 
 	sfc_log_init(sa, "done");
@@ -77,8 +256,33 @@ sfc_intr_stop(struct sfc_adapter *sa)
 int
 sfc_intr_init(struct sfc_adapter *sa)
 {
+	struct sfc_intr *intr = &sa->intr;
+
 	sfc_log_init(sa, "entry");
 
+	intr->handler = NULL;
+	intr->lsc_intr = (sa->eth_dev->data->dev_conf.intr_conf.lsc != 0);
+	if (!intr->lsc_intr) {
+		sfc_info(sa, "LSC tracking using interrupts is disabled");
+		goto done;
+	}
+
+	switch (intr->type) {
+	case EFX_INTR_MESSAGE:
+		intr->handler = sfc_intr_message_handler;
+		break;
+	case EFX_INTR_LINE:
+		intr->handler = sfc_intr_line_handler;
+		break;
+	case EFX_INTR_INVALID:
+		sfc_warn(sa, "interrupts are not supported");
+		break;
+	default:
+		sfc_panic(sa, "unexpected EFX interrupt type %u\n", intr->type);
+		break;
+	}
+
+done:
 	sfc_log_init(sa, "done");
 	return 0;
 }
-- 
2.5.5