[dpdk-dev] [PATCH 0/2] Event queue exception handling in Solarflare PMD

[dpdk-dev] [PATCH 0/2] Event queue exception handling in Solarflare PMD

[Andrew Rybchenko]

The patch series adds event queue exception handling.

It is not required when everything goes fine, but has chances to
recover if something goes wrong in either driver or HW.

The patch series should be applied after
[PATCH v2 00/55] Solarflare libefx-based PMD
(Message-ID: 1480436367-20749-1-git-send-email-arybchenko@solarflare.com)


Andrew Rybchenko (1):
  net/sfc: restart RxQ in case of exception on its event queue

Ivan Malov (1):
  net/sfc: restart TxQ in case of exception on its event queue

 drivers/net/sfc/sfc.h    |  6 ++++++
 drivers/net/sfc/sfc_ev.c | 40 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)

-- 
2.5.5

[dpdk-dev] [PATCH 1/2] net/sfc: restart RxQ in case of exception on its event queue

[Andrew Rybchenko]

Examples of recoverable exceptions are:
 - unexpected Rx event (Rx scatter abort with non-zero size,
   too big Rx descriptors batch completed)
 - Rx error due to invalid Rx descriptors push
 - Rx error due to Rx descriptor read error (e.g. unmapped Rx ring
   and denied by IOMMU)

Reviewed-by: Andrew Lee <alee@solarflare.com>
Reviewed-by: Robert Stonehouse <rstonehouse@solarflare.com>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
---
 drivers/net/sfc/sfc.h    |  6 ++++++
 drivers/net/sfc/sfc_ev.c | 26 ++++++++++++++++++++++++++
 2 files changed, 32 insertions(+)

diff --git a/drivers/net/sfc/sfc.h b/drivers/net/sfc/sfc.h
index 29d3a6b..995dfe6 100644
--- a/drivers/net/sfc/sfc.h
+++ b/drivers/net/sfc/sfc.h
@@ -187,6 +187,12 @@ sfc_adapter_lock(struct sfc_adapter *sa)
 	rte_spinlock_lock(&sa->lock);
 }
 
+static inline int
+sfc_adapter_trylock(struct sfc_adapter *sa)
+{
+	return rte_spinlock_trylock(&sa->lock);
+}
+
 static inline void
 sfc_adapter_unlock(struct sfc_adapter *sa)
 {
diff --git a/drivers/net/sfc/sfc_ev.c b/drivers/net/sfc/sfc_ev.c
index 96b95cc..36aede8 100644
--- a/drivers/net/sfc/sfc_ev.c
+++ b/drivers/net/sfc/sfc_ev.c
@@ -30,6 +30,7 @@
 #include <rte_debug.h>
 #include <rte_cycles.h>
 #include <rte_alarm.h>
+#include <rte_branch_prediction.h>
 
 #include "efx.h"
 
@@ -320,6 +321,31 @@ sfc_ev_qpoll(struct sfc_evq *evq)
 
 	efx_ev_qpoll(evq->common, &evq->read_ptr, &sfc_ev_callbacks, evq);
 
+	if (unlikely(evq->exception) && sfc_adapter_trylock(evq->sa)) {
+		struct sfc_adapter *sa = evq->sa;
+		int rc;
+
+		if ((evq->rxq != NULL) && (evq->rxq->state & SFC_RXQ_RUNNING)) {
+			unsigned int rxq_sw_index = sfc_rxq_sw_index(evq->rxq);
+
+			sfc_warn(sa,
+				 "restart RxQ %u because of exception on its EvQ %u",
+				 rxq_sw_index, evq->evq_index);
+
+			sfc_rx_qstop(sa, rxq_sw_index);
+			rc = sfc_rx_qstart(sa, rxq_sw_index);
+			if (rc != 0)
+				sfc_err(sa, "cannot restart RxQ %u",
+					rxq_sw_index);
+		}
+
+		if (evq->exception)
+			sfc_panic(sa, "unrecoverable exception on EvQ %u",
+				  evq->evq_index);
+
+		sfc_adapter_unlock(sa);
+	}
+
 	/* Poll-mode driver does not re-prime the event queue for interrupts */
 }
 
-- 
2.5.5

[dpdk-dev] [PATCH 2/2] net/sfc: restart TxQ in case of exception on its event queue

[Andrew Rybchenko]

From: Ivan Malov <ivan.malov@oktetlabs.ru>

Examples of recoverable exceptions because of Tx error are:
 - Tx descriptor PCI read error
 - invalid Tx option descriptor
 - Tx option descriptor not supported by the firmware variant
 - unexpected Tx option descriptor (e.g. missing FATSO2A before
   FATSO2B, missing FATSO2B after FATSO2A)
 - incomplete packet push (CONT bit set in the latest pushed DMA
   descriptor)

Reviewed-by: Andrew Lee <alee@solarflare.com>
Reviewed-by: Robert Stonehouse <rstonehouse@solarflare.com>
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
---
 drivers/net/sfc/sfc_ev.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/net/sfc/sfc_ev.c b/drivers/net/sfc/sfc_ev.c
index 36aede8..af3c7b2 100644
--- a/drivers/net/sfc/sfc_ev.c
+++ b/drivers/net/sfc/sfc_ev.c
@@ -339,6 +339,20 @@ sfc_ev_qpoll(struct sfc_evq *evq)
 					rxq_sw_index);
 		}
 
+		if (evq->txq != NULL) {
+			unsigned int txq_sw_index = sfc_txq_sw_index(evq->txq);
+
+			sfc_warn(sa,
+				 "restart TxQ %u because of exception on its EvQ %u",
+				 txq_sw_index, evq->evq_index);
+
+			sfc_tx_qstop(sa, txq_sw_index);
+			rc = sfc_tx_qstart(sa, txq_sw_index);
+			if (rc != 0)
+				sfc_err(sa, "cannot restart TxQ %u",
+					txq_sw_index);
+		}
+
 		if (evq->exception)
 			sfc_panic(sa, "unrecoverable exception on EvQ %u",
 				  evq->evq_index);
-- 
2.5.5

Re: [dpdk-dev] [PATCH 0/2] Event queue exception handling in Solarflare PMD

[Ferruh Yigit]

On 12/2/2016 7:41 AM, Andrew Rybchenko wrote:
> The patch series adds event queue exception handling.
> 
> It is not required when everything goes fine, but has chances to
> recover if something goes wrong in either driver or HW.
> 
> The patch series should be applied after
> [PATCH v2 00/55] Solarflare libefx-based PMD
> (Message-ID: 1480436367-20749-1-git-send-email-arybchenko@solarflare.com)
> 
> 
> Andrew Rybchenko (1):
>   net/sfc: restart RxQ in case of exception on its event queue
> 
> Ivan Malov (1):
>   net/sfc: restart TxQ in case of exception on its event queue
> 
>  drivers/net/sfc/sfc.h    |  6 ++++++
>  drivers/net/sfc/sfc_ev.c | 40 ++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 46 insertions(+)
> 

Series applied to dpdk-next-net/master, thanks.