IB/hfi1: Add AIP tx traces
authorMike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Mon, 29 Mar 2021 13:54:07 +0000 (09:54 -0400)
committerJason Gunthorpe <jgg@nvidia.com>
Wed, 7 Apr 2021 23:19:00 +0000 (20:19 -0300)
Add traces to allow for debugging issues with AIP tx.

Link: https://lore.kernel.org/r/1617026056-50483-2-git-send-email-dennis.dalessandro@cornelisnetworks.com
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
drivers/infiniband/hw/hfi1/ipoib_tx.c
drivers/infiniband/hw/hfi1/trace_tx.h

index edd4eea..5129dc9 100644 (file)
@@ -15,6 +15,7 @@
 #include "verbs.h"
 #include "trace_ibhdrs.h"
 #include "ipoib.h"
+#include "trace_tx.h"
 
 /* Add a convenience helper */
 #define CIRC_ADD(val, add, size) (((val) + (add)) & ((size) - 1))
@@ -63,12 +64,14 @@ static u64 hfi1_ipoib_used(struct hfi1_ipoib_txq *txq)
 
 static void hfi1_ipoib_stop_txq(struct hfi1_ipoib_txq *txq)
 {
+       trace_hfi1_txq_stop(txq);
        if (atomic_inc_return(&txq->stops) == 1)
                netif_stop_subqueue(txq->priv->netdev, txq->q_idx);
 }
 
 static void hfi1_ipoib_wake_txq(struct hfi1_ipoib_txq *txq)
 {
+       trace_hfi1_txq_wake(txq);
        if (atomic_dec_and_test(&txq->stops))
                netif_wake_subqueue(txq->priv->netdev, txq->q_idx);
 }
@@ -89,8 +92,10 @@ static void hfi1_ipoib_check_queue_depth(struct hfi1_ipoib_txq *txq)
 {
        ++txq->sent_txreqs;
        if (hfi1_ipoib_used(txq) >= hfi1_ipoib_ring_hwat(txq) &&
-           !atomic_xchg(&txq->ring_full, 1))
+           !atomic_xchg(&txq->ring_full, 1)) {
+               trace_hfi1_txq_full(txq);
                hfi1_ipoib_stop_txq(txq);
+       }
 }
 
 static void hfi1_ipoib_check_queue_stopped(struct hfi1_ipoib_txq *txq)
@@ -112,8 +117,10 @@ static void hfi1_ipoib_check_queue_stopped(struct hfi1_ipoib_txq *txq)
         * to protect against ring overflow.
         */
        if (hfi1_ipoib_used(txq) < hfi1_ipoib_ring_lwat(txq) &&
-           atomic_xchg(&txq->ring_full, 0))
+           atomic_xchg(&txq->ring_full, 0)) {
+               trace_hfi1_txq_xmit_unstopped(txq);
                hfi1_ipoib_wake_txq(txq);
+       }
 }
 
 static void hfi1_ipoib_free_tx(struct ipoib_txreq *tx, int budget)
@@ -405,6 +412,7 @@ static struct ipoib_txreq *hfi1_ipoib_send_dma_common(struct net_device *dev,
                                sdma_select_engine_sc(priv->dd,
                                                      txp->flow.tx_queue,
                                                      txp->flow.sc5);
+                       trace_hfi1_flow_switch(txp->txq);
                }
 
                return tx;
@@ -525,6 +533,7 @@ static int hfi1_ipoib_send_dma_list(struct net_device *dev,
        if (txq->flow.as_int != txp->flow.as_int) {
                int ret;
 
+               trace_hfi1_flow_flush(txq);
                ret = hfi1_ipoib_flush_tx_list(dev, txq);
                if (unlikely(ret)) {
                        if (ret == -EBUSY)
@@ -635,8 +644,10 @@ static int hfi1_ipoib_sdma_sleep(struct sdma_engine *sde,
                        /* came from non-list submit */
                        list_add_tail(&txreq->list, &txq->tx_list);
                if (list_empty(&txq->wait.list)) {
-                       if (!atomic_xchg(&txq->no_desc, 1))
+                       if (!atomic_xchg(&txq->no_desc, 1)) {
+                               trace_hfi1_txq_queued(txq);
                                hfi1_ipoib_stop_txq(txq);
+                       }
                        iowait_queue(pkts_sent, wait->iow, &sde->dmawait);
                }
 
@@ -659,6 +670,7 @@ static void hfi1_ipoib_sdma_wakeup(struct iowait *wait, int reason)
        struct hfi1_ipoib_txq *txq =
                container_of(wait, struct hfi1_ipoib_txq, wait);
 
+       trace_hfi1_txq_wakeup(txq);
        if (likely(txq->priv->netdev->reg_state == NETREG_REGISTERED))
                iowait_schedule(wait, system_highpri_wq, WORK_CPU_UNBOUND);
 }
index 769e5e4..8476541 100644 (file)
@@ -53,6 +53,7 @@
 #include "hfi.h"
 #include "mad.h"
 #include "sdma.h"
+#include "ipoib.h"
 
 const char *parse_sdma_flags(struct trace_seq *p, u64 desc0, u64 desc1);
 
@@ -858,6 +859,109 @@ DEFINE_EVENT(
        TP_ARGS(qp, flag)
 );
 
+DECLARE_EVENT_CLASS(/* AIP  */
+       hfi1_ipoib_txq_template,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq),
+       TP_STRUCT__entry(/* entry */
+               DD_DEV_ENTRY(txq->priv->dd)
+               __field(struct hfi1_ipoib_txq *, txq)
+               __field(struct sdma_engine *, sde)
+               __field(ulong, head)
+               __field(ulong, tail)
+               __field(uint, used)
+               __field(uint, flow)
+               __field(int, stops)
+               __field(int, no_desc)
+               __field(u8, idx)
+               __field(u8, stopped)
+       ),
+       TP_fast_assign(/* assign */
+               DD_DEV_ASSIGN(txq->priv->dd)
+               __entry->txq = txq;
+               __entry->sde = txq->sde;
+               __entry->head = txq->tx_ring.head;
+               __entry->tail = txq->tx_ring.tail;
+               __entry->idx = txq->q_idx;
+               __entry->used =
+                       txq->sent_txreqs -
+                       atomic64_read(&txq->complete_txreqs);
+               __entry->flow = txq->flow.as_int;
+               __entry->stops = atomic_read(&txq->stops);
+               __entry->no_desc = atomic_read(&txq->no_desc);
+               __entry->stopped =
+                __netif_subqueue_stopped(txq->priv->netdev, txq->q_idx);
+       ),
+       TP_printk(/* print  */
+               "[%s] txq %llx idx %u sde %llx head %lx tail %lx flow %x used %u stops %d no_desc %d stopped %u",
+               __get_str(dev),
+               (unsigned long long)__entry->txq,
+               __entry->idx,
+               (unsigned long long)__entry->sde,
+               __entry->head,
+               __entry->tail,
+               __entry->flow,
+               __entry->used,
+               __entry->stops,
+               __entry->no_desc,
+               __entry->stopped
+       )
+);
+
+DEFINE_EVENT(/* queue stop */
+       hfi1_ipoib_txq_template, hfi1_txq_stop,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* queue wake */
+       hfi1_ipoib_txq_template, hfi1_txq_wake,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* flow flush */
+       hfi1_ipoib_txq_template, hfi1_flow_flush,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* flow switch */
+       hfi1_ipoib_txq_template, hfi1_flow_switch,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* wakeup */
+       hfi1_ipoib_txq_template, hfi1_txq_wakeup,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* full */
+       hfi1_ipoib_txq_template, hfi1_txq_full,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* queued */
+       hfi1_ipoib_txq_template, hfi1_txq_queued,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* xmit_stopped */
+       hfi1_ipoib_txq_template, hfi1_txq_xmit_stopped,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
+DEFINE_EVENT(/* xmit_unstopped */
+       hfi1_ipoib_txq_template, hfi1_txq_xmit_unstopped,
+       TP_PROTO(struct hfi1_ipoib_txq *txq),
+       TP_ARGS(txq)
+);
+
 #endif /* __HFI1_TRACE_TX_H */
 
 #undef TRACE_INCLUDE_PATH