IB/hfi1, qib, rdmavt: Another round of patches for 4.11

index 121a4c9..44322c6 100644
@@ -1045,6 +1045,7 @@ static int wait_logical_linkstate(struct hfi1_pportdata *ppd, u32 state,
 static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
 			   unsigned int *np);
 static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd);
+static int wait_link_transfer_active(struct hfi1_devdata *dd, int wait_ms);
 
 /*
  * Error interrupt table entry.  This is used as input to the interrupt

@@ -8891,8 +8892,6 @@ int send_idle_sma(struct hfi1_devdata *dd, u64 message)
  */
 static int do_quick_linkup(struct hfi1_devdata *dd)
 {
-	u64 reg;
-	unsigned long timeout;
 	int ret;
 
 	lcb_shutdown(dd, 0);

@@ -8915,19 +8914,9 @@ static int do_quick_linkup(struct hfi1_devdata *dd)
 		write_csr(dd, DC_LCB_CFG_RUN,
 			  1ull << DC_LCB_CFG_RUN_EN_SHIFT);
 
-		/* watch LCB_STS_LINK_TRANSFER_ACTIVE */
-		timeout = jiffies + msecs_to_jiffies(10);
-		while (1) {
-			reg = read_csr(dd, DC_LCB_STS_LINK_TRANSFER_ACTIVE);
-			if (reg)
-				break;
-			if (time_after(jiffies, timeout)) {
-				dd_dev_err(dd,
-					   "timeout waiting for LINK_TRANSFER_ACTIVE\n");
-				return -ETIMEDOUT;
-			}
-			udelay(2);
-		}
+		ret = wait_link_transfer_active(dd, 10);
+		if (ret)
+			return ret;
 
 		write_csr(dd, DC_LCB_CFG_ALLOW_LINK_UP,
 			  1ull << DC_LCB_CFG_ALLOW_LINK_UP_VAL_SHIFT);

@@ -10082,6 +10071,64 @@ static void check_lni_states(struct hfi1_pportdata *ppd)
 	decode_state_complete(ppd, last_remote_state, "received");
 }
 
+/* wait for wait_ms for LINK_TRANSFER_ACTIVE to go to 1 */
+static int wait_link_transfer_active(struct hfi1_devdata *dd, int wait_ms)
+{
+	u64 reg;
+	unsigned long timeout;
+
+	/* watch LCB_STS_LINK_TRANSFER_ACTIVE */
+	timeout = jiffies + msecs_to_jiffies(wait_ms);
+	while (1) {
+		reg = read_csr(dd, DC_LCB_STS_LINK_TRANSFER_ACTIVE);
+		if (reg)
+			break;
+		if (time_after(jiffies, timeout)) {
+			dd_dev_err(dd,
+				   "timeout waiting for LINK_TRANSFER_ACTIVE\n");
+			return -ETIMEDOUT;
+		}
+		udelay(2);
+	}
+	return 0;
+}
+
+/* called when the logical link state is not down as it should be */
+static void force_logical_link_state_down(struct hfi1_pportdata *ppd)
+{
+	struct hfi1_devdata *dd = ppd->dd;
+
+	/*
+	 * Bring link up in LCB loopback
+	 */
+	write_csr(dd, DC_LCB_CFG_TX_FIFOS_RESET, 1);
+	write_csr(dd, DC_LCB_CFG_IGNORE_LOST_RCLK,
+		  DC_LCB_CFG_IGNORE_LOST_RCLK_EN_SMASK);
+
+	write_csr(dd, DC_LCB_CFG_LANE_WIDTH, 0);
+	write_csr(dd, DC_LCB_CFG_REINIT_AS_SLAVE, 0);
+	write_csr(dd, DC_LCB_CFG_CNT_FOR_SKIP_STALL, 0x110);
+	write_csr(dd, DC_LCB_CFG_LOOPBACK, 0x2);
+
+	write_csr(dd, DC_LCB_CFG_TX_FIFOS_RESET, 0);
+	(void)read_csr(dd, DC_LCB_CFG_TX_FIFOS_RESET);
+	udelay(3);
+	write_csr(dd, DC_LCB_CFG_ALLOW_LINK_UP, 1);
+	write_csr(dd, DC_LCB_CFG_RUN, 1ull << DC_LCB_CFG_RUN_EN_SHIFT);
+
+	wait_link_transfer_active(dd, 100);
+
+	/*
+	 * Bring the link down again.
+	 */
+	write_csr(dd, DC_LCB_CFG_TX_FIFOS_RESET, 1);
+	write_csr(dd, DC_LCB_CFG_ALLOW_LINK_UP, 0);
+	write_csr(dd, DC_LCB_CFG_IGNORE_LOST_RCLK, 0);
+
+	/* call again to adjust ppd->statusp, if needed */
+	get_logical_state(ppd);
+}
+
 /*
  * Helper for set_link_state().  Do not call except from that routine.
  * Expects ppd->hls_mutex to be held.

@@ -10135,15 +10182,18 @@ static int goto_offline(struct hfi1_pportdata *ppd, u8 rem_reason)
 			return ret;
 	}
 
-	/* make sure the logical state is also down */
-	wait_logical_linkstate(ppd, IB_PORT_DOWN, 1000);
-
 	/*
 	 * Now in charge of LCB - must be after the physical state is
 	 * offline.quiet and before host_link_state is changed.
 	 */
 	set_host_lcb_access(dd);
 	write_csr(dd, DC_LCB_ERR_EN, ~0ull); /* watch LCB errors */
+
+	/* make sure the logical state is also down */
+	ret = wait_logical_linkstate(ppd, IB_PORT_DOWN, 1000);
+	if (ret)
+		force_logical_link_state_down(ppd);
+
 	ppd->host_link_state = HLS_LINK_COOLDOWN; /* LCB access allowed */
 
 	if (ppd->port_type == PORT_TYPE_QSFP &&

index e6811c4..060e374 100644
@@ -1615,9 +1615,10 @@ static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *pq,
 {
 	hfi1_cdbg(SDMA, "[%u:%u:%u:%u] Setting completion status %u %d",
 		  pq->dd->unit, pq->ctxt, pq->subctxt, idx, state, ret);
-	cq->comps[idx].status = state;
 	if (state == ERROR)
 		cq->comps[idx].errcode = -ret;
+	smp_wmb(); /* make sure errcode is visible first */
+	cq->comps[idx].status = state;
 	trace_hfi1_sdma_user_completion(pq->dd, pq->ctxt, pq->subctxt,
 					idx, state, ret);
 }

index 44322c6..8b8840a 100644
@@ -8345,6 +8345,52 @@ static int read_lcb_via_8051(struct hfi1_devdata *dd, u32 addr, u64 *data)
 }
 
 /*
+ * Provide a cache for some of the LCB registers in case the LCB is
+ * unavailable.
+ * (The LCB is unavailable in certain link states, for example.)
+ */
+struct lcb_datum {
+	u32 off;
+	u64 val;
+};
+
+static struct lcb_datum lcb_cache[] = {
+	{ DC_LCB_ERR_INFO_RX_REPLAY_CNT, 0},
+	{ DC_LCB_ERR_INFO_SEQ_CRC_CNT, 0 },
+	{ DC_LCB_ERR_INFO_REINIT_FROM_PEER_CNT, 0 },
+};
+
+static void update_lcb_cache(struct hfi1_devdata *dd)
+{
+	int i;
+	int ret;
+	u64 val;
+
+	for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
+		ret = read_lcb_csr(dd, lcb_cache[i].off, &val);
+
+		/* Update if we get good data */
+		if (likely(ret != -EBUSY))
+			lcb_cache[i].val = val;
+	}
+}
+
+static int read_lcb_cache(u32 off, u64 *val)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
+		if (lcb_cache[i].off == off) {
+			*val = lcb_cache[i].val;
+			return 0;
+		}
+	}
+
+	pr_warn("%s bad offset 0x%x\n", __func__, off);
+	return -1;
+}
+
+/*
  * Read an LCB CSR.  Access may not be in host control, so check.
  * Return 0 on success, -EBUSY on failure.
  */

@@ -8355,9 +8401,13 @@ int read_lcb_csr(struct hfi1_devdata *dd, u32 addr, u64 *data)
 	/* if up, go through the 8051 for the value */
 	if (ppd->host_link_state & HLS_UP)
 		return read_lcb_via_8051(dd, addr, data);
-	/* if going up or down, no access */
-	if (ppd->host_link_state & (HLS_GOING_UP | HLS_GOING_OFFLINE))
-		return -EBUSY;
+	/* if going up or down, check the cache, otherwise, no access */
+	if (ppd->host_link_state & (HLS_GOING_UP | HLS_GOING_OFFLINE)) {
+		if (read_lcb_cache(addr, data))
+			return -EBUSY;
+		return 0;
+	}
+
 	/* otherwise, host has access */
 	*data = read_csr(dd, addr);
 	return 0;

@@ -10145,6 +10195,8 @@ static int goto_offline(struct hfi1_pportdata *ppd, u8 rem_reason)
 	int do_transition;
 	int do_wait;
 
+	update_lcb_cache(dd);
+
 	previous_state = ppd->host_link_state;
 	ppd->host_link_state = HLS_GOING_OFFLINE;
 	pstate = read_physical_state(dd);

index 0808e3c..b69ab47 100644
@@ -1167,7 +1167,7 @@ struct hfi1_devdata {
 	bool eprom_available;	/* true if EPROM is available for this device */
 	bool aspm_supported;	/* Does HW support ASPM */
 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
-	struct rhashtable sdma_rht;
+	struct rhashtable *sdma_rht;
 
 	struct kobject kobj;
 };

index 1d81cac..9bee28d 100644
@@ -868,7 +868,7 @@ struct sdma_engine *sdma_select_user_engine(struct hfi1_devdata *dd,
 
 	cpu_id = smp_processor_id();
 	rcu_read_lock();
-	rht_node = rhashtable_lookup_fast(&dd->sdma_rht, &cpu_id,
+	rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu_id,
 					  sdma_rht_params);
 
 	if (rht_node && rht_node->map[vl]) {

@@ -962,7 +962,7 @@ ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
 			continue;
 		}
 
-		rht_node = rhashtable_lookup_fast(&dd->sdma_rht, &cpu,
+		rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu,
 						  sdma_rht_params);
 		if (!rht_node) {
 			rht_node = kzalloc(sizeof(*rht_node), GFP_KERNEL);

@@ -982,7 +982,7 @@ ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
 			rht_node->map[vl]->ctr = 1;
 			rht_node->map[vl]->sde[0] = sde;
 
-			ret = rhashtable_insert_fast(&dd->sdma_rht,
+			ret = rhashtable_insert_fast(dd->sdma_rht,
 						     &rht_node->node,
 						     sdma_rht_params);
 			if (ret) {

@@ -1025,7 +1025,7 @@ ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
 		if (cpumask_test_cpu(cpu, mask))
 			continue;
 
-		rht_node = rhashtable_lookup_fast(&dd->sdma_rht, &cpu,
+		rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu,
 						  sdma_rht_params);
 		if (rht_node) {
 			bool empty = true;

@@ -1049,7 +1049,7 @@ ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
 			}
 
 			if (empty) {
-				ret = rhashtable_remove_fast(&dd->sdma_rht,
+				ret = rhashtable_remove_fast(dd->sdma_rht,
 							     &rht_node->node,
 							     sdma_rht_params);
 				WARN_ON(ret);

@@ -1108,7 +1108,7 @@ void sdma_seqfile_dump_cpu_list(struct seq_file *s,
 	struct sdma_rht_node *rht_node;
 	int i, j;
 
-	rht_node = rhashtable_lookup_fast(&dd->sdma_rht, &cpuid,
+	rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpuid,
 					  sdma_rht_params);
 	if (!rht_node)
 		return;

@@ -1322,6 +1322,12 @@ static void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
 	synchronize_rcu();
 	kfree(dd->per_sdma);
 	dd->per_sdma = NULL;
+
+	if (dd->sdma_rht) {
+		rhashtable_free_and_destroy(dd->sdma_rht, sdma_rht_free, NULL);
+		kfree(dd->sdma_rht);
+		dd->sdma_rht = NULL;
+	}
 }
 
 /**

@@ -1341,12 +1347,14 @@ int sdma_init(struct hfi1_devdata *dd, u8 port)
 {
 	unsigned this_idx;
 	struct sdma_engine *sde;
+	struct rhashtable *tmp_sdma_rht;
 	u16 descq_cnt;
 	void *curr_head;
 	struct hfi1_pportdata *ppd = dd->pport + port;
 	u32 per_sdma_credits;
 	uint idle_cnt = sdma_idle_cnt;
 	size_t num_engines = dd->chip_sdma_engines;
+	int ret = -ENOMEM;
 
 	if (!HFI1_CAP_IS_KSET(SDMA)) {
 		HFI1_CAP_CLEAR(SDMA_AHG);

@@ -1378,7 +1386,7 @@ int sdma_init(struct hfi1_devdata *dd, u8 port)
 	/* alloc memory for array of send engines */
 	dd->per_sdma = kcalloc(num_engines, sizeof(*dd->per_sdma), GFP_KERNEL);
 	if (!dd->per_sdma)
-		return -ENOMEM;
+		return ret;
 
 	idle_cnt = ns_to_cclock(dd, idle_cnt);
 	if (!sdma_desct_intr)

@@ -1507,18 +1515,27 @@ int sdma_init(struct hfi1_devdata *dd, u8 port)
 	dd->flags |= HFI1_HAS_SEND_DMA;
 	dd->flags |= idle_cnt ? HFI1_HAS_SDMA_TIMEOUT : 0;
 	dd->num_sdma = num_engines;
-	if (sdma_map_init(dd, port, ppd->vls_operational, NULL))
+	ret = sdma_map_init(dd, port, ppd->vls_operational, NULL);
+	if (ret < 0)
+		goto bail;
+
+	tmp_sdma_rht = kzalloc(sizeof(*tmp_sdma_rht), GFP_KERNEL);
+	if (!tmp_sdma_rht) {
+		ret = -ENOMEM;
 		goto bail;
+	}
 
-	if (rhashtable_init(&dd->sdma_rht, &sdma_rht_params))
+	ret = rhashtable_init(tmp_sdma_rht, &sdma_rht_params);
+	if (ret < 0)
 		goto bail;
+	dd->sdma_rht = tmp_sdma_rht;
 
 	dd_dev_info(dd, "SDMA num_sdma: %u\n", dd->num_sdma);
 	return 0;
 
 bail:
 	sdma_clean(dd, num_engines);
-	return -ENOMEM;
+	return ret;
 }
 
 /**

@@ -1604,7 +1621,6 @@ void sdma_exit(struct hfi1_devdata *dd)
 		sdma_finalput(&sde->state);
 	}
 	sdma_clean(dd, dd->num_sdma);
-	rhashtable_free_and_destroy(&dd->sdma_rht, sdma_rht_free, NULL);
 }
 
 /*

index 7382be1..4649530 100644
@@ -1034,7 +1034,10 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr)
 		/* see post_send() */
 		barrier();
 		rvt_put_swqe(wqe);
-		rvt_qp_swqe_complete(qp, wqe, IB_WC_SUCCESS);
+		rvt_qp_swqe_complete(qp,
+				     wqe,
+				     ib_hfi1_wc_opcode[wqe->wr.opcode],
+				     IB_WC_SUCCESS);
 	}
 	/*
 	 * If we were waiting for sends to complete before re-sending,

@@ -1081,7 +1084,10 @@ static inline void update_last_psn(struct rvt_qp *qp, u32 psn)
 		qp->s_last = s_last;
 		/* see post_send() */
 		barrier();
-		rvt_qp_swqe_complete(qp, wqe, IB_WC_SUCCESS);
+		rvt_qp_swqe_complete(qp,
+				     wqe,
+				     ib_hfi1_wc_opcode[wqe->wr.opcode],
+				     IB_WC_SUCCESS);
 	} else {
 		struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
 

index aa15bcb..d2eb793 100644
@@ -920,7 +920,10 @@ void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
 	    qp->ibqp.qp_type == IB_QPT_GSI)
 		atomic_dec(&ibah_to_rvtah(wqe->ud_wr.ah)->refcount);
 
-	rvt_qp_swqe_complete(qp, wqe, status);
+	rvt_qp_swqe_complete(qp,
+			     wqe,
+			     ib_hfi1_wc_opcode[wqe->wr.opcode],
+			     status);
 
 	if (qp->s_acked == old_last)
 		qp->s_acked = last;

index 222315f..815cb44 100644
@@ -297,6 +297,22 @@ static inline bool wss_exceeds_threshold(void)
 }
 
 /*
+ * Translate ib_wr_opcode into ib_wc_opcode.
+ */
+const enum ib_wc_opcode ib_hfi1_wc_opcode[] = {
+	[IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
+	[IB_WR_SEND] = IB_WC_SEND,
+	[IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
+	[IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
+	[IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD,
+	[IB_WR_SEND_WITH_INV] = IB_WC_SEND,
+	[IB_WR_LOCAL_INV] = IB_WC_LOCAL_INV,
+	[IB_WR_REG_MR] = IB_WC_REG_MR
+};
+
+/*
  * Length of header by opcode, 0 --> not supported
  */
 const u8 hdr_len_by_opcode[256] = {

index 12658e3..0234987 100644
@@ -938,7 +938,10 @@ void qib_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr)
 		/* see post_send() */
 		barrier();
 		rvt_put_swqe(wqe);
-		rvt_qp_swqe_complete(qp, wqe, IB_WC_SUCCESS);
+		rvt_qp_swqe_complete(qp,
+				     wqe,
+				     ib_qib_wc_opcode[wqe->wr.opcode],
+				     IB_WC_SUCCESS);
 	}
 	/*
 	 * If we were waiting for sends to complete before resending,

@@ -983,7 +986,10 @@ static inline void update_last_psn(struct rvt_qp *qp, u32 psn)
 		qp->s_last = s_last;
 		/* see post_send() */
 		barrier();
-		rvt_qp_swqe_complete(qp, wqe, IB_WC_SUCCESS);
+		rvt_qp_swqe_complete(qp,
+				     wqe,
+				     ib_qib_wc_opcode[wqe->wr.opcode],
+				     IB_WC_SUCCESS);
 	} else
 		this_cpu_inc(*ibp->rvp.rc_delayed_comp);
 

index 17655cc..6e1adf7 100644
@@ -769,7 +769,10 @@ void qib_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
 	    qp->ibqp.qp_type == IB_QPT_GSI)
 		atomic_dec(&ibah_to_rvtah(wqe->ud_wr.ah)->refcount);
 
-	rvt_qp_swqe_complete(qp, wqe, status);
+	rvt_qp_swqe_complete(qp,
+			     wqe,
+			     ib_qib_wc_opcode[wqe->wr.opcode],
+			     status);
 
 	if (qp->s_acked == old_last)
 		qp->s_acked = last;

index 83f8b5f..e120efe 100644
@@ -114,6 +114,19 @@
 MODULE_PARM_DESC(disable_sma, "Disable the SMA");
 
 /*
+ * Translate ib_wr_opcode into ib_wc_opcode.
+ */
+const enum ib_wc_opcode ib_qib_wc_opcode[] = {
+	[IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
+	[IB_WR_SEND] = IB_WC_SEND,
+	[IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
+	[IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
+	[IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
+};
+
+/*
  * System image GUID.
  */
 __be64 ib_qib_sys_image_guid;

index f5ad8d4..28fb724 100644
@@ -117,23 +117,6 @@
 };
 EXPORT_SYMBOL(ib_rvt_state_ops);
 
-/*
- * Translate ib_wr_opcode into ib_wc_opcode.
- */
-const enum ib_wc_opcode ib_rvt_wc_opcode[] = {
-	[IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
-	[IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
-	[IB_WR_SEND] = IB_WC_SEND,
-	[IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
-	[IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
-	[IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
-	[IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD,
-	[IB_WR_SEND_WITH_INV] = IB_WC_SEND,
-	[IB_WR_LOCAL_INV] = IB_WC_LOCAL_INV,
-	[IB_WR_REG_MR] = IB_WC_REG_MR
-};
-EXPORT_SYMBOL(ib_rvt_wc_opcode);
-
 static void get_map_page(struct rvt_qpn_table *qpt,
 			 struct rvt_qpn_map *map,
 			 gfp_t gfp)

index f381639..3cdd9e2 100644
@@ -574,6 +574,7 @@ static inline void rvt_qp_wqe_unreserve(
 static inline void rvt_qp_swqe_complete(
 	struct rvt_qp *qp,
 	struct rvt_swqe *wqe,
+	enum ib_wc_opcode opcode,
 	enum ib_wc_status status)
 {
 	if (unlikely(wqe->wr.send_flags & RVT_SEND_RESERVE_USED))

@@ -586,7 +587,7 @@ static inline void rvt_qp_swqe_complete(
 		memset(&wc, 0, sizeof(wc));
 		wc.wr_id = wqe->wr.wr_id;
 		wc.status = status;
-		wc.opcode = ib_rvt_wc_opcode[wqe->wr.opcode];
+		wc.opcode = opcode;
 		wc.qp = &qp->ibqp;
 		wc.byte_len = wqe->length;
 		rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.send_cq), &wc,

index 28fb724..3c55a8b 100644
@@ -1772,11 +1772,11 @@ static int rvt_post_one_wr(struct rvt_qp *qp,
 					0);
 		qp->s_next_psn = wqe->lpsn + 1;
 	}
-	trace_rvt_post_one_wr(qp, wqe);
 	if (unlikely(reserved_op))
 		rvt_qp_wqe_reserve(qp, wqe);
 	else
 		qp->s_avail--;
+	trace_rvt_post_one_wr(qp, wqe);
 	smp_wmb(); /* see request builders */
 	qp->s_head = next;
 

index 0e03173..a613a22 100644
@@ -71,10 +71,20 @@
 	wr_opcode_name(RDMA_READ_WITH_INV),                \
 	wr_opcode_name(LOCAL_INV),                         \
 	wr_opcode_name(MASKED_ATOMIC_CMP_AND_SWP),         \
-	wr_opcode_name(MASKED_ATOMIC_FETCH_AND_ADD))
+	wr_opcode_name(MASKED_ATOMIC_FETCH_AND_ADD),       \
+	wr_opcode_name(RESERVED1),                         \
+	wr_opcode_name(RESERVED2),                         \
+	wr_opcode_name(RESERVED3),                         \
+	wr_opcode_name(RESERVED4),                         \
+	wr_opcode_name(RESERVED5),                         \
+	wr_opcode_name(RESERVED6),                         \
+	wr_opcode_name(RESERVED7),                         \
+	wr_opcode_name(RESERVED8),                         \
+	wr_opcode_name(RESERVED9),                         \
+	wr_opcode_name(RESERVED10))
 
 #define POS_PRN \
-"[%s] wr_id %llx qpn %x psn 0x%x lpsn 0x%x length %u opcode 0x%.2x,%s size %u avail %u head %u last %u"
+"[%s] wqe %p wr_id %llx send_flags %x qpn %x qpt %u psn %x lpsn %x ssn %x length %u opcode 0x%.2x,%s size %u avail %u head %u last %u pid %u num_sge %u"
 
 TRACE_EVENT(
 	rvt_post_one_wr,

@@ -83,7 +93,9 @@
 	TP_STRUCT__entry(
 		RDI_DEV_ENTRY(ib_to_rvt(qp->ibqp.device))
 		__field(u64, wr_id)
+		__field(struct rvt_swqe *, wqe)
 		__field(u32, qpn)
+		__field(u32, qpt)
 		__field(u32, psn)
 		__field(u32, lpsn)
 		__field(u32, length)

@@ -92,11 +104,17 @@
 		__field(u32, avail)
 		__field(u32, head)
 		__field(u32, last)
+		__field(u32, ssn)
+		__field(int, send_flags)
+		__field(pid_t, pid)
+		__field(int, num_sge)
 	),
 	TP_fast_assign(
 		RDI_DEV_ASSIGN(ib_to_rvt(qp->ibqp.device))
+		__entry->wqe = wqe;
 		__entry->wr_id = wqe->wr.wr_id;
 		__entry->qpn = qp->ibqp.qp_num;
+		__entry->qpt = qp->ibqp.qp_type;
 		__entry->psn = wqe->psn;
 		__entry->lpsn = wqe->lpsn;
 		__entry->length = wqe->length;

@@ -105,20 +123,30 @@
 		__entry->avail = qp->s_avail;
 		__entry->head = qp->s_head;
 		__entry->last = qp->s_last;
+		__entry->pid = qp->pid;
+		__entry->ssn = wqe->ssn;
+		__entry->send_flags = wqe->wr.send_flags;
+		__entry->num_sge = wqe->wr.num_sge;
 	),
 	TP_printk(
 		POS_PRN,
 		__get_str(dev),
+		__entry->wqe,
 		__entry->wr_id,
+		__entry->send_flags,
 		__entry->qpn,
+		__entry->qpt,
 		__entry->psn,
 		__entry->lpsn,
+		__entry->ssn,
 		__entry->length,
 		__entry->opcode, show_wr_opcode(__entry->opcode),
 		__entry->size,
 		__entry->avail,
 		__entry->head,
-		__entry->last
+		__entry->last,
+		__entry->pid,
+		__entry->num_sge
 	)
 );
 

index 7aa7a4e..0ae2ff8 100644
@@ -50,6 +50,7 @@
 #include <linux/kthread.h>
 #include "cq.h"
 #include "vt.h"
+#include "trace.h"
 
 /**
  * rvt_cq_enter - add a new entry to the completion queue

@@ -93,6 +94,7 @@ void rvt_cq_enter(struct rvt_cq *cq, struct ib_wc *entry, bool solicited)
 		}
 		return;
 	}
+	trace_rvt_cq_enter(cq, entry, head);
 	if (cq->ip) {
 		wc->uqueue[head].wr_id = entry->wr_id;
 		wc->uqueue[head].status = entry->status;

@@ -482,6 +484,7 @@ int rvt_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 		if (tail == wc->head)
 			break;
 		/* The kernel doesn't need a RMB since it has the lock. */
+		trace_rvt_cq_poll(cq, &wc->kqueue[tail], npolled);
 		*entry = wc->kqueue[tail];
 		if (tail >= cq->ibcq.cqe)
 			tail = 0;

index e2d23ac..89554c0 100644
@@ -52,3 +52,4 @@
 #include "trace_qp.h"
 #include "trace_tx.h"
 #include "trace_mr.h"
+#include "trace_cq.h"

127 lines not on this page · raw message

index 4649530..0e56578 100644
@@ -1028,6 +1028,7 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr)
 		    cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)
 			break;
 		s_last = qp->s_last;
+		trace_hfi1_qp_send_completion(qp, wqe, s_last);
 		if (++s_last >= qp->s_size)
 			s_last = 0;
 		qp->s_last = s_last;

@@ -1079,6 +1080,7 @@ static inline void update_last_psn(struct rvt_qp *qp, u32 psn)
 
 		rvt_put_swqe(wqe);
 		s_last = qp->s_last;
+		trace_hfi1_qp_send_completion(qp, wqe, s_last);
 		if (++s_last >= qp->s_size)
 			s_last = 0;
 		qp->s_last = s_last;

index d2eb793..eeb650d 100644
@@ -909,8 +909,10 @@ void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
 
 	last = qp->s_last;
 	old_last = last;
+	trace_hfi1_qp_send_completion(qp, wqe, last);
 	if (++last >= qp->s_size)
 		last = 0;
+	trace_hfi1_qp_send_completion(qp, wqe, last);
 	qp->s_last = last;
 	/* See post_send() */
 	barrier();

49 lines not on this page · raw message

index f40864e..9bfb8eb 100644
@@ -1425,6 +1425,16 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	/* First, lock the non-writable module parameters */
 	HFI1_CAP_LOCK();
 
+	/* Validate dev ids */
+	if (!(ent->device == PCI_DEVICE_ID_INTEL0 ||
+	      ent->device == PCI_DEVICE_ID_INTEL1)) {
+		hfi1_early_err(&pdev->dev,
+			       "Failing on unknown Intel deviceid 0x%x\n",
+			       ent->device);
+		ret = -ENODEV;
+		goto bail;
+	}
+
 	/* Validate some global module parameters */
 	ret = init_validate_rcvhdrcnt(&pdev->dev, rcvhdrcnt);
 	if (ret)

@@ -1470,15 +1480,6 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	if (ret)
 		goto bail;
 
-	if (!(ent->device == PCI_DEVICE_ID_INTEL0 ||
-	      ent->device == PCI_DEVICE_ID_INTEL1)) {
-		hfi1_early_err(&pdev->dev,
-			       "Failing on unknown Intel deviceid 0x%x\n",
-			       ent->device);
-		ret = -ENODEV;
-		goto clean_bail;
-	}
-
 	/*
 	 * Do device-specific initialization, function table setup, dd
 	 * allocation, etc.

index 815cb44..8d71654 100644
@@ -1540,6 +1540,7 @@ static void hfi1_get_dev_fw_str(struct ib_device *ibdev, char *str,
 	"DRIVER_EgrHdrFull"
 };
 
+static DEFINE_MUTEX(cntr_names_lock); /* protects the *_cntr_names bufers */
 static const char **dev_cntr_names;
 static const char **port_cntr_names;
 static int num_driver_cntrs = ARRAY_SIZE(driver_cntr_names);

@@ -1594,6 +1595,7 @@ static int init_cntr_names(const char *names_in,
 {
 	int i, err;
 
+	mutex_lock(&cntr_names_lock);
 	if (!cntr_names_initialized) {
 		struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 

@@ -1602,8 +1604,10 @@ static int init_cntr_names(const char *names_in,
 				      num_driver_cntrs,
 				      &num_dev_cntrs,
 				      &dev_cntr_names);
-		if (err)
+		if (err) {
+			mutex_unlock(&cntr_names_lock);
 			return NULL;
+		}
 
 		for (i = 0; i < num_driver_cntrs; i++)
 			dev_cntr_names[num_dev_cntrs + i] =

@@ -1617,10 +1621,12 @@ static int init_cntr_names(const char *names_in,
 		if (err) {
 			kfree(dev_cntr_names);
 			dev_cntr_names = NULL;
+			mutex_unlock(&cntr_names_lock);
 			return NULL;
 		}
 		cntr_names_initialized = 1;
 	}
+	mutex_unlock(&cntr_names_lock);
 
 	if (!port_num)
 		return rdma_alloc_hw_stats_struct(

@@ -1839,9 +1845,13 @@ void hfi1_unregister_ib_device(struct hfi1_devdata *dd)
 	del_timer_sync(&dev->mem_timer);
 	verbs_txreq_exit(dev);
 
+	mutex_lock(&cntr_names_lock);
 	kfree(dev_cntr_names);
 	kfree(port_cntr_names);
+	dev_cntr_names = NULL;
+	port_cntr_names = NULL;
 	cntr_names_initialized = 0;
+	mutex_unlock(&cntr_names_lock);
 }
 
 void hfi1_cnp_rcv(struct hfi1_packet *packet)

index 8b8840a..f9d0d8c 100644
@@ -9533,8 +9533,11 @@ static int test_qsfp_read(struct hfi1_pportdata *ppd)
 	int ret;
 	u8 status;
 
-	/* report success if not a QSFP */
-	if (ppd->port_type != PORT_TYPE_QSFP)
+	/*
+	 * Report success if not a QSFP or, if it is a QSFP, but the cable is
+	 * not present
+	 */
+	if (ppd->port_type != PORT_TYPE_QSFP || !qsfp_mod_present(ppd))
 		return 0;
 
 	/* read byte 2, the status byte */

index f9d0d8c..77f4b41 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -7166,7 +7166,7 @@ static void get_link_widths(struct hfi1_devdata *dd, u16 *tx_width,
 	 * set the max_rate field in handle_verify_cap until v0.19.
 	 */
 	if ((dd->icode == ICODE_RTL_SILICON) &&
-	    (dd->dc8051_ver < dc8051_ver(0, 19))) {
+	    (dd->dc8051_ver < dc8051_ver(0, 19, 0))) {
 		/* max_rate: 0 = 12.5G, 1 = 25G */
 		switch (max_rate) {
 		case 0:

@@ -7351,7 +7351,7 @@ void handle_verify_cap(struct work_struct *work)
 	}
 
 	ppd->link_speed_active = 0;	/* invalid value */
-	if (dd->dc8051_ver < dc8051_ver(0, 20)) {
+	if (dd->dc8051_ver < dc8051_ver(0, 20, 0)) {
 		/* remote_tx_rate: 0 = 12.5G, 1 = 25G */
 		switch (remote_tx_rate) {
 		case 0:

@@ -8422,7 +8422,7 @@ static int write_lcb_via_8051(struct hfi1_devdata *dd, u32 addr, u64 data)
 	int ret;
 
 	if (dd->icode == ICODE_FUNCTIONAL_SIMULATOR ||
-	    (dd->dc8051_ver < dc8051_ver(0, 20))) {
+	    (dd->dc8051_ver < dc8051_ver(0, 20, 0))) {
 		if (acquire_lcb_access(dd, 0) == 0) {
 			write_csr(dd, addr, data);
 			release_lcb_access(dd, 0);

@@ -8728,13 +8728,20 @@ static void read_remote_device_id(struct hfi1_devdata *dd, u16 *device_id,
 			& REMOTE_DEVICE_REV_MASK;
 }
 
-void read_misc_status(struct hfi1_devdata *dd, u8 *ver_a, u8 *ver_b)
+void read_misc_status(struct hfi1_devdata *dd, u8 *ver_major, u8 *ver_minor,
+		      u8 *ver_patch)
 {
 	u32 frame;
 
 	read_8051_config(dd, MISC_STATUS, GENERAL_CONFIG, &frame);
-	*ver_a = (frame >> STS_FM_VERSION_A_SHIFT) & STS_FM_VERSION_A_MASK;
-	*ver_b = (frame >> STS_FM_VERSION_B_SHIFT) & STS_FM_VERSION_B_MASK;
+	*ver_major = (frame >> STS_FM_VERSION_MAJOR_SHIFT) &
+		STS_FM_VERSION_MAJOR_MASK;
+	*ver_minor = (frame >> STS_FM_VERSION_MINOR_SHIFT) &
+		STS_FM_VERSION_MINOR_MASK;
+
+	read_8051_config(dd, VERSION_PATCH, GENERAL_CONFIG, &frame);
+	*ver_patch = (frame >> STS_FM_VERSION_PATCH_SHIFT) &
+		STS_FM_VERSION_PATCH_MASK;
 }
 
 static void read_vc_remote_phy(struct hfi1_devdata *dd, u8 *power_management,

@@ -9130,7 +9137,7 @@ static int set_local_link_attributes(struct hfi1_pportdata *ppd)
 	if (ret)
 		goto set_local_link_attributes_fail;
 
-	if (dd->dc8051_ver < dc8051_ver(0, 20)) {
+	if (dd->dc8051_ver < dc8051_ver(0, 20, 0)) {
 		/* set the tx rate to the fastest enabled */
 		if (ppd->link_speed_enabled & OPA_LINK_SPEED_25G)
 			ppd->local_tx_rate = 1;

index 043fd21..24df45f 100644
@@ -1,7 +1,7 @@
 #ifndef _CHIP_H
 #define _CHIP_H
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -394,7 +394,8 @@
 #define LAST_REMOTE_STATE_COMPLETE   0x13
 #define LINK_QUALITY_INFO            0x14
 #define REMOTE_DEVICE_ID	     0x15
-#define LINK_DOWN_REASON	     0x16
+#define LINK_DOWN_REASON	     0x16 /* first byte of offset 0x16 */
+#define VERSION_PATCH		     0x16 /* last byte of offset 0x16 */
 
 /* 8051 lane specific register field IDs */
 #define TX_EQ_SETTINGS		0x00

@@ -524,10 +525,12 @@ enum {
 #define SUPPORTED_CRCS (CAP_CRC_14B | CAP_CRC_48B)
 
 /* misc status version fields */
-#define STS_FM_VERSION_A_SHIFT 16
-#define STS_FM_VERSION_A_MASK  0xff
-#define STS_FM_VERSION_B_SHIFT 24
-#define STS_FM_VERSION_B_MASK  0xff
+#define STS_FM_VERSION_MINOR_SHIFT 16
+#define STS_FM_VERSION_MINOR_MASK  0xff
+#define STS_FM_VERSION_MAJOR_SHIFT 24
+#define STS_FM_VERSION_MAJOR_MASK  0xff
+#define STS_FM_VERSION_PATCH_SHIFT 24
+#define STS_FM_VERSION_PATCH_MASK  0xff
 
 /* LCB_CFG_CRC_MODE TX_VAL and RX_VAL CRC mode values */
 #define LCB_CRC_16B			0x0	/* 16b CRC */

@@ -698,7 +701,8 @@ bool check_chip_resource(struct hfi1_devdata *dd, u32 resource,
 int read_8051_data(struct hfi1_devdata *dd, u32 addr, u32 len, u64 *result);
 
 /* chip.c */
-void read_misc_status(struct hfi1_devdata *dd, u8 *ver_a, u8 *ver_b);
+void read_misc_status(struct hfi1_devdata *dd, u8 *ver_major, u8 *ver_minor,
+		      u8 *ver_patch);
 void read_guid(struct hfi1_devdata *dd);
 int wait_fm_ready(struct hfi1_devdata *dd, u32 mstimeout);
 void set_link_down_reason(struct hfi1_pportdata *ppd, u8 lcl_reason,

index 0dd50cd..4042c11 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -1004,7 +1004,9 @@ static int load_8051_firmware(struct hfi1_devdata *dd,
 {
 	u64 reg;
 	int ret;
-	u8 ver_a, ver_b;
+	u8 ver_major;
+	u8 ver_minor;
+	u8 ver_patch;
 
 	/*
 	 * DC Reset sequence

@@ -1073,10 +1075,10 @@ static int load_8051_firmware(struct hfi1_devdata *dd,
 		return -ETIMEDOUT;
 	}
 
-	read_misc_status(dd, &ver_a, &ver_b);
-	dd_dev_info(dd, "8051 firmware version %d.%d\n",
-		    (int)ver_b, (int)ver_a);
-	dd->dc8051_ver = dc8051_ver(ver_b, ver_a);
+	read_misc_status(dd, &ver_major, &ver_minor, &ver_patch);
+	dd_dev_info(dd, "8051 firmware version %d.%d.%d\n",
+		    (int)ver_major, (int)ver_minor, (int)ver_patch);
+	dd->dc8051_ver = dc8051_ver(ver_major, ver_minor, ver_patch);
 
 	return 0;
 }

index b69ab47..a31638c 100644
@@ -1020,7 +1020,7 @@ struct hfi1_devdata {
 	u8 qos_shift;
 
 	u16 irev;	/* implementation revision */
-	u16 dc8051_ver; /* 8051 firmware version */
+	u32 dc8051_ver; /* 8051 firmware version */
 
 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
 	struct platform_config platform_config;

@@ -1173,9 +1173,10 @@ struct hfi1_devdata {
 };
 
 /* 8051 firmware version helper */
-#define dc8051_ver(a, b) ((a) << 8 | (b))
-#define dc8051_ver_maj(a) ((a & 0xff00) >> 8)
-#define dc8051_ver_min(a)  (a & 0x00ff)
+#define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
+#define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
+#define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
+#define dc8051_ver_patch(a) ((a) & 0x0000ff)
 
 /* f_put_tid types */
 #define PT_EXPECTED 0

index 8d71654..928918c 100644
@@ -1236,12 +1236,14 @@ int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 static void hfi1_fill_device_attr(struct hfi1_devdata *dd)
 {
 	struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
-	u16 ver = dd->dc8051_ver;
+	u32 ver = dd->dc8051_ver;
 
 	memset(&rdi->dparms.props, 0, sizeof(rdi->dparms.props));
 
-	rdi->dparms.props.fw_ver = ((u64)(dc8051_ver_maj(ver)) << 16) |
-				    (u64)dc8051_ver_min(ver);
+	rdi->dparms.props.fw_ver = ((u64)(dc8051_ver_maj(ver)) << 32) |
+		((u64)(dc8051_ver_min(ver)) << 16) |
+		(u64)dc8051_ver_patch(ver);
+
 	rdi->dparms.props.device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
 			IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
 			IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_RC_RNR_NAK_GEN |

@@ -1520,10 +1522,10 @@ static void hfi1_get_dev_fw_str(struct ib_device *ibdev, char *str,
 {
 	struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
 	struct hfi1_ibdev *dev = dev_from_rdi(rdi);
-	u16 ver = dd_from_dev(dev)->dc8051_ver;
+	u32 ver = dd_from_dev(dev)->dc8051_ver;
 
-	snprintf(str, str_len, "%u.%u", dc8051_ver_maj(ver),
-		 dc8051_ver_min(ver));
+	snprintf(str, str_len, "%u.%u.%u", dc8051_ver_maj(ver),
+		 dc8051_ver_min(ver), dc8051_ver_patch(ver));
 }
 
 static const char * const driver_cntr_names[] = {

index 5ea5005..8ce4765 100644
@@ -1,5 +1,5 @@
 /*
-* Copyright(c) 2015, 2016 Intel Corporation.
+* Copyright(c) 2015, 2016, 2017 Intel Corporation.
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.

@@ -104,11 +104,6 @@
 	     TP_ARGS(qp, psn)
 );
 
-DEFINE_EVENT(hfi1_rc_template, hfi1_timeout,
-	     TP_PROTO(struct rvt_qp *qp, u32 psn),
-	     TP_ARGS(qp, psn)
-);
-
 DEFINE_EVENT(hfi1_rc_template, hfi1_rcv_error,
 	     TP_PROTO(struct rvt_qp *qp, u32 psn),
 	     TP_ARGS(qp, psn)

index 3c55a8b..d7dabdf 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright(c) 2016 Intel Corporation.
+ * Copyright(c) 2016, 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -2052,8 +2052,12 @@ static void rvt_rc_timeout(unsigned long arg)
 	spin_lock_irqsave(&qp->r_lock, flags);
 	spin_lock(&qp->s_lock);
 	if (qp->s_flags & RVT_S_TIMER) {
+		struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
+
 		qp->s_flags &= ~RVT_S_TIMER;
+		rvp->n_rc_timeouts++;
 		del_timer(&qp->s_timer);
+		trace_rvt_rc_timeout(qp, qp->s_last_psn + 1);
 		if (rdi->driver_f.notify_restart_rc)
 			rdi->driver_f.notify_restart_rc(qp,
 							qp->s_last_psn + 1,

index 89554c0..bb4b1e7 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright(c) 2016 Intel Corporation.
+ * Copyright(c) 2016, 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -53,3 +53,4 @@
 #include "trace_tx.h"
 #include "trace_mr.h"
 #include "trace_cq.h"
+#include "trace_rc.h"

109 lines not on this page · raw message

index d7dabdf..728f5f1 100644
@@ -1772,10 +1772,13 @@ static int rvt_post_one_wr(struct rvt_qp *qp,
 					0);
 		qp->s_next_psn = wqe->lpsn + 1;
 	}
-	if (unlikely(reserved_op))
+	if (unlikely(reserved_op)) {
+		wqe->wr.send_flags |= RVT_SEND_RESERVE_USED;
 		rvt_qp_wqe_reserve(qp, wqe);
-	else
+	} else {
+		wqe->wr.send_flags &= ~RVT_SEND_RESERVE_USED;
 		qp->s_avail--;
+	}
 	trace_rvt_post_one_wr(qp, wqe);
 	smp_wmb(); /* see request builders */
 	qp->s_head = next;

index 3cdd9e2..e3bb312 100644
@@ -2,7 +2,7 @@
 #define DEF_RDMAVT_INCQP_H
 
 /*
- * Copyright(c) 2016 Intel Corporation.
+ * Copyright(c) 2016, 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.

@@ -526,7 +526,6 @@ static inline void rvt_qp_wqe_reserve(
 	struct rvt_qp *qp,
 	struct rvt_swqe *wqe)
 {
-	wqe->wr.send_flags |= RVT_SEND_RESERVE_USED;
 	atomic_inc(&qp->s_reserved_used);
 }
 

@@ -550,7 +549,6 @@ static inline void rvt_qp_wqe_unreserve(
 	struct rvt_swqe *wqe)
 {
 	if (unlikely(wqe->wr.send_flags & RVT_SEND_RESERVE_USED)) {
-		wqe->wr.send_flags &= ~RVT_SEND_RESERVE_USED;
 		atomic_dec(&qp->s_reserved_used);
 		/* insure no compiler re-order up to s_last change */
 		smp_mb__after_atomic();

index 9bee28d..1f7bf30 100644
@@ -962,6 +962,11 @@ ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
 			continue;
 		}
 
+		if (vl >= ARRAY_SIZE(rht_node->map)) {
+			ret = -EINVAL;
+			goto out;
+		}
+
 		rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu,
 						  sdma_rht_params);
 		if (!rht_node) {

index 7fe9dd8..cac6d52 100644
@@ -51,8 +51,12 @@
 #include <linux/export.h>
 #include <linux/module.h>
 #include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ratelimit.h>
+#include <linux/fault-inject.h>
 
 #include "hfi.h"
+#include "trace.h"
 #include "debugfs.h"
 #include "device.h"
 #include "qp.h"

@@ -1063,6 +1067,217 @@ static int _sdma_cpu_list_seq_show(struct seq_file *s, void *v)
 DEBUGFS_SEQ_FILE_OPEN(sdma_cpu_list)
 DEBUGFS_FILE_OPS(sdma_cpu_list);
 
+#ifdef CONFIG_FAULT_INJECTION
+static void *_fault_stats_seq_start(struct seq_file *s, loff_t *pos)
+{
+	struct hfi1_opcode_stats_perctx *opstats;
+
+	if (*pos >= ARRAY_SIZE(opstats->stats))
+		return NULL;
+	return pos;
+}
+
+static void *_fault_stats_seq_next(struct seq_file *s, void *v, loff_t *pos)
+{
+	struct hfi1_opcode_stats_perctx *opstats;
+
+	++*pos;
+	if (*pos >= ARRAY_SIZE(opstats->stats))
+		return NULL;
+	return pos;
+}
+
+static void _fault_stats_seq_stop(struct seq_file *s, void *v)
+{
+}
+
+static int _fault_stats_seq_show(struct seq_file *s, void *v)
+{
+	loff_t *spos = v;
+	loff_t i = *spos, j;
+	u64 n_packets = 0, n_bytes = 0;
+	struct hfi1_ibdev *ibd = (struct hfi1_ibdev *)s->private;
+	struct hfi1_devdata *dd = dd_from_dev(ibd);
+
+	for (j = 0; j < dd->first_user_ctxt; j++) {
+		if (!dd->rcd[j])
+			continue;
+		n_packets += dd->rcd[j]->opstats->stats[i].n_packets;
+		n_bytes += dd->rcd[j]->opstats->stats[i].n_bytes;
+	}
+	if (!n_packets && !n_bytes)
+		return SEQ_SKIP;
+	if (!ibd->fault_opcode->n_rxfaults[i] &&
+	    !ibd->fault_opcode->n_txfaults[i])
+		return SEQ_SKIP;
+	seq_printf(s, "%02llx %llu/%llu (faults rx:%llu faults: tx:%llu)\n", i,
+		   (unsigned long long)n_packets,
+		   (unsigned long long)n_bytes,
+		   (unsigned long long)ibd->fault_opcode->n_rxfaults[i],
+		   (unsigned long long)ibd->fault_opcode->n_txfaults[i]);
+	return 0;
+}
+
+DEBUGFS_SEQ_FILE_OPS(fault_stats);
+DEBUGFS_SEQ_FILE_OPEN(fault_stats);
+DEBUGFS_FILE_OPS(fault_stats);
+
+static void fault_exit_opcode_debugfs(struct hfi1_ibdev *ibd)
+{
+	debugfs_remove_recursive(ibd->fault_opcode->dir);
+	kfree(ibd->fault_opcode);
+	ibd->fault_opcode = NULL;
+}
+
+static int fault_init_opcode_debugfs(struct hfi1_ibdev *ibd)
+{
+	struct dentry *parent = ibd->hfi1_ibdev_dbg;
+
+	ibd->fault_opcode = kzalloc(sizeof(*ibd->fault_opcode), GFP_KERNEL);
+	if (!ibd->fault_opcode)
+		return -ENOMEM;
+
+	ibd->fault_opcode->attr.interval = 1;
+	ibd->fault_opcode->attr.require_end = ULONG_MAX;
+	ibd->fault_opcode->attr.stacktrace_depth = 32;
+	ibd->fault_opcode->attr.dname = NULL;
+	ibd->fault_opcode->attr.verbose = 0;
+	ibd->fault_opcode->fault_by_opcode = false;
+	ibd->fault_opcode->opcode = 0;
+	ibd->fault_opcode->mask = 0xff;
+
+	ibd->fault_opcode->dir =
+		fault_create_debugfs_attr("fault_opcode",
+					  parent,
+					  &ibd->fault_opcode->attr);
+	if (IS_ERR(ibd->fault_opcode->dir)) {
+		kfree(ibd->fault_opcode);
+		return -ENOENT;
+	}
+
+	DEBUGFS_SEQ_FILE_CREATE(fault_stats, ibd->fault_opcode->dir, ibd);
+	if (!debugfs_create_bool("fault_by_opcode", 0600,
+				 ibd->fault_opcode->dir,
+				 &ibd->fault_opcode->fault_by_opcode))
+		goto fail;
+	if (!debugfs_create_x8("opcode", 0600, ibd->fault_opcode->dir,
+			       &ibd->fault_opcode->opcode))
+		goto fail;
+	if (!debugfs_create_x8("mask", 0600, ibd->fault_opcode->dir,
+			       &ibd->fault_opcode->mask))
+		goto fail;
+
+	return 0;
+fail:
+	fault_exit_opcode_debugfs(ibd);
+	return -ENOMEM;
+}
+
+static void fault_exit_packet_debugfs(struct hfi1_ibdev *ibd)
+{
+	debugfs_remove_recursive(ibd->fault_packet->dir);
+	kfree(ibd->fault_packet);
+	ibd->fault_packet = NULL;
+}
+
+static int fault_init_packet_debugfs(struct hfi1_ibdev *ibd)
+{
+	struct dentry *parent = ibd->hfi1_ibdev_dbg;
+
+	ibd->fault_packet = kzalloc(sizeof(*ibd->fault_packet), GFP_KERNEL);
+	if (!ibd->fault_packet)
+		return -ENOMEM;
+
+	ibd->fault_packet->attr.interval = 1;
+	ibd->fault_packet->attr.require_end = ULONG_MAX;
+	ibd->fault_packet->attr.stacktrace_depth = 32;
+	ibd->fault_packet->attr.dname = NULL;
+	ibd->fault_packet->attr.verbose = 0;
+	ibd->fault_packet->fault_by_packet = false;
+
+	ibd->fault_packet->dir =
+		fault_create_debugfs_attr("fault_packet",
+					  parent,
+					  &ibd->fault_opcode->attr);
+	if (IS_ERR(ibd->fault_packet->dir)) {
+		kfree(ibd->fault_packet);
+		return -ENOENT;
+	}
+
+	if (!debugfs_create_bool("fault_by_packet", 0600,
+				 ibd->fault_packet->dir,
+				 &ibd->fault_packet->fault_by_packet))
+		goto fail;
+	if (!debugfs_create_u64("fault_stats", 0400,
+				ibd->fault_packet->dir,
+				&ibd->fault_packet->n_faults))
+		goto fail;
+
+	return 0;
+fail:
+	fault_exit_packet_debugfs(ibd);
+	return -ENOMEM;
+}
+
+static void fault_exit_debugfs(struct hfi1_ibdev *ibd)
+{
+	fault_exit_opcode_debugfs(ibd);
+	fault_exit_packet_debugfs(ibd);
+}
+
+static int fault_init_debugfs(struct hfi1_ibdev *ibd)
+{
+	int ret = 0;
+
+	ret = fault_init_opcode_debugfs(ibd);
+	if (ret)
+		return ret;
+
+	ret = fault_init_packet_debugfs(ibd);
+	if (ret)
+		fault_exit_opcode_debugfs(ibd);
+
+	return ret;
+}
+
+bool hfi1_dbg_fault_opcode(struct rvt_qp *qp, u32 opcode, bool rx)
+{
+	bool ret = false;
+	struct hfi1_ibdev *ibd = to_idev(qp->ibqp.device);
+
+	if (!ibd->fault_opcode || !ibd->fault_opcode->fault_by_opcode)
+		return false;
+	if (ibd->fault_opcode->opcode != (opcode & ibd->fault_opcode->mask))
+		return false;
+	ret = should_fail(&ibd->fault_opcode->attr, 1);
+	if (ret) {
+		trace_hfi1_fault_opcode(qp, opcode);
+		if (rx)
+			ibd->fault_opcode->n_rxfaults[opcode]++;
+		else
+			ibd->fault_opcode->n_txfaults[opcode]++;
+	}
+	return ret;
+}
+
+bool hfi1_dbg_fault_packet(struct hfi1_packet *packet)
+{
+	struct rvt_dev_info *rdi = &packet->rcd->ppd->dd->verbs_dev.rdi;
+	struct hfi1_ibdev *ibd = dev_from_rdi(rdi);
+	bool ret = false;
+
+	if (!ibd->fault_packet || !ibd->fault_packet->fault_by_packet)
+		return false;
+
+	ret = should_fail(&ibd->fault_packet->attr, 1);
+	if (ret) {
+		++ibd->fault_packet->n_faults;
+		trace_hfi1_fault_packet(packet);
+	}
+	return ret;
+}
+#endif
+
 void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd)
 {
 	char name[sizeof("port0counters") + 1];

@@ -1112,12 +1327,19 @@ void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd)
 					    !port_cntr_ops[i].ops.write ?
 					    S_IRUGO : S_IRUGO | S_IWUSR);
 		}
+
+#ifdef CONFIG_FAULT_INJECTION
+	fault_init_debugfs(ibd);
+#endif
 }
 
 void hfi1_dbg_ibdev_exit(struct hfi1_ibdev *ibd)
 {
 	if (!hfi1_dbg_root)
 		goto out;
+#ifdef CONFIG_FAULT_INJECTION
+	fault_exit_debugfs(ibd);
+#endif
 	debugfs_remove(ibd->hfi1_ibdev_link);
 	debugfs_remove_recursive(ibd->hfi1_ibdev_dbg);
 out:

71 lines not on this page · raw message

index 3881c95..c0b012f 100644
@@ -59,6 +59,7 @@
 #include "trace.h"
 #include "qp.h"
 #include "sdma.h"
+#include "debugfs.h"
 
 #undef pr_fmt
 #define pr_fmt(fmt) DRIVER_NAME ": " fmt

@@ -1354,6 +1355,9 @@ void handle_eflags(struct hfi1_packet *packet)
  */
 int process_receive_ib(struct hfi1_packet *packet)
 {
+	if (unlikely(hfi1_dbg_fault_packet(packet)))
+		return RHF_RCV_CONTINUE;
+
 	trace_hfi1_rcvhdr(packet->rcd->ppd->dd,
 			  packet->rcd->ctxt,
 			  rhf_err_flags(packet->rhf),

@@ -1409,6 +1413,8 @@ int process_receive_error(struct hfi1_packet *packet)
 
 int kdeth_process_expected(struct hfi1_packet *packet)
 {
+	if (unlikely(hfi1_dbg_fault_packet(packet)))
+		return RHF_RCV_CONTINUE;
 	if (unlikely(rhf_err_flags(packet->rhf)))
 		handle_eflags(packet);
 

@@ -1421,6 +1427,8 @@ int kdeth_process_eager(struct hfi1_packet *packet)
 {
 	if (unlikely(rhf_err_flags(packet->rhf)))
 		handle_eflags(packet);
+	if (unlikely(hfi1_dbg_fault_packet(packet)))
+		return RHF_RCV_CONTINUE;
 
 	dd_dev_err(packet->rcd->dd,
 		   "Unhandled eager packet received. Dropping.\n");

54 lines not on this page · raw message

index 928918c..9f016da 100644
@@ -60,6 +60,7 @@
 #include "trace.h"
 #include "qp.h"
 #include "verbs_txreq.h"
+#include "debugfs.h"
 
 static unsigned int hfi1_lkey_table_size = 16;
 module_param_named(lkey_table_size, hfi1_lkey_table_size, uint,

@@ -599,6 +600,11 @@ void hfi1_ib_rcv(struct hfi1_packet *packet)
 			rcu_read_unlock();
 			goto drop;
 		}
+		if (unlikely(hfi1_dbg_fault_opcode(packet->qp, opcode,
+						   true))) {
+			rcu_read_unlock();
+			goto drop;
+		}
 		spin_lock_irqsave(&packet->qp->r_lock, flags);
 		packet_handler = qp_ok(opcode, packet);
 		if (likely(packet_handler))

index 3a0b589..2756ec3 100644
@@ -195,6 +195,10 @@ struct hfi1_ibdev {
 	struct dentry *hfi1_ibdev_dbg;
 	/* per HFI symlinks to above */
 	struct dentry *hfi1_ibdev_link;
+#ifdef CONFIG_FAULT_INJECTION
+	struct fault_opcode *fault_opcode;
+	struct fault_packet *fault_packet;
+#endif
 #endif
 };
 

index 77f4b41..79a316a 100644
@@ -64,6 +64,7 @@
 #include "platform.h"
 #include "aspm.h"
 #include "affinity.h"
+#include "debugfs.h"
 
 #define NUM_IB_PORTS 1
 

@@ -7898,6 +7899,9 @@ static void handle_dcc_err(struct hfi1_devdata *dd, u32 unused, u64 reg)
 		reg &= ~DCC_ERR_FLG_EN_CSR_ACCESS_BLOCKED_HOST_SMASK;
 	}
 
+	if (unlikely(hfi1_dbg_fault_suppress_err(&dd->verbs_dev)))
+		reg &= ~DCC_ERR_FLG_LATE_EBP_ERR_SMASK;
+
 	/* report any remaining errors */
 	if (reg)
 		dd_dev_info_ratelimited(dd, "DCC Error: %s\n",

index cac6d52..dc2c1c9 100644
@@ -1240,6 +1240,11 @@ static int fault_init_debugfs(struct hfi1_ibdev *ibd)
 	return ret;
 }
 
+bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd)
+{
+	return ibd->fault_suppress_err;
+}
+
 bool hfi1_dbg_fault_opcode(struct rvt_qp *qp, u32 opcode, bool rx)
 {
 	bool ret = false;

@@ -1329,6 +1334,9 @@ void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd)
 		}
 
 #ifdef CONFIG_FAULT_INJECTION
+	debugfs_create_bool("fault_suppress_err", 0600,
+			    ibd->hfi1_ibdev_dbg,
+			    &ibd->fault_suppress_err);
 	fault_init_debugfs(ibd);
 #endif
 }

index 70be5ca..38c38a9 100644
@@ -75,6 +75,7 @@ struct fault_packet {
 
 bool hfi1_dbg_fault_opcode(struct rvt_qp *qp, u32 opcode, bool rx);
 bool hfi1_dbg_fault_packet(struct hfi1_packet *packet);
+bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd);
 #else
 static inline bool hfi1_dbg_fault_packet(struct hfi1_packet *packet)
 {

@@ -86,6 +87,11 @@ static inline bool hfi1_dbg_fault_opcode(struct rvt_qp *qp,
 {
 	return false;
 }
+
+static inline bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd)
+{
+	return false;
+}
 #endif
 
 #else

@@ -115,6 +121,11 @@ static inline bool hfi1_dbg_fault_opcode(struct rvt_qp *qp,
 {
 	return false;
 }
+
+static inline bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd)
+{
+	return false;
+}
 #endif
 
 #endif                          /* _HFI1_DEBUGFS_H */

index c0b012f..64bdbce 100644
@@ -1367,6 +1367,11 @@ int process_receive_ib(struct hfi1_packet *packet)
 			  packet->updegr,
 			  rhf_egr_index(packet->rhf));
 
+	if (unlikely(
+		 (hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
+		 (packet->rhf & RHF_DC_ERR))))
+		return RHF_RCV_CONTINUE;
+
 	if (unlikely(rhf_err_flags(packet->rhf))) {
 		handle_eflags(packet);
 		return RHF_RCV_CONTINUE;

@@ -1402,6 +1407,12 @@ int process_receive_bypass(struct hfi1_packet *packet)
 
 int process_receive_error(struct hfi1_packet *packet)
 {
+	/* KHdrHCRCErr -- KDETH packet with a bad HCRC */
+	if (unlikely(
+		 hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
+		 rhf_rcv_type_err(packet->rhf) == 3))
+		return RHF_RCV_CONTINUE;
+
 	handle_eflags(packet);
 
 	if (unlikely(rhf_err_flags(packet->rhf)))

90 lines not on this page · raw message

index 2756ec3..6c549e7 100644
@@ -198,6 +198,7 @@ struct hfi1_ibdev {
 #ifdef CONFIG_FAULT_INJECTION
 	struct fault_opcode *fault_opcode;
 	struct fault_packet *fault_packet;
+	bool fault_suppress_err;
 #endif
 #endif
 };

index b13419c..3665589 100644
@@ -80,6 +80,8 @@ enum {
 	IB_OPCODE_UD                                = 0x60,
 	/* per IBTA 1.3 vol 1 Table 38, A10.3.2 */
 	IB_OPCODE_CNP                               = 0x80,
+	/* Manufacturer specific */
+	IB_OPCODE_MSP                               = 0xe0,
 
 	/* operations -- just used to define real constants */
 	IB_OPCODE_SEND_FIRST                        = 0x00,

index ae30b68..7c86955 100644
@@ -191,8 +191,9 @@ static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region)
 
 		tmr = rcu_access_pointer(dev->dma_mr);
 		if (!tmr) {
-			rcu_assign_pointer(dev->dma_mr, mr);
 			mr->lkey_published = 1;
+			/* Insure published written first */
+			rcu_assign_pointer(dev->dma_mr, mr);
 			rvt_get_mr(mr);
 		}
 		goto success;

@@ -224,8 +225,9 @@ static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region)
 		mr->lkey |= 1 << 8;
 		rkt->gen++;
 	}
-	rcu_assign_pointer(rkt->table[r], mr);
 	mr->lkey_published = 1;
+	/* Insure published written first */
+	rcu_assign_pointer(rkt->table[r], mr);
 success:
 	spin_unlock_irqrestore(&rkt->lock, flags);
 out:

@@ -253,23 +255,24 @@ static void rvt_free_lkey(struct rvt_mregion *mr)
 	spin_lock_irqsave(&rkt->lock, flags);
 	if (!lkey) {
 		if (mr->lkey_published) {
-			RCU_INIT_POINTER(dev->dma_mr, NULL);
+			mr->lkey_published = 0;
+			/* insure published is written before pointer */
+			rcu_assign_pointer(dev->dma_mr, NULL);
 			rvt_put_mr(mr);
 		}
 	} else {
 		if (!mr->lkey_published)
 			goto out;
 		r = lkey >> (32 - dev->dparms.lkey_table_size);
-		RCU_INIT_POINTER(rkt->table[r], NULL);
+		mr->lkey_published = 0;
+		/* insure published is written before pointer */
+		rcu_assign_pointer(rkt->table[r], NULL);
 	}
-	mr->lkey_published = 0;
 	freed++;
 out:
 	spin_unlock_irqrestore(&rkt->lock, flags);
-	if (freed) {
-		synchronize_rcu();
+	if (freed)
 		percpu_ref_kill(&mr->refcount);
-	}
 }
 
 static struct rvt_mr *__rvt_alloc_mr(int count, struct ib_pd *pd)

@@ -822,16 +825,21 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 		goto ok;
 	}
 	mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]);
-	if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
-		     mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
+	if (!mr)
 		goto bail;
+	rvt_get_mr(mr);
+	if (!READ_ONCE(mr->lkey_published))
+		goto bail_unref;
+
+	if (unlikely(atomic_read(&mr->lkey_invalid) ||
+		     mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
+		goto bail_unref;
 
 	off = sge->addr - mr->user_base;
 	if (unlikely(sge->addr < mr->user_base ||
 		     off + sge->length > mr->length ||
 		     (mr->access_flags & acc) != acc))
-		goto bail;
-	rvt_get_mr(mr);
+		goto bail_unref;
 	rcu_read_unlock();
 
 	off += mr->offset;

@@ -867,6 +875,8 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 	isge->n = n;
 ok:
 	return 1;
+bail_unref:
+	rvt_put_mr(mr);
 bail:
 	rcu_read_unlock();
 	return 0;

@@ -922,15 +932,20 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 	}
 
 	mr = rcu_dereference(rkt->table[rkey >> rkt->shift]);
-	if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
-		     mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+	if (!mr)
 		goto bail;
+	rvt_get_mr(mr);
+	/* insure mr read is before test */
+	if (!READ_ONCE(mr->lkey_published))
+		goto bail_unref;
+	if (unlikely(atomic_read(&mr->lkey_invalid) ||
+		     mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+		goto bail_unref;
 
 	off = vaddr - mr->iova;
 	if (unlikely(vaddr < mr->iova || off + len > mr->length ||
 		     (mr->access_flags & acc) == 0))
-		goto bail;
-	rvt_get_mr(mr);
+		goto bail_unref;
 	rcu_read_unlock();
 
 	off += mr->offset;

@@ -966,6 +981,8 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 	sge->n = n;
 ok:
 	return 1;
+bail_unref:
+	rvt_put_mr(mr);
 bail:
 	rcu_read_unlock();
 	return 0;

index 5e7e577..552b26d 100644
@@ -68,6 +68,12 @@
 MODULE_PARM_DESC(lkey_table_size,
 		 "LKEY table size in bits (2^n, 1 <= n <= 23)");
 
+static unsigned int hfi1_no_user_mr_percpu;
+module_param_named(no_user_mr_percpu, hfi1_no_user_mr_percpu, uint,
+		   S_IRUGO);
+MODULE_PARM_DESC(no_user_mr_percpu,
+		 "Avoid percpu refcount for user MRs (default 0)");
+
 static unsigned int hfi1_max_pds = 0xFFFF;
 module_param_named(max_pds, hfi1_max_pds, uint, S_IRUGO);
 MODULE_PARM_DESC(max_pds,

@@ -1841,6 +1847,7 @@ int hfi1_register_ib_device(struct hfi1_devdata *dd)
 	/* misc settings */
 	dd->verbs_dev.rdi.flags = 0; /* Let rdmavt handle it all */
 	dd->verbs_dev.rdi.dparms.lkey_table_size = hfi1_lkey_table_size;
+	dd->verbs_dev.rdi.dparms.no_user_mr_percpu = hfi1_no_user_mr_percpu;
 	dd->verbs_dev.rdi.dparms.nports = dd->num_pports;
 	dd->verbs_dev.rdi.dparms.npkeys = hfi1_get_npkeys(dd);
 

index e120efe..6c718cd 100644
@@ -56,6 +56,12 @@
 MODULE_PARM_DESC(lkey_table_size,
 		 "LKEY table size in bits (2^n, 1 <= n <= 23)");
 
+static unsigned int qib_no_user_mr_percpu;
+module_param_named(no_user_mr_percpu, qib_no_user_mr_percpu, uint,
+		   S_IRUGO);
+MODULE_PARM_DESC(no_user_mr_percpu,
+		 "Avoid percpu refcount for user MRs (default 0)");
+
 static unsigned int ib_qib_max_pds = 0xFFFF;
 module_param_named(max_pds, ib_qib_max_pds, uint, S_IRUGO);
 MODULE_PARM_DESC(max_pds,

@@ -1606,6 +1612,7 @@ int qib_register_ib_device(struct qib_devdata *dd)
 	dd->verbs_dev.rdi.dparms.max_rdma_atomic = QIB_MAX_RDMA_ATOMIC;
 	dd->verbs_dev.rdi.driver_f.get_guid_be = qib_get_guid_be;
 	dd->verbs_dev.rdi.dparms.lkey_table_size = qib_lkey_table_size;
+	dd->verbs_dev.rdi.dparms.no_user_mr_percpu = qib_no_user_mr_percpu;
 	dd->verbs_dev.rdi.dparms.qp_table_size = ib_qib_qp_table_size;
 	dd->verbs_dev.rdi.dparms.qpn_start = 1;
 	dd->verbs_dev.rdi.dparms.qpn_res_start = QIB_KD_QP;

index 7c86955..bbcc31f 100644
@@ -280,6 +280,7 @@ static void rvt_free_lkey(struct rvt_mregion *mr)
 	struct rvt_mr *mr;
 	int rval = -ENOMEM;
 	int m;
+	struct rvt_dev_info *dev = ib_to_rvt(pd->device);
 
 	/* Allocate struct plus pointers to first level page tables. */
 	m = (count + RVT_SEGSZ - 1) / RVT_SEGSZ;

@@ -287,7 +288,10 @@ static void rvt_free_lkey(struct rvt_mregion *mr)
 	if (!mr)
 		goto bail;
 
-	rval = rvt_init_mregion(&mr->mr, pd, count, 0);
+	rval = rvt_init_mregion(&mr->mr, pd, count,
+				ibpd_to_rvtpd(pd)->user &&
+				dev->dparms.no_user_mr_percpu ?
+					PERCPU_REF_INIT_ATOMIC : 0);
 	if (rval)
 		goto bail;
 	/*

index 8fc1ca7..d60a41e 100644
@@ -142,6 +142,7 @@ struct rvt_driver_params {
 	 * For instance special module parameters. Goes here.
 	 */
 	unsigned int lkey_table_size;
+	unsigned int no_user_mr_percpu;
 	unsigned int qp_table_size;
 	int qpn_start;
 	int qpn_inc;

index 7b7a76e..40cd335 100644
@@ -3186,6 +3186,7 @@ ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
 	struct ib_qp                 *qp;
 	struct ib_uverbs_mcast_entry *mcast;
 	int                           ret = -EINVAL;
+	bool                          found = false;
 
 	if (copy_from_user(&cmd, buf, sizeof cmd))
 		return -EFAULT;

@@ -3194,10 +3195,6 @@ ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
 	if (!qp)
 		return -EINVAL;
 
-	ret = ib_detach_mcast(qp, (union ib_gid *) cmd.gid, cmd.mlid);
-	if (ret)
-		goto out_put;
-
 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
 
 	list_for_each_entry(mcast, &obj->mcast_list, list)

@@ -3205,9 +3202,17 @@ ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
 			list_del(&mcast->list);
 			kfree(mcast);
+			found = true;
 			break;
 		}
 
+	if (!found) {
+		ret = -EINVAL;
+		goto out_put;
+	}
+
+	ret = ib_detach_mcast(qp, (union ib_gid *)cmd.gid, cmd.mlid);
+
 out_put:
 	put_qp_write(qp);