Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma updates from Jason Gunthorpe:
 "This has been a smaller cycle than normal. One new driver was
  accepted, which is unusual, and at least one more driver remains in
  review on the list.

  Summary:

   - Driver fixes for hns, hfi1, nes, rxe, i40iw, mlx5, cxgb4,
     vmw_pvrdma

   - Many patches from MatthewW converting radix tree and IDR users to
     use xarray

   - Introduction of tracepoints to the MAD layer

   - Build large SGLs at the start for DMA mapping and get the driver to
     split them

   - Generally clean SGL handling code throughout the subsystem

   - Support for restricting RDMA devices to net namespaces for
     containers

   - Progress to remove object allocation boilerplate code from drivers

   - Change in how the mlx5 driver shows representor ports linked to VFs

   - mlx5 uapi feature to access the on chip SW ICM memory

   - Add a new driver for 'EFA'. This is HW that supports user space
     packet processing through QPs in Amazon's cloud"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (186 commits)
  RDMA/ipoib: Allow user space differentiate between valid dev_port
  IB/core, ipoib: Do not overreact to SM LID change event
  RDMA/device: Don't fire uevent before device is fully initialized
  lib/scatterlist: Remove leftover from sg_page_iter comment
  RDMA/efa: Add driver to Kconfig/Makefile
  RDMA/efa: Add the efa module
  RDMA/efa: Add EFA verbs implementation
  RDMA/efa: Add common command handlers
  RDMA/efa: Implement functions that submit and complete admin commands
  RDMA/efa: Add the ABI definitions
  RDMA/efa: Add the com service API definitions
  RDMA/efa: Add the efa_com.h file
  RDMA/efa: Add the efa.h header file
  RDMA/efa: Add EFA device definitions
  RDMA: Add EFA related definitions
  RDMA/umem: Remove hugetlb flag
  RDMA/bnxt_re: Use core helpers to get aligned DMA address
  RDMA/i40iw: Use core helpers to get aligned DMA address within a supported page size
  RDMA/verbs: Add a DMA iterator to return aligned contiguous memory blocks
  RDMA/umem: Add API to find best driver supported page size in an MR
  ...

21 files changed, 381 insertions, 221 deletions

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index addefae16c9c..310105d4e3de 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -4104,6 +4104,9 @@ def_access_ibp_counter(seq_naks);
 
 static struct cntr_entry dev_cntrs[DEV_CNTR_LAST] = {
 [C_RCV_OVF] = RXE32_DEV_CNTR_ELEM(RcvOverflow, RCV_BUF_OVFL_CNT, CNTR_SYNTH),
+[C_RX_LEN_ERR] = RXE32_DEV_CNTR_ELEM(RxLenErr, RCV_LENGTH_ERR_CNT, CNTR_SYNTH),
+[C_RX_ICRC_ERR] = RXE32_DEV_CNTR_ELEM(RxICrcErr, RCV_ICRC_ERR_CNT, CNTR_SYNTH),
+[C_RX_EBP] = RXE32_DEV_CNTR_ELEM(RxEbpCnt, RCV_EBP_CNT, CNTR_SYNTH),
 [C_RX_TID_FULL] = RXE32_DEV_CNTR_ELEM(RxTIDFullEr, RCV_TID_FULL_ERR_CNT,
 			CNTR_NORMAL),
 [C_RX_TID_INVALID] = RXE32_DEV_CNTR_ELEM(RxTIDInvalid, RCV_TID_VALID_ERR_CNT,
@@ -13294,15 +13297,18 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 	/*
 	 * The RMT entries are currently allocated as shown below:
 	 * 1. QOS (0 to 128 entries);
-	 * 2. FECN for PSM (num_user_contexts + num_vnic_contexts);
+	 * 2. FECN (num_kernel_context - 1 + num_user_contexts +
+	 *    num_vnic_contexts);
 	 * 3. VNIC (num_vnic_contexts).
-	 * It should be noted that PSM FECN oversubscribe num_vnic_contexts
+	 * It should be noted that FECN oversubscribe num_vnic_contexts
 	 * entries of RMT because both VNIC and PSM could allocate any receive
 	 * context between dd->first_dyn_alloc_text and dd->num_rcv_contexts,
 	 * and PSM FECN must reserve an RMT entry for each possible PSM receive
 	 * context.
 	 */
 	rmt_count = qos_rmt_entries(dd, NULL, NULL) + (num_vnic_contexts * 2);
+	if (HFI1_CAP_IS_KSET(TID_RDMA))
+		rmt_count += num_kernel_contexts - 1;
 	if (rmt_count + n_usr_ctxts > NUM_MAP_ENTRIES) {
 		user_rmt_reduced = NUM_MAP_ENTRIES - rmt_count;
 		dd_dev_err(dd,
@@ -14285,37 +14291,43 @@ bail:
 	init_qpmap_table(dd, FIRST_KERNEL_KCTXT, dd->n_krcv_queues - 1);
 }
 
-static void init_user_fecn_handling(struct hfi1_devdata *dd,
-				    struct rsm_map_table *rmt)
+static void init_fecn_handling(struct hfi1_devdata *dd,
+			       struct rsm_map_table *rmt)
 {
 	struct rsm_rule_data rrd;
 	u64 reg;
-	int i, idx, regoff, regidx;
+	int i, idx, regoff, regidx, start;
 	u8 offset;
 	u32 total_cnt;
 
+	if (HFI1_CAP_IS_KSET(TID_RDMA))
+		/* Exclude context 0 */
+		start = 1;
+	else
+		start = dd->first_dyn_alloc_ctxt;
+
+	total_cnt = dd->num_rcv_contexts - start;
+
 	/* there needs to be enough room in the map table */
-	total_cnt = dd->num_rcv_contexts - dd->first_dyn_alloc_ctxt;
 	if (rmt->used + total_cnt >= NUM_MAP_ENTRIES) {
-		dd_dev_err(dd, "User FECN handling disabled - too many user contexts allocated\n");
+		dd_dev_err(dd, "FECN handling disabled - too many contexts allocated\n");
 		return;
 	}
 
 	/*
 	 * RSM will extract the destination context as an index into the
 	 * map table.  The destination contexts are a sequential block
-	 * in the range first_dyn_alloc_ctxt...num_rcv_contexts-1 (inclusive).
+	 * in the range start...num_rcv_contexts-1 (inclusive).
 	 * Map entries are accessed as offset + extracted value.  Adjust
 	 * the added offset so this sequence can be placed anywhere in
 	 * the table - as long as the entries themselves do not wrap.
 	 * There are only enough bits in offset for the table size, so
 	 * start with that to allow for a "negative" offset.
 	 */
-	offset = (u8)(NUM_MAP_ENTRIES + (int)rmt->used -
-						(int)dd->first_dyn_alloc_ctxt);
+	offset = (u8)(NUM_MAP_ENTRIES + rmt->used - start);
 
-	for (i = dd->first_dyn_alloc_ctxt, idx = rmt->used;
-				i < dd->num_rcv_contexts; i++, idx++) {
+	for (i = start, idx = rmt->used; i < dd->num_rcv_contexts;
+	     i++, idx++) {
 		/* replace with identity mapping */
 		regoff = (idx % 8) * 8;
 		regidx = idx / 8;
@@ -14437,7 +14449,7 @@ static void init_rxe(struct hfi1_devdata *dd)
 	rmt = alloc_rsm_map_table(dd);
 	/* set up QOS, including the QPN map table */
 	init_qos(dd, rmt);
-	init_user_fecn_handling(dd, rmt);
+	init_fecn_handling(dd, rmt);
 	complete_rsm_map_table(dd, rmt);
 	/* record number of used rsm map entries for vnic */
 	dd->vnic.rmt_start = rmt->used;
@@ -14663,8 +14675,8 @@ void hfi1_start_cleanup(struct hfi1_devdata *dd)
  */
 static int init_asic_data(struct hfi1_devdata *dd)
 {
-	unsigned long flags;
-	struct hfi1_devdata *tmp, *peer = NULL;
+	unsigned long index;
+	struct hfi1_devdata *peer;
 	struct hfi1_asic_data *asic_data;
 	int ret = 0;
 
@@ -14673,14 +14685,12 @@ static int init_asic_data(struct hfi1_devdata *dd)
 	if (!asic_data)
 		return -ENOMEM;
 
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
+	xa_lock_irq(&hfi1_dev_table);
 	/* Find our peer device */
-	list_for_each_entry(tmp, &hfi1_dev_list, list) {
-		if ((HFI_BASE_GUID(dd) == HFI_BASE_GUID(tmp)) &&
-		    dd->unit != tmp->unit) {
-			peer = tmp;
+	xa_for_each(&hfi1_dev_table, index, peer) {
+		if ((HFI_BASE_GUID(dd) == HFI_BASE_GUID(peer)) &&
+		    dd->unit != peer->unit)
 			break;
-		}
 	}
 
 	if (peer) {
@@ -14692,7 +14702,7 @@ static int init_asic_data(struct hfi1_devdata *dd)
 		mutex_init(&dd->asic_data->asic_resource_mutex);
 	}
 	dd->asic_data->dds[dd->hfi1_id] = dd; /* self back-pointer */
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
+	xa_unlock_irq(&hfi1_dev_table);
 
 	/* first one through - set up i2c devices */
 	if (!peer)
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index 6c27c1c6a868..4e6c3556ec48 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -858,6 +858,9 @@ static inline int idx_from_vl(int vl)
 /* Per device counter indexes */
 enum {
 	C_RCV_OVF = 0,
+	C_RX_LEN_ERR,
+	C_RX_ICRC_ERR,
+	C_RX_EBP,
 	C_RX_TID_FULL,
 	C_RX_TID_INVALID,
 	C_RX_TID_FLGMS,
diff --git a/drivers/infiniband/hw/hfi1/chip_registers.h b/drivers/infiniband/hw/hfi1/chip_registers.h
index c0800ea5a3f8..ab3589d17aee 100644
--- a/drivers/infiniband/hw/hfi1/chip_registers.h
+++ b/drivers/infiniband/hw/hfi1/chip_registers.h
@@ -380,6 +380,9 @@
 #define DC_LCB_PRF_TX_FLIT_CNT (DC_LCB_CSRS + 0x000000000418)
 #define DC_LCB_STS_LINK_TRANSFER_ACTIVE (DC_LCB_CSRS + 0x000000000468)
 #define DC_LCB_STS_ROUND_TRIP_LTP_CNT (DC_LCB_CSRS + 0x0000000004B0)
+#define RCV_LENGTH_ERR_CNT 0
+#define RCV_ICRC_ERR_CNT 6
+#define RCV_EBP_CNT 9
 #define RCV_BUF_OVFL_CNT 10
 #define RCV_CONTEXT_EGR_STALL 22
 #define RCV_DATA_PKT_CNT 0
diff --git a/drivers/infiniband/hw/hfi1/common.h b/drivers/infiniband/hw/hfi1/common.h
index 7310a5dba420..d47da7b0438f 100644
--- a/drivers/infiniband/hw/hfi1/common.h
+++ b/drivers/infiniband/hw/hfi1/common.h
@@ -286,7 +286,7 @@ struct diag_pkt {
 #define RHF_TID_ERR		(0x1ull << 59)
 #define RHF_LEN_ERR		(0x1ull << 60)
 #define RHF_ECC_ERR		(0x1ull << 61)
-#define RHF_VCRC_ERR		(0x1ull << 62)
+#define RHF_RESERVED		(0x1ull << 62)
 #define RHF_ICRC_ERR		(0x1ull << 63)
 
 #define RHF_ERROR_SMASK 0xffe0000000000000ull		/* bits 63:53 */
diff --git a/drivers/infiniband/hw/hfi1/debugfs.c b/drivers/infiniband/hw/hfi1/debugfs.c
index 427ba0ce74a5..15efb4a380b2 100644
--- a/drivers/infiniband/hw/hfi1/debugfs.c
+++ b/drivers/infiniband/hw/hfi1/debugfs.c
@@ -1080,6 +1080,77 @@ static int qsfp2_debugfs_release(struct inode *in, struct file *fp)
 	return __qsfp_debugfs_release(in, fp, 1);
 }
 
+#define EXPROM_WRITE_ENABLE BIT_ULL(14)
+
+static bool exprom_wp_disabled;
+
+static int exprom_wp_set(struct hfi1_devdata *dd, bool disable)
+{
+	u64 gpio_val = 0;
+
+	if (disable) {
+		gpio_val = EXPROM_WRITE_ENABLE;
+		exprom_wp_disabled = true;
+		dd_dev_info(dd, "Disable Expansion ROM Write Protection\n");
+	} else {
+		exprom_wp_disabled = false;
+		dd_dev_info(dd, "Enable Expansion ROM Write Protection\n");
+	}
+
+	write_csr(dd, ASIC_GPIO_OUT, gpio_val);
+	write_csr(dd, ASIC_GPIO_OE, gpio_val);
+
+	return 0;
+}
+
+static ssize_t exprom_wp_debugfs_read(struct file *file, char __user *buf,
+				      size_t count, loff_t *ppos)
+{
+	return 0;
+}
+
+static ssize_t exprom_wp_debugfs_write(struct file *file,
+				       const char __user *buf, size_t count,
+				       loff_t *ppos)
+{
+	struct hfi1_pportdata *ppd = private2ppd(file);
+	char cdata;
+
+	if (count != 1)
+		return -EINVAL;
+	if (get_user(cdata, buf))
+		return -EFAULT;
+	if (cdata == '0')
+		exprom_wp_set(ppd->dd, false);
+	else if (cdata == '1')
+		exprom_wp_set(ppd->dd, true);
+	else
+		return -EINVAL;
+
+	return 1;
+}
+
+static unsigned long exprom_in_use;
+
+static int exprom_wp_debugfs_open(struct inode *in, struct file *fp)
+{
+	if (test_and_set_bit(0, &exprom_in_use))
+		return -EBUSY;
+
+	return 0;
+}
+
+static int exprom_wp_debugfs_release(struct inode *in, struct file *fp)
+{
+	struct hfi1_pportdata *ppd = private2ppd(fp);
+
+	if (exprom_wp_disabled)
+		exprom_wp_set(ppd->dd, false);
+	clear_bit(0, &exprom_in_use);
+
+	return 0;
+}
+
 #define DEBUGFS_OPS(nm, readroutine, writeroutine)	\
 { \
 	.name = nm, \
@@ -1119,6 +1190,9 @@ static const struct counter_info port_cntr_ops[] = {
 		     qsfp1_debugfs_open, qsfp1_debugfs_release),
 	DEBUGFS_XOPS("qsfp2", qsfp2_debugfs_read, qsfp2_debugfs_write,
 		     qsfp2_debugfs_open, qsfp2_debugfs_release),
+	DEBUGFS_XOPS("exprom_wp", exprom_wp_debugfs_read,
+		     exprom_wp_debugfs_write, exprom_wp_debugfs_open,
+		     exprom_wp_debugfs_release),
 	DEBUGFS_OPS("asic_flags", asic_flags_read, asic_flags_write),
 	DEBUGFS_OPS("dc8051_memory", dc8051_memory_read, NULL),
 	DEBUGFS_OPS("lcb", debugfs_lcb_read, debugfs_lcb_write),
@@ -1302,15 +1376,15 @@ static void _driver_stats_seq_stop(struct seq_file *s, void *v)
 
 static u64 hfi1_sps_ints(void)
 {
-	unsigned long flags;
+	unsigned long index, flags;
 	struct hfi1_devdata *dd;
 	u64 sps_ints = 0;
 
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-	list_for_each_entry(dd, &hfi1_dev_list, list) {
+	xa_lock_irqsave(&hfi1_dev_table, flags);
+	xa_for_each(&hfi1_dev_table, index, dd) {
 		sps_ints += get_all_cpu_total(dd->int_counter);
 	}
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
+	xa_unlock_irqrestore(&hfi1_dev_table, flags);
 	return sps_ints;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index 2a9d2912f5db..01aa1f132f55 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -72,8 +72,6 @@
  */
 const char ib_hfi1_version[] = HFI1_DRIVER_VERSION "\n";
 
-DEFINE_SPINLOCK(hfi1_devs_lock);
-LIST_HEAD(hfi1_dev_list);
 DEFINE_MUTEX(hfi1_mutex);	/* general driver use */
 
 unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
@@ -175,11 +173,11 @@ int hfi1_count_active_units(void)
 {
 	struct hfi1_devdata *dd;
 	struct hfi1_pportdata *ppd;
-	unsigned long flags;
+	unsigned long index, flags;
 	int pidx, nunits_active = 0;
 
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-	list_for_each_entry(dd, &hfi1_dev_list, list) {
+	xa_lock_irqsave(&hfi1_dev_table, flags);
+	xa_for_each(&hfi1_dev_table, index, dd) {
 		if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase1)
 			continue;
 		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
@@ -190,7 +188,7 @@ int hfi1_count_active_units(void)
 			}
 		}
 	}
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
+	xa_unlock_irqrestore(&hfi1_dev_table, flags);
 	return nunits_active;
 }
 
@@ -264,7 +262,7 @@ static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
 	    hfi1_dbg_fault_suppress_err(verbs_dev))
 		return;
 
-	if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
+	if (packet->rhf & RHF_ICRC_ERR)
 		return;
 
 	if (packet->etype == RHF_RCV_TYPE_BYPASS) {
@@ -516,7 +514,9 @@ bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
 	 */
 	do_cnp = prescan ||
 		(opcode >= IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST &&
-		 opcode <= IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE);
+		 opcode <= IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE) ||
+		opcode == TID_OP(READ_RESP) ||
+		opcode == TID_OP(ACK);
 
 	/* Call appropriate CNP handler */
 	if (!ignore_fecn && do_cnp && fecn)
@@ -1581,7 +1581,7 @@ static void show_eflags_errs(struct hfi1_packet *packet)
 	u32 rte = rhf_rcv_type_err(packet->rhf);
 
 	dd_dev_err(rcd->dd,
-		   "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n",
+		   "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s] rte 0x%x\n",
 		   rcd->ctxt, packet->rhf,
 		   packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "",
 		   packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "",
@@ -1589,7 +1589,6 @@ static void show_eflags_errs(struct hfi1_packet *packet)
 		   packet->rhf & RHF_TID_ERR ? "tid " : "",
 		   packet->rhf & RHF_LEN_ERR ? "len " : "",
 		   packet->rhf & RHF_ECC_ERR ? "ecc " : "",
-		   packet->rhf & RHF_VCRC_ERR ? "vcrc " : "",
 		   packet->rhf & RHF_ICRC_ERR ? "icrc " : "",
 		   rte);
 }
diff --git a/drivers/infiniband/hw/hfi1/exp_rcv.c b/drivers/infiniband/hw/hfi1/exp_rcv.c
index 1be49a0d9c11..e9d5cc8b771a 100644
--- a/drivers/infiniband/hw/hfi1/exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/exp_rcv.c
@@ -112,9 +112,6 @@ int hfi1_alloc_ctxt_rcv_groups(struct hfi1_ctxtdata *rcd)
  */
 void hfi1_free_ctxt_rcv_groups(struct hfi1_ctxtdata *rcd)
 {
-	WARN_ON(!EXP_TID_SET_EMPTY(rcd->tid_full_list));
-	WARN_ON(!EXP_TID_SET_EMPTY(rcd->tid_used_list));
-
 	kfree(rcd->groups);
 	rcd->groups = NULL;
 	hfi1_exp_tid_group_init(rcd);
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index 048b5d73ba39..b458c218842b 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -54,7 +54,6 @@
 #include <linux/list.h>
 #include <linux/scatterlist.h>
 #include <linux/slab.h>
-#include <linux/idr.h>
 #include <linux/io.h>
 #include <linux/fs.h>
 #include <linux/completion.h>
@@ -65,6 +64,7 @@
 #include <linux/kthread.h>
 #include <linux/i2c.h>
 #include <linux/i2c-algo-bit.h>
+#include <linux/xarray.h>
 #include <rdma/ib_hdrs.h>
 #include <rdma/opa_addr.h>
 #include <linux/rhashtable.h>
@@ -1021,8 +1021,8 @@ struct hfi1_asic_data {
 struct hfi1_vnic_data {
 	struct hfi1_ctxtdata *ctxt[HFI1_NUM_VNIC_CTXT];
 	struct kmem_cache *txreq_cache;
+	struct xarray vesws;
 	u8 num_vports;
-	struct idr vesw_idr;
 	u8 rmt_start;
 	u8 num_ctxt;
 };
@@ -1041,7 +1041,6 @@ struct sdma_vl_map;
 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
 struct hfi1_devdata {
 	struct hfi1_ibdev verbs_dev;     /* must be first */
-	struct list_head list;
 	/* pointers to related structs for this device */
 	/* pci access data structure */
 	struct pci_dev *pcidev;
@@ -1426,8 +1425,7 @@ struct hfi1_filedata {
 	struct mm_struct *mm;
 };
 
-extern struct list_head hfi1_dev_list;
-extern spinlock_t hfi1_devs_lock;
+extern struct xarray hfi1_dev_table;
 struct hfi1_devdata *hfi1_lookup(int unit);
 
 static inline unsigned long uctxt_offset(struct hfi1_ctxtdata *uctxt)
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index faaaac8fbc55..71cb9525c074 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -49,7 +49,7 @@
 #include <linux/netdevice.h>
 #include <linux/vmalloc.h>
 #include <linux/delay.h>
-#include <linux/idr.h>
+#include <linux/xarray.h>
 #include <linux/module.h>
 #include <linux/printk.h>
 #include <linux/hrtimer.h>
@@ -124,7 +124,7 @@ MODULE_PARM_DESC(user_credit_return_threshold, "Credit return threshold for user
 
 static inline u64 encode_rcv_header_entry_size(u16 size);
 
-static struct idr hfi1_unit_table;
+DEFINE_XARRAY_FLAGS(hfi1_dev_table, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ);
 
 static int hfi1_create_kctxt(struct hfi1_devdata *dd,
 			     struct hfi1_pportdata *ppd)
@@ -469,7 +469,7 @@ int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa,
 		if (rcd->egrbufs.size < hfi1_max_mtu) {
 			rcd->egrbufs.size = __roundup_pow_of_two(hfi1_max_mtu);
 			hfi1_cdbg(PROC,
-				  "ctxt%u: eager bufs size too small. Adjusting to %zu\n",
+				  "ctxt%u: eager bufs size too small. Adjusting to %u\n",
 				    rcd->ctxt, rcd->egrbufs.size);
 		}
 		rcd->egrbufs.rcvtid_size = HFI1_MAX_EAGER_BUFFER_SIZE;
@@ -805,7 +805,8 @@ static int create_workqueues(struct hfi1_devdata *dd)
 			ppd->hfi1_wq =
 				alloc_workqueue(
 				    "hfi%d_%d",
-				    WQ_SYSFS | WQ_HIGHPRI | WQ_CPU_INTENSIVE,
+				    WQ_SYSFS | WQ_HIGHPRI | WQ_CPU_INTENSIVE |
+				    WQ_MEM_RECLAIM,
 				    HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES,
 				    dd->unit, pidx);
 			if (!ppd->hfi1_wq)
@@ -1018,21 +1019,9 @@ done:
 	return ret;
 }
 
-static inline struct hfi1_devdata *__hfi1_lookup(int unit)
-{
-	return idr_find(&hfi1_unit_table, unit);
-}
-
 struct hfi1_devdata *hfi1_lookup(int unit)
 {
-	struct hfi1_devdata *dd;
-	unsigned long flags;
-
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-	dd = __hfi1_lookup(unit);
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
-
-	return dd;
+	return xa_load(&hfi1_dev_table, unit);
 }
 
 /*
@@ -1200,7 +1189,7 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
 /*
  * Release our hold on the shared asic data.  If we are the last one,
  * return the structure to be finalized outside the lock.  Must be
- * holding hfi1_devs_lock.
+ * holding hfi1_dev_table lock.
  */
 static struct hfi1_asic_data *release_asic_data(struct hfi1_devdata *dd)
 {
@@ -1236,13 +1225,10 @@ static void hfi1_clean_devdata(struct hfi1_devdata *dd)
 	struct hfi1_asic_data *ad;
 	unsigned long flags;
 
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-	if (!list_empty(&dd->list)) {
-		idr_remove(&hfi1_unit_table, dd->unit);
-		list_del_init(&dd->list);
-	}
+	xa_lock_irqsave(&hfi1_dev_table, flags);
+	__xa_erase(&hfi1_dev_table, dd->unit);
 	ad = release_asic_data(dd);
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
+	xa_unlock_irqrestore(&hfi1_dev_table, flags);
 
 	finalize_asic_data(dd, ad);
 	free_platform_config(dd);
@@ -1286,13 +1272,10 @@ void hfi1_free_devdata(struct hfi1_devdata *dd)
  * Must be done via verbs allocator, because the verbs cleanup process
  * both does cleanup and free of the data structure.
  * "extra" is for chip-specific data.
- *
- * Use the idr mechanism to get a unit number for this unit.
  */
 static struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev,
 					       size_t extra)
 {
-	unsigned long flags;
 	struct hfi1_devdata *dd;
 	int ret, nports;
 
@@ -1307,21 +1290,10 @@ static struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev,
 	dd->pport = (struct hfi1_pportdata *)(dd + 1);
 	dd->pcidev = pdev;
 	pci_set_drvdata(pdev, dd);
-
-	INIT_LIST_HEAD(&dd->list);
-	idr_preload(GFP_KERNEL);
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-
-	ret = idr_alloc(&hfi1_unit_table, dd, 0, 0, GFP_NOWAIT);
-	if (ret >= 0) {
-		dd->unit = ret;
-		list_add(&dd->list, &hfi1_dev_list);
-	}
 	dd->node = NUMA_NO_NODE;
 
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
-	idr_preload_end();
-
+	ret = xa_alloc_irq(&hfi1_dev_table, &dd->unit, dd, xa_limit_32b,
+			GFP_KERNEL);
 	if (ret < 0) {
 		dev_err(&pdev->dev,
 			"Could not allocate unit ID: error %d\n", -ret);
@@ -1522,8 +1494,6 @@ static int __init hfi1_mod_init(void)
 	 * These must be called before the driver is registered with
 	 * the PCI subsystem.
 	 */
-	idr_init(&hfi1_unit_table);
-
 	hfi1_dbg_init();
 	ret = pci_register_driver(&hfi1_pci_driver);
 	if (ret < 0) {
@@ -1534,7 +1504,6 @@ static int __init hfi1_mod_init(void)
 
 bail_dev:
 	hfi1_dbg_exit();
-	idr_destroy(&hfi1_unit_table);
 	dev_cleanup();
 bail:
 	return ret;
@@ -1552,7 +1521,7 @@ static void __exit hfi1_mod_cleanup(void)
 	node_affinity_destroy_all();
 	hfi1_dbg_exit();
 
-	idr_destroy(&hfi1_unit_table);
+	WARN_ON(!xa_empty(&hfi1_dev_table));
 	dispose_firmware();	/* asymmetric with obtain_firmware() */
 	dev_cleanup();
 }
@@ -2071,7 +2040,7 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
 	rcd->egrbufs.size = alloced_bytes;
 
 	hfi1_cdbg(PROC,
-		  "ctxt%u: Alloced %u rcv tid entries @ %uKB, total %zuKB\n",
+		  "ctxt%u: Alloced %u rcv tid entries @ %uKB, total %uKB\n",
 		  rcd->ctxt, rcd->egrbufs.alloced,
 		  rcd->egrbufs.rcvtid_size / 1024, rcd->egrbufs.size / 1024);
 
diff --git a/drivers/infiniband/hw/hfi1/opfn.h b/drivers/infiniband/hw/hfi1/opfn.h
index 5f2011cabc25..62f93c1dc082 100644
--- a/drivers/infiniband/hw/hfi1/opfn.h
+++ b/drivers/infiniband/hw/hfi1/opfn.h
@@ -47,12 +47,14 @@
  * for future transactions
  */
 
+#include <linux/workqueue.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/rdmavt_qp.h>
+
 /* STL Verbs Extended */
 #define IB_BTHE_E_SHIFT           24
 #define HFI1_VERBS_E_ATOMIC_VADDR U64_MAX
 
-struct ib_atomic_eth;
-
 enum hfi1_opfn_codes {
 	STL_VERBS_EXTD_NONE = 0,
 	STL_VERBS_EXTD_TID_RDMA,
diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c
index eba300330a02..4e0e9fc0a777 100644
--- a/drivers/infiniband/hw/hfi1/qp.c
+++ b/drivers/infiniband/hw/hfi1/qp.c
@@ -742,6 +742,8 @@ void *qp_priv_alloc(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 		iowait_wakeup,
 		iowait_sdma_drained,
 		hfi1_init_priority);
+	/* Init to a value to start the running average correctly */
+	priv->s_running_pkt_size = piothreshold / 2;
 	return priv;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 5991211d72bd..a922edcf23d6 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -140,10 +140,7 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
 	case OP(RDMA_READ_RESPONSE_LAST):
 	case OP(RDMA_READ_RESPONSE_ONLY):
 		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* FALLTHROUGH */
 	case OP(ATOMIC_ACKNOWLEDGE):
 		/*
@@ -343,7 +340,8 @@ write_resp:
 			break;
 
 		e->sent = 1;
-		qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+		/* Do not free e->rdma_sge until all data are received */
+		qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
 		break;
 
 	case TID_OP(READ_RESP):
@@ -1836,7 +1834,7 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah)
 		qp->s_last = s_last;
 		/* see post_send() */
 		barrier();
-		rvt_put_swqe(wqe);
+		rvt_put_qp_swqe(qp, wqe);
 		rvt_qp_swqe_complete(qp,
 				     wqe,
 				     ib_hfi1_wc_opcode[wqe->wr.opcode],
@@ -1884,7 +1882,7 @@ struct rvt_swqe *do_rc_completion(struct rvt_qp *qp,
 		u32 s_last;
 
 		trdma_clean_swqe(qp, wqe);
-		rvt_put_swqe(wqe);
+		rvt_put_qp_swqe(qp, wqe);
 		rvt_qp_wqe_unreserve(qp, wqe);
 		s_last = qp->s_last;
 		trace_hfi1_qp_send_completion(qp, wqe, s_last);
@@ -2643,10 +2641,7 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data,
 		len = be32_to_cpu(reth->length);
 		if (unlikely(offset + len != e->rdma_sge.sge_length))
 			goto unlock_done;
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		if (len != 0) {
 			u32 rkey = be32_to_cpu(reth->rkey);
 			u64 vaddr = get_ib_reth_vaddr(reth);
@@ -3088,10 +3083,7 @@ send_last:
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		reth = &ohdr->u.rc.reth;
 		len = be32_to_cpu(reth->length);
 		if (len) {
@@ -3166,10 +3158,7 @@ send_last:
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* Process OPFN special virtual address */
 		if (opfn) {
 			opfn_conn_response(qp, e, ateth);
diff --git a/drivers/infiniband/hw/hfi1/rc.h b/drivers/infiniband/hw/hfi1/rc.h
index 8e0935b9bf2a..5ed5e85d5841 100644
--- a/drivers/infiniband/hw/hfi1/rc.h
+++ b/drivers/infiniband/hw/hfi1/rc.h
@@ -41,6 +41,14 @@ static inline u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe,
 	return rvt_restart_sge(ss, wqe, len);
 }
 
+static inline void release_rdma_sge_mr(struct rvt_ack_entry *e)
+{
+	if (e->rdma_sge.mr) {
+		rvt_put_mr(e->rdma_sge.mr);
+		e->rdma_sge.mr = NULL;
+	}
+}
+
 struct rvt_ack_entry *find_prev_entry(struct rvt_qp *qp, u32 psn, u8 *prev,
 				      u8 *prev_ack, bool *scheduled);
 int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, u64 val,
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index 124a3ec1e15c..23ac6057b211 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -524,7 +524,7 @@ void _hfi1_do_send(struct work_struct *work)
 
 /**
  * hfi1_do_send - perform a send on a QP
- * @work: contains a pointer to the QP
+ * @qp: a pointer to the QP
  * @in_thread: true if in a workqueue thread
  *
  * Process entries in the send work queue until credit or queue is
diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.c b/drivers/infiniband/hw/hfi1/tid_rdma.c
index 43cbce7a19ea..6fb93032fbef 100644
--- a/drivers/infiniband/hw/hfi1/tid_rdma.c
+++ b/drivers/infiniband/hw/hfi1/tid_rdma.c
@@ -67,8 +67,6 @@ static u32 mask_generation(u32 a)
 #define TID_RDMA_DESTQP_FLOW_SHIFT      11
 #define TID_RDMA_DESTQP_FLOW_MASK       0x1f
 
-#define TID_FLOW_SW_PSN BIT(0)
-
 #define TID_OPFN_QP_CTXT_MASK 0xff
 #define TID_OPFN_QP_CTXT_SHIFT 56
 #define TID_OPFN_QP_KDETH_MASK 0xff
@@ -128,6 +126,15 @@ static int make_tid_rdma_ack(struct rvt_qp *qp,
 			     struct ib_other_headers *ohdr,
 			     struct hfi1_pkt_state *ps);
 static void hfi1_do_tid_send(struct rvt_qp *qp);
+static u32 read_r_next_psn(struct hfi1_devdata *dd, u8 ctxt, u8 fidx);
+static void tid_rdma_rcv_err(struct hfi1_packet *packet,
+			     struct ib_other_headers *ohdr,
+			     struct rvt_qp *qp, u32 psn, int diff, bool fecn);
+static void update_r_next_psn_fecn(struct hfi1_packet *packet,
+				   struct hfi1_qp_priv *priv,
+				   struct hfi1_ctxtdata *rcd,
+				   struct tid_rdma_flow *flow,
+				   bool fecn);
 
 static u64 tid_rdma_opfn_encode(struct tid_rdma_params *p)
 {
@@ -776,7 +783,6 @@ int hfi1_kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp)
 		rcd->flows[fs->index].generation = fs->generation;
 	fs->generation = kern_setup_hw_flow(rcd, fs->index);
 	fs->psn = 0;
-	fs->flags = 0;
 	dequeue_tid_waiter(rcd, &rcd->flow_queue, qp);
 	/* get head before dropping lock */
 	fqp = first_qp(rcd, &rcd->flow_queue);
@@ -1807,6 +1813,7 @@ sync_check:
 			goto done;
 
 		hfi1_kern_clear_hw_flow(req->rcd, qp);
+		qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
 		req->state = TID_REQUEST_ACTIVE;
 	}
 
@@ -2036,10 +2043,7 @@ static int tid_rdma_rcv_error(struct hfi1_packet *packet,
 		if (psn != e->psn || len != req->total_len)
 			goto unlock;
 
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 
 		rkey = be32_to_cpu(reth->rkey);
 		vaddr = get_ib_reth_vaddr(reth);
@@ -2238,7 +2242,7 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 	struct ib_reth *reth;
 	struct hfi1_qp_priv *qpriv = qp->priv;
 	u32 bth0, psn, len, rkey;
-	bool is_fecn;
+	bool fecn;
 	u8 next;
 	u64 vaddr;
 	int diff;
@@ -2248,7 +2252,7 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 	if (hfi1_ruc_check_hdr(ibp, packet))
 		return;
 
-	is_fecn = process_ecn(qp, packet);
+	fecn = process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	trace_hfi1_rsp_rcv_tid_read_req(qp, psn);
 
@@ -2267,9 +2271,8 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 
 	diff = delta_psn(psn, qp->r_psn);
 	if (unlikely(diff)) {
-		if (tid_rdma_rcv_error(packet, ohdr, qp, psn, diff))
-			return;
-		goto send_ack;
+		tid_rdma_rcv_err(packet, ohdr, qp, psn, diff, fecn);
+		return;
 	}
 
 	/* We've verified the request, insert it into the ack queue. */
@@ -2285,10 +2288,7 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 		update_ack_queue(qp, next);
 	}
 	e = &qp->s_ack_queue[qp->r_head_ack_queue];
-	if (e->rdma_sge.mr) {
-		rvt_put_mr(e->rdma_sge.mr);
-		e->rdma_sge.mr = NULL;
-	}
+	release_rdma_sge_mr(e);
 
 	rkey = be32_to_cpu(reth->rkey);
 	qp->r_len = len;
@@ -2324,11 +2324,11 @@ void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
 
 	/* Schedule the send tasklet. */
 	qp->s_flags |= RVT_S_RESP_PENDING;
+	if (fecn)
+		qp->s_flags |= RVT_S_ECN;
 	hfi1_schedule_send(qp);
 
 	spin_unlock_irqrestore(&qp->s_lock, flags);
-	if (is_fecn)
-		goto send_ack;
 	return;
 
 nack_inv_unlock:
@@ -2345,8 +2345,6 @@ nack_acc:
 	rvt_rc_error(qp, IB_WC_LOC_PROT_ERR);
 	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
 	qp->r_ack_psn = qp->r_psn;
-send_ack:
-	hfi1_send_rc_ack(packet, is_fecn);
 }
 
 u32 hfi1_build_tid_rdma_read_resp(struct rvt_qp *qp, struct rvt_ack_entry *e,
@@ -2463,12 +2461,12 @@ void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet)
 	struct tid_rdma_request *req;
 	struct tid_rdma_flow *flow;
 	u32 opcode, aeth;
-	bool is_fecn;
+	bool fecn;
 	unsigned long flags;
 	u32 kpsn, ipsn;
 
 	trace_hfi1_sender_rcv_tid_read_resp(qp);
-	is_fecn = process_ecn(qp, packet);
+	fecn = process_ecn(qp, packet);
 	kpsn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	aeth = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.aeth);
 	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;
@@ -2481,8 +2479,43 @@ void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet)
 
 	flow = &req->flows[req->clear_tail];
 	/* When header suppression is disabled */
-	if (cmp_psn(ipsn, flow->flow_state.ib_lpsn))
+	if (cmp_psn(ipsn, flow->flow_state.ib_lpsn)) {
+		update_r_next_psn_fecn(packet, priv, rcd, flow, fecn);
+
+		if (cmp_psn(kpsn, flow->flow_state.r_next_psn))
+			goto ack_done;
+		flow->flow_state.r_next_psn = mask_psn(kpsn + 1);
+		/*
+		 * Copy the payload to destination buffer if this packet is
+		 * delivered as an eager packet due to RSM rule and FECN.
+		 * The RSM rule selects FECN bit in BTH and SH bit in
+		 * KDETH header and therefore will not match the last
+		 * packet of each segment that has SH bit cleared.
+		 */
+		if (fecn && packet->etype == RHF_RCV_TYPE_EAGER) {
+			struct rvt_sge_state ss;
+			u32 len;
+			u32 tlen = packet->tlen;
+			u16 hdrsize = packet->hlen;
+			u8 pad = packet->pad;
+			u8 extra_bytes = pad + packet->extra_byte +
+				(SIZE_OF_CRC << 2);
+			u32 pmtu = qp->pmtu;
+
+			if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
+				goto ack_op_err;
+			len = restart_sge(&ss, req->e.swqe, ipsn, pmtu);
+			if (unlikely(len < pmtu))
+				goto ack_op_err;
+			rvt_copy_sge(qp, &ss, packet->payload, pmtu, false,
+				     false);
+			/* Raise the sw sequence check flag for next packet */
+			priv->s_flags |= HFI1_R_TID_SW_PSN;
+		}
+
 		goto ack_done;
+	}
+	flow->flow_state.r_next_psn = mask_psn(kpsn + 1);
 	req->ack_pending--;
 	priv->pending_tid_r_segs--;
 	qp->s_num_rd_atomic--;
@@ -2524,6 +2557,7 @@ void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet)
 	     req->comp_seg == req->cur_seg) ||
 	    priv->tid_r_comp == priv->tid_r_reqs) {
 		hfi1_kern_clear_hw_flow(priv->rcd, qp);
+		priv->s_flags &= ~HFI1_R_TID_SW_PSN;
 		if (req->state == TID_REQUEST_SYNC)
 			req->state = TID_REQUEST_ACTIVE;
 	}
@@ -2545,8 +2579,6 @@ ack_op_err:
 
 ack_done:
 	spin_unlock_irqrestore(&qp->s_lock, flags);
-	if (is_fecn)
-		hfi1_send_rc_ack(packet, is_fecn);
 }
 
 void hfi1_kern_read_tid_flow_free(struct rvt_qp *qp)
@@ -2773,9 +2805,9 @@ static bool handle_read_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 				rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
 				return ret;
 			}
-			if (priv->flow_state.flags & TID_FLOW_SW_PSN) {
+			if (priv->s_flags & HFI1_R_TID_SW_PSN) {
 				diff = cmp_psn(psn,
-					       priv->flow_state.r_next_psn);
+					       flow->flow_state.r_next_psn);
 				if (diff > 0) {
 					if (!(qp->r_flags & RVT_R_RDMAR_SEQ))
 						restart_tid_rdma_read_req(rcd,
@@ -2811,22 +2843,15 @@ static bool handle_read_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 						qp->r_flags &=
 							~RVT_R_RDMAR_SEQ;
 				}
-				priv->flow_state.r_next_psn++;
+				flow->flow_state.r_next_psn =
+					mask_psn(psn + 1);
 			} else {
-				u64 reg;
 				u32 last_psn;
 
-				/*
-				 * The only sane way to get the amount of
-				 * progress is to read the HW flow state.
-				 */
-				reg = read_uctxt_csr(dd, rcd->ctxt,
-						     RCV_TID_FLOW_TABLE +
-						     (8 * flow->idx));
-				last_psn = mask_psn(reg);
-
-				priv->flow_state.r_next_psn = last_psn;
-				priv->flow_state.flags |= TID_FLOW_SW_PSN;
+				last_psn = read_r_next_psn(dd, rcd->ctxt,
+							   flow->idx);
+				flow->flow_state.r_next_psn = last_psn;
+				priv->s_flags |= HFI1_R_TID_SW_PSN;
 				/*
 				 * If no request has been restarted yet,
 				 * restart the current one.
@@ -2891,10 +2916,11 @@ bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 	struct rvt_ack_entry *e;
 	struct tid_rdma_request *req;
 	struct tid_rdma_flow *flow;
+	int diff = 0;
 
 	trace_hfi1_msg_handle_kdeth_eflags(NULL, "Kdeth error: rhf ",
 					   packet->rhf);
-	if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
+	if (packet->rhf & RHF_ICRC_ERR)
 		return ret;
 
 	packet->ohdr = &hdr->u.oth;
@@ -2974,17 +3000,10 @@ bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 		switch (rte) {
 		case RHF_RTE_EXPECTED_FLOW_SEQ_ERR:
 			if (!(qpriv->s_flags & HFI1_R_TID_SW_PSN)) {
-				u64 reg;
-
 				qpriv->s_flags |= HFI1_R_TID_SW_PSN;
-				/*
-				 * The only sane way to get the amount of
-				 * progress is to read the HW flow state.
-				 */
-				reg = read_uctxt_csr(dd, rcd->ctxt,
-						     RCV_TID_FLOW_TABLE +
-						     (8 * flow->idx));
-				flow->flow_state.r_next_psn = mask_psn(reg);
+				flow->flow_state.r_next_psn =
+					read_r_next_psn(dd, rcd->ctxt,
+							flow->idx);
 				qpriv->r_next_psn_kdeth =
 					flow->flow_state.r_next_psn;
 				goto nak_psn;
@@ -2997,10 +3016,12 @@ bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 				 * mismatch could be due to packets that were
 				 * already in flight.
 				 */
-				if (psn != flow->flow_state.r_next_psn) {
-					psn = flow->flow_state.r_next_psn;
+				diff = cmp_psn(psn,
+					       flow->flow_state.r_next_psn);
+				if (diff > 0)
 					goto nak_psn;
-				}
+				else if (diff < 0)
+					break;
 
 				qpriv->s_nak_state = 0;
 				/*
@@ -3011,8 +3032,10 @@ bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd,
 				if (psn == full_flow_psn(flow,
 							 flow->flow_state.lpsn))
 					ret = false;
+				flow->flow_state.r_next_psn =
+					mask_psn(psn + 1);
 				qpriv->r_next_psn_kdeth =
-					++flow->flow_state.r_next_psn;
+					flow->flow_state.r_next_psn;
 			}
 			break;
 
@@ -3517,8 +3540,10 @@ static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx)
 		if (qpriv->r_tid_alloc == qpriv->r_tid_head) {
 			/* If all data has been received, clear the flow */
 			if (qpriv->flow_state.index < RXE_NUM_TID_FLOWS &&
-			    !qpriv->alloc_w_segs)
+			    !qpriv->alloc_w_segs) {
 				hfi1_kern_clear_hw_flow(rcd, qp);
+				qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
+			}
 			break;
 		}
 
@@ -3544,8 +3569,7 @@ static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx)
 		if (qpriv->sync_pt && !qpriv->alloc_w_segs) {
 			hfi1_kern_clear_hw_flow(rcd, qp);
 			qpriv->sync_pt = false;
-			if (qpriv->s_flags & HFI1_R_TID_SW_PSN)
-				qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
+			qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
 		}
 
 		/* Allocate flow if we don't have one */
@@ -3687,7 +3711,7 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
 	struct hfi1_qp_priv *qpriv = qp->priv;
 	struct tid_rdma_request *req;
 	u32 bth0, psn, len, rkey, num_segs;
-	bool is_fecn;
+	bool fecn;
 	u8 next;
 	u64 vaddr;
 	int diff;
@@ -3696,7 +3720,7 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
 	if (hfi1_ruc_check_hdr(ibp, packet))
 		return;
 
-	is_fecn = process_ecn(qp, packet);
+	fecn = process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	trace_hfi1_rsp_rcv_tid_write_req(qp, psn);
 
@@ -3713,9 +3737,8 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
 	num_segs = DIV_ROUND_UP(len, qpriv->tid_rdma.local.max_len);
 	diff = delta_psn(psn, qp->r_psn);
 	if (unlikely(diff)) {
-		if (tid_rdma_rcv_error(packet, ohdr, qp, psn, diff))
-			return;
-		goto send_ack;
+		tid_rdma_rcv_err(packet, ohdr, qp, psn, diff, fecn);
+		return;
 	}
 
 	/*
@@ -3751,10 +3774,7 @@ void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
 		goto update_head;
 	}
 
-	if (e->rdma_sge.mr) {
-		rvt_put_mr(e->rdma_sge.mr);
-		e->rdma_sge.mr = NULL;
-	}
+	release_rdma_sge_mr(e);
 
 	/* The length needs to be in multiples of PAGE_SIZE */
 	if (!len || len & ~PAGE_MASK)
@@ -3834,11 +3854,11 @@ update_head:
 
 	/* Schedule the send tasklet. */
 	qp->s_flags |= RVT_S_RESP_PENDING;
+	if (fecn)
+		qp->s_flags |= RVT_S_ECN;
 	hfi1_schedule_send(qp);
 
 	spin_unlock_irqrestore(&qp->s_lock, flags);
-	if (is_fecn)
-		goto send_ack;
 	return;
 
 nack_inv_unlock:
@@ -3855,8 +3875,6 @@ nack_acc:
 	rvt_rc_error(qp, IB_WC_LOC_PROT_ERR);
 	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
 	qp->r_ack_psn = qp->r_psn;
-send_ack:
-	hfi1_send_rc_ack(packet, is_fecn);
 }
 
 u32 hfi1_build_tid_rdma_write_resp(struct rvt_qp *qp, struct rvt_ack_entry *e,
@@ -4073,10 +4091,10 @@ void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet)
 	struct tid_rdma_flow *flow;
 	enum ib_wc_status status;
 	u32 opcode, aeth, psn, flow_psn, i, tidlen = 0, pktlen;
-	bool is_fecn;
+	bool fecn;
 	unsigned long flags;
 
-	is_fecn = process_ecn(qp, packet);
+	fecn = process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	aeth = be32_to_cpu(ohdr->u.tid_rdma.w_rsp.aeth);
 	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;
@@ -4216,7 +4234,6 @@ void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet)
 		qpriv->s_tid_cur = i;
 	}
 	qp->s_flags &= ~HFI1_S_WAIT_TID_RESP;
-
 	hfi1_schedule_tid_send(qp);
 	goto ack_done;
 
@@ -4225,9 +4242,9 @@ ack_op_err:
 ack_err:
 	rvt_error_qp(qp, status);
 ack_done:
+	if (fecn)
+		qp->s_flags |= RVT_S_ECN;
 	spin_unlock_irqrestore(&qp->s_lock, flags);
-	if (is_fecn)
-		hfi1_send_rc_ack(packet, is_fecn);
 }
 
 bool hfi1_build_tid_rdma_packet(struct rvt_swqe *wqe,
@@ -4307,7 +4324,9 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet)
 	unsigned long flags;
 	u32 psn, next;
 	u8 opcode;
+	bool fecn;
 
+	fecn = process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;
 
@@ -4320,9 +4339,53 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet)
 	req = ack_to_tid_req(e);
 	flow = &req->flows[req->clear_tail];
 	if (cmp_psn(psn, full_flow_psn(flow, flow->flow_state.lpsn))) {
+		update_r_next_psn_fecn(packet, priv, rcd, flow, fecn);
+
 		if (cmp_psn(psn, flow->flow_state.r_next_psn))
 			goto send_nak;
-		flow->flow_state.r_next_psn++;
+
+		flow->flow_state.r_next_psn = mask_psn(psn + 1);
+		/*
+		 * Copy the payload to destination buffer if this packet is
+		 * delivered as an eager packet due to RSM rule and FECN.
+		 * The RSM rule selects FECN bit in BTH and SH bit in
+		 * KDETH header and therefore will not match the last
+		 * packet of each segment that has SH bit cleared.
+		 */
+		if (fecn && packet->etype == RHF_RCV_TYPE_EAGER) {
+			struct rvt_sge_state ss;
+			u32 len;
+			u32 tlen = packet->tlen;
+			u16 hdrsize = packet->hlen;
+			u8 pad = packet->pad;
+			u8 extra_bytes = pad + packet->extra_byte +
+				(SIZE_OF_CRC << 2);
+			u32 pmtu = qp->pmtu;
+
+			if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
+				goto send_nak;
+			len = req->comp_seg * req->seg_len;
+			len += delta_psn(psn,
+				full_flow_psn(flow, flow->flow_state.spsn)) *
+				pmtu;
+			if (unlikely(req->total_len - len < pmtu))
+				goto send_nak;
+
+			/*
+			 * The e->rdma_sge field is set when TID RDMA WRITE REQ
+			 * is first received and is never modified thereafter.
+			 */
+			ss.sge = e->rdma_sge;
+			ss.sg_list = NULL;
+			ss.num_sge = 1;
+			ss.total_len = req->total_len;
+			rvt_skip_sge(&ss, len, false);
+			rvt_copy_sge(qp, &ss, packet->payload, pmtu, false,
+				     false);
+			/* Raise the sw sequence check flag for next packet */
+			priv->r_next_psn_kdeth = mask_psn(psn + 1);
+			priv->s_flags |= HFI1_R_TID_SW_PSN;
+		}
 		goto exit;
 	}
 	flow->flow_state.r_next_psn = mask_psn(psn + 1);
@@ -4347,6 +4410,7 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet)
 		priv->r_tid_ack = priv->r_tid_tail;
 
 	if (opcode == TID_OP(WRITE_DATA_LAST)) {
+		release_rdma_sge_mr(e);
 		for (next = priv->r_tid_tail + 1; ; next++) {
 			if (next > rvt_size_atomic(&dev->rdi))
 				next = 0;
@@ -4386,6 +4450,8 @@ done:
 	hfi1_schedule_tid_send(qp);
 exit:
 	priv->r_next_psn_kdeth = flow->flow_state.r_next_psn;
+	if (fecn)
+		qp->s_flags |= RVT_S_ECN;
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 	return;
 
@@ -4487,12 +4553,11 @@ void hfi1_rc_rcv_tid_rdma_ack(struct hfi1_packet *packet)
 	struct tid_rdma_request *req;
 	struct tid_rdma_flow *flow;
 	u32 aeth, psn, req_psn, ack_psn, fspsn, resync_psn, ack_kpsn;
-	bool is_fecn;
 	unsigned long flags;
 	u16 fidx;
 
 	trace_hfi1_tid_write_sender_rcv_tid_ack(qp, 0);
-	is_fecn = process_ecn(qp, packet);
+	process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 	aeth = be32_to_cpu(ohdr->u.tid_rdma.ack.aeth);
 	req_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.ack.verbs_psn));
@@ -4846,10 +4911,10 @@ void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet)
 	struct tid_rdma_flow *flow;
 	struct tid_flow_state *fs = &qpriv->flow_state;
 	u32 psn, generation, idx, gen_next;
-	bool is_fecn;
+	bool fecn;
 	unsigned long flags;
 
-	is_fecn = process_ecn(qp, packet);
+	fecn = process_ecn(qp, packet);
 	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
 
 	generation = mask_psn(psn + 1) >> HFI1_KDETH_BTH_SEQ_SHIFT;
@@ -4940,6 +5005,8 @@ void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet)
 	qpriv->s_flags |= RVT_S_ACK_PENDING;
 	hfi1_schedule_tid_send(qp);
 bail:
+	if (fecn)
+		qp->s_flags |= RVT_S_ECN;
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 }
 
@@ -5449,3 +5516,48 @@ bool hfi1_tid_rdma_ack_interlock(struct rvt_qp *qp, struct rvt_ack_entry *e)
 	}
 	return false;
 }
+
+static u32 read_r_next_psn(struct hfi1_devdata *dd, u8 ctxt, u8 fidx)
+{
+	u64 reg;
+
+	/*
+	 * The only sane way to get the amount of
+	 * progress is to read the HW flow state.
+	 */
+	reg = read_uctxt_csr(dd, ctxt, RCV_TID_FLOW_TABLE + (8 * fidx));
+	return mask_psn(reg);
+}
+
+static void tid_rdma_rcv_err(struct hfi1_packet *packet,
+			     struct ib_other_headers *ohdr,
+			     struct rvt_qp *qp, u32 psn, int diff, bool fecn)
+{
+	unsigned long flags;
+
+	tid_rdma_rcv_error(packet, ohdr, qp, psn, diff);
+	if (fecn) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qp->s_flags |= RVT_S_ECN;
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	}
+}
+
+static void update_r_next_psn_fecn(struct hfi1_packet *packet,
+				   struct hfi1_qp_priv *priv,
+				   struct hfi1_ctxtdata *rcd,
+				   struct tid_rdma_flow *flow,
+				   bool fecn)
+{
+	/*
+	 * If a start/middle packet is delivered here due to
+	 * RSM rule and FECN, we need to update the r_next_psn.
+	 */
+	if (fecn && packet->etype == RHF_RCV_TYPE_EAGER &&
+	    !(priv->s_flags & HFI1_R_TID_SW_PSN)) {
+		struct hfi1_devdata *dd = rcd->dd;
+
+		flow->flow_state.r_next_psn =
+			read_r_next_psn(dd, rcd->ctxt, flow->idx);
+	}
+}
diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.h b/drivers/infiniband/hw/hfi1/tid_rdma.h
index 53ab24ef4f02..1c536185261e 100644
--- a/drivers/infiniband/hw/hfi1/tid_rdma.h
+++ b/drivers/infiniband/hw/hfi1/tid_rdma.h
@@ -76,10 +76,8 @@ struct tid_rdma_qp_params {
 struct tid_flow_state {
 	u32 generation;
 	u32 psn;
-	u32 r_next_psn;      /* next PSN to be received (in TID space) */
 	u8 index;
 	u8 last_index;
-	u8 flags;
 };
 
 enum tid_rdma_req_state {
diff --git a/drivers/infiniband/hw/hfi1/trace_dbg.h b/drivers/infiniband/hw/hfi1/trace_dbg.h
index e62171fb7379..de7a87392b8d 100644
--- a/drivers/infiniband/hw/hfi1/trace_dbg.h
+++ b/drivers/infiniband/hw/hfi1/trace_dbg.h
@@ -86,14 +86,14 @@ DECLARE_EVENT_CLASS(hfi1_trace_template,
  * actual function to work and can not be in a macro.
  */
 #define __hfi1_trace_def(lvl) \
-void __hfi1_trace_##lvl(const char *funct, char *fmt, ...);		\
+void __printf(2, 3) __hfi1_trace_##lvl(const char *funct, char *fmt, ...); \
 									\
 DEFINE_EVENT(hfi1_trace_template, hfi1_ ##lvl,				\
 	TP_PROTO(const char *function, struct va_format *vaf),		\
 	TP_ARGS(function, vaf))
 
 #define __hfi1_trace_fn(lvl) \
-void __hfi1_trace_##lvl(const char *func, char *fmt, ...)		\
+void __printf(2, 3) __hfi1_trace_##lvl(const char *func, char *fmt, ...)\
 {									\
 	struct va_format vaf = {					\
 		.fmt = fmt,						\
diff --git a/drivers/infiniband/hw/hfi1/trace_tid.h b/drivers/infiniband/hw/hfi1/trace_tid.h
index 548dfc45a407..4388b594ed1b 100644
--- a/drivers/infiniband/hw/hfi1/trace_tid.h
+++ b/drivers/infiniband/hw/hfi1/trace_tid.h
@@ -53,7 +53,7 @@ u16 hfi1_trace_get_tid_idx(u32 ent);
 			    "tid_r_comp %u pending_tid_r_segs %u " \
 			    "s_flags 0x%x ps_flags 0x%x iow_flags 0x%lx " \
 			    "s_state 0x%x hw_flow_index %u generation 0x%x " \
-			    "fpsn 0x%x flow_flags 0x%x"
+			    "fpsn 0x%x"
 
 #define TID_REQ_PRN "[%s] qpn 0x%x newreq %u opcode 0x%x psn 0x%x lpsn 0x%x " \
 		    "cur_seg %u comp_seg %u ack_seg %u alloc_seg %u " \
@@ -71,7 +71,7 @@ u16 hfi1_trace_get_tid_idx(u32 ent);
 			    "pending_tid_w_segs %u sync_pt %s " \
 			    "ps_nak_psn 0x%x ps_nak_state 0x%x " \
 			    "prnr_nak_state 0x%x hw_flow_index %u generation "\
-			    "0x%x fpsn 0x%x flow_flags 0x%x resync %s" \
+			    "0x%x fpsn 0x%x resync %s" \
 			    "r_next_psn_kdeth 0x%x"
 
 #define TID_WRITE_SENDER_PRN "[%s] qpn 0x%x newreq %u s_tid_cur %u " \
@@ -973,7 +973,6 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__field(u32, hw_flow_index)
 		__field(u32, generation)
 		__field(u32, fpsn)
-		__field(u32, flow_flags)
 	),
 	TP_fast_assign(/* assign */
 		struct hfi1_qp_priv *priv = qp->priv;
@@ -991,7 +990,6 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__entry->hw_flow_index = priv->flow_state.index;
 		__entry->generation = priv->flow_state.generation;
 		__entry->fpsn = priv->flow_state.psn;
-		__entry->flow_flags = priv->flow_state.flags;
 	),
 	TP_printk(/* print */
 		TID_READ_SENDER_PRN,
@@ -1007,8 +1005,7 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__entry->s_state,
 		__entry->hw_flow_index,
 		__entry->generation,
-		__entry->fpsn,
-		__entry->flow_flags
+		__entry->fpsn
 	)
 );
 
@@ -1338,7 +1335,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__field(u32, hw_flow_index)
 		__field(u32, generation)
 		__field(u32, fpsn)
-		__field(u32, flow_flags)
 		__field(bool, resync)
 		__field(u32, r_next_psn_kdeth)
 	),
@@ -1360,7 +1356,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__entry->hw_flow_index = priv->flow_state.index;
 		__entry->generation = priv->flow_state.generation;
 		__entry->fpsn = priv->flow_state.psn;
-		__entry->flow_flags = priv->flow_state.flags;
 		__entry->resync = priv->resync;
 		__entry->r_next_psn_kdeth = priv->r_next_psn_kdeth;
 	),
@@ -1381,7 +1376,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__entry->hw_flow_index,
 		__entry->generation,
 		__entry->fpsn,
-		__entry->flow_flags,
 		__entry->resync ? "yes" : "no",
 		__entry->r_next_psn_kdeth
 	)
diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c
index 55a56b3d7f83..1eb4105b2d22 100644
--- a/drivers/infiniband/hw/hfi1/verbs.c
+++ b/drivers/infiniband/hw/hfi1/verbs.c
@@ -1223,15 +1223,16 @@ static inline send_routine get_send_routine(struct rvt_qp *qp,
 	case IB_QPT_UD:
 		break;
 	case IB_QPT_UC:
-	case IB_QPT_RC: {
+	case IB_QPT_RC:
+		priv->s_running_pkt_size =
+			(tx->s_cur_size + priv->s_running_pkt_size) / 2;
 		if (piothreshold &&
-		    tx->s_cur_size <= min(piothreshold, qp->pmtu) &&
+		    priv->s_running_pkt_size <= min(piothreshold, qp->pmtu) &&
 		    (BIT(ps->opcode & OPMASK) & pio_opmask[ps->opcode >> 5]) &&
 		    iowait_sdma_pending(&priv->s_iowait) == 0 &&
 		    !sdma_txreq_built(&tx->txreq))
 			return dd->process_pio_send;
 		break;
-	}
 	default:
 		break;
 	}
@@ -1739,15 +1740,15 @@ static struct rdma_hw_stats *alloc_hw_stats(struct ib_device *ibdev,
 
 static u64 hfi1_sps_ints(void)
 {
-	unsigned long flags;
+	unsigned long index, flags;
 	struct hfi1_devdata *dd;
 	u64 sps_ints = 0;
 
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-	list_for_each_entry(dd, &hfi1_dev_list, list) {
+	xa_lock_irqsave(&hfi1_dev_table, flags);
+	xa_for_each(&hfi1_dev_table, index, dd) {
 		sps_ints += get_all_cpu_total(dd->int_counter);
 	}
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
+	xa_unlock_irqrestore(&hfi1_dev_table, flags);
 	return sps_ints;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 62ace0b2d17a..7ecb8ed4a1d9 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -170,6 +170,7 @@ struct hfi1_qp_priv {
 	struct tid_flow_state flow_state;
 	struct tid_rdma_qp_params tid_rdma;
 	struct rvt_qp *owner;
+	u16 s_running_pkt_size;
 	u8 hdr_type; /* 9B or 16B */
 	struct rvt_sge_state tid_ss;       /* SGE state pointer for 2nd leg */
 	atomic_t n_requests;               /* # of TID RDMA requests in the */
diff --git a/drivers/infiniband/hw/hfi1/vnic_main.c b/drivers/infiniband/hw/hfi1/vnic_main.c
index 2b07032dbdda..b49e60e8397d 100644
--- a/drivers/infiniband/hw/hfi1/vnic_main.c
+++ b/drivers/infiniband/hw/hfi1/vnic_main.c
@@ -162,12 +162,12 @@ static void deallocate_vnic_ctxt(struct hfi1_devdata *dd,
 
 void hfi1_vnic_setup(struct hfi1_devdata *dd)
 {
-	idr_init(&dd->vnic.vesw_idr);
+	xa_init(&dd->vnic.vesws);
 }
 
 void hfi1_vnic_cleanup(struct hfi1_devdata *dd)
 {
-	idr_destroy(&dd->vnic.vesw_idr);
+	WARN_ON(!xa_empty(&dd->vnic.vesws));
 }
 
 #define SUM_GRP_COUNTERS(stats, qstats, x_grp) do {            \
@@ -533,7 +533,7 @@ void hfi1_vnic_bypass_rcv(struct hfi1_packet *packet)
 	l4_type = hfi1_16B_get_l4(packet->ebuf);
 	if (likely(l4_type == OPA_16B_L4_ETHR)) {
 		vesw_id = HFI1_VNIC_GET_VESWID(packet->ebuf);
-		vinfo = idr_find(&dd->vnic.vesw_idr, vesw_id);
+		vinfo = xa_load(&dd->vnic.vesws, vesw_id);
 
 		/*
 		 * In case of invalid vesw id, count the error on
@@ -541,9 +541,10 @@ void hfi1_vnic_bypass_rcv(struct hfi1_packet *packet)
 		 */
 		if (unlikely(!vinfo)) {
 			struct hfi1_vnic_vport_info *vinfo_tmp;
-			int id_tmp = 0;
+			unsigned long index = 0;
 
-			vinfo_tmp =  idr_get_next(&dd->vnic.vesw_idr, &id_tmp);
+			vinfo_tmp = xa_find(&dd->vnic.vesws, &index, ULONG_MAX,
+					XA_PRESENT);
 			if (vinfo_tmp) {
 				spin_lock(&vport_cntr_lock);
 				vinfo_tmp->stats[0].netstats.rx_nohandler++;
@@ -597,8 +598,7 @@ static int hfi1_vnic_up(struct hfi1_vnic_vport_info *vinfo)
 	if (!vinfo->vesw_id)
 		return -EINVAL;
 
-	rc = idr_alloc(&dd->vnic.vesw_idr, vinfo, vinfo->vesw_id,
-		       vinfo->vesw_id + 1, GFP_NOWAIT);
+	rc = xa_insert(&dd->vnic.vesws, vinfo->vesw_id, vinfo, GFP_KERNEL);
 	if (rc < 0)
 		return rc;
 
@@ -624,7 +624,7 @@ static void hfi1_vnic_down(struct hfi1_vnic_vport_info *vinfo)
 	clear_bit(HFI1_VNIC_UP, &vinfo->flags);
 	netif_carrier_off(vinfo->netdev);
 	netif_tx_disable(vinfo->netdev);
-	idr_remove(&dd->vnic.vesw_idr, vinfo->vesw_id);
+	xa_erase(&dd->vnic.vesws, vinfo->vesw_id);
 
 	/* ensure irqs see the change */
 	msix_vnic_synchronize_irq(dd);