summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/sw/rdmavt
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-11 08:27:49 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-11 08:27:49 +0000
commitace9429bb58fd418f0c81d4c2835699bddf6bde6 (patch)
treeb2d64bc10158fdd5497876388cd68142ca374ed3 /drivers/infiniband/sw/rdmavt
parentInitial commit. (diff)
downloadlinux-ace9429bb58fd418f0c81d4c2835699bddf6bde6.tar.xz
linux-ace9429bb58fd418f0c81d4c2835699bddf6bde6.zip
Adding upstream version 6.6.15.upstream/6.6.15
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'drivers/infiniband/sw/rdmavt')
-rw-r--r--drivers/infiniband/sw/rdmavt/Kconfig8
-rw-r--r--drivers/infiniband/sw/rdmavt/Makefile14
-rw-r--r--drivers/infiniband/sw/rdmavt/ah.c139
-rw-r--r--drivers/infiniband/sw/rdmavt/ah.h17
-rw-r--r--drivers/infiniband/sw/rdmavt/cq.c534
-rw-r--r--drivers/infiniband/sw/rdmavt/cq.h20
-rw-r--r--drivers/infiniband/sw/rdmavt/mad.c130
-rw-r--r--drivers/infiniband/sw/rdmavt/mad.h18
-rw-r--r--drivers/infiniband/sw/rdmavt/mcast.c401
-rw-r--r--drivers/infiniband/sw/rdmavt/mcast.h16
-rw-r--r--drivers/infiniband/sw/rdmavt/mmap.c169
-rw-r--r--drivers/infiniband/sw/rdmavt/mmap.h19
-rw-r--r--drivers/infiniband/sw/rdmavt/mr.c917
-rw-r--r--drivers/infiniband/sw/rdmavt/mr.h36
-rw-r--r--drivers/infiniband/sw/rdmavt/pd.c64
-rw-r--r--drivers/infiniband/sw/rdmavt/pd.h14
-rw-r--r--drivers/infiniband/sw/rdmavt/qp.c3220
-rw-r--r--drivers/infiniband/sw/rdmavt/qp.h30
-rw-r--r--drivers/infiniband/sw/rdmavt/rc.c172
-rw-r--r--drivers/infiniband/sw/rdmavt/srq.c306
-rw-r--r--drivers/infiniband/sw/rdmavt/srq.h19
-rw-r--r--drivers/infiniband/sw/rdmavt/trace.c7
-rw-r--r--drivers/infiniband/sw/rdmavt/trace.h14
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_cq.h124
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_mr.h182
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_qp.h96
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_rc.h67
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_rvt.h39
-rw-r--r--drivers/infiniband/sw/rdmavt/trace_tx.h163
-rw-r--r--drivers/infiniband/sw/rdmavt/vt.c619
-rw-r--r--drivers/infiniband/sw/rdmavt/vt.h62
31 files changed, 7636 insertions, 0 deletions
diff --git a/drivers/infiniband/sw/rdmavt/Kconfig b/drivers/infiniband/sw/rdmavt/Kconfig
new file mode 100644
index 0000000000..0df48b3a6b
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/Kconfig
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config INFINIBAND_RDMAVT
+ tristate "RDMA verbs transport library"
+ depends on INFINIBAND_VIRT_DMA
+ depends on X86_64
+ depends on PCI
+ help
+ This is a common software verbs provider for RDMA networks.
diff --git a/drivers/infiniband/sw/rdmavt/Makefile b/drivers/infiniband/sw/rdmavt/Makefile
new file mode 100644
index 0000000000..b21962dafc
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/Makefile
@@ -0,0 +1,14 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# rdmavt driver
+#
+#
+#
+# Called from the kernel module build system.
+#
+obj-$(CONFIG_INFINIBAND_RDMAVT) += rdmavt.o
+
+rdmavt-y := vt.o ah.o cq.o mad.o mcast.o mmap.o mr.o pd.o qp.o \
+ rc.o srq.o trace.o
+
+CFLAGS_trace.o = -I$(src)
diff --git a/drivers/infiniband/sw/rdmavt/ah.c b/drivers/infiniband/sw/rdmavt/ah.c
new file mode 100644
index 0000000000..63999239ed
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/ah.c
@@ -0,0 +1,139 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 - 2019 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include "ah.h"
+#include "vt.h" /* for prints */
+
+/**
+ * rvt_check_ah - validate the attributes of AH
+ * @ibdev: the ib device
+ * @ah_attr: the attributes of the AH
+ *
+ * If driver supports a more detailed check_ah function call back to it
+ * otherwise just check the basics.
+ *
+ * Return: 0 on success
+ */
+int rvt_check_ah(struct ib_device *ibdev,
+ struct rdma_ah_attr *ah_attr)
+{
+ int err;
+ int port_num = rdma_ah_get_port_num(ah_attr);
+ struct ib_port_attr port_attr;
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ u8 ah_flags = rdma_ah_get_ah_flags(ah_attr);
+ u8 static_rate = rdma_ah_get_static_rate(ah_attr);
+
+ err = ib_query_port(ibdev, port_num, &port_attr);
+ if (err)
+ return -EINVAL;
+ if (port_num < 1 ||
+ port_num > ibdev->phys_port_cnt)
+ return -EINVAL;
+ if (static_rate != IB_RATE_PORT_CURRENT &&
+ ib_rate_to_mbps(static_rate) < 0)
+ return -EINVAL;
+ if ((ah_flags & IB_AH_GRH) &&
+ rdma_ah_read_grh(ah_attr)->sgid_index >= port_attr.gid_tbl_len)
+ return -EINVAL;
+ if (rdi->driver_f.check_ah)
+ return rdi->driver_f.check_ah(ibdev, ah_attr);
+ return 0;
+}
+EXPORT_SYMBOL(rvt_check_ah);
+
+/**
+ * rvt_create_ah - create an address handle
+ * @ibah: the IB address handle
+ * @init_attr: the attributes of the AH
+ * @udata: pointer to user's input output buffer information.
+ *
+ * This may be called from interrupt context.
+ *
+ * Return: 0 on success
+ */
+int rvt_create_ah(struct ib_ah *ibah, struct rdma_ah_init_attr *init_attr,
+ struct ib_udata *udata)
+{
+ struct rvt_ah *ah = ibah_to_rvtah(ibah);
+ struct rvt_dev_info *dev = ib_to_rvt(ibah->device);
+ unsigned long flags;
+
+ if (rvt_check_ah(ibah->device, init_attr->ah_attr))
+ return -EINVAL;
+
+ spin_lock_irqsave(&dev->n_ahs_lock, flags);
+ if (dev->n_ahs_allocated == dev->dparms.props.max_ah) {
+ spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+ return -ENOMEM;
+ }
+
+ dev->n_ahs_allocated++;
+ spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+
+ rdma_copy_ah_attr(&ah->attr, init_attr->ah_attr);
+
+ if (dev->driver_f.notify_new_ah)
+ dev->driver_f.notify_new_ah(ibah->device,
+ init_attr->ah_attr, ah);
+
+ return 0;
+}
+
+/**
+ * rvt_destroy_ah - Destroy an address handle
+ * @ibah: address handle
+ * @destroy_flags: destroy address handle flags (see enum rdma_destroy_ah_flags)
+ * Return: 0 on success
+ */
+int rvt_destroy_ah(struct ib_ah *ibah, u32 destroy_flags)
+{
+ struct rvt_dev_info *dev = ib_to_rvt(ibah->device);
+ struct rvt_ah *ah = ibah_to_rvtah(ibah);
+ unsigned long flags;
+
+ spin_lock_irqsave(&dev->n_ahs_lock, flags);
+ dev->n_ahs_allocated--;
+ spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+
+ rdma_destroy_ah_attr(&ah->attr);
+ return 0;
+}
+
+/**
+ * rvt_modify_ah - modify an ah with given attrs
+ * @ibah: address handle to modify
+ * @ah_attr: attrs to apply
+ *
+ * Return: 0 on success
+ */
+int rvt_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr)
+{
+ struct rvt_ah *ah = ibah_to_rvtah(ibah);
+
+ if (rvt_check_ah(ibah->device, ah_attr))
+ return -EINVAL;
+
+ ah->attr = *ah_attr;
+
+ return 0;
+}
+
+/**
+ * rvt_query_ah - return attrs for ah
+ * @ibah: address handle to query
+ * @ah_attr: return info in this
+ *
+ * Return: always 0
+ */
+int rvt_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr)
+{
+ struct rvt_ah *ah = ibah_to_rvtah(ibah);
+
+ *ah_attr = ah->attr;
+
+ return 0;
+}
diff --git a/drivers/infiniband/sw/rdmavt/ah.h b/drivers/infiniband/sw/rdmavt/ah.h
new file mode 100644
index 0000000000..c11fdf637d
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/ah.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTAH_H
+#define DEF_RVTAH_H
+
+#include <rdma/rdma_vt.h>
+
+int rvt_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
+ struct ib_udata *udata);
+int rvt_destroy_ah(struct ib_ah *ibah, u32 destroy_flags);
+int rvt_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
+int rvt_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
+
+#endif /* DEF_RVTAH_H */
diff --git a/drivers/infiniband/sw/rdmavt/cq.c b/drivers/infiniband/sw/rdmavt/cq.c
new file mode 100644
index 0000000000..9fe4dcaa04
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/cq.c
@@ -0,0 +1,534 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 - 2018 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include "cq.h"
+#include "vt.h"
+#include "trace.h"
+
+static struct workqueue_struct *comp_vector_wq;
+
+/**
+ * rvt_cq_enter - add a new entry to the completion queue
+ * @cq: completion queue
+ * @entry: work completion entry to add
+ * @solicited: true if @entry is solicited
+ *
+ * This may be called with qp->s_lock held.
+ *
+ * Return: return true on success, else return
+ * false if cq is full.
+ */
+bool rvt_cq_enter(struct rvt_cq *cq, struct ib_wc *entry, bool solicited)
+{
+ struct ib_uverbs_wc *uqueue = NULL;
+ struct ib_wc *kqueue = NULL;
+ struct rvt_cq_wc *u_wc = NULL;
+ struct rvt_k_cq_wc *k_wc = NULL;
+ unsigned long flags;
+ u32 head;
+ u32 next;
+ u32 tail;
+
+ spin_lock_irqsave(&cq->lock, flags);
+
+ if (cq->ip) {
+ u_wc = cq->queue;
+ uqueue = &u_wc->uqueue[0];
+ head = RDMA_READ_UAPI_ATOMIC(u_wc->head);
+ tail = RDMA_READ_UAPI_ATOMIC(u_wc->tail);
+ } else {
+ k_wc = cq->kqueue;
+ kqueue = &k_wc->kqueue[0];
+ head = k_wc->head;
+ tail = k_wc->tail;
+ }
+
+ /*
+ * Note that the head pointer might be writable by
+ * user processes.Take care to verify it is a sane value.
+ */
+ if (head >= (unsigned)cq->ibcq.cqe) {
+ head = cq->ibcq.cqe;
+ next = 0;
+ } else {
+ next = head + 1;
+ }
+
+ if (unlikely(next == tail || cq->cq_full)) {
+ struct rvt_dev_info *rdi = cq->rdi;
+
+ if (!cq->cq_full)
+ rvt_pr_err_ratelimited(rdi, "CQ is full!\n");
+ cq->cq_full = true;
+ spin_unlock_irqrestore(&cq->lock, flags);
+ if (cq->ibcq.event_handler) {
+ struct ib_event ev;
+
+ ev.device = cq->ibcq.device;
+ ev.element.cq = &cq->ibcq;
+ ev.event = IB_EVENT_CQ_ERR;
+ cq->ibcq.event_handler(&ev, cq->ibcq.cq_context);
+ }
+ return false;
+ }
+ trace_rvt_cq_enter(cq, entry, head);
+ if (uqueue) {
+ uqueue[head].wr_id = entry->wr_id;
+ uqueue[head].status = entry->status;
+ uqueue[head].opcode = entry->opcode;
+ uqueue[head].vendor_err = entry->vendor_err;
+ uqueue[head].byte_len = entry->byte_len;
+ uqueue[head].ex.imm_data = entry->ex.imm_data;
+ uqueue[head].qp_num = entry->qp->qp_num;
+ uqueue[head].src_qp = entry->src_qp;
+ uqueue[head].wc_flags = entry->wc_flags;
+ uqueue[head].pkey_index = entry->pkey_index;
+ uqueue[head].slid = ib_lid_cpu16(entry->slid);
+ uqueue[head].sl = entry->sl;
+ uqueue[head].dlid_path_bits = entry->dlid_path_bits;
+ uqueue[head].port_num = entry->port_num;
+ /* Make sure entry is written before the head index. */
+ RDMA_WRITE_UAPI_ATOMIC(u_wc->head, next);
+ } else {
+ kqueue[head] = *entry;
+ k_wc->head = next;
+ }
+
+ if (cq->notify == IB_CQ_NEXT_COMP ||
+ (cq->notify == IB_CQ_SOLICITED &&
+ (solicited || entry->status != IB_WC_SUCCESS))) {
+ /*
+ * This will cause send_complete() to be called in
+ * another thread.
+ */
+ cq->notify = RVT_CQ_NONE;
+ cq->triggered++;
+ queue_work_on(cq->comp_vector_cpu, comp_vector_wq,
+ &cq->comptask);
+ }
+
+ spin_unlock_irqrestore(&cq->lock, flags);
+ return true;
+}
+EXPORT_SYMBOL(rvt_cq_enter);
+
+static void send_complete(struct work_struct *work)
+{
+ struct rvt_cq *cq = container_of(work, struct rvt_cq, comptask);
+
+ /*
+ * The completion handler will most likely rearm the notification
+ * and poll for all pending entries. If a new completion entry
+ * is added while we are in this routine, queue_work()
+ * won't call us again until we return so we check triggered to
+ * see if we need to call the handler again.
+ */
+ for (;;) {
+ u8 triggered = cq->triggered;
+
+ /*
+ * IPoIB connected mode assumes the callback is from a
+ * soft IRQ. We simulate this by blocking "bottom halves".
+ * See the implementation for ipoib_cm_handle_tx_wc(),
+ * netif_tx_lock_bh() and netif_tx_lock().
+ */
+ local_bh_disable();
+ cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+ local_bh_enable();
+
+ if (cq->triggered == triggered)
+ return;
+ }
+}
+
+/**
+ * rvt_create_cq - create a completion queue
+ * @ibcq: Allocated CQ
+ * @attr: creation attributes
+ * @udata: user data for libibverbs.so
+ *
+ * Called by ib_create_cq() in the generic verbs code.
+ *
+ * Return: 0 on success
+ */
+int rvt_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
+ struct ib_udata *udata)
+{
+ struct ib_device *ibdev = ibcq->device;
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ struct rvt_cq *cq = ibcq_to_rvtcq(ibcq);
+ struct rvt_cq_wc *u_wc = NULL;
+ struct rvt_k_cq_wc *k_wc = NULL;
+ u32 sz;
+ unsigned int entries = attr->cqe;
+ int comp_vector = attr->comp_vector;
+ int err;
+
+ if (attr->flags)
+ return -EOPNOTSUPP;
+
+ if (entries < 1 || entries > rdi->dparms.props.max_cqe)
+ return -EINVAL;
+
+ if (comp_vector < 0)
+ comp_vector = 0;
+
+ comp_vector = comp_vector % rdi->ibdev.num_comp_vectors;
+
+ /*
+ * Allocate the completion queue entries and head/tail pointers.
+ * This is allocated separately so that it can be resized and
+ * also mapped into user space.
+ * We need to use vmalloc() in order to support mmap and large
+ * numbers of entries.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ sz = sizeof(struct ib_uverbs_wc) * (entries + 1);
+ sz += sizeof(*u_wc);
+ u_wc = vmalloc_user(sz);
+ if (!u_wc)
+ return -ENOMEM;
+ } else {
+ sz = sizeof(struct ib_wc) * (entries + 1);
+ sz += sizeof(*k_wc);
+ k_wc = vzalloc_node(sz, rdi->dparms.node);
+ if (!k_wc)
+ return -ENOMEM;
+ }
+
+ /*
+ * Return the address of the WC as the offset to mmap.
+ * See rvt_mmap() for details.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ cq->ip = rvt_create_mmap_info(rdi, sz, udata, u_wc);
+ if (IS_ERR(cq->ip)) {
+ err = PTR_ERR(cq->ip);
+ goto bail_wc;
+ }
+
+ err = ib_copy_to_udata(udata, &cq->ip->offset,
+ sizeof(cq->ip->offset));
+ if (err)
+ goto bail_ip;
+ }
+
+ spin_lock_irq(&rdi->n_cqs_lock);
+ if (rdi->n_cqs_allocated == rdi->dparms.props.max_cq) {
+ spin_unlock_irq(&rdi->n_cqs_lock);
+ err = -ENOMEM;
+ goto bail_ip;
+ }
+
+ rdi->n_cqs_allocated++;
+ spin_unlock_irq(&rdi->n_cqs_lock);
+
+ if (cq->ip) {
+ spin_lock_irq(&rdi->pending_lock);
+ list_add(&cq->ip->pending_mmaps, &rdi->pending_mmaps);
+ spin_unlock_irq(&rdi->pending_lock);
+ }
+
+ /*
+ * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
+ * The number of entries should be >= the number requested or return
+ * an error.
+ */
+ cq->rdi = rdi;
+ if (rdi->driver_f.comp_vect_cpu_lookup)
+ cq->comp_vector_cpu =
+ rdi->driver_f.comp_vect_cpu_lookup(rdi, comp_vector);
+ else
+ cq->comp_vector_cpu =
+ cpumask_first(cpumask_of_node(rdi->dparms.node));
+
+ cq->ibcq.cqe = entries;
+ cq->notify = RVT_CQ_NONE;
+ spin_lock_init(&cq->lock);
+ INIT_WORK(&cq->comptask, send_complete);
+ if (u_wc)
+ cq->queue = u_wc;
+ else
+ cq->kqueue = k_wc;
+
+ trace_rvt_create_cq(cq, attr);
+ return 0;
+
+bail_ip:
+ kfree(cq->ip);
+bail_wc:
+ vfree(u_wc);
+ vfree(k_wc);
+ return err;
+}
+
+/**
+ * rvt_destroy_cq - destroy a completion queue
+ * @ibcq: the completion queue to destroy.
+ * @udata: user data or NULL for kernel object
+ *
+ * Called by ib_destroy_cq() in the generic verbs code.
+ */
+int rvt_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
+{
+ struct rvt_cq *cq = ibcq_to_rvtcq(ibcq);
+ struct rvt_dev_info *rdi = cq->rdi;
+
+ flush_work(&cq->comptask);
+ spin_lock_irq(&rdi->n_cqs_lock);
+ rdi->n_cqs_allocated--;
+ spin_unlock_irq(&rdi->n_cqs_lock);
+ if (cq->ip)
+ kref_put(&cq->ip->ref, rvt_release_mmap_info);
+ else
+ vfree(cq->kqueue);
+ return 0;
+}
+
+/**
+ * rvt_req_notify_cq - change the notification type for a completion queue
+ * @ibcq: the completion queue
+ * @notify_flags: the type of notification to request
+ *
+ * This may be called from interrupt context. Also called by
+ * ib_req_notify_cq() in the generic verbs code.
+ *
+ * Return: 0 for success.
+ */
+int rvt_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
+{
+ struct rvt_cq *cq = ibcq_to_rvtcq(ibcq);
+ unsigned long flags;
+ int ret = 0;
+
+ spin_lock_irqsave(&cq->lock, flags);
+ /*
+ * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow
+ * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2).
+ */
+ if (cq->notify != IB_CQ_NEXT_COMP)
+ cq->notify = notify_flags & IB_CQ_SOLICITED_MASK;
+
+ if (notify_flags & IB_CQ_REPORT_MISSED_EVENTS) {
+ if (cq->queue) {
+ if (RDMA_READ_UAPI_ATOMIC(cq->queue->head) !=
+ RDMA_READ_UAPI_ATOMIC(cq->queue->tail))
+ ret = 1;
+ } else {
+ if (cq->kqueue->head != cq->kqueue->tail)
+ ret = 1;
+ }
+ }
+
+ spin_unlock_irqrestore(&cq->lock, flags);
+
+ return ret;
+}
+
+/*
+ * rvt_resize_cq - change the size of the CQ
+ * @ibcq: the completion queue
+ *
+ * Return: 0 for success.
+ */
+int rvt_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
+{
+ struct rvt_cq *cq = ibcq_to_rvtcq(ibcq);
+ u32 head, tail, n;
+ int ret;
+ u32 sz;
+ struct rvt_dev_info *rdi = cq->rdi;
+ struct rvt_cq_wc *u_wc = NULL;
+ struct rvt_cq_wc *old_u_wc = NULL;
+ struct rvt_k_cq_wc *k_wc = NULL;
+ struct rvt_k_cq_wc *old_k_wc = NULL;
+
+ if (cqe < 1 || cqe > rdi->dparms.props.max_cqe)
+ return -EINVAL;
+
+ /*
+ * Need to use vmalloc() if we want to support large #s of entries.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ sz = sizeof(struct ib_uverbs_wc) * (cqe + 1);
+ sz += sizeof(*u_wc);
+ u_wc = vmalloc_user(sz);
+ if (!u_wc)
+ return -ENOMEM;
+ } else {
+ sz = sizeof(struct ib_wc) * (cqe + 1);
+ sz += sizeof(*k_wc);
+ k_wc = vzalloc_node(sz, rdi->dparms.node);
+ if (!k_wc)
+ return -ENOMEM;
+ }
+ /* Check that we can write the offset to mmap. */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ __u64 offset = 0;
+
+ ret = ib_copy_to_udata(udata, &offset, sizeof(offset));
+ if (ret)
+ goto bail_free;
+ }
+
+ spin_lock_irq(&cq->lock);
+ /*
+ * Make sure head and tail are sane since they
+ * might be user writable.
+ */
+ if (u_wc) {
+ old_u_wc = cq->queue;
+ head = RDMA_READ_UAPI_ATOMIC(old_u_wc->head);
+ tail = RDMA_READ_UAPI_ATOMIC(old_u_wc->tail);
+ } else {
+ old_k_wc = cq->kqueue;
+ head = old_k_wc->head;
+ tail = old_k_wc->tail;
+ }
+
+ if (head > (u32)cq->ibcq.cqe)
+ head = (u32)cq->ibcq.cqe;
+ if (tail > (u32)cq->ibcq.cqe)
+ tail = (u32)cq->ibcq.cqe;
+ if (head < tail)
+ n = cq->ibcq.cqe + 1 + head - tail;
+ else
+ n = head - tail;
+ if (unlikely((u32)cqe < n)) {
+ ret = -EINVAL;
+ goto bail_unlock;
+ }
+ for (n = 0; tail != head; n++) {
+ if (u_wc)
+ u_wc->uqueue[n] = old_u_wc->uqueue[tail];
+ else
+ k_wc->kqueue[n] = old_k_wc->kqueue[tail];
+ if (tail == (u32)cq->ibcq.cqe)
+ tail = 0;
+ else
+ tail++;
+ }
+ cq->ibcq.cqe = cqe;
+ if (u_wc) {
+ RDMA_WRITE_UAPI_ATOMIC(u_wc->head, n);
+ RDMA_WRITE_UAPI_ATOMIC(u_wc->tail, 0);
+ cq->queue = u_wc;
+ } else {
+ k_wc->head = n;
+ k_wc->tail = 0;
+ cq->kqueue = k_wc;
+ }
+ spin_unlock_irq(&cq->lock);
+
+ if (u_wc)
+ vfree(old_u_wc);
+ else
+ vfree(old_k_wc);
+
+ if (cq->ip) {
+ struct rvt_mmap_info *ip = cq->ip;
+
+ rvt_update_mmap_info(rdi, ip, sz, u_wc);
+
+ /*
+ * Return the offset to mmap.
+ * See rvt_mmap() for details.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ ret = ib_copy_to_udata(udata, &ip->offset,
+ sizeof(ip->offset));
+ if (ret)
+ return ret;
+ }
+
+ spin_lock_irq(&rdi->pending_lock);
+ if (list_empty(&ip->pending_mmaps))
+ list_add(&ip->pending_mmaps, &rdi->pending_mmaps);
+ spin_unlock_irq(&rdi->pending_lock);
+ }
+
+ return 0;
+
+bail_unlock:
+ spin_unlock_irq(&cq->lock);
+bail_free:
+ vfree(u_wc);
+ vfree(k_wc);
+
+ return ret;
+}
+
+/**
+ * rvt_poll_cq - poll for work completion entries
+ * @ibcq: the completion queue to poll
+ * @num_entries: the maximum number of entries to return
+ * @entry: pointer to array where work completions are placed
+ *
+ * This may be called from interrupt context. Also called by ib_poll_cq()
+ * in the generic verbs code.
+ *
+ * Return: the number of completion entries polled.
+ */
+int rvt_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
+{
+ struct rvt_cq *cq = ibcq_to_rvtcq(ibcq);
+ struct rvt_k_cq_wc *wc;
+ unsigned long flags;
+ int npolled;
+ u32 tail;
+
+ /* The kernel can only poll a kernel completion queue */
+ if (cq->ip)
+ return -EINVAL;
+
+ spin_lock_irqsave(&cq->lock, flags);
+
+ wc = cq->kqueue;
+ tail = wc->tail;
+ if (tail > (u32)cq->ibcq.cqe)
+ tail = (u32)cq->ibcq.cqe;
+ for (npolled = 0; npolled < num_entries; ++npolled, ++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;
+ else
+ tail++;
+ }
+ wc->tail = tail;
+
+ spin_unlock_irqrestore(&cq->lock, flags);
+
+ return npolled;
+}
+
+/**
+ * rvt_driver_cq_init - Init cq resources on behalf of driver
+ *
+ * Return: 0 on success
+ */
+int rvt_driver_cq_init(void)
+{
+ comp_vector_wq = alloc_workqueue("%s", WQ_HIGHPRI | WQ_CPU_INTENSIVE,
+ 0, "rdmavt_cq");
+ if (!comp_vector_wq)
+ return -ENOMEM;
+
+ return 0;
+}
+
+/**
+ * rvt_cq_exit - tear down cq reources
+ */
+void rvt_cq_exit(void)
+{
+ destroy_workqueue(comp_vector_wq);
+ comp_vector_wq = NULL;
+}
diff --git a/drivers/infiniband/sw/rdmavt/cq.h b/drivers/infiniband/sw/rdmavt/cq.h
new file mode 100644
index 0000000000..b0a948ec76
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/cq.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 - 2018 Intel Corporation.
+ */
+
+#ifndef DEF_RVTCQ_H
+#define DEF_RVTCQ_H
+
+#include <rdma/rdma_vt.h>
+#include <rdma/rdmavt_cq.h>
+
+int rvt_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
+ struct ib_udata *udata);
+int rvt_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata);
+int rvt_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
+int rvt_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
+int rvt_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
+int rvt_driver_cq_init(void);
+void rvt_cq_exit(void);
+#endif /* DEF_RVTCQ_H */
diff --git a/drivers/infiniband/sw/rdmavt/mad.c b/drivers/infiniband/sw/rdmavt/mad.c
new file mode 100644
index 0000000000..98a8fe3b04
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mad.c
@@ -0,0 +1,130 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <rdma/ib_mad.h>
+#include "mad.h"
+#include "vt.h"
+
+/**
+ * rvt_process_mad - process an incoming MAD packet
+ * @ibdev: the infiniband device this packet came in on
+ * @mad_flags: MAD flags
+ * @port_num: the port number this packet came in on, 1 based from ib core
+ * @in_wc: the work completion entry for this packet
+ * @in_grh: the global route header for this packet
+ * @in: the incoming MAD
+ * @in_mad_size: size of the incoming MAD reply
+ * @out: any outgoing MAD reply
+ * @out_mad_size: size of the outgoing MAD reply
+ * @out_mad_pkey_index: unused
+ *
+ * Note that the verbs framework has already done the MAD sanity checks,
+ * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
+ * MADs.
+ *
+ * This is called by the ib_mad module.
+ *
+ * Return: IB_MAD_RESULT_SUCCESS or error
+ */
+int rvt_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
+ const struct ib_wc *in_wc, const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *in, size_t in_mad_size,
+ struct ib_mad_hdr *out, size_t *out_mad_size,
+ u16 *out_mad_pkey_index)
+{
+ /*
+ * MAD processing is quite different between hfi1 and qib. Therefore
+ * this is expected to be provided by the driver. Other drivers in the
+ * future may choose to implement this but it should not be made into a
+ * requirement.
+ */
+ return IB_MAD_RESULT_FAILURE;
+}
+
+static void rvt_send_mad_handler(struct ib_mad_agent *agent,
+ struct ib_mad_send_wc *mad_send_wc)
+{
+ ib_free_send_mad(mad_send_wc->send_buf);
+}
+
+/**
+ * rvt_create_mad_agents - create mad agents
+ * @rdi: rvt dev struct
+ *
+ * If driver needs to be notified of mad agent creation then call back
+ *
+ * Return 0 on success
+ */
+int rvt_create_mad_agents(struct rvt_dev_info *rdi)
+{
+ struct ib_mad_agent *agent;
+ struct rvt_ibport *rvp;
+ int p;
+ int ret;
+
+ for (p = 0; p < rdi->dparms.nports; p++) {
+ rvp = rdi->ports[p];
+ agent = ib_register_mad_agent(&rdi->ibdev, p + 1,
+ IB_QPT_SMI,
+ NULL, 0, rvt_send_mad_handler,
+ NULL, NULL, 0);
+ if (IS_ERR(agent)) {
+ ret = PTR_ERR(agent);
+ goto err;
+ }
+
+ rvp->send_agent = agent;
+
+ if (rdi->driver_f.notify_create_mad_agent)
+ rdi->driver_f.notify_create_mad_agent(rdi, p);
+ }
+
+ return 0;
+
+err:
+ for (p = 0; p < rdi->dparms.nports; p++) {
+ rvp = rdi->ports[p];
+ if (rvp->send_agent) {
+ agent = rvp->send_agent;
+ rvp->send_agent = NULL;
+ ib_unregister_mad_agent(agent);
+ if (rdi->driver_f.notify_free_mad_agent)
+ rdi->driver_f.notify_free_mad_agent(rdi, p);
+ }
+ }
+
+ return ret;
+}
+
+/**
+ * rvt_free_mad_agents - free up mad agents
+ * @rdi: rvt dev struct
+ *
+ * If driver needs notification of mad agent removal make the call back
+ */
+void rvt_free_mad_agents(struct rvt_dev_info *rdi)
+{
+ struct ib_mad_agent *agent;
+ struct rvt_ibport *rvp;
+ int p;
+
+ for (p = 0; p < rdi->dparms.nports; p++) {
+ rvp = rdi->ports[p];
+ if (rvp->send_agent) {
+ agent = rvp->send_agent;
+ rvp->send_agent = NULL;
+ ib_unregister_mad_agent(agent);
+ }
+ if (rvp->sm_ah) {
+ rdma_destroy_ah(&rvp->sm_ah->ibah,
+ RDMA_DESTROY_AH_SLEEPABLE);
+ rvp->sm_ah = NULL;
+ }
+
+ if (rdi->driver_f.notify_free_mad_agent)
+ rdi->driver_f.notify_free_mad_agent(rdi, p);
+ }
+}
+
diff --git a/drivers/infiniband/sw/rdmavt/mad.h b/drivers/infiniband/sw/rdmavt/mad.h
new file mode 100644
index 0000000000..368be29eab
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mad.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTMAD_H
+#define DEF_RVTMAD_H
+
+#include <rdma/rdma_vt.h>
+
+int rvt_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
+ const struct ib_wc *in_wc, const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *in, size_t in_mad_size,
+ struct ib_mad_hdr *out, size_t *out_mad_size,
+ u16 *out_mad_pkey_index);
+int rvt_create_mad_agents(struct rvt_dev_info *rdi);
+void rvt_free_mad_agents(struct rvt_dev_info *rdi);
+#endif /* DEF_RVTMAD_H */
diff --git a/drivers/infiniband/sw/rdmavt/mcast.c b/drivers/infiniband/sw/rdmavt/mcast.c
new file mode 100644
index 0000000000..a123874e1c
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mcast.c
@@ -0,0 +1,401 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/rculist.h>
+#include <rdma/rdma_vt.h>
+#include <rdma/rdmavt_qp.h>
+
+#include "mcast.h"
+
+/**
+ * rvt_driver_mcast_init - init resources for multicast
+ * @rdi: rvt dev struct
+ *
+ * This is per device that registers with rdmavt
+ */
+void rvt_driver_mcast_init(struct rvt_dev_info *rdi)
+{
+ /*
+ * Anything that needs setup for multicast on a per driver or per rdi
+ * basis should be done in here.
+ */
+ spin_lock_init(&rdi->n_mcast_grps_lock);
+}
+
+/**
+ * rvt_mcast_qp_alloc - alloc a struct to link a QP to mcast GID struct
+ * @qp: the QP to link
+ */
+static struct rvt_mcast_qp *rvt_mcast_qp_alloc(struct rvt_qp *qp)
+{
+ struct rvt_mcast_qp *mqp;
+
+ mqp = kmalloc(sizeof(*mqp), GFP_KERNEL);
+ if (!mqp)
+ goto bail;
+
+ mqp->qp = qp;
+ rvt_get_qp(qp);
+
+bail:
+ return mqp;
+}
+
+static void rvt_mcast_qp_free(struct rvt_mcast_qp *mqp)
+{
+ struct rvt_qp *qp = mqp->qp;
+
+ /* Notify hfi1_destroy_qp() if it is waiting. */
+ rvt_put_qp(qp);
+
+ kfree(mqp);
+}
+
+/**
+ * rvt_mcast_alloc - allocate the multicast GID structure
+ * @mgid: the multicast GID
+ * @lid: the muilticast LID (host order)
+ *
+ * A list of QPs will be attached to this structure.
+ */
+static struct rvt_mcast *rvt_mcast_alloc(union ib_gid *mgid, u16 lid)
+{
+ struct rvt_mcast *mcast;
+
+ mcast = kzalloc(sizeof(*mcast), GFP_KERNEL);
+ if (!mcast)
+ goto bail;
+
+ mcast->mcast_addr.mgid = *mgid;
+ mcast->mcast_addr.lid = lid;
+
+ INIT_LIST_HEAD(&mcast->qp_list);
+ init_waitqueue_head(&mcast->wait);
+ atomic_set(&mcast->refcount, 0);
+
+bail:
+ return mcast;
+}
+
+static void rvt_mcast_free(struct rvt_mcast *mcast)
+{
+ struct rvt_mcast_qp *p, *tmp;
+
+ list_for_each_entry_safe(p, tmp, &mcast->qp_list, list)
+ rvt_mcast_qp_free(p);
+
+ kfree(mcast);
+}
+
+/**
+ * rvt_mcast_find - search the global table for the given multicast GID/LID
+ * NOTE: It is valid to have 1 MLID with multiple MGIDs. It is not valid
+ * to have 1 MGID with multiple MLIDs.
+ * @ibp: the IB port structure
+ * @mgid: the multicast GID to search for
+ * @lid: the multicast LID portion of the multicast address (host order)
+ *
+ * The caller is responsible for decrementing the reference count if found.
+ *
+ * Return: NULL if not found.
+ */
+struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid,
+ u16 lid)
+{
+ struct rb_node *n;
+ unsigned long flags;
+ struct rvt_mcast *found = NULL;
+
+ spin_lock_irqsave(&ibp->lock, flags);
+ n = ibp->mcast_tree.rb_node;
+ while (n) {
+ int ret;
+ struct rvt_mcast *mcast;
+
+ mcast = rb_entry(n, struct rvt_mcast, rb_node);
+
+ ret = memcmp(mgid->raw, mcast->mcast_addr.mgid.raw,
+ sizeof(*mgid));
+ if (ret < 0) {
+ n = n->rb_left;
+ } else if (ret > 0) {
+ n = n->rb_right;
+ } else {
+ /* MGID/MLID must match */
+ if (mcast->mcast_addr.lid == lid) {
+ atomic_inc(&mcast->refcount);
+ found = mcast;
+ }
+ break;
+ }
+ }
+ spin_unlock_irqrestore(&ibp->lock, flags);
+ return found;
+}
+EXPORT_SYMBOL(rvt_mcast_find);
+
+/*
+ * rvt_mcast_add - insert mcast GID into table and attach QP struct
+ * @mcast: the mcast GID table
+ * @mqp: the QP to attach
+ *
+ * Return: zero if both were added. Return EEXIST if the GID was already in
+ * the table but the QP was added. Return ESRCH if the QP was already
+ * attached and neither structure was added. Return EINVAL if the MGID was
+ * found, but the MLID did NOT match.
+ */
+static int rvt_mcast_add(struct rvt_dev_info *rdi, struct rvt_ibport *ibp,
+ struct rvt_mcast *mcast, struct rvt_mcast_qp *mqp)
+{
+ struct rb_node **n = &ibp->mcast_tree.rb_node;
+ struct rb_node *pn = NULL;
+ int ret;
+
+ spin_lock_irq(&ibp->lock);
+
+ while (*n) {
+ struct rvt_mcast *tmcast;
+ struct rvt_mcast_qp *p;
+
+ pn = *n;
+ tmcast = rb_entry(pn, struct rvt_mcast, rb_node);
+
+ ret = memcmp(mcast->mcast_addr.mgid.raw,
+ tmcast->mcast_addr.mgid.raw,
+ sizeof(mcast->mcast_addr.mgid));
+ if (ret < 0) {
+ n = &pn->rb_left;
+ continue;
+ }
+ if (ret > 0) {
+ n = &pn->rb_right;
+ continue;
+ }
+
+ if (tmcast->mcast_addr.lid != mcast->mcast_addr.lid) {
+ ret = EINVAL;
+ goto bail;
+ }
+
+ /* Search the QP list to see if this is already there. */
+ list_for_each_entry_rcu(p, &tmcast->qp_list, list) {
+ if (p->qp == mqp->qp) {
+ ret = ESRCH;
+ goto bail;
+ }
+ }
+ if (tmcast->n_attached ==
+ rdi->dparms.props.max_mcast_qp_attach) {
+ ret = ENOMEM;
+ goto bail;
+ }
+
+ tmcast->n_attached++;
+
+ list_add_tail_rcu(&mqp->list, &tmcast->qp_list);
+ ret = EEXIST;
+ goto bail;
+ }
+
+ spin_lock(&rdi->n_mcast_grps_lock);
+ if (rdi->n_mcast_grps_allocated == rdi->dparms.props.max_mcast_grp) {
+ spin_unlock(&rdi->n_mcast_grps_lock);
+ ret = ENOMEM;
+ goto bail;
+ }
+
+ rdi->n_mcast_grps_allocated++;
+ spin_unlock(&rdi->n_mcast_grps_lock);
+
+ mcast->n_attached++;
+
+ list_add_tail_rcu(&mqp->list, &mcast->qp_list);
+
+ atomic_inc(&mcast->refcount);
+ rb_link_node(&mcast->rb_node, pn, n);
+ rb_insert_color(&mcast->rb_node, &ibp->mcast_tree);
+
+ ret = 0;
+
+bail:
+ spin_unlock_irq(&ibp->lock);
+
+ return ret;
+}
+
+/**
+ * rvt_attach_mcast - attach a qp to a multicast group
+ * @ibqp: Infiniband qp
+ * @gid: multicast guid
+ * @lid: multicast lid
+ *
+ * Return: 0 on success
+ */
+int rvt_attach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+ struct rvt_ibport *ibp = rdi->ports[qp->port_num - 1];
+ struct rvt_mcast *mcast;
+ struct rvt_mcast_qp *mqp;
+ int ret = -ENOMEM;
+
+ if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET)
+ return -EINVAL;
+
+ /*
+ * Allocate data structures since its better to do this outside of
+ * spin locks and it will most likely be needed.
+ */
+ mcast = rvt_mcast_alloc(gid, lid);
+ if (!mcast)
+ return -ENOMEM;
+
+ mqp = rvt_mcast_qp_alloc(qp);
+ if (!mqp)
+ goto bail_mcast;
+
+ switch (rvt_mcast_add(rdi, ibp, mcast, mqp)) {
+ case ESRCH:
+ /* Neither was used: OK to attach the same QP twice. */
+ ret = 0;
+ goto bail_mqp;
+ case EEXIST: /* The mcast wasn't used */
+ ret = 0;
+ goto bail_mcast;
+ case ENOMEM:
+ /* Exceeded the maximum number of mcast groups. */
+ ret = -ENOMEM;
+ goto bail_mqp;
+ case EINVAL:
+ /* Invalid MGID/MLID pair */
+ ret = -EINVAL;
+ goto bail_mqp;
+ default:
+ break;
+ }
+
+ return 0;
+
+bail_mqp:
+ rvt_mcast_qp_free(mqp);
+
+bail_mcast:
+ rvt_mcast_free(mcast);
+
+ return ret;
+}
+
+/**
+ * rvt_detach_mcast - remove a qp from a multicast group
+ * @ibqp: Infiniband qp
+ * @gid: multicast guid
+ * @lid: multicast lid
+ *
+ * Return: 0 on success
+ */
+int rvt_detach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+ struct rvt_ibport *ibp = rdi->ports[qp->port_num - 1];
+ struct rvt_mcast *mcast = NULL;
+ struct rvt_mcast_qp *p, *tmp, *delp = NULL;
+ struct rb_node *n;
+ int last = 0;
+ int ret = 0;
+
+ if (ibqp->qp_num <= 1)
+ return -EINVAL;
+
+ spin_lock_irq(&ibp->lock);
+
+ /* Find the GID in the mcast table. */
+ n = ibp->mcast_tree.rb_node;
+ while (1) {
+ if (!n) {
+ spin_unlock_irq(&ibp->lock);
+ return -EINVAL;
+ }
+
+ mcast = rb_entry(n, struct rvt_mcast, rb_node);
+ ret = memcmp(gid->raw, mcast->mcast_addr.mgid.raw,
+ sizeof(*gid));
+ if (ret < 0) {
+ n = n->rb_left;
+ } else if (ret > 0) {
+ n = n->rb_right;
+ } else {
+ /* MGID/MLID must match */
+ if (mcast->mcast_addr.lid != lid) {
+ spin_unlock_irq(&ibp->lock);
+ return -EINVAL;
+ }
+ break;
+ }
+ }
+
+ /* Search the QP list. */
+ list_for_each_entry_safe(p, tmp, &mcast->qp_list, list) {
+ if (p->qp != qp)
+ continue;
+ /*
+ * We found it, so remove it, but don't poison the forward
+ * link until we are sure there are no list walkers.
+ */
+ list_del_rcu(&p->list);
+ mcast->n_attached--;
+ delp = p;
+
+ /* If this was the last attached QP, remove the GID too. */
+ if (list_empty(&mcast->qp_list)) {
+ rb_erase(&mcast->rb_node, &ibp->mcast_tree);
+ last = 1;
+ }
+ break;
+ }
+
+ spin_unlock_irq(&ibp->lock);
+ /* QP not attached */
+ if (!delp)
+ return -EINVAL;
+
+ /*
+ * Wait for any list walkers to finish before freeing the
+ * list element.
+ */
+ wait_event(mcast->wait, atomic_read(&mcast->refcount) <= 1);
+ rvt_mcast_qp_free(delp);
+
+ if (last) {
+ atomic_dec(&mcast->refcount);
+ wait_event(mcast->wait, !atomic_read(&mcast->refcount));
+ rvt_mcast_free(mcast);
+ spin_lock_irq(&rdi->n_mcast_grps_lock);
+ rdi->n_mcast_grps_allocated--;
+ spin_unlock_irq(&rdi->n_mcast_grps_lock);
+ }
+
+ return 0;
+}
+
+/**
+ * rvt_mcast_tree_empty - determine if any qps are attached to any mcast group
+ * @rdi: rvt dev struct
+ *
+ * Return: in use count
+ */
+int rvt_mcast_tree_empty(struct rvt_dev_info *rdi)
+{
+ int i;
+ int in_use = 0;
+
+ for (i = 0; i < rdi->dparms.nports; i++)
+ if (rdi->ports[i]->mcast_tree.rb_node)
+ in_use++;
+ return in_use;
+}
diff --git a/drivers/infiniband/sw/rdmavt/mcast.h b/drivers/infiniband/sw/rdmavt/mcast.h
new file mode 100644
index 0000000000..b96d86f962
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mcast.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTMCAST_H
+#define DEF_RVTMCAST_H
+
+#include <rdma/rdma_vt.h>
+
+void rvt_driver_mcast_init(struct rvt_dev_info *rdi);
+int rvt_attach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+int rvt_detach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+int rvt_mcast_tree_empty(struct rvt_dev_info *rdi);
+
+#endif /* DEF_RVTMCAST_H */
diff --git a/drivers/infiniband/sw/rdmavt/mmap.c b/drivers/infiniband/sw/rdmavt/mmap.c
new file mode 100644
index 0000000000..4d2238f3f3
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mmap.c
@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <rdma/uverbs_ioctl.h>
+#include "mmap.h"
+
+/**
+ * rvt_mmap_init - init link list and lock for mem map
+ * @rdi: rvt dev struct
+ */
+void rvt_mmap_init(struct rvt_dev_info *rdi)
+{
+ INIT_LIST_HEAD(&rdi->pending_mmaps);
+ spin_lock_init(&rdi->pending_lock);
+ rdi->mmap_offset = PAGE_SIZE;
+ spin_lock_init(&rdi->mmap_offset_lock);
+}
+
+/**
+ * rvt_release_mmap_info - free mmap info structure
+ * @ref: a pointer to the kref within struct rvt_mmap_info
+ */
+void rvt_release_mmap_info(struct kref *ref)
+{
+ struct rvt_mmap_info *ip =
+ container_of(ref, struct rvt_mmap_info, ref);
+ struct rvt_dev_info *rdi = ib_to_rvt(ip->context->device);
+
+ spin_lock_irq(&rdi->pending_lock);
+ list_del(&ip->pending_mmaps);
+ spin_unlock_irq(&rdi->pending_lock);
+
+ vfree(ip->obj);
+ kfree(ip);
+}
+
+static void rvt_vma_open(struct vm_area_struct *vma)
+{
+ struct rvt_mmap_info *ip = vma->vm_private_data;
+
+ kref_get(&ip->ref);
+}
+
+static void rvt_vma_close(struct vm_area_struct *vma)
+{
+ struct rvt_mmap_info *ip = vma->vm_private_data;
+
+ kref_put(&ip->ref, rvt_release_mmap_info);
+}
+
+static const struct vm_operations_struct rvt_vm_ops = {
+ .open = rvt_vma_open,
+ .close = rvt_vma_close,
+};
+
+/**
+ * rvt_mmap - create a new mmap region
+ * @context: the IB user context of the process making the mmap() call
+ * @vma: the VMA to be initialized
+ *
+ * Return: zero if the mmap is OK. Otherwise, return an errno.
+ */
+int rvt_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(context->device);
+ unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+ unsigned long size = vma->vm_end - vma->vm_start;
+ struct rvt_mmap_info *ip, *pp;
+ int ret = -EINVAL;
+
+ /*
+ * Search the device's list of objects waiting for a mmap call.
+ * Normally, this list is very short since a call to create a
+ * CQ, QP, or SRQ is soon followed by a call to mmap().
+ */
+ spin_lock_irq(&rdi->pending_lock);
+ list_for_each_entry_safe(ip, pp, &rdi->pending_mmaps,
+ pending_mmaps) {
+ /* Only the creator is allowed to mmap the object */
+ if (context != ip->context || (__u64)offset != ip->offset)
+ continue;
+ /* Don't allow a mmap larger than the object. */
+ if (size > ip->size)
+ break;
+
+ list_del_init(&ip->pending_mmaps);
+ spin_unlock_irq(&rdi->pending_lock);
+
+ ret = remap_vmalloc_range(vma, ip->obj, 0);
+ if (ret)
+ goto done;
+ vma->vm_ops = &rvt_vm_ops;
+ vma->vm_private_data = ip;
+ rvt_vma_open(vma);
+ goto done;
+ }
+ spin_unlock_irq(&rdi->pending_lock);
+done:
+ return ret;
+}
+
+/**
+ * rvt_create_mmap_info - allocate information for hfi1_mmap
+ * @rdi: rvt dev struct
+ * @size: size in bytes to map
+ * @udata: user data (must be valid!)
+ * @obj: opaque pointer to a cq, wq etc
+ *
+ * Return: rvt_mmap struct on success, ERR_PTR on failure
+ */
+struct rvt_mmap_info *rvt_create_mmap_info(struct rvt_dev_info *rdi, u32 size,
+ struct ib_udata *udata, void *obj)
+{
+ struct rvt_mmap_info *ip;
+
+ if (!udata)
+ return ERR_PTR(-EINVAL);
+
+ ip = kmalloc_node(sizeof(*ip), GFP_KERNEL, rdi->dparms.node);
+ if (!ip)
+ return ERR_PTR(-ENOMEM);
+
+ size = PAGE_ALIGN(size);
+
+ spin_lock_irq(&rdi->mmap_offset_lock);
+ if (rdi->mmap_offset == 0)
+ rdi->mmap_offset = ALIGN(PAGE_SIZE, SHMLBA);
+ ip->offset = rdi->mmap_offset;
+ rdi->mmap_offset += ALIGN(size, SHMLBA);
+ spin_unlock_irq(&rdi->mmap_offset_lock);
+
+ INIT_LIST_HEAD(&ip->pending_mmaps);
+ ip->size = size;
+ ip->context =
+ container_of(udata, struct uverbs_attr_bundle, driver_udata)
+ ->context;
+ ip->obj = obj;
+ kref_init(&ip->ref);
+
+ return ip;
+}
+
+/**
+ * rvt_update_mmap_info - update a mem map
+ * @rdi: rvt dev struct
+ * @ip: mmap info pointer
+ * @size: size to grow by
+ * @obj: opaque pointer to cq, wq, etc.
+ */
+void rvt_update_mmap_info(struct rvt_dev_info *rdi, struct rvt_mmap_info *ip,
+ u32 size, void *obj)
+{
+ size = PAGE_ALIGN(size);
+
+ spin_lock_irq(&rdi->mmap_offset_lock);
+ if (rdi->mmap_offset == 0)
+ rdi->mmap_offset = PAGE_SIZE;
+ ip->offset = rdi->mmap_offset;
+ rdi->mmap_offset += size;
+ spin_unlock_irq(&rdi->mmap_offset_lock);
+
+ ip->size = size;
+ ip->obj = obj;
+}
diff --git a/drivers/infiniband/sw/rdmavt/mmap.h b/drivers/infiniband/sw/rdmavt/mmap.h
new file mode 100644
index 0000000000..7e92cf28e0
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mmap.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RDMAVTMMAP_H
+#define DEF_RDMAVTMMAP_H
+
+#include <rdma/rdma_vt.h>
+
+void rvt_mmap_init(struct rvt_dev_info *rdi);
+void rvt_release_mmap_info(struct kref *ref);
+int rvt_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
+struct rvt_mmap_info *rvt_create_mmap_info(struct rvt_dev_info *rdi, u32 size,
+ struct ib_udata *udata, void *obj);
+void rvt_update_mmap_info(struct rvt_dev_info *rdi, struct rvt_mmap_info *ip,
+ u32 size, void *obj);
+
+#endif /* DEF_RDMAVTMMAP_H */
diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
new file mode 100644
index 0000000000..8a1f2e2851
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -0,0 +1,917 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <rdma/ib_umem.h>
+#include <rdma/rdma_vt.h>
+#include "vt.h"
+#include "mr.h"
+#include "trace.h"
+
+/**
+ * rvt_driver_mr_init - Init MR resources per driver
+ * @rdi: rvt dev struct
+ *
+ * Do any intilization needed when a driver registers with rdmavt.
+ *
+ * Return: 0 on success or errno on failure
+ */
+int rvt_driver_mr_init(struct rvt_dev_info *rdi)
+{
+ unsigned int lkey_table_size = rdi->dparms.lkey_table_size;
+ unsigned lk_tab_size;
+ int i;
+
+ /*
+ * The top hfi1_lkey_table_size bits are used to index the
+ * table. The lower 8 bits can be owned by the user (copied from
+ * the LKEY). The remaining bits act as a generation number or tag.
+ */
+ if (!lkey_table_size)
+ return -EINVAL;
+
+ spin_lock_init(&rdi->lkey_table.lock);
+
+ /* ensure generation is at least 4 bits */
+ if (lkey_table_size > RVT_MAX_LKEY_TABLE_BITS) {
+ rvt_pr_warn(rdi, "lkey bits %u too large, reduced to %u\n",
+ lkey_table_size, RVT_MAX_LKEY_TABLE_BITS);
+ rdi->dparms.lkey_table_size = RVT_MAX_LKEY_TABLE_BITS;
+ lkey_table_size = rdi->dparms.lkey_table_size;
+ }
+ rdi->lkey_table.max = 1 << lkey_table_size;
+ rdi->lkey_table.shift = 32 - lkey_table_size;
+ lk_tab_size = rdi->lkey_table.max * sizeof(*rdi->lkey_table.table);
+ rdi->lkey_table.table = (struct rvt_mregion __rcu **)
+ vmalloc_node(lk_tab_size, rdi->dparms.node);
+ if (!rdi->lkey_table.table)
+ return -ENOMEM;
+
+ RCU_INIT_POINTER(rdi->dma_mr, NULL);
+ for (i = 0; i < rdi->lkey_table.max; i++)
+ RCU_INIT_POINTER(rdi->lkey_table.table[i], NULL);
+
+ rdi->dparms.props.max_mr = rdi->lkey_table.max;
+ return 0;
+}
+
+/**
+ * rvt_mr_exit - clean up MR
+ * @rdi: rvt dev structure
+ *
+ * called when drivers have unregistered or perhaps failed to register with us
+ */
+void rvt_mr_exit(struct rvt_dev_info *rdi)
+{
+ if (rdi->dma_mr)
+ rvt_pr_err(rdi, "DMA MR not null!\n");
+
+ vfree(rdi->lkey_table.table);
+}
+
+static void rvt_deinit_mregion(struct rvt_mregion *mr)
+{
+ int i = mr->mapsz;
+
+ mr->mapsz = 0;
+ while (i)
+ kfree(mr->map[--i]);
+ percpu_ref_exit(&mr->refcount);
+}
+
+static void __rvt_mregion_complete(struct percpu_ref *ref)
+{
+ struct rvt_mregion *mr = container_of(ref, struct rvt_mregion,
+ refcount);
+
+ complete(&mr->comp);
+}
+
+static int rvt_init_mregion(struct rvt_mregion *mr, struct ib_pd *pd,
+ int count, unsigned int percpu_flags)
+{
+ int m, i = 0;
+ struct rvt_dev_info *dev = ib_to_rvt(pd->device);
+
+ mr->mapsz = 0;
+ m = (count + RVT_SEGSZ - 1) / RVT_SEGSZ;
+ for (; i < m; i++) {
+ mr->map[i] = kzalloc_node(sizeof(*mr->map[0]), GFP_KERNEL,
+ dev->dparms.node);
+ if (!mr->map[i])
+ goto bail;
+ mr->mapsz++;
+ }
+ init_completion(&mr->comp);
+ /* count returning the ptr to user */
+ if (percpu_ref_init(&mr->refcount, &__rvt_mregion_complete,
+ percpu_flags, GFP_KERNEL))
+ goto bail;
+
+ atomic_set(&mr->lkey_invalid, 0);
+ mr->pd = pd;
+ mr->max_segs = count;
+ return 0;
+bail:
+ rvt_deinit_mregion(mr);
+ return -ENOMEM;
+}
+
+/**
+ * rvt_alloc_lkey - allocate an lkey
+ * @mr: memory region that this lkey protects
+ * @dma_region: 0->normal key, 1->restricted DMA key
+ *
+ * Returns 0 if successful, otherwise returns -errno.
+ *
+ * Increments mr reference count as required.
+ *
+ * Sets the lkey field mr for non-dma regions.
+ *
+ */
+static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region)
+{
+ unsigned long flags;
+ u32 r;
+ u32 n;
+ int ret = 0;
+ struct rvt_dev_info *dev = ib_to_rvt(mr->pd->device);
+ struct rvt_lkey_table *rkt = &dev->lkey_table;
+
+ rvt_get_mr(mr);
+ spin_lock_irqsave(&rkt->lock, flags);
+
+ /* special case for dma_mr lkey == 0 */
+ if (dma_region) {
+ struct rvt_mregion *tmr;
+
+ tmr = rcu_access_pointer(dev->dma_mr);
+ if (!tmr) {
+ mr->lkey_published = 1;
+ /* Insure published written first */
+ rcu_assign_pointer(dev->dma_mr, mr);
+ rvt_get_mr(mr);
+ }
+ goto success;
+ }
+
+ /* Find the next available LKEY */
+ r = rkt->next;
+ n = r;
+ for (;;) {
+ if (!rcu_access_pointer(rkt->table[r]))
+ break;
+ r = (r + 1) & (rkt->max - 1);
+ if (r == n)
+ goto bail;
+ }
+ rkt->next = (r + 1) & (rkt->max - 1);
+ /*
+ * Make sure lkey is never zero which is reserved to indicate an
+ * unrestricted LKEY.
+ */
+ rkt->gen++;
+ /*
+ * bits are capped to ensure enough bits for generation number
+ */
+ mr->lkey = (r << (32 - dev->dparms.lkey_table_size)) |
+ ((((1 << (24 - dev->dparms.lkey_table_size)) - 1) & rkt->gen)
+ << 8);
+ if (mr->lkey == 0) {
+ mr->lkey |= 1 << 8;
+ rkt->gen++;
+ }
+ mr->lkey_published = 1;
+ /* Insure published written first */
+ rcu_assign_pointer(rkt->table[r], mr);
+success:
+ spin_unlock_irqrestore(&rkt->lock, flags);
+out:
+ return ret;
+bail:
+ rvt_put_mr(mr);
+ spin_unlock_irqrestore(&rkt->lock, flags);
+ ret = -ENOMEM;
+ goto out;
+}
+
+/**
+ * rvt_free_lkey - free an lkey
+ * @mr: mr to free from tables
+ */
+static void rvt_free_lkey(struct rvt_mregion *mr)
+{
+ unsigned long flags;
+ u32 lkey = mr->lkey;
+ u32 r;
+ struct rvt_dev_info *dev = ib_to_rvt(mr->pd->device);
+ struct rvt_lkey_table *rkt = &dev->lkey_table;
+ int freed = 0;
+
+ spin_lock_irqsave(&rkt->lock, flags);
+ if (!lkey) {
+ if (mr->lkey_published) {
+ 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);
+ mr->lkey_published = 0;
+ /* insure published is written before pointer */
+ rcu_assign_pointer(rkt->table[r], NULL);
+ }
+ freed++;
+out:
+ spin_unlock_irqrestore(&rkt->lock, flags);
+ if (freed)
+ percpu_ref_kill(&mr->refcount);
+}
+
+static struct rvt_mr *__rvt_alloc_mr(int count, struct ib_pd *pd)
+{
+ struct rvt_mr *mr;
+ int rval = -ENOMEM;
+ int m;
+
+ /* Allocate struct plus pointers to first level page tables. */
+ m = (count + RVT_SEGSZ - 1) / RVT_SEGSZ;
+ mr = kzalloc(struct_size(mr, mr.map, m), GFP_KERNEL);
+ if (!mr)
+ goto bail;
+
+ rval = rvt_init_mregion(&mr->mr, pd, count, 0);
+ if (rval)
+ goto bail;
+ /*
+ * ib_reg_phys_mr() will initialize mr->ibmr except for
+ * lkey and rkey.
+ */
+ rval = rvt_alloc_lkey(&mr->mr, 0);
+ if (rval)
+ goto bail_mregion;
+ mr->ibmr.lkey = mr->mr.lkey;
+ mr->ibmr.rkey = mr->mr.lkey;
+done:
+ return mr;
+
+bail_mregion:
+ rvt_deinit_mregion(&mr->mr);
+bail:
+ kfree(mr);
+ mr = ERR_PTR(rval);
+ goto done;
+}
+
+static void __rvt_free_mr(struct rvt_mr *mr)
+{
+ rvt_free_lkey(&mr->mr);
+ rvt_deinit_mregion(&mr->mr);
+ kfree(mr);
+}
+
+/**
+ * rvt_get_dma_mr - get a DMA memory region
+ * @pd: protection domain for this memory region
+ * @acc: access flags
+ *
+ * Return: the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *rvt_get_dma_mr(struct ib_pd *pd, int acc)
+{
+ struct rvt_mr *mr;
+ struct ib_mr *ret;
+ int rval;
+
+ if (ibpd_to_rvtpd(pd)->user)
+ return ERR_PTR(-EPERM);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr) {
+ ret = ERR_PTR(-ENOMEM);
+ goto bail;
+ }
+
+ rval = rvt_init_mregion(&mr->mr, pd, 0, 0);
+ if (rval) {
+ ret = ERR_PTR(rval);
+ goto bail;
+ }
+
+ rval = rvt_alloc_lkey(&mr->mr, 1);
+ if (rval) {
+ ret = ERR_PTR(rval);
+ goto bail_mregion;
+ }
+
+ mr->mr.access_flags = acc;
+ ret = &mr->ibmr;
+done:
+ return ret;
+
+bail_mregion:
+ rvt_deinit_mregion(&mr->mr);
+bail:
+ kfree(mr);
+ goto done;
+}
+
+/**
+ * rvt_reg_user_mr - register a userspace memory region
+ * @pd: protection domain for this memory region
+ * @start: starting userspace address
+ * @length: length of region to register
+ * @virt_addr: associated virtual address
+ * @mr_access_flags: access flags for this memory region
+ * @udata: unused by the driver
+ *
+ * Return: the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *rvt_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+ u64 virt_addr, int mr_access_flags,
+ struct ib_udata *udata)
+{
+ struct rvt_mr *mr;
+ struct ib_umem *umem;
+ struct sg_page_iter sg_iter;
+ int n, m;
+ struct ib_mr *ret;
+
+ if (length == 0)
+ return ERR_PTR(-EINVAL);
+
+ umem = ib_umem_get(pd->device, start, length, mr_access_flags);
+ if (IS_ERR(umem))
+ return (void *)umem;
+
+ n = ib_umem_num_pages(umem);
+
+ mr = __rvt_alloc_mr(n, pd);
+ if (IS_ERR(mr)) {
+ ret = (struct ib_mr *)mr;
+ goto bail_umem;
+ }
+
+ mr->mr.user_base = start;
+ mr->mr.iova = virt_addr;
+ mr->mr.length = length;
+ mr->mr.offset = ib_umem_offset(umem);
+ mr->mr.access_flags = mr_access_flags;
+ mr->umem = umem;
+
+ mr->mr.page_shift = PAGE_SHIFT;
+ m = 0;
+ n = 0;
+ for_each_sgtable_page (&umem->sgt_append.sgt, &sg_iter, 0) {
+ void *vaddr;
+
+ vaddr = page_address(sg_page_iter_page(&sg_iter));
+ if (!vaddr) {
+ ret = ERR_PTR(-EINVAL);
+ goto bail_inval;
+ }
+ mr->mr.map[m]->segs[n].vaddr = vaddr;
+ mr->mr.map[m]->segs[n].length = PAGE_SIZE;
+ trace_rvt_mr_user_seg(&mr->mr, m, n, vaddr, PAGE_SIZE);
+ if (++n == RVT_SEGSZ) {
+ m++;
+ n = 0;
+ }
+ }
+ return &mr->ibmr;
+
+bail_inval:
+ __rvt_free_mr(mr);
+
+bail_umem:
+ ib_umem_release(umem);
+
+ return ret;
+}
+
+/**
+ * rvt_dereg_clean_qp_cb - callback from iterator
+ * @qp: the qp
+ * @v: the mregion (as u64)
+ *
+ * This routine fields the callback for all QPs and
+ * for QPs in the same PD as the MR will call the
+ * rvt_qp_mr_clean() to potentially cleanup references.
+ */
+static void rvt_dereg_clean_qp_cb(struct rvt_qp *qp, u64 v)
+{
+ struct rvt_mregion *mr = (struct rvt_mregion *)v;
+
+ /* skip PDs that are not ours */
+ if (mr->pd != qp->ibqp.pd)
+ return;
+ rvt_qp_mr_clean(qp, mr->lkey);
+}
+
+/**
+ * rvt_dereg_clean_qps - find QPs for reference cleanup
+ * @mr: the MR that is being deregistered
+ *
+ * This routine iterates RC QPs looking for references
+ * to the lkey noted in mr.
+ */
+static void rvt_dereg_clean_qps(struct rvt_mregion *mr)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(mr->pd->device);
+
+ rvt_qp_iter(rdi, (u64)mr, rvt_dereg_clean_qp_cb);
+}
+
+/**
+ * rvt_check_refs - check references
+ * @mr: the megion
+ * @t: the caller identification
+ *
+ * This routine checks MRs holding a reference during
+ * when being de-registered.
+ *
+ * If the count is non-zero, the code calls a clean routine then
+ * waits for the timeout for the count to zero.
+ */
+static int rvt_check_refs(struct rvt_mregion *mr, const char *t)
+{
+ unsigned long timeout;
+ struct rvt_dev_info *rdi = ib_to_rvt(mr->pd->device);
+
+ if (mr->lkey) {
+ /* avoid dma mr */
+ rvt_dereg_clean_qps(mr);
+ /* @mr was indexed on rcu protected @lkey_table */
+ synchronize_rcu();
+ }
+
+ timeout = wait_for_completion_timeout(&mr->comp, 5 * HZ);
+ if (!timeout) {
+ rvt_pr_err(rdi,
+ "%s timeout mr %p pd %p lkey %x refcount %ld\n",
+ t, mr, mr->pd, mr->lkey,
+ atomic_long_read(&mr->refcount.data->count));
+ rvt_get_mr(mr);
+ return -EBUSY;
+ }
+ return 0;
+}
+
+/**
+ * rvt_mr_has_lkey - is MR
+ * @mr: the mregion
+ * @lkey: the lkey
+ */
+bool rvt_mr_has_lkey(struct rvt_mregion *mr, u32 lkey)
+{
+ return mr && lkey == mr->lkey;
+}
+
+/**
+ * rvt_ss_has_lkey - is mr in sge tests
+ * @ss: the sge state
+ * @lkey: the lkey
+ *
+ * This code tests for an MR in the indicated
+ * sge state.
+ */
+bool rvt_ss_has_lkey(struct rvt_sge_state *ss, u32 lkey)
+{
+ int i;
+ bool rval = false;
+
+ if (!ss->num_sge)
+ return rval;
+ /* first one */
+ rval = rvt_mr_has_lkey(ss->sge.mr, lkey);
+ /* any others */
+ for (i = 0; !rval && i < ss->num_sge - 1; i++)
+ rval = rvt_mr_has_lkey(ss->sg_list[i].mr, lkey);
+ return rval;
+}
+
+/**
+ * rvt_dereg_mr - unregister and free a memory region
+ * @ibmr: the memory region to free
+ * @udata: unused by the driver
+ *
+ * Note that this is called to free MRs created by rvt_get_dma_mr()
+ * or rvt_reg_user_mr().
+ *
+ * Returns 0 on success.
+ */
+int rvt_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata)
+{
+ struct rvt_mr *mr = to_imr(ibmr);
+ int ret;
+
+ rvt_free_lkey(&mr->mr);
+
+ rvt_put_mr(&mr->mr); /* will set completion if last */
+ ret = rvt_check_refs(&mr->mr, __func__);
+ if (ret)
+ goto out;
+ rvt_deinit_mregion(&mr->mr);
+ ib_umem_release(mr->umem);
+ kfree(mr);
+out:
+ return ret;
+}
+
+/**
+ * rvt_alloc_mr - Allocate a memory region usable with the
+ * @pd: protection domain for this memory region
+ * @mr_type: mem region type
+ * @max_num_sg: Max number of segments allowed
+ *
+ * Return: the memory region on success, otherwise return an errno.
+ */
+struct ib_mr *rvt_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+ u32 max_num_sg)
+{
+ struct rvt_mr *mr;
+
+ if (mr_type != IB_MR_TYPE_MEM_REG)
+ return ERR_PTR(-EINVAL);
+
+ mr = __rvt_alloc_mr(max_num_sg, pd);
+ if (IS_ERR(mr))
+ return (struct ib_mr *)mr;
+
+ return &mr->ibmr;
+}
+
+/**
+ * rvt_set_page - page assignment function called by ib_sg_to_pages
+ * @ibmr: memory region
+ * @addr: dma address of mapped page
+ *
+ * Return: 0 on success
+ */
+static int rvt_set_page(struct ib_mr *ibmr, u64 addr)
+{
+ struct rvt_mr *mr = to_imr(ibmr);
+ u32 ps = 1 << mr->mr.page_shift;
+ u32 mapped_segs = mr->mr.length >> mr->mr.page_shift;
+ int m, n;
+
+ if (unlikely(mapped_segs == mr->mr.max_segs))
+ return -ENOMEM;
+
+ m = mapped_segs / RVT_SEGSZ;
+ n = mapped_segs % RVT_SEGSZ;
+ mr->mr.map[m]->segs[n].vaddr = (void *)addr;
+ mr->mr.map[m]->segs[n].length = ps;
+ mr->mr.length += ps;
+ trace_rvt_mr_page_seg(&mr->mr, m, n, (void *)addr, ps);
+
+ return 0;
+}
+
+/**
+ * rvt_map_mr_sg - map sg list and set it the memory region
+ * @ibmr: memory region
+ * @sg: dma mapped scatterlist
+ * @sg_nents: number of entries in sg
+ * @sg_offset: offset in bytes into sg
+ *
+ * Overwrite rvt_mr length with mr length calculated by ib_sg_to_pages.
+ *
+ * Return: number of sg elements mapped to the memory region
+ */
+int rvt_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset)
+{
+ struct rvt_mr *mr = to_imr(ibmr);
+ int ret;
+
+ mr->mr.length = 0;
+ mr->mr.page_shift = PAGE_SHIFT;
+ ret = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rvt_set_page);
+ mr->mr.user_base = ibmr->iova;
+ mr->mr.iova = ibmr->iova;
+ mr->mr.offset = ibmr->iova - (u64)mr->mr.map[0]->segs[0].vaddr;
+ mr->mr.length = (size_t)ibmr->length;
+ trace_rvt_map_mr_sg(ibmr, sg_nents, sg_offset);
+ return ret;
+}
+
+/**
+ * rvt_fast_reg_mr - fast register physical MR
+ * @qp: the queue pair where the work request comes from
+ * @ibmr: the memory region to be registered
+ * @key: updated key for this memory region
+ * @access: access flags for this memory region
+ *
+ * Returns 0 on success.
+ */
+int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key,
+ int access)
+{
+ struct rvt_mr *mr = to_imr(ibmr);
+
+ if (qp->ibqp.pd != mr->mr.pd)
+ return -EACCES;
+
+ /* not applicable to dma MR or user MR */
+ if (!mr->mr.lkey || mr->umem)
+ return -EINVAL;
+
+ if ((key & 0xFFFFFF00) != (mr->mr.lkey & 0xFFFFFF00))
+ return -EINVAL;
+
+ ibmr->lkey = key;
+ ibmr->rkey = key;
+ mr->mr.lkey = key;
+ mr->mr.access_flags = access;
+ mr->mr.iova = ibmr->iova;
+ atomic_set(&mr->mr.lkey_invalid, 0);
+
+ return 0;
+}
+EXPORT_SYMBOL(rvt_fast_reg_mr);
+
+/**
+ * rvt_invalidate_rkey - invalidate an MR rkey
+ * @qp: queue pair associated with the invalidate op
+ * @rkey: rkey to invalidate
+ *
+ * Returns 0 on success.
+ */
+int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey)
+{
+ struct rvt_dev_info *dev = ib_to_rvt(qp->ibqp.device);
+ struct rvt_lkey_table *rkt = &dev->lkey_table;
+ struct rvt_mregion *mr;
+
+ if (rkey == 0)
+ return -EINVAL;
+
+ rcu_read_lock();
+ mr = rcu_dereference(
+ rkt->table[(rkey >> (32 - dev->dparms.lkey_table_size))]);
+ if (unlikely(!mr || mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+ goto bail;
+
+ atomic_set(&mr->lkey_invalid, 1);
+ rcu_read_unlock();
+ return 0;
+
+bail:
+ rcu_read_unlock();
+ return -EINVAL;
+}
+EXPORT_SYMBOL(rvt_invalidate_rkey);
+
+/**
+ * rvt_sge_adjacent - is isge compressible
+ * @last_sge: last outgoing SGE written
+ * @sge: SGE to check
+ *
+ * If adjacent will update last_sge to add length.
+ *
+ * Return: true if isge is adjacent to last sge
+ */
+static inline bool rvt_sge_adjacent(struct rvt_sge *last_sge,
+ struct ib_sge *sge)
+{
+ if (last_sge && sge->lkey == last_sge->mr->lkey &&
+ ((uint64_t)(last_sge->vaddr + last_sge->length) == sge->addr)) {
+ if (sge->lkey) {
+ if (unlikely((sge->addr - last_sge->mr->user_base +
+ sge->length > last_sge->mr->length)))
+ return false; /* overrun, caller will catch */
+ } else {
+ last_sge->length += sge->length;
+ }
+ last_sge->sge_length += sge->length;
+ trace_rvt_sge_adjacent(last_sge, sge);
+ return true;
+ }
+ return false;
+}
+
+/**
+ * rvt_lkey_ok - check IB SGE for validity and initialize
+ * @rkt: table containing lkey to check SGE against
+ * @pd: protection domain
+ * @isge: outgoing internal SGE
+ * @last_sge: last outgoing SGE written
+ * @sge: SGE to check
+ * @acc: access flags
+ *
+ * Check the IB SGE for validity and initialize our internal version
+ * of it.
+ *
+ * Increments the reference count when a new sge is stored.
+ *
+ * Return: 0 if compressed, 1 if added , otherwise returns -errno.
+ */
+int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
+ struct rvt_sge *isge, struct rvt_sge *last_sge,
+ struct ib_sge *sge, int acc)
+{
+ struct rvt_mregion *mr;
+ unsigned n, m;
+ size_t off;
+
+ /*
+ * We use LKEY == zero for kernel virtual addresses
+ * (see rvt_get_dma_mr()).
+ */
+ if (sge->lkey == 0) {
+ struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
+
+ if (pd->user)
+ return -EINVAL;
+ if (rvt_sge_adjacent(last_sge, sge))
+ return 0;
+ rcu_read_lock();
+ mr = rcu_dereference(dev->dma_mr);
+ if (!mr)
+ goto bail;
+ rvt_get_mr(mr);
+ rcu_read_unlock();
+
+ isge->mr = mr;
+ isge->vaddr = (void *)sge->addr;
+ isge->length = sge->length;
+ isge->sge_length = sge->length;
+ isge->m = 0;
+ isge->n = 0;
+ goto ok;
+ }
+ if (rvt_sge_adjacent(last_sge, sge))
+ return 0;
+ rcu_read_lock();
+ mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]);
+ 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_unref;
+ rcu_read_unlock();
+
+ off += mr->offset;
+ if (mr->page_shift) {
+ /*
+ * page sizes are uniform power of 2 so no loop is necessary
+ * entries_spanned_by_off is the number of times the loop below
+ * would have executed.
+ */
+ size_t entries_spanned_by_off;
+
+ entries_spanned_by_off = off >> mr->page_shift;
+ off -= (entries_spanned_by_off << mr->page_shift);
+ m = entries_spanned_by_off / RVT_SEGSZ;
+ n = entries_spanned_by_off % RVT_SEGSZ;
+ } else {
+ m = 0;
+ n = 0;
+ while (off >= mr->map[m]->segs[n].length) {
+ off -= mr->map[m]->segs[n].length;
+ n++;
+ if (n >= RVT_SEGSZ) {
+ m++;
+ n = 0;
+ }
+ }
+ }
+ isge->mr = mr;
+ isge->vaddr = mr->map[m]->segs[n].vaddr + off;
+ isge->length = mr->map[m]->segs[n].length - off;
+ isge->sge_length = sge->length;
+ isge->m = m;
+ isge->n = n;
+ok:
+ trace_rvt_sge_new(isge, sge);
+ return 1;
+bail_unref:
+ rvt_put_mr(mr);
+bail:
+ rcu_read_unlock();
+ return -EINVAL;
+}
+EXPORT_SYMBOL(rvt_lkey_ok);
+
+/**
+ * rvt_rkey_ok - check the IB virtual address, length, and RKEY
+ * @qp: qp for validation
+ * @sge: SGE state
+ * @len: length of data
+ * @vaddr: virtual address to place data
+ * @rkey: rkey to check
+ * @acc: access flags
+ *
+ * Return: 1 if successful, otherwise 0.
+ *
+ * increments the reference count upon success
+ */
+int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
+ u32 len, u64 vaddr, u32 rkey, int acc)
+{
+ struct rvt_dev_info *dev = ib_to_rvt(qp->ibqp.device);
+ struct rvt_lkey_table *rkt = &dev->lkey_table;
+ struct rvt_mregion *mr;
+ unsigned n, m;
+ size_t off;
+
+ /*
+ * We use RKEY == zero for kernel virtual addresses
+ * (see rvt_get_dma_mr()).
+ */
+ rcu_read_lock();
+ if (rkey == 0) {
+ struct rvt_pd *pd = ibpd_to_rvtpd(qp->ibqp.pd);
+ struct rvt_dev_info *rdi = ib_to_rvt(pd->ibpd.device);
+
+ if (pd->user)
+ goto bail;
+ mr = rcu_dereference(rdi->dma_mr);
+ if (!mr)
+ goto bail;
+ rvt_get_mr(mr);
+ rcu_read_unlock();
+
+ sge->mr = mr;
+ sge->vaddr = (void *)vaddr;
+ sge->length = len;
+ sge->sge_length = len;
+ sge->m = 0;
+ sge->n = 0;
+ goto ok;
+ }
+
+ mr = rcu_dereference(rkt->table[rkey >> rkt->shift]);
+ 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_unref;
+ rcu_read_unlock();
+
+ off += mr->offset;
+ if (mr->page_shift) {
+ /*
+ * page sizes are uniform power of 2 so no loop is necessary
+ * entries_spanned_by_off is the number of times the loop below
+ * would have executed.
+ */
+ size_t entries_spanned_by_off;
+
+ entries_spanned_by_off = off >> mr->page_shift;
+ off -= (entries_spanned_by_off << mr->page_shift);
+ m = entries_spanned_by_off / RVT_SEGSZ;
+ n = entries_spanned_by_off % RVT_SEGSZ;
+ } else {
+ m = 0;
+ n = 0;
+ while (off >= mr->map[m]->segs[n].length) {
+ off -= mr->map[m]->segs[n].length;
+ n++;
+ if (n >= RVT_SEGSZ) {
+ m++;
+ n = 0;
+ }
+ }
+ }
+ sge->mr = mr;
+ sge->vaddr = mr->map[m]->segs[n].vaddr + off;
+ sge->length = mr->map[m]->segs[n].length - off;
+ sge->sge_length = len;
+ sge->m = m;
+ sge->n = n;
+ok:
+ return 1;
+bail_unref:
+ rvt_put_mr(mr);
+bail:
+ rcu_read_unlock();
+ return 0;
+}
+EXPORT_SYMBOL(rvt_rkey_ok);
diff --git a/drivers/infiniband/sw/rdmavt/mr.h b/drivers/infiniband/sw/rdmavt/mr.h
new file mode 100644
index 0000000000..d17f1400b5
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/mr.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTMR_H
+#define DEF_RVTMR_H
+
+#include <rdma/rdma_vt.h>
+
+struct rvt_mr {
+ struct ib_mr ibmr;
+ struct ib_umem *umem;
+ struct rvt_mregion mr; /* must be last */
+};
+
+static inline struct rvt_mr *to_imr(struct ib_mr *ibmr)
+{
+ return container_of(ibmr, struct rvt_mr, ibmr);
+}
+
+int rvt_driver_mr_init(struct rvt_dev_info *rdi);
+void rvt_mr_exit(struct rvt_dev_info *rdi);
+
+/* Mem Regions */
+struct ib_mr *rvt_get_dma_mr(struct ib_pd *pd, int acc);
+struct ib_mr *rvt_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+ u64 virt_addr, int mr_access_flags,
+ struct ib_udata *udata);
+int rvt_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
+struct ib_mr *rvt_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+ u32 max_num_sg);
+int rvt_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset);
+
+#endif /* DEF_RVTMR_H */
diff --git a/drivers/infiniband/sw/rdmavt/pd.c b/drivers/infiniband/sw/rdmavt/pd.c
new file mode 100644
index 0000000000..ae62071969
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/pd.c
@@ -0,0 +1,64 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <linux/slab.h>
+#include "pd.h"
+
+/**
+ * rvt_alloc_pd - allocate a protection domain
+ * @ibpd: PD
+ * @udata: optional user data
+ *
+ * Allocate and keep track of a PD.
+ *
+ * Return: 0 on success
+ */
+int rvt_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
+{
+ struct ib_device *ibdev = ibpd->device;
+ struct rvt_dev_info *dev = ib_to_rvt(ibdev);
+ struct rvt_pd *pd = ibpd_to_rvtpd(ibpd);
+ int ret = 0;
+
+ /*
+ * While we could continue allocating protecetion domains, being
+ * constrained only by system resources. The IBTA spec defines that
+ * there is a max_pd limit that can be set and we need to check for
+ * that.
+ */
+
+ spin_lock(&dev->n_pds_lock);
+ if (dev->n_pds_allocated == dev->dparms.props.max_pd) {
+ spin_unlock(&dev->n_pds_lock);
+ ret = -ENOMEM;
+ goto bail;
+ }
+
+ dev->n_pds_allocated++;
+ spin_unlock(&dev->n_pds_lock);
+
+ /* ib_alloc_pd() will initialize pd->ibpd. */
+ pd->user = !!udata;
+
+bail:
+ return ret;
+}
+
+/**
+ * rvt_dealloc_pd - Free PD
+ * @ibpd: Free up PD
+ * @udata: Valid user data or NULL for kernel object
+ *
+ * Return: always 0
+ */
+int rvt_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
+{
+ struct rvt_dev_info *dev = ib_to_rvt(ibpd->device);
+
+ spin_lock(&dev->n_pds_lock);
+ dev->n_pds_allocated--;
+ spin_unlock(&dev->n_pds_lock);
+ return 0;
+}
diff --git a/drivers/infiniband/sw/rdmavt/pd.h b/drivers/infiniband/sw/rdmavt/pd.h
new file mode 100644
index 0000000000..42a0ef3b7d
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/pd.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RDMAVTPD_H
+#define DEF_RDMAVTPD_H
+
+#include <rdma/rdma_vt.h>
+
+int rvt_alloc_pd(struct ib_pd *pd, struct ib_udata *udata);
+int rvt_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata);
+
+#endif /* DEF_RDMAVTPD_H */
diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c
new file mode 100644
index 0000000000..dc83d0ac6a
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/qp.c
@@ -0,0 +1,3220 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 - 2020 Intel Corporation.
+ */
+
+#include <linux/hash.h>
+#include <linux/bitops.h>
+#include <linux/lockdep.h>
+#include <linux/vmalloc.h>
+#include <linux/slab.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_hdrs.h>
+#include <rdma/opa_addr.h>
+#include <rdma/uverbs_ioctl.h>
+#include "qp.h"
+#include "vt.h"
+#include "trace.h"
+
+#define RVT_RWQ_COUNT_THRESHOLD 16
+
+static void rvt_rc_timeout(struct timer_list *t);
+static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+ enum ib_qp_type type);
+
+/*
+ * Convert the AETH RNR timeout code into the number of microseconds.
+ */
+static const u32 ib_rvt_rnr_table[32] = {
+ 655360, /* 00: 655.36 */
+ 10, /* 01: .01 */
+ 20, /* 02 .02 */
+ 30, /* 03: .03 */
+ 40, /* 04: .04 */
+ 60, /* 05: .06 */
+ 80, /* 06: .08 */
+ 120, /* 07: .12 */
+ 160, /* 08: .16 */
+ 240, /* 09: .24 */
+ 320, /* 0A: .32 */
+ 480, /* 0B: .48 */
+ 640, /* 0C: .64 */
+ 960, /* 0D: .96 */
+ 1280, /* 0E: 1.28 */
+ 1920, /* 0F: 1.92 */
+ 2560, /* 10: 2.56 */
+ 3840, /* 11: 3.84 */
+ 5120, /* 12: 5.12 */
+ 7680, /* 13: 7.68 */
+ 10240, /* 14: 10.24 */
+ 15360, /* 15: 15.36 */
+ 20480, /* 16: 20.48 */
+ 30720, /* 17: 30.72 */
+ 40960, /* 18: 40.96 */
+ 61440, /* 19: 61.44 */
+ 81920, /* 1A: 81.92 */
+ 122880, /* 1B: 122.88 */
+ 163840, /* 1C: 163.84 */
+ 245760, /* 1D: 245.76 */
+ 327680, /* 1E: 327.68 */
+ 491520 /* 1F: 491.52 */
+};
+
+/*
+ * Note that it is OK to post send work requests in the SQE and ERR
+ * states; rvt_do_send() will process them and generate error
+ * completions as per IB 1.2 C10-96.
+ */
+const int ib_rvt_state_ops[IB_QPS_ERR + 1] = {
+ [IB_QPS_RESET] = 0,
+ [IB_QPS_INIT] = RVT_POST_RECV_OK,
+ [IB_QPS_RTR] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK,
+ [IB_QPS_RTS] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
+ RVT_POST_SEND_OK | RVT_PROCESS_SEND_OK |
+ RVT_PROCESS_NEXT_SEND_OK,
+ [IB_QPS_SQD] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
+ RVT_POST_SEND_OK | RVT_PROCESS_SEND_OK,
+ [IB_QPS_SQE] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
+ RVT_POST_SEND_OK | RVT_FLUSH_SEND,
+ [IB_QPS_ERR] = RVT_POST_RECV_OK | RVT_FLUSH_RECV |
+ RVT_POST_SEND_OK | RVT_FLUSH_SEND,
+};
+EXPORT_SYMBOL(ib_rvt_state_ops);
+
+/* platform specific: return the last level cache (llc) size, in KiB */
+static int rvt_wss_llc_size(void)
+{
+ /* assume that the boot CPU value is universal for all CPUs */
+ return boot_cpu_data.x86_cache_size;
+}
+
+/* platform specific: cacheless copy */
+static void cacheless_memcpy(void *dst, void *src, size_t n)
+{
+ /*
+ * Use the only available X64 cacheless copy. Add a __user cast
+ * to quiet sparse. The src agument is already in the kernel so
+ * there are no security issues. The extra fault recovery machinery
+ * is not invoked.
+ */
+ __copy_user_nocache(dst, (void __user *)src, n);
+}
+
+void rvt_wss_exit(struct rvt_dev_info *rdi)
+{
+ struct rvt_wss *wss = rdi->wss;
+
+ if (!wss)
+ return;
+
+ /* coded to handle partially initialized and repeat callers */
+ kfree(wss->entries);
+ wss->entries = NULL;
+ kfree(rdi->wss);
+ rdi->wss = NULL;
+}
+
+/*
+ * rvt_wss_init - Init wss data structures
+ *
+ * Return: 0 on success
+ */
+int rvt_wss_init(struct rvt_dev_info *rdi)
+{
+ unsigned int sge_copy_mode = rdi->dparms.sge_copy_mode;
+ unsigned int wss_threshold = rdi->dparms.wss_threshold;
+ unsigned int wss_clean_period = rdi->dparms.wss_clean_period;
+ long llc_size;
+ long llc_bits;
+ long table_size;
+ long table_bits;
+ struct rvt_wss *wss;
+ int node = rdi->dparms.node;
+
+ if (sge_copy_mode != RVT_SGE_COPY_ADAPTIVE) {
+ rdi->wss = NULL;
+ return 0;
+ }
+
+ rdi->wss = kzalloc_node(sizeof(*rdi->wss), GFP_KERNEL, node);
+ if (!rdi->wss)
+ return -ENOMEM;
+ wss = rdi->wss;
+
+ /* check for a valid percent range - default to 80 if none or invalid */
+ if (wss_threshold < 1 || wss_threshold > 100)
+ wss_threshold = 80;
+
+ /* reject a wildly large period */
+ if (wss_clean_period > 1000000)
+ wss_clean_period = 256;
+
+ /* reject a zero period */
+ if (wss_clean_period == 0)
+ wss_clean_period = 1;
+
+ /*
+ * Calculate the table size - the next power of 2 larger than the
+ * LLC size. LLC size is in KiB.
+ */
+ llc_size = rvt_wss_llc_size() * 1024;
+ table_size = roundup_pow_of_two(llc_size);
+
+ /* one bit per page in rounded up table */
+ llc_bits = llc_size / PAGE_SIZE;
+ table_bits = table_size / PAGE_SIZE;
+ wss->pages_mask = table_bits - 1;
+ wss->num_entries = table_bits / BITS_PER_LONG;
+
+ wss->threshold = (llc_bits * wss_threshold) / 100;
+ if (wss->threshold == 0)
+ wss->threshold = 1;
+
+ wss->clean_period = wss_clean_period;
+ atomic_set(&wss->clean_counter, wss_clean_period);
+
+ wss->entries = kcalloc_node(wss->num_entries, sizeof(*wss->entries),
+ GFP_KERNEL, node);
+ if (!wss->entries) {
+ rvt_wss_exit(rdi);
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+/*
+ * Advance the clean counter. When the clean period has expired,
+ * clean an entry.
+ *
+ * This is implemented in atomics to avoid locking. Because multiple
+ * variables are involved, it can be racy which can lead to slightly
+ * inaccurate information. Since this is only a heuristic, this is
+ * OK. Any innaccuracies will clean themselves out as the counter
+ * advances. That said, it is unlikely the entry clean operation will
+ * race - the next possible racer will not start until the next clean
+ * period.
+ *
+ * The clean counter is implemented as a decrement to zero. When zero
+ * is reached an entry is cleaned.
+ */
+static void wss_advance_clean_counter(struct rvt_wss *wss)
+{
+ int entry;
+ int weight;
+ unsigned long bits;
+
+ /* become the cleaner if we decrement the counter to zero */
+ if (atomic_dec_and_test(&wss->clean_counter)) {
+ /*
+ * Set, not add, the clean period. This avoids an issue
+ * where the counter could decrement below the clean period.
+ * Doing a set can result in lost decrements, slowing the
+ * clean advance. Since this a heuristic, this possible
+ * slowdown is OK.
+ *
+ * An alternative is to loop, advancing the counter by a
+ * clean period until the result is > 0. However, this could
+ * lead to several threads keeping another in the clean loop.
+ * This could be mitigated by limiting the number of times
+ * we stay in the loop.
+ */
+ atomic_set(&wss->clean_counter, wss->clean_period);
+
+ /*
+ * Uniquely grab the entry to clean and move to next.
+ * The current entry is always the lower bits of
+ * wss.clean_entry. The table size, wss.num_entries,
+ * is always a power-of-2.
+ */
+ entry = (atomic_inc_return(&wss->clean_entry) - 1)
+ & (wss->num_entries - 1);
+
+ /* clear the entry and count the bits */
+ bits = xchg(&wss->entries[entry], 0);
+ weight = hweight64((u64)bits);
+ /* only adjust the contended total count if needed */
+ if (weight)
+ atomic_sub(weight, &wss->total_count);
+ }
+}
+
+/*
+ * Insert the given address into the working set array.
+ */
+static void wss_insert(struct rvt_wss *wss, void *address)
+{
+ u32 page = ((unsigned long)address >> PAGE_SHIFT) & wss->pages_mask;
+ u32 entry = page / BITS_PER_LONG; /* assumes this ends up a shift */
+ u32 nr = page & (BITS_PER_LONG - 1);
+
+ if (!test_and_set_bit(nr, &wss->entries[entry]))
+ atomic_inc(&wss->total_count);
+
+ wss_advance_clean_counter(wss);
+}
+
+/*
+ * Is the working set larger than the threshold?
+ */
+static inline bool wss_exceeds_threshold(struct rvt_wss *wss)
+{
+ return atomic_read(&wss->total_count) >= wss->threshold;
+}
+
+static void get_map_page(struct rvt_qpn_table *qpt,
+ struct rvt_qpn_map *map)
+{
+ unsigned long page = get_zeroed_page(GFP_KERNEL);
+
+ /*
+ * Free the page if someone raced with us installing it.
+ */
+
+ spin_lock(&qpt->lock);
+ if (map->page)
+ free_page(page);
+ else
+ map->page = (void *)page;
+ spin_unlock(&qpt->lock);
+}
+
+/**
+ * init_qpn_table - initialize the QP number table for a device
+ * @rdi: rvt dev struct
+ * @qpt: the QPN table
+ */
+static int init_qpn_table(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt)
+{
+ u32 offset, i;
+ struct rvt_qpn_map *map;
+ int ret = 0;
+
+ if (!(rdi->dparms.qpn_res_end >= rdi->dparms.qpn_res_start))
+ return -EINVAL;
+
+ spin_lock_init(&qpt->lock);
+
+ qpt->last = rdi->dparms.qpn_start;
+ qpt->incr = rdi->dparms.qpn_inc << rdi->dparms.qos_shift;
+
+ /*
+ * Drivers may want some QPs beyond what we need for verbs let them use
+ * our qpn table. No need for two. Lets go ahead and mark the bitmaps
+ * for those. The reserved range must be *after* the range which verbs
+ * will pick from.
+ */
+
+ /* Figure out number of bit maps needed before reserved range */
+ qpt->nmaps = rdi->dparms.qpn_res_start / RVT_BITS_PER_PAGE;
+
+ /* This should always be zero */
+ offset = rdi->dparms.qpn_res_start & RVT_BITS_PER_PAGE_MASK;
+
+ /* Starting with the first reserved bit map */
+ map = &qpt->map[qpt->nmaps];
+
+ rvt_pr_info(rdi, "Reserving QPNs from 0x%x to 0x%x for non-verbs use\n",
+ rdi->dparms.qpn_res_start, rdi->dparms.qpn_res_end);
+ for (i = rdi->dparms.qpn_res_start; i <= rdi->dparms.qpn_res_end; i++) {
+ if (!map->page) {
+ get_map_page(qpt, map);
+ if (!map->page) {
+ ret = -ENOMEM;
+ break;
+ }
+ }
+ set_bit(offset, map->page);
+ offset++;
+ if (offset == RVT_BITS_PER_PAGE) {
+ /* next page */
+ qpt->nmaps++;
+ map++;
+ offset = 0;
+ }
+ }
+ return ret;
+}
+
+/**
+ * free_qpn_table - free the QP number table for a device
+ * @qpt: the QPN table
+ */
+static void free_qpn_table(struct rvt_qpn_table *qpt)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(qpt->map); i++)
+ free_page((unsigned long)qpt->map[i].page);
+}
+
+/**
+ * rvt_driver_qp_init - Init driver qp resources
+ * @rdi: rvt dev strucutre
+ *
+ * Return: 0 on success
+ */
+int rvt_driver_qp_init(struct rvt_dev_info *rdi)
+{
+ int i;
+ int ret = -ENOMEM;
+
+ if (!rdi->dparms.qp_table_size)
+ return -EINVAL;
+
+ /*
+ * If driver is not doing any QP allocation then make sure it is
+ * providing the necessary QP functions.
+ */
+ if (!rdi->driver_f.free_all_qps ||
+ !rdi->driver_f.qp_priv_alloc ||
+ !rdi->driver_f.qp_priv_free ||
+ !rdi->driver_f.notify_qp_reset ||
+ !rdi->driver_f.notify_restart_rc)
+ return -EINVAL;
+
+ /* allocate parent object */
+ rdi->qp_dev = kzalloc_node(sizeof(*rdi->qp_dev), GFP_KERNEL,
+ rdi->dparms.node);
+ if (!rdi->qp_dev)
+ return -ENOMEM;
+
+ /* allocate hash table */
+ rdi->qp_dev->qp_table_size = rdi->dparms.qp_table_size;
+ rdi->qp_dev->qp_table_bits = ilog2(rdi->dparms.qp_table_size);
+ rdi->qp_dev->qp_table =
+ kmalloc_array_node(rdi->qp_dev->qp_table_size,
+ sizeof(*rdi->qp_dev->qp_table),
+ GFP_KERNEL, rdi->dparms.node);
+ if (!rdi->qp_dev->qp_table)
+ goto no_qp_table;
+
+ for (i = 0; i < rdi->qp_dev->qp_table_size; i++)
+ RCU_INIT_POINTER(rdi->qp_dev->qp_table[i], NULL);
+
+ spin_lock_init(&rdi->qp_dev->qpt_lock);
+
+ /* initialize qpn map */
+ if (init_qpn_table(rdi, &rdi->qp_dev->qpn_table))
+ goto fail_table;
+
+ spin_lock_init(&rdi->n_qps_lock);
+
+ return 0;
+
+fail_table:
+ kfree(rdi->qp_dev->qp_table);
+ free_qpn_table(&rdi->qp_dev->qpn_table);
+
+no_qp_table:
+ kfree(rdi->qp_dev);
+
+ return ret;
+}
+
+/**
+ * rvt_free_qp_cb - callback function to reset a qp
+ * @qp: the qp to reset
+ * @v: a 64-bit value
+ *
+ * This function resets the qp and removes it from the
+ * qp hash table.
+ */
+static void rvt_free_qp_cb(struct rvt_qp *qp, u64 v)
+{
+ unsigned int *qp_inuse = (unsigned int *)v;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+
+ /* Reset the qp and remove it from the qp hash list */
+ rvt_reset_qp(rdi, qp, qp->ibqp.qp_type);
+
+ /* Increment the qp_inuse count */
+ (*qp_inuse)++;
+}
+
+/**
+ * rvt_free_all_qps - check for QPs still in use
+ * @rdi: rvt device info structure
+ *
+ * There should not be any QPs still in use.
+ * Free memory for table.
+ * Return the number of QPs still in use.
+ */
+static unsigned rvt_free_all_qps(struct rvt_dev_info *rdi)
+{
+ unsigned int qp_inuse = 0;
+
+ qp_inuse += rvt_mcast_tree_empty(rdi);
+
+ rvt_qp_iter(rdi, (u64)&qp_inuse, rvt_free_qp_cb);
+
+ return qp_inuse;
+}
+
+/**
+ * rvt_qp_exit - clean up qps on device exit
+ * @rdi: rvt dev structure
+ *
+ * Check for qp leaks and free resources.
+ */
+void rvt_qp_exit(struct rvt_dev_info *rdi)
+{
+ u32 qps_inuse = rvt_free_all_qps(rdi);
+
+ if (qps_inuse)
+ rvt_pr_err(rdi, "QP memory leak! %u still in use\n",
+ qps_inuse);
+
+ kfree(rdi->qp_dev->qp_table);
+ free_qpn_table(&rdi->qp_dev->qpn_table);
+ kfree(rdi->qp_dev);
+}
+
+static inline unsigned mk_qpn(struct rvt_qpn_table *qpt,
+ struct rvt_qpn_map *map, unsigned off)
+{
+ return (map - qpt->map) * RVT_BITS_PER_PAGE + off;
+}
+
+/**
+ * alloc_qpn - Allocate the next available qpn or zero/one for QP type
+ * IB_QPT_SMI/IB_QPT_GSI
+ * @rdi: rvt device info structure
+ * @qpt: queue pair number table pointer
+ * @type: the QP type
+ * @port_num: IB port number, 1 based, comes from core
+ * @exclude_prefix: prefix of special queue pair number being allocated
+ *
+ * Return: The queue pair number
+ */
+static int alloc_qpn(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
+ enum ib_qp_type type, u8 port_num, u8 exclude_prefix)
+{
+ u32 i, offset, max_scan, qpn;
+ struct rvt_qpn_map *map;
+ u32 ret;
+ u32 max_qpn = exclude_prefix == RVT_AIP_QP_PREFIX ?
+ RVT_AIP_QPN_MAX : RVT_QPN_MAX;
+
+ if (rdi->driver_f.alloc_qpn)
+ return rdi->driver_f.alloc_qpn(rdi, qpt, type, port_num);
+
+ if (type == IB_QPT_SMI || type == IB_QPT_GSI) {
+ unsigned n;
+
+ ret = type == IB_QPT_GSI;
+ n = 1 << (ret + 2 * (port_num - 1));
+ spin_lock(&qpt->lock);
+ if (qpt->flags & n)
+ ret = -EINVAL;
+ else
+ qpt->flags |= n;
+ spin_unlock(&qpt->lock);
+ goto bail;
+ }
+
+ qpn = qpt->last + qpt->incr;
+ if (qpn >= max_qpn)
+ qpn = qpt->incr | ((qpt->last & 1) ^ 1);
+ /* offset carries bit 0 */
+ offset = qpn & RVT_BITS_PER_PAGE_MASK;
+ map = &qpt->map[qpn / RVT_BITS_PER_PAGE];
+ max_scan = qpt->nmaps - !offset;
+ for (i = 0;;) {
+ if (unlikely(!map->page)) {
+ get_map_page(qpt, map);
+ if (unlikely(!map->page))
+ break;
+ }
+ do {
+ if (!test_and_set_bit(offset, map->page)) {
+ qpt->last = qpn;
+ ret = qpn;
+ goto bail;
+ }
+ offset += qpt->incr;
+ /*
+ * This qpn might be bogus if offset >= BITS_PER_PAGE.
+ * That is OK. It gets re-assigned below
+ */
+ qpn = mk_qpn(qpt, map, offset);
+ } while (offset < RVT_BITS_PER_PAGE && qpn < RVT_QPN_MAX);
+ /*
+ * In order to keep the number of pages allocated to a
+ * minimum, we scan the all existing pages before increasing
+ * the size of the bitmap table.
+ */
+ if (++i > max_scan) {
+ if (qpt->nmaps == RVT_QPNMAP_ENTRIES)
+ break;
+ map = &qpt->map[qpt->nmaps++];
+ /* start at incr with current bit 0 */
+ offset = qpt->incr | (offset & 1);
+ } else if (map < &qpt->map[qpt->nmaps]) {
+ ++map;
+ /* start at incr with current bit 0 */
+ offset = qpt->incr | (offset & 1);
+ } else {
+ map = &qpt->map[0];
+ /* wrap to first map page, invert bit 0 */
+ offset = qpt->incr | ((offset & 1) ^ 1);
+ }
+ /* there can be no set bits in low-order QoS bits */
+ WARN_ON(rdi->dparms.qos_shift > 1 &&
+ offset & ((BIT(rdi->dparms.qos_shift - 1) - 1) << 1));
+ qpn = mk_qpn(qpt, map, offset);
+ }
+
+ ret = -ENOMEM;
+
+bail:
+ return ret;
+}
+
+/**
+ * rvt_clear_mr_refs - Drop help mr refs
+ * @qp: rvt qp data structure
+ * @clr_sends: If shoudl clear send side or not
+ */
+static void rvt_clear_mr_refs(struct rvt_qp *qp, int clr_sends)
+{
+ unsigned n;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+
+ if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
+ rvt_put_ss(&qp->s_rdma_read_sge);
+
+ rvt_put_ss(&qp->r_sge);
+
+ if (clr_sends) {
+ while (qp->s_last != qp->s_head) {
+ struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, qp->s_last);
+
+ rvt_put_qp_swqe(qp, wqe);
+ if (++qp->s_last >= qp->s_size)
+ qp->s_last = 0;
+ smp_wmb(); /* see qp_set_savail */
+ }
+ if (qp->s_rdma_mr) {
+ rvt_put_mr(qp->s_rdma_mr);
+ qp->s_rdma_mr = NULL;
+ }
+ }
+
+ for (n = 0; qp->s_ack_queue && n < rvt_max_atomic(rdi); n++) {
+ struct rvt_ack_entry *e = &qp->s_ack_queue[n];
+
+ if (e->rdma_sge.mr) {
+ rvt_put_mr(e->rdma_sge.mr);
+ e->rdma_sge.mr = NULL;
+ }
+ }
+}
+
+/**
+ * rvt_swqe_has_lkey - return true if lkey is used by swqe
+ * @wqe: the send wqe
+ * @lkey: the lkey
+ *
+ * Test the swqe for using lkey
+ */
+static bool rvt_swqe_has_lkey(struct rvt_swqe *wqe, u32 lkey)
+{
+ int i;
+
+ for (i = 0; i < wqe->wr.num_sge; i++) {
+ struct rvt_sge *sge = &wqe->sg_list[i];
+
+ if (rvt_mr_has_lkey(sge->mr, lkey))
+ return true;
+ }
+ return false;
+}
+
+/**
+ * rvt_qp_sends_has_lkey - return true is qp sends use lkey
+ * @qp: the rvt_qp
+ * @lkey: the lkey
+ */
+static bool rvt_qp_sends_has_lkey(struct rvt_qp *qp, u32 lkey)
+{
+ u32 s_last = qp->s_last;
+
+ while (s_last != qp->s_head) {
+ struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, s_last);
+
+ if (rvt_swqe_has_lkey(wqe, lkey))
+ return true;
+
+ if (++s_last >= qp->s_size)
+ s_last = 0;
+ }
+ if (qp->s_rdma_mr)
+ if (rvt_mr_has_lkey(qp->s_rdma_mr, lkey))
+ return true;
+ return false;
+}
+
+/**
+ * rvt_qp_acks_has_lkey - return true if acks have lkey
+ * @qp: the qp
+ * @lkey: the lkey
+ */
+static bool rvt_qp_acks_has_lkey(struct rvt_qp *qp, u32 lkey)
+{
+ int i;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+
+ for (i = 0; qp->s_ack_queue && i < rvt_max_atomic(rdi); i++) {
+ struct rvt_ack_entry *e = &qp->s_ack_queue[i];
+
+ if (rvt_mr_has_lkey(e->rdma_sge.mr, lkey))
+ return true;
+ }
+ return false;
+}
+
+/**
+ * rvt_qp_mr_clean - clean up remote ops for lkey
+ * @qp: the qp
+ * @lkey: the lkey that is being de-registered
+ *
+ * This routine checks if the lkey is being used by
+ * the qp.
+ *
+ * If so, the qp is put into an error state to elminate
+ * any references from the qp.
+ */
+void rvt_qp_mr_clean(struct rvt_qp *qp, u32 lkey)
+{
+ bool lastwqe = false;
+
+ if (qp->ibqp.qp_type == IB_QPT_SMI ||
+ qp->ibqp.qp_type == IB_QPT_GSI)
+ /* avoid special QPs */
+ return;
+ spin_lock_irq(&qp->r_lock);
+ spin_lock(&qp->s_hlock);
+ spin_lock(&qp->s_lock);
+
+ if (qp->state == IB_QPS_ERR || qp->state == IB_QPS_RESET)
+ goto check_lwqe;
+
+ if (rvt_ss_has_lkey(&qp->r_sge, lkey) ||
+ rvt_qp_sends_has_lkey(qp, lkey) ||
+ rvt_qp_acks_has_lkey(qp, lkey))
+ lastwqe = rvt_error_qp(qp, IB_WC_LOC_PROT_ERR);
+check_lwqe:
+ spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->s_hlock);
+ spin_unlock_irq(&qp->r_lock);
+ if (lastwqe) {
+ struct ib_event ev;
+
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
+}
+
+/**
+ * rvt_remove_qp - remove qp form table
+ * @rdi: rvt dev struct
+ * @qp: qp to remove
+ *
+ * Remove the QP from the table so it can't be found asynchronously by
+ * the receive routine.
+ */
+static void rvt_remove_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
+{
+ struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
+ u32 n = hash_32(qp->ibqp.qp_num, rdi->qp_dev->qp_table_bits);
+ unsigned long flags;
+ int removed = 1;
+
+ spin_lock_irqsave(&rdi->qp_dev->qpt_lock, flags);
+
+ if (rcu_dereference_protected(rvp->qp[0],
+ lockdep_is_held(&rdi->qp_dev->qpt_lock)) == qp) {
+ RCU_INIT_POINTER(rvp->qp[0], NULL);
+ } else if (rcu_dereference_protected(rvp->qp[1],
+ lockdep_is_held(&rdi->qp_dev->qpt_lock)) == qp) {
+ RCU_INIT_POINTER(rvp->qp[1], NULL);
+ } else {
+ struct rvt_qp *q;
+ struct rvt_qp __rcu **qpp;
+
+ removed = 0;
+ qpp = &rdi->qp_dev->qp_table[n];
+ for (; (q = rcu_dereference_protected(*qpp,
+ lockdep_is_held(&rdi->qp_dev->qpt_lock))) != NULL;
+ qpp = &q->next) {
+ if (q == qp) {
+ RCU_INIT_POINTER(*qpp,
+ rcu_dereference_protected(qp->next,
+ lockdep_is_held(&rdi->qp_dev->qpt_lock)));
+ removed = 1;
+ trace_rvt_qpremove(qp, n);
+ break;
+ }
+ }
+ }
+
+ spin_unlock_irqrestore(&rdi->qp_dev->qpt_lock, flags);
+ if (removed) {
+ synchronize_rcu();
+ rvt_put_qp(qp);
+ }
+}
+
+/**
+ * rvt_alloc_rq - allocate memory for user or kernel buffer
+ * @rq: receive queue data structure
+ * @size: number of request queue entries
+ * @node: The NUMA node
+ * @udata: True if user data is available or not false
+ *
+ * Return: If memory allocation failed, return -ENONEM
+ * This function is used by both shared receive
+ * queues and non-shared receive queues to allocate
+ * memory.
+ */
+int rvt_alloc_rq(struct rvt_rq *rq, u32 size, int node,
+ struct ib_udata *udata)
+{
+ if (udata) {
+ rq->wq = vmalloc_user(sizeof(struct rvt_rwq) + size);
+ if (!rq->wq)
+ goto bail;
+ /* need kwq with no buffers */
+ rq->kwq = kzalloc_node(sizeof(*rq->kwq), GFP_KERNEL, node);
+ if (!rq->kwq)
+ goto bail;
+ rq->kwq->curr_wq = rq->wq->wq;
+ } else {
+ /* need kwq with buffers */
+ rq->kwq =
+ vzalloc_node(sizeof(struct rvt_krwq) + size, node);
+ if (!rq->kwq)
+ goto bail;
+ rq->kwq->curr_wq = rq->kwq->wq;
+ }
+
+ spin_lock_init(&rq->kwq->p_lock);
+ spin_lock_init(&rq->kwq->c_lock);
+ return 0;
+bail:
+ rvt_free_rq(rq);
+ return -ENOMEM;
+}
+
+/**
+ * rvt_init_qp - initialize the QP state to the reset state
+ * @rdi: rvt dev struct
+ * @qp: the QP to init or reinit
+ * @type: the QP type
+ *
+ * This function is called from both rvt_create_qp() and
+ * rvt_reset_qp(). The difference is that the reset
+ * patch the necessary locks to protect against concurent
+ * access.
+ */
+static void rvt_init_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+ enum ib_qp_type type)
+{
+ qp->remote_qpn = 0;
+ qp->qkey = 0;
+ qp->qp_access_flags = 0;
+ qp->s_flags &= RVT_S_SIGNAL_REQ_WR;
+ qp->s_hdrwords = 0;
+ qp->s_wqe = NULL;
+ qp->s_draining = 0;
+ qp->s_next_psn = 0;
+ qp->s_last_psn = 0;
+ qp->s_sending_psn = 0;
+ qp->s_sending_hpsn = 0;
+ qp->s_psn = 0;
+ qp->r_psn = 0;
+ qp->r_msn = 0;
+ if (type == IB_QPT_RC) {
+ qp->s_state = IB_OPCODE_RC_SEND_LAST;
+ qp->r_state = IB_OPCODE_RC_SEND_LAST;
+ } else {
+ qp->s_state = IB_OPCODE_UC_SEND_LAST;
+ qp->r_state = IB_OPCODE_UC_SEND_LAST;
+ }
+ qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
+ qp->r_nak_state = 0;
+ qp->r_aflags = 0;
+ qp->r_flags = 0;
+ qp->s_head = 0;
+ qp->s_tail = 0;
+ qp->s_cur = 0;
+ qp->s_acked = 0;
+ qp->s_last = 0;
+ qp->s_ssn = 1;
+ qp->s_lsn = 0;
+ qp->s_mig_state = IB_MIG_MIGRATED;
+ qp->r_head_ack_queue = 0;
+ qp->s_tail_ack_queue = 0;
+ qp->s_acked_ack_queue = 0;
+ qp->s_num_rd_atomic = 0;
+ qp->r_sge.num_sge = 0;
+ atomic_set(&qp->s_reserved_used, 0);
+}
+
+/**
+ * _rvt_reset_qp - initialize the QP state to the reset state
+ * @rdi: rvt dev struct
+ * @qp: the QP to reset
+ * @type: the QP type
+ *
+ * r_lock, s_hlock, and s_lock are required to be held by the caller
+ */
+static void _rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+ enum ib_qp_type type)
+ __must_hold(&qp->s_lock)
+ __must_hold(&qp->s_hlock)
+ __must_hold(&qp->r_lock)
+{
+ lockdep_assert_held(&qp->r_lock);
+ lockdep_assert_held(&qp->s_hlock);
+ lockdep_assert_held(&qp->s_lock);
+ if (qp->state != IB_QPS_RESET) {
+ qp->state = IB_QPS_RESET;
+
+ /* Let drivers flush their waitlist */
+ rdi->driver_f.flush_qp_waiters(qp);
+ rvt_stop_rc_timers(qp);
+ qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
+ spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->s_hlock);
+ spin_unlock_irq(&qp->r_lock);
+
+ /* Stop the send queue and the retry timer */
+ rdi->driver_f.stop_send_queue(qp);
+ rvt_del_timers_sync(qp);
+ /* Wait for things to stop */
+ rdi->driver_f.quiesce_qp(qp);
+
+ /* take qp out the hash and wait for it to be unused */
+ rvt_remove_qp(rdi, qp);
+
+ /* grab the lock b/c it was locked at call time */
+ spin_lock_irq(&qp->r_lock);
+ spin_lock(&qp->s_hlock);
+ spin_lock(&qp->s_lock);
+
+ rvt_clear_mr_refs(qp, 1);
+ /*
+ * Let the driver do any tear down or re-init it needs to for
+ * a qp that has been reset
+ */
+ rdi->driver_f.notify_qp_reset(qp);
+ }
+ rvt_init_qp(rdi, qp, type);
+ lockdep_assert_held(&qp->r_lock);
+ lockdep_assert_held(&qp->s_hlock);
+ lockdep_assert_held(&qp->s_lock);
+}
+
+/**
+ * rvt_reset_qp - initialize the QP state to the reset state
+ * @rdi: the device info
+ * @qp: the QP to reset
+ * @type: the QP type
+ *
+ * This is the wrapper function to acquire the r_lock, s_hlock, and s_lock
+ * before calling _rvt_reset_qp().
+ */
+static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+ enum ib_qp_type type)
+{
+ spin_lock_irq(&qp->r_lock);
+ spin_lock(&qp->s_hlock);
+ spin_lock(&qp->s_lock);
+ _rvt_reset_qp(rdi, qp, type);
+ spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->s_hlock);
+ spin_unlock_irq(&qp->r_lock);
+}
+
+/**
+ * rvt_free_qpn - Free a qpn from the bit map
+ * @qpt: QP table
+ * @qpn: queue pair number to free
+ */
+static void rvt_free_qpn(struct rvt_qpn_table *qpt, u32 qpn)
+{
+ struct rvt_qpn_map *map;
+
+ if ((qpn & RVT_AIP_QP_PREFIX_MASK) == RVT_AIP_QP_BASE)
+ qpn &= RVT_AIP_QP_SUFFIX;
+
+ map = qpt->map + (qpn & RVT_QPN_MASK) / RVT_BITS_PER_PAGE;
+ if (map->page)
+ clear_bit(qpn & RVT_BITS_PER_PAGE_MASK, map->page);
+}
+
+/**
+ * get_allowed_ops - Given a QP type return the appropriate allowed OP
+ * @type: valid, supported, QP type
+ */
+static u8 get_allowed_ops(enum ib_qp_type type)
+{
+ return type == IB_QPT_RC ? IB_OPCODE_RC : type == IB_QPT_UC ?
+ IB_OPCODE_UC : IB_OPCODE_UD;
+}
+
+/**
+ * free_ud_wq_attr - Clean up AH attribute cache for UD QPs
+ * @qp: Valid QP with allowed_ops set
+ *
+ * The rvt_swqe data structure being used is a union, so this is
+ * only valid for UD QPs.
+ */
+static void free_ud_wq_attr(struct rvt_qp *qp)
+{
+ struct rvt_swqe *wqe;
+ int i;
+
+ for (i = 0; qp->allowed_ops == IB_OPCODE_UD && i < qp->s_size; i++) {
+ wqe = rvt_get_swqe_ptr(qp, i);
+ kfree(wqe->ud_wr.attr);
+ wqe->ud_wr.attr = NULL;
+ }
+}
+
+/**
+ * alloc_ud_wq_attr - AH attribute cache for UD QPs
+ * @qp: Valid QP with allowed_ops set
+ * @node: Numa node for allocation
+ *
+ * The rvt_swqe data structure being used is a union, so this is
+ * only valid for UD QPs.
+ */
+static int alloc_ud_wq_attr(struct rvt_qp *qp, int node)
+{
+ struct rvt_swqe *wqe;
+ int i;
+
+ for (i = 0; qp->allowed_ops == IB_OPCODE_UD && i < qp->s_size; i++) {
+ wqe = rvt_get_swqe_ptr(qp, i);
+ wqe->ud_wr.attr = kzalloc_node(sizeof(*wqe->ud_wr.attr),
+ GFP_KERNEL, node);
+ if (!wqe->ud_wr.attr) {
+ free_ud_wq_attr(qp);
+ return -ENOMEM;
+ }
+ }
+
+ return 0;
+}
+
+/**
+ * rvt_create_qp - create a queue pair for a device
+ * @ibqp: the queue pair
+ * @init_attr: the attributes of the queue pair
+ * @udata: user data for libibverbs.so
+ *
+ * Queue pair creation is mostly an rvt issue. However, drivers have their own
+ * unique idea of what queue pair numbers mean. For instance there is a reserved
+ * range for PSM.
+ *
+ * Return: 0 on success, otherwise returns an errno.
+ *
+ * Called by the ib_create_qp() core verbs function.
+ */
+int rvt_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *init_attr,
+ struct ib_udata *udata)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ int ret = -ENOMEM;
+ struct rvt_swqe *swq = NULL;
+ size_t sz;
+ size_t sg_list_sz = 0;
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+ void *priv = NULL;
+ size_t sqsize;
+ u8 exclude_prefix = 0;
+
+ if (!rdi)
+ return -EINVAL;
+
+ if (init_attr->create_flags & ~IB_QP_CREATE_NETDEV_USE)
+ return -EOPNOTSUPP;
+
+ if (init_attr->cap.max_send_sge > rdi->dparms.props.max_send_sge ||
+ init_attr->cap.max_send_wr > rdi->dparms.props.max_qp_wr)
+ return -EINVAL;
+
+ /* Check receive queue parameters if no SRQ is specified. */
+ if (!init_attr->srq) {
+ if (init_attr->cap.max_recv_sge >
+ rdi->dparms.props.max_recv_sge ||
+ init_attr->cap.max_recv_wr > rdi->dparms.props.max_qp_wr)
+ return -EINVAL;
+
+ if (init_attr->cap.max_send_sge +
+ init_attr->cap.max_send_wr +
+ init_attr->cap.max_recv_sge +
+ init_attr->cap.max_recv_wr == 0)
+ return -EINVAL;
+ }
+ sqsize =
+ init_attr->cap.max_send_wr + 1 +
+ rdi->dparms.reserved_operations;
+ switch (init_attr->qp_type) {
+ case IB_QPT_SMI:
+ case IB_QPT_GSI:
+ if (init_attr->port_num == 0 ||
+ init_attr->port_num > ibqp->device->phys_port_cnt)
+ return -EINVAL;
+ fallthrough;
+ case IB_QPT_UC:
+ case IB_QPT_RC:
+ case IB_QPT_UD:
+ sz = struct_size(swq, sg_list, init_attr->cap.max_send_sge);
+ swq = vzalloc_node(array_size(sz, sqsize), rdi->dparms.node);
+ if (!swq)
+ return -ENOMEM;
+
+ if (init_attr->srq) {
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(init_attr->srq);
+
+ if (srq->rq.max_sge > 1)
+ sg_list_sz = sizeof(*qp->r_sg_list) *
+ (srq->rq.max_sge - 1);
+ } else if (init_attr->cap.max_recv_sge > 1)
+ sg_list_sz = sizeof(*qp->r_sg_list) *
+ (init_attr->cap.max_recv_sge - 1);
+ qp->r_sg_list =
+ kzalloc_node(sg_list_sz, GFP_KERNEL, rdi->dparms.node);
+ if (!qp->r_sg_list)
+ goto bail_qp;
+ qp->allowed_ops = get_allowed_ops(init_attr->qp_type);
+
+ RCU_INIT_POINTER(qp->next, NULL);
+ if (init_attr->qp_type == IB_QPT_RC) {
+ qp->s_ack_queue =
+ kcalloc_node(rvt_max_atomic(rdi),
+ sizeof(*qp->s_ack_queue),
+ GFP_KERNEL,
+ rdi->dparms.node);
+ if (!qp->s_ack_queue)
+ goto bail_qp;
+ }
+ /* initialize timers needed for rc qp */
+ timer_setup(&qp->s_timer, rvt_rc_timeout, 0);
+ hrtimer_init(&qp->s_rnr_timer, CLOCK_MONOTONIC,
+ HRTIMER_MODE_REL);
+ qp->s_rnr_timer.function = rvt_rc_rnr_retry;
+
+ /*
+ * Driver needs to set up it's private QP structure and do any
+ * initialization that is needed.
+ */
+ priv = rdi->driver_f.qp_priv_alloc(rdi, qp);
+ if (IS_ERR(priv)) {
+ ret = PTR_ERR(priv);
+ goto bail_qp;
+ }
+ qp->priv = priv;
+ qp->timeout_jiffies =
+ usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+ 1000UL);
+ if (init_attr->srq) {
+ sz = 0;
+ } else {
+ qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
+ qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
+ sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
+ sizeof(struct rvt_rwqe);
+ ret = rvt_alloc_rq(&qp->r_rq, qp->r_rq.size * sz,
+ rdi->dparms.node, udata);
+ if (ret)
+ goto bail_driver_priv;
+ }
+
+ /*
+ * ib_create_qp() will initialize qp->ibqp
+ * except for qp->ibqp.qp_num.
+ */
+ spin_lock_init(&qp->r_lock);
+ spin_lock_init(&qp->s_hlock);
+ spin_lock_init(&qp->s_lock);
+ atomic_set(&qp->refcount, 0);
+ atomic_set(&qp->local_ops_pending, 0);
+ init_waitqueue_head(&qp->wait);
+ INIT_LIST_HEAD(&qp->rspwait);
+ qp->state = IB_QPS_RESET;
+ qp->s_wq = swq;
+ qp->s_size = sqsize;
+ qp->s_avail = init_attr->cap.max_send_wr;
+ qp->s_max_sge = init_attr->cap.max_send_sge;
+ if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
+ qp->s_flags = RVT_S_SIGNAL_REQ_WR;
+ ret = alloc_ud_wq_attr(qp, rdi->dparms.node);
+ if (ret)
+ goto bail_rq_rvt;
+
+ if (init_attr->create_flags & IB_QP_CREATE_NETDEV_USE)
+ exclude_prefix = RVT_AIP_QP_PREFIX;
+
+ ret = alloc_qpn(rdi, &rdi->qp_dev->qpn_table,
+ init_attr->qp_type,
+ init_attr->port_num,
+ exclude_prefix);
+ if (ret < 0)
+ goto bail_rq_wq;
+
+ qp->ibqp.qp_num = ret;
+ if (init_attr->create_flags & IB_QP_CREATE_NETDEV_USE)
+ qp->ibqp.qp_num |= RVT_AIP_QP_BASE;
+ qp->port_num = init_attr->port_num;
+ rvt_init_qp(rdi, qp, init_attr->qp_type);
+ if (rdi->driver_f.qp_priv_init) {
+ ret = rdi->driver_f.qp_priv_init(rdi, qp, init_attr);
+ if (ret)
+ goto bail_rq_wq;
+ }
+ break;
+
+ default:
+ /* Don't support raw QPs */
+ return -EOPNOTSUPP;
+ }
+
+ init_attr->cap.max_inline_data = 0;
+
+ /*
+ * Return the address of the RWQ as the offset to mmap.
+ * See rvt_mmap() for details.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ if (!qp->r_rq.wq) {
+ __u64 offset = 0;
+
+ ret = ib_copy_to_udata(udata, &offset,
+ sizeof(offset));
+ if (ret)
+ goto bail_qpn;
+ } else {
+ u32 s = sizeof(struct rvt_rwq) + qp->r_rq.size * sz;
+
+ qp->ip = rvt_create_mmap_info(rdi, s, udata,
+ qp->r_rq.wq);
+ if (IS_ERR(qp->ip)) {
+ ret = PTR_ERR(qp->ip);
+ goto bail_qpn;
+ }
+
+ ret = ib_copy_to_udata(udata, &qp->ip->offset,
+ sizeof(qp->ip->offset));
+ if (ret)
+ goto bail_ip;
+ }
+ qp->pid = current->pid;
+ }
+
+ spin_lock(&rdi->n_qps_lock);
+ if (rdi->n_qps_allocated == rdi->dparms.props.max_qp) {
+ spin_unlock(&rdi->n_qps_lock);
+ ret = -ENOMEM;
+ goto bail_ip;
+ }
+
+ rdi->n_qps_allocated++;
+ /*
+ * Maintain a busy_jiffies variable that will be added to the timeout
+ * period in mod_retry_timer and add_retry_timer. This busy jiffies
+ * is scaled by the number of rc qps created for the device to reduce
+ * the number of timeouts occurring when there is a large number of
+ * qps. busy_jiffies is incremented every rc qp scaling interval.
+ * The scaling interval is selected based on extensive performance
+ * evaluation of targeted workloads.
+ */
+ if (init_attr->qp_type == IB_QPT_RC) {
+ rdi->n_rc_qps++;
+ rdi->busy_jiffies = rdi->n_rc_qps / RC_QP_SCALING_INTERVAL;
+ }
+ spin_unlock(&rdi->n_qps_lock);
+
+ if (qp->ip) {
+ spin_lock_irq(&rdi->pending_lock);
+ list_add(&qp->ip->pending_mmaps, &rdi->pending_mmaps);
+ spin_unlock_irq(&rdi->pending_lock);
+ }
+
+ return 0;
+
+bail_ip:
+ if (qp->ip)
+ kref_put(&qp->ip->ref, rvt_release_mmap_info);
+
+bail_qpn:
+ rvt_free_qpn(&rdi->qp_dev->qpn_table, qp->ibqp.qp_num);
+
+bail_rq_wq:
+ free_ud_wq_attr(qp);
+
+bail_rq_rvt:
+ rvt_free_rq(&qp->r_rq);
+
+bail_driver_priv:
+ rdi->driver_f.qp_priv_free(rdi, qp);
+
+bail_qp:
+ kfree(qp->s_ack_queue);
+ kfree(qp->r_sg_list);
+ vfree(swq);
+ return ret;
+}
+
+/**
+ * rvt_error_qp - put a QP into the error state
+ * @qp: the QP to put into the error state
+ * @err: the receive completion error to signal if a RWQE is active
+ *
+ * Flushes both send and receive work queues.
+ *
+ * Return: true if last WQE event should be generated.
+ * The QP r_lock and s_lock should be held and interrupts disabled.
+ * If we are already in error state, just return.
+ */
+int rvt_error_qp(struct rvt_qp *qp, enum ib_wc_status err)
+{
+ struct ib_wc wc;
+ int ret = 0;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+
+ lockdep_assert_held(&qp->r_lock);
+ lockdep_assert_held(&qp->s_lock);
+ if (qp->state == IB_QPS_ERR || qp->state == IB_QPS_RESET)
+ goto bail;
+
+ qp->state = IB_QPS_ERR;
+
+ if (qp->s_flags & (RVT_S_TIMER | RVT_S_WAIT_RNR)) {
+ qp->s_flags &= ~(RVT_S_TIMER | RVT_S_WAIT_RNR);
+ del_timer(&qp->s_timer);
+ }
+
+ if (qp->s_flags & RVT_S_ANY_WAIT_SEND)
+ qp->s_flags &= ~RVT_S_ANY_WAIT_SEND;
+
+ rdi->driver_f.notify_error_qp(qp);
+
+ /* Schedule the sending tasklet to drain the send work queue. */
+ if (READ_ONCE(qp->s_last) != qp->s_head)
+ rdi->driver_f.schedule_send(qp);
+
+ rvt_clear_mr_refs(qp, 0);
+
+ memset(&wc, 0, sizeof(wc));
+ wc.qp = &qp->ibqp;
+ wc.opcode = IB_WC_RECV;
+
+ if (test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags)) {
+ wc.wr_id = qp->r_wr_id;
+ wc.status = err;
+ rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
+ }
+ wc.status = IB_WC_WR_FLUSH_ERR;
+
+ if (qp->r_rq.kwq) {
+ u32 head;
+ u32 tail;
+ struct rvt_rwq *wq = NULL;
+ struct rvt_krwq *kwq = NULL;
+
+ spin_lock(&qp->r_rq.kwq->c_lock);
+ /* qp->ip used to validate if there is a user buffer mmaped */
+ if (qp->ip) {
+ wq = qp->r_rq.wq;
+ head = RDMA_READ_UAPI_ATOMIC(wq->head);
+ tail = RDMA_READ_UAPI_ATOMIC(wq->tail);
+ } else {
+ kwq = qp->r_rq.kwq;
+ head = kwq->head;
+ tail = kwq->tail;
+ }
+ /* sanity check pointers before trusting them */
+ if (head >= qp->r_rq.size)
+ head = 0;
+ if (tail >= qp->r_rq.size)
+ tail = 0;
+ while (tail != head) {
+ wc.wr_id = rvt_get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
+ if (++tail >= qp->r_rq.size)
+ tail = 0;
+ rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
+ }
+ if (qp->ip)
+ RDMA_WRITE_UAPI_ATOMIC(wq->tail, tail);
+ else
+ kwq->tail = tail;
+ spin_unlock(&qp->r_rq.kwq->c_lock);
+ } else if (qp->ibqp.event_handler) {
+ ret = 1;
+ }
+
+bail:
+ return ret;
+}
+EXPORT_SYMBOL(rvt_error_qp);
+
+/*
+ * Put the QP into the hash table.
+ * The hash table holds a reference to the QP.
+ */
+static void rvt_insert_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
+{
+ struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
+ unsigned long flags;
+
+ rvt_get_qp(qp);
+ spin_lock_irqsave(&rdi->qp_dev->qpt_lock, flags);
+
+ if (qp->ibqp.qp_num <= 1) {
+ rcu_assign_pointer(rvp->qp[qp->ibqp.qp_num], qp);
+ } else {
+ u32 n = hash_32(qp->ibqp.qp_num, rdi->qp_dev->qp_table_bits);
+
+ qp->next = rdi->qp_dev->qp_table[n];
+ rcu_assign_pointer(rdi->qp_dev->qp_table[n], qp);
+ trace_rvt_qpinsert(qp, n);
+ }
+
+ spin_unlock_irqrestore(&rdi->qp_dev->qpt_lock, flags);
+}
+
+/**
+ * rvt_modify_qp - modify the attributes of a queue pair
+ * @ibqp: the queue pair who's attributes we're modifying
+ * @attr: the new attributes
+ * @attr_mask: the mask of attributes to modify
+ * @udata: user data for libibverbs.so
+ *
+ * Return: 0 on success, otherwise returns an errno.
+ */
+int rvt_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ enum ib_qp_state cur_state, new_state;
+ struct ib_event ev;
+ int lastwqe = 0;
+ int mig = 0;
+ int pmtu = 0; /* for gcc warning only */
+ int opa_ah;
+
+ if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
+ return -EOPNOTSUPP;
+
+ spin_lock_irq(&qp->r_lock);
+ spin_lock(&qp->s_hlock);
+ spin_lock(&qp->s_lock);
+
+ cur_state = attr_mask & IB_QP_CUR_STATE ?
+ attr->cur_qp_state : qp->state;
+ new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
+ opa_ah = rdma_cap_opa_ah(ibqp->device, qp->port_num);
+
+ if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
+ attr_mask))
+ goto inval;
+
+ if (rdi->driver_f.check_modify_qp &&
+ rdi->driver_f.check_modify_qp(qp, attr, attr_mask, udata))
+ goto inval;
+
+ if (attr_mask & IB_QP_AV) {
+ if (opa_ah) {
+ if (rdma_ah_get_dlid(&attr->ah_attr) >=
+ opa_get_mcast_base(OPA_MCAST_NR))
+ goto inval;
+ } else {
+ if (rdma_ah_get_dlid(&attr->ah_attr) >=
+ be16_to_cpu(IB_MULTICAST_LID_BASE))
+ goto inval;
+ }
+
+ if (rvt_check_ah(qp->ibqp.device, &attr->ah_attr))
+ goto inval;
+ }
+
+ if (attr_mask & IB_QP_ALT_PATH) {
+ if (opa_ah) {
+ if (rdma_ah_get_dlid(&attr->alt_ah_attr) >=
+ opa_get_mcast_base(OPA_MCAST_NR))
+ goto inval;
+ } else {
+ if (rdma_ah_get_dlid(&attr->alt_ah_attr) >=
+ be16_to_cpu(IB_MULTICAST_LID_BASE))
+ goto inval;
+ }
+
+ if (rvt_check_ah(qp->ibqp.device, &attr->alt_ah_attr))
+ goto inval;
+ if (attr->alt_pkey_index >= rvt_get_npkeys(rdi))
+ goto inval;
+ }
+
+ if (attr_mask & IB_QP_PKEY_INDEX)
+ if (attr->pkey_index >= rvt_get_npkeys(rdi))
+ goto inval;
+
+ if (attr_mask & IB_QP_MIN_RNR_TIMER)
+ if (attr->min_rnr_timer > 31)
+ goto inval;
+
+ if (attr_mask & IB_QP_PORT)
+ if (qp->ibqp.qp_type == IB_QPT_SMI ||
+ qp->ibqp.qp_type == IB_QPT_GSI ||
+ attr->port_num == 0 ||
+ attr->port_num > ibqp->device->phys_port_cnt)
+ goto inval;
+
+ if (attr_mask & IB_QP_DEST_QPN)
+ if (attr->dest_qp_num > RVT_QPN_MASK)
+ goto inval;
+
+ if (attr_mask & IB_QP_RETRY_CNT)
+ if (attr->retry_cnt > 7)
+ goto inval;
+
+ if (attr_mask & IB_QP_RNR_RETRY)
+ if (attr->rnr_retry > 7)
+ goto inval;
+
+ /*
+ * Don't allow invalid path_mtu values. OK to set greater
+ * than the active mtu (or even the max_cap, if we have tuned
+ * that to a small mtu. We'll set qp->path_mtu
+ * to the lesser of requested attribute mtu and active,
+ * for packetizing messages.
+ * Note that the QP port has to be set in INIT and MTU in RTR.
+ */
+ if (attr_mask & IB_QP_PATH_MTU) {
+ pmtu = rdi->driver_f.get_pmtu_from_attr(rdi, qp, attr);
+ if (pmtu < 0)
+ goto inval;
+ }
+
+ if (attr_mask & IB_QP_PATH_MIG_STATE) {
+ if (attr->path_mig_state == IB_MIG_REARM) {
+ if (qp->s_mig_state == IB_MIG_ARMED)
+ goto inval;
+ if (new_state != IB_QPS_RTS)
+ goto inval;
+ } else if (attr->path_mig_state == IB_MIG_MIGRATED) {
+ if (qp->s_mig_state == IB_MIG_REARM)
+ goto inval;
+ if (new_state != IB_QPS_RTS && new_state != IB_QPS_SQD)
+ goto inval;
+ if (qp->s_mig_state == IB_MIG_ARMED)
+ mig = 1;
+ } else {
+ goto inval;
+ }
+ }
+
+ if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+ if (attr->max_dest_rd_atomic > rdi->dparms.max_rdma_atomic)
+ goto inval;
+
+ switch (new_state) {
+ case IB_QPS_RESET:
+ if (qp->state != IB_QPS_RESET)
+ _rvt_reset_qp(rdi, qp, ibqp->qp_type);
+ break;
+
+ case IB_QPS_RTR:
+ /* Allow event to re-trigger if QP set to RTR more than once */
+ qp->r_flags &= ~RVT_R_COMM_EST;
+ qp->state = new_state;
+ break;
+
+ case IB_QPS_SQD:
+ qp->s_draining = qp->s_last != qp->s_cur;
+ qp->state = new_state;
+ break;
+
+ case IB_QPS_SQE:
+ if (qp->ibqp.qp_type == IB_QPT_RC)
+ goto inval;
+ qp->state = new_state;
+ break;
+
+ case IB_QPS_ERR:
+ lastwqe = rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+ break;
+
+ default:
+ qp->state = new_state;
+ break;
+ }
+
+ if (attr_mask & IB_QP_PKEY_INDEX)
+ qp->s_pkey_index = attr->pkey_index;
+
+ if (attr_mask & IB_QP_PORT)
+ qp->port_num = attr->port_num;
+
+ if (attr_mask & IB_QP_DEST_QPN)
+ qp->remote_qpn = attr->dest_qp_num;
+
+ if (attr_mask & IB_QP_SQ_PSN) {
+ qp->s_next_psn = attr->sq_psn & rdi->dparms.psn_modify_mask;
+ qp->s_psn = qp->s_next_psn;
+ qp->s_sending_psn = qp->s_next_psn;
+ qp->s_last_psn = qp->s_next_psn - 1;
+ qp->s_sending_hpsn = qp->s_last_psn;
+ }
+
+ if (attr_mask & IB_QP_RQ_PSN)
+ qp->r_psn = attr->rq_psn & rdi->dparms.psn_modify_mask;
+
+ if (attr_mask & IB_QP_ACCESS_FLAGS)
+ qp->qp_access_flags = attr->qp_access_flags;
+
+ if (attr_mask & IB_QP_AV) {
+ rdma_replace_ah_attr(&qp->remote_ah_attr, &attr->ah_attr);
+ qp->s_srate = rdma_ah_get_static_rate(&attr->ah_attr);
+ qp->srate_mbps = ib_rate_to_mbps(qp->s_srate);
+ }
+
+ if (attr_mask & IB_QP_ALT_PATH) {
+ rdma_replace_ah_attr(&qp->alt_ah_attr, &attr->alt_ah_attr);
+ qp->s_alt_pkey_index = attr->alt_pkey_index;
+ }
+
+ if (attr_mask & IB_QP_PATH_MIG_STATE) {
+ qp->s_mig_state = attr->path_mig_state;
+ if (mig) {
+ qp->remote_ah_attr = qp->alt_ah_attr;
+ qp->port_num = rdma_ah_get_port_num(&qp->alt_ah_attr);
+ qp->s_pkey_index = qp->s_alt_pkey_index;
+ }
+ }
+
+ if (attr_mask & IB_QP_PATH_MTU) {
+ qp->pmtu = rdi->driver_f.mtu_from_qp(rdi, qp, pmtu);
+ qp->log_pmtu = ilog2(qp->pmtu);
+ }
+
+ if (attr_mask & IB_QP_RETRY_CNT) {
+ qp->s_retry_cnt = attr->retry_cnt;
+ qp->s_retry = attr->retry_cnt;
+ }
+
+ if (attr_mask & IB_QP_RNR_RETRY) {
+ qp->s_rnr_retry_cnt = attr->rnr_retry;
+ qp->s_rnr_retry = attr->rnr_retry;
+ }
+
+ if (attr_mask & IB_QP_MIN_RNR_TIMER)
+ qp->r_min_rnr_timer = attr->min_rnr_timer;
+
+ if (attr_mask & IB_QP_TIMEOUT) {
+ qp->timeout = attr->timeout;
+ qp->timeout_jiffies = rvt_timeout_to_jiffies(qp->timeout);
+ }
+
+ if (attr_mask & IB_QP_QKEY)
+ qp->qkey = attr->qkey;
+
+ if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+ qp->r_max_rd_atomic = attr->max_dest_rd_atomic;
+
+ if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
+ qp->s_max_rd_atomic = attr->max_rd_atomic;
+
+ if (rdi->driver_f.modify_qp)
+ rdi->driver_f.modify_qp(qp, attr, attr_mask, udata);
+
+ spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->s_hlock);
+ spin_unlock_irq(&qp->r_lock);
+
+ if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+ rvt_insert_qp(rdi, qp);
+
+ if (lastwqe) {
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
+ if (mig) {
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_PATH_MIG;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
+ return 0;
+
+inval:
+ spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->s_hlock);
+ spin_unlock_irq(&qp->r_lock);
+ return -EINVAL;
+}
+
+/**
+ * rvt_destroy_qp - destroy a queue pair
+ * @ibqp: the queue pair to destroy
+ * @udata: unused by the driver
+ *
+ * Note that this can be called while the QP is actively sending or
+ * receiving!
+ *
+ * Return: 0 on success.
+ */
+int rvt_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+
+ rvt_reset_qp(rdi, qp, ibqp->qp_type);
+
+ wait_event(qp->wait, !atomic_read(&qp->refcount));
+ /* qpn is now available for use again */
+ rvt_free_qpn(&rdi->qp_dev->qpn_table, qp->ibqp.qp_num);
+
+ spin_lock(&rdi->n_qps_lock);
+ rdi->n_qps_allocated--;
+ if (qp->ibqp.qp_type == IB_QPT_RC) {
+ rdi->n_rc_qps--;
+ rdi->busy_jiffies = rdi->n_rc_qps / RC_QP_SCALING_INTERVAL;
+ }
+ spin_unlock(&rdi->n_qps_lock);
+
+ if (qp->ip)
+ kref_put(&qp->ip->ref, rvt_release_mmap_info);
+ kvfree(qp->r_rq.kwq);
+ rdi->driver_f.qp_priv_free(rdi, qp);
+ kfree(qp->s_ack_queue);
+ kfree(qp->r_sg_list);
+ rdma_destroy_ah_attr(&qp->remote_ah_attr);
+ rdma_destroy_ah_attr(&qp->alt_ah_attr);
+ free_ud_wq_attr(qp);
+ vfree(qp->s_wq);
+ return 0;
+}
+
+/**
+ * rvt_query_qp - query an ipbq
+ * @ibqp: IB qp to query
+ * @attr: attr struct to fill in
+ * @attr_mask: attr mask ignored
+ * @init_attr: struct to fill in
+ *
+ * Return: always 0
+ */
+int rvt_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+
+ attr->qp_state = qp->state;
+ attr->cur_qp_state = attr->qp_state;
+ attr->path_mtu = rdi->driver_f.mtu_to_path_mtu(qp->pmtu);
+ attr->path_mig_state = qp->s_mig_state;
+ attr->qkey = qp->qkey;
+ attr->rq_psn = qp->r_psn & rdi->dparms.psn_mask;
+ attr->sq_psn = qp->s_next_psn & rdi->dparms.psn_mask;
+ attr->dest_qp_num = qp->remote_qpn;
+ attr->qp_access_flags = qp->qp_access_flags;
+ attr->cap.max_send_wr = qp->s_size - 1 -
+ rdi->dparms.reserved_operations;
+ attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1;
+ attr->cap.max_send_sge = qp->s_max_sge;
+ attr->cap.max_recv_sge = qp->r_rq.max_sge;
+ attr->cap.max_inline_data = 0;
+ attr->ah_attr = qp->remote_ah_attr;
+ attr->alt_ah_attr = qp->alt_ah_attr;
+ attr->pkey_index = qp->s_pkey_index;
+ attr->alt_pkey_index = qp->s_alt_pkey_index;
+ attr->en_sqd_async_notify = 0;
+ attr->sq_draining = qp->s_draining;
+ attr->max_rd_atomic = qp->s_max_rd_atomic;
+ attr->max_dest_rd_atomic = qp->r_max_rd_atomic;
+ attr->min_rnr_timer = qp->r_min_rnr_timer;
+ attr->port_num = qp->port_num;
+ attr->timeout = qp->timeout;
+ attr->retry_cnt = qp->s_retry_cnt;
+ attr->rnr_retry = qp->s_rnr_retry_cnt;
+ attr->alt_port_num =
+ rdma_ah_get_port_num(&qp->alt_ah_attr);
+ attr->alt_timeout = qp->alt_timeout;
+
+ init_attr->event_handler = qp->ibqp.event_handler;
+ init_attr->qp_context = qp->ibqp.qp_context;
+ init_attr->send_cq = qp->ibqp.send_cq;
+ init_attr->recv_cq = qp->ibqp.recv_cq;
+ init_attr->srq = qp->ibqp.srq;
+ init_attr->cap = attr->cap;
+ if (qp->s_flags & RVT_S_SIGNAL_REQ_WR)
+ init_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
+ else
+ init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
+ init_attr->qp_type = qp->ibqp.qp_type;
+ init_attr->port_num = qp->port_num;
+ return 0;
+}
+
+/**
+ * rvt_post_recv - post a receive on a QP
+ * @ibqp: the QP to post the receive on
+ * @wr: the WR to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ *
+ * Return: 0 on success otherwise errno
+ */
+int rvt_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
+ const struct ib_recv_wr **bad_wr)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_krwq *wq = qp->r_rq.kwq;
+ unsigned long flags;
+ int qp_err_flush = (ib_rvt_state_ops[qp->state] & RVT_FLUSH_RECV) &&
+ !qp->ibqp.srq;
+
+ /* Check that state is OK to post receive. */
+ if (!(ib_rvt_state_ops[qp->state] & RVT_POST_RECV_OK) || !wq) {
+ *bad_wr = wr;
+ return -EINVAL;
+ }
+
+ for (; wr; wr = wr->next) {
+ struct rvt_rwqe *wqe;
+ u32 next;
+ int i;
+
+ if ((unsigned)wr->num_sge > qp->r_rq.max_sge) {
+ *bad_wr = wr;
+ return -EINVAL;
+ }
+
+ spin_lock_irqsave(&qp->r_rq.kwq->p_lock, flags);
+ next = wq->head + 1;
+ if (next >= qp->r_rq.size)
+ next = 0;
+ if (next == READ_ONCE(wq->tail)) {
+ spin_unlock_irqrestore(&qp->r_rq.kwq->p_lock, flags);
+ *bad_wr = wr;
+ return -ENOMEM;
+ }
+ if (unlikely(qp_err_flush)) {
+ struct ib_wc wc;
+
+ memset(&wc, 0, sizeof(wc));
+ wc.qp = &qp->ibqp;
+ wc.opcode = IB_WC_RECV;
+ wc.wr_id = wr->wr_id;
+ wc.status = IB_WC_WR_FLUSH_ERR;
+ rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
+ } else {
+ wqe = rvt_get_rwqe_ptr(&qp->r_rq, wq->head);
+ wqe->wr_id = wr->wr_id;
+ wqe->num_sge = wr->num_sge;
+ for (i = 0; i < wr->num_sge; i++) {
+ wqe->sg_list[i].addr = wr->sg_list[i].addr;
+ wqe->sg_list[i].length = wr->sg_list[i].length;
+ wqe->sg_list[i].lkey = wr->sg_list[i].lkey;
+ }
+ /*
+ * Make sure queue entry is written
+ * before the head index.
+ */
+ smp_store_release(&wq->head, next);
+ }
+ spin_unlock_irqrestore(&qp->r_rq.kwq->p_lock, flags);
+ }
+ return 0;
+}
+
+/**
+ * rvt_qp_valid_operation - validate post send wr request
+ * @qp: the qp
+ * @post_parms: the post send table for the driver
+ * @wr: the work request
+ *
+ * The routine validates the operation based on the
+ * validation table an returns the length of the operation
+ * which can extend beyond the ib_send_bw. Operation
+ * dependent flags key atomic operation validation.
+ *
+ * There is an exception for UD qps that validates the pd and
+ * overrides the length to include the additional UD specific
+ * length.
+ *
+ * Returns a negative error or the length of the work request
+ * for building the swqe.
+ */
+static inline int rvt_qp_valid_operation(
+ struct rvt_qp *qp,
+ const struct rvt_operation_params *post_parms,
+ const struct ib_send_wr *wr)
+{
+ int len;
+
+ if (wr->opcode >= RVT_OPERATION_MAX || !post_parms[wr->opcode].length)
+ return -EINVAL;
+ if (!(post_parms[wr->opcode].qpt_support & BIT(qp->ibqp.qp_type)))
+ return -EINVAL;
+ if ((post_parms[wr->opcode].flags & RVT_OPERATION_PRIV) &&
+ ibpd_to_rvtpd(qp->ibqp.pd)->user)
+ return -EINVAL;
+ if (post_parms[wr->opcode].flags & RVT_OPERATION_ATOMIC_SGE &&
+ (wr->num_sge == 0 ||
+ wr->sg_list[0].length < sizeof(u64) ||
+ wr->sg_list[0].addr & (sizeof(u64) - 1)))
+ return -EINVAL;
+ if (post_parms[wr->opcode].flags & RVT_OPERATION_ATOMIC &&
+ !qp->s_max_rd_atomic)
+ return -EINVAL;
+ len = post_parms[wr->opcode].length;
+ /* UD specific */
+ if (qp->ibqp.qp_type != IB_QPT_UC &&
+ qp->ibqp.qp_type != IB_QPT_RC) {
+ if (qp->ibqp.pd != ud_wr(wr)->ah->pd)
+ return -EINVAL;
+ len = sizeof(struct ib_ud_wr);
+ }
+ return len;
+}
+
+/**
+ * rvt_qp_is_avail - determine queue capacity
+ * @qp: the qp
+ * @rdi: the rdmavt device
+ * @reserved_op: is reserved operation
+ *
+ * This assumes the s_hlock is held but the s_last
+ * qp variable is uncontrolled.
+ *
+ * For non reserved operations, the qp->s_avail
+ * may be changed.
+ *
+ * The return value is zero or a -ENOMEM.
+ */
+static inline int rvt_qp_is_avail(
+ struct rvt_qp *qp,
+ struct rvt_dev_info *rdi,
+ bool reserved_op)
+{
+ u32 slast;
+ u32 avail;
+ u32 reserved_used;
+
+ /* see rvt_qp_wqe_unreserve() */
+ smp_mb__before_atomic();
+ if (unlikely(reserved_op)) {
+ /* see rvt_qp_wqe_unreserve() */
+ reserved_used = atomic_read(&qp->s_reserved_used);
+ if (reserved_used >= rdi->dparms.reserved_operations)
+ return -ENOMEM;
+ return 0;
+ }
+ /* non-reserved operations */
+ if (likely(qp->s_avail))
+ return 0;
+ /* See rvt_qp_complete_swqe() */
+ slast = smp_load_acquire(&qp->s_last);
+ if (qp->s_head >= slast)
+ avail = qp->s_size - (qp->s_head - slast);
+ else
+ avail = slast - qp->s_head;
+
+ reserved_used = atomic_read(&qp->s_reserved_used);
+ avail = avail - 1 -
+ (rdi->dparms.reserved_operations - reserved_used);
+ /* insure we don't assign a negative s_avail */
+ if ((s32)avail <= 0)
+ return -ENOMEM;
+ qp->s_avail = avail;
+ if (WARN_ON(qp->s_avail >
+ (qp->s_size - 1 - rdi->dparms.reserved_operations)))
+ rvt_pr_err(rdi,
+ "More avail entries than QP RB size.\nQP: %u, size: %u, avail: %u\nhead: %u, tail: %u, cur: %u, acked: %u, last: %u",
+ qp->ibqp.qp_num, qp->s_size, qp->s_avail,
+ qp->s_head, qp->s_tail, qp->s_cur,
+ qp->s_acked, qp->s_last);
+ return 0;
+}
+
+/**
+ * rvt_post_one_wr - post one RC, UC, or UD send work request
+ * @qp: the QP to post on
+ * @wr: the work request to send
+ * @call_send: kick the send engine into gear
+ */
+static int rvt_post_one_wr(struct rvt_qp *qp,
+ const struct ib_send_wr *wr,
+ bool *call_send)
+{
+ struct rvt_swqe *wqe;
+ u32 next;
+ int i;
+ int j;
+ int acc;
+ struct rvt_lkey_table *rkt;
+ struct rvt_pd *pd;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+ u8 log_pmtu;
+ int ret;
+ size_t cplen;
+ bool reserved_op;
+ int local_ops_delayed = 0;
+
+ BUILD_BUG_ON(IB_QPT_MAX >= (sizeof(u32) * BITS_PER_BYTE));
+
+ /* IB spec says that num_sge == 0 is OK. */
+ if (unlikely(wr->num_sge > qp->s_max_sge))
+ return -EINVAL;
+
+ ret = rvt_qp_valid_operation(qp, rdi->post_parms, wr);
+ if (ret < 0)
+ return ret;
+ cplen = ret;
+
+ /*
+ * Local operations include fast register and local invalidate.
+ * Fast register needs to be processed immediately because the
+ * registered lkey may be used by following work requests and the
+ * lkey needs to be valid at the time those requests are posted.
+ * Local invalidate can be processed immediately if fencing is
+ * not required and no previous local invalidate ops are pending.
+ * Signaled local operations that have been processed immediately
+ * need to have requests with "completion only" flags set posted
+ * to the send queue in order to generate completions.
+ */
+ if ((rdi->post_parms[wr->opcode].flags & RVT_OPERATION_LOCAL)) {
+ switch (wr->opcode) {
+ case IB_WR_REG_MR:
+ ret = rvt_fast_reg_mr(qp,
+ reg_wr(wr)->mr,
+ reg_wr(wr)->key,
+ reg_wr(wr)->access);
+ if (ret || !(wr->send_flags & IB_SEND_SIGNALED))
+ return ret;
+ break;
+ case IB_WR_LOCAL_INV:
+ if ((wr->send_flags & IB_SEND_FENCE) ||
+ atomic_read(&qp->local_ops_pending)) {
+ local_ops_delayed = 1;
+ } else {
+ ret = rvt_invalidate_rkey(
+ qp, wr->ex.invalidate_rkey);
+ if (ret || !(wr->send_flags & IB_SEND_SIGNALED))
+ return ret;
+ }
+ break;
+ default:
+ return -EINVAL;
+ }
+ }
+
+ reserved_op = rdi->post_parms[wr->opcode].flags &
+ RVT_OPERATION_USE_RESERVE;
+ /* check for avail */
+ ret = rvt_qp_is_avail(qp, rdi, reserved_op);
+ if (ret)
+ return ret;
+ next = qp->s_head + 1;
+ if (next >= qp->s_size)
+ next = 0;
+
+ rkt = &rdi->lkey_table;
+ pd = ibpd_to_rvtpd(qp->ibqp.pd);
+ wqe = rvt_get_swqe_ptr(qp, qp->s_head);
+
+ /* cplen has length from above */
+ memcpy(&wqe->ud_wr, wr, cplen);
+
+ wqe->length = 0;
+ j = 0;
+ if (wr->num_sge) {
+ struct rvt_sge *last_sge = NULL;
+
+ acc = wr->opcode >= IB_WR_RDMA_READ ?
+ IB_ACCESS_LOCAL_WRITE : 0;
+ for (i = 0; i < wr->num_sge; i++) {
+ u32 length = wr->sg_list[i].length;
+
+ if (length == 0)
+ continue;
+ ret = rvt_lkey_ok(rkt, pd, &wqe->sg_list[j], last_sge,
+ &wr->sg_list[i], acc);
+ if (unlikely(ret < 0))
+ goto bail_inval_free;
+ wqe->length += length;
+ if (ret)
+ last_sge = &wqe->sg_list[j];
+ j += ret;
+ }
+ wqe->wr.num_sge = j;
+ }
+
+ /*
+ * Calculate and set SWQE PSN values prior to handing it off
+ * to the driver's check routine. This give the driver the
+ * opportunity to adjust PSN values based on internal checks.
+ */
+ log_pmtu = qp->log_pmtu;
+ if (qp->allowed_ops == IB_OPCODE_UD) {
+ struct rvt_ah *ah = rvt_get_swqe_ah(wqe);
+
+ log_pmtu = ah->log_pmtu;
+ rdma_copy_ah_attr(wqe->ud_wr.attr, &ah->attr);
+ }
+
+ if (rdi->post_parms[wr->opcode].flags & RVT_OPERATION_LOCAL) {
+ if (local_ops_delayed)
+ atomic_inc(&qp->local_ops_pending);
+ else
+ wqe->wr.send_flags |= RVT_SEND_COMPLETION_ONLY;
+ wqe->ssn = 0;
+ wqe->psn = 0;
+ wqe->lpsn = 0;
+ } else {
+ wqe->ssn = qp->s_ssn++;
+ wqe->psn = qp->s_next_psn;
+ wqe->lpsn = wqe->psn +
+ (wqe->length ?
+ ((wqe->length - 1) >> log_pmtu) :
+ 0);
+ }
+
+ /* general part of wqe valid - allow for driver checks */
+ if (rdi->driver_f.setup_wqe) {
+ ret = rdi->driver_f.setup_wqe(qp, wqe, call_send);
+ if (ret < 0)
+ goto bail_inval_free_ref;
+ }
+
+ if (!(rdi->post_parms[wr->opcode].flags & RVT_OPERATION_LOCAL))
+ qp->s_next_psn = wqe->lpsn + 1;
+
+ if (unlikely(reserved_op)) {
+ wqe->wr.send_flags |= RVT_SEND_RESERVE_USED;
+ rvt_qp_wqe_reserve(qp, wqe);
+ } else {
+ wqe->wr.send_flags &= ~RVT_SEND_RESERVE_USED;
+ qp->s_avail--;
+ }
+ trace_rvt_post_one_wr(qp, wqe, wr->num_sge);
+ smp_wmb(); /* see request builders */
+ qp->s_head = next;
+
+ return 0;
+
+bail_inval_free_ref:
+ if (qp->allowed_ops == IB_OPCODE_UD)
+ rdma_destroy_ah_attr(wqe->ud_wr.attr);
+bail_inval_free:
+ /* release mr holds */
+ while (j) {
+ struct rvt_sge *sge = &wqe->sg_list[--j];
+
+ rvt_put_mr(sge->mr);
+ }
+ return ret;
+}
+
+/**
+ * rvt_post_send - post a send on a QP
+ * @ibqp: the QP to post the send on
+ * @wr: the list of work requests to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ *
+ * Return: 0 on success else errno
+ */
+int rvt_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
+ const struct ib_send_wr **bad_wr)
+{
+ struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+ unsigned long flags = 0;
+ bool call_send;
+ unsigned nreq = 0;
+ int err = 0;
+
+ spin_lock_irqsave(&qp->s_hlock, flags);
+
+ /*
+ * Ensure QP state is such that we can send. If not bail out early,
+ * there is no need to do this every time we post a send.
+ */
+ if (unlikely(!(ib_rvt_state_ops[qp->state] & RVT_POST_SEND_OK))) {
+ spin_unlock_irqrestore(&qp->s_hlock, flags);
+ return -EINVAL;
+ }
+
+ /*
+ * If the send queue is empty, and we only have a single WR then just go
+ * ahead and kick the send engine into gear. Otherwise we will always
+ * just schedule the send to happen later.
+ */
+ call_send = qp->s_head == READ_ONCE(qp->s_last) && !wr->next;
+
+ for (; wr; wr = wr->next) {
+ err = rvt_post_one_wr(qp, wr, &call_send);
+ if (unlikely(err)) {
+ *bad_wr = wr;
+ goto bail;
+ }
+ nreq++;
+ }
+bail:
+ spin_unlock_irqrestore(&qp->s_hlock, flags);
+ if (nreq) {
+ /*
+ * Only call do_send if there is exactly one packet, and the
+ * driver said it was ok.
+ */
+ if (nreq == 1 && call_send)
+ rdi->driver_f.do_send(qp);
+ else
+ rdi->driver_f.schedule_send_no_lock(qp);
+ }
+ return err;
+}
+
+/**
+ * rvt_post_srq_recv - post a receive on a shared receive queue
+ * @ibsrq: the SRQ to post the receive on
+ * @wr: the list of work requests to post
+ * @bad_wr: A pointer to the first WR to cause a problem is put here
+ *
+ * This may be called from interrupt context.
+ *
+ * Return: 0 on success else errno
+ */
+int rvt_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
+ const struct ib_recv_wr **bad_wr)
+{
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
+ struct rvt_krwq *wq;
+ unsigned long flags;
+
+ for (; wr; wr = wr->next) {
+ struct rvt_rwqe *wqe;
+ u32 next;
+ int i;
+
+ if ((unsigned)wr->num_sge > srq->rq.max_sge) {
+ *bad_wr = wr;
+ return -EINVAL;
+ }
+
+ spin_lock_irqsave(&srq->rq.kwq->p_lock, flags);
+ wq = srq->rq.kwq;
+ next = wq->head + 1;
+ if (next >= srq->rq.size)
+ next = 0;
+ if (next == READ_ONCE(wq->tail)) {
+ spin_unlock_irqrestore(&srq->rq.kwq->p_lock, flags);
+ *bad_wr = wr;
+ return -ENOMEM;
+ }
+
+ wqe = rvt_get_rwqe_ptr(&srq->rq, wq->head);
+ wqe->wr_id = wr->wr_id;
+ wqe->num_sge = wr->num_sge;
+ for (i = 0; i < wr->num_sge; i++) {
+ wqe->sg_list[i].addr = wr->sg_list[i].addr;
+ wqe->sg_list[i].length = wr->sg_list[i].length;
+ wqe->sg_list[i].lkey = wr->sg_list[i].lkey;
+ }
+ /* Make sure queue entry is written before the head index. */
+ smp_store_release(&wq->head, next);
+ spin_unlock_irqrestore(&srq->rq.kwq->p_lock, flags);
+ }
+ return 0;
+}
+
+/*
+ * rvt used the internal kernel struct as part of its ABI, for now make sure
+ * the kernel struct does not change layout. FIXME: rvt should never cast the
+ * user struct to a kernel struct.
+ */
+static struct ib_sge *rvt_cast_sge(struct rvt_wqe_sge *sge)
+{
+ BUILD_BUG_ON(offsetof(struct ib_sge, addr) !=
+ offsetof(struct rvt_wqe_sge, addr));
+ BUILD_BUG_ON(offsetof(struct ib_sge, length) !=
+ offsetof(struct rvt_wqe_sge, length));
+ BUILD_BUG_ON(offsetof(struct ib_sge, lkey) !=
+ offsetof(struct rvt_wqe_sge, lkey));
+ return (struct ib_sge *)sge;
+}
+
+/*
+ * Validate a RWQE and fill in the SGE state.
+ * Return 1 if OK.
+ */
+static int init_sge(struct rvt_qp *qp, struct rvt_rwqe *wqe)
+{
+ int i, j, ret;
+ struct ib_wc wc;
+ struct rvt_lkey_table *rkt;
+ struct rvt_pd *pd;
+ struct rvt_sge_state *ss;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+
+ rkt = &rdi->lkey_table;
+ pd = ibpd_to_rvtpd(qp->ibqp.srq ? qp->ibqp.srq->pd : qp->ibqp.pd);
+ ss = &qp->r_sge;
+ ss->sg_list = qp->r_sg_list;
+ qp->r_len = 0;
+ for (i = j = 0; i < wqe->num_sge; i++) {
+ if (wqe->sg_list[i].length == 0)
+ continue;
+ /* Check LKEY */
+ ret = rvt_lkey_ok(rkt, pd, j ? &ss->sg_list[j - 1] : &ss->sge,
+ NULL, rvt_cast_sge(&wqe->sg_list[i]),
+ IB_ACCESS_LOCAL_WRITE);
+ if (unlikely(ret <= 0))
+ goto bad_lkey;
+ qp->r_len += wqe->sg_list[i].length;
+ j++;
+ }
+ ss->num_sge = j;
+ ss->total_len = qp->r_len;
+ return 1;
+
+bad_lkey:
+ while (j) {
+ struct rvt_sge *sge = --j ? &ss->sg_list[j - 1] : &ss->sge;
+
+ rvt_put_mr(sge->mr);
+ }
+ ss->num_sge = 0;
+ memset(&wc, 0, sizeof(wc));
+ wc.wr_id = wqe->wr_id;
+ wc.status = IB_WC_LOC_PROT_ERR;
+ wc.opcode = IB_WC_RECV;
+ wc.qp = &qp->ibqp;
+ /* Signal solicited completion event. */
+ rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
+ return 0;
+}
+
+/**
+ * get_rvt_head - get head indices of the circular buffer
+ * @rq: data structure for request queue entry
+ * @ip: the QP
+ *
+ * Return - head index value
+ */
+static inline u32 get_rvt_head(struct rvt_rq *rq, void *ip)
+{
+ u32 head;
+
+ if (ip)
+ head = RDMA_READ_UAPI_ATOMIC(rq->wq->head);
+ else
+ head = rq->kwq->head;
+
+ return head;
+}
+
+/**
+ * rvt_get_rwqe - copy the next RWQE into the QP's RWQE
+ * @qp: the QP
+ * @wr_id_only: update qp->r_wr_id only, not qp->r_sge
+ *
+ * Return -1 if there is a local error, 0 if no RWQE is available,
+ * otherwise return 1.
+ *
+ * Can be called from interrupt level.
+ */
+int rvt_get_rwqe(struct rvt_qp *qp, bool wr_id_only)
+{
+ unsigned long flags;
+ struct rvt_rq *rq;
+ struct rvt_krwq *kwq = NULL;
+ struct rvt_rwq *wq;
+ struct rvt_srq *srq;
+ struct rvt_rwqe *wqe;
+ void (*handler)(struct ib_event *, void *);
+ u32 tail;
+ u32 head;
+ int ret;
+ void *ip = NULL;
+
+ if (qp->ibqp.srq) {
+ srq = ibsrq_to_rvtsrq(qp->ibqp.srq);
+ handler = srq->ibsrq.event_handler;
+ rq = &srq->rq;
+ ip = srq->ip;
+ } else {
+ srq = NULL;
+ handler = NULL;
+ rq = &qp->r_rq;
+ ip = qp->ip;
+ }
+
+ spin_lock_irqsave(&rq->kwq->c_lock, flags);
+ if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
+ ret = 0;
+ goto unlock;
+ }
+ kwq = rq->kwq;
+ if (ip) {
+ wq = rq->wq;
+ tail = RDMA_READ_UAPI_ATOMIC(wq->tail);
+ } else {
+ tail = kwq->tail;
+ }
+
+ /* Validate tail before using it since it is user writable. */
+ if (tail >= rq->size)
+ tail = 0;
+
+ if (kwq->count < RVT_RWQ_COUNT_THRESHOLD) {
+ head = get_rvt_head(rq, ip);
+ kwq->count = rvt_get_rq_count(rq, head, tail);
+ }
+ if (unlikely(kwq->count == 0)) {
+ ret = 0;
+ goto unlock;
+ }
+ /* Make sure entry is read after the count is read. */
+ smp_rmb();
+ wqe = rvt_get_rwqe_ptr(rq, tail);
+ /*
+ * Even though we update the tail index in memory, the verbs
+ * consumer is not supposed to post more entries until a
+ * completion is generated.
+ */
+ if (++tail >= rq->size)
+ tail = 0;
+ if (ip)
+ RDMA_WRITE_UAPI_ATOMIC(wq->tail, tail);
+ else
+ kwq->tail = tail;
+ if (!wr_id_only && !init_sge(qp, wqe)) {
+ ret = -1;
+ goto unlock;
+ }
+ qp->r_wr_id = wqe->wr_id;
+
+ kwq->count--;
+ ret = 1;
+ set_bit(RVT_R_WRID_VALID, &qp->r_aflags);
+ if (handler) {
+ /*
+ * Validate head pointer value and compute
+ * the number of remaining WQEs.
+ */
+ if (kwq->count < srq->limit) {
+ kwq->count =
+ rvt_get_rq_count(rq,
+ get_rvt_head(rq, ip), tail);
+ if (kwq->count < srq->limit) {
+ struct ib_event ev;
+
+ srq->limit = 0;
+ spin_unlock_irqrestore(&rq->kwq->c_lock, flags);
+ ev.device = qp->ibqp.device;
+ ev.element.srq = qp->ibqp.srq;
+ ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
+ handler(&ev, srq->ibsrq.srq_context);
+ goto bail;
+ }
+ }
+ }
+unlock:
+ spin_unlock_irqrestore(&rq->kwq->c_lock, flags);
+bail:
+ return ret;
+}
+EXPORT_SYMBOL(rvt_get_rwqe);
+
+/**
+ * rvt_comm_est - handle trap with QP established
+ * @qp: the QP
+ */
+void rvt_comm_est(struct rvt_qp *qp)
+{
+ qp->r_flags |= RVT_R_COMM_EST;
+ if (qp->ibqp.event_handler) {
+ struct ib_event ev;
+
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_COMM_EST;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
+}
+EXPORT_SYMBOL(rvt_comm_est);
+
+void rvt_rc_error(struct rvt_qp *qp, enum ib_wc_status err)
+{
+ unsigned long flags;
+ int lastwqe;
+
+ spin_lock_irqsave(&qp->s_lock, flags);
+ lastwqe = rvt_error_qp(qp, err);
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+
+ if (lastwqe) {
+ struct ib_event ev;
+
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
+}
+EXPORT_SYMBOL(rvt_rc_error);
+
+/*
+ * rvt_rnr_tbl_to_usec - return index into ib_rvt_rnr_table
+ * @index - the index
+ * return usec from an index into ib_rvt_rnr_table
+ */
+unsigned long rvt_rnr_tbl_to_usec(u32 index)
+{
+ return ib_rvt_rnr_table[(index & IB_AETH_CREDIT_MASK)];
+}
+EXPORT_SYMBOL(rvt_rnr_tbl_to_usec);
+
+static inline unsigned long rvt_aeth_to_usec(u32 aeth)
+{
+ return ib_rvt_rnr_table[(aeth >> IB_AETH_CREDIT_SHIFT) &
+ IB_AETH_CREDIT_MASK];
+}
+
+/*
+ * rvt_add_retry_timer_ext - add/start a retry timer
+ * @qp - the QP
+ * @shift - timeout shift to wait for multiple packets
+ * add a retry timer on the QP
+ */
+void rvt_add_retry_timer_ext(struct rvt_qp *qp, u8 shift)
+{
+ struct ib_qp *ibqp = &qp->ibqp;
+ struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
+
+ lockdep_assert_held(&qp->s_lock);
+ qp->s_flags |= RVT_S_TIMER;
+ /* 4.096 usec. * (1 << qp->timeout) */
+ qp->s_timer.expires = jiffies + rdi->busy_jiffies +
+ (qp->timeout_jiffies << shift);
+ add_timer(&qp->s_timer);
+}
+EXPORT_SYMBOL(rvt_add_retry_timer_ext);
+
+/**
+ * rvt_add_rnr_timer - add/start an rnr timer on the QP
+ * @qp: the QP
+ * @aeth: aeth of RNR timeout, simulated aeth for loopback
+ */
+void rvt_add_rnr_timer(struct rvt_qp *qp, u32 aeth)
+{
+ u32 to;
+
+ lockdep_assert_held(&qp->s_lock);
+ qp->s_flags |= RVT_S_WAIT_RNR;
+ to = rvt_aeth_to_usec(aeth);
+ trace_rvt_rnrnak_add(qp, to);
+ hrtimer_start(&qp->s_rnr_timer,
+ ns_to_ktime(1000 * to), HRTIMER_MODE_REL_PINNED);
+}
+EXPORT_SYMBOL(rvt_add_rnr_timer);
+
+/**
+ * rvt_stop_rc_timers - stop all timers
+ * @qp: the QP
+ * stop any pending timers
+ */
+void rvt_stop_rc_timers(struct rvt_qp *qp)
+{
+ lockdep_assert_held(&qp->s_lock);
+ /* Remove QP from all timers */
+ if (qp->s_flags & (RVT_S_TIMER | RVT_S_WAIT_RNR)) {
+ qp->s_flags &= ~(RVT_S_TIMER | RVT_S_WAIT_RNR);
+ del_timer(&qp->s_timer);
+ hrtimer_try_to_cancel(&qp->s_rnr_timer);
+ }
+}
+EXPORT_SYMBOL(rvt_stop_rc_timers);
+
+/**
+ * rvt_stop_rnr_timer - stop an rnr timer
+ * @qp: the QP
+ *
+ * stop an rnr timer and return if the timer
+ * had been pending.
+ */
+static void rvt_stop_rnr_timer(struct rvt_qp *qp)
+{
+ lockdep_assert_held(&qp->s_lock);
+ /* Remove QP from rnr timer */
+ if (qp->s_flags & RVT_S_WAIT_RNR) {
+ qp->s_flags &= ~RVT_S_WAIT_RNR;
+ trace_rvt_rnrnak_stop(qp, 0);
+ }
+}
+
+/**
+ * rvt_del_timers_sync - wait for any timeout routines to exit
+ * @qp: the QP
+ */
+void rvt_del_timers_sync(struct rvt_qp *qp)
+{
+ del_timer_sync(&qp->s_timer);
+ hrtimer_cancel(&qp->s_rnr_timer);
+}
+EXPORT_SYMBOL(rvt_del_timers_sync);
+
+/*
+ * This is called from s_timer for missing responses.
+ */
+static void rvt_rc_timeout(struct timer_list *t)
+{
+ struct rvt_qp *qp = from_timer(qp, t, s_timer);
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+ unsigned long flags;
+
+ 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,
+ 1);
+ rdi->driver_f.schedule_send(qp);
+ }
+ spin_unlock(&qp->s_lock);
+ spin_unlock_irqrestore(&qp->r_lock, flags);
+}
+
+/*
+ * This is called from s_timer for RNR timeouts.
+ */
+enum hrtimer_restart rvt_rc_rnr_retry(struct hrtimer *t)
+{
+ struct rvt_qp *qp = container_of(t, struct rvt_qp, s_rnr_timer);
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+ unsigned long flags;
+
+ spin_lock_irqsave(&qp->s_lock, flags);
+ rvt_stop_rnr_timer(qp);
+ trace_rvt_rnrnak_timeout(qp, 0);
+ rdi->driver_f.schedule_send(qp);
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ return HRTIMER_NORESTART;
+}
+EXPORT_SYMBOL(rvt_rc_rnr_retry);
+
+/**
+ * rvt_qp_iter_init - initial for QP iteration
+ * @rdi: rvt devinfo
+ * @v: u64 value
+ * @cb: user-defined callback
+ *
+ * This returns an iterator suitable for iterating QPs
+ * in the system.
+ *
+ * The @cb is a user-defined callback and @v is a 64-bit
+ * value passed to and relevant for processing in the
+ * @cb. An example use case would be to alter QP processing
+ * based on criteria not part of the rvt_qp.
+ *
+ * Use cases that require memory allocation to succeed
+ * must preallocate appropriately.
+ *
+ * Return: a pointer to an rvt_qp_iter or NULL
+ */
+struct rvt_qp_iter *rvt_qp_iter_init(struct rvt_dev_info *rdi,
+ u64 v,
+ void (*cb)(struct rvt_qp *qp, u64 v))
+{
+ struct rvt_qp_iter *i;
+
+ i = kzalloc(sizeof(*i), GFP_KERNEL);
+ if (!i)
+ return NULL;
+
+ i->rdi = rdi;
+ /* number of special QPs (SMI/GSI) for device */
+ i->specials = rdi->ibdev.phys_port_cnt * 2;
+ i->v = v;
+ i->cb = cb;
+
+ return i;
+}
+EXPORT_SYMBOL(rvt_qp_iter_init);
+
+/**
+ * rvt_qp_iter_next - return the next QP in iter
+ * @iter: the iterator
+ *
+ * Fine grained QP iterator suitable for use
+ * with debugfs seq_file mechanisms.
+ *
+ * Updates iter->qp with the current QP when the return
+ * value is 0.
+ *
+ * Return: 0 - iter->qp is valid 1 - no more QPs
+ */
+int rvt_qp_iter_next(struct rvt_qp_iter *iter)
+ __must_hold(RCU)
+{
+ int n = iter->n;
+ int ret = 1;
+ struct rvt_qp *pqp = iter->qp;
+ struct rvt_qp *qp;
+ struct rvt_dev_info *rdi = iter->rdi;
+
+ /*
+ * The approach is to consider the special qps
+ * as additional table entries before the
+ * real hash table. Since the qp code sets
+ * the qp->next hash link to NULL, this works just fine.
+ *
+ * iter->specials is 2 * # ports
+ *
+ * n = 0..iter->specials is the special qp indices
+ *
+ * n = iter->specials..rdi->qp_dev->qp_table_size+iter->specials are
+ * the potential hash bucket entries
+ *
+ */
+ for (; n < rdi->qp_dev->qp_table_size + iter->specials; n++) {
+ if (pqp) {
+ qp = rcu_dereference(pqp->next);
+ } else {
+ if (n < iter->specials) {
+ struct rvt_ibport *rvp;
+ int pidx;
+
+ pidx = n % rdi->ibdev.phys_port_cnt;
+ rvp = rdi->ports[pidx];
+ qp = rcu_dereference(rvp->qp[n & 1]);
+ } else {
+ qp = rcu_dereference(
+ rdi->qp_dev->qp_table[
+ (n - iter->specials)]);
+ }
+ }
+ pqp = qp;
+ if (qp) {
+ iter->qp = qp;
+ iter->n = n;
+ return 0;
+ }
+ }
+ return ret;
+}
+EXPORT_SYMBOL(rvt_qp_iter_next);
+
+/**
+ * rvt_qp_iter - iterate all QPs
+ * @rdi: rvt devinfo
+ * @v: a 64-bit value
+ * @cb: a callback
+ *
+ * This provides a way for iterating all QPs.
+ *
+ * The @cb is a user-defined callback and @v is a 64-bit
+ * value passed to and relevant for processing in the
+ * cb. An example use case would be to alter QP processing
+ * based on criteria not part of the rvt_qp.
+ *
+ * The code has an internal iterator to simplify
+ * non seq_file use cases.
+ */
+void rvt_qp_iter(struct rvt_dev_info *rdi,
+ u64 v,
+ void (*cb)(struct rvt_qp *qp, u64 v))
+{
+ int ret;
+ struct rvt_qp_iter i = {
+ .rdi = rdi,
+ .specials = rdi->ibdev.phys_port_cnt * 2,
+ .v = v,
+ .cb = cb
+ };
+
+ rcu_read_lock();
+ do {
+ ret = rvt_qp_iter_next(&i);
+ if (!ret) {
+ rvt_get_qp(i.qp);
+ rcu_read_unlock();
+ i.cb(i.qp, i.v);
+ rcu_read_lock();
+ rvt_put_qp(i.qp);
+ }
+ } while (!ret);
+ rcu_read_unlock();
+}
+EXPORT_SYMBOL(rvt_qp_iter);
+
+/*
+ * This should be called with s_lock and r_lock held.
+ */
+void rvt_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
+ enum ib_wc_status status)
+{
+ u32 old_last, last;
+ struct rvt_dev_info *rdi;
+
+ if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_OR_FLUSH_SEND))
+ return;
+ rdi = ib_to_rvt(qp->ibqp.device);
+
+ old_last = qp->s_last;
+ trace_rvt_qp_send_completion(qp, wqe, old_last);
+ last = rvt_qp_complete_swqe(qp, wqe, rdi->wc_opcode[wqe->wr.opcode],
+ status);
+ if (qp->s_acked == old_last)
+ qp->s_acked = last;
+ if (qp->s_cur == old_last)
+ qp->s_cur = last;
+ if (qp->s_tail == old_last)
+ qp->s_tail = last;
+ if (qp->state == IB_QPS_SQD && last == qp->s_cur)
+ qp->s_draining = 0;
+}
+EXPORT_SYMBOL(rvt_send_complete);
+
+/**
+ * rvt_copy_sge - copy data to SGE memory
+ * @qp: associated QP
+ * @ss: the SGE state
+ * @data: the data to copy
+ * @length: the length of the data
+ * @release: boolean to release MR
+ * @copy_last: do a separate copy of the last 8 bytes
+ */
+void rvt_copy_sge(struct rvt_qp *qp, struct rvt_sge_state *ss,
+ void *data, u32 length,
+ bool release, bool copy_last)
+{
+ struct rvt_sge *sge = &ss->sge;
+ int i;
+ bool in_last = false;
+ bool cacheless_copy = false;
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+ struct rvt_wss *wss = rdi->wss;
+ unsigned int sge_copy_mode = rdi->dparms.sge_copy_mode;
+
+ if (sge_copy_mode == RVT_SGE_COPY_CACHELESS) {
+ cacheless_copy = length >= PAGE_SIZE;
+ } else if (sge_copy_mode == RVT_SGE_COPY_ADAPTIVE) {
+ if (length >= PAGE_SIZE) {
+ /*
+ * NOTE: this *assumes*:
+ * o The first vaddr is the dest.
+ * o If multiple pages, then vaddr is sequential.
+ */
+ wss_insert(wss, sge->vaddr);
+ if (length >= (2 * PAGE_SIZE))
+ wss_insert(wss, (sge->vaddr + PAGE_SIZE));
+
+ cacheless_copy = wss_exceeds_threshold(wss);
+ } else {
+ wss_advance_clean_counter(wss);
+ }
+ }
+
+ if (copy_last) {
+ if (length > 8) {
+ length -= 8;
+ } else {
+ copy_last = false;
+ in_last = true;
+ }
+ }
+
+again:
+ while (length) {
+ u32 len = rvt_get_sge_length(sge, length);
+
+ WARN_ON_ONCE(len == 0);
+ if (unlikely(in_last)) {
+ /* enforce byte transfer ordering */
+ for (i = 0; i < len; i++)
+ ((u8 *)sge->vaddr)[i] = ((u8 *)data)[i];
+ } else if (cacheless_copy) {
+ cacheless_memcpy(sge->vaddr, data, len);
+ } else {
+ memcpy(sge->vaddr, data, len);
+ }
+ rvt_update_sge(ss, len, release);
+ data += len;
+ length -= len;
+ }
+
+ if (copy_last) {
+ copy_last = false;
+ in_last = true;
+ length = 8;
+ goto again;
+ }
+}
+EXPORT_SYMBOL(rvt_copy_sge);
+
+static enum ib_wc_status loopback_qp_drop(struct rvt_ibport *rvp,
+ struct rvt_qp *sqp)
+{
+ rvp->n_pkt_drops++;
+ /*
+ * For RC, the requester would timeout and retry so
+ * shortcut the timeouts and just signal too many retries.
+ */
+ return sqp->ibqp.qp_type == IB_QPT_RC ?
+ IB_WC_RETRY_EXC_ERR : IB_WC_SUCCESS;
+}
+
+/**
+ * rvt_ruc_loopback - handle UC and RC loopback requests
+ * @sqp: the sending QP
+ *
+ * This is called from rvt_do_send() to forward a WQE addressed to the same HFI
+ * Note that although we are single threaded due to the send engine, we still
+ * have to protect against post_send(). We don't have to worry about
+ * receive interrupts since this is a connected protocol and all packets
+ * will pass through here.
+ */
+void rvt_ruc_loopback(struct rvt_qp *sqp)
+{
+ struct rvt_ibport *rvp = NULL;
+ struct rvt_dev_info *rdi = ib_to_rvt(sqp->ibqp.device);
+ struct rvt_qp *qp;
+ struct rvt_swqe *wqe;
+ struct rvt_sge *sge;
+ unsigned long flags;
+ struct ib_wc wc;
+ u64 sdata;
+ atomic64_t *maddr;
+ enum ib_wc_status send_status;
+ bool release;
+ int ret;
+ bool copy_last = false;
+ int local_ops = 0;
+
+ rcu_read_lock();
+ rvp = rdi->ports[sqp->port_num - 1];
+
+ /*
+ * Note that we check the responder QP state after
+ * checking the requester's state.
+ */
+
+ qp = rvt_lookup_qpn(ib_to_rvt(sqp->ibqp.device), rvp,
+ sqp->remote_qpn);
+
+ spin_lock_irqsave(&sqp->s_lock, flags);
+
+ /* Return if we are already busy processing a work request. */
+ if ((sqp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT)) ||
+ !(ib_rvt_state_ops[sqp->state] & RVT_PROCESS_OR_FLUSH_SEND))
+ goto unlock;
+
+ sqp->s_flags |= RVT_S_BUSY;
+
+again:
+ if (sqp->s_last == READ_ONCE(sqp->s_head))
+ goto clr_busy;
+ wqe = rvt_get_swqe_ptr(sqp, sqp->s_last);
+
+ /* Return if it is not OK to start a new work request. */
+ if (!(ib_rvt_state_ops[sqp->state] & RVT_PROCESS_NEXT_SEND_OK)) {
+ if (!(ib_rvt_state_ops[sqp->state] & RVT_FLUSH_SEND))
+ goto clr_busy;
+ /* We are in the error state, flush the work request. */
+ send_status = IB_WC_WR_FLUSH_ERR;
+ goto flush_send;
+ }
+
+ /*
+ * We can rely on the entry not changing without the s_lock
+ * being held until we update s_last.
+ * We increment s_cur to indicate s_last is in progress.
+ */
+ if (sqp->s_last == sqp->s_cur) {
+ if (++sqp->s_cur >= sqp->s_size)
+ sqp->s_cur = 0;
+ }
+ spin_unlock_irqrestore(&sqp->s_lock, flags);
+
+ if (!qp) {
+ send_status = loopback_qp_drop(rvp, sqp);
+ goto serr_no_r_lock;
+ }
+ spin_lock_irqsave(&qp->r_lock, flags);
+ if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) ||
+ qp->ibqp.qp_type != sqp->ibqp.qp_type) {
+ send_status = loopback_qp_drop(rvp, sqp);
+ goto serr;
+ }
+
+ memset(&wc, 0, sizeof(wc));
+ send_status = IB_WC_SUCCESS;
+
+ release = true;
+ sqp->s_sge.sge = wqe->sg_list[0];
+ sqp->s_sge.sg_list = wqe->sg_list + 1;
+ sqp->s_sge.num_sge = wqe->wr.num_sge;
+ sqp->s_len = wqe->length;
+ switch (wqe->wr.opcode) {
+ case IB_WR_REG_MR:
+ goto send_comp;
+
+ case IB_WR_LOCAL_INV:
+ if (!(wqe->wr.send_flags & RVT_SEND_COMPLETION_ONLY)) {
+ if (rvt_invalidate_rkey(sqp,
+ wqe->wr.ex.invalidate_rkey))
+ send_status = IB_WC_LOC_PROT_ERR;
+ local_ops = 1;
+ }
+ goto send_comp;
+
+ case IB_WR_SEND_WITH_INV:
+ case IB_WR_SEND_WITH_IMM:
+ case IB_WR_SEND:
+ ret = rvt_get_rwqe(qp, false);
+ if (ret < 0)
+ goto op_err;
+ if (!ret)
+ goto rnr_nak;
+ if (wqe->length > qp->r_len)
+ goto inv_err;
+ switch (wqe->wr.opcode) {
+ case IB_WR_SEND_WITH_INV:
+ if (!rvt_invalidate_rkey(qp,
+ wqe->wr.ex.invalidate_rkey)) {
+ wc.wc_flags = IB_WC_WITH_INVALIDATE;
+ wc.ex.invalidate_rkey =
+ wqe->wr.ex.invalidate_rkey;
+ }
+ break;
+ case IB_WR_SEND_WITH_IMM:
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.ex.imm_data = wqe->wr.ex.imm_data;
+ break;
+ default:
+ break;
+ }
+ break;
+
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+ goto inv_err;
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.ex.imm_data = wqe->wr.ex.imm_data;
+ ret = rvt_get_rwqe(qp, true);
+ if (ret < 0)
+ goto op_err;
+ if (!ret)
+ goto rnr_nak;
+ /* skip copy_last set and qp_access_flags recheck */
+ goto do_write;
+ case IB_WR_RDMA_WRITE:
+ copy_last = rvt_is_user_qp(qp);
+ if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+ goto inv_err;
+do_write:
+ if (wqe->length == 0)
+ break;
+ if (unlikely(!rvt_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
+ wqe->rdma_wr.remote_addr,
+ wqe->rdma_wr.rkey,
+ IB_ACCESS_REMOTE_WRITE)))
+ goto acc_err;
+ qp->r_sge.sg_list = NULL;
+ qp->r_sge.num_sge = 1;
+ qp->r_sge.total_len = wqe->length;
+ break;
+
+ case IB_WR_RDMA_READ:
+ if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+ goto inv_err;
+ if (unlikely(!rvt_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
+ wqe->rdma_wr.remote_addr,
+ wqe->rdma_wr.rkey,
+ IB_ACCESS_REMOTE_READ)))
+ goto acc_err;
+ release = false;
+ sqp->s_sge.sg_list = NULL;
+ sqp->s_sge.num_sge = 1;
+ qp->r_sge.sge = wqe->sg_list[0];
+ qp->r_sge.sg_list = wqe->sg_list + 1;
+ qp->r_sge.num_sge = wqe->wr.num_sge;
+ qp->r_sge.total_len = wqe->length;
+ break;
+
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+ goto inv_err;
+ if (unlikely(wqe->atomic_wr.remote_addr & (sizeof(u64) - 1)))
+ goto inv_err;
+ if (unlikely(!rvt_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+ wqe->atomic_wr.remote_addr,
+ wqe->atomic_wr.rkey,
+ IB_ACCESS_REMOTE_ATOMIC)))
+ goto acc_err;
+ /* Perform atomic OP and save result. */
+ maddr = (atomic64_t *)qp->r_sge.sge.vaddr;
+ sdata = wqe->atomic_wr.compare_add;
+ *(u64 *)sqp->s_sge.sge.vaddr =
+ (wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
+ (u64)atomic64_add_return(sdata, maddr) - sdata :
+ (u64)cmpxchg((u64 *)qp->r_sge.sge.vaddr,
+ sdata, wqe->atomic_wr.swap);
+ rvt_put_mr(qp->r_sge.sge.mr);
+ qp->r_sge.num_sge = 0;
+ goto send_comp;
+
+ default:
+ send_status = IB_WC_LOC_QP_OP_ERR;
+ goto serr;
+ }
+
+ sge = &sqp->s_sge.sge;
+ while (sqp->s_len) {
+ u32 len = rvt_get_sge_length(sge, sqp->s_len);
+
+ WARN_ON_ONCE(len == 0);
+ rvt_copy_sge(qp, &qp->r_sge, sge->vaddr,
+ len, release, copy_last);
+ rvt_update_sge(&sqp->s_sge, len, !release);
+ sqp->s_len -= len;
+ }
+ if (release)
+ rvt_put_ss(&qp->r_sge);
+
+ if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
+ goto send_comp;
+
+ if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
+ wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+ else
+ wc.opcode = IB_WC_RECV;
+ wc.wr_id = qp->r_wr_id;
+ wc.status = IB_WC_SUCCESS;
+ wc.byte_len = wqe->length;
+ wc.qp = &qp->ibqp;
+ wc.src_qp = qp->remote_qpn;
+ wc.slid = rdma_ah_get_dlid(&qp->remote_ah_attr) & U16_MAX;
+ wc.sl = rdma_ah_get_sl(&qp->remote_ah_attr);
+ wc.port_num = 1;
+ /* Signal completion event if the solicited bit is set. */
+ rvt_recv_cq(qp, &wc, wqe->wr.send_flags & IB_SEND_SOLICITED);
+
+send_comp:
+ spin_unlock_irqrestore(&qp->r_lock, flags);
+ spin_lock_irqsave(&sqp->s_lock, flags);
+ rvp->n_loop_pkts++;
+flush_send:
+ sqp->s_rnr_retry = sqp->s_rnr_retry_cnt;
+ spin_lock(&sqp->r_lock);
+ rvt_send_complete(sqp, wqe, send_status);
+ spin_unlock(&sqp->r_lock);
+ if (local_ops) {
+ atomic_dec(&sqp->local_ops_pending);
+ local_ops = 0;
+ }
+ goto again;
+
+rnr_nak:
+ /* Handle RNR NAK */
+ if (qp->ibqp.qp_type == IB_QPT_UC)
+ goto send_comp;
+ rvp->n_rnr_naks++;
+ /*
+ * Note: we don't need the s_lock held since the BUSY flag
+ * makes this single threaded.
+ */
+ if (sqp->s_rnr_retry == 0) {
+ send_status = IB_WC_RNR_RETRY_EXC_ERR;
+ goto serr;
+ }
+ if (sqp->s_rnr_retry_cnt < 7)
+ sqp->s_rnr_retry--;
+ spin_unlock_irqrestore(&qp->r_lock, flags);
+ spin_lock_irqsave(&sqp->s_lock, flags);
+ if (!(ib_rvt_state_ops[sqp->state] & RVT_PROCESS_RECV_OK))
+ goto clr_busy;
+ rvt_add_rnr_timer(sqp, qp->r_min_rnr_timer <<
+ IB_AETH_CREDIT_SHIFT);
+ goto clr_busy;
+
+op_err:
+ send_status = IB_WC_REM_OP_ERR;
+ wc.status = IB_WC_LOC_QP_OP_ERR;
+ goto err;
+
+inv_err:
+ send_status =
+ sqp->ibqp.qp_type == IB_QPT_RC ?
+ IB_WC_REM_INV_REQ_ERR :
+ IB_WC_SUCCESS;
+ wc.status = IB_WC_LOC_QP_OP_ERR;
+ goto err;
+
+acc_err:
+ send_status = IB_WC_REM_ACCESS_ERR;
+ wc.status = IB_WC_LOC_PROT_ERR;
+err:
+ /* responder goes to error state */
+ rvt_rc_error(qp, wc.status);
+
+serr:
+ spin_unlock_irqrestore(&qp->r_lock, flags);
+serr_no_r_lock:
+ spin_lock_irqsave(&sqp->s_lock, flags);
+ spin_lock(&sqp->r_lock);
+ rvt_send_complete(sqp, wqe, send_status);
+ spin_unlock(&sqp->r_lock);
+ if (sqp->ibqp.qp_type == IB_QPT_RC) {
+ int lastwqe;
+
+ spin_lock(&sqp->r_lock);
+ lastwqe = rvt_error_qp(sqp, IB_WC_WR_FLUSH_ERR);
+ spin_unlock(&sqp->r_lock);
+
+ sqp->s_flags &= ~RVT_S_BUSY;
+ spin_unlock_irqrestore(&sqp->s_lock, flags);
+ if (lastwqe) {
+ struct ib_event ev;
+
+ ev.device = sqp->ibqp.device;
+ ev.element.qp = &sqp->ibqp;
+ ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+ sqp->ibqp.event_handler(&ev, sqp->ibqp.qp_context);
+ }
+ goto done;
+ }
+clr_busy:
+ sqp->s_flags &= ~RVT_S_BUSY;
+unlock:
+ spin_unlock_irqrestore(&sqp->s_lock, flags);
+done:
+ rcu_read_unlock();
+}
+EXPORT_SYMBOL(rvt_ruc_loopback);
diff --git a/drivers/infiniband/sw/rdmavt/qp.h b/drivers/infiniband/sw/rdmavt/qp.h
new file mode 100644
index 0000000000..bd04be8072
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/qp.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTQP_H
+#define DEF_RVTQP_H
+
+#include <rdma/rdmavt_qp.h>
+
+int rvt_driver_qp_init(struct rvt_dev_info *rdi);
+void rvt_qp_exit(struct rvt_dev_info *rdi);
+int rvt_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *init_attr,
+ struct ib_udata *udata);
+int rvt_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata);
+int rvt_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata);
+int rvt_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_qp_init_attr *init_attr);
+int rvt_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
+ const struct ib_recv_wr **bad_wr);
+int rvt_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
+ const struct ib_send_wr **bad_wr);
+int rvt_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
+ const struct ib_recv_wr **bad_wr);
+int rvt_wss_init(struct rvt_dev_info *rdi);
+void rvt_wss_exit(struct rvt_dev_info *rdi);
+int rvt_alloc_rq(struct rvt_rq *rq, u32 size, int node,
+ struct ib_udata *udata);
+#endif /* DEF_RVTQP_H */
diff --git a/drivers/infiniband/sw/rdmavt/rc.c b/drivers/infiniband/sw/rdmavt/rc.c
new file mode 100644
index 0000000000..4e5d4a2763
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/rc.c
@@ -0,0 +1,172 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <rdma/rdmavt_qp.h>
+#include <rdma/ib_hdrs.h>
+
+/*
+ * Convert the AETH credit code into the number of credits.
+ */
+static const u16 credit_table[31] = {
+ 0, /* 0 */
+ 1, /* 1 */
+ 2, /* 2 */
+ 3, /* 3 */
+ 4, /* 4 */
+ 6, /* 5 */
+ 8, /* 6 */
+ 12, /* 7 */
+ 16, /* 8 */
+ 24, /* 9 */
+ 32, /* A */
+ 48, /* B */
+ 64, /* C */
+ 96, /* D */
+ 128, /* E */
+ 192, /* F */
+ 256, /* 10 */
+ 384, /* 11 */
+ 512, /* 12 */
+ 768, /* 13 */
+ 1024, /* 14 */
+ 1536, /* 15 */
+ 2048, /* 16 */
+ 3072, /* 17 */
+ 4096, /* 18 */
+ 6144, /* 19 */
+ 8192, /* 1A */
+ 12288, /* 1B */
+ 16384, /* 1C */
+ 24576, /* 1D */
+ 32768 /* 1E */
+};
+
+/**
+ * rvt_compute_aeth - compute the AETH (syndrome + MSN)
+ * @qp: the queue pair to compute the AETH for
+ *
+ * Returns the AETH.
+ */
+__be32 rvt_compute_aeth(struct rvt_qp *qp)
+{
+ u32 aeth = qp->r_msn & IB_MSN_MASK;
+
+ if (qp->ibqp.srq) {
+ /*
+ * Shared receive queues don't generate credits.
+ * Set the credit field to the invalid value.
+ */
+ aeth |= IB_AETH_CREDIT_INVAL << IB_AETH_CREDIT_SHIFT;
+ } else {
+ u32 min, max, x;
+ u32 credits;
+ u32 head;
+ u32 tail;
+
+ credits = READ_ONCE(qp->r_rq.kwq->count);
+ if (credits == 0) {
+ /* sanity check pointers before trusting them */
+ if (qp->ip) {
+ head = RDMA_READ_UAPI_ATOMIC(qp->r_rq.wq->head);
+ tail = RDMA_READ_UAPI_ATOMIC(qp->r_rq.wq->tail);
+ } else {
+ head = READ_ONCE(qp->r_rq.kwq->head);
+ tail = READ_ONCE(qp->r_rq.kwq->tail);
+ }
+ if (head >= qp->r_rq.size)
+ head = 0;
+ if (tail >= qp->r_rq.size)
+ tail = 0;
+ /*
+ * Compute the number of credits available (RWQEs).
+ * There is a small chance that the pair of reads are
+ * not atomic, which is OK, since the fuzziness is
+ * resolved as further ACKs go out.
+ */
+ credits = rvt_get_rq_count(&qp->r_rq, head, tail);
+ }
+ /*
+ * Binary search the credit table to find the code to
+ * use.
+ */
+ min = 0;
+ max = 31;
+ for (;;) {
+ x = (min + max) / 2;
+ if (credit_table[x] == credits)
+ break;
+ if (credit_table[x] > credits) {
+ max = x;
+ } else {
+ if (min == x)
+ break;
+ min = x;
+ }
+ }
+ aeth |= x << IB_AETH_CREDIT_SHIFT;
+ }
+ return cpu_to_be32(aeth);
+}
+EXPORT_SYMBOL(rvt_compute_aeth);
+
+/**
+ * rvt_get_credit - flush the send work queue of a QP
+ * @qp: the qp who's send work queue to flush
+ * @aeth: the Acknowledge Extended Transport Header
+ *
+ * The QP s_lock should be held.
+ */
+void rvt_get_credit(struct rvt_qp *qp, u32 aeth)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
+ u32 credit = (aeth >> IB_AETH_CREDIT_SHIFT) & IB_AETH_CREDIT_MASK;
+
+ lockdep_assert_held(&qp->s_lock);
+ /*
+ * If the credit is invalid, we can send
+ * as many packets as we like. Otherwise, we have to
+ * honor the credit field.
+ */
+ if (credit == IB_AETH_CREDIT_INVAL) {
+ if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) {
+ qp->s_flags |= RVT_S_UNLIMITED_CREDIT;
+ if (qp->s_flags & RVT_S_WAIT_SSN_CREDIT) {
+ qp->s_flags &= ~RVT_S_WAIT_SSN_CREDIT;
+ rdi->driver_f.schedule_send(qp);
+ }
+ }
+ } else if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) {
+ /* Compute new LSN (i.e., MSN + credit) */
+ credit = (aeth + credit_table[credit]) & IB_MSN_MASK;
+ if (rvt_cmp_msn(credit, qp->s_lsn) > 0) {
+ qp->s_lsn = credit;
+ if (qp->s_flags & RVT_S_WAIT_SSN_CREDIT) {
+ qp->s_flags &= ~RVT_S_WAIT_SSN_CREDIT;
+ rdi->driver_f.schedule_send(qp);
+ }
+ }
+ }
+}
+EXPORT_SYMBOL(rvt_get_credit);
+
+/**
+ * rvt_restart_sge - rewind the sge state for a wqe
+ * @ss: the sge state pointer
+ * @wqe: the wqe to rewind
+ * @len: the data length from the start of the wqe in bytes
+ *
+ * Returns the remaining data length.
+ */
+u32 rvt_restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, u32 len)
+{
+ ss->sge = wqe->sg_list[0];
+ ss->sg_list = wqe->sg_list + 1;
+ ss->num_sge = wqe->wr.num_sge;
+ ss->total_len = wqe->length;
+ rvt_skip_sge(ss, len, false);
+ return wqe->length - len;
+}
+EXPORT_SYMBOL(rvt_restart_sge);
+
diff --git a/drivers/infiniband/sw/rdmavt/srq.c b/drivers/infiniband/sw/rdmavt/srq.c
new file mode 100644
index 0000000000..14d196bde2
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/srq.c
@@ -0,0 +1,306 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <rdma/uverbs_ioctl.h>
+
+#include "srq.h"
+#include "vt.h"
+#include "qp.h"
+/**
+ * rvt_driver_srq_init - init srq resources on a per driver basis
+ * @rdi: rvt dev structure
+ *
+ * Do any initialization needed when a driver registers with rdmavt.
+ */
+void rvt_driver_srq_init(struct rvt_dev_info *rdi)
+{
+ spin_lock_init(&rdi->n_srqs_lock);
+ rdi->n_srqs_allocated = 0;
+}
+
+/**
+ * rvt_create_srq - create a shared receive queue
+ * @ibsrq: the protection domain of the SRQ to create
+ * @srq_init_attr: the attributes of the SRQ
+ * @udata: data from libibverbs when creating a user SRQ
+ *
+ * Return: 0 on success
+ */
+int rvt_create_srq(struct ib_srq *ibsrq, struct ib_srq_init_attr *srq_init_attr,
+ struct ib_udata *udata)
+{
+ struct rvt_dev_info *dev = ib_to_rvt(ibsrq->device);
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
+ u32 sz;
+ int ret;
+
+ if (srq_init_attr->srq_type != IB_SRQT_BASIC)
+ return -EOPNOTSUPP;
+
+ if (srq_init_attr->attr.max_sge == 0 ||
+ srq_init_attr->attr.max_sge > dev->dparms.props.max_srq_sge ||
+ srq_init_attr->attr.max_wr == 0 ||
+ srq_init_attr->attr.max_wr > dev->dparms.props.max_srq_wr)
+ return -EINVAL;
+
+ /*
+ * Need to use vmalloc() if we want to support large #s of entries.
+ */
+ srq->rq.size = srq_init_attr->attr.max_wr + 1;
+ srq->rq.max_sge = srq_init_attr->attr.max_sge;
+ sz = sizeof(struct ib_sge) * srq->rq.max_sge +
+ sizeof(struct rvt_rwqe);
+ if (rvt_alloc_rq(&srq->rq, srq->rq.size * sz,
+ dev->dparms.node, udata)) {
+ ret = -ENOMEM;
+ goto bail_srq;
+ }
+
+ /*
+ * Return the address of the RWQ as the offset to mmap.
+ * See rvt_mmap() for details.
+ */
+ if (udata && udata->outlen >= sizeof(__u64)) {
+ u32 s = sizeof(struct rvt_rwq) + srq->rq.size * sz;
+
+ srq->ip = rvt_create_mmap_info(dev, s, udata, srq->rq.wq);
+ if (IS_ERR(srq->ip)) {
+ ret = PTR_ERR(srq->ip);
+ goto bail_wq;
+ }
+
+ ret = ib_copy_to_udata(udata, &srq->ip->offset,
+ sizeof(srq->ip->offset));
+ if (ret)
+ goto bail_ip;
+ }
+
+ /*
+ * ib_create_srq() will initialize srq->ibsrq.
+ */
+ spin_lock_init(&srq->rq.lock);
+ srq->limit = srq_init_attr->attr.srq_limit;
+
+ spin_lock(&dev->n_srqs_lock);
+ if (dev->n_srqs_allocated == dev->dparms.props.max_srq) {
+ spin_unlock(&dev->n_srqs_lock);
+ ret = -ENOMEM;
+ goto bail_ip;
+ }
+
+ dev->n_srqs_allocated++;
+ spin_unlock(&dev->n_srqs_lock);
+
+ if (srq->ip) {
+ spin_lock_irq(&dev->pending_lock);
+ list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
+ spin_unlock_irq(&dev->pending_lock);
+ }
+
+ return 0;
+
+bail_ip:
+ kfree(srq->ip);
+bail_wq:
+ rvt_free_rq(&srq->rq);
+bail_srq:
+ return ret;
+}
+
+/**
+ * rvt_modify_srq - modify a shared receive queue
+ * @ibsrq: the SRQ to modify
+ * @attr: the new attributes of the SRQ
+ * @attr_mask: indicates which attributes to modify
+ * @udata: user data for libibverbs.so
+ *
+ * Return: 0 on success
+ */
+int rvt_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+ enum ib_srq_attr_mask attr_mask,
+ struct ib_udata *udata)
+{
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
+ struct rvt_dev_info *dev = ib_to_rvt(ibsrq->device);
+ struct rvt_rq tmp_rq = {};
+ int ret = 0;
+
+ if (attr_mask & IB_SRQ_MAX_WR) {
+ struct rvt_krwq *okwq = NULL;
+ struct rvt_rwq *owq = NULL;
+ struct rvt_rwqe *p;
+ u32 sz, size, n, head, tail;
+
+ /* Check that the requested sizes are below the limits. */
+ if ((attr->max_wr > dev->dparms.props.max_srq_wr) ||
+ ((attr_mask & IB_SRQ_LIMIT) ?
+ attr->srq_limit : srq->limit) > attr->max_wr)
+ return -EINVAL;
+ sz = sizeof(struct rvt_rwqe) +
+ srq->rq.max_sge * sizeof(struct ib_sge);
+ size = attr->max_wr + 1;
+ if (rvt_alloc_rq(&tmp_rq, size * sz, dev->dparms.node,
+ udata))
+ return -ENOMEM;
+ /* Check that we can write the offset to mmap. */
+ if (udata && udata->inlen >= sizeof(__u64)) {
+ __u64 offset_addr;
+ __u64 offset = 0;
+
+ ret = ib_copy_from_udata(&offset_addr, udata,
+ sizeof(offset_addr));
+ if (ret)
+ goto bail_free;
+ udata->outbuf = (void __user *)
+ (unsigned long)offset_addr;
+ ret = ib_copy_to_udata(udata, &offset,
+ sizeof(offset));
+ if (ret)
+ goto bail_free;
+ }
+
+ spin_lock_irq(&srq->rq.kwq->c_lock);
+ /*
+ * validate head and tail pointer values and compute
+ * the number of remaining WQEs.
+ */
+ if (udata) {
+ owq = srq->rq.wq;
+ head = RDMA_READ_UAPI_ATOMIC(owq->head);
+ tail = RDMA_READ_UAPI_ATOMIC(owq->tail);
+ } else {
+ okwq = srq->rq.kwq;
+ head = okwq->head;
+ tail = okwq->tail;
+ }
+ if (head >= srq->rq.size || tail >= srq->rq.size) {
+ ret = -EINVAL;
+ goto bail_unlock;
+ }
+ n = head;
+ if (n < tail)
+ n += srq->rq.size - tail;
+ else
+ n -= tail;
+ if (size <= n) {
+ ret = -EINVAL;
+ goto bail_unlock;
+ }
+ n = 0;
+ p = tmp_rq.kwq->curr_wq;
+ while (tail != head) {
+ struct rvt_rwqe *wqe;
+ int i;
+
+ wqe = rvt_get_rwqe_ptr(&srq->rq, tail);
+ p->wr_id = wqe->wr_id;
+ p->num_sge = wqe->num_sge;
+ for (i = 0; i < wqe->num_sge; i++)
+ p->sg_list[i] = wqe->sg_list[i];
+ n++;
+ p = (struct rvt_rwqe *)((char *)p + sz);
+ if (++tail >= srq->rq.size)
+ tail = 0;
+ }
+ srq->rq.kwq = tmp_rq.kwq;
+ if (udata) {
+ srq->rq.wq = tmp_rq.wq;
+ RDMA_WRITE_UAPI_ATOMIC(tmp_rq.wq->head, n);
+ RDMA_WRITE_UAPI_ATOMIC(tmp_rq.wq->tail, 0);
+ } else {
+ tmp_rq.kwq->head = n;
+ tmp_rq.kwq->tail = 0;
+ }
+ srq->rq.size = size;
+ if (attr_mask & IB_SRQ_LIMIT)
+ srq->limit = attr->srq_limit;
+ spin_unlock_irq(&srq->rq.kwq->c_lock);
+
+ vfree(owq);
+ kvfree(okwq);
+
+ if (srq->ip) {
+ struct rvt_mmap_info *ip = srq->ip;
+ struct rvt_dev_info *dev = ib_to_rvt(srq->ibsrq.device);
+ u32 s = sizeof(struct rvt_rwq) + size * sz;
+
+ rvt_update_mmap_info(dev, ip, s, tmp_rq.wq);
+
+ /*
+ * Return the offset to mmap.
+ * See rvt_mmap() for details.
+ */
+ if (udata && udata->inlen >= sizeof(__u64)) {
+ ret = ib_copy_to_udata(udata, &ip->offset,
+ sizeof(ip->offset));
+ if (ret)
+ return ret;
+ }
+
+ /*
+ * Put user mapping info onto the pending list
+ * unless it already is on the list.
+ */
+ spin_lock_irq(&dev->pending_lock);
+ if (list_empty(&ip->pending_mmaps))
+ list_add(&ip->pending_mmaps,
+ &dev->pending_mmaps);
+ spin_unlock_irq(&dev->pending_lock);
+ }
+ } else if (attr_mask & IB_SRQ_LIMIT) {
+ spin_lock_irq(&srq->rq.kwq->c_lock);
+ if (attr->srq_limit >= srq->rq.size)
+ ret = -EINVAL;
+ else
+ srq->limit = attr->srq_limit;
+ spin_unlock_irq(&srq->rq.kwq->c_lock);
+ }
+ return ret;
+
+bail_unlock:
+ spin_unlock_irq(&srq->rq.kwq->c_lock);
+bail_free:
+ rvt_free_rq(&tmp_rq);
+ return ret;
+}
+
+/**
+ * rvt_query_srq - query srq data
+ * @ibsrq: srq to query
+ * @attr: return info in attr
+ *
+ * Return: always 0
+ */
+int rvt_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
+{
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
+
+ attr->max_wr = srq->rq.size - 1;
+ attr->max_sge = srq->rq.max_sge;
+ attr->srq_limit = srq->limit;
+ return 0;
+}
+
+/**
+ * rvt_destroy_srq - destory an srq
+ * @ibsrq: srq object to destroy
+ * @udata: user data for libibverbs.so
+ */
+int rvt_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata)
+{
+ struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
+ struct rvt_dev_info *dev = ib_to_rvt(ibsrq->device);
+
+ spin_lock(&dev->n_srqs_lock);
+ dev->n_srqs_allocated--;
+ spin_unlock(&dev->n_srqs_lock);
+ if (srq->ip)
+ kref_put(&srq->ip->ref, rvt_release_mmap_info);
+ kvfree(srq->rq.kwq);
+ return 0;
+}
diff --git a/drivers/infiniband/sw/rdmavt/srq.h b/drivers/infiniband/sw/rdmavt/srq.h
new file mode 100644
index 0000000000..7d17372cd2
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/srq.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RVTSRQ_H
+#define DEF_RVTSRQ_H
+
+#include <rdma/rdma_vt.h>
+void rvt_driver_srq_init(struct rvt_dev_info *rdi);
+int rvt_create_srq(struct ib_srq *ibsrq, struct ib_srq_init_attr *srq_init_attr,
+ struct ib_udata *udata);
+int rvt_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+ enum ib_srq_attr_mask attr_mask,
+ struct ib_udata *udata);
+int rvt_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
+int rvt_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata);
+
+#endif /* DEF_RVTSRQ_H */
diff --git a/drivers/infiniband/sw/rdmavt/trace.c b/drivers/infiniband/sw/rdmavt/trace.c
new file mode 100644
index 0000000000..01704b8dd6
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace.c
@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#define CREATE_TRACE_POINTS
+#include "trace.h"
diff --git a/drivers/infiniband/sw/rdmavt/trace.h b/drivers/infiniband/sw/rdmavt/trace.h
new file mode 100644
index 0000000000..30eb4a72ea
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016, 2017 Intel Corporation.
+ */
+
+#define RDI_DEV_ENTRY(rdi) __string(dev, rvt_get_ibdev_name(rdi))
+#define RDI_DEV_ASSIGN(rdi) __assign_str(dev, rvt_get_ibdev_name(rdi))
+
+#include "trace_rvt.h"
+#include "trace_qp.h"
+#include "trace_tx.h"
+#include "trace_mr.h"
+#include "trace_cq.h"
+#include "trace_rc.h"
diff --git a/drivers/infiniband/sw/rdmavt/trace_cq.h b/drivers/infiniband/sw/rdmavt/trace_cq.h
new file mode 100644
index 0000000000..30dd1d9bae
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_cq.h
@@ -0,0 +1,124 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 - 2018 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_CQ_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_CQ_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdmavt_cq.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt_cq
+
+#define wc_opcode_name(opcode) { IB_WC_##opcode, #opcode }
+#define show_wc_opcode(opcode) \
+__print_symbolic(opcode, \
+ wc_opcode_name(SEND), \
+ wc_opcode_name(RDMA_WRITE), \
+ wc_opcode_name(RDMA_READ), \
+ wc_opcode_name(COMP_SWAP), \
+ wc_opcode_name(FETCH_ADD), \
+ wc_opcode_name(LSO), \
+ wc_opcode_name(LOCAL_INV), \
+ wc_opcode_name(REG_MR), \
+ wc_opcode_name(MASKED_COMP_SWAP), \
+ wc_opcode_name(RECV), \
+ wc_opcode_name(RECV_RDMA_WITH_IMM))
+
+#define CQ_ATTR_PRINT \
+"[%s] user cq %s cqe %u comp_vector %d comp_vector_cpu %d flags %x"
+
+DECLARE_EVENT_CLASS(rvt_cq_template,
+ TP_PROTO(struct rvt_cq *cq,
+ const struct ib_cq_init_attr *attr),
+ TP_ARGS(cq, attr),
+ TP_STRUCT__entry(RDI_DEV_ENTRY(cq->rdi)
+ __field(struct rvt_mmap_info *, ip)
+ __field(unsigned int, cqe)
+ __field(int, comp_vector)
+ __field(int, comp_vector_cpu)
+ __field(u32, flags)
+ ),
+ TP_fast_assign(RDI_DEV_ASSIGN(cq->rdi);
+ __entry->ip = cq->ip;
+ __entry->cqe = attr->cqe;
+ __entry->comp_vector = attr->comp_vector;
+ __entry->comp_vector_cpu =
+ cq->comp_vector_cpu;
+ __entry->flags = attr->flags;
+ ),
+ TP_printk(CQ_ATTR_PRINT, __get_str(dev),
+ __entry->ip ? "true" : "false", __entry->cqe,
+ __entry->comp_vector, __entry->comp_vector_cpu,
+ __entry->flags
+ )
+);
+
+DEFINE_EVENT(rvt_cq_template, rvt_create_cq,
+ TP_PROTO(struct rvt_cq *cq, const struct ib_cq_init_attr *attr),
+ TP_ARGS(cq, attr));
+
+#define CQ_PRN \
+"[%s] idx %u wr_id %llx status %u opcode %u,%s length %u qpn %x flags %x imm %x"
+
+DECLARE_EVENT_CLASS(
+ rvt_cq_entry_template,
+ TP_PROTO(struct rvt_cq *cq, struct ib_wc *wc, u32 idx),
+ TP_ARGS(cq, wc, idx),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(cq->rdi)
+ __field(u64, wr_id)
+ __field(u32, status)
+ __field(u32, opcode)
+ __field(u32, qpn)
+ __field(u32, length)
+ __field(u32, idx)
+ __field(u32, flags)
+ __field(u32, imm)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(cq->rdi);
+ __entry->wr_id = wc->wr_id;
+ __entry->status = wc->status;
+ __entry->opcode = wc->opcode;
+ __entry->length = wc->byte_len;
+ __entry->qpn = wc->qp->qp_num;
+ __entry->idx = idx;
+ __entry->flags = wc->wc_flags;
+ __entry->imm = be32_to_cpu(wc->ex.imm_data);
+ ),
+ TP_printk(
+ CQ_PRN,
+ __get_str(dev),
+ __entry->idx,
+ __entry->wr_id,
+ __entry->status,
+ __entry->opcode, show_wc_opcode(__entry->opcode),
+ __entry->length,
+ __entry->qpn,
+ __entry->flags,
+ __entry->imm
+ )
+);
+
+DEFINE_EVENT(
+ rvt_cq_entry_template, rvt_cq_enter,
+ TP_PROTO(struct rvt_cq *cq, struct ib_wc *wc, u32 idx),
+ TP_ARGS(cq, wc, idx));
+
+DEFINE_EVENT(
+ rvt_cq_entry_template, rvt_cq_poll,
+ TP_PROTO(struct rvt_cq *cq, struct ib_wc *wc, u32 idx),
+ TP_ARGS(cq, wc, idx));
+
+#endif /* __RVT_TRACE_CQ_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_cq
+#include <trace/define_trace.h>
diff --git a/drivers/infiniband/sw/rdmavt/trace_mr.h b/drivers/infiniband/sw/rdmavt/trace_mr.h
new file mode 100644
index 0000000000..1de7012000
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_mr.h
@@ -0,0 +1,182 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_MR_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_MR_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_vt.h>
+#include <rdma/rdmavt_mr.h>
+
+#include "mr.h"
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt_mr
+DECLARE_EVENT_CLASS(
+ rvt_mr_template,
+ TP_PROTO(struct rvt_mregion *mr, u16 m, u16 n, void *v, size_t len),
+ TP_ARGS(mr, m, n, v, len),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(mr->pd->device))
+ __field(void *, vaddr)
+ __field(struct page *, page)
+ __field(u64, iova)
+ __field(u64, user_base)
+ __field(size_t, len)
+ __field(size_t, length)
+ __field(u32, lkey)
+ __field(u32, offset)
+ __field(u16, m)
+ __field(u16, n)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(mr->pd->device));
+ __entry->vaddr = v;
+ __entry->page = virt_to_page(v);
+ __entry->iova = mr->iova;
+ __entry->user_base = mr->user_base;
+ __entry->lkey = mr->lkey;
+ __entry->m = m;
+ __entry->n = n;
+ __entry->len = len;
+ __entry->length = mr->length;
+ __entry->offset = mr->offset;
+ ),
+ TP_printk(
+ "[%s] lkey %x iova %llx user_base %llx mr_len %lu vaddr %llx page %p m %u n %u len %lu off %u",
+ __get_str(dev),
+ __entry->lkey,
+ __entry->iova,
+ __entry->user_base,
+ __entry->length,
+ (unsigned long long)__entry->vaddr,
+ __entry->page,
+ __entry->m,
+ __entry->n,
+ __entry->len,
+ __entry->offset
+ )
+);
+
+DEFINE_EVENT(
+ rvt_mr_template, rvt_mr_page_seg,
+ TP_PROTO(struct rvt_mregion *mr, u16 m, u16 n, void *v, size_t len),
+ TP_ARGS(mr, m, n, v, len));
+
+DEFINE_EVENT(
+ rvt_mr_template, rvt_mr_fmr_seg,
+ TP_PROTO(struct rvt_mregion *mr, u16 m, u16 n, void *v, size_t len),
+ TP_ARGS(mr, m, n, v, len));
+
+DEFINE_EVENT(
+ rvt_mr_template, rvt_mr_user_seg,
+ TP_PROTO(struct rvt_mregion *mr, u16 m, u16 n, void *v, size_t len),
+ TP_ARGS(mr, m, n, v, len));
+
+DECLARE_EVENT_CLASS(
+ rvt_sge_template,
+ TP_PROTO(struct rvt_sge *sge, struct ib_sge *isge),
+ TP_ARGS(sge, isge),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(sge->mr->pd->device))
+ __field(struct rvt_mregion *, mr)
+ __field(struct rvt_sge *, sge)
+ __field(struct ib_sge *, isge)
+ __field(void *, vaddr)
+ __field(u64, ivaddr)
+ __field(u32, lkey)
+ __field(u32, sge_length)
+ __field(u32, length)
+ __field(u32, ilength)
+ __field(int, user)
+ __field(u16, m)
+ __field(u16, n)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(sge->mr->pd->device));
+ __entry->mr = sge->mr;
+ __entry->sge = sge;
+ __entry->isge = isge;
+ __entry->vaddr = sge->vaddr;
+ __entry->ivaddr = isge->addr;
+ __entry->lkey = sge->mr->lkey;
+ __entry->sge_length = sge->sge_length;
+ __entry->length = sge->length;
+ __entry->ilength = isge->length;
+ __entry->m = sge->m;
+ __entry->n = sge->m;
+ __entry->user = ibpd_to_rvtpd(sge->mr->pd)->user;
+ ),
+ TP_printk(
+ "[%s] mr %p sge %p isge %p vaddr %p ivaddr %llx lkey %x sge_length %u length %u ilength %u m %u n %u user %u",
+ __get_str(dev),
+ __entry->mr,
+ __entry->sge,
+ __entry->isge,
+ __entry->vaddr,
+ __entry->ivaddr,
+ __entry->lkey,
+ __entry->sge_length,
+ __entry->length,
+ __entry->ilength,
+ __entry->m,
+ __entry->n,
+ __entry->user
+ )
+);
+
+DEFINE_EVENT(
+ rvt_sge_template, rvt_sge_adjacent,
+ TP_PROTO(struct rvt_sge *sge, struct ib_sge *isge),
+ TP_ARGS(sge, isge));
+
+DEFINE_EVENT(
+ rvt_sge_template, rvt_sge_new,
+ TP_PROTO(struct rvt_sge *sge, struct ib_sge *isge),
+ TP_ARGS(sge, isge));
+
+TRACE_EVENT(
+ rvt_map_mr_sg,
+ TP_PROTO(struct ib_mr *ibmr, int sg_nents, unsigned int *sg_offset),
+ TP_ARGS(ibmr, sg_nents, sg_offset),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(to_imr(ibmr)->mr.pd->device))
+ __field(u64, iova)
+ __field(u64, ibmr_iova)
+ __field(u64, user_base)
+ __field(u64, ibmr_length)
+ __field(int, sg_nents)
+ __field(uint, sg_offset)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(to_imr(ibmr)->mr.pd->device));
+ __entry->ibmr_iova = ibmr->iova;
+ __entry->iova = to_imr(ibmr)->mr.iova;
+ __entry->user_base = to_imr(ibmr)->mr.user_base;
+ __entry->ibmr_length = to_imr(ibmr)->mr.length;
+ __entry->sg_nents = sg_nents;
+ __entry->sg_offset = sg_offset ? *sg_offset : 0;
+ ),
+ TP_printk(
+ "[%s] ibmr_iova %llx iova %llx user_base %llx length %llx sg_nents %d sg_offset %u",
+ __get_str(dev),
+ __entry->ibmr_iova,
+ __entry->iova,
+ __entry->user_base,
+ __entry->ibmr_length,
+ __entry->sg_nents,
+ __entry->sg_offset
+ )
+);
+
+#endif /* __RVT_TRACE_MR_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_mr
+#include <trace/define_trace.h>
diff --git a/drivers/infiniband/sw/rdmavt/trace_qp.h b/drivers/infiniband/sw/rdmavt/trace_qp.h
new file mode 100644
index 0000000000..c28c81fcb3
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_qp.h
@@ -0,0 +1,96 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_QP_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_QP_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdmavt_qp.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt_qp
+
+DECLARE_EVENT_CLASS(rvt_qphash_template,
+ TP_PROTO(struct rvt_qp *qp, u32 bucket),
+ TP_ARGS(qp, bucket),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(qp->ibqp.device))
+ __field(u32, qpn)
+ __field(u32, bucket)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(qp->ibqp.device));
+ __entry->qpn = qp->ibqp.qp_num;
+ __entry->bucket = bucket;
+ ),
+ TP_printk(
+ "[%s] qpn 0x%x bucket %u",
+ __get_str(dev),
+ __entry->qpn,
+ __entry->bucket
+ )
+);
+
+DEFINE_EVENT(rvt_qphash_template, rvt_qpinsert,
+ TP_PROTO(struct rvt_qp *qp, u32 bucket),
+ TP_ARGS(qp, bucket));
+
+DEFINE_EVENT(rvt_qphash_template, rvt_qpremove,
+ TP_PROTO(struct rvt_qp *qp, u32 bucket),
+ TP_ARGS(qp, bucket));
+
+DECLARE_EVENT_CLASS(
+ rvt_rnrnak_template,
+ TP_PROTO(struct rvt_qp *qp, u32 to),
+ TP_ARGS(qp, to),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(qp->ibqp.device))
+ __field(u32, qpn)
+ __field(void *, hrtimer)
+ __field(u32, s_flags)
+ __field(u32, to)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(qp->ibqp.device));
+ __entry->qpn = qp->ibqp.qp_num;
+ __entry->hrtimer = &qp->s_rnr_timer;
+ __entry->s_flags = qp->s_flags;
+ __entry->to = to;
+ ),
+ TP_printk(
+ "[%s] qpn 0x%x hrtimer 0x%p s_flags 0x%x timeout %u us",
+ __get_str(dev),
+ __entry->qpn,
+ __entry->hrtimer,
+ __entry->s_flags,
+ __entry->to
+ )
+);
+
+DEFINE_EVENT(
+ rvt_rnrnak_template, rvt_rnrnak_add,
+ TP_PROTO(struct rvt_qp *qp, u32 to),
+ TP_ARGS(qp, to));
+
+DEFINE_EVENT(
+ rvt_rnrnak_template, rvt_rnrnak_timeout,
+ TP_PROTO(struct rvt_qp *qp, u32 to),
+ TP_ARGS(qp, to));
+
+DEFINE_EVENT(
+ rvt_rnrnak_template, rvt_rnrnak_stop,
+ TP_PROTO(struct rvt_qp *qp, u32 to),
+ TP_ARGS(qp, to));
+
+#endif /* __RVT_TRACE_QP_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_qp
+#include <trace/define_trace.h>
+
diff --git a/drivers/infiniband/sw/rdmavt/trace_rc.h b/drivers/infiniband/sw/rdmavt/trace_rc.h
new file mode 100644
index 0000000000..833bf778b0
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_rc.h
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2017 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_RC_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_RC_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdmavt_qp.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt_rc
+
+DECLARE_EVENT_CLASS(rvt_rc_template,
+ TP_PROTO(struct rvt_qp *qp, u32 psn),
+ TP_ARGS(qp, psn),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(qp->ibqp.device))
+ __field(u32, qpn)
+ __field(u32, s_flags)
+ __field(u32, psn)
+ __field(u32, s_psn)
+ __field(u32, s_next_psn)
+ __field(u32, s_sending_psn)
+ __field(u32, s_sending_hpsn)
+ __field(u32, r_psn)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(ib_to_rvt(qp->ibqp.device));
+ __entry->qpn = qp->ibqp.qp_num;
+ __entry->s_flags = qp->s_flags;
+ __entry->psn = psn;
+ __entry->s_psn = qp->s_psn;
+ __entry->s_next_psn = qp->s_next_psn;
+ __entry->s_sending_psn = qp->s_sending_psn;
+ __entry->s_sending_hpsn = qp->s_sending_hpsn;
+ __entry->r_psn = qp->r_psn;
+ ),
+ TP_printk(
+ "[%s] qpn 0x%x s_flags 0x%x psn 0x%x s_psn 0x%x s_next_psn 0x%x s_sending_psn 0x%x sending_hpsn 0x%x r_psn 0x%x",
+ __get_str(dev),
+ __entry->qpn,
+ __entry->s_flags,
+ __entry->psn,
+ __entry->s_psn,
+ __entry->s_next_psn,
+ __entry->s_sending_psn,
+ __entry->s_sending_hpsn,
+ __entry->r_psn
+ )
+);
+
+DEFINE_EVENT(rvt_rc_template, rvt_rc_timeout,
+ TP_PROTO(struct rvt_qp *qp, u32 psn),
+ TP_ARGS(qp, psn)
+);
+
+#endif /* __RVT_TRACE_RC_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_rc
+#include <trace/define_trace.h>
diff --git a/drivers/infiniband/sw/rdmavt/trace_rvt.h b/drivers/infiniband/sw/rdmavt/trace_rvt.h
new file mode 100644
index 0000000000..9df6b0b826
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_rvt.h
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_RVT_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_RVT_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_vt.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt
+
+TRACE_EVENT(rvt_dbg,
+ TP_PROTO(struct rvt_dev_info *rdi,
+ const char *msg),
+ TP_ARGS(rdi, msg),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(rdi)
+ __string(msg, msg)
+ ),
+ TP_fast_assign(
+ RDI_DEV_ASSIGN(rdi);
+ __assign_str(msg, msg);
+ ),
+ TP_printk("[%s]: %s", __get_str(dev), __get_str(msg))
+);
+
+#endif /* __RVT_TRACE_MISC_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_rvt
+#include <trace/define_trace.h>
+
diff --git a/drivers/infiniband/sw/rdmavt/trace_tx.h b/drivers/infiniband/sw/rdmavt/trace_tx.h
new file mode 100644
index 0000000000..ff7d39a307
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/trace_tx.h
@@ -0,0 +1,163 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+#if !defined(__RVT_TRACE_TX_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __RVT_TRACE_TX_H
+
+#include <linux/tracepoint.h>
+#include <linux/trace_seq.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdmavt_qp.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM rvt_tx
+
+#define wr_opcode_name(opcode) { IB_WR_##opcode, #opcode }
+#define show_wr_opcode(opcode) \
+__print_symbolic(opcode, \
+ wr_opcode_name(RDMA_WRITE), \
+ wr_opcode_name(RDMA_WRITE_WITH_IMM), \
+ wr_opcode_name(SEND), \
+ wr_opcode_name(SEND_WITH_IMM), \
+ wr_opcode_name(RDMA_READ), \
+ wr_opcode_name(ATOMIC_CMP_AND_SWP), \
+ wr_opcode_name(ATOMIC_FETCH_AND_ADD), \
+ wr_opcode_name(LSO), \
+ wr_opcode_name(SEND_WITH_INV), \
+ 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(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] 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 wr_num_sge %u"
+
+TRACE_EVENT(
+ rvt_post_one_wr,
+ TP_PROTO(struct rvt_qp *qp, struct rvt_swqe *wqe, int wr_num_sge),
+ TP_ARGS(qp, wqe, wr_num_sge),
+ 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)
+ __field(u32, opcode)
+ __field(u32, size)
+ __field(u32, avail)
+ __field(u32, head)
+ __field(u32, last)
+ __field(u32, ssn)
+ __field(int, send_flags)
+ __field(pid_t, pid)
+ __field(int, num_sge)
+ __field(int, wr_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;
+ __entry->opcode = wqe->wr.opcode;
+ __entry->size = qp->s_size;
+ __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;
+ __entry->wr_num_sge = 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->pid,
+ __entry->num_sge,
+ __entry->wr_num_sge
+ )
+);
+
+TRACE_EVENT(
+ rvt_qp_send_completion,
+ TP_PROTO(struct rvt_qp *qp, struct rvt_swqe *wqe, u32 idx),
+ TP_ARGS(qp, wqe, idx),
+ TP_STRUCT__entry(
+ RDI_DEV_ENTRY(ib_to_rvt(qp->ibqp.device))
+ __field(struct rvt_swqe *, wqe)
+ __field(u64, wr_id)
+ __field(u32, qpn)
+ __field(u32, qpt)
+ __field(u32, length)
+ __field(u32, idx)
+ __field(u32, ssn)
+ __field(enum ib_wr_opcode, opcode)
+ __field(int, send_flags)
+ ),
+ 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->length = wqe->length;
+ __entry->idx = idx;
+ __entry->ssn = wqe->ssn;
+ __entry->opcode = wqe->wr.opcode;
+ __entry->send_flags = wqe->wr.send_flags;
+ ),
+ TP_printk(
+ "[%s] qpn 0x%x qpt %u wqe %p idx %u wr_id %llx length %u ssn %u opcode %x send_flags %x",
+ __get_str(dev),
+ __entry->qpn,
+ __entry->qpt,
+ __entry->wqe,
+ __entry->idx,
+ __entry->wr_id,
+ __entry->length,
+ __entry->ssn,
+ __entry->opcode,
+ __entry->send_flags
+ )
+);
+#endif /* __RVT_TRACE_TX_H */
+
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE trace_tx
+#include <trace/define_trace.h>
+
diff --git a/drivers/infiniband/sw/rdmavt/vt.c b/drivers/infiniband/sw/rdmavt/vt.c
new file mode 100644
index 0000000000..d61f8de7f2
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/vt.c
@@ -0,0 +1,619 @@
+// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
+/*
+ * Copyright(c) 2016 - 2018 Intel Corporation.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/dma-mapping.h>
+#include "vt.h"
+#include "cq.h"
+#include "trace.h"
+
+#define RVT_UVERBS_ABI_VERSION 2
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("RDMA Verbs Transport Library");
+
+static int __init rvt_init(void)
+{
+ int ret = rvt_driver_cq_init();
+
+ if (ret)
+ pr_err("Error in driver CQ init.\n");
+
+ return ret;
+}
+module_init(rvt_init);
+
+static void __exit rvt_cleanup(void)
+{
+ rvt_cq_exit();
+}
+module_exit(rvt_cleanup);
+
+/**
+ * rvt_alloc_device - allocate rdi
+ * @size: how big of a structure to allocate
+ * @nports: number of ports to allocate array slots for
+ *
+ * Use IB core device alloc to allocate space for the rdi which is assumed to be
+ * inside of the ib_device. Any extra space that drivers require should be
+ * included in size.
+ *
+ * We also allocate a port array based on the number of ports.
+ *
+ * Return: pointer to allocated rdi
+ */
+struct rvt_dev_info *rvt_alloc_device(size_t size, int nports)
+{
+ struct rvt_dev_info *rdi;
+
+ rdi = container_of(_ib_alloc_device(size), struct rvt_dev_info, ibdev);
+ if (!rdi)
+ return rdi;
+
+ rdi->ports = kcalloc(nports, sizeof(*rdi->ports), GFP_KERNEL);
+ if (!rdi->ports)
+ ib_dealloc_device(&rdi->ibdev);
+
+ return rdi;
+}
+EXPORT_SYMBOL(rvt_alloc_device);
+
+/**
+ * rvt_dealloc_device - deallocate rdi
+ * @rdi: structure to free
+ *
+ * Free a structure allocated with rvt_alloc_device()
+ */
+void rvt_dealloc_device(struct rvt_dev_info *rdi)
+{
+ kfree(rdi->ports);
+ ib_dealloc_device(&rdi->ibdev);
+}
+EXPORT_SYMBOL(rvt_dealloc_device);
+
+static int rvt_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *props,
+ struct ib_udata *uhw)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+
+ if (uhw->inlen || uhw->outlen)
+ return -EINVAL;
+ /*
+ * Return rvt_dev_info.dparms.props contents
+ */
+ *props = rdi->dparms.props;
+ return 0;
+}
+
+static int rvt_get_numa_node(struct ib_device *ibdev)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+
+ return rdi->dparms.node;
+}
+
+static int rvt_modify_device(struct ib_device *device,
+ int device_modify_mask,
+ struct ib_device_modify *device_modify)
+{
+ /*
+ * There is currently no need to supply this based on qib and hfi1.
+ * Future drivers may need to implement this though.
+ */
+
+ return -EOPNOTSUPP;
+}
+
+/**
+ * rvt_query_port - Passes the query port call to the driver
+ * @ibdev: Verbs IB dev
+ * @port_num: port number, 1 based from ib core
+ * @props: structure to hold returned properties
+ *
+ * Return: 0 on success
+ */
+static int rvt_query_port(struct ib_device *ibdev, u32 port_num,
+ struct ib_port_attr *props)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ struct rvt_ibport *rvp;
+ u32 port_index = ibport_num_to_idx(ibdev, port_num);
+
+ rvp = rdi->ports[port_index];
+ /* props being zeroed by the caller, avoid zeroing it here */
+ props->sm_lid = rvp->sm_lid;
+ props->sm_sl = rvp->sm_sl;
+ props->port_cap_flags = rvp->port_cap_flags;
+ props->max_msg_sz = 0x80000000;
+ props->pkey_tbl_len = rvt_get_npkeys(rdi);
+ props->bad_pkey_cntr = rvp->pkey_violations;
+ props->qkey_viol_cntr = rvp->qkey_violations;
+ props->subnet_timeout = rvp->subnet_timeout;
+ props->init_type_reply = 0;
+
+ /* Populate the remaining ib_port_attr elements */
+ return rdi->driver_f.query_port_state(rdi, port_num, props);
+}
+
+/**
+ * rvt_modify_port - modify port
+ * @ibdev: Verbs IB dev
+ * @port_num: Port number, 1 based from ib core
+ * @port_modify_mask: How to change the port
+ * @props: Structure to fill in
+ *
+ * Return: 0 on success
+ */
+static int rvt_modify_port(struct ib_device *ibdev, u32 port_num,
+ int port_modify_mask, struct ib_port_modify *props)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ struct rvt_ibport *rvp;
+ int ret = 0;
+ u32 port_index = ibport_num_to_idx(ibdev, port_num);
+
+ rvp = rdi->ports[port_index];
+ if (port_modify_mask & IB_PORT_OPA_MASK_CHG) {
+ rvp->port_cap3_flags |= props->set_port_cap_mask;
+ rvp->port_cap3_flags &= ~props->clr_port_cap_mask;
+ } else {
+ rvp->port_cap_flags |= props->set_port_cap_mask;
+ rvp->port_cap_flags &= ~props->clr_port_cap_mask;
+ }
+
+ if (props->set_port_cap_mask || props->clr_port_cap_mask)
+ rdi->driver_f.cap_mask_chg(rdi, port_num);
+ if (port_modify_mask & IB_PORT_SHUTDOWN)
+ ret = rdi->driver_f.shut_down_port(rdi, port_num);
+ if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
+ rvp->qkey_violations = 0;
+
+ return ret;
+}
+
+/**
+ * rvt_query_pkey - Return a pkey from the table at a given index
+ * @ibdev: Verbs IB dev
+ * @port_num: Port number, 1 based from ib core
+ * @index: Index into pkey table
+ * @pkey: returned pkey from the port pkey table
+ *
+ * Return: 0 on failure pkey otherwise
+ */
+static int rvt_query_pkey(struct ib_device *ibdev, u32 port_num, u16 index,
+ u16 *pkey)
+{
+ /*
+ * Driver will be responsible for keeping rvt_dev_info.pkey_table up to
+ * date. This function will just return that value. There is no need to
+ * lock, if a stale value is read and sent to the user so be it there is
+ * no way to protect against that anyway.
+ */
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ u32 port_index;
+
+ port_index = ibport_num_to_idx(ibdev, port_num);
+
+ if (index >= rvt_get_npkeys(rdi))
+ return -EINVAL;
+
+ *pkey = rvt_get_pkey(rdi, port_index, index);
+ return 0;
+}
+
+/**
+ * rvt_query_gid - Return a gid from the table
+ * @ibdev: Verbs IB dev
+ * @port_num: Port number, 1 based from ib core
+ * @guid_index: Index in table
+ * @gid: Gid to return
+ *
+ * Return: 0 on success
+ */
+static int rvt_query_gid(struct ib_device *ibdev, u32 port_num,
+ int guid_index, union ib_gid *gid)
+{
+ struct rvt_dev_info *rdi;
+ struct rvt_ibport *rvp;
+ u32 port_index;
+
+ /*
+ * Driver is responsible for updating the guid table. Which will be used
+ * to craft the return value. This will work similar to how query_pkey()
+ * is being done.
+ */
+ port_index = ibport_num_to_idx(ibdev, port_num);
+
+ rdi = ib_to_rvt(ibdev);
+ rvp = rdi->ports[port_index];
+
+ gid->global.subnet_prefix = rvp->gid_prefix;
+
+ return rdi->driver_f.get_guid_be(rdi, rvp, guid_index,
+ &gid->global.interface_id);
+}
+
+/**
+ * rvt_alloc_ucontext - Allocate a user context
+ * @uctx: Verbs context
+ * @udata: User data allocated
+ */
+static int rvt_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata)
+{
+ return 0;
+}
+
+/**
+ * rvt_dealloc_ucontext - Free a user context
+ * @context: Unused
+ */
+static void rvt_dealloc_ucontext(struct ib_ucontext *context)
+{
+ return;
+}
+
+static int rvt_get_port_immutable(struct ib_device *ibdev, u32 port_num,
+ struct ib_port_immutable *immutable)
+{
+ struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
+ struct ib_port_attr attr;
+ int err;
+
+ immutable->core_cap_flags = rdi->dparms.core_cap_flags;
+
+ err = ib_query_port(ibdev, port_num, &attr);
+ if (err)
+ return err;
+
+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
+ immutable->gid_tbl_len = attr.gid_tbl_len;
+ immutable->max_mad_size = rdi->dparms.max_mad_size;
+
+ return 0;
+}
+
+enum {
+ MISC,
+ QUERY_DEVICE,
+ MODIFY_DEVICE,
+ QUERY_PORT,
+ MODIFY_PORT,
+ QUERY_PKEY,
+ QUERY_GID,
+ ALLOC_UCONTEXT,
+ DEALLOC_UCONTEXT,
+ GET_PORT_IMMUTABLE,
+ CREATE_QP,
+ MODIFY_QP,
+ DESTROY_QP,
+ QUERY_QP,
+ POST_SEND,
+ POST_RECV,
+ POST_SRQ_RECV,
+ CREATE_AH,
+ DESTROY_AH,
+ MODIFY_AH,
+ QUERY_AH,
+ CREATE_SRQ,
+ MODIFY_SRQ,
+ DESTROY_SRQ,
+ QUERY_SRQ,
+ ATTACH_MCAST,
+ DETACH_MCAST,
+ GET_DMA_MR,
+ REG_USER_MR,
+ DEREG_MR,
+ ALLOC_MR,
+ MAP_MR_SG,
+ ALLOC_FMR,
+ MAP_PHYS_FMR,
+ UNMAP_FMR,
+ DEALLOC_FMR,
+ MMAP,
+ CREATE_CQ,
+ DESTROY_CQ,
+ POLL_CQ,
+ REQ_NOTFIY_CQ,
+ RESIZE_CQ,
+ ALLOC_PD,
+ DEALLOC_PD,
+ _VERB_IDX_MAX /* Must always be last! */
+};
+
+static const struct ib_device_ops rvt_dev_ops = {
+ .uverbs_abi_ver = RVT_UVERBS_ABI_VERSION,
+
+ .alloc_mr = rvt_alloc_mr,
+ .alloc_pd = rvt_alloc_pd,
+ .alloc_ucontext = rvt_alloc_ucontext,
+ .attach_mcast = rvt_attach_mcast,
+ .create_ah = rvt_create_ah,
+ .create_cq = rvt_create_cq,
+ .create_qp = rvt_create_qp,
+ .create_srq = rvt_create_srq,
+ .create_user_ah = rvt_create_ah,
+ .dealloc_pd = rvt_dealloc_pd,
+ .dealloc_ucontext = rvt_dealloc_ucontext,
+ .dereg_mr = rvt_dereg_mr,
+ .destroy_ah = rvt_destroy_ah,
+ .destroy_cq = rvt_destroy_cq,
+ .destroy_qp = rvt_destroy_qp,
+ .destroy_srq = rvt_destroy_srq,
+ .detach_mcast = rvt_detach_mcast,
+ .get_dma_mr = rvt_get_dma_mr,
+ .get_numa_node = rvt_get_numa_node,
+ .get_port_immutable = rvt_get_port_immutable,
+ .map_mr_sg = rvt_map_mr_sg,
+ .mmap = rvt_mmap,
+ .modify_ah = rvt_modify_ah,
+ .modify_device = rvt_modify_device,
+ .modify_port = rvt_modify_port,
+ .modify_qp = rvt_modify_qp,
+ .modify_srq = rvt_modify_srq,
+ .poll_cq = rvt_poll_cq,
+ .post_recv = rvt_post_recv,
+ .post_send = rvt_post_send,
+ .post_srq_recv = rvt_post_srq_recv,
+ .query_ah = rvt_query_ah,
+ .query_device = rvt_query_device,
+ .query_gid = rvt_query_gid,
+ .query_pkey = rvt_query_pkey,
+ .query_port = rvt_query_port,
+ .query_qp = rvt_query_qp,
+ .query_srq = rvt_query_srq,
+ .reg_user_mr = rvt_reg_user_mr,
+ .req_notify_cq = rvt_req_notify_cq,
+ .resize_cq = rvt_resize_cq,
+
+ INIT_RDMA_OBJ_SIZE(ib_ah, rvt_ah, ibah),
+ INIT_RDMA_OBJ_SIZE(ib_cq, rvt_cq, ibcq),
+ INIT_RDMA_OBJ_SIZE(ib_pd, rvt_pd, ibpd),
+ INIT_RDMA_OBJ_SIZE(ib_qp, rvt_qp, ibqp),
+ INIT_RDMA_OBJ_SIZE(ib_srq, rvt_srq, ibsrq),
+ INIT_RDMA_OBJ_SIZE(ib_ucontext, rvt_ucontext, ibucontext),
+};
+
+static noinline int check_support(struct rvt_dev_info *rdi, int verb)
+{
+ switch (verb) {
+ case MISC:
+ /*
+ * These functions are not part of verbs specifically but are
+ * required for rdmavt to function.
+ */
+ if ((!rdi->ibdev.ops.port_groups) ||
+ (!rdi->driver_f.get_pci_dev))
+ return -EINVAL;
+ break;
+
+ case MODIFY_DEVICE:
+ /*
+ * rdmavt does not support modify device currently drivers must
+ * provide.
+ */
+ if (!rdi->ibdev.ops.modify_device)
+ return -EOPNOTSUPP;
+ break;
+
+ case QUERY_PORT:
+ if (!rdi->ibdev.ops.query_port)
+ if (!rdi->driver_f.query_port_state)
+ return -EINVAL;
+ break;
+
+ case MODIFY_PORT:
+ if (!rdi->ibdev.ops.modify_port)
+ if (!rdi->driver_f.cap_mask_chg ||
+ !rdi->driver_f.shut_down_port)
+ return -EINVAL;
+ break;
+
+ case QUERY_GID:
+ if (!rdi->ibdev.ops.query_gid)
+ if (!rdi->driver_f.get_guid_be)
+ return -EINVAL;
+ break;
+
+ case CREATE_QP:
+ if (!rdi->ibdev.ops.create_qp)
+ if (!rdi->driver_f.qp_priv_alloc ||
+ !rdi->driver_f.qp_priv_free ||
+ !rdi->driver_f.notify_qp_reset ||
+ !rdi->driver_f.flush_qp_waiters ||
+ !rdi->driver_f.stop_send_queue ||
+ !rdi->driver_f.quiesce_qp)
+ return -EINVAL;
+ break;
+
+ case MODIFY_QP:
+ if (!rdi->ibdev.ops.modify_qp)
+ if (!rdi->driver_f.notify_qp_reset ||
+ !rdi->driver_f.schedule_send ||
+ !rdi->driver_f.get_pmtu_from_attr ||
+ !rdi->driver_f.flush_qp_waiters ||
+ !rdi->driver_f.stop_send_queue ||
+ !rdi->driver_f.quiesce_qp ||
+ !rdi->driver_f.notify_error_qp ||
+ !rdi->driver_f.mtu_from_qp ||
+ !rdi->driver_f.mtu_to_path_mtu)
+ return -EINVAL;
+ break;
+
+ case DESTROY_QP:
+ if (!rdi->ibdev.ops.destroy_qp)
+ if (!rdi->driver_f.qp_priv_free ||
+ !rdi->driver_f.notify_qp_reset ||
+ !rdi->driver_f.flush_qp_waiters ||
+ !rdi->driver_f.stop_send_queue ||
+ !rdi->driver_f.quiesce_qp)
+ return -EINVAL;
+ break;
+
+ case POST_SEND:
+ if (!rdi->ibdev.ops.post_send)
+ if (!rdi->driver_f.schedule_send ||
+ !rdi->driver_f.do_send ||
+ !rdi->post_parms)
+ return -EINVAL;
+ break;
+
+ }
+
+ return 0;
+}
+
+/**
+ * rvt_register_device - register a driver
+ * @rdi: main dev structure for all of rdmavt operations
+ *
+ * It is up to drivers to allocate the rdi and fill in the appropriate
+ * information.
+ *
+ * Return: 0 on success otherwise an errno.
+ */
+int rvt_register_device(struct rvt_dev_info *rdi)
+{
+ int ret = 0, i;
+
+ if (!rdi)
+ return -EINVAL;
+
+ /*
+ * Check to ensure drivers have setup the required helpers for the verbs
+ * they want rdmavt to handle
+ */
+ for (i = 0; i < _VERB_IDX_MAX; i++)
+ if (check_support(rdi, i)) {
+ pr_err("Driver support req not met at %d\n", i);
+ return -EINVAL;
+ }
+
+ ib_set_device_ops(&rdi->ibdev, &rvt_dev_ops);
+
+ /* Once we get past here we can use rvt_pr macros and tracepoints */
+ trace_rvt_dbg(rdi, "Driver attempting registration");
+ rvt_mmap_init(rdi);
+
+ /* Queue Pairs */
+ ret = rvt_driver_qp_init(rdi);
+ if (ret) {
+ pr_err("Error in driver QP init.\n");
+ return -EINVAL;
+ }
+
+ /* Address Handle */
+ spin_lock_init(&rdi->n_ahs_lock);
+ rdi->n_ahs_allocated = 0;
+
+ /* Shared Receive Queue */
+ rvt_driver_srq_init(rdi);
+
+ /* Multicast */
+ rvt_driver_mcast_init(rdi);
+
+ /* Mem Region */
+ ret = rvt_driver_mr_init(rdi);
+ if (ret) {
+ pr_err("Error in driver MR init.\n");
+ goto bail_no_mr;
+ }
+
+ /* Memory Working Set Size */
+ ret = rvt_wss_init(rdi);
+ if (ret) {
+ rvt_pr_err(rdi, "Error in WSS init.\n");
+ goto bail_mr;
+ }
+
+ /* Completion queues */
+ spin_lock_init(&rdi->n_cqs_lock);
+
+ /* Protection Domain */
+ spin_lock_init(&rdi->n_pds_lock);
+ rdi->n_pds_allocated = 0;
+
+ /*
+ * There are some things which could be set by underlying drivers but
+ * really should be up to rdmavt to set. For instance drivers can't know
+ * exactly which functions rdmavt supports, nor do they know the ABI
+ * version, so we do all of this sort of stuff here.
+ */
+ rdi->ibdev.uverbs_cmd_mask |=
+ (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
+ (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
+ (1ull << IB_USER_VERBS_CMD_POST_SEND) |
+ (1ull << IB_USER_VERBS_CMD_POST_RECV) |
+ (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
+ rdi->ibdev.node_type = RDMA_NODE_IB_CA;
+ if (!rdi->ibdev.num_comp_vectors)
+ rdi->ibdev.num_comp_vectors = 1;
+
+ /* We are now good to announce we exist */
+ ret = ib_register_device(&rdi->ibdev, dev_name(&rdi->ibdev.dev), NULL);
+ if (ret) {
+ rvt_pr_err(rdi, "Failed to register driver with ib core.\n");
+ goto bail_wss;
+ }
+
+ rvt_create_mad_agents(rdi);
+
+ rvt_pr_info(rdi, "Registration with rdmavt done.\n");
+ return ret;
+
+bail_wss:
+ rvt_wss_exit(rdi);
+bail_mr:
+ rvt_mr_exit(rdi);
+
+bail_no_mr:
+ rvt_qp_exit(rdi);
+
+ return ret;
+}
+EXPORT_SYMBOL(rvt_register_device);
+
+/**
+ * rvt_unregister_device - remove a driver
+ * @rdi: rvt dev struct
+ */
+void rvt_unregister_device(struct rvt_dev_info *rdi)
+{
+ trace_rvt_dbg(rdi, "Driver is unregistering.");
+ if (!rdi)
+ return;
+
+ rvt_free_mad_agents(rdi);
+
+ ib_unregister_device(&rdi->ibdev);
+ rvt_wss_exit(rdi);
+ rvt_mr_exit(rdi);
+ rvt_qp_exit(rdi);
+}
+EXPORT_SYMBOL(rvt_unregister_device);
+
+/**
+ * rvt_init_port - init internal data for driver port
+ * @rdi: rvt_dev_info struct
+ * @port: rvt port
+ * @port_index: 0 based index of ports, different from IB core port num
+ * @pkey_table: pkey_table for @port
+ *
+ * Keep track of a list of ports. No need to have a detach port.
+ * They persist until the driver goes away.
+ *
+ * Return: always 0
+ */
+int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
+ int port_index, u16 *pkey_table)
+{
+
+ rdi->ports[port_index] = port;
+ rdi->ports[port_index]->pkey_table = pkey_table;
+
+ return 0;
+}
+EXPORT_SYMBOL(rvt_init_port);
diff --git a/drivers/infiniband/sw/rdmavt/vt.h b/drivers/infiniband/sw/rdmavt/vt.h
new file mode 100644
index 0000000000..461574e3f6
--- /dev/null
+++ b/drivers/infiniband/sw/rdmavt/vt.h
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
+/*
+ * Copyright(c) 2016 Intel Corporation.
+ */
+
+#ifndef DEF_RDMAVT_H
+#define DEF_RDMAVT_H
+
+#include <rdma/rdma_vt.h>
+#include <linux/pci.h>
+#include "pd.h"
+#include "qp.h"
+#include "ah.h"
+#include "mr.h"
+#include "srq.h"
+#include "mcast.h"
+#include "mmap.h"
+#include "cq.h"
+#include "mad.h"
+
+#define rvt_pr_info(rdi, fmt, ...) \
+ __rvt_pr_info(rdi->driver_f.get_pci_dev(rdi), \
+ rvt_get_ibdev_name(rdi), \
+ fmt, \
+ ##__VA_ARGS__)
+
+#define rvt_pr_warn(rdi, fmt, ...) \
+ __rvt_pr_warn(rdi->driver_f.get_pci_dev(rdi), \
+ rvt_get_ibdev_name(rdi), \
+ fmt, \
+ ##__VA_ARGS__)
+
+#define rvt_pr_err(rdi, fmt, ...) \
+ __rvt_pr_err(rdi->driver_f.get_pci_dev(rdi), \
+ rvt_get_ibdev_name(rdi), \
+ fmt, \
+ ##__VA_ARGS__)
+
+#define rvt_pr_err_ratelimited(rdi, fmt, ...) \
+ __rvt_pr_err_ratelimited((rdi)->driver_f.get_pci_dev(rdi), \
+ rvt_get_ibdev_name(rdi), \
+ fmt, \
+ ##__VA_ARGS__)
+
+#define __rvt_pr_info(pdev, name, fmt, ...) \
+ dev_info(&pdev->dev, "%s: " fmt, name, ##__VA_ARGS__)
+
+#define __rvt_pr_warn(pdev, name, fmt, ...) \
+ dev_warn(&pdev->dev, "%s: " fmt, name, ##__VA_ARGS__)
+
+#define __rvt_pr_err(pdev, name, fmt, ...) \
+ dev_err(&pdev->dev, "%s: " fmt, name, ##__VA_ARGS__)
+
+#define __rvt_pr_err_ratelimited(pdev, name, fmt, ...) \
+ dev_err_ratelimited(&(pdev)->dev, "%s: " fmt, name, ##__VA_ARGS__)
+
+static inline u32 ibport_num_to_idx(struct ib_device *ibdev, u32 port_num)
+{
+ return port_num - 1; /* IB ports start at 1 our arrays at 0 */
+}
+
+#endif /* DEF_RDMAVT_H */