Update Linux to v5.4.148
Sourced from [1]
[1] https://cdn.kernel.org/pub/linux/kernel/v5.x/linux-5.4.148.tar.gz
Change-Id: Ib3d26c5ba9b022e2e03533005c4fed4d7c30b61b
Signed-off-by: Olivier Deprez <olivier.deprez@arm.com>
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index dc71b6e..93040c9 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -357,9 +357,10 @@
return -ENOMEM;
spin_lock_init(&dev->sgid_lock);
+ xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
if (IS_IWARP(dev)) {
- xa_init_flags(&dev->qps, XA_FLAGS_LOCK_IRQ);
+ xa_init(&dev->qps);
dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
}
@@ -600,7 +601,7 @@
qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
/* Part 2 - check capabilities */
- page_size = ~dev->attr.page_size_caps + 1;
+ page_size = ~qed_attr->page_size_caps + 1;
if (page_size > PAGE_SIZE) {
DP_ERR(dev,
"Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 0cfd849..ed56df3 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -40,6 +40,7 @@
#include <linux/qed/qed_rdma_if.h>
#include <linux/qed/qede_rdma.h>
#include <linux/qed/roce_common.h>
+#include <linux/completion.h>
#include "qedr_hsi_rdma.h"
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
@@ -348,10 +349,10 @@
u32 wqe_prod;
u32 sge_prod;
u32 wr_prod_cnt;
- u32 wr_cons_cnt;
+ atomic_t wr_cons_cnt;
u32 num_elems;
- u32 *virt_prod_pair_addr;
+ struct rdma_srq_producers *virt_prod_pair_addr;
dma_addr_t phy_prod_pair_addr;
};
@@ -377,10 +378,20 @@
QEDR_QP_ERR_RQ_PBL_FULL = 32,
};
+enum qedr_qp_create_type {
+ QEDR_QP_CREATE_NONE,
+ QEDR_QP_CREATE_USER,
+ QEDR_QP_CREATE_KERNEL,
+};
+
+enum qedr_iwarp_cm_flags {
+ QEDR_IWARP_CM_WAIT_FOR_CONNECT = BIT(0),
+ QEDR_IWARP_CM_WAIT_FOR_DISCONNECT = BIT(1),
+};
+
struct qedr_qp {
struct ib_qp ibqp; /* must be first */
struct qedr_dev *dev;
- struct qedr_iw_ep *ep;
struct qedr_qp_hwq_info sq;
struct qedr_qp_hwq_info rq;
@@ -395,6 +406,7 @@
u32 id;
struct qedr_pd *pd;
enum ib_qp_type qp_type;
+ enum qedr_qp_create_type create_type;
struct qed_rdma_qp *qed_qp;
u32 qp_id;
u16 icid;
@@ -437,8 +449,11 @@
/* Relevant to qps created from user space only (applications) */
struct qedr_userq usq;
struct qedr_userq urq;
- atomic_t refcnt;
- bool destroyed;
+
+ /* synchronization objects used with iwarp ep */
+ struct kref refcnt;
+ struct completion iwarp_cm_comp;
+ unsigned long iwarp_cm_flags; /* enum iwarp_cm_flags */
};
struct qedr_ah {
@@ -531,7 +546,7 @@
struct iw_cm_id *cm_id;
struct qedr_qp *qp;
void *qed_context;
- u8 during_connect;
+ struct kref refcnt;
};
static inline
diff --git a/drivers/infiniband/hw/qedr/qedr_iw_cm.c b/drivers/infiniband/hw/qedr/qedr_iw_cm.c
index 22881d4..b9b46a0 100644
--- a/drivers/infiniband/hw/qedr/qedr_iw_cm.c
+++ b/drivers/infiniband/hw/qedr/qedr_iw_cm.c
@@ -79,6 +79,27 @@
}
}
+static void qedr_iw_free_qp(struct kref *ref)
+{
+ struct qedr_qp *qp = container_of(ref, struct qedr_qp, refcnt);
+
+ kfree(qp);
+}
+
+static void
+qedr_iw_free_ep(struct kref *ref)
+{
+ struct qedr_iw_ep *ep = container_of(ref, struct qedr_iw_ep, refcnt);
+
+ if (ep->qp)
+ kref_put(&ep->qp->refcnt, qedr_iw_free_qp);
+
+ if (ep->cm_id)
+ ep->cm_id->rem_ref(ep->cm_id);
+
+ kfree(ep);
+}
+
static void
qedr_iw_mpa_request(void *context, struct qed_iwarp_cm_event_params *params)
{
@@ -93,6 +114,7 @@
ep->dev = dev;
ep->qed_context = params->ep_context;
+ kref_init(&ep->refcnt);
memset(&event, 0, sizeof(event));
event.event = IW_CM_EVENT_CONNECT_REQUEST;
@@ -128,8 +150,17 @@
if (params->cm_info) {
event.ird = params->cm_info->ird;
event.ord = params->cm_info->ord;
- event.private_data_len = params->cm_info->private_data_len;
- event.private_data = (void *)params->cm_info->private_data;
+ /* Only connect_request and reply have valid private data
+ * the rest of the events this may be left overs from
+ * connection establishment. CONNECT_REQUEST is issued via
+ * qedr_iw_mpa_request
+ */
+ if (event_type == IW_CM_EVENT_CONNECT_REPLY) {
+ event.private_data_len =
+ params->cm_info->private_data_len;
+ event.private_data =
+ (void *)params->cm_info->private_data;
+ }
}
if (ep->cm_id)
@@ -141,12 +172,10 @@
{
struct qedr_iw_ep *ep = (struct qedr_iw_ep *)context;
- if (ep->cm_id) {
+ if (ep->cm_id)
qedr_iw_issue_event(context, params, IW_CM_EVENT_CLOSE);
- ep->cm_id->rem_ref(ep->cm_id);
- ep->cm_id = NULL;
- }
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
}
static void
@@ -186,11 +215,13 @@
struct qedr_qp *qp = ep->qp;
struct iw_cm_event event;
- if (qp->destroyed) {
- kfree(dwork);
- qedr_iw_qp_rem_ref(&qp->ibqp);
- return;
- }
+ /* The qp won't be released until we release the ep.
+ * the ep's refcnt was increased before calling this
+ * function, therefore it is safe to access qp
+ */
+ if (test_and_set_bit(QEDR_IWARP_CM_WAIT_FOR_DISCONNECT,
+ &qp->iwarp_cm_flags))
+ goto out;
memset(&event, 0, sizeof(event));
event.status = dwork->status;
@@ -204,7 +235,6 @@
else
qp_params.new_state = QED_ROCE_QP_STATE_SQD;
- kfree(dwork);
if (ep->cm_id)
ep->cm_id->event_handler(ep->cm_id, &event);
@@ -214,7 +244,10 @@
dev->ops->rdma_modify_qp(dev->rdma_ctx, qp->qed_qp, &qp_params);
- qedr_iw_qp_rem_ref(&qp->ibqp);
+ complete(&ep->qp->iwarp_cm_comp);
+out:
+ kfree(dwork);
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
}
static void
@@ -224,13 +257,17 @@
struct qedr_discon_work *work;
struct qedr_iw_ep *ep = (struct qedr_iw_ep *)context;
struct qedr_dev *dev = ep->dev;
- struct qedr_qp *qp = ep->qp;
work = kzalloc(sizeof(*work), GFP_ATOMIC);
if (!work)
return;
- qedr_iw_qp_add_ref(&qp->ibqp);
+ /* We can't get a close event before disconnect, but since
+ * we're scheduling a work queue we need to make sure close
+ * won't delete the ep, so we increase the refcnt
+ */
+ kref_get(&ep->refcnt);
+
work->ep = ep;
work->event = params->event;
work->status = params->status;
@@ -252,16 +289,30 @@
if ((params->status == -ECONNREFUSED) && (!ep->qp)) {
DP_DEBUG(dev, QEDR_MSG_IWARP,
"PASSIVE connection refused releasing ep...\n");
- kfree(ep);
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
return;
}
+ complete(&ep->qp->iwarp_cm_comp);
qedr_iw_issue_event(context, params, IW_CM_EVENT_ESTABLISHED);
if (params->status < 0)
qedr_iw_close_event(context, params);
}
+static void
+qedr_iw_active_complete(void *context,
+ struct qed_iwarp_cm_event_params *params)
+{
+ struct qedr_iw_ep *ep = (struct qedr_iw_ep *)context;
+
+ complete(&ep->qp->iwarp_cm_comp);
+ qedr_iw_issue_event(context, params, IW_CM_EVENT_CONNECT_REPLY);
+
+ if (params->status < 0)
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
+}
+
static int
qedr_iw_mpa_reply(void *context, struct qed_iwarp_cm_event_params *params)
{
@@ -288,27 +339,15 @@
qedr_iw_mpa_reply(context, params);
break;
case QED_IWARP_EVENT_PASSIVE_COMPLETE:
- ep->during_connect = 0;
qedr_iw_passive_complete(context, params);
break;
-
case QED_IWARP_EVENT_ACTIVE_COMPLETE:
- ep->during_connect = 0;
- qedr_iw_issue_event(context,
- params,
- IW_CM_EVENT_CONNECT_REPLY);
- if (params->status < 0) {
- struct qedr_iw_ep *ep = (struct qedr_iw_ep *)context;
-
- ep->cm_id->rem_ref(ep->cm_id);
- ep->cm_id = NULL;
- }
+ qedr_iw_active_complete(context, params);
break;
case QED_IWARP_EVENT_DISCONNECT:
qedr_iw_disconnect_event(context, params);
break;
case QED_IWARP_EVENT_CLOSE:
- ep->during_connect = 0;
qedr_iw_close_event(context, params);
break;
case QED_IWARP_EVENT_RQ_EMPTY:
@@ -451,10 +490,10 @@
if ((!dst) || dst->error) {
if (dst) {
- dst_release(dst);
DP_ERR(dev,
"ip6_route_output returned dst->error = %d\n",
dst->error);
+ dst_release(dst);
}
return -EINVAL;
}
@@ -476,6 +515,19 @@
return rc;
}
+struct qedr_qp *qedr_iw_load_qp(struct qedr_dev *dev, u32 qpn)
+{
+ struct qedr_qp *qp;
+
+ xa_lock(&dev->qps);
+ qp = xa_load(&dev->qps, qpn);
+ if (qp)
+ kref_get(&qp->refcnt);
+ xa_unlock(&dev->qps);
+
+ return qp;
+}
+
int qedr_iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
{
struct qedr_dev *dev = get_qedr_dev(cm_id->device);
@@ -491,10 +543,6 @@
int rc = 0;
int i;
- qp = xa_load(&dev->qps, conn_param->qpn);
- if (unlikely(!qp))
- return -EINVAL;
-
laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
@@ -516,8 +564,15 @@
return -ENOMEM;
ep->dev = dev;
+ kref_init(&ep->refcnt);
+
+ qp = qedr_iw_load_qp(dev, conn_param->qpn);
+ if (!qp) {
+ rc = -EINVAL;
+ goto err;
+ }
+
ep->qp = qp;
- qp->ep = ep;
cm_id->add_ref(cm_id);
ep->cm_id = cm_id;
@@ -580,16 +635,22 @@
in_params.qp = qp->qed_qp;
memcpy(in_params.local_mac_addr, dev->ndev->dev_addr, ETH_ALEN);
- ep->during_connect = 1;
+ if (test_and_set_bit(QEDR_IWARP_CM_WAIT_FOR_CONNECT,
+ &qp->iwarp_cm_flags)) {
+ rc = -ENODEV;
+ goto err; /* QP already being destroyed */
+ }
+
rc = dev->ops->iwarp_connect(dev->rdma_ctx, &in_params, &out_params);
- if (rc)
+ if (rc) {
+ complete(&qp->iwarp_cm_comp);
goto err;
+ }
return rc;
err:
- cm_id->rem_ref(cm_id);
- kfree(ep);
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
return rc;
}
@@ -668,6 +729,7 @@
listener->qed_handle);
cm_id->rem_ref(cm_id);
+ kfree(listener);
return rc;
}
@@ -681,14 +743,13 @@
DP_DEBUG(dev, QEDR_MSG_IWARP, "Accept on qpid=%d\n", conn_param->qpn);
- qp = xa_load(&dev->qps, conn_param->qpn);
+ qp = qedr_iw_load_qp(dev, conn_param->qpn);
if (!qp) {
DP_ERR(dev, "Invalid QP number %d\n", conn_param->qpn);
return -EINVAL;
}
ep->qp = qp;
- qp->ep = ep;
cm_id->add_ref(cm_id);
ep->cm_id = cm_id;
@@ -700,15 +761,23 @@
params.ird = conn_param->ird;
params.ord = conn_param->ord;
- ep->during_connect = 1;
+ if (test_and_set_bit(QEDR_IWARP_CM_WAIT_FOR_CONNECT,
+ &qp->iwarp_cm_flags)) {
+ rc = -EINVAL;
+ goto err; /* QP already destroyed */
+ }
+
rc = dev->ops->iwarp_accept(dev->rdma_ctx, ¶ms);
- if (rc)
+ if (rc) {
+ complete(&qp->iwarp_cm_comp);
goto err;
+ }
return rc;
+
err:
- ep->during_connect = 0;
- cm_id->rem_ref(cm_id);
+ kref_put(&ep->refcnt, qedr_iw_free_ep);
+
return rc;
}
@@ -731,17 +800,14 @@
{
struct qedr_qp *qp = get_qedr_qp(ibqp);
- atomic_inc(&qp->refcnt);
+ kref_get(&qp->refcnt);
}
void qedr_iw_qp_rem_ref(struct ib_qp *ibqp)
{
struct qedr_qp *qp = get_qedr_qp(ibqp);
- if (atomic_dec_and_test(&qp->refcnt)) {
- xa_erase_irq(&qp->dev->qps, qp->qp_id);
- kfree(qp);
- }
+ kref_put(&qp->refcnt, qedr_iw_free_qp);
}
struct ib_qp *qedr_iw_get_qp(struct ib_device *ibdev, int qpn)
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 6f3ce86..4408d33 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -51,6 +51,7 @@
#include "verbs.h"
#include <rdma/qedr-abi.h>
#include "qedr_roce_cm.h"
+#include "qedr_iw_cm.h"
#define QEDR_SRQ_WQE_ELEM_SIZE sizeof(union rdma_srq_elm)
#define RDMA_MAX_SGE_PER_SRQ (4)
@@ -1193,7 +1194,10 @@
struct ib_qp_init_attr *attrs)
{
spin_lock_init(&qp->q_lock);
- atomic_set(&qp->refcnt, 1);
+ if (rdma_protocol_iwarp(&dev->ibdev, 1)) {
+ kref_init(&qp->refcnt);
+ init_completion(&qp->iwarp_cm_comp);
+ }
qp->pd = pd;
qp->qp_type = attrs->qp_type;
qp->max_inline_data = attrs->cap.max_inline_data;
@@ -1577,6 +1581,14 @@
ib_umem_release(qp->urq.umem);
qp->urq.umem = NULL;
+
+ if (rdma_protocol_roce(&dev->ibdev, 1)) {
+ qedr_free_pbl(dev, &qp->usq.pbl_info, qp->usq.pbl_tbl);
+ qedr_free_pbl(dev, &qp->urq.pbl_info, qp->urq.pbl_tbl);
+ } else {
+ kfree(qp->usq.pbl_tbl);
+ kfree(qp->urq.pbl_tbl);
+ }
}
static int qedr_create_user_qp(struct qedr_dev *dev,
@@ -1592,6 +1604,7 @@
int alloc_and_init = rdma_protocol_roce(&dev->ibdev, 1);
int rc = -EINVAL;
+ qp->create_type = QEDR_QP_CREATE_USER;
memset(&ureq, 0, sizeof(ureq));
rc = ib_copy_from_udata(&ureq, udata, sizeof(ureq));
if (rc) {
@@ -1805,6 +1818,7 @@
u32 n_sq_entries;
memset(&in_params, 0, sizeof(in_params));
+ qp->create_type = QEDR_QP_CREATE_KERNEL;
/* A single work request may take up to QEDR_MAX_SQ_WQE_SIZE elements in
* the ring. The ring should allow at least a single WR, even if the
@@ -1918,7 +1932,7 @@
qp->ibqp.qp_num = qp->qp_id;
if (rdma_protocol_iwarp(&dev->ibdev, 1)) {
- rc = xa_insert_irq(&dev->qps, qp->qp_id, qp, GFP_KERNEL);
+ rc = xa_insert(&dev->qps, qp->qp_id, qp, GFP_KERNEL);
if (rc)
goto err;
}
@@ -2391,7 +2405,7 @@
qp_attr->cap.max_recv_wr = qp->rq.max_wr;
qp_attr->cap.max_send_sge = qp->sq.max_sges;
qp_attr->cap.max_recv_sge = qp->rq.max_sges;
- qp_attr->cap.max_inline_data = ROCE_REQ_MAX_INLINE_DATA_SIZE;
+ qp_attr->cap.max_inline_data = dev->attr.max_inline;
qp_init_attr->cap = qp_attr->cap;
qp_attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
@@ -2437,7 +2451,7 @@
return rc;
}
- if (udata)
+ if (qp->create_type == QEDR_QP_CREATE_USER)
qedr_cleanup_user(dev, qp);
else
qedr_cleanup_kernel(dev, qp);
@@ -2467,34 +2481,46 @@
qedr_modify_qp(ibqp, &attr, attr_mask, NULL);
}
} else {
- /* Wait for the connect/accept to complete */
- if (qp->ep) {
- int wait_count = 1;
+ /* If connection establishment started the WAIT_FOR_CONNECT
+ * bit will be on and we need to Wait for the establishment
+ * to complete before destroying the qp.
+ */
+ if (test_and_set_bit(QEDR_IWARP_CM_WAIT_FOR_CONNECT,
+ &qp->iwarp_cm_flags))
+ wait_for_completion(&qp->iwarp_cm_comp);
- while (qp->ep->during_connect) {
- DP_DEBUG(dev, QEDR_MSG_QP,
- "Still in during connect/accept\n");
-
- msleep(100);
- if (wait_count++ > 200) {
- DP_NOTICE(dev,
- "during connect timeout\n");
- break;
- }
- }
- }
+ /* If graceful disconnect started, the WAIT_FOR_DISCONNECT
+ * bit will be on, and we need to wait for the disconnect to
+ * complete before continuing. We can use the same completion,
+ * iwarp_cm_comp, since this is the only place that waits for
+ * this completion and it is sequential. In addition,
+ * disconnect can't occur before the connection is fully
+ * established, therefore if WAIT_FOR_DISCONNECT is on it
+ * means WAIT_FOR_CONNECT is also on and the completion for
+ * CONNECT already occurred.
+ */
+ if (test_and_set_bit(QEDR_IWARP_CM_WAIT_FOR_DISCONNECT,
+ &qp->iwarp_cm_flags))
+ wait_for_completion(&qp->iwarp_cm_comp);
}
if (qp->qp_type == IB_QPT_GSI)
qedr_destroy_gsi_qp(dev);
+ /* We need to remove the entry from the xarray before we release the
+ * qp_id to avoid a race of the qp_id being reallocated and failing
+ * on xa_insert
+ */
+ if (rdma_protocol_iwarp(&dev->ibdev, 1))
+ xa_erase(&dev->qps, qp->qp_id);
+
qedr_free_qp_resources(dev, qp, udata);
- if (atomic_dec_and_test(&qp->refcnt) &&
- rdma_protocol_iwarp(&dev->ibdev, 1)) {
- xa_erase_irq(&dev->qps, qp->qp_id);
+ if (rdma_protocol_iwarp(&dev->ibdev, 1))
+ qedr_iw_qp_rem_ref(&qp->ibqp);
+ else
kfree(qp);
- }
+
return 0;
}
@@ -2673,8 +2699,8 @@
dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
- if ((mr->type != QEDR_MR_DMA) && (mr->type != QEDR_MR_FRMR))
- qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+ if (mr->type != QEDR_MR_DMA)
+ free_mr_info(dev, &mr->info);
/* it could be user registered memory. */
ib_umem_release(mr->umem);
@@ -3436,7 +3462,7 @@
* count and consumer count and subtract it from max
* work request supported so that we get elements left.
*/
- used = hw_srq->wr_prod_cnt - hw_srq->wr_cons_cnt;
+ used = hw_srq->wr_prod_cnt - (u32)atomic_read(&hw_srq->wr_cons_cnt);
return hw_srq->max_wr - used;
}
@@ -3451,7 +3477,6 @@
unsigned long flags;
int status = 0;
u32 num_sge;
- u32 offset;
spin_lock_irqsave(&srq->lock, flags);
@@ -3464,7 +3489,8 @@
if (!qedr_srq_elem_left(hw_srq) ||
wr->num_sge > srq->hw_srq.max_sges) {
DP_ERR(dev, "Can't post WR (%d,%d) || (%d > %d)\n",
- hw_srq->wr_prod_cnt, hw_srq->wr_cons_cnt,
+ hw_srq->wr_prod_cnt,
+ atomic_read(&hw_srq->wr_cons_cnt),
wr->num_sge, srq->hw_srq.max_sges);
status = -ENOMEM;
*bad_wr = wr;
@@ -3498,22 +3524,20 @@
hw_srq->sge_prod++;
}
- /* Flush WQE and SGE information before
+ /* Update WQE and SGE information before
* updating producer.
*/
- wmb();
+ dma_wmb();
/* SRQ producer is 8 bytes. Need to update SGE producer index
* in first 4 bytes and need to update WQE producer in
* next 4 bytes.
*/
- *srq->hw_srq.virt_prod_pair_addr = hw_srq->sge_prod;
- offset = offsetof(struct rdma_srq_producers, wqe_prod);
- *((u8 *)srq->hw_srq.virt_prod_pair_addr + offset) =
- hw_srq->wqe_prod;
+ srq->hw_srq.virt_prod_pair_addr->sge_prod = hw_srq->sge_prod;
+ /* Make sure sge producer is updated first */
+ dma_wmb();
+ srq->hw_srq.virt_prod_pair_addr->wqe_prod = hw_srq->wqe_prod;
- /* Flush producer after updating it. */
- wmb();
wr = wr->next;
}
@@ -3932,7 +3956,7 @@
} else {
__process_resp_one(dev, qp, cq, wc, resp, wr_id);
}
- srq->hw_srq.wr_cons_cnt++;
+ atomic_inc(&srq->hw_srq.wr_cons_cnt);
return 1;
}