aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
diff options
context:
space:
mode:
authorXi Wang <wangxi11@huawei.com>2020-01-09 20:20:12 +0800
committerJason Gunthorpe <jgg@mellanox.com>2020-01-10 11:17:37 -0400
commit626903e9355bdf8d401fc0ac7e7407862c642710 (patch)
tree5abc505c557112abb7f420ec8890ce06ea0b5bd0 /drivers/infiniband/hw/hns/hns_roce_hw_v2.c
parentRDMA/hns: Bugfix for posting a wqe with sge (diff)
downloadlinux-dev-626903e9355bdf8d401fc0ac7e7407862c642710.tar.xz
linux-dev-626903e9355bdf8d401fc0ac7e7407862c642710.zip
RDMA/hns: Add support for reporting wc as software mode
When hardware is in resetting stage, we may can't poll back all the expected work completions as the hardware won't generate cqe anymore. This patch allows the driver to compose the expected wc instead of the hardware during resetting stage. Once the hardware finished resetting, we can poll cq from hardware again. Link: https://lore.kernel.org/r/1578572412-25756-1-git-send-email-liweihang@huawei.com Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Weihang Li <liweihang@huawei.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Diffstat (limited to 'drivers/infiniband/hw/hns/hns_roce_hw_v2.c')
-rw-r--r--drivers/infiniband/hw/hns/hns_roce_hw_v2.c158
1 files changed, 131 insertions, 27 deletions
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index fb16908652e7..f1e0ba6a11d5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -225,6 +225,30 @@ static int hns_roce_v2_modify_qp(struct ib_qp *ibqp,
int attr_mask, enum ib_qp_state cur_state,
enum ib_qp_state new_state);
+static int check_send_valid(struct hns_roce_dev *hr_dev,
+ struct hns_roce_qp *hr_qp)
+{
+ struct ib_qp *ibqp = &hr_qp->ibqp;
+ struct device *dev = hr_dev->dev;
+
+ if (unlikely(ibqp->qp_type != IB_QPT_RC &&
+ ibqp->qp_type != IB_QPT_GSI &&
+ ibqp->qp_type != IB_QPT_UD)) {
+ dev_err(dev, "Not supported QP(0x%x)type!\n", ibqp->qp_type);
+ return -EOPNOTSUPP;
+ } else if (unlikely(hr_qp->state == IB_QPS_RESET ||
+ hr_qp->state == IB_QPS_INIT ||
+ hr_qp->state == IB_QPS_RTR)) {
+ dev_err(dev, "Post WQE fail, QP state %d!\n", hr_qp->state);
+ return -EINVAL;
+ } else if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN)) {
+ dev_err(dev, "Post WQE fail, dev state %d!\n", hr_dev->state);
+ return -EIO;
+ }
+
+ return 0;
+}
+
static int hns_roce_v2_post_send(struct ib_qp *ibqp,
const struct ib_send_wr *wr,
const struct ib_send_wr **bad_wr)
@@ -247,28 +271,21 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
bool loopback;
int attr_mask;
u32 tmp_len;
- int ret = 0;
u32 hr_op;
u8 *smac;
int nreq;
+ int ret;
int i;
- if (unlikely(ibqp->qp_type != IB_QPT_RC &&
- ibqp->qp_type != IB_QPT_GSI &&
- ibqp->qp_type != IB_QPT_UD)) {
- dev_err(dev, "Not supported QP(0x%x)type!\n", ibqp->qp_type);
- *bad_wr = wr;
- return -EOPNOTSUPP;
- }
+ spin_lock_irqsave(&qp->sq.lock, flags);
- if (unlikely(qp->state == IB_QPS_RESET || qp->state == IB_QPS_INIT ||
- qp->state == IB_QPS_RTR)) {
- dev_err(dev, "Post WQE fail, QP state %d err!\n", qp->state);
+ ret = check_send_valid(hr_dev, qp);
+ if (ret) {
*bad_wr = wr;
- return -EINVAL;
+ nreq = 0;
+ goto out;
}
- spin_lock_irqsave(&qp->sq.lock, flags);
sge_idx = qp->next_sge;
for (nreq = 0; wr; ++nreq, wr = wr->next) {
@@ -609,6 +626,17 @@ out:
return ret;
}
+static int check_recv_valid(struct hns_roce_dev *hr_dev,
+ struct hns_roce_qp *hr_qp)
+{
+ if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN))
+ return -EIO;
+ else if (hr_qp->state == IB_QPS_RESET)
+ return -EINVAL;
+
+ return 0;
+}
+
static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
const struct ib_recv_wr *wr,
const struct ib_recv_wr **bad_wr)
@@ -623,16 +651,17 @@ static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
void *wqe = NULL;
int attr_mask;
u32 wqe_idx;
- int ret = 0;
int nreq;
+ int ret;
int i;
spin_lock_irqsave(&hr_qp->rq.lock, flags);
- if (hr_qp->state == IB_QPS_RESET) {
- spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
+ ret = check_recv_valid(hr_dev, hr_qp);
+ if (ret) {
*bad_wr = wr;
- return -EINVAL;
+ nreq = 0;
+ goto out;
}
for (nreq = 0; wr; ++nreq, wr = wr->next) {
@@ -2673,6 +2702,55 @@ static int hns_roce_handle_recv_inl_wqe(struct hns_roce_v2_cqe *cqe,
return 0;
}
+static int sw_comp(struct hns_roce_qp *hr_qp, struct hns_roce_wq *wq,
+ int num_entries, struct ib_wc *wc)
+{
+ unsigned int left;
+ int npolled = 0;
+
+ left = wq->head - wq->tail;
+ if (left == 0)
+ return 0;
+
+ left = min_t(unsigned int, (unsigned int)num_entries, left);
+ while (npolled < left) {
+ wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+ wc->status = IB_WC_WR_FLUSH_ERR;
+ wc->vendor_err = 0;
+ wc->qp = &hr_qp->ibqp;
+
+ wq->tail++;
+ wc++;
+ npolled++;
+ }
+
+ return npolled;
+}
+
+static int hns_roce_v2_sw_poll_cq(struct hns_roce_cq *hr_cq, int num_entries,
+ struct ib_wc *wc)
+{
+ struct hns_roce_qp *hr_qp;
+ int npolled = 0;
+
+ list_for_each_entry(hr_qp, &hr_cq->sq_list, sq_node) {
+ npolled += sw_comp(hr_qp, &hr_qp->sq,
+ num_entries - npolled, wc + npolled);
+ if (npolled >= num_entries)
+ goto out;
+ }
+
+ list_for_each_entry(hr_qp, &hr_cq->rq_list, rq_node) {
+ npolled += sw_comp(hr_qp, &hr_qp->rq,
+ num_entries - npolled, wc + npolled);
+ if (npolled >= num_entries)
+ goto out;
+ }
+
+out:
+ return npolled;
+}
+
static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
struct hns_roce_qp **cur_qp, struct ib_wc *wc)
{
@@ -2953,6 +3031,7 @@ static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
struct ib_wc *wc)
{
+ struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
struct hns_roce_qp *cur_qp = NULL;
unsigned long flags;
@@ -2960,6 +3039,18 @@ static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
spin_lock_irqsave(&hr_cq->lock, flags);
+ /*
+ * When the device starts to reset, the state is RST_DOWN. At this time,
+ * there may still be some valid CQEs in the hardware that are not
+ * polled. Therefore, it is not allowed to switch to the software mode
+ * immediately. When the state changes to UNINIT, CQE no longer exists
+ * in the hardware, and then switch to software mode.
+ */
+ if (hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT) {
+ npolled = hns_roce_v2_sw_poll_cq(hr_cq, num_entries, wc);
+ goto out;
+ }
+
for (npolled = 0; npolled < num_entries; ++npolled) {
if (hns_roce_v2_poll_one(hr_cq, &cur_qp, wc + npolled))
break;
@@ -2971,6 +3062,7 @@ static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
hns_roce_v2_cq_set_ci(hr_cq, hr_cq->cons_index);
}
+out:
spin_unlock_irqrestore(&hr_cq->lock, flags);
return npolled;
@@ -4629,6 +4721,7 @@ static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
{
struct hns_roce_cq *send_cq, *recv_cq;
struct ib_device *ibdev = &hr_dev->ib_dev;
+ unsigned long flags;
int ret = 0;
if (hr_qp->ibqp.qp_type == IB_QPT_RC && hr_qp->state != IB_QPS_RESET) {
@@ -4639,21 +4732,32 @@ static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
ibdev_err(ibdev, "modify QP to Reset failed.\n");
}
- send_cq = to_hr_cq(hr_qp->ibqp.send_cq);
- recv_cq = to_hr_cq(hr_qp->ibqp.recv_cq);
+ send_cq = hr_qp->ibqp.send_cq ? to_hr_cq(hr_qp->ibqp.send_cq) : NULL;
+ recv_cq = hr_qp->ibqp.recv_cq ? to_hr_cq(hr_qp->ibqp.recv_cq) : NULL;
+ spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
hns_roce_lock_cqs(send_cq, recv_cq);
+ list_del(&hr_qp->node);
+ list_del(&hr_qp->sq_node);
+ list_del(&hr_qp->rq_node);
+
if (!udata) {
- __hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn, hr_qp->ibqp.srq ?
- to_hr_srq(hr_qp->ibqp.srq) : NULL);
- if (send_cq != recv_cq)
+ if (recv_cq)
+ __hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn,
+ (hr_qp->ibqp.srq ?
+ to_hr_srq(hr_qp->ibqp.srq) :
+ NULL));
+
+ if (send_cq && send_cq != recv_cq)
__hns_roce_v2_cq_clean(send_cq, hr_qp->qpn, NULL);
+
}
hns_roce_qp_remove(hr_dev, hr_qp);
hns_roce_unlock_cqs(send_cq, recv_cq);
+ spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
hns_roce_qp_free(hr_dev, hr_qp);
@@ -6397,6 +6501,10 @@ static void __hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
return;
handle->priv = NULL;
+
+ hr_dev->state = HNS_ROCE_DEVICE_STATE_UNINIT;
+ hns_roce_handle_device_err(hr_dev);
+
hns_roce_exit(hr_dev);
kfree(hr_dev->priv);
ib_dealloc_device(&hr_dev->ib_dev);
@@ -6458,7 +6566,6 @@ static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
{
struct hns_roce_dev *hr_dev;
- struct ib_event event;
if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) {
set_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
@@ -6476,10 +6583,7 @@ static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
hr_dev->active = false;
hr_dev->dis_db = true;
- event.event = IB_EVENT_DEVICE_FATAL;
- event.device = &hr_dev->ib_dev;
- event.element.port_num = 1;
- ib_dispatch_event(&event);
+ hr_dev->state = HNS_ROCE_DEVICE_STATE_RST_DOWN;
return 0;
}