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

Pull rdma fixes from Jason Gunthorpe:
 "A small set of late-rc patches, mostly fixes for various crashers,
  some syzkaller fixes and a mlx5 HW limitation:

   - Several MAINTAINERS updates

   - Memory leak regression in ODP

   - Several fixes for syzkaller related crashes. Google recently taught
     syzkaller to create the software RDMA devices

   - Crash fixes for HFI1

   - Several fixes for mlx5 crashes

   - Prevent unprivileged access to an unsafe mlx5 HW resource"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/mlx5: Block delay drop to unprivileged users
  RDMA/mlx5: Fix access to wrong pointer while performing flush due to error
  RDMA/core: Ensure security pkey modify is not lost
  MAINTAINERS: Clean RXE section and add Zhu as RXE maintainer
  IB/hfi1: Ensure pq is not left on waitlist
  IB/rdmavt: Free kernel completion queue when done
  RDMA/mad: Do not crash if the rdma device does not have a umad interface
  RDMA/core: Fix missing error check on dev_set_name()
  RDMA/nl: Do not permit empty devices names during RDMA_NLDEV_CMD_NEWLINK/SET
  RDMA/mlx5: Fix the number of hwcounters of a dynamic counter
  MAINTAINERS: Update maintainers for HISILICON ROCE DRIVER
  RDMA/odp: Fix leaking the tgid for implicit ODP

1 files changed, 25 insertions, 2 deletions

diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index 367a71bc5f4b..3dec3de903b7 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -330,6 +330,22 @@ static void mlx5_handle_error_cqe(struct mlx5_ib_dev *dev,
 		dump_cqe(dev, cqe);
 }
 
+static void handle_atomics(struct mlx5_ib_qp *qp, struct mlx5_cqe64 *cqe64,
+			   u16 tail, u16 head)
+{
+	u16 idx;
+
+	do {
+		idx = tail & (qp->sq.wqe_cnt - 1);
+		if (idx == head)
+			break;
+
+		tail = qp->sq.w_list[idx].next;
+	} while (1);
+	tail = qp->sq.w_list[idx].next;
+	qp->sq.last_poll = tail;
+}
+
 static void free_cq_buf(struct mlx5_ib_dev *dev, struct mlx5_ib_cq_buf *buf)
 {
 	mlx5_frag_buf_free(dev->mdev, &buf->frag_buf);
@@ -368,7 +384,7 @@ static void get_sig_err_item(struct mlx5_sig_err_cqe *cqe,
 }
 
 static void sw_comp(struct mlx5_ib_qp *qp, int num_entries, struct ib_wc *wc,
-		    int *npolled, int is_send)
+		    int *npolled, bool is_send)
 {
 	struct mlx5_ib_wq *wq;
 	unsigned int cur;
@@ -383,10 +399,16 @@ static void sw_comp(struct mlx5_ib_qp *qp, int num_entries, struct ib_wc *wc,
 		return;
 
 	for (i = 0;  i < cur && np < num_entries; i++) {
-		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		unsigned int idx;
+
+		idx = (is_send) ? wq->last_poll : wq->tail;
+		idx &= (wq->wqe_cnt - 1);
+		wc->wr_id = wq->wrid[idx];
 		wc->status = IB_WC_WR_FLUSH_ERR;
 		wc->vendor_err = MLX5_CQE_SYNDROME_WR_FLUSH_ERR;
 		wq->tail++;
+		if (is_send)
+			wq->last_poll = wq->w_list[idx].next;
 		np++;
 		wc->qp = &qp->ibqp;
 		wc++;
@@ -473,6 +495,7 @@ repoll:
 		wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
 		idx = wqe_ctr & (wq->wqe_cnt - 1);
 		handle_good_req(wc, cqe64, wq, idx);
+		handle_atomics(*cur_qp, cqe64, wq->last_poll, idx);
 		wc->wr_id = wq->wrid[idx];
 		wq->tail = wq->wqe_head[idx] + 1;
 		wc->status = IB_WC_SUCCESS;