[PATCH 5.2 133/162] IB/hfi1: Drop stale TID RDMA packets that cause TIDErr

From: Greg Kroah-Hartman
Date: Tue Aug 27 2019 - 04:05:44 EST


From: Kaike Wan <kaike.wan@xxxxxxxxx>

commit d9d1f5e7bb82415591e8b62b222cbb88c4797ef3 upstream.

In a congested fabric with adaptive routing enabled, traces show that
packets could be delivered out of order. A stale TID RDMA data packet
could lead to TidErr if the TID entries have been released by duplicate
data packets generated from retries, and subsequently erroneously force
the qp into error state in the current implementation.

Since the payload has already been dropped by hardware, the packet can
be simply dropped and it is no longer necessary to put the qp into
error state.

Fixes: 9905bf06e890 ("IB/hfi1: Add functions to receive TID RDMA READ response")
Cc: <stable@xxxxxxxxxxxxxxx>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@xxxxxxxxx>
Signed-off-by: Kaike Wan <kaike.wan@xxxxxxxxx>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@xxxxxxxxx>
Link: https://lore.kernel.org/r/20190815192058.105923.72324.stgit@xxxxxxxxxxxxxxxxxxxx
Signed-off-by: Doug Ledford <dledford@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
drivers/infiniband/hw/hfi1/tid_rdma.c | 47 ++--------------------------------
1 file changed, 3 insertions(+), 44 deletions(-)

--- a/drivers/infiniband/hw/hfi1/tid_rdma.c
+++ b/drivers/infiniband/hw/hfi1/tid_rdma.c
@@ -2576,18 +2576,9 @@ void hfi1_kern_read_tid_flow_free(struct
hfi1_kern_clear_hw_flow(priv->rcd, qp);
}

-static bool tid_rdma_tid_err(struct hfi1_ctxtdata *rcd,
- struct hfi1_packet *packet, u8 rcv_type,
- u8 opcode)
+static bool tid_rdma_tid_err(struct hfi1_packet *packet, u8 rcv_type)
{
struct rvt_qp *qp = packet->qp;
- struct hfi1_qp_priv *qpriv = qp->priv;
- u32 ipsn;
- struct ib_other_headers *ohdr = packet->ohdr;
- struct rvt_ack_entry *e;
- struct tid_rdma_request *req;
- struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
- u32 i;

if (rcv_type >= RHF_RCV_TYPE_IB)
goto done;
@@ -2604,41 +2595,9 @@ static bool tid_rdma_tid_err(struct hfi1
if (rcv_type == RHF_RCV_TYPE_EAGER) {
hfi1_restart_rc(qp, qp->s_last_psn + 1, 1);
hfi1_schedule_send(qp);
- goto done_unlock;
- }
-
- /*
- * For TID READ response, error out QP after freeing the tid
- * resources.
- */
- if (opcode == TID_OP(READ_RESP)) {
- ipsn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn));
- if (cmp_psn(ipsn, qp->s_last_psn) > 0 &&
- cmp_psn(ipsn, qp->s_psn) < 0) {
- hfi1_kern_read_tid_flow_free(qp);
- spin_unlock(&qp->s_lock);
- rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
- goto done;
- }
- goto done_unlock;
}

- /*
- * Error out the qp for TID RDMA WRITE
- */
- hfi1_kern_clear_hw_flow(qpriv->rcd, qp);
- for (i = 0; i < rvt_max_atomic(rdi); i++) {
- e = &qp->s_ack_queue[i];
- if (e->opcode == TID_OP(WRITE_REQ)) {
- req = ack_to_tid_req(e);
- hfi1_kern_exp_rcv_clear_all(req);
- }
- }
- spin_unlock(&qp->s_lock);
- rvt_rc_error(qp, IB_WC_LOC_LEN_ERR);
- goto done;
-
-done_unlock:
+ /* Since no payload is delivered, just drop the packet */
spin_unlock(&qp->s_lock);
done:
return true;
@@ -2927,7 +2886,7 @@ bool hfi1_handle_kdeth_eflags(struct hfi
if (lnh == HFI1_LRH_GRH)
goto r_unlock;

- if (tid_rdma_tid_err(rcd, packet, rcv_type, opcode))
+ if (tid_rdma_tid_err(packet, rcv_type))
goto r_unlock;
}