d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ibcore: Fix sleeping in atomic when RoCE is used

Browse Source 
A couple of places in the CM do

    spin_lock_irq(&cm_id_priv->lock);
    ...
    if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))

However when the underlying transport is RoCE, this leads to a sleeping function
being called with the lock held - the callchain is

    cm_alloc_response_msg() ->
      ib_create_ah_from_wc() ->
        ib_init_ah_from_wc() ->
          rdma_addr_find_l2_eth_by_grh() ->
            rdma_resolve_ip()

and rdma_resolve_ip() starts out by doing

    req = kzalloc(sizeof *req, GFP_KERNEL);

not to mention rdma_addr_find_l2_eth_by_grh() doing

    wait_for_completion(&ctx.comp);

to wait for the task that rdma_resolve_ip() queues up.

Fix this by moving the AH creation out of the lock.

Linux commit:
c76161181193985087cd716fdf69b5cb6cf9ee85

Submitted by:   hselasky@
Approved by:    hselasky (mentor)
MFC after:      1 week
Sponsored by:   Mellanox Technologies
This commit is contained in:
slavash 2018-12-05 13:24:12 +00:00
parent cc6a289dfb
commit ed5f1f49e7
1 changed files with 44 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

63
sys/ofed/drivers/infiniband/core/ib_cm.c
View File

@ -332,11 +332,19 @@ out:
return ret;
}
static int cm_alloc_response_msg(struct cm_port *port,
struct ib_mad_recv_wc *mad_recv_wc,
struct ib_mad_send_buf **msg)
static struct ib_mad_send_buf *cm_alloc_response_msg_no_ah(struct cm_port *port,
struct ib_mad_recv_wc *mad_recv_wc)
{
return ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index,
0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
GFP_ATOMIC,
IB_MGMT_BASE_VERSION);
}
static int cm_create_response_msg_ah(struct cm_port *port,
struct ib_mad_recv_wc *mad_recv_wc,
struct ib_mad_send_buf *msg)
{
struct ib_mad_send_buf *m;
struct ib_ah *ah;
ah = ib_create_ah_from_wc(port->mad_agent->qp->pd, mad_recv_wc->wc,
@ -344,27 +352,40 @@ static int cm_alloc_response_msg(struct cm_port *port,
if (IS_ERR(ah))
return PTR_ERR(ah);
m = ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index,
0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
GFP_ATOMIC,
IB_MGMT_BASE_VERSION);
if (IS_ERR(m)) {
ib_destroy_ah(ah);
return PTR_ERR(m);
}
m->ah = ah;
*msg = m;
msg->ah = ah;
return 0;
}
static void cm_free_msg(struct ib_mad_send_buf *msg)
{
ib_destroy_ah(msg->ah);
if (msg->ah)
ib_destroy_ah(msg->ah);
if (msg->context[0])
cm_deref_id(msg->context[0]);
ib_free_send_mad(msg);
}
static int cm_alloc_response_msg(struct cm_port *port,
struct ib_mad_recv_wc *mad_recv_wc,
struct ib_mad_send_buf **msg)
{
struct ib_mad_send_buf *m;
int ret;
m = cm_alloc_response_msg_no_ah(port, mad_recv_wc);
if (IS_ERR(m))
return PTR_ERR(m);
ret = cm_create_response_msg_ah(port, mad_recv_wc, m);
if (ret) {
cm_free_msg(m);
return ret;
}
*msg = m;
return 0;
}
static void * cm_copy_private_data(const void *private_data,
u8 private_data_len)
{
@ -2331,7 +2352,8 @@ static int cm_dreq_handler(struct cm_work *work)
case IB_CM_TIMEWAIT:
atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
counter[CM_DREQ_COUNTER]);
if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
msg = cm_alloc_response_msg_no_ah(work->port, work->mad_recv_wc);
if (IS_ERR(msg))
goto unlock;
cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
@ -2339,7 +2361,8 @@ static int cm_dreq_handler(struct cm_work *work)
cm_id_priv->private_data_len);
spin_unlock_irq(&cm_id_priv->lock);
if (ib_post_send_mad(msg, NULL))
if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) ||
ib_post_send_mad(msg, NULL))
cm_free_msg(msg);
goto deref;
case IB_CM_DREQ_RCVD:
@ -2884,7 +2907,8 @@ static int cm_lap_handler(struct cm_work *work)
case IB_CM_MRA_LAP_SENT:
atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
counter[CM_LAP_COUNTER]);
if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
msg = cm_alloc_response_msg_no_ah(work->port, work->mad_recv_wc);
if (IS_ERR(msg))
goto unlock;
cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
@ -2894,7 +2918,8 @@ static int cm_lap_handler(struct cm_work *work)
cm_id_priv->private_data_len);
spin_unlock_irq(&cm_id_priv->lock);
if (ib_post_send_mad(msg, NULL))
if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) ||
ib_post_send_mad(msg, NULL))
cm_free_msg(msg);
goto deref;
case IB_CM_LAP_RCVD: