diff options
author | Hariprasad Shenai <hariprasad@chelsio.com> | 2014-07-14 21:34:52 +0530 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2014-07-15 16:25:16 -0700 |
commit | 4c2c5763227a14ce111d6f35df708459d2443cc3 (patch) | |
tree | b261cf8eab4279d5a4162b7417e10162eeb2d755 /drivers/infiniband/hw/cxgb4/qp.c | |
parent | 04e10e2164fcfa05e14eff3c2757a5097f11d258 (diff) | |
download | talos-op-linux-4c2c5763227a14ce111d6f35df708459d2443cc3.tar.gz talos-op-linux-4c2c5763227a14ce111d6f35df708459d2443cc3.zip |
cxgb4/iw_cxgb4: use firmware ord/ird resource limits
Advertise a larger max read queue depth for qps, and gather the resource limits
from fw and use them to avoid exhaustinq all the resources.
Design:
cxgb4:
Obtain the max_ordird_qp and max_ird_adapter device params from FW
at init time and pass them up to the ULDs when they attach. If these
parameters are not available, due to older firmware, then hard-code
the values based on the known values for older firmware.
iw_cxgb4:
Fix the c4iw_query_device() to report these correct values based on
adapter parameters. ibv_query_device() will always return:
max_qp_rd_atom = max_qp_init_rd_atom = min(module_max, max_ordird_qp)
max_res_rd_atom = max_ird_adapter
Bump up the per qp max module option to 32, allowing it to be increased
by the user up to the device max of max_ordird_qp. 32 seems to be
sufficient to maximize throughput for streaming read benchmarks.
Fail connection setup if the negotiated IRD exhausts the available
adapter ird resources. So the driver will track the amount of ird
resource in use and not send an RI_WR/INIT to FW that would reduce the
available ird resources below zero.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'drivers/infiniband/hw/cxgb4/qp.c')
-rw-r--r-- | drivers/infiniband/hw/cxgb4/qp.c | 54 |
1 files changed, 46 insertions, 8 deletions
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index 6f74e0e9a022..0de3cf64eb5e 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -58,6 +58,31 @@ static int max_fr_immd = T4_MAX_FR_IMMD; module_param(max_fr_immd, int, 0644); MODULE_PARM_DESC(max_fr_immd, "fastreg threshold for using DSGL instead of immedate"); +static int alloc_ird(struct c4iw_dev *dev, u32 ird) +{ + int ret = 0; + + spin_lock_irq(&dev->lock); + if (ird <= dev->avail_ird) + dev->avail_ird -= ird; + else + ret = -ENOMEM; + spin_unlock_irq(&dev->lock); + + if (ret) + dev_warn(&dev->rdev.lldi.pdev->dev, + "device IRD resources exhausted\n"); + + return ret; +} + +static void free_ird(struct c4iw_dev *dev, int ird) +{ + spin_lock_irq(&dev->lock); + dev->avail_ird += ird; + spin_unlock_irq(&dev->lock); +} + static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state) { unsigned long flag; @@ -1204,12 +1229,20 @@ static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp) int ret; struct sk_buff *skb; - PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid, - qhp->ep->hwtid); + PDBG("%s qhp %p qid 0x%x tid %u ird %u ord %u\n", __func__, qhp, + qhp->wq.sq.qid, qhp->ep->hwtid, qhp->ep->ird, qhp->ep->ord); skb = alloc_skb(sizeof *wqe, GFP_KERNEL); - if (!skb) - return -ENOMEM; + if (!skb) { + ret = -ENOMEM; + goto out; + } + ret = alloc_ird(rhp, qhp->attr.max_ird); + if (ret) { + qhp->attr.max_ird = 0; + kfree_skb(skb); + goto out; + } set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx); wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe)); @@ -1260,10 +1293,14 @@ static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp) ret = c4iw_ofld_send(&rhp->rdev, skb); if (ret) - goto out; + goto err1; ret = c4iw_wait_for_reply(&rhp->rdev, &qhp->ep->com.wr_wait, qhp->ep->hwtid, qhp->wq.sq.qid, __func__); + if (!ret) + goto out; +err1: + free_ird(rhp, qhp->attr.max_ird); out: PDBG("%s ret %d\n", __func__, ret); return ret; @@ -1308,7 +1345,7 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp, newattr.max_ord = attrs->max_ord; } if (mask & C4IW_QP_ATTR_MAX_IRD) { - if (attrs->max_ird > c4iw_max_read_depth) { + if (attrs->max_ird > cur_max_read_depth(rhp)) { ret = -EINVAL; goto out; } @@ -1531,6 +1568,7 @@ int c4iw_destroy_qp(struct ib_qp *ib_qp) if (!list_empty(&qhp->db_fc_entry)) list_del_init(&qhp->db_fc_entry); spin_unlock_irq(&rhp->lock); + free_ird(rhp, qhp->attr.max_ird); ucontext = ib_qp->uobject ? to_c4iw_ucontext(ib_qp->uobject->context) : NULL; @@ -1621,8 +1659,8 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, qhp->attr.enable_rdma_read = 1; qhp->attr.enable_rdma_write = 1; qhp->attr.enable_bind = 1; - qhp->attr.max_ord = 1; - qhp->attr.max_ird = 1; + qhp->attr.max_ord = 0; + qhp->attr.max_ird = 0; qhp->sq_sig_all = attrs->sq_sig_type == IB_SIGNAL_ALL_WR; spin_lock_init(&qhp->lock); mutex_init(&qhp->mutex); |