iw_cxgb4: Allocate and use IQs specifically for indirect interrupts
Currently indirect interrupts for RDMA CQs funnel through the LLD's RDMA
RXQs, which also handle direct interrupts for offload CPLs during RDMA
connection setup/teardown. The intended T4 usage model, however, is to
have indirect interrupts flow through dedicated IQs. IE not to mix
indirect interrupts with CPL messages in an IQ. This patch adds the
concept of RDMA concentrator IQs, or CIQs, setup and maintained by the
LLD and exported to iw_cxgb4 for use when creating CQs. RDMA CPLs will
flow through the LLD's RDMA RXQs, and CQ interrupts flow through the
CIQs.
Design:
cxgb4 creates and exports an array of CIQs for the RDMA ULD. These IQs
are sized according to the max available CQs available at adapter init.
In addition, these IQs don't need FL buffers since they only service
indirect interrupts. One CIQ is setup per RX channel similar to the
RDMA RXQs.
iw_cxgb4 will utilize these CIQs based on the vector value passed into
create_cq(). The num_comp_vectors advertised by iw_cxgb4 will be the
number of CIQs configured, and thus the vector value will be the index
into the array of CIQs.
Based on original work by Steve Wise <swise@opengridcomputing.com>
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>
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 8cf6be9..c26c3f8 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -818,12 +818,17 @@
for_each_rdmarxq(&adap->sge, i)
snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
adap->port[0]->name, i);
+
+ for_each_rdmaciq(&adap->sge, i)
+ snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma-ciq%d",
+ adap->port[0]->name, i);
}
static int request_msix_queue_irqs(struct adapter *adap)
{
struct sge *s = &adap->sge;
- int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
+ int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, rdmaciqqidx = 0;
+ int msi_index = 2;
err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
adap->msix_info[1].desc, &s->fw_evtq);
@@ -857,9 +862,21 @@
goto unwind;
msi_index++;
}
+ for_each_rdmaciq(s, rdmaciqqidx) {
+ err = request_irq(adap->msix_info[msi_index].vec,
+ t4_sge_intr_msix, 0,
+ adap->msix_info[msi_index].desc,
+ &s->rdmaciq[rdmaciqqidx].rspq);
+ if (err)
+ goto unwind;
+ msi_index++;
+ }
return 0;
unwind:
+ while (--rdmaciqqidx >= 0)
+ free_irq(adap->msix_info[--msi_index].vec,
+ &s->rdmaciq[rdmaciqqidx].rspq);
while (--rdmaqidx >= 0)
free_irq(adap->msix_info[--msi_index].vec,
&s->rdmarxq[rdmaqidx].rspq);
@@ -885,6 +902,8 @@
free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
for_each_rdmarxq(s, i)
free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
+ for_each_rdmaciq(s, i)
+ free_irq(adap->msix_info[msi_index++].vec, &s->rdmaciq[i].rspq);
}
/**
@@ -1047,7 +1066,8 @@
if (msi_idx > 0)
msi_idx++;
err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
- &q->fl, uldrx_handler);
+ q->fl.size ? &q->fl : NULL,
+ uldrx_handler);
if (err)
goto freeout;
memset(&q->stats, 0, sizeof(q->stats));
@@ -1064,13 +1084,28 @@
if (msi_idx > 0)
msi_idx++;
err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
- msi_idx, &q->fl, uldrx_handler);
+ msi_idx, q->fl.size ? &q->fl : NULL,
+ uldrx_handler);
if (err)
goto freeout;
memset(&q->stats, 0, sizeof(q->stats));
s->rdma_rxq[i] = q->rspq.abs_id;
}
+ for_each_rdmaciq(s, i) {
+ struct sge_ofld_rxq *q = &s->rdmaciq[i];
+
+ if (msi_idx > 0)
+ msi_idx++;
+ err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
+ msi_idx, q->fl.size ? &q->fl : NULL,
+ uldrx_handler);
+ if (err)
+ goto freeout;
+ memset(&q->stats, 0, sizeof(q->stats));
+ s->rdma_ciq[i] = q->rspq.abs_id;
+ }
+
for_each_port(adap, i) {
/*
* Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
@@ -3789,7 +3824,9 @@
lli.mtus = adap->params.mtus;
if (uld == CXGB4_ULD_RDMA) {
lli.rxq_ids = adap->sge.rdma_rxq;
+ lli.ciq_ids = adap->sge.rdma_ciq;
lli.nrxq = adap->sge.rdmaqs;
+ lli.nciq = adap->sge.rdmaciqs;
} else if (uld == CXGB4_ULD_ISCSI) {
lli.rxq_ids = adap->sge.ofld_rxq;
lli.nrxq = adap->sge.ofldqsets;
@@ -5695,6 +5732,7 @@
{
struct sge *s = &adap->sge;
int i, q10g = 0, n10g = 0, qidx = 0;
+ int ciq_size;
for_each_port(adap, i)
n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
@@ -5733,6 +5771,7 @@
s->ofldqsets = adap->params.nports;
/* For RDMA one Rx queue per channel suffices */
s->rdmaqs = adap->params.nports;
+ s->rdmaciqs = adap->params.nports;
}
for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
@@ -5767,6 +5806,19 @@
r->fl.size = 72;
}
+ ciq_size = 64 + adap->vres.cq.size + adap->tids.nftids;
+ if (ciq_size > SGE_MAX_IQ_SIZE) {
+ CH_WARN(adap, "CIQ size too small for available IQs\n");
+ ciq_size = SGE_MAX_IQ_SIZE;
+ }
+
+ for (i = 0; i < ARRAY_SIZE(s->rdmaciq); i++) {
+ struct sge_ofld_rxq *r = &s->rdmaciq[i];
+
+ init_rspq(&r->rspq, 0, 0, ciq_size, 64);
+ r->rspq.uld = CXGB4_ULD_RDMA;
+ }
+
init_rspq(&s->fw_evtq, 6, 0, 512, 64);
init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
}
@@ -5815,9 +5867,9 @@
want = s->max_ethqsets + EXTRA_VECS;
if (is_offload(adap)) {
- want += s->rdmaqs + s->ofldqsets;
+ want += s->rdmaqs + s->rdmaciqs + s->ofldqsets;
/* need nchan for each possible ULD */
- ofld_need = 2 * nchan;
+ ofld_need = 3 * nchan;
}
need = adap->params.nports + EXTRA_VECS + ofld_need;