To optimize NVME-oF READ IOPs, use a specialized WQE that combines
the RDMA WRITE and SEND_INV WR chain submitted by the NVME-oF target
driver.
This reduces uP overhead per NVME-oF IO, and results in over 10%
improvement in NVME-oF 4K READ IOPs.
Signed-off-by: Potnuri Bharat Teja <bharat@xxxxxxxxxxx>
Signed-off-by: Steve Wise <swise@xxxxxxxxxxxxxxxxxxxxx>
---
drivers/infiniband/hw/cxgb4/qp.c | 147 +++++++++++++++++++++++++++++-
drivers/infiniband/hw/cxgb4/t4.h | 4 +
drivers/infiniband/hw/cxgb4/t4fw_ri_api.h | 31 +++++++
3 files changed, 181 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index dbfa450771fc..622c433181b1 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -456,7 +456,12 @@ static int build_isgl(__be64 *queue_start, __be64 *queue_end,
{
int i;
u32 plen = 0;
- __be64 *flitp = (__be64 *)isglp->sge;
+ __be64 *flitp;
+
+ if ((__be64 *)isglp == queue_end)
+ isglp = (struct fw_ri_isgl *)queue_start;
+
+ flitp = (__be64 *)isglp->sge;
for (i = 0; i < num_sge; i++) {
if ((plen + sg_list[i].length) < plen)
@@ -593,6 +598,56 @@ static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe,
return 0;
}
+static void build_immd_cmpl(struct t4_sq *sq, struct fw_ri_immd_cmpl *immdp,
+ struct ib_send_wr *wr)
+{
+ memcpy((u8 *)immdp->data, (u8 *)(uintptr_t)wr->sg_list->addr, 16);
+ memset(immdp->r1, 0, 6);
+ immdp->op = FW_RI_DATA_IMMD;
+ immdp->immdlen = 16;
+}
+
+static void build_rdma_write_cmpl(struct t4_sq *sq,
+ struct fw_ri_rdma_write_cmpl_wr *wcwr,
+ struct ib_send_wr *wr, u8 *len16)
+{
+ u32 plen;
+ int size;
+
+ /*
+ * This code assumes the struct fields preceding the write isgl
+ * fit in one 64B WR slot. This is because the WQE is built
+ * directly in the dma queue, and wrapping is only handled
+ * by the code buildling sgls. IE the "fixed part" of the wr
+ * structs must all fit in 64B. The WQE build code should probably be
+ * redesigned to avoid this restriction, but for now just add
+ * the BUILD_BUG_ON() to catch if this WQE struct gets too big.
+ */
+ BUILD_BUG_ON(offsetof(struct fw_ri_rdma_write_cmpl_wr, u) > 64);
+
+ wcwr->stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
+ wcwr->to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
+ wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey);
+ wcwr->r2 = 0;
+ wcwr->r3 = 0;
+
+ /* SEND_INV SGL */
+ if (wr->next->send_flags & IB_SEND_INLINE)
+ build_immd_cmpl(sq, &wcwr->u_cmpl.immd_src, wr->next);
+ else
+ build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size],
+ &wcwr->u_cmpl.isgl_src, wr->next->sg_list, 1, NULL);
+
+ /* WRITE SGL */
+ build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size],
+ wcwr->u.isgl_src, wr->sg_list, wr->num_sge, &plen);
+
+ size = sizeof(*wcwr) + sizeof(struct fw_ri_isgl) +
+ wr->num_sge * sizeof(struct fw_ri_sge);
+ wcwr->plen = cpu_to_be32(plen);
+ *len16 = DIV_ROUND_UP(size, 16);
+}
+
static int build_rdma_read(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16)
{
if (wr->num_sge > 1)
@@ -622,6 +677,72 @@ static int build_rdma_read(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16)
return 0;
}
+static void post_write_cmpl(struct c4iw_qp *qhp, struct ib_send_wr *wr)
+{
+ bool send_signaled = (wr->next->send_flags & IB_SEND_SIGNALED) ||
+ qhp->sq_sig_all;
+ bool write_signaled = (wr->send_flags & IB_SEND_SIGNALED) ||
+ qhp->sq_sig_all;
+ struct t4_swsqe *swsqe;
+ union t4_wr *wqe;
+ u16 write_wrid;
+ u8 len16;
+ u16 idx;
+
+ /*
+ * The sw_sq entries still look like a WRITE and a SEND and consume
+ * 2 slots. The FW WR, however, will be a single uber-WR.
+ */
+ wqe = (union t4_wr *)((u8 *)qhp->wq.sq.queue +
+ qhp->wq.sq.wq_pidx * T4_EQ_ENTRY_SIZE);
+ build_rdma_write_cmpl(&qhp->wq.sq, &wqe->write_cmpl, wr, &len16);
+
+ /* WRITE swsqe */
+ swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
+ swsqe->opcode = FW_RI_RDMA_WRITE;
+ swsqe->idx = qhp->wq.sq.pidx;
+ swsqe->complete = 0;
+ swsqe->signaled = write_signaled;
+ swsqe->flushed = 0;
+ swsqe->wr_id = wr->wr_id;
+ if (c4iw_wr_log) {
+ swsqe->sge_ts =
+ cxgb4_read_sge_timestamp(qhp->rhp->rdev.lldi.ports[0]);
+ swsqe->host_time = ktime_get();
+ }
+
+ write_wrid = qhp->wq.sq.pidx;
+
+ /* just bump the sw_sq */
+ qhp->wq.sq.in_use++;
+ if (++qhp->wq.sq.pidx == qhp->wq.sq.size)
+ qhp->wq.sq.pidx = 0;
+
+ /* SEND_WITH_INV swsqe */
+ swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
+ swsqe->opcode = FW_RI_SEND_WITH_INV;
+ swsqe->idx = qhp->wq.sq.pidx;
+ swsqe->complete = 0;
+ swsqe->signaled = send_signaled;
+ swsqe->flushed = 0;
+ swsqe->wr_id = wr->next->wr_id;
+ if (c4iw_wr_log) {
+ swsqe->sge_ts =
+ cxgb4_read_sge_timestamp(qhp->rhp->rdev.lldi.ports[0]);
+ swsqe->host_time = ktime_get();
+ }
+
+ wqe->write_cmpl.flags_send = send_signaled ? FW_RI_COMPLETION_FLAG : 0;
+ wqe->write_cmpl.wrid_send = qhp->wq.sq.pidx;
+
+ init_wr_hdr(wqe, write_wrid, FW_RI_RDMA_WRITE_CMPL_WR,
+ write_signaled ? FW_RI_COMPLETION_FLAG : 0, len16);
+ t4_sq_produce(&qhp->wq, len16);
+ idx = DIV_ROUND_UP(len16 * 16, T4_EQ_ENTRY_SIZE);
+
+ t4_ring_sq_db(&qhp->wq, idx, wqe);
+}
+
static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe,
struct ib_recv_wr *wr, u8 *len16)
{
@@ -997,6 +1118,30 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
*bad_wr = wr;
return -ENOMEM;
}
+
+ /*
+ * Fastpath for NVMe-oF target WRITE + SEND_WITH_INV wr chain which is
+ * the response for small NVMEe-oF READ requests. If the chain is
+ * exactly a WRITE->SEND_WITH_INV and the sgl depths and lengths
+ * meet the requirements of the fw_ri_write_cmpl_wr work request,
+ * then build and post the write_cmpl WR. If any of the tests
+ * below are not true, then we continue on with the tradtional WRITE
+ * and SEND WRs.
+ */
+ if (qhp->rhp->rdev.lldi.write_cmpl_support &&
+ CHELSIO_CHIP_VERSION(qhp->rhp->rdev.lldi.adapter_type) >=
+ CHELSIO_T5 &&
+ wr && wr->next && !wr->next->next &&
+ wr->opcode == IB_WR_RDMA_WRITE &&
+ wr->sg_list[0].length && wr->num_sge <= T4_WRITE_CMPL_MAX_SGL &&
+ wr->next->opcode == IB_WR_SEND_WITH_INV &&
+ wr->next->sg_list[0].length == T4_WRITE_CMPL_MAX_CQE &&
+ wr->next->num_sge == 1 && num_wrs >= 2) {
+ post_write_cmpl(qhp, wr);
+ spin_unlock_irqrestore(&qhp->lock, flag);
+ return 0;
+ }
+
while (wr) {
if (num_wrs == 0) {
err = -ENOMEM;
diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h
index 5fff9a685ca8..2027e4c2342d 100644
--- a/drivers/infiniband/hw/cxgb4/t4.h
+++ b/drivers/infiniband/hw/cxgb4/t4.h
@@ -91,6 +91,9 @@ static inline int t4_max_fr_depth(int use_dsgl)
#define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)
#define T4_MAX_RECV_SGE 4
+#define T4_WRITE_CMPL_MAX_SGL 4
+#define T4_WRITE_CMPL_MAX_CQE 16
+
union t4_wr {
struct fw_ri_res_wr res;
struct fw_ri_wr ri;
@@ -101,6 +104,7 @@ union t4_wr {
struct fw_ri_fr_nsmr_wr fr;
struct fw_ri_fr_nsmr_tpte_wr fr_tpte;
struct fw_ri_inv_lstag_wr inv;
+ struct fw_ri_rdma_write_cmpl_wr write_cmpl;
struct t4_status_page status;
__be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_SQ_NUM_SLOTS];
};
diff --git a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
index 48c187690921..5eb97dfc9284 100644
--- a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
+++ b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
@@ -585,6 +585,37 @@ struct fw_ri_send_wr {
#define FW_RI_SEND_WR_SENDOP_G(x) \
(((x) >> FW_RI_SEND_WR_SENDOP_S) & FW_RI_SEND_WR_SENDOP_M)
+struct fw_ri_rdma_write_cmpl_wr {
+ __u8 opcode;
+ __u8 flags;
+ __u16 wrid;
+ __u8 r1[3];
+ __u8 len16;
+ __u8 r2;
+ __u8 flags_send;
+ __u16 wrid_send;
+ __be32 stag_inv;
+ __be32 plen;
+ __be32 stag_sink;
+ __be64 to_sink;
+ union fw_ri_cmpl {
+ struct fw_ri_immd_cmpl {
+ __u8 op;
+ __u8 r1[6];
+ __u8 immdlen;
+ __u8 data[16];
+ } immd_src;
+ struct fw_ri_isgl isgl_src;
+ } u_cmpl;
+ __be64 r3;
+#ifndef C99_NOT_SUPPORTED
+ union fw_ri_write {
+ struct fw_ri_immd immd_src[0];
+ struct fw_ri_isgl isgl_src[0];
+ } u;
+#endif
+};
+
struct fw_ri_rdma_read_wr {
__u8 opcode;
__u8 flags;
--
2.9.3
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
