> Hey, I appreciate it, but I really do have to warn you that I did this all
> blind, and just meant for it to be a "I think this kind of direction is
> more productive" thing. I'm not going to guarantee that it works at all.
Oh, understood, and I'm definitely planning on taking your patch as
inspiration and a kick in the pants to finally clean up the code
rather than trying to use it as is. Obviously it was a quick hack
from someone that doesn't know the hardware -- for example
handle_raddr_seg() is at least misnamed, since UD work requests don't
have raddr segments.
But if you're going to take the time to work on the code, then I feel
like it's only polite not to drop your work on the floor. Just like I
hope someone who works on IB and who is usually sending me stuff to
merge would give a patch from me some attention.
BTW, I noticed one interesting thing while starting on this cleanup.
I wanted to make sure that the generated code didn't change with the
first step, and I actually discovered that the patch below seems to
make the generated code *better*, maybe because some gcc alias
analysis doesn't get as paranoid without the void *:
add/remove: 0/0 grow/shrink: 0/7 up/down: 0/-85 (-85)
function old new delta
mthca_arbel_post_srq_recv 373 369 -4
mthca_arbel_post_receive 570 562 -8
mthca_tavor_post_srq_recv 520 508 -12
mthca_tavor_post_send 1344 1330 -14
mthca_arbel_post_send 1481 1467 -14
mlx4_ib_post_send 1598 1582 -16
mthca_tavor_post_receive 792 775 -17
And here's the patch itself -- I think this is a reasonable size step
to break things up into. I assume that this is the basic form of the
cleanup that you're proposing?
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 8d09aa3..a59c7f0 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1183,6 +1183,14 @@ static int mlx4_wq_overflow(struct mlx4_ib_wq *wq, int nreq, struct ib_cq *ib_cq
return cur + nreq >= wq->max_post;
}
+static void set_data_seg(struct mlx4_wqe_data_seg *dseg,
+ struct ib_sge *sg)
+{
+ dseg->byte_count = cpu_to_be32(sg->length);
+ dseg->lkey = cpu_to_be32(sg->lkey);
+ dseg->addr = cpu_to_be64(sg->addr);
+}
+
int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
struct ib_send_wr **bad_wr)
{
@@ -1313,12 +1321,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mlx4_wqe_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mlx4_wqe_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mlx4_wqe_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
+ set_data_seg(wqe, wr->sg_list + i);
wqe += sizeof (struct mlx4_wqe_data_seg);
size += sizeof (struct mlx4_wqe_data_seg) / 16;
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index 0e9ef24..2548250 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1740,13 +1740,8 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
@@ -1869,13 +1864,8 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
@@ -2125,13 +2115,8 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
@@ -2253,20 +2238,12 @@ int mthca_arbel_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
wqe += sizeof (struct mthca_data_seg);
}
- if (i < qp->rq.max_gs) {
- ((struct mthca_data_seg *) wqe)->byte_count = 0;
- ((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
- ((struct mthca_data_seg *) wqe)->addr = 0;
- }
+ if (i < qp->rq.max_gs)
+ mthca_set_data_seg_inval(wqe);
qp->wrid[ind] = wr->wr_id;
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index b8f05a5..88d219e 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -543,20 +543,12 @@ int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
wqe += sizeof (struct mthca_data_seg);
}
- if (i < srq->max_gs) {
- ((struct mthca_data_seg *) wqe)->byte_count = 0;
- ((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
- ((struct mthca_data_seg *) wqe)->addr = 0;
- }
+ if (i < srq->max_gs)
+ mthca_set_data_seg_inval(wqe);
((struct mthca_next_seg *) prev_wqe)->nda_op =
cpu_to_be32((ind << srq->wqe_shift) | 1);
@@ -662,20 +654,12 @@ int mthca_arbel_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
wqe += sizeof (struct mthca_data_seg);
}
- if (i < srq->max_gs) {
- ((struct mthca_data_seg *) wqe)->byte_count = 0;
- ((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
- ((struct mthca_data_seg *) wqe)->addr = 0;
- }
+ if (i < srq->max_gs)
+ mthca_set_data_seg_inval(wqe);
srq->wrid[ind] = wr->wr_id;
srq->first_free = next_ind;
diff --git a/drivers/infiniband/hw/mthca/mthca_wqe.h b/drivers/infiniband/hw/mthca/mthca_wqe.h
index e7d2c1e..f6a66fe 100644
--- a/drivers/infiniband/hw/mthca/mthca_wqe.h
+++ b/drivers/infiniband/hw/mthca/mthca_wqe.h
@@ -113,4 +113,19 @@ struct mthca_mlx_seg {
__be16 vcrc;
};
+static __always_inline void mthca_set_data_seg(struct mthca_data_seg *dseg,
+ struct ib_sge *sg)
+{
+ dseg->byte_count = cpu_to_be32(sg->length);
+ dseg->lkey = cpu_to_be32(sg->lkey);
+ dseg->addr = cpu_to_be64(sg->addr);
+}
+
+static __always_inline void mthca_set_data_seg_inval(struct mthca_data_seg *dseg)
+{
+ dseg->byte_count = 0;
+ dseg->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
+ dseg->addr = 0;
+}
+
#endif /* MTHCA_WQE_H */
-
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
