Support UD on hip08 #587
Support UD on hip08 #587
Conversation
providers/hns/hns_roce_u.h
Outdated
| @@ -68,6 +68,8 @@ | |||
| #define HNS_ROCE_TPTR_OFFSET 0x1000 | |||
| #define HNS_ROCE_STATIC_RATE 3 /* Gbps */ | |||
|
|
|||
| #define ETH_ALEN 6 | |||
There was a problem hiding this comment.
It is declared in "linux/if_ether.h"
providers/hns/hns_roce_u_hw_v1.c
Outdated
|
|
||
| wqe_idx = (qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1); | ||
|
|
There was a problem hiding this comment.
I didn't understand your problem and solution, but a couple of lines above you have hns_roce_wq_overflow() function which supposes to protect from overflow.
There was a problem hiding this comment.
plus the "a %= b" construction is the better way to write it.
There was a problem hiding this comment.
I didn't understand your problem and solution, but a couple of lines above you have hns_roce_wq_overflow() function which supposes to protect from overflow.
Hi, Leon, thanks. You are right, the commit message is uncorrect, I will rewrite it.
There was a problem hiding this comment.
plus the "a %= b" construction is the better way to write it.
Hi Leon, do you mean:
wqe_idx = (qp->sq.head + nreq) % qp->sq.wqe_cnt;
providers/hns/hns_roce_u_hw_v2.c
Outdated
| switch (wr->opcode) { | ||
| case IBV_WR_SEND: | ||
| hr_op = HNS_ROCE_WQE_OP_SEND; | ||
| ud_sq_wqe->immtdata = 0; |
There was a problem hiding this comment.
OK, will remove it, thank you.
providers/hns/hns_roce_u_hw_v2.c
Outdated
| sge_info->total_len = 0; | ||
|
|
||
| for (i = 0; i < wr->num_sge; i++) | ||
| if (likely(wr->sg_list[i].length)) { |
There was a problem hiding this comment.
success-oriented flow, please
if (unlikely(,,,))
continue;
..
There was a problem hiding this comment.
OK, thank you for your advice :)
providers/hns/hns_roce_u_hw_v2.c
Outdated
|
|
||
| if (wr->send_flags & IBV_SEND_INLINE) { | ||
| if (wr->opcode == IBV_WR_RDMA_READ) { | ||
| fprintf(stderr, "Not supported inline data!\n"); |
There was a problem hiding this comment.
No fprints in libraries
Hi Leon,
Do you means that we should print informations only when debug in libraries? Or it's not reasonable to print message here in inline mode?
I saw some fprintf in other drivers, should I replace all printf/fprintf in hns with a macro that only show message when debug?
providers/hns/hns_roce_u_hw_v2.c
Outdated
|
|
||
| if (sge_info->total_len > qp->max_inline_data) { | ||
| fprintf(stderr, | ||
| "Failed to inline, data len=%d, max inline len=%d!\n", |
There was a problem hiding this comment.
ditto
Hi Leon,
I have modified code as suggested and pushed a new version, could you please review it again? Thank you :)
Li-Weihang
force-pushed
the
ud_support
branch
2 times, most recently
from
e347970
to
e193d07
Compare
This patch achieves two verbs create_ah and destroy_ah to support allocation and destruction of Address Handle. Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Currently, the wqe idx is calculated repeatly everywhere it is used. This patch defines wqe_idx and calculated it only once, then just use it as needed. Signed-off-by: Yixian Liu <liuyixian@huawei.com> Signed-off-by: Weihang Li <liweihang@hisilicon.com>
This patch refactors the interface of hns_roce_u_v2_post_send, which is now very complicated. We reduce the complexity with following points: 1. Separate RC server into a function. 2. Simplify and separate the process of sge. 3. Keep the logic and consistence of all operations. Signed-off-by: Yixian Liu <liuyixian@huawei.com> Signed-off-by: Weihang Li <liweihang@hisilicon.com>
User Application can use Unreliable Datagram on hip08 now. Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Li-Weihang
force-pushed
the
ud_support
branch
from
e193d07
to
90c68be
Compare
This patchset enables hip08 to support Unreliable Datagram.