d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

libcxgb4: Use memcpy instead of copying WRs 8B at a time in the userspace

Browse Source 
RDMA library for cxgbe(4).

MFC after:	3 days
Sponsored by:	Chelsio Communications
This commit is contained in:
Navdeep Parhar 2017-05-27 02:05:21 +00:00
parent 41cf0d54a2
commit 7de5a71e79
1 changed files with 32 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
contrib/ofed/libcxgb4/src/qp.c
View File

@ -47,39 +47,49 @@ struct c4iw_stats c4iw_stats;
static void copy_wr_to_sq(struct t4_wq *wq, union t4_wr *wqe, u8 len16) static void copy_wr_to_sq(struct t4_wq *wq, union t4_wr *wqe, u8 len16)
{ {
u64 *src, *dst; void *src, *dst;
uintptr_t end;
int total, len;
src = (u64 *)wqe; src = &wqe->flits[0];
dst = (u64 *)((u8 *)wq->sq.queue + wq->sq.wq_pidx * T4_EQ_ENTRY_SIZE); dst = &wq->sq.queue->flits[wq->sq.wq_pidx *
(T4_EQ_ENTRY_SIZE / sizeof(__be64))];
if (t4_sq_onchip(wq)) { if (t4_sq_onchip(wq)) {
len16 = align(len16, 4); len16 = align(len16, 4);
wc_wmb(); wc_wmb();
} }
while (len16) {
*dst++ = *src++; total = len16 * 16;
if (dst == (u64 *)&wq->sq.queue[wq->sq.size]) end = (uintptr_t)&wq->sq.queue[wq->sq.size];
dst = (u64 *)wq->sq.queue; if (__predict_true((uintptr_t)dst + total <= end)) {
*dst++ = *src++; /* Won't wrap around. */
if (dst == (u64 *)&wq->sq.queue[wq->sq.size]) memcpy(dst, src, total);
dst = (u64 *)wq->sq.queue; } else {
len16--; len = end - (uintptr_t)dst;
memcpy(dst, src, len);
memcpy(wq->sq.queue, src + len, total - len);
} }
} }
static void copy_wr_to_rq(struct t4_wq *wq, union t4_recv_wr *wqe, u8 len16) static void copy_wr_to_rq(struct t4_wq *wq, union t4_recv_wr *wqe, u8 len16)
{ {
u64 *src, *dst; void *src, *dst;
uintptr_t end;
int total, len;
src = (u64 *)wqe; src = &wqe->flits[0];
dst = (u64 *)((u8 *)wq->rq.queue + wq->rq.wq_pidx * T4_EQ_ENTRY_SIZE); dst = &wq->rq.queue->flits[wq->rq.wq_pidx *
while (len16) { (T4_EQ_ENTRY_SIZE / sizeof(__be64))];
*dst++ = *src++;
if (dst >= (u64 *)&wq->rq.queue[wq->rq.size]) total = len16 * 16;
dst = (u64 *)wq->rq.queue; end = (uintptr_t)&wq->rq.queue[wq->rq.size];
*dst++ = *src++; if (__predict_true((uintptr_t)dst + total <= end)) {
if (dst >= (u64 *)&wq->rq.queue[wq->rq.size]) /* Won't wrap around. */
dst = (u64 *)wq->rq.queue; memcpy(dst, src, total);
len16--; } else {
len = end - (uintptr_t)dst;
memcpy(dst, src, len);
memcpy(wq->rq.queue, src + len, total - len);
} }
} }