In commit e5eefe77435a ("tcp: Refactor to use events instead of
states, split out spliced implementation"), this:
if (!bitmap_isset(rcvlowat_set, conn - ts) &&
readlen > (long)c->tcp.pipe_size / 10) {
(note the !) became:
if (conn->flags & lowat_set_flag &&
readlen > (long)c->tcp.pipe_size / 10) {
in the new tcp_splice_sock_handler().
We want to check, there, if we should set SO_RCVLOWAT, only if we
haven't set it already.
But, instead, we're checking if it's already set before we set it, so
we'll never set it, of course.
Fix the check and re-enable the functionality, which should give us
improved CPU utilisation in non-interactive cases where we are not
transferring at full pipe capacity.
Fixes: e5eefe77435a ("tcp: Refactor to use events instead of states, split out
spliced implementation")
Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com>
---
tcp_splice.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tcp_splice.c b/tcp_splice.c
index 8a39a6f..5d845c9 100644
--- a/tcp_splice.c
+++ b/tcp_splice.c
@@ -556,7 +556,7 @@ eintr:
if (readlen >= (long)c->tcp.pipe_size * 10 / 100)
continue;
- if (conn->flags & lowat_set_flag &&
+ if (!(conn->flags & lowat_set_flag) &&
readlen > (long)c->tcp.pipe_size / 10) {
int lowat = c->tcp.pipe_size / 4;
--
2.43.0