diff --git a/src/hal/drivers/mesa-hostmot2/hm2_eth_net_posix.c b/src/hal/drivers/mesa-hostmot2/hm2_eth_net_posix.c index 4e8ddb48d68..2c5f882cab0 100644 --- a/src/hal/drivers/mesa-hostmot2/hm2_eth_net_posix.c +++ b/src/hal/drivers/mesa-hostmot2/hm2_eth_net_posix.c @@ -148,30 +148,16 @@ int hm2_posix_eth_socket_send(hm2_eth_t *board, const void *buffer, int len) { } int hm2_posix_eth_socket_recv(hm2_eth_t *board, void *buffer, int len, int recv_timeout_ns) { - struct pollfd pfd; - struct timespec ts; int ret; - - //SO_RCVTIMEO only delivers a timeout down to ~10ms - //while ppoll() works down to 100us - pfd.fd=board->sockfd; - pfd.events = POLLIN; - ts.tv_sec = 0; - ts.tv_nsec = recv_timeout_ns; - while (ts.tv_nsec >= 1000000000) { - ts.tv_nsec -= 1000000000; - ts.tv_sec ++; - } - ret = ppoll(&pfd, 1, &ts, NULL); - - if (ret < 0) { - LL_PRINT("ERROR: ppoll() failed: %m\n"); - } else if(ret) { - ret = recv(board->sockfd, buffer, len, 0); - } else { - errno = EAGAIN; - ret = -1; - } - + //Note: Do not use ppoll here. It increases the read time on + //badly tuned PC's. Most probably due to the CPU goes to sleep + //states or clocks down. + const long poll_time_ns = 10000; + long long read_deadline = rtapi_get_time() + recv_timeout_ns; + do { + errno = 0; + ret = recv(board->sockfd, buffer, len, MSG_DONTWAIT); + if(ret < 0) rtapi_delay(poll_time_ns); + } while (ret < 0 && rtapi_get_time() < read_deadline); return ret; }