This is an automated email from Gerrit.
Jan Matyas (mat...@codasip.com) just uploaded a new patch set to Gerrit, which
you can find at http://openocd.zylin.com/5412
-- gerrit
commit 45c400bb2a3e8c20374e87986e391f7175bcd2b7
Author: Jan Matyas <mat...@codasip.com>
Date: Thu Jan 23 07:54:34 2020 +0100
jtag_vpi: added timeout for socket receive
Adding timeout for socket receive operations in jtag_vpi driver.
Should the jtag_vpi server got broken and stopped responding (for
any reason), we need to have a means how to detect the situation
and report to the user - hence the introduction of the timeout.
The default jtag_vpi recv timeout is set to a very large (safe)
value - 15 sec - which is many orders of magnitude greater than
the typical exchange rate over the jtag_vpi socket (milliseconds).
Exceeding such an timeout always means a problem - either
the jtag_vpi server (RTL simulation) is terribly slow, or it
broke down and stopped responding for good.
TCL command "jtag_vpi_set_recv_timeout" has been also added to
make the timeout configurable, should the user ever need to
change it.
Change-Id: Ide2c5a52c86f1fc72febcc4bd115fa87f3e307ad
Signed-off-by: Jan Matyas <mat...@codasip.com>
diff --git a/src/jtag/drivers/jtag_vpi.c b/src/jtag/drivers/jtag_vpi.c
index 6abc702..4b5795a 100644
--- a/src/jtag/drivers/jtag_vpi.c
+++ b/src/jtag/drivers/jtag_vpi.c
@@ -33,6 +33,7 @@
#include <netinet/tcp.h>
#endif
+#include <helper/time_support.h>
#include <string.h>
#define NO_TAP_SHIFT 0
@@ -49,6 +50,10 @@
#define CMD_SCAN_CHAIN_FLIP_TMS 3
#define CMD_STOP_SIMU 4
+#define RECV_DEFAULT_TIMEOUT_SEC 15
+#define RECV_TIMEOUT_WARN_PERCENT 30 /* warn if XX % of the timeout is
reached */
+#define RECV_TIMEOUT_MAX_WARNINGS 16
+
/* jtag_vpi server port and address to connect to */
static int server_port = SERVER_PORT;
static char *server_address;
@@ -56,6 +61,16 @@ static char *server_address;
/* Send CMD_STOP_SIMU to server when OpenOCD exits? */
static bool stop_sim_on_exit;
+/* Maximum wait time for reception of a message from jtag_vpi server */
+static int recv_timeout_sec = RECV_DEFAULT_TIMEOUT_SEC;
+
+/* Threshold for wait time after which a warning will be printed. */
+static int64_t recv_warn_threshold_ms = (RECV_DEFAULT_TIMEOUT_SEC * 1000)
+ * RECV_TIMEOUT_WARN_PERCENT / 100;
+
+/* Number of timeout warnings emitted so far (avoid printing too many of
them). */
+static int num_recv_warnings;
+
static int sockfd;
static struct sockaddr_in serv_addr;
@@ -172,8 +187,71 @@ static int jtag_vpi_receive_cmd(struct vpi_cmd *vpi)
{
unsigned bytes_buffered = 0;
while (bytes_buffered < sizeof(struct vpi_cmd)) {
+ /* How many bytes left to receive? */
int bytes_to_receive = sizeof(struct vpi_cmd) - bytes_buffered;
- int retval = read_socket(sockfd, ((char *)vpi) +
bytes_buffered, bytes_to_receive);
+
+ /* Use select() to wait for incoming data just for a limited
period of time. */
+ const int64_t wait_start = timeval_ms();
+ fd_set read_fds;
+ FD_ZERO(&read_fds);
+ FD_SET(sockfd, &read_fds);
+ struct timeval tv;
+ tv.tv_usec = 0;
+ tv.tv_sec = recv_timeout_sec;
+ int retval = socket_select(sockfd + 1, &read_fds, NULL, NULL,
&tv);
+ if (retval < 0) {
+ /* select() failed */
+#ifdef _WIN32
+ int wsa_err = WSAGetLastError();
+ if (wsa_err == WSAEINTR) {
+ /* Wait interrupted by WSACancelBlockingCall().
Keep waiting. */
+ continue;
+ }
+#else
+ if (errno == EINTR) {
+ /* Wait interrupted by a signal. Keep waiting.
*/
+ continue;
+ }
+#endif
+ else {
+ /* This should not happen. It would mean that
select() failed miserably.
+ * Jtag_vpi protocol cannot recover from it, so
exit. */
+ LOG_ERROR("Error when waiting for jtag_vpi
socket readiness "
+ "(select failed, errno=%d)",
errno);
+ exit(-1);
+ }
+ } else if (retval == 0) {
+ /* select() timed out.
+ This should hardly ever happen and if it does, it
always indicates a problem.
+ Most likely the performance of the jtag_vpi server
(RTL simulation)
+ is extremely low. Or the server may have broken down
and stopped
+ responding for good. */
+ LOG_ERROR("Timeout when waiting for response from
jtag_vpi server. "
+ "(>%d seconds)", recv_timeout_sec);
+ LOG_INFO("Hint: Command 'jtag_vpi_set_recv_timeout' can
be used to adjust the "
+ "jtag_vpi timeout.");
+
+ /* Jtag_vpi protocol, as designed, cannot recover from
this.
+ * The best we can do is to terminate and notify the
user. */
+ exit(-1);
+ }
+ /* Otherwise select() succeeded. New data is available at the
socket. */
+ /* Just warn the user if the data arrived with a long delay. */
+ int64_t wait_duration = timeval_ms() - wait_start;
+ if (wait_duration > recv_warn_threshold_ms) {
+ if (num_recv_warnings < RECV_TIMEOUT_MAX_WARNINGS) {
+ /* Early warning to the user that the
performance of jtag_vpi server
+ * is too low. */
+ LOG_WARNING("Slow response from jtag_vpi: %.1f
sec(s). (Low performance of "
+ "jtag_vpi server?)",
(double)(wait_duration/1000.0));
+ num_recv_warnings++;
+ if (num_recv_warnings ==
RECV_TIMEOUT_MAX_WARNINGS)
+ LOG_WARNING("Too many slow responses
about jtag_vpi. Suppressing further warnings.");
+ }
+ }
+
+ /* Read the data from the socket. */
+ retval = read_socket(sockfd, ((char *)vpi) + bytes_buffered,
bytes_to_receive);
if (retval < 0) {
#ifdef _WIN32
int wsa_err = WSAGetLastError();
@@ -199,7 +277,8 @@ static int jtag_vpi_receive_cmd(struct vpi_cmd *vpi)
bytes_buffered += retval;
}
- /* Use little endian when transmitting/receiving jtag_vpi cmds. */
+ /* Handle endianness of received data.
+ * Jtag_vpi datagrams are transmitted in little endian over network. */
vpi->cmd = le_to_h_u32(vpi->cmd_buf);
vpi->length = le_to_h_u32(vpi->length_buf);
vpi->nb_bits = le_to_h_u32(vpi->nb_bits_buf);
@@ -638,6 +717,27 @@ COMMAND_HANDLER(jtag_vpi_stop_sim_on_exit_handler)
return ERROR_OK;
}
+COMMAND_HANDLER(jtag_vpi_set_recv_timeout_handler)
+{
+ if (CMD_ARGC != 1) {
+ LOG_ERROR("jtag_vpi_set_recv_timeout expects 1 argument");
+ return ERROR_COMMAND_SYNTAX_ERROR;
+ } else {
+ int val;
+ COMMAND_PARSE_NUMBER(int, CMD_ARGV[0], val);
+ if (val < 1) {
+ LOG_ERROR("The argument of
\"jtag_vpi_set_recv_timeout\" must be a positive integer");
+ return ERROR_COMMAND_SYNTAX_ERROR;
+ }
+ /* Hard receive timeout - store the new value given by user */
+ recv_timeout_sec = val;
+ /* Soft receive timeout (just for warnings) - recompute the
value */
+ recv_warn_threshold_ms = (recv_timeout_sec * 1000)
+ * RECV_TIMEOUT_WARN_PERCENT / 100;
+ }
+ return ERROR_OK;
+}
+
static const struct command_registration jtag_vpi_command_handlers[] = {
{
.name = "jtag_vpi_set_port",
@@ -661,6 +761,13 @@ static const struct command_registration
jtag_vpi_command_handlers[] = {
"before OpenOCD exits (default: off)",
.usage = "<on|off>",
},
+ {
+ .name = "jtag_vpi_set_recv_timeout",
+ .handler = &jtag_vpi_set_recv_timeout_handler,
+ .mode = COMMAND_CONFIG,
+ .help = "Set the timeout for responses from jtag_vpi
connection",
+ .usage = "<seconds>",
+ },
COMMAND_REGISTRATION_DONE
};
--
_______________________________________________
OpenOCD-devel mailing list
OpenOCD-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/openocd-devel
