*/
#include <config.h>
+
+#include <string.h>
+#include <unistd.h>
+
#include "protocol.h"
-SR_PRIV int asix_omega_rtm_cli_receive_data(int fd, int revents, void *cb_data)
+/*
+ * Start the external acquisition process (vendor's CLI application).
+ * Get the initial response to verify its operation.
+ */
+SR_PRIV int omega_rtm_cli_open(const struct sr_dev_inst *sdi)
{
- const struct sr_dev_inst *sdi;
struct dev_context *devc;
+ gboolean ok;
+ gchar **argv;
+ GSpawnFlags flags;
+ GPid pid;
+ gint fd_in, fd_out;
+ GError *error;
+ GString *txt;
+ ssize_t rcvd;
+ uint8_t rsp[3 * sizeof(uint16_t)];
+ const uint8_t *rdptr;
+ uint16_t stamp, sample1, sample2;
- (void)fd;
+ if (!sdi)
+ return SR_ERR_ARG;
+ devc = sdi->priv;
+ if (!devc)
+ return SR_ERR_ARG;
- if (!(sdi = cb_data))
- return TRUE;
+ if (devc->child.running) {
+ sr_err("Vendor application already running.");
+ return SR_ERR_BUG;
+ }
+
+ /* Prepare to feed sample data to the session. */
+ memset(&devc->rawdata, 0, sizeof(devc->rawdata));
+ memset(&devc->samples, 0, sizeof(devc->samples));
+ devc->samples.queue = feed_queue_logic_alloc(sdi,
+ FEED_QUEUE_DEPTH, sizeof(devc->samples.last_sample));
+
+ /*
+ * Start the background process. May take considerable time
+ * before actual acquisition starts.
+ */
+ sr_dbg("Starting vendor application");
+ argv = devc->child.argv;
+ flags = devc->child.flags;
+ error = NULL;
+ ok = g_spawn_async_with_pipes(NULL, argv, NULL, flags, NULL, NULL,
+ &pid, &fd_in, &fd_out, NULL, &error);
+ if (error) {
+ sr_err("Cannot execute RTM CLI process: %s", error->message);
+ g_error_free(error);
+ ok = FALSE;
+ }
+ if (fd_in < 0 || fd_out < 0)
+ ok = FALSE;
+ if (!ok) {
+ sr_err("Vendor application start failed.");
+ return SR_ERR_IO;
+ }
+ devc->child.pid = pid;
+ devc->child.fd_stdin_write = fd_in;
+ devc->child.fd_stdout_read = fd_out;
+ devc->child.running = TRUE;
+ sr_dbg("Started vendor application, in %d, out %d", fd_in, fd_out);
+ txt = sr_hexdump_new((const uint8_t *)&pid, sizeof(pid));
+ sr_dbg("Vendor application PID (OS dependent): %s", txt->str);
+ sr_hexdump_free(txt);
+
+ /*
+ * Get the initial response. Verifies its operation, and only
+ * returns with success when acquisition became operational.
+ */
+ rcvd = read(fd_out, rsp, sizeof(rsp));
+ sr_dbg("Read from vendor application, ret %zd", rcvd);
+ if (rcvd < 0)
+ ok = FALSE;
+ if (ok && (size_t)rcvd != sizeof(rsp))
+ ok = FALSE;
+ if (!ok) {
+ omega_rtm_cli_close(sdi);
+ return SR_ERR_IO;
+ }
+
+ /*
+ * Ignore the first timestamp. Grab the most recent sample data
+ * to feed the session from it upon later repetition.
+ */
+ rdptr = rsp;
+ stamp = read_u16le_inc(&rdptr);
+ sample1 = read_u16le_inc(&rdptr);
+ sample2 = read_u16le_inc(&rdptr);
+ sr_dbg("stamp %u, samples %x %x", stamp, sample1, sample2);
+ write_u16le(devc->samples.last_sample, sample2);
+
+ return SR_OK;
+}
+
+/*
+ * Terminate the external acquisition process (vendor's CLI application).
+ */
+SR_PRIV int omega_rtm_cli_close(const struct sr_dev_inst *sdi)
+{
+ struct dev_context *devc;
+
+ if (!sdi)
+ return SR_ERR_ARG;
+ devc = sdi->priv;
+ if (!devc)
+ return SR_ERR_ARG;
+
+ /* Close the external process' stdin. Discard its stdout. */
+ sr_dbg("Closing vendor application file descriptors.");
+ if (devc->child.fd_stdin_write >= 0) {
+ sr_dbg("Closing vendor application stdin descriptor.");
+ close(devc->child.fd_stdin_write);
+ devc->child.fd_stdin_write = -1;
+ }
+ if (devc->child.fd_stdout_read >= 0) {
+ sr_dbg("Closing vendor application stdout descriptor.");
+ close(devc->child.fd_stdout_read);
+ devc->child.fd_stdout_read = -1;
+ }
+
+ /* Terminate the external process. */
+ if (devc->child.running) {
+ sr_dbg("Closing vendor application process.");
+ (void)g_spawn_close_pid(devc->child.pid);
+ memset(&devc->child.pid, 0, sizeof(devc->child.pid));
+ devc->child.running = FALSE;
+ }
+
+ /* Release the session feed queue. */
+ if (devc->samples.queue) {
+ feed_queue_logic_free(devc->samples.queue);
+ devc->samples.queue = NULL;
+ }
+
+ sr_dbg("Done closing vendor application.");
- if (!(devc = sdi->priv))
+ return SR_OK;
+}
+
+/*
+ * Process received sample data, which comes in 6-byte chunks.
+ * Uncompress the RLE stream. Strictly enforce user specified sample
+ * count limits in the process, cap the submission when an uncompressed
+ * chunk would exceed the limit.
+ */
+static int omega_rtm_cli_process_rawdata(const struct sr_dev_inst *sdi)
+{
+ static const size_t chunk_size = 3 * sizeof(uint16_t);
+
+ struct dev_context *devc;
+ const uint8_t *rdptr;
+ size_t avail, taken, count;
+ uint16_t stamp, sample1, sample2;
+ int ret;
+
+ devc = sdi->priv;
+ rdptr = &devc->rawdata.buff[0];
+ avail = devc->rawdata.fill;
+ taken = 0;
+ ret = SR_OK;
+
+ /* Cope with previous errors, silently discard RX data. */
+ if (!devc->samples.queue)
+ ret = SR_ERR_DATA;
+
+ /* Process those chunks whose reception has completed. */
+ while (ret == SR_OK && avail >= chunk_size) {
+ stamp = read_u16le_inc(&rdptr);
+ sample1 = read_u16le_inc(&rdptr);
+ sample2 = read_u16le_inc(&rdptr);
+ avail -= chunk_size;
+ taken += chunk_size;
+
+ /*
+ * Uncompress the RLE stream by repeating the last
+ * sample value when necessary. Notice that the stamp
+ * has a resolution of 10ns and thus covers two times
+ * the number of samples, these are taken each 5ns (at
+ * 200MHz rate). A stamp value of 1 is immediately
+ * adjacent to the last chunk.
+ */
+ if (stamp)
+ stamp--;
+ count = stamp * 2;
+ if (devc->samples.check_count) {
+ if (count > devc->samples.remain_count)
+ count = devc->samples.remain_count;
+ devc->samples.remain_count -= count;
+ }
+ if (count) {
+ ret = feed_queue_logic_submit(devc->samples.queue,
+ devc->samples.last_sample, count);
+ if (ret != SR_OK)
+ break;
+ sr_sw_limits_update_samples_read(&devc->limits, count);
+ }
+ if (devc->samples.check_count && !devc->samples.remain_count)
+ break;
+
+ /*
+ * Also send the current samples. Keep the last value at
+ * hand because future chunks might repeat it.
+ */
+ write_u16le(devc->samples.last_sample, sample1);
+ ret = feed_queue_logic_submit(devc->samples.queue,
+ devc->samples.last_sample, 1);
+ if (ret != SR_OK)
+ break;
+
+ write_u16le(devc->samples.last_sample, sample2);
+ ret = feed_queue_logic_submit(devc->samples.queue,
+ devc->samples.last_sample, 1);
+ if (ret != SR_OK)
+ break;
+
+ count = 2;
+ sr_sw_limits_update_samples_read(&devc->limits, count);
+ if (devc->samples.check_count) {
+ if (count > devc->samples.remain_count)
+ count = devc->samples.remain_count;
+ devc->samples.remain_count -= count;
+ if (!devc->samples.remain_count)
+ break;
+ }
+ }
+
+ /*
+ * Silently consume all chunks which were successfully received.
+ * These either completely got processed, or we are in an error
+ * path and discard unprocessed but complete sample data before
+ * propagating the error condition. This simplifies the logic
+ * above, and allows to keep draining the acquisition process'
+ * output, perhaps even resynchronize to it in a later attempt.
+ * The cost of this rare operation does not matter, robustness
+ * does.
+ */
+ while (avail >= chunk_size) {
+ avail -= chunk_size;
+ taken += chunk_size;
+ }
+
+ /*
+ * Shift remainders (incomplete chunks) down to the start of the
+ * receive buffer.
+ */
+ if (taken && avail) {
+ memmove(&devc->rawdata.buff[0],
+ &devc->rawdata.buff[taken], avail);
+ }
+ devc->rawdata.fill -= taken;
+
+ return ret;
+}
+
+SR_PRIV int omega_rtm_cli_receive_data(int fd, int revents, void *cb_data)
+{
+ const struct sr_dev_inst *sdi;
+ struct dev_context *devc;
+ uint8_t *buff;
+ size_t space;
+ ssize_t rcvd;
+ int ret;
+
+ sdi = cb_data;
+ if (!sdi)
+ return TRUE;
+ devc = sdi->priv;
+ if (!devc)
return TRUE;
- if (revents == G_IO_IN) {
- /* TODO */
+ /* Process receive data when available. */
+ if (revents & G_IO_IN) do {
+ buff = &devc->rawdata.buff[devc->rawdata.fill];
+ space = sizeof(devc->rawdata.buff) - devc->rawdata.fill;
+ rcvd = read(fd, buff, space);
+ sr_spew("Read from vendor application, ret %zd", rcvd);
+ if (rcvd <= 0)
+ break;
+ devc->rawdata.fill += (size_t)rcvd;
+ ret = omega_rtm_cli_process_rawdata(sdi);
+ if (ret != SR_OK) {
+ sr_err("Could not process sample data.");
+ }
+ } while (0);
+
+ /* Handle receive errors. */
+ if (revents & G_IO_ERR) {
+ (void)feed_queue_logic_flush(devc->samples.queue);
+ (void)sr_dev_acquisition_stop((struct sr_dev_inst *)sdi);
+ }
+
+ /* Handle optional acquisition limits. */
+ if (sr_sw_limits_check(&devc->limits)) {
+ (void)feed_queue_logic_flush(devc->samples.queue);
+ (void)sr_dev_acquisition_stop((struct sr_dev_inst *)sdi);
}
return TRUE;