Signed-off-by: Michael Walle <michael@walle.cc>

That way we have more registers addresses left for the SIE module.

Signed-off-by: Michael Walle <michael@walle.cc>

Signed-off-by: Michael Walle <michael@walle.cc>

Signed-off-by: Michael Walle <michael@walle.cc>

Signed-off-by: Michael Walle <michael@walle.cc>

Don't silently ignore errors while parsing the parameters.

Signed-off-by: Michael Walle <michael@walle.cc>

Drop doublerate parameter and replace it with a generic speed parameter.
Additionally, on linux you can supply any (non-standard) baudrate.

Signed-off-by: Michael Walle <michael@walle.cc>

This brings the LV3 success rate to 50/50 = 100%.

If hitting persistent timeouts or an unresponsive device, we could
loop forever. This would hang the entire USB subsystem, also for any
devices on the other port. Also unplugging wouldn't help. Only a
reboot or - lately - reloading the USB firmware would get out of this.

Parameter stability:

Retry limit for main data phase
|    Retry limit for the IN/OUT ending it
|    |    LV3 successes
|    |    |
 50   10  13/25 = 52%
100   10  23/50 = 46%
250   10  12/25 = 48%
250  100  29/50 = 58%

If the OUT token to end a data phase wasn't ack'ed, we didn't notice.

This still leaves the LV3 success rate at 13/25 = 52%.

We started sending SOF a little too early, making it impossible to
reset the bus without physical unplugging.

Note that resetting an LV3 via unplug-usb_load-plug leads to
successful enumeration with probability 40/75 = 53% while only
usb_load yields 23/50 = 46%.

This also introduces a failure mode not seen before where the
LV3 enumerates ("Found MIDI") but still doesn't communicate.
Subsequent bus resets are unsuccessful. Plug cycling helps.

We got a HUGE number of these with full-speed devices, badly spamming
the diagnostic output.

Note that this change may reduce the probability of successful
enumeration: with the timeout message, 36/50 = 72% unplug-reset-plug
cycles succeeded with LV3, while without the message, only 40/75 =
53% did.

This doesn't affect the code to currently generate. Just removes a
bit of potentially misleading information.

…ingly

The problem were:

- missing "void" in function declarations in main.c

- debug.c didn't include debug.h

None of these caused actual malfunction.

This is for debugging. Removed reading back the reset status via CSR,
since it seems not very useful and we don't have to decode the address
this way.

….bin to M1

Uses ftp-upload from the eponymous package.

Following discussion on #milkymist, the names are now "official".

  so far validate_report_descriptor is empty