My new debugging setup, along with increasing the logging levels in lwip, quickly revealed some of the problems I am having. I managed to tackle a few of them last night.
First, I noticed my lwip ppp interface is not attempting to re-negotiate a connection. So, if it fails the first time, you have to quit the program and start again. This isn't very user-friendly, so I added a timed function to check the ppp status and, if necessary, alert the user and restart the ppp negotiation. This makes recovery from a timing problem much easier.
Next, during a failed connection to the webserver, I would get a warning like:
pppos_input: Dropping bad fcs 0xc540 proto=0x21
Basically, lwip was dropping ppp packets. The bad fcs was misleading and caused me some headaches. Finally, I noticed that every time this happened, about 4 lines before the packet was dropped would be a log entry: pppos_input: got 63 bytes Now, lwip ppp reports got xx bytes whenever it gets data from the serial port. It is the 63 bytes that is significant, and was associated with the dropped packet. I happened to retain some information from my book, Inside Macintosh, Vol. II:
Each input driver's buffer can initially hold up to 64 characters, but your application can specify a larger buffer if necessary.Ah! So, lwip was getting the opening part of a packet, which was cutoff by the buffer overrun, then getting another packet, ending with the checksum. The checksum wouldn't add up, so it dumped the packet (actually two packets). I think it reports 63 bytes rather than 64 bytes because the buffer is a pascal string with the first byte representing the string length(?). Fortunately, Inside Macintosh, Vol. II also provided a solution:
SerSetBuf allows you to specify a new input buffer, replacing the driver's 64-character default buffer.So, I added a 256 byte buffer and the connection is much more reliable. It even significantly improved the rate of initial ppp negotiation.