Re: [Etherboot-discuss] PXENV_TFTP_OPEN locks up
Brought to you by:
marty_connor,
stefanhajnoczi
From: Brendan T. <btr...@gm...> - 2006-09-23 02:45:33
|
Hi, On 9/22/06, Michael Brown <mb...@fe...> wrote: > The UDP port is definitely meant to be little endian as far as the PXE > spec is concerned (even though everything else in the known universe uses > network endianness). I'm using "dw 69 << 8" on a (little endian) 80x86 rather than "dw 69", and etherboot says it's right.. ;-) > How long did you leave it to time out? I think the timeouts in Etherboot > 5.4 can be set rather high (up to around an hour), depending on > configuration options. In gpxe-0.5 it's reduced to something more > sensible. I probably only left it for around 15 minutes at most... To be honest I was expecting a 15 second time out (or less, considering TFTP isn't often used on a wide area network), which would give my little NBP the chance to display a "waiting for TFTP server" message (possibly with an additional '.' every 5 minutes) or perhaps the occasional beep from the speaker (for headless systems). I would've also assumed that this is the reason why "TFTP_READ_FILE" includes a "UINT16 TFTPOpenTimeout" field while "TFTP_OPEN_FILE" doesn't (the assumption being that "TFTP_READ_FILE" would continually retry until the NBP specified time out expires, while "TFTP_OPEN_FILE" never retries and only waits long enough to be reasonably sure that it's initial TFTP request isn't going to receive an ACK or error packet in reply). In this case, regardless of which function is used the client can have some control over how "lack of service" is handled. Of course it's unlikely that my interpretation of the "less than specific" specification corresponds to anyone else's interpretation, or common practice... :-) Cheers, Brendan |