Dave Huang wrote:
> Hmm, well here's what I've been able to figure out so far... if I put a
> printf() in adb_reinit(), after the call to adb_op_sync in the loop that
> does the initial scan through the devices, things seems to work. A
> DELAY(1000) or DELAY(1000000) instead doesn't help. It's stuck in
> send_adb_cuda(), in the loop at the end (the one that says "poll until
> byte done"). The interrupt level or whatever that thing is called is 7;
> i.e. everything except NMI is blocked. Is that okay?

that's pretty standard, but i guess whether or not it's wrong depends on
whether or not polling is actually working on the cuda-based machines.

> It looks like the
> ADB_INTR_IS_ON macro is waiting for something interrupt-related...

probably so.
 
> I hacked ddb to print a stack trace when it's invoked, since I can't
> type "t" to do it myself, but there's nothing particularly interesting
> in it... adbattach -> ADBReInit -> adb_reinit -> adb_op_sync -> adb_op
> -> send_adb_cuda.

cool!  i've been meaning to do this for some time.  could you send me the
patch?  i'd appreciate it.

> BTW, anyone know if the problem only occurs on Cuda machines? I've got a
> 660av...

i've only seen it on my 840av, not on any of my other machines (like my
q700 or my se/30).  so far, i've only seen it reported on cuda-based
machines...

perhaps something's a little screwed up there.

later.

colin