I thought I had eliminated the allocation from interrupt context, but I'll look at
   it again.

[snip]

   > Ok...  I set a breakpoint with ddb and the following stack trace was
   > returned [1.4L kernel]:
   >
   > _usb_block_allocmem(f026dcc4,1000,0,f0568c3c,f0460000) at _usb_block_allocmem + 0x15
   > _usb_allocmem(f0460000,29b,0,f0568c3c,f4e62bdc) at _usb_allocmem + 0x39
   > _uhci_allocm(f0460000,f0568c3c,29b,e000de42,f4c34390) at _usb_allocm + 0x13
   > _usbd_transfer(f056800) at _usbd_transfer + 0x52
   > _umodemstart(f4c34390) at _umodemstart + 0xd0
   > _pppstart(f4c34390) at _pppstart + 0x1b
   > _pppsyncstart(f02908cc) at _pppsyncstart + 0x484
   > _pppintr(10,10,f4e1bdc8,f4e1bdc8,f4e62ed4) at _pppintr + 0x89
   >
   > [ddb complained about corruption at this point]
   >
   > I could not make the system panic from ddb, so I do not have a kernel core
   > dump to verify the above which was copied on to paper and then into email.
   >
   > Brad Spencer - brad@anduin.eldar.org   http://anduin.eldar.org
   > [finger brad@anduin.eldar.org for PGP public key]

   --

	   -- Lennart


You may very well have...  the above was a 1.4L kernel.  I compiled up
1.4P on another system with the same uhci USB controller and it didn't
appear to have the above problem.  However, I just tried it on a 1.4L ohci
system and managed to cause it, so that particular issue might have been
delt with.

However, I get the following on the uhci system after pppd exits and I use
'cu' to talk to the modem:

uhci_device_bulk_transfer: not done, ii=0xf0530ac0
uhci_device_bulk_transfer: not done, ii=0xf05320a0
uhci_device_bulk_transfer: not done, ii=0xf0530ac0
uhci_device_bulk_transfer: not done, ii=0xf05320a0
uhci_device_bulk_transfer: not done, ii=0xf0530ac0


The ohci system didn't appear to exibit anything like this.


There appears to be some issues with running ppp over a umodem.  Stuff
like ftp seems to work reasonably well, but if ftp is running, interactive
telnet session appear to get starved.  Normal dialup session, with 'cu',
seem fine, also.  [or at the very least, the behavior seems somewhat
different between the umodem and a normal serial modem].


Brad Spencer - brad@anduin.eldar.org   http://anduin.eldar.org
[finger brad@anduin.eldar.org for PGP public key]