home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.amiga
- Path: sparky!uunet!sun-barr!cs.utexas.edu!usc!sol.ctr.columbia.edu!The-Star.honeywell.com!umn.edu!kksys.com!quest!digibd!kas!rhealey
- From: rhealey@kas.helios.mn.org (Rob Healey)
- Subject: Re: Kernel Panics
- Message-ID: <1992Aug26.145201.18842@kas.helios.mn.org>
- Date: Wed, 26 Aug 1992 14:52:01 GMT
- References: <DON.92Aug25011100@blkhole.resun.com>
- Organization: Rob's home system, Hopkins, MN
- Lines: 40
-
- In article <DON.92Aug25011100@blkhole.resun.com> don@blkhole.resun.com (Don Phillips) writes:
- =I guess I really shouldn't read this news group. No sooner did I read
- =the article about kernel panics, than I had the pleasure of having
- =three occur.
- =
-
- Looks like you've hit the STREAMS "feature"... From my experience
- it was a ql serial port getting flooded with serial data but
- hardware flow control was held low and no program could consume
- the data coming into the port. When I hardwired RTS<->CTS and
- had a program running on the port to suck up the data the panics
- stopped.
-
- All I can suggest is to make VERY sure peripherals attached to serial
- ports PROPERLY obey full duplex hardware flow control and stop
- sending data when RTS goes low.
-
- There appears to be a STREAMS bottleneck in buffer
- allocation/deallocation that can degenerate into a deadlock and
- either silent death of all STREAMS I/O, serial ports, network
- ports and console screens, or a panic. The first routines after
- the 5 panic rountines in your dump seem to point to the STREAMS
- subsystem.
-
- The only way to avoid the STREAMS problems is to make sure you
- don't create excessive amounts of data sitting in STREAMS buffers
- and not being processed. The modem situation triggers this because
- the incoming data keeps causing new buffers to be needed and
- the system seems to assume that if it lowers RTS as a warning to
- please stop the peripheral will stop. If the peripheral keeps
- going it exhausts STREAMS resources and bad things happen. When
- the peripheral obeys the flow control, buffers aren't used excessively
- and when programs suck off the data again, the buffers aren't
- used excessively.
-
- This is all observation, reality maybe an entirely different reason...
-
- Hope this helps find a workaround for you,
-
- -Rob
-