diff options
Diffstat (limited to 'TODO-2.1/gibbs')
| -rw-r--r-- | TODO-2.1/gibbs | 65 |
1 files changed, 65 insertions, 0 deletions
diff --git a/TODO-2.1/gibbs b/TODO-2.1/gibbs new file mode 100644 index 000000000000..234e8405b50b --- /dev/null +++ b/TODO-2.1/gibbs @@ -0,0 +1,65 @@ +# TODO File for core-team member: Justin Gibbs +# Revision: $Id: gibbs,v 1.1 1995/07/03 22:04:37 gibbs Exp $ +# +# This list contains my medium term goals (three month time frames). Feel +# free to contact me on any of these issues. +# +-*----------------------------------------------------------------------------*- + +Update: Mon Jul 3, 1995 +Due: Mid Jul +Prio: High +Task: Adaptec 27/28/29/3940 driver house cleaning. + +The current aic7xxx driver, although quite stable, still has many weak +points. + +1) The abort code is incomplete and insufficient to handle most errors. +The code should first perform a bus device reset and resort to a full +bus reset only if that fails. The clean up after a bus reset currently +doesn't account for commands other than the command that was the cause +of the abort, so they are not reported properly to the upper level +SCSI code. + +2) The scatter gather sequencer code was an abomination. I've rewritten +it, but in making it faster, I've exposed some race conditions in the +Twin channel support that I'm in the process of diagnosing. + +3) I'd like to be able to have active selections on both busses of a twin +channeled adapter at a time. This may come about while I fix #2. + +4) Add support for the 3940 controller. I've already got this in my "alpha" +driver I'm working on, but I need to test being able to handle 255 commands. + +Update: Mon Jul 3, 1995 +Due: 2.1 +Prio: High +Task: Implement proper tagged queuing functionality and better error + handling in the SCSI subsystem. + +A major week point in the SCSI system is how it handles tape devices. It +doesn't allow any retries irregardless of the type of error encountered. +For example, a "device busy" statis is not a failure (many devices give +this statis when they are repositioning the tape), but will abort the I/O. +Whenever this type of abort occurs (this is the "Oops, not queued" error), +the VM system for some reason believes that buffer is already run down even +though is hasn't been touched at all. Other weak points are parity errors, +and sense keys that signal command aborts. + +Update: Mon Jul 3, 1995 +Due: 2.1 +Prio: High +Task: Better support for removable r/w SCSI devices. + +Chuck Robey has sent me a Zip drive to work on this with, but I haven't +had a chance to look at how difficult this task will be. + +Update: Mon Jul 3, 1995 +Due: 2.2 possibly 2.1 +Prio: Medium +Task: Get LFS to work + +Subject says it all. This entailes updating LFS to the new VM system and +killing many places in the code where it "doesn't play by the rules." How +soon I get to start working on this task will depend on how well the other +tasks go on my list. |