I have updated diffs for the replacement scsi queueing code on
http://www.andante.org/scsi_queue.html. I now have diffs against 2.3.15,
2.3.16, and 2.3.17.
I still have only done light testing, but I have corrected a couple of
module related breakages which were slowing down my testing. The only
outstanding problem that I have is that my aic7xxx driver is hanging during
load, and this is because the interrupt line never gets cleared, and the
driver is stuck in an infinite loop in the interrupt handler. Now that the
merging is done, I will try a different card for more extensive testing.
If anyone has any ideas as to why the interrupt line is getting stuck,
drop me a line. The card is an AIC7850, and the test machine is a K5, if
that makes any difference. FWIW, the same card works fine in the same
machine with a 2.0.?? kernel. I may have to go in and try and debug this
further. I have the same problem with an unmodified kernels, so the
problem is not related to anything I have done.
I have a fairly high degree of certainty that if you do not select the
new queueing code that the behavior should be identical to the current
distribution kernel. I basically made a local copy of my changes, stripped
out the new code and diffed it against the original, and I am only seeing
changes related to comments or a few trivial formatting changes.
Nonetheless, people probably should try this to make sure that I am in fact
correct. Note that I have identified some bugs in the existing code which
are only fixed in the new queueing code (it would be relatively easy to make
these fixes unconditional).
Anyone who tries the new queueing code should of course be prepared for
possible system crashes and data corruption, especially at these early
stages. Also, I realized last night that I still haven't provided support
for "single lun" in the new queueing code, so don't bother trying the new
queueing code if you are using a CD changer.
I suspect that with the new queueing code that people will see an
improved performance (mainly by streamlining and the elimination of the rats
nest known as do_sd_request()). I saw a very minor improvement on my
laptop when running the simulator, but that's obviously not a representative
test. If you are doing benchmarking, please turn off the additional
consistency checking for the new queueing code :-).
If anyone has any questions, comments, bug reports, etc, I would be
interested in hearing them.
-Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
