Hi Finn,
> Thanks for testing. > > "No regressions over v1" means "no regressions", right? > > > > Well, what would I compare the driver performance to? With your patches to > sort out locking races, the driver is more stable than I've ever seen it in > years. That's a definite win. Big improvement over the driver in its current > state in m68k and mainstream (which locks up quite reliably with even > moderate concurrent IDE and SCSI I/O for me). No regression over v1 or > patches that you sent for me to test off-list. > > On the other hand, I've seen warnings about lost bytes (stuck in the DMA > fifo) for the first time _ever_ with the new driver - we've discussed that > at length, and it is still unclear why these happen. This is a known NCR5380 > issue, and pretty much anything could have precipitated that. Must have > happened for other Falcon users before in the past, because the interrupt > handler explicitly checks for this condition. I went back and checked test logs from driver tests before any of your patches - no regressions over vanilla 3.17 or 3.16. The 'lost bytes' error is rare enough that it would not have shown in those tests. Vanilla kernels run into the ST-DMA locking race at about 7% of the total test set. The last kernel that performed with a similar stability as with your patches was 2.4.30 or thereabouts. Cheers, Michael -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
