Most recent kernel where this bug did not occur: 2.6.10 Distribution: Gentoo Hardware Environment: x86, AHA19160, WD18310 Software Environment: Problem Description: In 2.6.14 and 2.6.16 I get domain validation errors on my WD18310 connected to a 19160, causing it to drop back to asynchronous rather than 40 or 80MHz wide. Domain validation reports parity errors, write buffer failures, performs resets and generally stuffs around for a few minutes before deciding it will allow async. I have 2 other discs (Seagate SX118202LS) attached to the same chain which still work. My scsi system works perfectly in 2.6.10 with no data corruption. I have a very similar issue with a dual SYM53C896 in a different machine: it works in 2.6.10 but produces occasional noise in dmesg: sym0:9:0:phase change 6-7 11@17cd5f84 resid=6. On newer (>= 2.6.14) kernels, it completely fails to boot, giving the same sort of parity errors I'm having with the aic7xxx driver. When I checked at the time (a while ago now), there was NO change in the relevant driver between kernel versions. The only changes are to the scsi architecture. This leads me to believe the bug is not in the aic7xxx driver but the new scsi domain validation code that was overhauled somewhere around 2.6.13-14. Some of my error messages look a lot like those in #5268. Steps to reproduce: Use kernel >=2.6.14 with aic7892 or 53C896.
Created attachment 8267 [details] dmesg output (first part missing, sorry) dmesg has truncated the early messages, sorry. You can see the tail end of the final reset phase, which is about 10% of what it had to say for itself.
Created attachment 8268 [details] /proc/scsi/aic7xxx/0
Created attachment 8325 [details] dmesg output from working 2.6.10 kernel.
Created attachment 8326 [details] patch from James Bottomley for debug purposes
Created attachment 8327 [details] dmesg from 2.6.16 after applying James' patch (attached above) Note the parity errors on data-in, even after skipping the write tests.
Is this issue still present in kernel 2.6.19?
Please reopen this bug if it's still present with kernel 2.6.20.