On Fri, 9 Jun 2006, Lev A. Melnikovsky wrote: > Dear Guennadi, > > > Return the active request for retry to the upper layer in case of > > "reselect without disconnect" instead of just losing it. > Oh, thank you for the patch. Now the scan is finished even if several > "reselect: w/o disconnected cmds <02-0> with active srb" happened during the > operation. Wow, really! So, we are on the right track. > A funny artifact though can be seen at the picture (this is a > ~800KB map of Japan http://kapitza.ras.ru/~leva/Japan.jpg ). It looks like > during each reselect a chunk of data is lost (or maybe transferred twice?) and > the data stream gets an offset. My understanding is that the scan data are > 8bits per color component which gives 24bits/pixel. Nice pic, if you took a bit further West you'd scan Vladivostok - that's where I come from:-) > Any idea why it happens and how to fix it? Well, I think, we just have to chose the right error-cods to return, this controls the mid-layer bahaviour - whether to abort, or to resubmit, or whatever. The error I returned (DID_SOFT_ERROR) means "retry". Honestly, I don't know what's the correct error-code here. I can only guess. Well, please, try DID_ABORT. Also, you can strace the application with complete data log (strace -xx -s 1024 -o/tmp/scan.trace) and try to find if there are duplicated blocks... Maybe we should just report an error to the application and let it re-try? It seemed logical to me to ask for a retry. James? > On Wed, 7 Jun 2006 at 11:28pm, Guennadi Liakhovetski wrote: > > > Warning, Lev - this patch may explode your scanner, or upload your bank > > account info to p2p servers, or send your gpg keys to microsoft, or... > > You've been warned:-) You might want to wait for some comments to the > > patch... > Thanks for the warning :-) Thanks for your courage:-) Guennadi --- Guennadi Liakhovetski - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
