On Thu, 19 Jul 2007 16:04:45 +0300 Boaz Harrosh wrote: > Randy Dunlap wrote: > > On Wed, 18 Jul 2007 19:43:35 -0700 Randy Dunlap wrote: > > > >> On Wed, 18 Jul 2007 14:58:18 -0700 Randy Dunlap wrote: > >> > >>> On Wed, 18 Jul 2007 10:33:20 -0700 Randy Dunlap wrote: > >>> > >>>> Hi, > >>>> > >>>> I don't see the reset problem after applying this patch, but I'm > >>>> still seeing an Oops in datao_run. Log is attached. > OK 1 bug down. > > >> Looks to me like this driver doesn't handle highmem virtual addresses > >> very well. I expect that it would work OK if CONFIG_HIGHMEM=n or > >> it had a slave_alloc function to limit the memory addresses that it > >> sees (like ppa & imm do). As it is on a machine with HIGHMEM=y, > >> it gets sglist virtual addresses of 0, e.g.: > > > > Confirmed that limiting the blk queue bounce limit to > > BLK_BOUNCE_HIGH runs as expected (copy files, compare them successfully). > > > This is grate! Thanks for checking and debugging this. > Am I right in assuming that the remaining bug has been there > before my patchset? Has my patchset made it any worse? Yes, this problem has been around forever AFAIK. You didn't add to it. > > include/asm-i386/scatterlist.h talks about using pci_map_sg() > > to get bus addresses from the scatterlist table. However, this driver > > knows nothing about PCI. Should it use dma sg mapping instead? > > or what is the desired fix here? > > I'm afraid this is out of my league. James, Christoph, anybody? > What would be the right thing to do here? > > > > > Thanks. > > --- > > Thank you. Please forgive me with the debug thing. One thing I learned in > my 17 years of programing is never send code you have not compiled & tested. > I mean what can go wrong with turning debugging on, right?;) > > I will wait out the solution to above remaining problem and send another > round of patches. > > One more thing if I may? Do you have at hand an old Zip cartridge that > has bad sectors or other errors on it. My motivation is to exercise the > Sense return code-path. I would like to make sure that its doing the > expected. It is the part that changed the most. > I will try to look if there is a way to send a scsi_read using sg3_utils that > will read passed the media capacity, and will trigger a short read. I guess > on an unmounted device. Douglas? No, I don't have any bad media that I know of. I suggest that you look into the "Fault-injection framework" (CONFIG_FAULT_INJECTION) to see if it can inject some errors for you. --- ~Randy *** Remember to use Documentation/SubmitChecklist when testing your code *** - To unsubscribe from this list: 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
