On Wednesday 26 June 2013, James Bottomley wrote: > On Mon, 2013-06-24 at 08:18 -0400, Mark Lord wrote: > > And the SCSI stack in Linux has rather atrocious error handling. > > It lumps multiple requests together, and can fail the entire lot even > > if only a single sector is bad. > > That's rather misleading. SCSI doesn't lump anything together; it > handles the requests it was passed. For reads and writes through the > page cache, block will aggregate in the elevators, but you avoid that by > not using the page cache (O_DIRECT or SG_IO). Yes, it works fine with O_DIRECT - that's why hdd_realloc reads sector-by-sector when an error was detected. I'd also like to disable read retries but that does not seem to be possible. > For devices which report failing sectors correctly data up to the failed > sector is returned and the request is shortened and retried from the > failed sector on. If we get a second failure at the beginning (where > the previous bad sector was), then we give up. > > James -- Ondrej Zary -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
