On Thu, 2012-10-11 at 15:49 +0200, asronchetti@libero.it wrote: > Hi, > i did some test with other distros (slackware/fedora) and the corruption > problem seems to persist. > Yesterday i was coming back to debian , dumping a dd backup image from an > external WD hard disk, using a fedora usb-live, this way: > > dc3dd if=./debian-backup-image of=/dev/sda1 hash=sha256 > > the partition image is ~2gB and before the end of the copy the system crashed. > (sda1 is on the new SSD disk i bought ~ on May) > > so i rebooted fedora and i launched the command again , using a sequence of > Ctrl-Z, 'sync', and 'fg' , each 500mB. > the copy was successful this time. > So i think the problem could be the new ram (8gB) i bought together with the > SSD disk. > > So i replaced the hard disk and the ram with the old ones. (normal hd + 4gB > ram. It's the original set of this notebook ). > > i rebooted fedora live-usb and i repeated the operation using as destination > the sda1 partition of the original disk. It went ok with no problem, it copied > 2gB in a time, without crashing. > > Now i'm trying debian wheezy on this non-SSD hard disk, with bumblebee enabled > for a week (hoping it doesnt crash this time). After that , i'll dump the > debian stable backup on sda1 of this same disk, to see how it is . > > If the tests will be successful in the next few days, it could mean that the > problem was caused by some caching problem in the SSD , right? Could be. > I still dont understand why the combination of those SSD + 8gB ram caused > problems. > > I've just checked my syslog and i see that with this setup (wheezy + non-SSD + > the old 4gB RAM) the AMI BIOS memory corruption message doesnt appear anymore. > > i'll upgrade the bug in the next few days with the outcomes of these tests. Some internal devices e.g. disk or network controller can only write to a limited memory range (e.g. up to 4 GB). If the driver is not aware of the actual limit then it may tell the device to write to memory outside this range, and the device will then write to the wrong area of memory (and similarly when reading). This can result in severe data corruption and crashes. This would normally be considered a driver bug, as it is supposed to tell the kernel which areas of memory to use with the device. I wouldn't expect many recently designed chips to have such problems, though. Ben. -- Ben Hutchings Kids! Bringing about Armageddon can be dangerous. Do not attempt it in your own home. - Terry Pratchett and Neil Gaiman, `Good Omens'
Attachment:
signature.asc
Description: This is a digitally signed message part