[EMAIL PROTECTED] wrote on 10/15/2008 05:36:12 PM: > [lpi-examdev] Half-OT: is parted reliable? > > Bryan J. Smith > Please respond to b.j.smith, "This is the lpi-examdev mailing list." > > > I'm not sure if this report still applies > > http://osdir.com/ml/file-systems.ntfs.devel/2005-08/msg00224.html > > "BTW2, I'm afraid [lib]parted can still corrupt partition > > tables if the number of heads are 240 (mainly laptops). > > When kernel 2.6 was released, it strictly followed the > Extended Interrupt 13h Disk Services specification. This > includes LBA and other geometry standards, including support > for heads of 255+ (for ATA 48-bit sectors), as well as 16 or > 32 (some NAND EEPROM options). > > The "240" issue comes from _buggy_ BIOSes, namely those > of IBM-Levono: > > Return 5-bit/16 heads (10000), ?lower 3 bits? from > sectors (63 = 111111) = an erroneous 240 (11110000) > heads as the 8-bit/0-255 heads value
I've messed with disk drives since dropping one on your foot would have broken it. I've also messed with partitioning hard drives and booting multiple OSes on PC drives since OS/2 was pretty new. The cylinder/head/sector limitations in MS DOS represented the same kind of thinking as that of the MS founder who said that 640K should be enough for anyone. I've used Partition Magic fdisk, parted and assorted other techniques for copying and resizing partitions over many years. I've seen disks with a jumper to set the number of heads to 15 instead of 16, but never had to use that jumper. I've also owned and used a number of IBM and Lenovo systems (disclaimer I happen to work for IBM) since the PC was introduced and I can't say I've ever run into a problem partitioning with an IBM/Lenovo BIOS, other than one tool or another reporting that some partition doesn't end on a cylinder boundary. Since 'cylinders' are now a totally mythincal beast, I'm not sure this matters. > > Since they have left it that way for so long, it's almost > a legacy requirement they "leave it buggy" for > compatibility with their own tools. > > > They could perhaps test and definitely keep in mind this > > (the problem manifests only in some rare case)." > > The problem is that following the Extended Interrupt 13h > Disk Services specifications is in conflict with much of > this. > > Personally, for maximum compatibility across all PC/OSes > (except DOS 7/Win9x, which absolutely requires the BIOS > and disk label/partition table geometry to match), I > first fire off this: > > # fdisk -H 255 -S 63 /dev/sda > > I initialize any new partition table with 255/63. That > is what Microsoft actually does when it sees geometry issues > as well in NT4SP4+ too, so it's a good one to follow (to > follow their assumptions). In fact, it's because parted > did not always "look" to check that Windows might have forced > the 255/63 geometry on a "buggy" 240/63 that often caused the > issue. > > Unfortunately it's not ideal going forward, although you can > use GPT and other things. > > Frankly, the sooner we get to EFI, the better. Unfortunately > Microsoft keeps dragging their feet on universal support of > EFI, even though Linux has had 32-bit and 64-bit EFI support > for some time (at least Apple uses 32-bit EFI). > > The nice thing about fdisk, versus parted, is that it does > _not_ initial any slices/volumes/filesystems. Parted will > go off and do all sorts of things for you. So if I don't > like how the geometry ends up, I can always quite without > saving with fdisk. With parted, by default, it's writing > everytime you change something. > > -- Bryan > > P.S. Yes, I'm still around. ;) Now THAT's the best news in this post. :-) Ian Shields Ph.D. Linux Technologist, ISV & Developer Relations IBM Corp Research Triangle Park, NC [EMAIL PROTECTED]
_______________________________________________ lpi-examdev mailing list [email protected] http://list.lpi.org/cgi-bin/mailman/listinfo/lpi-examdev
