Date: Sun, 17 Oct 1999 11:27:54 -0700 From: "Ronald F. Guilmette" <rfg@monkeys.com> To: Mark Ovens <mark@dogma.freebsd-uk.eu.org> Cc: freebsd-questions@freebsd.org Subject: Re: Some serious gripes about `fdisk' and also `booteasy'. Message-ID: <24065.940184874@monkeys.com> In-Reply-To: Your message of Sun, 17 Oct 1999 09:47:27 %2B0100. <19991017094727.B319@marder-1>
next in thread | previous in thread | raw e-mail | index | archive | help
In message <19991017094727.B319@marder-1>, you wrote: >On Sat, Oct 16, 1999 at 08:35:11PM -0700, Annelise Anderson wrote: >> >> >> On Sat, 16 Oct 1999, Ronald F. Guilmette wrote: >> >> > >> > This had bitten me twice now. >> > >> > Will whoever is supporting/maintaining the `fdisk' program *PLEASE* fix it >> > so that it displays some sort of a warning when the user is just about to >> > partition a SCSI drive for which the on/off status of BIOS address trans- >> > lation (for disks larger than 1GB) _cannot_ be automatically and properly >> > determined by fdisk itself? >> >> One piece of advice often given is to put a dos partition on the disk >> (it can be deleted during install if wish) to assist with determining >> the correct geometry. I've installed FreeBSD on probably a close to >> a dozen scsi hard drives and never had this problem. I didn't think >> bios address translation was relevant for scsi drives, in any case. >> > >Neither did I. That's the main reason I moved to all-SCSI, it "just >works", especially with *nix. None of these "only 4 devices", "1024 >cyls/504MB/LBA", "can't have a CD as primary slave with a disk on >secondary master" problems that IDE seems to be plagued with. I'd >be interested in knowing the "truth" about this for future reference. Just for your future reference, the "truth" about this is that yes, the backwards (harware) compatability constraints associated with PeeCee hardware... in which they all _still_ try to remain somewhat compatible with the very earliest IBM PeeCees that supported disk drives... DOES indeed still make us all play funny games when trying to address specific disk sectors, regardless of whether you have an (E)IDE drive or a SCSI drive. The attempts to be backward compatible mean that in the address that gets sent to the disk drive (in the form cylinder/head/sector) has to get munged around in weird ways, in particular for any disk that has more than 1 gigabyte capacity. Again, this bit of screwyness affects _both_ (E)IDE drives and also SCSI drives. But in the case of (E)IDE drives, the software driver can (I think) always correctly figure out the proper set of rules it has to use when forming the cylinder/head/sector disk address. But in the case of SCSI drives, it can't always do that. Currently, unless you have some extra hints (e.g. a DOS partition) already lying around that the FreeBSD `fdisk' program can use to get a clue as to the current ``disk geometry'' (i.e. the rules for forming a cyl/head/sector address from a raw sector address) the `fdisk' program is likely to guess wrong about your disk geometry, at least if you are using a SCSI disk. In fact, now that a majority of disk drives have capacities in excess of 1 GB, the `fdisk' program will *mostly* guess wrong. And when it does, you *must* manually override its guess or else you will end up with a partition that you cannot actually use for anything. (And unless I'm mistaken, you _may_ even end up improperly overwriting parts of _other_ existing partitions on that same SCSI disk... which would be really Bad News if you had other partitions on there that contained stuff that you wanted to keep.) Obviously this is a dangerous situation. Equally obvious (to me at least) is that the FreeBSD `fdisk' program should start by checking to see if it is running on some sort of x86 (PeeCee) type system. If it is, then it should proceed to try to determine what sort of disk drive is being partitioned, SCSI or (E)IDE. (In fact, its possible that the first of these checks isn't even necessary because, unless I'm mistaken, `fdisk' is _only_ used on PeeCee type systems, and even on those, it is only used in cases where FreeBSD isn't being given the whole disk to itself, but is instead being asked to ``co-reside'' on the disk with 1, 2, or 3 other traditional PeeCee (BIOS level) partitions.) Anyway, if `fdisk' is running on a PeeCee type system, and if it is dealing with a SCSI drive that may or may not have the so-called ``BIOS address translation'' feature endabled in the SCSI controller card, then it SHOULD NOT just guess about the disk geometry... because it can easily guess wrong, with potentially tragic results. Instead, it should have two mutually- exclusive command line options like: --translation and: --no-translation and if the user supplies neither of those, *and* if `fdisk' is unable to find any useful ``hints'' (e.g. an existing DOS partition) lying around already on the disk, then upon startup it should *prompt* the user with a question like: Is BIOS address translation enabled on your SCSI controller card for this specific SCSI disk drive (y/n)? I hope that this change will get made in the FreeBSD `fdisk' program. That will make it a lot safer to use. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?24065.940184874>