Fw: Re: Fw: medley raid problem
Heinz Mauelshagen
mauelshagen at redhat.com
Thu Nov 3 11:31:37 UTC 2005
James,
it works but it breaks my design goal to avoid hw ties in dmraid code :(
Can you send me your metadata dd'ed into files hd[efg].dat, so that I can try
figure another PCI* dependency avoiding approach ?
Thanks,
Heinz
On Sun, Oct 30, 2005 at 01:15:58AM -0500, James Olson wrote:
> Hi Molle,
> Thanks for your help. Anyway, the hpa message turned out to be a red herring. The problem was actually in the detection of the magic number in the metadata. From sil.c the magic number is defined:
> #define SIL_MAGIC 0x2F000000
> and on my 3 drives I read:
> magic: hdf 0x0F000000 disk_number: 2
> magic: hdg 0x2F000000 disk_number: 1
> magic: hde 0x0B000000 disk_number: 0
>
> so it is finding the metadata but the routine:
> return sil->magic == SIL_MAGIC && sil->disk_number < 8;
> fails for hdf and hde.
>
> I wasn't sure how to fix the magic number detection so I wrote a patch to use the PCI vendor and product ID instead, like they do in the working ataraid, medley 2.4 kernel modules, and it works! :
>
> --- dmraid.orig/dmraid/1.0.0.rc9/lib/format/ataraid/sil.c 2005-09-22 06:09:02.000000000 -0700
> +++ dmraid/1.0.0.rc9/lib/format/ataraid/sil.c 2005-10-29 22:50:32.000000000 -0700
> @@ -128,7 +128,11 @@
>
> static int is_sil(struct sil *sil)
> {
> - return sil->magic == SIL_MAGIC && sil->disk_number < 8;
> +#define SIL_680_VENDOR_ID 0x1095
> +#define SIL_680_PRODUCT_ID 0x0680
> + if ((sil->vendor_id == SIL_680_VENDOR_ID) && (sil->product_id == SIL_680_PRODUCT_ID))
> + return 1;
> + else return 0;
> }
>
> static int sil_valid(struct lib_context *lc, struct dev_info *di,
>
> ----- Original Message -----
> From: "Molle Bestefich" <molle.bestefich at gmail.com>
> To: "ATARAID (eg, Promise Fasttrak, Highpoint 370) related discussions" <ataraid-list at redhat.com>
> Subject: Re: Fw: Re: Fw: medley raid problem
> Date: Sat, 29 Oct 2005 22:54:28 +0200
>
> >
> > James Olson wrote:
> > > 2.4 kernel dmesg log:
> > > hde: 6835952 sectors (3500 MB), CHS=6781/16/63, UDMA(33)
> > > hdf: 6346368 sectors (3249 MB) w/96KiB Cache, CHS=6296/16/63, DMA
> > > hdg: host protected area => 1
> > > hdg: 6346368 sectors (3249 MB) w/256KiB Cache, CHS=6296/16/63, UDMA(33)
> >
> > Too bad it doesn't tell you how large the HPA is, otherwise you could
> > subtract that from the total size and look for metadata at the
> > resulting sector...
> >
> > > Here is the 2.6 log:
> > > hde: FUJITSU MPA3035ATU, ATA DISK drive
> > > hdf: IBM-DAQA-33240, ATA DISK drive
> > > hdg: WDC AC23200L, ATA DISK drive
> > > hde: 6835952 sectors (3500 MB), CHS=6781/16/63, UDMA(33)
> > > hdf: 6346368 sectors (3249 MB) w/96KiB Cache, CHS=6296/16/63, DMA
> > > hdg: 6346368 sectors (3249 MB) w/256KiB Cache, CHS=6296/16/63, UDMA(33)
> >
> > I've seen 2.6 versions that do tell about HPA areas, I find it suspect
> > that yours doesn't.
> > Perhaps it's a bug (but it really does disable the HPA), perhaps there
> > really is no HPA or perhaps the information has been intentionally
> > removed from newer 2.6 kernels.
> >
> > Could you compare the number of sectors that Linux tells you above to
> > the number reported by your BIOS?
> >
> > Just to get you up to speed on HPA:
> > There was some discussion on linux-kernel, and the powers that be
> > (Alan Cox) does not want to change the kernel's current faulty
> > behaviour of automatically disabling the HPA, since that would break
> > things for the few people that has partitions which span the HPA if
> > they upgrade the kernel and do not manually enable the "HPA disable"
> > thing. Most of those people supposedly have BIOSs where they can
> > disable HPA the proper way (the ThinkPad users do, at least), but that
> > argument didn't seem to stick.
> >
> > The only way forward right now that would pass the Cox barrier is a
> > patch that change all the partition detection code to:
> > * Know about HPA/non-HPA size of physical devices.
> > * Detect partitions that are out-of-bounds of the non-HPA size.
> > * Call IDE code to disable HPA when such partitions are found.
> >
> > Or patches that:
> > * Change all user code to be aware of HPA versus non-HPA IDE disks.
> >
> > Both seems overly complex solutions compared to:
> > * Fixing the kernel to behave itself and not disable HPA.
> > * Provide a kernel command line flag or configuration option for
> > those who need to disable HPA manually because they have a horrible
> > BIOS.
> >
> > But that's how things are right now.
> >
> > > Also there is some question of the drive geometry between the different kernels.
> > > The hdparm -g outputs differ.
> >
> > Maybe 2.4 does not disable HPA?
> > 2.6 does (my best guess is that it does so because some IDE developer
> > had a laptop which didn't allow him to disable HPA).
> >
> > Seems odd though, since the two logs you've given indicate that both
> > kernels see the same number of sectors. What sector count do you get
> > if you run fdisk on the two kernels?
> >
> > Blah blah. Sorry for the lengthy mail.
> >
> > _______________________________________________
> > Ataraid-list mailing list
> > Ataraid-list at redhat.com
> > https://www.redhat.com/mailman/listinfo/ataraid-list
>
>
> --
> _______________________________________________
>
> Search for businesses by name, location, or phone number. -Lycos Yellow Pages
>
> http://r.lycos.com/r/yp_emailfooter/http://yellowpages.lycos.com/default.asp?SRC=lycos10
>
>
> _______________________________________________
> Ataraid-list mailing list
> Ataraid-list at redhat.com
> https://www.redhat.com/mailman/listinfo/ataraid-list
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Heinz Mauelshagen Red Hat GmbH
Consulting Development Engineer Am Sonnenhang 11
Cluster and Storage Development 56242 Marienrachdorf
Germany
Mauelshagen at RedHat.com +49 2626 141200
FAX 924446
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
More information about the Ataraid-list
mailing list