-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Thursday 2007-08-23 at 22:34 +0200, Marcus Meissner wrote:

It is a feature. hda is ... mostly history.

We now use libata for IDE drives which is a SCSI based library.

Is there any hope of increasing the 16 partition limit some time in the future? Alternatively, will the ata support remain for ever, with maintenance? I have disks with 20 partitions. I suspect I'm not the only one using more than 16 partitions. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGzseStTMYHG2NR9URAu8zAJ9k+JF8utZYN/BPnZ8AIzb2Bi2AyACfWPq3 QVGgQDZbvLSTKEk2tng/TaY= =nKDL -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 14:41 +0200, Andreas Jaeger wrote:

...

Is there any hope of increasing the 16 partition limit some time in the future?

No - but we added a workaround for this. If you have more than 16 partitions, just give it a try!

...

Alternatively, will the ata support remain for ever, with maintenance?

I guess for 10.4, we will remove it.

...

I have disks with 20 partitions. I suspect I'm not the only one using more than 16 partitions.

So, please test whether it our workaround makes you happy ;-)

I'll try it, but unfortunately, I don't think I'll be able to till mid September, at least. Holiday season means intermittent PC access for me. Question: Can 10.2 coexist in the same PC with 10.3 beta? Regarding this problem, I mean. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGztdJtTMYHG2NR9URAiwRAJwJnNUqmxlOdtLg0uPN0qWYuyE78ACeJeJh A/g5kCx+v40JfB340opeC0c= =J2AV -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On 2007/08/24 15:18 (GMT0200) Andreas Jaeger apparently typed:

"Carlos E. R." writes:

...

Question: Can 10.2 coexist in the same PC with 10.3 beta? Regarding this problem, I mean.

It should - Hannes, can you comment?

Dunno why Carlos should ask. I have 4-5 systems with both Factory and 10.2, 10.1 and/or 10.0, along with several other distros, and windoz, and OS/2, and DOS, and more than 15 partitions per PATA HD, each. As long as partition tables remain standard, no OS depends on how any other is configured. -- " It is impossible to rightly govern the world without God and the Bible." George Washington Team OS/2 ** Reg. Linux User #211409 Felix Miata *** http://mrmazda.no-ip.com/ -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

On 2007/08/24 20:46 (GMT+0200) Carlos E. R. apparently typed:

Friday 2007-08-24 at 13:51 -0400, Felix Miata wrote:

...

On 2007/08/24 15:18 (GMT0200) Andreas Jaeger apparently typed:

...

...

"Carlos E. R." writes:

...

...

...

Question: Can 10.2 coexist in the same PC with 10.3 beta? Regarding this problem, I mean.

...

...

It should - Hannes, can you comment?

...

Dunno why Carlos should ask. I have 4-5 systems with both Factory and 10.2, 10.1 and/or 10.0, along with several other distros, and windoz, and OS/2, and DOS, and more than 15 partitions per PATA HD, each. As long as partition tables remain standard, no OS depends on how any other is configured.

I ask because I don't know and I'm afraid to try. I have no idea what will happen to a disk with 18 partitions when the system does not see the last two. Will it think that is empty space?

Using brokenmodules=ata_piix to install causes traditional drivers to be used instead of libata, which means everything works as before. If Andreas is referring to some other kind of workaround, then I don't know. Compare to Fedora 7, where its installer is incapable of proceeding beyond the partitioning phase on a system with a disk having more than 15 partitions, unless it is allowed to erase all current partitioning.

Are scsi partitions defined differently, assigning a nibble (4 bits) to the partition number, and so when it finds a partition table needing a byte will everything in the table be pushed over another 4 bits? Because the only reason I can think of limiting to 16 items is because a record somewhere is limited to 4 bits.

IIUI, the SCSI major of 8 uses too many bits, such that insufficent bits remain to count past 15, a combined record limit of one byte.

Will it overwrite the grub code with references to sda somehow, so that later my main system wanting to use hda will fail?

AFAIK, Grub's only installation facility limitation here is the existence of a device. If the system can't provide one for a partition >15, Grub can't install to it. As long as you have something booted that can provide access to the device (pre-libata, a pre-10.2 live CD, Knoppix, etc.), then you can use that to install Grub, presuming there's any reason to install it again on a system that already has it installed somewhere. I keep generic MBR code installed, and only put Grub on partitions. I have one partition per system I use as /boot for one distro, either as an active primary partition, or as the first logical, to which I chainload from another boot loader. I avoid letting any installer or installed OS mess with the grub directory on that partition. Additional installations get their boot loader installed to their /, and I either chainload from my primary loader, or use my /boot partition's Grub installation to load its kernel and initrd directly. IOW, there's no good reason to have an additional installation write to the MBR, or to any existing partition that it isn't told to use.

Notice that the same grub instance will have to boot both 10.2 and 10.3 (or rather, the grub in the mbr will have to start another mbr for the other system). Ie, a 10.2 grub in the mbr will start one of two other grubs in different boot partitions, each starting 10.2 (main) or 10.3 (beta).

Use 10.2 as your primary boot loader. Install 10.3's Grub on its own /, and use 10.2 to chainload to 10.3. Optionally, have 10.3 install no boot loader at all, and just load the 10.3 kernel and initrd directly from 10.2's Grub.

You see, I'm completely ignorant on this issue, and because I'm ignorant I'm afraid. Just the same as a village in my country have rejected the installation of a new cellular phone aerial because they think it is dangerous and "radioactive" (and thus, they have no phone service at all). Ignorance breeds fear. and I'm afraid about this new scsi issue.

It's easier with spare systems and/or spare HDs to experiment on. I have many, and do a lot of experimenting. -- " It is impossible to rightly govern the world without God and the Bible." George Washington Team OS/2 ** Reg. Linux User #211409 Felix Miata *** http://mrmazda.no-ip.com/ -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 13:44 -0400, Felix Miata wrote:

On 2007/08/24 14:41 (GMT+0200) Andreas Jaeger apparently typed:

...

"Carlos E. R." writes:

...

...

Is there any hope of increasing the 16 partition limit some time in the future?

...

No -

Because it's a kernel design issue, something SUSE/Novell has little or no control over.

Well, they can assign developers to scsi and develop patches that add support for more than 16 partitions to scsi modules. If technically possible, of course (I don't know what is the origin of that limit). Or invent something different than scsi for both scsi and ata.

...

but we added a workaround for this. If you have more than 16 partitions, just give it a try!

brokenmodules=ata_piix? If not, what, and where documented?

...

...

Alternatively, will the ata support remain for ever, with maintenance?

...

I guess for 10.4, we will remove it.

Sounds ominous.

Quite so!

...

...

I have disks with 20 partitions. I suspect I'm not the only one using more than 16 partitions.

...

So, please test whether it our workaround makes you happy ;-)

I have disks approaching the 63 PATA partition count limit.

And I thought I had a lot... :-o

A related issue is optional mounts. When exist upwards of 30 lines in fstab, with labels or worse for partition identification, it's a new and large headache to figure out what to type to mount partitions that are noauto in fstab. When using devicenames of 4-5 characters this was minor thing to remember.

Could you give a sample? I don't understand the issue (I do have long fstab files) - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGzyD0tTMYHG2NR9URAlkrAJ4p/0Sz1+T+wdul9vSscM/W9sU+UACghE7/ Y2ftshrTxF9ViOtC7MeO0v8= =oSq+ -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 15:04 -0400, Patrick Shanahan wrote:

* Carlos E. R. [08-24-07 14:20]:

...

Well, they can assign developers to scsi and develop patches that add support for more than 16 partitions to scsi modules. If technically possible, of course (I don't know what is the origin of that limit). Or invent something different than scsi for both scsi and ata.

iirc, original scsi design had 8 channels and a later extended version had 16, but you were only able to utilize 7/15 of those as the scsi system needed the 0 channel for itself.

But... aren't channels the devices connected to the same cable (bus)? Ugh, just looked it up in the wikipedia and there is no mention of what "channels" are. I can see a "reasonable reason" to limit devices to 8 or 16 (limit the number of cables), but not for the number of partitions inside a device. For small disks, maybe, but with the size of disks nowdays, 16 partitions is absurd.

BUT, I *am* approaching the age where facts may not necessarily be facts anymore :^)

Heh! - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGzzPqtTMYHG2NR9URAmIYAKCYP7VUjfWvJAYyKcLcZdk18Dul3gCfYSL8 TNqs8emDuKTeLKLYHNRCkSY= =jxyL -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 17:07 -0400, Patrick Shanahan wrote:

* Carlos E. R. [08-24-07 15:42]: [...]

...

But... aren't channels the devices connected to the same cable (bus)? Ugh, just looked it up in the wikipedia and there is no mention of what "channels" are.

as usual, this came to mind after posting. The moment of greatest revelation :^) This is *probably* correct, your understanding. It has been some time since I spent much time/money on scsi. As I recall it was when scanning required scsi or a proprietary card, I had an HP 4c and paid ~us$800 for a discounted item.

Me too... but my boss paid, not me. And those scsi cards only worked for one "thing", very often you could not plug a disk or another scanner to the card. Time changes! I'm grateful for usb over this one, at least ;-)

...

I can see a "reasonable reason" to limit devices to 8 or 16 (limit the number of cables), but not for the number of partitions inside a device. For small disks, maybe, but with the size of disks nowdays, 16 partitions is absurd.

agreed.

It appears to be a software limit. And very difficult to overcome :-( - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGz10itTMYHG2NR9URAsUEAJ9dMCn09vlqw7Dd8tl0BvGHJJGkeACfQGKC 55E5FjETKUBEUBjwt+ybdFk= =/F0s -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 15:42 -0400, Felix Miata wrote:

On 2007/08/24 14:18 (GMT-0400) Carlos E. R. apparently typed:

...

Friday 2007-08-24 at 13:44 -0400, Felix Miata wrote:

...

...

A related issue is optional mounts. When exist upwards of 30 lines in fstab, with labels or worse for partition identification, it's a new and large headache to figure out what to type to mount partitions that are noauto in fstab. When using devicenames of 4-5 characters this was minor thing to remember.

...

Could you give a sample? I don't understand the issue (I do have long fstab files)

Consider all these fstab entries that might cause the exact same partition to mount to the exact same location (note that in Fedora 7, the first of the 5 is invalid-not using libata for PATA is not an option):

/dev/hda21 /disks/hda/fedora ext3 noauto,acl 0 0 LABEL=S12A-21fedora /disks/hda/fedora ext3 noauto,acl 0 0 UUID=7e303340-8ab6-4839-847e-260b9717c4a6 /disks/hda/fedora ext3 noauto,acl 0 0 ID=ata-ST3120814A_5LS0EPFV-part21 /disks/hda/fedora ext3 noauto,acl 0 0 PATH=pci-0000:00:1f.1-ide-0:0-part21 /disks/hda/fedora ext3 noauto,acl 0 0

Ah, I see the confusion. Well, /disks/hda/fedora refers to the same device and partition, but could not. Yes, a big source of problems. It would need an fstab parser/verification program. Not easy... I suppose the first match would mandate.

...

From CLI: 1-The old way, how fast could you mount the partition you remember is #21 on the first PATA disk? (e.g. mount /dev/hda21)

Well, I give mount points based on the intended usage of the partition, not based on the partition number. In the case above I would use "mount /disks/hda/fedora" and let the system figure! :-p

2-The libata way, how fast can you mount the partition you remember is #21 on the first PATA disk? How would you do it?

I dunno what is the libata way :-? Ah! Perhaps you mean that you wouldn't know ahead what name (sda, sdb, sdc...) it would have?

In addition, which one of those 5 possible entries to you prefer to have in your own (long) fstab(s)?

Label. If I give a partition a label it is because I want to use it. Otherwise, like for vfat partitions, I use the id: /dev/disk/by-id/usb-Kingston_DataTravelerCR_ad053f40c9f050-part1 /mnt/usb/Kingston vfat noauto,user,users,uid=cer,gid=users,fmask=0117,dm Mmmm...! Looking at your sample, I have discovered I could use "ID=/usb-Kingston_DataTravelerCR_ad053f40c9f050-part1" instead. A few letters shorter ;-) Perhaps I should use uuid, that seems to be not related to the partition number: it means that repartitioning doesn't affect it, same as a label. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGz07EtTMYHG2NR9URAiRGAJ9BDQsCZ9tzhU8mehvC9GlJa+9fDwCffpf3 AisDO/SzNL073QnUScQetmw= =DYWQ -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Friday 2007-08-24 at 16:31 -0400, Felix Miata wrote:

...

...

Because it's a kernel design issue, something SUSE/Novell has little or no control over.

...

Well, they can assign developers to scsi and develop patches that add support for more than 16 partitions to scsi modules. If technically possible, of course (I don't know what is the origin of that limit). Or invent something different than scsi for both scsi and ata.

This is nothing ordinary patching can solve. It's a fundamental kernel design problem. SCSI has always had the 15 partition limit because its device node major is 8. To change it to something less, which would be required to allow more than 15 SCSI partitions, would require major backward compatibility breakage in kernel design. Novell developers wouldn't likely try anything significant without major participation from the kernel development community and the Linux community generally.

Huh? Mmm? pata uses larger major numbers. What...? :-? [...] [reading] Ah, from what you said I looked at "/usr/src/linux/Documentation/devices.txt": 3 block First MFM, RLL and IDE hard disk/CD-ROM interface 0 = /dev/hda Master: whole disk (or CD-ROM) 64 = /dev/hdb Slave: whole disk (or CD-ROM) For partitions, add to the whole disk device number: 0 = /dev/hd? Whole disk 1 = /dev/hd?1 First partition 2 = /dev/hd?2 Second partition ... 63 = /dev/hd?63 63rd partition 22 block Second IDE hard disk/CD-ROM interface 0 = /dev/hdc Master: whole disk (or CD-ROM) 64 = /dev/hdd Slave: whole disk (or CD-ROM) 8 block SCSI disk devices (0-15) 0 = /dev/sda First SCSI disk whole disk 16 = /dev/sdb Second SCSI disk whole disk 32 = /dev/sdc Third SCSI disk whole disk ... 240 = /dev/sdp Sixteenth SCSI disk whole disk Partitions are handled in the same way as for IDE disks (see major number 3) except that the limit on partitions is 15. I think I understand. The minor device number is used to distinguish both the device (16 of them) and the partitions (16 of them). If I understand correctly, it is one byte. But pata distributes that byte differently, allowing for more partitions and less disks. Very curious! Indeed, a kernel design decision (or unix design?) having a long reach. Reminds me of the "1 megabyte is enough memory" decision :-p Still, I suppose they could use a different device name, like nsd (new s d), with a different number scheme (like the /dev/sg? devices). But that would be a wild speculation on my part.

Kernel people have said a solution should come from userspace, and I think this is where Novell developers are mostly spending some effort quietly. I found some also from Mandriva, while none from Fedora.

Let's hope! :-)

The problem as I see it is little or no use cases for systems the Distro makers are paid to support to require more than 15 partitions. Outside experimentation and development environments, I can't think of a use case for more than 15. This makes it hard for a developer to get paid to work on a solution, meaning fewer resources available to task this problem.

I understand. I don't really know about current linux enterprise usage, but I guess somebody should do a good, serious, poll. But I used a big unix machine, very serious (not a pc, but with scsi disks) that used several partitions. If there was one for the system, there would be another one replicating the system (kept umounted). Several programs had dedicated partitions (raw), including a database. Things that could grow had their own partition, like the logs, for instance. However, I don't remember how many partitions it had. Several, for sure. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFGz1t+tTMYHG2NR9URApDAAJ9FZWS67Bgi7dXHPJRwToR9Q3UbcwCfdkst cjnYcKPX1zW8LrrgXmaI9L4= =+mwo -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org