Bug 305095

I think in the meantime I could further narrow down the problem.

Since the partitions I need to use for installing openSuse almost all have partition numbers >15, the installer informed me :-) about the new libata and its ridiculous low count of only 15 partitions it can handle. So I got advised, to use the kernel argument "hwprobe=-modules.pata" to fall back to the previous type, which can handle many more partitions.

Apparently this very kernel argument now is causing trouble when it comes to Configuring Installation Packages (or similar), right after the NIC page and the selection of the Installation Type page.

For some reason, the installer now seems to lose its ability to still read from the CD drive with the installation CD during the process.

So that is where the trouble seems to be rooted.

So, now I either can't install because of lack of support for more than 15 partitions or I cannot install, because the installer loses access to the installation CD :-(

Beta2 worked fine for / target on hda22 for me using brokenmodules=ata_piix rather than hwprobe=-modules.pata. Where did you find instructions to use the latter?

I'm using: http://download.opensuse.org/distribution/10.3-Beta2/iso/cd/openSUSE-10.3-Beta2-GNOME-i386.iso

My initial start screen reads:

F2: Language: Deutsch | English
F3: Video Mode: 1280 × 1024
F4: Source: CD-Rom
F5: Kernel: Default
F6: Driver: No

=> Installation, no kernel parameters

-Skip Media Check (why don't have I here an Alt-F4 console anymore?)
- Accept license
- Now at the next page "System Probing" I have a popup Info window saying:

"Your disk /dev/sda" <well that's odd. I definitively have a pATA HDD> "contains 17 partitions. The maximum number of partitions that the kernel driver of the disk can handle is 15. Partitions numbered above 15 cannot be accessed." <What an insane blunder!! Wasn't the previous limit some 63 partitions?>
"openSuSE is switching to the new IDE drivers using the libata modules. These do only support partitions" <?? guess that should be drives> " with up to 15 partitions. 
You haave the following options with openSuSE 10.3:
- Use the old IDE drivers: Boot again and add 'hwprobe=-modules.pata' as argument to the kernel
- Repartition your system ......
- Use LVM/EMS since they can provide an arbitrary and flexible number of block deevices partitions. This needs a repartition as well"

Does this coexist peacefully with eCS/OS/2 systems using IBM's LVM?

Note: <...> are my comments

Which IBM LVM? My Factory is on /dev/hda22, with eCS 2.0Beta on HPFS /dev/hda24 and eCS 1.1.4 on HPFS /dev/hda10 on an AMD/NForce2 system. Try again using brokenmodules=ata_piix rather than hwprobe=-modules.pata.

Then LVM from MCP1|2, respectively eCS 1.2/2.0RC1

I'm currently on it. So far, I can tell, that the warning about exceeding the 15 supported partitions still comes up. I ignored it and moved on now to "User Defined Partitioning (Experts)", "Preparing HDD: expert mode" <those terms might be not quite correct. My language setting is German>

Well, I only get to see 14 real partitions: /dev/sda1.../sda3 and /sda5.../sda15.
/sda1, even though at the very end of my HDD, is the eCS/OS/2 LVM BM, /sda6.../sda11 are two OS/2 Warp4 HPFS and the rest eCS 2.0xx JFS partitions, also on an Athlon-XP/nForce2 system.

The ones > /sda15, created by IBM's LVM and intended for the current installation, aren't displayed here.

Are you using the Linux LVM additionally?

I erred in comment 2. I used the KDE Beta1 iso to install to my AMD/NForce2 system, then upgraded to Beta2 using a Factory mirror. I did not burn that CD, but rather placed it on a partition the installer could find, and used Grub to load copies of the initrd and kernel from it. Using Gnome rather than KDE shouldn't have made a difference for this problem, but maybe something changed for the worse in Beta2. According to my menu.lst, and not my recollection, I did not use brokenmodules=ata_piix, but rather as you did, hwprobe=-modules.pata, which made all partitions /dev/hdxxx, not /dev/sd*. OTOH, my other Factory system was installed from the same iso several days later, and subsequently upgraded Factory/Beta2 using command line Smart, but did according to menu.lst on its (Debian) /boot use brokenmodules=ata_piix to wind up with no /dev/sda*.

As the installer's partitioner is showing you sd instead of hd, maybe something else is going on on your system. Are you sure you typed brokenmodules=ata_piix and hwprobe=-modules.pata correctly? Are there any SATA controllers in your system?

One other difference is all my partitioning is done exclusively using DFSee, usually the OS/2 version, but often the DOS from CD version.

Can you run Knoppix 5.x and access your PATA as /dev/hd*?

Following your advice in comment #4, I only applied brokenmodules=ata_piix, but not also in addition hwprobe=-modules.pata the last time.

Now with 'hwprobe=-modules.pata' additionally being back in the game, the "Insert openSUSE 10.3 CD 1" blocker now is back.

Well, I remain being stuck. There is nothing left for me to do but wait until a fix or workaround becomes available, to reliable avoid that unfortunate new libata module with its ridiculous low partition count restriction.

And yes, I do have a SiI3112A sATA controller on my mobo.
I create and maintain all partitions with the eCS LVM.

I still had a Knoppix 5.0 laying around here and with it, I get to access *h*da1...17 and also can browse with Konquerer through the contents of my former /home (/dev/hda17). 

(In reply to comment #7 from Frustriert Frustriert)
> Now with 'hwprobe=-modules.pata' additionally being back in the game, the
> "Insert openSUSE 10.3 CD 1" blocker now is back.
 
Try installing using an installation method other than from DVD/CD and see if you get around that block. e.g. - from HD. http://en.opensuse.org/Installation

I've submitted a new kpartx rpm which includes a script called 'activate_dm_linear'. It takes the device name as an argument and will put a device-mapper table on top of the given disk. This will invoke kpartx which in turn will generate device-mapper tables for all partitions on that drive. You should then be able to install on any of these partitions using the persistent device names for it.

Steffen, can you please include '/sbin/activate_dm_linear' from the kpartx rpm into the installation image?

added

I just tried to install the latest openSUSE-10.3-Beta3-GNOME-i386.iso, but unfortunately, I find myself in exactly the same situation as with the previous Beta2. 
Not using any kernel arguments, I get the message again about the only 15 partitions the kernel can handle.
Rebooting and using "hwprobe=-modules.pata", I find myself stuck once more at "Systemanalyse -> Initialisierung... -> Initialisiere Paket-Manager..."when the installer yet again loses touch with the ATAPI CD drive containing the installation CD.
Whatever the patch was supposed to do, doesn't work for me in Beta3.

(In reply to comment #11 from Frustriert Frustriert)

I have the exact same problem.

With no parameters I'm asked to reboot with hwprobe=-modules.pata". If I do, I'm asked for CD 1, and I'm using the DVD, in a loop. If I cancel, installation exits to text installation screen.

My system is fully PATA, the mobo doesn't have any SATA support. I have more than 16 partitions, and can not rearrange to less.

I'm afraid I reported it again as Bug #309070 following advise in the factory list to report the problem, didn't know it was already reported.

There was a problem with accessing the CD media.  This is fixed now (don't know the exact bugnumber).  I have setup a system with 22 partitions myself now and will check that installation and update works fine on it with RC1 and the goldmaster (using "hwprobe=-modules.pata").  Yesterday's test with the current internal version worked fine for me.

A general solution for supporting more than 15 partitions should get investigated but this might need to be driven by the upstream Linux kernel community.

Note that this hits all libata devices - including new SATA systems.

Carlos, sorry for asking you to report a new bug - but thanks for your insistance so that I checked it myself on my testmachine.

Hannes, I guess this can be closed since the CD installation works now.

The general problem with more than 15 partitions might be a separate bug that we should consider for the next release.

(In reply to comment #13 from Andreas Jaeger)

Thank you, Andreas. 

I'll test the next beta/rc when it comes, then.

And I'll wait eagerly for a definitive solution to the 15 partition limit. 


I suppose we are a small percentage of users having this situation, but there are some out there, and fewer of them are aware of the situation and speak out. I hope they come out now!

I suppose scsi HD tend to be small and expensive, so they get fewer partitions. But ata are bigger, and the availability of that partitioning space allows for more partitions. Having now to limit to 15 is a regression for us. I hope a good solution for all can be found. (I heard they wanted to push the limit to millions of scsi devices, too).

(In reply to comment #13 from Andreas Jaeger)
> The general problem with more than 15 partitions might be a separate bug that
> we should consider for the next release.
 
You're not forgetting bug 218122 are you?

(In reply to comment #13 from Andreas Jaeger)

Do I understand this correctly so far, that with RC1 an installation from ATAPI CD now should be possible, if more than 15 partitions exist, using the advised kernel argument "hwprobe=-modules.pata", but it still only works, if no partitions numbered higher than 15 are being used?

Are there any daily/nightly CD ISO builds available for testing prior to RC1?

I can't help myself, but how could the people in (Linux) charge agree to make a mandatory switch to libata, when it turns out to be such a huge step backwards and comes with such severe restrictions in the number of usable HD partitions?

An installation with RC1 from ATAPI CD with a PATA disk using more than 15 partitions will work if the argument "hwprobe=-modules.pata" is used.  There are no restrictions.  You will be able to use partitions higher than 15.

If you do *not* use "hwprobe=-modules.pata", you can only use the first 15 partitions.

We're not syncing out our internal ISOs to reduce bandwidth - I tested yesterday with one here and will test today again.

Felix, thanks for reminding be about 218122.  I reopened it now.

Karl, can you document this in the release notes?

(In reply to comment #17 from Andreas Jaeger)

Andreas, if it's not asked for too much, during your RC1 tests, could you please also try specifically to install GRUB on a JFs and/or ReiserFS partition (should be failing for both, / and a separate /boot)?

Still being a very beginner with Linux, based on my very recent starter's experience with Ubuntu 7.04 and 7.10 tribe-X and Debian Etch, I already had to learn, that partitions >15 aren't getting mounted and that apparently in addition, grub-install has a severe bug, preventing its very installation, if the targeted FS isn't ext2|3. 

This, in addition to the partition >15 issue ultimately brought me to try openSuSE, but since so far the installation also failed me due to this bug, I don't know yet about how the rest would turn out.

But I reckon, grub-install only working on ext2|3 FS should also be severe enough tor SUSE/Novell to further investigate, unless you are not utilising it at all, that is.
If desired, I can provide further details and the related Ubuntu bug-#.

Did you file bugreports for these issues?  Please do so separately.

I'll do testinstall with different filesystem starting with ReiserFS.  Btw. JFS is really unsupported.

ReiserFS works fine for me.

Is it enough to re-word the old entry about libata as follows:

  <!-- FATE 301746 and bug 305095 -->
  <sect3 id="libata" status="2007-09-14">
   <title>libata for IDE Devices</title>

   <para>
   libata uses <filename>/dev/sda</filename> for the first harddisk instead of
   <filename>/dev/hda</filename>.  Disks with more than 15 partitions are not
   handled right now.
   To use the old scheme, boot with the following kernel parameter:</para>

   <screen
>hwprobe=-modules.pata</screen>
  </sect3>

  <para>Then you will be able to see partitions > 15 again and can access them
  for installation.</para>
 </sect2>

Hannes, please add further information.

<para>
   libata uses <filename>/dev/sda</filename> for the first harddisk instead of
   <filename>/dev/hda</filename>.  Disks with more than 15 partitions are not
   handled automatically right now.
   To enable support for more than 15 partitions, please switch to a virtual
   console before the partitioner starts and issue the command:

   <screen>activate_dm_linear /dev/sdX</screen>

   where <filename>/dev/sdX</filename> has to be replaced with the device
   node of the disk in question.
   This command will create device-mapper devices for the disk and each
   partition, so the system can installed on the device-mapper devices.

   Alternatively you can disable libata support by booting with the
   following kernel parameters:</para>
  

(In reply to comment #25 from Andreas Jaeger)
Having been a very Linux newbie until just a few weeks ago, I would like to suggest to a add the hotkeys, which allow switching to a virtual console and back, in the second paragraph. A beginner can type in what is suggested to him, but he most definitely doesn't know yet, how to get to and back from another console yet.

(In reply to comment #21 from Andreas Jaeger)
No, not here, since so far I couldn't end up with a working installation yet, to see how openSuSE is behaving. With Etch f.e., I had to fall back to LILO, in order to get a Linux bootloader installed at all.
I chose (IBM'S) JFS because of the compatibility with my existing main OS OS/2, respectively eCS.
This a bug I was pointed to for Debian: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=239111
and these are the ones I know of for Ubuntu:
https://bugs.launchpad.net/ubuntu/+source/ubiquity/+bug/123102
https://bugs.launchpad.net/ubuntu/+source/grub-installer/+bug/14010

I changed the RN text accordingly in SVN.

RNs submitted to the build.

Re-assigning to aj for further tracking or closing.

I doubt that activate_dm_linear works at all.  Let me test it...

release notes only -> no blocker

I just tested - and the activate_dm_linear support does not work as documented.
This needs some more work for 11.0.

Karl, please use again the text from comment #23.

  <!-- FATE 301746 and bug 305095 -->
  <sect3 id="libata" status="2007-09-18">
   <title>libata for IDE Devices</title>

   <para>
    libata uses <filename>/dev/sda</filename> for the first harddisk instead of
    <filename>/dev/hda</filename>.  Disks with more than 15 partitions are not
    handled automatically right now.  You can disable libata support by booting
    with the following kernel parameter:</para>

   <screen
    >hwprobe=-modules.pata</screen>

   <para>Then you see all the partitions > 15 again and can access them for
    installation.</para>
  </sect3>

Submitted to /work .