2008-06-14 18:48:16

by werner

[permalink] [raw]
Subject: Regressions in the last kernels

During preparation and own testing for my distro; installation on computers for neighbours, by reclamations by users, I learned about the following problems of 2.6.26-rcX :

1) When compiling for i486, one cannot select/configurate more memory than 4 GB, menuconfig don't show items with higher values. My own computer/server has 8 GB, but when I compile the kernel with 4 GB and i486 then 'free' shows only 3.2 GB (no shared grafics, the grafic card has its own memory) ...

2) On some laptops, especially IBM/Centrinho laptops, 2.6.26-rcX dont find the CD devices, nor that from what was booted. This problem is independend on the edd problem. It wasn't present at the 2.6.25-rc9 kernel or earliers (i didn't test it on 2.6.25 defin.), but it's present at 2.6.26-rc4 and -rc5 (-rc6 i didnt test). Just because of this problem it wasnt possible to install something on this kind laptop, and I had yesterday to make a new Install .iso with an older kernel as alternative: ftp://ftp.uni-siegen.de/pub/sys-linux/SYS_Linux-0.23-rc3.iso

Exactly happens the following: BIOS finds correctly the CD drive for booting and boots, but later Linux dont find it (only older kernels), on some laptops. I insert in the drive the Install DVD: It boots, loads the kernel and the initrd correctly. This is the small install / rescue system. However, although everything present in /dev, inclusive the device nodes /dev/hdb etc (but empty), Linux don't find /dev/hdb f.ex. for mount it or calling with hdparm or anyhow else. Thus, also the system installation fails, because the packages or lzma-zipped system cannot be readed from the same CD what was used for boot and what BIOS found/managed ... With exactly the same system/initrd, configuration etc everything the same but just change to an older kernel, everything works normally; thus this is a kernel regression

*** At this point, I want to suggest, that Linux should have a special devide node /dev/<anyname> on what one can mount ANYWAY and ANYHOW that boot device what BIOS used and understood to boot from, even when the hardware detection of Linux fails completely. When bios managed it to read that device and to load the kernel and the initrd , then Linux should manage it too, to read from the same CD other files. This is necessary in order to perform the installation anyway, even when Linux dont find hardware or the boot CD. ***


Werner Landgraf
werner#guyane.dyn-o-saur.com
[email protected]

Linux SYS
0.23-rc3


2008-06-14 20:24:38

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: Regressions in the last kernels

On Saturday, 14 of June 2008, werner wrote:
> During preparation and own testing for my distro; installation on computers
> for neighbours, by reclamations by users, I learned about the following
> problems of 2.6.26-rcX :
>
> 1) When compiling for i486, one cannot select/configurate more memory than
> 4 GB, menuconfig don't show items with higher values. My own computer/server
> has 8 GB, but when I compile the kernel with 4 GB and i486 then 'free' shows
> only 3.2 GB (no shared grafics, the grafic card has its own memory) ...

I've just created the Bugzilla entry at:
http://bugzilla.kernel.org/show_bug.cgi?id=10912
for this problem. Please add yourself to the CC list in there and please file
a separate bug for the CD issue.

> 2) On some laptops, especially IBM/Centrinho laptops, 2.6.26-rcX dont find
> the CD devices, nor that from what was booted. This problem is independend
> on the edd problem. It wasn't present at the 2.6.25-rc9 kernel or earliers
> (i didn't test it on 2.6.25 defin.), but it's present at 2.6.26-rc4 and -rc5
> (-rc6 i didnt test). Just because of this problem it wasnt possible to
> install something on this kind laptop, and I had yesterday to make a new
> Install .iso with an older kernel as alternative:
> ftp://ftp.uni-siegen.de/pub/sys-linux/SYS_Linux-0.23-rc3.iso
>
> Exactly happens the following: BIOS finds correctly the CD drive for
> booting and boots, but later Linux dont find it (only older kernels), on some
> laptops. I insert in the drive the Install DVD: It boots, loads
> the kernel and the initrd correctly. This is the small install / rescue
> system. However, although everything present in /dev, inclusive the device
> nodes /dev/hdb etc (but empty), Linux don't find /dev/hdb f.ex. for mount
> it or calling with hdparm or anyhow else. Thus, also the system installation
> fails, because the packages or lzma-zipped system cannot be readed from the
> same CD what was used for boot and what BIOS found/managed ... With
> exactly the same system/initrd, configuration etc everything the same but
> just change to an older kernel, everything works normally; thus this is a
> kernel regression

This looks like it might have been fixed already. Which -rc kernels have you
tried so far?

> *** At this point, I want to suggest, that Linux should have a special devide
> node /dev/<anyname> on what one can mount ANYWAY and ANYHOW that boot device
> what BIOS used and understood to boot from, even when the hardware detection
> of Linux fails completely. When bios managed it to read that device and to
> load the kernel and the initrd , then Linux should manage it too, to read
> from the same CD other files. This is necessary in order to perform the
> installation anyway, even when Linux dont find hardware or the boot CD. ***

You can use the /dev/disk/by-id/* things for that I think.

Thanks,
Rafael

2008-06-14 21:18:24

by Arjan van de Ven

[permalink] [raw]
Subject: Re: Regressions in the last kernels

On Sat, 14 Jun 2008 15:26:09 -0300 (GFT)
"werner" <[email protected]> wrote:

> During preparation and own testing for my distro; installation on
> computers for neighbours, by reclamations by users, I learned about
> the following problems of 2.6.26-rcX :
>
> 1) When compiling for i486, one cannot select/configurate more
> memory than 4 GB, menuconfig don't show items with higher values. My
> own computer/server has 8 GB, but when I compile the kernel with 4 GB
> and i486 then 'free' shows only 3.2 GB (no shared grafics, the grafic
> card has its own memory) ...


interesting problem... 486's didn't have PAE so are limited to 4Gb of
memory... makes this a policy question more than a bug...
(in fact, if you configure for > 4Gb a 486 CPU will not boot... so
arguably the new behavior is correct!!!)


--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-15 05:03:44

by Rene Herman

[permalink] [raw]
Subject: Re: Regressions in the last kernels

On 14-06-08 23:18, Arjan van de Ven wrote:

> On Sat, 14 Jun 2008 15:26:09 -0300 (GFT)
> "werner" <[email protected]> wrote:

>> 1) When compiling for i486, one cannot select/configurate more
>> memory than 4 GB, menuconfig don't show items with higher values. My
>> own computer/server has 8 GB, but when I compile the kernel with 4 GB
>> and i486 then 'free' shows only 3.2 GB (no shared grafics, the grafic
>> card has its own memory) ...
>
>
> interesting problem... 486's didn't have PAE so are limited to 4Gb of
> memory... makes this a policy question more than a bug...
> (in fact, if you configure for > 4Gb a 486 CPU will not boot... so
> arguably the new behavior is correct!!!)

What's arguable about it?

Rene.

2008-06-15 06:29:48

by David Newall

[permalink] [raw]
Subject: Re: Regressions in the last kernels

werner wrote:
> During preparation and own testing for my distro; installation on computers for neighbours, by reclamations by users, I learned about the following problems of 2.6.26-rcX :
>
> 1) When compiling for i486, one cannot select/configurate more memory than 4 GB, menuconfig don't show items with higher values.
>

Is not the i486 limited to a 4GB address space?

2008-06-15 10:36:54

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: Regressions in the last kernels

Please don't drop the CC list from your replies and please don't top-post.

On Sunday, 15 of June 2008, werner wrote:
> The 2nd problem anyway is present with the 2.6.26-rc5 and -rc4 kernels.
> It was NOT presend with 2.6.25-rc9, at least not on 1 IBM laptop what I had
> before-yesterday, which had the problem with 2.6.26-rc5/4 but not with
> 2.6.25-rc9 (all other config was the same). It did NOT help change edd from
> off to on, so that is a new problem. With -rc6, I donk know because I couldnt
> test it yet, the man had gone away with his laptop already.

Okay, so this issue seems to be a recent regression.

> Many people bring me old computers to repair. I always throw out Windows and
> install Linux. So I see plenty kernel errors. But I dont have time to report them
> all, only some. And normally I dont have long time the computers because
> later the people come and go away with them computers.
>
> I make all my kernel packages for i486 because here in the 3rd world many
> people have old computers ftp://ftp5.gwdg.de/pub/linux/install/sys/
> However these kernels should work on newer computers too, like on my server
> what has 8 GB memory. With i486, all the last kernels 2.6.26 cannot be
> configured to more than 4 GB and, running, nor that is working correctly.

This is intentional, as i486 processors cannot handle more than 4 GB of RAM.

Thanks,
Rafael

2008-06-15 11:01:15

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: Regressions in the last kernels

On Sunday, 15 of June 2008, David Newall wrote:
> werner wrote:
> > During preparation and own testing for my distro; installation on computers
> > for neighbours, by reclamations by users, I learned about the following
> > problems of 2.6.26-rcX :
> >
> > 1) When compiling for i486, one cannot select/configurate more memory than
> > 4 GB, menuconfig don't show items with higher values.
> >
>
> Is not the i486 limited to a 4GB address space?

The point, if I understand it correctly, is that if you compile the kernel for
i486 and then run it on something that does support PAE, it will only see
4 GB of RAM.

Thanks,
Rafael

2008-06-15 14:52:26

by David Newall

[permalink] [raw]
Subject: Re: Regressions in the last kernels

Rafael J. Wysocki wrote:
> On Sunday, 15 of June 2008, David Newall wrote:
>
>> werner wrote:
>>
>>> During preparation and own testing for my distro; installation on computers
>>> for neighbours, by reclamations by users, I learned about the following
>>> problems of 2.6.26-rcX :
>>>
>>> 1) When compiling for i486, one cannot select/configurate more memory than
>>> 4 GB, menuconfig don't show items with higher values.
>>>
>>>
>> Is not the i486 limited to a 4GB address space?
>>
>
> The point, if I understand it correctly, is that if you compile the kernel for
> i486 and then run it on something that does support PAE, it will only see
> 4 GB of RAM.

Rather like compiling it for i486 and running it on ia64: you ask for an
i486, you don't get to use features beyond that CPU.

2008-06-15 17:30:28

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: Regressions in the last kernels

On Sunday, 15 of June 2008, David Newall wrote:
> Rafael J. Wysocki wrote:
> > On Sunday, 15 of June 2008, David Newall wrote:
> >
> >> werner wrote:
> >>
> >>> During preparation and own testing for my distro; installation on computers
> >>> for neighbours, by reclamations by users, I learned about the following
> >>> problems of 2.6.26-rcX :
> >>>
> >>> 1) When compiling for i486, one cannot select/configurate more memory than
> >>> 4 GB, menuconfig don't show items with higher values.
> >>>
> >>>
> >> Is not the i486 limited to a 4GB address space?
> >>
> >
> > The point, if I understand it correctly, is that if you compile the kernel for
> > i486 and then run it on something that does support PAE, it will only see
> > 4 GB of RAM.
>
> Rather like compiling it for i486 and running it on ia64: you ask for an
> i486, you don't get to use features beyond that CPU.

Sure, that's reasonable.

Thanks,
Rafael