Bug 1303804

I am unable to reproduce the problem, it works on all of my i686 test machines without problem.

Could you find out which version introduced the problem for you? 5.01-10 or 5.01-11?

Where can I find 5.01-10? It's not in koji.

The machine is a Dell Dimension B110 I bought in 2006, with a 32-bit Celeron D CPU. I'm planning to test Fedora-Live-Workstation-i686-rawhide-20160204.iso as soon as I finish downloading, both in a VM, and if that works, after burned to media.

Fedora-Live-Workstation-i686-rawhide-20160204.iso memtest86+ works, both in a VirtualBox VM on a 64-bit host, and when burned to media, run on the same 32-bit machine where the F23 version fails when run from the grub menu.

(In reply to Andre Robatino from comment #2)
> Where can I find 5.01-10? It's not in koji.

Interesting, it seems it is not in the build system. I built scratch build for testing:

http://koji.fedoraproject.org/koji/taskinfo?taskID=12890206

(In reply to Andre Robatino from comment #4)
> Fedora-Live-Workstation-i686-rawhide-20160204.iso memtest86+ works, both in
> a VirtualBox VM on a 64-bit host, and when burned to media, run on the same
> 32-bit machine where the F23 version fails when run from the grub menu.

This is even more interesting, IIRC the only difference between rawhide and f23 memtest86+ versions are different version of compiler used.

(In reply to Jaroslav Škarvada from comment #5)
> (In reply to Andre Robatino from comment #2)
> > Where can I find 5.01-10? It's not in koji.
> 
> Interesting, it seems it is not in the build system. I built scratch build
> for testing:
> 
> http://koji.fedoraproject.org/koji/taskinfo?taskID=12890206

Just tested, this fails the same way as 5.01-11.

(In reply to Jaroslav Škarvada from comment #6)
> (In reply to Andre Robatino from comment #4)
> > Fedora-Live-Workstation-i686-rawhide-20160204.iso memtest86+ works, both in
> > a VirtualBox VM on a 64-bit host, and when burned to media, run on the same
> > 32-bit machine where the F23 version fails when run from the grub menu.
> 
> This is even more interesting, IIRC the only difference between rawhide and
> f23 memtest86+ versions are different version of compiler used.

Does it matter whether you run memtest86+ from the live image's troubleshooting menu, versus running it from the grub menu on an installed system with memtest86+ installed?

(In reply to Andre Robatino from comment #8)
> Does it matter whether you run memtest86+ from the live image's
> troubleshooting menu, versus running it from the grub menu on an installed
> system with memtest86+ installed?

I think yes, it is different loader, IIRC isolinux vs grub-2.

(In reply to Jaroslav Škarvada from comment #9)
> (In reply to Andre Robatino from comment #8)
> > Does it matter whether you run memtest86+ from the live image's
> > troubleshooting menu, versus running it from the grub menu on an installed
> > system with memtest86+ installed?
> 
> I think yes, it is different loader, IIRC isolinux vs grub-2.

OK, I want to keep F23 installed on this machine (until F24 goes Gold), so the only way for me to test the F23 memtest version is running it from the grub menu (unless there's a F23 live image available which includes the new memtest) and the only way for me to test the Rawhide version is running it from the live troubleshooting menu (since I don't want to install Rawhide on it).

memtest86+-5.01-10 - fixed relocation routine which has been used for long in older memtests, but was broken since rebase to version 5 release. The relocation is required by some bootloaders, because they cannot load ELF image bellow 1MB (e.g. grub-legacy), relocated the code to address used in memtest86+-4.20.

memtest86+-5.01-11 - Relocated the code to the address introduced in 5 release, because the old address introduced problems with some firmwares when booted binary image from floppy/USB.

Unfortunately I haven't found Dell Dimension B110 in our pool, nor any machine to reproduce the problem on, so I don't know what's going there.

Created attachment 1121462 [details]
dmesg for Dell Dimension B110

I'll be happy to provide any logfiles or hardware info I can. For example, here is dmesg output. There is some basic info regarding memory layout early in the file.

Unfortunately such issues are hard to debug and mostly requires physical access to the machine. Does the binary image works for you (instead of the ELF?). To check it just change in your grub.cfg:

knetbsd /elf-memtest86+-5.01

to

linux16 /memtest86+-5.01

(In reply to Jaroslav Škarvada from comment #13)
> Unfortunately such issues are hard to debug and mostly requires physical
> access to the machine. Does the binary image works for you (instead of the
> ELF?). To check it just change in your grub.cfg:
> 
> knetbsd /elf-memtest86+-5.01
> 
> to
> 
> linux16 /memtest86+-5.01

After making the change, it still fails, but the displayed message is now

Loading Fedora Memtest ...Loading memtest86+-5.01  ...
error: invalid magic number.

Press any key to continue...

(This is with 5.01-11. I updated back to -11 after testing -10.)

Testing 5.01-10 with the same grub.cfg change, the result is exactly the same.

Sorry, just realized I made the wrong change (instead of "linux16 /memtest86+-5.01", I had "linux16 /elf-memtest86+-5.01"). Now that I have fixed it, memtest86+-5.01-11 works properly. But with -10, I get the original message

Loading Fedora Memtest ...Loading memtest86+-5.01  ...

and after 10 seconds, it reboots by itself to the grub menu. I don't believe that happened before, it was just hung.

(In reply to Andre Robatino from comment #16)
> Sorry, just realized I made the wrong change (instead of "linux16
> /memtest86+-5.01", I had "linux16 /elf-memtest86+-5.01"). Now that I have
> fixed it, memtest86+-5.01-11 works properly. But with -10, I get the
> original message
> 
> Loading Fedora Memtest ...Loading memtest86+-5.01  ...
> 
> and after 10 seconds, it reboots by itself to the grub menu. I don't believe
> that happened before, it was just hung.

Thanks for info. I will probably update the memtest-setup script to have an option which version to install.

Another person had this problem, but on a 64-bit machine. See https://lists.fedoraproject.org/pipermail/users/2016-February/468857.html . So I don't know what the trigger is.

Hello, everyone.  For me, the new memtest86+ (memtest86+-5.01-11.fc23.x86_64) works on my Core2 Duo machines, but not on my Core i5 machines (first and second generation).  Let me know if you'd like more details.

OK, so the trigger definitely isn't old hardware. (BTW, one of my 64-bit machines that work is a Core 2 Duo as well.) I also checked that the F23 release version, which worked for me (memtest86+-5.01-8.fc22) used knetbsd instead of linux16, and that version worked on all my machines. So it's not just using knetbsd either.

Added option and configuration file and updated the doc:

Install ELF version:
# memtest-setup
# grub2-mkconfig -o /boot/grub2/grub.cfg

Install non-ELF version:
# memtest-setup -b
# grub2-mkconfig -o /boot/grub2/grub.cfg

It would be great to go with only one variant of the image sometimes in the future, but it is not yet ready.

Regarding machines failing with the ELF image, I have no idea - both images do reloc - this looks like black magic :)

Grub2 don't requires reloc and AFAIK it is the only non-UEFI bootloader in Fedora (not counting syslinux), so I dropped the reloc in rawhide and let's see.

memtest86+-5.01-14.fc23 has been submitted as an update to Fedora 23. https://bodhi.fedoraproject.org/updates/FEDORA-2016-5a0ae108e4

memtest86+-5.01-14.fc23 has been pushed to the Fedora 23 testing repository. If problems still persist, please make note of it in this bug report.
See https://fedoraproject.org/wiki/QA:Updates_Testing for
instructions on how to install test updates.
You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2016-5a0ae108e4

The default ELF version of memtest86+-5.01-14.fc23 works on the same machine where it failed with -10 and -11.

memtest86+-5.01-14.fc23 has been pushed to the Fedora 23 stable repository. If problems still persist, please make note of it in this bug report.

I think the resolution is supposed to be CURRENTRELEASE since this was reported for and fixed in F23.

CURRENTRELEASE 	The problem described has been fixed in the current supported release.

ERRATA 	The problem described has been fixed and is available as an update from our support web site.

Oh, I see it was set automatically. Still not sure if that's correct.

(In reply to Andre Robatino from comment #26)
> I think the resolution is supposed to be CURRENTRELEASE since this was
> reported for and fixed in F23.
> 
> CURRENTRELEASE 	The problem described has been fixed in the current
> supported release.
> 
> ERRATA 	The problem described has been fixed and is available as an update
> from our support web site.

Sorry, I think you are wrong, it was resolved by and an errata (tracked update in updates), thus I think it should be closed as ERRATA (as it automatically was by the infrastructure).