Recently I've had cause to run memtest on some relatively modern systems (Haswell, Haswell refresh, and Kaby Lake R). On all three systems the memtest86+ provided with the Fedora Workstation Live media failed.
On the Haswell and Haswell refresh machines it froze shortly after starting. It didn't matter if the test was run in 'Safe Mode', or with SMP.
On the Kaby Lake R machine it causes the machine to reboot before memtest86+ launches.
It works fine on an older Wolfdale Core 2 Duo.
I've tried this with both the Workstation Live 31 and a 32 Beta image. Both verified.
Has anyone else seen this? Should there be a test case? Should it just be dropped since memtest86+ appears to be dead upstream, and doesn't support UEFI (and possibly doesn't work)?
My experience is that you need to change the bios to legacy and then it works fine.
Thanks Richard
On Tue, Mar 31, 2020 at 4:06 PM Brandon Nielsen nielsenb@jetfuse.net wrote:
Recently I've had cause to run memtest on some relatively modern systems (Haswell, Haswell refresh, and Kaby Lake R). On all three systems the memtest86+ provided with the Fedora Workstation Live media failed.
On the Haswell and Haswell refresh machines it froze shortly after starting. It didn't matter if the test was run in 'Safe Mode', or with SMP.
I can confirm the issues. I have Haswell (i5 4570): * With F32 Workstation Beta, memtest freezes at 35% (of the first test) reliably if I keep the defaults (which defaults to SMP:disabled) or force safe mode with F1. When I force SMP using F2, it reboots quickly. * With F31 Workstation, memtest reboots at 35% (of the first test) reliably if I keep the defaults (SMP: disabled) or force safe mode. When I force SMP, it reboots quickly. * With F30 Workstation, memtest seems to work fine (I waited only a few minutes) in all configurations.
With a libvirt VM, memtest seems to work fine (I waited only a few minutes) with safe mode with F30-F32, but reboots immediately in default configuration or SMP (F30-F32).
I used to have problems with memtest when I forced SMP in the past, but never with the default or safe mode configuration. Something changed recently (F31) for the worse. Interestingly, memtest always reports the same version (5.01) so I wonder what affects this.
On 3/31/20 10:23 AM, Kamil Paral wrote:
I can confirm the issues. I have Haswell (i5 4570):
- With F32 Workstation Beta, memtest freezes at 35% (of the first test)
reliably if I keep the defaults (which defaults to SMP:disabled) or force safe mode with F1. When I force SMP using F2, it reboots quickly.
- With F31 Workstation, memtest reboots at 35% (of the first test)
reliably if I keep the defaults (SMP: disabled) or force safe mode. When I force SMP, it reboots quickly.
- With F30 Workstation, memtest seems to work fine (I waited only a few
minutes) in all configurations.
With a libvirt VM, memtest seems to work fine (I waited only a few minutes) with safe mode with F30-F32, but reboots immediately in default configuration or SMP (F30-F32).
I used to have problems with memtest when I forced SMP in the past, but never with the default or safe mode configuration. Something changed recently (F31) for the worse. Interestingly, memtest always reports the same version (5.01) so I wonder what affects this.
I can confirm F30 Workstation Live works. I don't see any obvious difference in how memtest86+ is invoked.
The binary itself is obviously different however. Starting with Fedora 31, the memtest executable became 182.7 kB, previously it was 186.8 kB. Is it no longer compiling correctly?
On Tue, Mar 31, 2020 at 9:24 AM Kamil Paral kparal@redhat.com wrote:
On Tue, Mar 31, 2020 at 4:06 PM Brandon Nielsen nielsenb@jetfuse.net wrote:
Recently I've had cause to run memtest on some relatively modern systems (Haswell, Haswell refresh, and Kaby Lake R). On all three systems the memtest86+ provided with the Fedora Workstation Live media failed.
On the Haswell and Haswell refresh machines it froze shortly after starting. It didn't matter if the test was run in 'Safe Mode', or with SMP.
I can confirm the issues. I have Haswell (i5 4570):
- With F32 Workstation Beta, memtest freezes at 35% (of the first test) reliably if I keep the defaults (which defaults to SMP:disabled) or force safe mode with F1. When I force SMP using F2, it reboots quickly.
- With F31 Workstation, memtest reboots at 35% (of the first test) reliably if I keep the defaults (SMP: disabled) or force safe mode. When I force SMP, it reboots quickly.
- With F30 Workstation, memtest seems to work fine (I waited only a few minutes) in all configurations.
With a libvirt VM, memtest seems to work fine (I waited only a few minutes) with safe mode with F30-F32, but reboots immediately in default configuration or SMP (F30-F32).
I used to have problems with memtest when I forced SMP in the past, but never with the default or safe mode configuration. Something changed recently (F31) for the worse. Interestingly, memtest always reports the same version (5.01) so I wonder what affects this.
gcc version?
On 3/31/20 3:53 PM, Chris Murphy wrote:
gcc version?
I don't think an application written mostly in assembly would change that significantly in size with different compiler versions.
Anyway, it looks like it broke when gcc-compat was dropped:
On 3/31/20 4:03 PM, Brandon Nielsen wrote:
On 3/31/20 3:53 PM, Chris Murphy wrote:
gcc version?
I don't think an application written mostly in assembly would change that significantly in size with different compiler versions.
Anyway, it looks like it broke when gcc-compat was dropped:
Which exactly matches your point about compiler version, it's also written mostly in C. I should learn to read...
Brandon Nielsen composed on 2020-03-31 09:05 (UTC-0500):
Recently I've had cause to run memtest on some relatively modern systems (Haswell, Haswell refresh, and Kaby Lake R). On all three systems the memtest86+ provided with the Fedora Workstation Live media failed.
On the Haswell and Haswell refresh machines it froze shortly after starting. It didn't matter if the test was run in 'Safe Mode', or with SMP.
On the Kaby Lake R machine it causes the machine to reboot before memtest86+ launches.
It works fine on an older Wolfdale Core 2 Duo.
I've tried this with both the Workstation Live 31 and a 32 Beta image. Both verified.
Has anyone else seen this? Should there be a test case? Should it just be dropped since memtest86+ appears to be dead upstream, and doesn't support UEFI (and possibly doesn't work)?
I long ago switched to free version of https://www.memtest86.com/ for anything with DDR4 or UEFI.
-
Brandon Nielsen -
Chris Murphy -
Felix Miata -
Kamil Paral -
Richard Shaw