The Leading Cause Of The Recent Linux Kernel Power Problems

[Xake]

Someone needs to do the same here, because obviously the MB manufacturers are just hacking things up until they get it to work with Windows.

Sorry to say but that would require a huge whitelist/blacklist, and would need to be dependent on BIOS version. Mostly because these are functions that never worked in Windows in the first place, since for example Windows XP did not support ASPM and thus manufacturers did not support it in their BIOS or had broken stuff in there. So if your MoBo has not gotten a BIOS update since WindowsXP but acctually supports this, then it needs to be added to the whitelist. And that would soon become a HUGE whitelist.

[Rabauke]

max@t410:~$ dmesg | grep -i aspm
[    0.000000] Command line: BOOT_IMAGE=/boot/vmlinuz-2.6.39-2-amd64 root=UUID=9ff997f7-ee34-433b-b5d9-6123b29fb119 ro quiet pcie_aspm=force
[    0.000000] Kernel command line: BOOT_IMAGE=/boot/vmlinuz-2.6.39-2-amd64 root=UUID=9ff997f7-ee34-433b-b5d9-6123b29fb119 ro quiet pcie_aspm=force
[    0.000000] PCIe ASPM is forcedly enabled
[    0.509091] ACPI FADT declares the system doesn't support PCIe ASPM, so disable it
[    0.800123] Unable to assume _OSC PCIe control. Disabling ASPM
[    0.811896] Unable to assume _OSC PCIe control. Disabling ASPM
max@t410:~$ uname -a
Linux t410 2.6.39-2-amd64 #1 SMP Wed Jun 8 11:01:04 UTC 2011 x86_64 GNU/Linux

Any thoughts on that? I'm not affected by that bug, am I?

[Drago]

I am eager to see, who is gonna be "the large oem supplier to embrace coreboot", so I will build my next rig with his mobo. I fact this is what stopping me now!

[diebels]

I'm getting it on a pre 38 kernel:

@ThinkPad-X201:~$ grep -i aspm /var/log/dmesg
[    0.288969] ACPI FADT declares the system doesn't support PCIe ASPM, so disable it
@ThinkPad-X201:~$ uname -a
Linux ThinkPad-X201 2.6.35-28-generic-pae #50-Ubuntu SMP Fri Mar 18 20:43:15 UTC 2011 i686 GNU/Linux

[vitalyb]

ASPM was a major headache on one of the servers we use (randomly failing Intel NICs) so we had to replace the server. Although original server is still not in production various tests (including the one with all different hardware but original motherboard) shown that this is motherboards fault. ASPM was under suspicion almost all the time but no BIOS options, kernel options, etc. changed anything. Until 2.6.38, which suddenly just worked. Not sure if system became fully stable, but difference was huge.

[ahlaht]

With pcie_aspm=force I have:
PCIe ASPM is forcedly enabled
ACPI FADT declares the system doesn't support PCIe ASPM, so disable it

I wonder if it's enabled now or not...

To answer my own question:

The source code revealed "ACPI FADT declares the system doesn't support PCIe
ASPM, so disable it" is always printed on my system even if it's actually enabled.

[darkbasic]

[ 0.000000] PCIe ASPM is forcedly enabled
[ 0.229555] ACPI _OSC control for PCIe not granted, disabling ASPM

Is it enabled or not?

[allquixotic]

Lol, I have an ASUS P6T Deluxe motherboard where the IOMMU (a device for remapping DMA-capable devices such as PCI-E video cards) is completely broken. I get random kernel panics when I have Intel VT-D enabled (VT-D is a prerequisite for the IOMMU). It's pathetic. Motherboard manufacturers don't pay attention to product quality anymore. They just throw shit together, put a bunch of hacks into the Windows driver so that it works semi-decently, and let it go. Even if whole entire features are permanently broken at the hardware level.

[ahlaht]

I'm getting it on a pre 38 kernel:

The message is printed on earlier kernels too, but it does things differently.

[ 0.000000] PCIe ASPM is forcedly enabled
[ 0.229555] ACPI _OSC control for PCIe not granted, disabling ASPM

Is it enabled or not?

Might be enabled. It calls "pcie_no_aspm()" and it does nothing if pcie_aspm=force is set.

[zhark]

Great work Michael!

Looking forward for details on more Linux power regressions, and workarounds.