If you have an affected motherboard to the ASPM power regression in the Linux kernel and it's from Gigabyte, don't expect a BIOS update from them to correct the ASPM semantics in the BIOS. Gigabyte recommends you just use Microsoft Windows.
This shouldn't be a huge surprise with a motherboard vendor outright denying Linux support and not correcting their BIOS to properly support a PCI Express power management feature, since it works fine in Microsoft Windows regardless. It's all too often that motherboard vendors won't make BIOS changes to their products as long as it works under Windows, even if their BIOS engineers took a short-cut in the first place or incorrectly implemented a feature. The response that Gigabyte had in regards to the ASPM Linux power issues on their products is posted below, which appeared this morning in the Phoronix Forums
Thank you for your kindly mail and inquiry. About the issue you mentioned, since our products only support Windows OS, we do not receive proper driver from chipset vender, we cannot guarantee Linux to work on our system. We suggest you to install Windows OS to prevent having problems. If you install the Windows OS and still have any problems, please provide the error message screenshot for us, so we can try to see how to help. Sorry for the inconvenience.
It's really not a big surprise in general for the motherboard industry to act this way or Gigabyte recommending the use of Windows instead (there's a reason why Phoronix hasn't had a review of a Gigabyte product in over three years
). ASUS is another vendor that sometimes is notorious for their poor BIOS implementations, just read the experiences about how ASUS has very poor BIOS support
as said by Dean Hilkewich, a veteran member of the Phoronix Forums. Even the new ZaReason Chimera "Sandy Bridge" notebook (from the OEM Compal) shipping with Ubuntu that I am currently testing doesn't even advertise Active-State Power Management from the BIOS.
For those not up to speed on the situation, what's being talked about is the major power regression in the Linux 2.6.38 kernel
that I spotted earlier this year that's significantly decreasing the battery life for many Linux systems. I subsequently traced the issue to a change in PCI-E ASPM behavior
in the Linux kernel where Active-State Power Management is no longer being enabled unless the system's BIOS/firmware properly advertises it's supported. Many motherboards though aren't setting the ASPM bit since Windows seems to work without it. There's some ways around it
by having drivers properly white-list hardware that's known to work regardless of what the BIOS says or determining how Microsoft Windows is dealing with toggling the ASPM support, but nothing major has yet to happen upstream in the Linux kernel to fix the situation. (At least you can override the ASPM support by booting the kernel with pcie_aspm=force
Getting the major motherboard vendors to put out new BIOSes for their affected products is sadly unrealistic. Earlier this month I published a partial list of affected motherboards
to this ASPM issue based upon data collected on OpenBenchmarking.org
. Among the affected brands are ASUS,Acer, Dell, Gigabyte, HP, Intel, Lenovo, MSI, and Pegatron. There's even some affected motherboards from Intel.
What motherboard brand generally offers the best BIOS that's properly compliant and shouldn't have any functional issues with Linux? Tyan is one of the best motherboard brands I have dealt with in this regard, but Tyan mainly targets the workstation and server markets. Abit also used to be great going back many years (the good ol' NF7-S and IC7-MAX3 days), but sadly they are no longer around. SuperMicro boards also usually have well-developed BIOS as well, but like Tyan you won't find much consumer hardware from them. Lenovo notebooks tend to be good overall and semi-frequent BIOS updates, but there's still some ASPM problems on ThinkPads in my possession and I don't have much experience with their very recent models. Aside from that, it's mostly a mixed bag with the other vendors that depends upon the particular model. (To help going forward, there are more OpenBenchmarking.org
features on the way to deal with this crowd-sourcing of other non-performance metrics.)