[GTALUG] Intel Baytrail systems may become stable under Linux!
D. Hugh Redelmeier
hugh at mimosa.com
Sun Feb 12 10:56:12 EST 2017
For two years, Intel Baytrail processors have been hanging under
Linux.
Baytrail is a generation of Atom processors and includes many low-end
processors that I think of as Atoms but are called things like
Pentium and Celeron. But not all Pentium and Celeron processors.
<http://ark.intel.com/products/codename/55844/Bay-Trail>
The bug report is
<https://bugzilla.kernel.org/show_bug.cgi?id=109051>
but started elsewhere. A few days ago, entry 724 (!) in that bug
report noted the submission of a kernel patch that strongly reduces
the problem. The thread for the patch is interesting:
<https://lists.freedesktop.org/archives/intel-gfx/2017-January/117932.html>
- the fix doesn't fix all the underlying problems but very
significantly improves stability. Perhaps to the point that
the instability fades into the background noise of crashes.
- the fix slightly cripples power management on these systems but much
less than the only known work-around (preventing the use of cstates
higher than 1)
- this may slow down the search for residual bugs, but how can it be
slower than the current process has been?
- The fix was already in the kernel for Cherryview processors. I'm
not sure what those are: it certainly includes Cherry Trail processors.
<http://ark.intel.com/products/codename/46629/Cherry-Trail>
- the bug started to become visible in Linux 3.17-rc1 and should be
ameliorated in 4.11. What a long horrible run. Most of the life of
the chips!
I have a number of Atom systems that I never got around to Linuxifying
due to fear of this bug and, for some, annoyance with 32-bit UEFI
firmware. I will say that I've had a great experience with a netbook
with a Celeron N2840 (Baytrail). Perhaps because I don't often play
video content with it.
Some of these Baytrail systems are still a nightmare for Linux due to
non-systematic and undocumented connection of the SoCs to the peripheral
circuits. Here's a sample thread for one such system (more than 200
entries!):
<https://bugzilla.kernel.org/show_bug.cgi?id=95681>
More information about the talk
mailing list