Comment 21 for bug 1332409

Revision history for this message
Steve (lp-z) wrote :

fixed in: v3.16-rc3
fixed in: v3.14.5
fixed in: v3.14.0
fixed in: v3.14.0rc1
broken in: v3.13.11.4 [the most recent v3.13 kernel]
broken in v3.13.0 [the earliest v3.13 kernel, without going into the rc versions]
fixed in: 3.12.22 [one of the more recent 3.12 kernels; was the most recent at the time I had tried it]
[and broken in all other v3.13 versions I had tried]

At this point it looks like it's something that is special only to 3.13 (possibly some commit that was applied into 3.13 at an early time, never made it into 3.14 branch or was rewritten in 3.14...?) At this point, I'm not planning on testing more kernels since it seems like this is the pattern.

Looking at http://kernelnewbies.org/Linux_3.13 - there were some major changes around both "Improved performance in NUMA systems" (this is a multi-processor system) and "Improved page table access scalability in hugepage workloads" (this does have 256gb of memory); one guess is that there could be something in those changes that was redesigned in 3.14 to work better(?) (though scanning through the top features in 3.14 nothing is jumping out to me...?)