@Tom Ellis: can you try reverting the commit cdaa050bd1c658c5b808a35a1fc061038ad19bc5 ("x86: Flush TLB if PGD entry is changed in i386 PAE mode") from the Ubuntu-2.6.32-32.62 kernel, build a -pae flavour and ask for testing?
I think it's likely this commit play a role in this bug, since it's inside the last bad/good window from your last bisect log (f1809971f60bf42ff9b47eac395e8729ed6bd525..1198ae36cd30092718563b42e2f1d847516a5e45), and this commit affects the PAE path, adding a flush_tlb_mm call that also appears on the soft lockup trace.
@Tom Ellis: can you try reverting the commit cdaa050bd1c658c 5b808a35a1fc061 038ad19bc5 ("x86: Flush TLB if PGD entry is changed in i386 PAE mode") from the Ubuntu-2.6.32-32.62 kernel, build a -pae flavour and ask for testing?
I think it's likely this commit play a role in this bug, since it's inside the last bad/good window from your last bisect log (f1809971f60bf4 2ff9b47eac395e8 729ed6bd525. .1198ae36cd3009 2718563b42e2f1d 847516a5e45) , and this commit affects the PAE path, adding a flush_tlb_mm call that also appears on the soft lockup trace.