On 08/05/2016 04:26 PM, Ben Hutchings wrote:

On Sun, 2016-07-31 at 10:34 -0700, Guenter Roeck wrote:

quoted

Now I get a crash with the nosmp qemu test, and

Did you work out what introduced that?

Bisect says:

# bad: [c85164130167a51006876477e6b4044312960478] ecryptfs: don't allow mmap when the lower fs doesn't support it
# good: [c5d5c18b211dabbfac73670c65dfa4243800adf7] Linux 3.2.81
git bisect start 'HEAD' 'v3.2.81'
# bad: [0d5b7d090ee63c7ab5530fea29beffe989ae385a] parisc: Fix pagefault crash in unaligned __get_user() call
git bisect bad 0d5b7d090ee63c7ab5530fea29beffe989ae385a
# bad: [db17a77a32fe7703a433f428f65f28fddb20d5ba] MIPS: Sync icache & dcache in set_pte_at
git bisect bad db17a77a32fe7703a433f428f65f28fddb20d5ba
# good: [9f4faf23b3c3d719a1005364ac3cca8cb1ca07ca] char: Drop bogus dependency of DEVPORT on !M68K
git bisect good 9f4faf23b3c3d719a1005364ac3cca8cb1ca07ca
# good: [180309d63478b762234b229abf37b0ed4d95fef7] MIPS: Adjust set_pte() SMP fix to handle R10000_LLSC_WAR
git bisect good 180309d63478b762234b229abf37b0ed4d95fef7
# good: [dcbaa9382ed2ded2de553633e58791d057e76ad2] powerpc/mm/hash64: Factor out hash preload psize check
git bisect good dcbaa9382ed2ded2de553633e58791d057e76ad2
# good: [ad1e677bf950f3943e8a3cdd546b2c6202ea7ed4] sched/loadavg: Fix loadavg artifacts on fully idle and on fully loaded systems
git bisect good ad1e677bf950f3943e8a3cdd546b2c6202ea7ed4
# first bad commit: [db17a77a32fe7703a433f428f65f28fddb20d5ba] MIPS: Sync icache & dcache in set_pte_at

After reverting that patch, the problem is gone.

Turns out 3.16 is affected by the same problem, and reverting the patch there fixes it as well.

Guenter