Comment 39 for bug 1674838

Thomas M Steenholdt (tmus) wrote :

There must be a better way to do this...

All of these issues seem to arise from a BUG event in swapops.h:129. That particular spot is a section that's only active, when the kernel was built with CONFIG_MIGRATION=y. So first step is probably to verify that CONFIG_MIGRATION is even enabled for the mainline kernel (the configs are not the same, I'm told). So for all we know, the bug could still be upstream.
If somebody running the mainline kernel could post the output of the following command, that would be useful:

cat /boot/config-$(uname -r) |grep CONFIG_MIGRATION

If CONFIG_MIGRATION is enabled on mainline (CONFIG_MIGRATION=y in the output above), next step should be to check if some of the Ubuntu modifications touch in the source in any relevant places. The BUG event in swapops.h:129 seems to be hit if migration_entry_to_page() is called with an unlocked page. Grepping through the source, this function is only called from a handful of places, so it should be possible cross-reference with the Ubuntu modifications.

Perhaps this will bring us closer to the problem a bit faster?