On 8/3/19 1:29 AM, Petr Vandrovec wrote:
> On Fri, Aug 2, 2019, 3:59 PM Matthew Wilcox <willy@infradead.org
> <mailto:willy@infradead.org>> wrote:
> 
>     That doesn't help because we call reset_page_owner() in the free
>     page path.
> 
>     We could turn on tracing because we call trace_mm_page_free() in this
>     path.  That requires the reporter to be able to reproduce the problem,
>     and it's not clear to me whether this is a "happened once" or "every
>     time I do this, it happens" problem.
> 
> 
> It happened on 3 of the boots with that kernel.  4th time box either
> spontaneously rebooted when X started, or watchdog restarted box shortly
> after starting X server.
> 
> So I believe I should be able to reproduce it with additional patches or
> extra flags enabled.

Does the issue still happen with rc4? Could you apply the 3 attached
patches (work in progress), configure-enable CONFIG_DEBUG_PAGEALLOC and
CONFIG_PAGE_OWNER and boot kernel with debug_pagealloc=on page_owner=on
parameters? That should print stacktraces of allocation and first
freeing (assuming this is a double free).

Vlastimil