Hi Robin 

During some of the stress tests we also came across a different warning from the arm64  page management code
It looks like a race is detected between HW and SW marking a bit in the PTE

Not sure it's really related but I thought it might give a clue on the issue
http://pastebin.com/ASv19vZP

Thanks

Yehuda 

-----Original Message-----
From: Marcin Wojtas [mailto:mw at semihalf.com]
Sent: Tuesday, May 31, 2016 13:30
To: Robin Murphy
Cc: linux-mm at kvack.org; linux-kernel at vger.kernel.org; linux-arm-
kernel at lists.infradead.org; Lior Amsalem; Thomas Petazzoni; Yehuda
Yitschak; Catalin Marinas; Arnd Bergmann; Grzegorz Jaszczyk; Will Deacon;
Nadav Haklai; Tomasz Nowicki; Gregory Cl?ment
Subject: Re: [BUG] Page allocation failures with newest kernels

Hi Robin,

quoted

I remember there were some issues around 4.2 with the revision of the
arm64 atomic implementations affecting the cmpxchg_double() in SLUB,
but those should all be fixed (and the symptoms tended to be

considerably more fatal).

quoted

A stronger candidate would be 97303480753e (which landed in 4.4),
which has various knock-on effects on the layout of SLUB internals -
does fiddling with L1_CACHE_SHIFT make any difference?

I'll check the commits, thanks. I forgot to add L1_CACHE_SHIFT was my first
suspect - I had spent a long time debugging network controller, which
stopped working because of this change - L1_CACHE_BYTES (and hence
NET_SKB_PAD) not fitting HW constraints. Anyway reverting it didn't help at
all for page alloc issue.

Best regards,
Marcin