From mboxrd@z Thu Jan 1 00:00:00 1970 From: Bob Picco Date: Tue, 08 Oct 2013 13:42:02 +0000 Subject: Re: [PATCH v2 0/9] sparc64: Support 47-bit physical addresses. Message-Id: <20131008134202.GO29427@zareason> List-Id: References: <20130930.123130.354302574443157859.davem@davemloft.net> In-Reply-To: <20130930.123130.354302574443157859.davem@davemloft.net> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: sparclinux@vger.kernel.org Hi, David Miller wrote: [Mon Oct 07 2013, 12:08:50PM EDT] > From: Bob Picco > Date: Mon, 7 Oct 2013 12:07:45 -0400 > > > David Miller wrote: [Fri Oct 04 2013, 02:00:59PM EDT] > >> It's the trickiest change, as it changes the PMD layout completely. > >> > >> If the problem persists, the next point to check is where the huge > >> page size gets adjusted to 8MB from 4MB. Try reverted to just before > >> that change and test again. That's the other major possible source of > >> errors. > > Yes, I suspected this and the reason for bottom up. > > > > It seems to be here. With 9/9 and 8/9 reverted, the issue persists. Virgin > > 3.11 is solid. Of course there is always the possibility of mistake(s). > > > > I've obviously missed an issue. Looking... > > Please, explicitly test both at the point where huge pages go from > 4MB to 8MB and exactly right before that change. I've done this on 3.12-rc4. The point (7/9) where we transition HPAGE_SHIFT there are the issues already described with THP. > > Testing just a vanilla tree leaves too many variables in between. I agree.