Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190410101947.8603-1-juha-matti.tilli@foreca.com>
 <CANn89iK9DcDi68WQn0z2R65L1fQpa88nUnpe_di1OWGhhoqAQw@mail.gmail.com>
 <CAA1_w3_4ZeOA-viD4W0QH1rUYPczcwS2KpnsvRhCX2hQniV-kQ@mail.gmail.com> <CANn89iLmzADQF5=htGsjNwXrUZxMUYaYZB+6WOM_vo6zG+JB9A@mail.gmail.com>
In-Reply-To: <CANn89iLmzADQF5=htGsjNwXrUZxMUYaYZB+6WOM_vo6zG+JB9A@mail.gmail.com>
From:   Juha-Matti Tilli <juha-matti.tilli@foreca.com>
Date:   Thu, 11 Apr 2019 09:26:02 +0300
Message-ID: <CAA1_w39v3sjg9qOoX68bSNx_WeWVJxEh3MKrx7G-uJTO7UxhiQ@mail.gmail.com>
Subject: Re: [PATCH] net: add big honking pfmemalloc OOM warning
To:     Eric Dumazet <edumazet@google.com>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        netdev <netdev@vger.kernel.org>,
        Rafael Aquini <aquini@redhat.com>,
        Murphy Zhou <xzhou@redhat.com>,
        Yongcheng Yang <yoyang@redhat.com>,
        Jianhong Yin <jiyin@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Apr 10, 2019 at 6:36 PM Eric Dumazet <edumazet@google.com> wrote:
> Before diving into the details, can we first double check which exact
> kernel version you are using ?
>
> In the past some pfmemalloc bugs have been solved, I do not want spend
> time finding them a second time.

Of course I can tell the kernel version.

The NFS server is 3.10.0-862.14.4.el7.foreca.x86_64 where "foreca"
means we've applied a custom Samsung SSD TRIM patch, nothing else
different from regular CentOS / RHEL kernel.

The NFS client is 3.10.0-957.5.1.el7.x86_64.

The pfmemalloc problem happens at the server, not at the client.

The exact patches in this kernel may be a bit complicated to
determine, because RHEL apparently likes to maintain their own kernel
version. They do not have the SNMP counter even, which is why I opened
a bug report to RHEL Bugzilla. I have extracted source code, so if you
can tell a list of git commit SHA1 hashes that could fix the issue, I
can check manually whether they're there, and open bug reports to RHEL
bugzilla to have them backported if not there.

I at least see this in the exact kernel version I'm using:

static inline bool ixgbe_page_is_reserved(struct page *page)
{
        return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
}

...so ixgbe shouldn't be reusing pfmemalloc pages.

BR, Juha-Matti