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: <1543892757-4323-1-git-send-email-kernelfans@gmail.com>
 <20181204065453.4rsyhtsk2aej4vim@master> <CAFgQCTvQBC11=4eGQ6T9vyB+KOznFCr4hjdg1wwD8GotSRWpDg@mail.gmail.com>
 <20181204083428.emgcaomg6vulknaq@master> <CAFgQCTtY28w=9LLgOMT5J-SfKqz-Cirv-bNvXQvjazn_Ev4F4g@mail.gmail.com>
 <20181204090950.ql3zbnbjjbfnvhdg@master>
In-Reply-To: <20181204090950.ql3zbnbjjbfnvhdg@master>
From:   Pingfan Liu <kernelfans@gmail.com>
Date:   Wed, 5 Dec 2018 13:50:28 +0800
Message-ID: <CAFgQCTtKud54LL+afhf+j4axzho9MNErDN1DCNZLbOtHG-gHWQ@mail.gmail.com>
Subject: Re: [PATCH] mm/alloc: fallback to first node if the wanted node offline
To:     richard.weiyang@gmail.com
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        Andrew Morton <akpm@linux-foundation.org>,
        Michal Hocko <mhocko@suse.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Bjorn Helgaas <bhelgaas@google.com>,
        Jonathan Cameron <Jonathan.Cameron@huawei.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Dec 4, 2018 at 5:09 PM Wei Yang <richard.weiyang@gmail.com> wrote:
>
> On Tue, Dec 04, 2018 at 04:52:52PM +0800, Pingfan Liu wrote:
> >On Tue, Dec 4, 2018 at 4:34 PM Wei Yang <richard.weiyang@gmail.com> wrote:
> >>
> >> On Tue, Dec 04, 2018 at 03:20:13PM +0800, Pingfan Liu wrote:
> >> >On Tue, Dec 4, 2018 at 2:54 PM Wei Yang <richard.weiyang@gmail.com> wrote:
> >> >>
> >> >> On Tue, Dec 04, 2018 at 11:05:57AM +0800, Pingfan Liu wrote:
> >> >> >During my test on some AMD machine, with kexec -l nr_cpus=x option, the
> >> >> >kernel failed to bootup, because some node's data struct can not be allocated,
> >> >> >e.g, on x86, initialized by init_cpu_to_node()->init_memory_less_node(). But
> >> >> >device->numa_node info is used as preferred_nid param for
> >> >>
> >> >> could we fix the preferred_nid before passed to
> >> >> __alloc_pages_nodemask()?
> >> >>
> >> >Yes, we can doit too, but what is the gain?
> >>
> >> node_zonelist() is used some places. If we are sure where the problem
> >> is, it is not necessary to spread to other places.
> >>
> >> >
> >> >> BTW, I don't catch the function call flow to this point. Would you mind
> >> >> giving me some hint?
> >> >>
> >> >You can track the code along slab_alloc() ->...->__alloc_pages_nodemask()
> >>
> >> slab_alloc() pass NUMA_NO_NODE down, so I am lost in where the
> >> preferred_nid is assigned.
> >>
> >You can follow:
> >[    5.773618]  new_slab+0xa9/0x570
> >[    5.773618]  ___slab_alloc+0x375/0x540
> >[    5.773618]  ? pinctrl_bind_pins+0x2b/0x2a0
> >where static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
> >
>
> Well, thanks for your patience, but I still don't get it.
>
> new_slab(node)
>     allocate_slab(node)
>        alloc_slab_page(node)
>            if (node == NUMA_NO_NODE)
>                alloc_pages()
>            eles
>                __alloc_pages_node(node)
>
> As you mentioned, this starts from slab_alloc() which pass NUMA_NO_NODE.
> This means it goes to alloc_pages() and then alloc_pages_current() ->
> __alloc_pages_nodemask(). Here we use policy_node() to get the
> preferred_nid.
>
> I didn't catch the relathionship between policy_node() and
> device->numa_node. Maybe I got wrong in some place. Would you minding
> sharing more?
>
Have uploaded the full panic log. Enjoy it.

Regards,
Pingfan