| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 6 Dec 2018 11:34:30 +0800
From: Pingfan Liu <kernelfans@...il.com>
To: mhocko@...nel.org
Cc: linux-mm@...ck.org, linux-kernel@...r.kernel.org,
Andrew Morton <akpm@...ux-foundation.org>,
Vlastimil Babka <vbabka@...e.cz>,
Mike Rapoport <rppt@...ux.vnet.ibm.com>,
Bjorn Helgaas <bhelgaas@...gle.com>,
Jonathan Cameron <Jonathan.Cameron@...wei.com>,
"Rafael J. Wysocki" <rjw@...ysocki.net>,
Len Brown <lenb@...nel.org>, linux-acpi@...r.kernel.org
Subject: Re: [PATCH] mm/alloc: fallback to first node if the wanted node offline
On Wed, Dec 5, 2018 at 5:43 PM Michal Hocko <mhocko@...nel.org> wrote:
>
> On Wed 05-12-18 17:29:31, Pingfan Liu wrote:
> > On Wed, Dec 5, 2018 at 5:21 PM Michal Hocko <mhocko@...nel.org> wrote:
> > >
> > > On Wed 05-12-18 13:38:17, Pingfan Liu wrote:
> > > > On Tue, Dec 4, 2018 at 4:56 PM Michal Hocko <mhocko@...nel.org> wrote:
> > > > >
> > > > > On Tue 04-12-18 16:20:32, Pingfan Liu wrote:
> > > > > > On Tue, Dec 4, 2018 at 3:22 PM Michal Hocko <mhocko@...nel.org> wrote:
> > > > > > >
> > > > > > > On Tue 04-12-18 11:05:57, 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
> > > > > > > > __alloc_pages_nodemask(), which causes NULL reference
> > > > > > > > ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
> > > > > > > > This patch tries to fix the issue by falling back to the first online node,
> > > > > > > > when encountering such corner case.
> > > > > > >
> > > > > > > We have seen similar issues already and the bug was usually that the
> > > > > > > zonelists were not initialized yet or the node is completely bogus.
> > > > > > > Zonelists should be initialized by build_all_zonelists quite early so I
> > > > > > > am wondering whether the later is the case. What is the actual node
> > > > > > > number the device is associated with?
> > > > > > >
> > > > > > The device's node num is 2. And in my case, I used nr_cpus param. Due
> > > > > > to init_cpu_to_node() initialize all the possible node. It is hard
> > > > > > for me to figure out without this param, how zonelists is accessed
> > > > > > before page allocator works.
> > > > >
> > > > > I believe we should focus on this. Why does the node have no zonelist
> > > > > even though all zonelists should be initialized already? Maybe this is
> > > > > nr_cpus pecularity and we do not initialize all the existing numa nodes.
> > > > > Or maybe the device is associated to a non-existing node with that
> > > > > setup. A full dmesg might help us here.
> > > > >
> > > > Requiring the machine again, and I got the following without nr_cpus option
> > > > [root@...l-per7425-03 ~]# cd /sys/devices/system/node/
> > > > [root@...l-per7425-03 node]# ls
> > > > has_cpu has_memory has_normal_memory node0 node1 node2 node3
> > > > node4 node5 node6 node7 online possible power uevent
> > > > [root@...l-per7425-03 node]# cat has_cpu
> > > > 0-7
> > > > [root@...l-per7425-03 node]# cat has_memory
> > > > 1,5
> > > > [root@...l-per7425-03 node]# cat online
> > > > 0-7
> > > > [root@...l-per7425-03 node]# cat possible
> > > > 0-7
> > > > And lscpu shows the following numa-cpu info:
> > > > NUMA node0 CPU(s): 0,8,16,24
> > > > NUMA node1 CPU(s): 2,10,18,26
> > > > NUMA node2 CPU(s): 4,12,20,28
> > > > NUMA node3 CPU(s): 6,14,22,30
> > > > NUMA node4 CPU(s): 1,9,17,25
> > > > NUMA node5 CPU(s): 3,11,19,27
> > > > NUMA node6 CPU(s): 5,13,21,29
> > > > NUMA node7 CPU(s): 7,15,23,31
> > > >
> > > > For the full panic message (I masked some hostname info with xx),
> > > > please see the attachment.
> > > > In a short word, it seems a problem with nr_cpus, if without this
> > > > option, the kernel can bootup correctly.
> > >
> > > Yep.
> > > [ 0.007418] Early memory node ranges
> > > [ 0.007419] node 1: [mem 0x0000000000001000-0x000000000008efff]
> > > [ 0.007420] node 1: [mem 0x0000000000090000-0x000000000009ffff]
> > > [ 0.007422] node 1: [mem 0x0000000000100000-0x000000005c3d6fff]
> > > [ 0.007422] node 1: [mem 0x00000000643df000-0x0000000068ff7fff]
> > > [ 0.007423] node 1: [mem 0x000000006c528000-0x000000006fffffff]
> > > [ 0.007424] node 1: [mem 0x0000000100000000-0x000000047fffffff]
> > > [ 0.007425] node 5: [mem 0x0000000480000000-0x000000087effffff]
> > >
> > > There is clearly no node2. Where did the driver get the node2 from?
> > Since using nr_cpus=4 , the node2 is not be instanced by x86 initalizing code.
> > For the normal bootup, having the following:
> > [ 0.007704] Movable zone start for each node
> > [ 0.007707] Early memory node ranges
> > [ 0.007708] node 1: [mem 0x0000000000001000-0x000000000008efff]
> > [ 0.007709] node 1: [mem 0x0000000000090000-0x000000000009ffff]
> > [ 0.007711] node 1: [mem 0x0000000000100000-0x000000005c3d6fff]
> > [ 0.007712] node 1: [mem 0x00000000643df000-0x0000000068ff7fff]
> > [ 0.007712] node 1: [mem 0x000000006c528000-0x000000006fffffff]
> > [ 0.007713] node 1: [mem 0x0000000100000000-0x000000047fffffff]
> > [ 0.007714] node 5: [mem 0x0000000480000000-0x000000087effffff]
> > [ 0.008434] Zeroed struct page in unavailable ranges: 46490 pages
>
> Hmm, this is even more interesting. So even a normal boot doesn't have
> node 2. So where exactly does the device get its affinity from?
>
I am afraid that there is maybe some misunderstanding, but for the
normal bootup, the full boot msg shows the existence of node 2
First, the following can be observed:
[ 0.007385] SRAT: PXM 0 -> APIC 0x00 -> Node 0
[ 0.007386] SRAT: PXM 0 -> APIC 0x01 -> Node 0
[ 0.007387] SRAT: PXM 0 -> APIC 0x08 -> Node 0
[ 0.007388] SRAT: PXM 0 -> APIC 0x09 -> Node 0
[ 0.007389] SRAT: PXM 1 -> APIC 0x10 -> Node 1
[ 0.007390] SRAT: PXM 1 -> APIC 0x11 -> Node 1
[ 0.007391] SRAT: PXM 1 -> APIC 0x18 -> Node 1
[ 0.007392] SRAT: PXM 1 -> APIC 0x19 -> Node 1
[ 0.007393] SRAT: PXM 2 -> APIC 0x20 -> Node 2
[ 0.007394] SRAT: PXM 2 -> APIC 0x21 -> Node 2
[ 0.007395] SRAT: PXM 2 -> APIC 0x28 -> Node 2
[ 0.007396] SRAT: PXM 2 -> APIC 0x29 -> Node 2
[ 0.007397] SRAT: PXM 3 -> APIC 0x30 -> Node 3
[ 0.007398] SRAT: PXM 3 -> APIC 0x31 -> Node 3
[ 0.007399] SRAT: PXM 3 -> APIC 0x38 -> Node 3
[ 0.007400] SRAT: PXM 3 -> APIC 0x39 -> Node 3
[ 0.007401] SRAT: PXM 4 -> APIC 0x40 -> Node 4
[ 0.007402] SRAT: PXM 4 -> APIC 0x41 -> Node 4
[ 0.007403] SRAT: PXM 4 -> APIC 0x48 -> Node 4
[ 0.007403] SRAT: PXM 4 -> APIC 0x49 -> Node 4
[ 0.007404] SRAT: PXM 5 -> APIC 0x50 -> Node 5
[ 0.007405] SRAT: PXM 5 -> APIC 0x51 -> Node 5
[ 0.007406] SRAT: PXM 5 -> APIC 0x58 -> Node 5
[ 0.007407] SRAT: PXM 5 -> APIC 0x59 -> Node 5
[ 0.007408] SRAT: PXM 6 -> APIC 0x60 -> Node 6
[ 0.007409] SRAT: PXM 6 -> APIC 0x61 -> Node 6
[ 0.007410] SRAT: PXM 6 -> APIC 0x68 -> Node 6
[ 0.007411] SRAT: PXM 6 -> APIC 0x69 -> Node 6
[ 0.007412] SRAT: PXM 7 -> APIC 0x70 -> Node 7
[ 0.007413] SRAT: PXM 7 -> APIC 0x71 -> Node 7
[ 0.007414] SRAT: PXM 7 -> APIC 0x78 -> Node 7
[ 0.007414] SRAT: PXM 7 -> APIC 0x79 -> Node 7
Second:
[ 0.024497] NODE_DATA(2) allocated [mem 0x87ef4b000-0x87ef75fff]
[ 0.024498] NODE_DATA(2) on node 5
[ 0.024530] Initmem setup node 2 [mem 0x0000000000000000-0x0000000000000000]
Besides these, as I pasted, when normal bootup, lscpu shows the: NUMA
node2 CPU(s): 4,12,20,28
and
[root@...l-per7425-03 node]# cat has_cpu
0-7
[root@...l-per7425-03 node]# cat has_memory
1,5
So I think node 2 exists, which has cpus but no memory.
I attach SRAT, which show there are eight Proximity Domain. As my
understanding, each Proximity Domain will correspond to a numa node.
Cced acpi guys and hope them can give some hints.
> I suspect we are looking at two issues here. The first one, and a more
> important one is that there is a NUMA affinity configured for the device
> to a non-existing node. The second one is that nr_cpus affects
> initialization of possible nodes.
The dev->numa_node info is extracted from acpi table, not depends on
the instance of numa-node, which may be limited by nr_cpus. Hence the
node is existing, just not instanced.
Thanks,
Pingfan
View attachment "SRAT.dsl" of type "text/x-dsl" (23325 bytes)
Powered by blists - more mailing lists