Warning after memory hotplug then online.

Warning after memory hotplug then online.

[Jonathan Cameron]

Hi All,

This is with some additional patches on top of the mm tree to support
arm64 memory hot plug, but this particular issue doesn't (at first glance)
seem to be connected to that.  It's not a recent issue as IIRC I
disabled Kconfig for cgroups when starting to work on this some time ago
as a quick and dirty work around for this.

arm64 defconfig + the bits and pieces needed for the hot plug patchset.

[  125.865690] WARNING: CPU: 13 PID: 294 at kernel/cgroup/cgroup.c:4346 
css_task_iter_start+0xb0/0xb8
[  125.883686] Modules linked in:
[  125.889808] CPU: 13 PID: 294 Comm: kworker/13:1 Not tainted 
4.19.0-rc4-mm1-00209-g70dc260f963a #912
[  125.907978] Hardware name: Huawei Taishan 2280 /D05, BIOS Hisilicon 
D05 IT20 Nemo 2.0 RC0 03/30/2018
[  125.926329] Workqueue: events cpuset_hotplug_workfn
[  125.945728] pc : css_task_iter_start+0xb0/0xb8
[  125.954641] lr : update_tasks_nodemask+0x78/0x128
[  125.964077] sp : ffff00000ae2bc40
[  125.970718] x29: ffff00000ae2bc40 x28: 0000000000000000
[  125.981379] x27: 0000000000000000 x26: ffff0000091579a0
[  125.992040] x25: ffff000009157aa8 x24: 0000000000000000
[  126.002700] x23: ffff0000092c5b58 x22: 0000000000000000
[  126.013361] x21: ffff0000091579a0 x20: ffff0000091579a0
[  126.026018] x19: ffff00000ae2bcc8 x18: 0000000000000400
[  126.036679] x17: 0000000000000000 x16: 0000000000000000
[  126.047339] x15: 0000000000000400 x14: 0000000000000400
[  126.057999] x13: 0000000000000000 x12: 0000000000000001
[  126.068659] x11: 0000000000000001 x10: 0000000000000960
[  126.079319] x9 : ffff00000ae2bd70 x8 : ffff8011f39d6680
[  126.089980] x7 : fefefefefefefeff x6 : 0000000000000000
[  126.100640] x5 : ffff000009139000 x4 : 0000000000000000
[  126.111300] x3 : ffff000009139000 x2 : ffff00000ae2bcc8
[  126.121960] x1 : 0000000000000000 x0 : 0000000000000000
[  126.132621] Call trace:
[  126.137517]  css_task_iter_start+0xb0/0xb8
[  126.145730]  update_tasks_nodemask+0x78/0x128
[  126.154469]  cpuset_hotplug_workfn+0x1b4/0x6b8
[  126.163383]  process_one_work+0x1e0/0x318
[  126.171422]  worker_thread+0x40/0x450
[  126.178763]  kthread+0x128/0x130
[  126.185232]  ret_from_fork+0x10/0x18
[  126.192397] ---[ end trace 08fa9eb01e348a8b ]---

I'm running with an initrd only and very minimal setup indeed.

There superficially doesn't seem to be anything to stop this path being called
after memory hotplug and we clearly shouldn't be doing this.

The cpuset_hotplug_workfn correctly identifies that we have some new memory.

Thoughts? (Or am I doing something stupid?)

Thanks,

Jonathan

Re: Warning after memory hotplug then online.

[Oscar Salvador]

On Mon, Sep 24, 2018 at 01:07:01PM +0100, Jonathan Cameron wrote:
> 
> Hi All,
> 
> This is with some additional patches on top of the mm tree to support
> arm64 memory hot plug, but this particular issue doesn't (at first glance)
> seem to be connected to that.  It's not a recent issue as IIRC I
> disabled Kconfig for cgroups when starting to work on this some time ago
> as a quick and dirty work around for this.

Hi Jonathan,

would you mind to describe the steps you are taking?
You are adding the memory, and then you online it?

-- 
Oscar Salvador
SUSE L3

Re: Warning after memory hotplug then online.

[Oscar Salvador]

On Mon, Sep 24, 2018 at 02:39:17PM +0200, Oscar Salvador wrote:
> On Mon, Sep 24, 2018 at 01:07:01PM +0100, Jonathan Cameron wrote:
> > 
> > Hi All,
> > 
> > This is with some additional patches on top of the mm tree to support
> > arm64 memory hot plug, but this particular issue doesn't (at first glance)
> > seem to be connected to that.  It's not a recent issue as IIRC I
> > disabled Kconfig for cgroups when starting to work on this some time ago
> > as a quick and dirty work around for this.
> 
> Hi Jonathan,
> 
> would you mind to describe the steps you are taking?
> You are adding the memory, and then you online it?

I forgot to ask.
Does this warning only show up with 4.19.0-rc4-mm1-00209-g70dc260f963a, or you can
trigger it with an older version?
Do you happen to know the last one that did not trigger that warning?

Thanks
-- 
Oscar Salvador
SUSE L3

Re: Warning after memory hotplug then online.

[Jonathan Cameron]

On Mon, 24 Sep 2018 14:39:17 +0200
Oscar Salvador <osalvador@techadventures.net> wrote:

> On Mon, Sep 24, 2018 at 01:07:01PM +0100, Jonathan Cameron wrote:
> > 
> > Hi All,
> > 
> > This is with some additional patches on top of the mm tree to support
> > arm64 memory hot plug, but this particular issue doesn't (at first glance)
> > seem to be connected to that.  It's not a recent issue as IIRC I
> > disabled Kconfig for cgroups when starting to work on this some time ago
> > as a quick and dirty work around for this.  
> 
> Hi Jonathan,
> 
> would you mind to describe the steps you are taking?
> You are adding the memory, and then you online it?
> 

Yes. Exactly that. 

I've hacked the efi memory map to give me 8GB of memory to play with.
I then use /sys/devices/system/memory/probe to hot add a section and
online via /sys/devices/nodes/devices/node3/memory80/online

Everything 'works', but this warning occurs.

This is with a forward ported version of the first
lot of hotplug patches that Andrea Reale, Maciej Bielski and Scott Branden 
wrote... (with NUMA and ACPI support added).

https://lkml.org/lkml/2017/4/11/536

Jonathan

Re: Warning after memory hotplug then online.

[Jonathan Cameron]

On Mon, 24 Sep 2018 14:42:03 +0200
Oscar Salvador <osalvador@techadventures.net> wrote:

> On Mon, Sep 24, 2018 at 02:39:17PM +0200, Oscar Salvador wrote:
> > On Mon, Sep 24, 2018 at 01:07:01PM +0100, Jonathan Cameron wrote:  
> > > 
> > > Hi All,
> > > 
> > > This is with some additional patches on top of the mm tree to support
> > > arm64 memory hot plug, but this particular issue doesn't (at first glance)
> > > seem to be connected to that.  It's not a recent issue as IIRC I
> > > disabled Kconfig for cgroups when starting to work on this some time ago
> > > as a quick and dirty work around for this.  
> > 
> > Hi Jonathan,
> > 
> > would you mind to describe the steps you are taking?
> > You are adding the memory, and then you online it?  
> 
> I forgot to ask.
> Does this warning only show up with 4.19.0-rc4-mm1-00209-g70dc260f963a, or you can
> trigger it with an older version?
> Do you happen to know the last one that did not trigger that warning?
> 
> Thanks

From memory perhaps 6 months back. Will take a while to backport the patch stack
far enough to try anything old now.

I'll look at this later in the week.

Thanks,

Jonathan