"blk-mq: Use llist_head for blk_cpu_done" causing warnings

"blk-mq: Use llist_head for blk_cpu_done" causing warnings

[Jens Axboe]

Hi Sebastian,

I'm running into these when that last patch is applied:

NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #10!!!

on at least my kvm test bed on the laptop. I'm going to revert this
series for now.

-- 
Jens Axboe

Re: "blk-mq: Use llist_head for blk_cpu_done" causing warnings

[Sebastian Andrzej Siewior]

On 2020-12-18 08:02:59 [-0700], Jens Axboe wrote:
> Hi Sebastian,
Hi Jens,

> I'm running into these when that last patch is applied:
> 
> NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #10!!!
> 
> on at least my kvm test bed on the laptop. I'm going to revert this
> series for now.

Okay. Can you keep #1 from the series?

Anyway, could you please throw a .config and the test in my inbox so I
can reproduce it and then I will look at this next year.

Sebastian

Re: "blk-mq: Use llist_head for blk_cpu_done" causing warnings

[Jens Axboe]

[-- Attachment #1: Type: text/plain, Size: 846 bytes --]

On 12/18/20 8:18 AM, Sebastian Andrzej Siewior wrote:
> On 2020-12-18 08:02:59 [-0700], Jens Axboe wrote:
>> Hi Sebastian,
> Hi Jens,
> 
>> I'm running into these when that last patch is applied:
>>
>> NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #10!!!
>>
>> on at least my kvm test bed on the laptop. I'm going to revert this
>> series for now.
> 
> Okay. Can you keep #1 from the series?

Sure, we can keep this one. I did verify that only the last part of
that series caused issues.

> Anyway, could you please throw a .config and the test in my inbox so I
> can reproduce it and then I will look at this next year.

Attached. The test is just booting my vm, haven't even started running
anything when it triggers. It also seems to be happening throughout
the vm running, at a pretty low rate though.

-- 
Jens Axboe


[-- Attachment #2: qemu-config.gz --]
[-- Type: application/gzip, Size: 20045 bytes --]

Re: "blk-mq: Use llist_head for blk_cpu_done" causing warnings

[Sebastian Andrzej Siewior]

On 2020-12-18 08:22:29 [-0700], Jens Axboe wrote:

Hi,

> > Anyway, could you please throw a .config and the test in my inbox so I
> > can reproduce it and then I will look at this next year.
> 
> Attached. The test is just booting my vm, haven't even started running
> anything when it triggers. It also seems to be happening throughout
> the vm running, at a pretty low rate though.

I tried those patches on my old -next tree, v5.11-rc1 and today's -next
tree. I booted your config with:
| exec qemu-system-x86_64  \
|         -m 4G \
|         -machine q35,accel=kvm \
|         -cpu host \
|         -smp 8 \
|         -kernel "$KERNEL" \
|         -drive file=Deb_sid_amd64-playground_xfs.img,if=virtio,id=diskroot,media=disk,format=raw,discard=on,snapshot=on \
|         -nographic \
|         -append "earlyprintk=ttyS0,115200 console=ttyS0,115200n8 root=/dev/vda1 sysrq_always_enabled ftrace_dump_on_oops nokaslr ignore_loglevel"
|

on an AMD and Intel (due to only CONFIG_CPU_SUP_INTEL=y). Nothing.
Is there anything special in your commandline / disk-setup? After a
complete disk-upgrade I have almost no block-softirqs:

|~# grep BLOCK /proc/softirqs
|       BLOCK:          0          0          0          0          4          0          8          0

Sebastian