Date:	Wed, 11 May 2016 15:39:17 +0200
From:	Hannes Frederic Sowa <hannes@...essinduktion.org>
To:	Eric Dumazet <eric.dumazet@...il.com>,
	Paolo Abeni <pabeni@...hat.com>
Cc:	Eric Dumazet <edumazet@...gle.com>,
	netdev <netdev@...r.kernel.org>,
	"David S. Miller" <davem@...emloft.net>,
	Jiri Pirko <jiri@...lanox.com>,
	Daniel Borkmann <daniel@...earbox.net>,
	Alexei Starovoitov <ast@...mgrid.com>,
	Alexander Duyck <aduyck@...antis.com>,
	Tom Herbert <tom@...bertland.com>,
	Peter Zijlstra <peterz@...radead.org>,
	Ingo Molnar <mingo@...nel.org>, Rik van Riel <riel@...hat.com>,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH 0/2] net: threadable napi poll loop

Hi all,

On 11.05.2016 15:08, Eric Dumazet wrote:
> On Wed, 2016-05-11 at 11:48 +0200, Paolo Abeni wrote:
>> Hi Eric,
>> On Tue, 2016-05-10 at 15:51 -0700, Eric Dumazet wrote:
>>> On Wed, 2016-05-11 at 00:32 +0200, Hannes Frederic Sowa wrote:
>>>
>>>> Not only did we want to present this solely as a bugfix but also as as
>>>> performance enhancements in case of virtio (as you can see in the cover
>>>> letter). Given that a long time ago there was a tendency to remove
>>>> softirqs completely, we thought it might be very interesting, that a
>>>> threaded napi in general seems to be absolutely viable nowadays and
>>>> might offer new features.
>>>
>>> Well, you did not fix the bug, you worked around by adding yet another
>>> layer, with another sysctl that admins or programs have to manage.
>>>
>>> If you have a special need for virtio, do not hide it behind a 'bug fix'
>>> but add it as a features request.
>>>
>>> This ksoftirqd issue is real and a fix looks very reasonable.
>>>
>>> Please try this patch, as I had very good success with it.
>>
>> Thank you for your time and your effort.
>>
>> I tested your patch on the bare metal "single core" scenario, disabling
>> the unneeded cores with:
>> CPUS=`nproc`
>> for I in `seq 1 $CPUS`; do echo 0  >  /sys/devices/system/node/node0/cpu$I/online; done
>>
>> And I got a:
>>
>> [   86.925249] Broke affinity for irq <num>
>>
> 
> Was it fatal, or simply a warning that you are removing the cpu that was
> the only allowed cpu in an affinity_mask ?
> 
> Looks another bug to fix then ? We disabled CPU hotplug here at Google
> for our production, as it was notoriously buggy. No time to fix dozens
> of issues added by a crowd of developers that do not even know a cpu can
> be unplugged.
> 
> Maybe some caller of local_bh_disable()/local_bh_enable() expected that
> current softirq would be processed. Obviously flaky even before the
> patches.

Yes, I fear this could come up. If we want to target net or stable maybe
we should maybe special case this patch specifically for net-rx?

>> for each irq number generated by a network device.
>>
>> In this scenario, your patch solves the ksoftirqd issue, performing
>> comparable to the napi threaded patches (with a negative delta in the
>> noise range) and introducing a minor regression with a single flow, in
>> the noise range (3%).
>>
>> As said in a previous mail, we actually experimented something similar,
>> but it felt quite hackish.
> 
> Right, we are networking guys, and we feel that messing with such core
> infra is not for us. So we feel comfortable adding a pure networking
> patch.

We posted this patch as an RFC. My initial internal proposal only had a
check in ___napi_schedule and completely relied on threaded irqs and
didn't spawn a thread per napi instance in the networking stack. I think
this is the better approach long term, as it allows to configure
threaded irqs per device and doesn't specifically deal with networking
only. NAPI must be aware of when to schedule, obviously, so we need
another check in napi_schedule.

My plan was definitely to go with something more generic, but we didn't
yet know how to express that in a generic way, but relied on the forced
threaded irqs kernel parameter.

>> AFAICS this patch adds three more tests in the fast path and affect all
>> other softirq use case. I'm not sure how to check for regression there.
> 
> It is obvious to me that ksoftird mechanism is not working as intended.

Yes.

> Fixing it might uncover bugs from parts of the kernel relying on the
> bug, indirectly or directly. Is it a good thing ?
> 
> I can not tell before trying.
> 
> Just by looking at /proc/{ksoftirqs_pid}/sched you can see the problem,
> as we normally schedule ksoftird under stress but most of the time,
> the softirq items were processed by another tasks as you found out.

Exactly, the pending mask gets reset by the task handling the softirq
inline and ksoftirqd runs dry too early not processing any more softirq
notifications.

>> The napi thread patches are actually a new feature, that also fixes the
>> ksoftirqd issue: hunting the ksoftirqd issue has been the initial
>> trigger for this work. I'm sorry for not being clear enough in the cover
>> letter.
>>
>> The napi thread patches offer additional benefits, i.e. an additional
>> relevant gain in the described test scenario, and do not impact on other
>> subsystems/kernel entities. 
>>
>> I still think they are worthy, and I bet you would disagree, but could
>> you please articulate more which parts concern you most and/or are more
>> bloated ?
> 
> Just look at the added code. napi_threaded_poll() is very buggy, but
> honestly I do not want to fix the bugs you added there. If you have only
> one vcpu, how jiffies can ever change since you block BH ?

I think the local_bh_disable/enable needs to be more fine granular,
correct (inside the loop).

> I was planning to remove cond_resched_softirq() that we no longer use
> after my recent changes to TCP stack,
> and you call it again (while it is obviously buggy since it does not
> check if a BH is pending, only if a thread needs the cpu)

Good point, thanks!

> I prefer fixing the existing code, really. It took us years to
> understand it and maybe fix it.

I agree, we should find a simple way to let ksoftirqd and netrx behave
better and target threaded napi as a new feature.

> Just think of what will happen if you have 10 devices (10 new threads in
> your model) and one cpu.
> 
> Instead of the nice existing netif_rx() doing 64 packets per device
> rounds, you'll now rely on process scheduler behavior that has no such
> granularity.

We didn't inspect all kinds of workloads right now but I hoped to see
benefits in forwarding with multiple interfaces. This is also why we
want to have some "easy" way to configure that for admins. Also
integration with RPS is on the todo list.

> Adding more threads is the natural answer of userland programmers, but
> in the kernel it is not the right answer. We already have mechanism,
> just use them and fix them if they are broken.
> 
> Sorry, I really do not think your patches are the way to go.
> But this thread is definitely interesting.

I am fine with that. It is us to show clear benefits or use cases for
that. If we fail with that, no problem at all that the patches get
rejected, we don't want to add bloat to the kernel, for sure! At this
point I still think a possibility to run napi in kthreads will allow
specific workloads to see an improvement. Maybe the simple branch in
napi_schedule is just worth so people can play around with it. As it
shouldn't change behavior we can later on simply remove it.

Thanks,
Hannes

Powered by blists - more mailing lists