All of lore.kernel.org
 help / color / mirror / Atom feed
From: Felix Fietkau <nbd@nbd.name>
To: Rajkumar Manoharan <rmanohar@codeaurora.org>,
	Rakesh Pillai <pillair@codeaurora.org>
Cc: ath10k@lists.infradead.org, linux-wireless@vger.kernel.org,
	linux-kernel@vger.kernel.org, kvalo@codeaurora.org,
	johannes@sipsolutions.net, davem@davemloft.net, kuba@kernel.org,
	netdev@vger.kernel.org, dianders@chromium.org,
	evgreen@chromium.org
Subject: Re: [RFC 2/7] ath10k: Add support to process rx packet in thread
Date: Wed, 22 Jul 2020 14:27:42 +0200	[thread overview]
Message-ID: <d79ae05e-e75a-de2f-f2e3-bc73637e1501@nbd.name> (raw)
In-Reply-To: <13573549c277b34d4c87c471ff1a7060@codeaurora.org>

On 2020-07-21 23:53, Rajkumar Manoharan wrote:
> On 2020-07-21 10:14, Rakesh Pillai wrote:
>> NAPI instance gets scheduled on a CPU core on which
>> the IRQ was triggered. The processing of rx packets
>> can be CPU intensive and since NAPI cannot be moved
>> to a different CPU core, to get better performance,
>> its better to move the gist of rx packet processing
>> in a high priority thread.
>> 
>> Add the init/deinit part for a thread to process the
>> receive packets.
>> 
> IMHO this defeat the whole purpose of NAPI. Originally in ath10k
> irq processing happened in tasklet (high priority) context which in
> turn push more data to net core even though net is unable to process
> driver data as both happen in different context (fast producer - slow 
> consumer)
> issue. Why can't CPU governor schedule the interrupts in less loaded CPU 
> core?
> Otherwise you can play with different RPS and affinity settings to meet 
> your
> requirement.
> 
> IMO introducing high priority tasklets/threads is not viable solution.
I'm beginning to think that the main problem with NAPI here is that the
work done by poll functions on 802.11 drivers is significantly more CPU
intensive compared to ethernet drivers, possibly more than what NAPI was
designed for.

I'm considering testing a different approach (with mt76 initially):
- Add a mac80211 rx function that puts processed skbs into a list
instead of handing them to the network stack directly.
- Move all rx processing to a high priority thread, keep a driver
internal queue for fully processed packets.
- Schedule NAPI poll on completion.
- NAPI poll function pulls from the internal queue and passes to the
network stack.

With this approach, the network stack retains some control over the
processing rate of rx packets, while the scheduler can move the CPU
intensive processing around to where it fits best.

What do you think?

- Felix

WARNING: multiple messages have this Message-ID (diff)
From: Felix Fietkau <nbd@nbd.name>
To: Rajkumar Manoharan <rmanohar@codeaurora.org>,
	Rakesh Pillai <pillair@codeaurora.org>
Cc: netdev@vger.kernel.org, linux-wireless@vger.kernel.org,
	linux-kernel@vger.kernel.org, ath10k@lists.infradead.org,
	dianders@chromium.org, evgreen@chromium.org, kuba@kernel.org,
	johannes@sipsolutions.net, davem@davemloft.net,
	kvalo@codeaurora.org
Subject: Re: [RFC 2/7] ath10k: Add support to process rx packet in thread
Date: Wed, 22 Jul 2020 14:27:42 +0200	[thread overview]
Message-ID: <d79ae05e-e75a-de2f-f2e3-bc73637e1501@nbd.name> (raw)
In-Reply-To: <13573549c277b34d4c87c471ff1a7060@codeaurora.org>

On 2020-07-21 23:53, Rajkumar Manoharan wrote:
> On 2020-07-21 10:14, Rakesh Pillai wrote:
>> NAPI instance gets scheduled on a CPU core on which
>> the IRQ was triggered. The processing of rx packets
>> can be CPU intensive and since NAPI cannot be moved
>> to a different CPU core, to get better performance,
>> its better to move the gist of rx packet processing
>> in a high priority thread.
>> 
>> Add the init/deinit part for a thread to process the
>> receive packets.
>> 
> IMHO this defeat the whole purpose of NAPI. Originally in ath10k
> irq processing happened in tasklet (high priority) context which in
> turn push more data to net core even though net is unable to process
> driver data as both happen in different context (fast producer - slow 
> consumer)
> issue. Why can't CPU governor schedule the interrupts in less loaded CPU 
> core?
> Otherwise you can play with different RPS and affinity settings to meet 
> your
> requirement.
> 
> IMO introducing high priority tasklets/threads is not viable solution.
I'm beginning to think that the main problem with NAPI here is that the
work done by poll functions on 802.11 drivers is significantly more CPU
intensive compared to ethernet drivers, possibly more than what NAPI was
designed for.

I'm considering testing a different approach (with mt76 initially):
- Add a mac80211 rx function that puts processed skbs into a list
instead of handing them to the network stack directly.
- Move all rx processing to a high priority thread, keep a driver
internal queue for fully processed packets.
- Schedule NAPI poll on completion.
- NAPI poll function pulls from the internal queue and passes to the
network stack.

With this approach, the network stack retains some control over the
processing rate of rx packets, while the scheduler can move the CPU
intensive processing around to where it fits best.

What do you think?

- Felix

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

  reply	other threads:[~2020-07-22 12:28 UTC|newest]

Thread overview: 96+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-07-21 17:14 [RFC 0/7] Add support to process rx packets in thread Rakesh Pillai
2020-07-21 17:14 ` Rakesh Pillai
2020-07-21 17:14 ` [RFC 1/7] mac80211: Add check for napi handle before WARN_ON Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-22 12:56   ` Johannes Berg
2020-07-22 12:56     ` Johannes Berg
2020-07-23 18:26     ` Rakesh Pillai
2020-07-23 18:26       ` Rakesh Pillai
2020-07-23 20:06       ` Johannes Berg
2020-07-23 20:06         ` Johannes Berg
2020-07-24  6:21         ` Rakesh Pillai
2020-07-24  6:21           ` Rakesh Pillai
2020-07-26 16:19         ` Rakesh Pillai
2020-07-26 16:19           ` Rakesh Pillai
2020-07-30 12:40           ` Johannes Berg
2020-07-30 12:40             ` Johannes Berg
2020-07-21 17:14 ` [RFC 2/7] ath10k: Add support to process rx packet in thread Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-21 21:53   ` Rajkumar Manoharan
2020-07-21 21:53     ` Rajkumar Manoharan
2020-07-22 12:27     ` Felix Fietkau [this message]
2020-07-22 12:27       ` Felix Fietkau
2020-07-22 12:55       ` Johannes Berg
2020-07-22 12:55         ` Johannes Berg
2020-07-22 13:00         ` Felix Fietkau
2020-07-22 13:00           ` Felix Fietkau
2020-07-23  6:09           ` Rajkumar Manoharan
2020-07-23  6:09             ` Rajkumar Manoharan
2021-03-22 23:57           ` Ben Greear
2021-03-22 23:57             ` Ben Greear
2021-03-23  1:20             ` Brian Norris
2021-03-23  1:20               ` Brian Norris
2021-03-23  3:01               ` Ben Greear
2021-03-23  3:01                 ` Ben Greear
2021-03-23  7:45                 ` Felix Fietkau
2021-03-23  7:45                   ` Felix Fietkau
2021-03-25  9:45                   ` Rakesh Pillai
2021-03-25  9:45                     ` Rakesh Pillai
2021-03-25 10:33                     ` Felix Fietkau
2021-03-25 10:33                       ` Felix Fietkau
2020-07-23 18:25     ` Rakesh Pillai
2020-07-23 18:25       ` Rakesh Pillai
2020-07-24 23:11       ` Jacob Keller
2020-07-24 23:11         ` Jacob Keller
2020-07-21 17:14 ` [RFC 3/7] ath10k: Add module param to enable rx thread Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-21 17:14 ` [RFC 4/7] ath10k: Do not exhaust budget on process tx completion Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-21 17:14 ` [RFC 5/7] ath10k: Handle the rx packet processing in thread Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-21 17:14 ` [RFC 6/7] ath10k: Add deliver to stack from thread context Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-21 17:14 ` [RFC 7/7] ath10k: Handle rx thread suspend and resume Rakesh Pillai
2020-07-21 17:14   ` Rakesh Pillai
2020-07-23 23:06   ` Sebastian Gottschall
2020-07-23 23:06     ` Sebastian Gottschall
2020-07-24  6:19     ` Rakesh Pillai
2020-07-24  6:19       ` Rakesh Pillai
2020-07-21 17:25 ` [RFC 0/7] Add support to process rx packets in thread Andrew Lunn
2020-07-21 17:25   ` Andrew Lunn
2020-07-21 18:05   ` Florian Fainelli
2020-07-21 18:05     ` Florian Fainelli
2020-07-23 18:21     ` Rakesh Pillai
2020-07-23 18:21       ` Rakesh Pillai
2020-07-23 19:02       ` Florian Fainelli
2020-07-23 19:02         ` Florian Fainelli
2020-07-24  6:20         ` Rakesh Pillai
2020-07-24  6:20           ` Rakesh Pillai
2020-07-24 22:28           ` Florian Fainelli
2020-07-24 22:28             ` Florian Fainelli
2020-07-22  9:12   ` David Laight
2020-07-22  9:12     ` David Laight
2020-07-25  8:16     ` Hillf Danton
2020-07-25 10:38       ` Sebastian Gottschall
2020-07-25 10:38         ` Sebastian Gottschall
2020-07-25 12:25         ` Hillf Danton
2020-07-25 14:08         ` Sebastian Gottschall
2020-07-25 14:08           ` Sebastian Gottschall
2020-07-25 14:57           ` Hillf Danton
2020-07-25 15:41             ` Sebastian Gottschall
2020-07-25 15:41               ` Sebastian Gottschall
2020-07-26 11:16               ` David Laight
2020-07-26 11:16                 ` David Laight
2020-07-28 16:59                 ` Rakesh Pillai
2020-07-28 16:59                   ` Rakesh Pillai
2020-07-29  1:34                   ` Hillf Danton
2020-07-25 17:57       ` Felix Fietkau
2020-07-25 17:57         ` Felix Fietkau
2020-07-26  1:22         ` Hillf Danton
2020-07-26  8:10           ` Felix Fietkau
2020-07-26  8:10             ` Felix Fietkau
2020-07-26  8:32             ` Hillf Danton
2020-07-26  8:59               ` Felix Fietkau
2020-07-26  8:59                 ` Felix Fietkau
2020-07-22 16:20   ` Jakub Kicinski
2020-07-22 16:20     ` Jakub Kicinski

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=d79ae05e-e75a-de2f-f2e3-bc73637e1501@nbd.name \
    --to=nbd@nbd.name \
    --cc=ath10k@lists.infradead.org \
    --cc=davem@davemloft.net \
    --cc=dianders@chromium.org \
    --cc=evgreen@chromium.org \
    --cc=johannes@sipsolutions.net \
    --cc=kuba@kernel.org \
    --cc=kvalo@codeaurora.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-wireless@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=pillair@codeaurora.org \
    --cc=rmanohar@codeaurora.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.