Netdev Archive on lore.kernel.org
 help / color / Atom feed
From: Eric Dumazet <eric.dumazet@gmail.com>
To: Hayes Wang <hayeswang@realtek.com>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>
Cc: nic_swsd <nic_swsd@realtek.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH net-next] r8152: divide the tx and rx bottom functions
Date: Fri, 16 Aug 2019 10:19:33 +0200
Message-ID: <68015004-fb60-f6c6-05b0-610466223cf5@gmail.com> (raw)
In-Reply-To: <0835B3720019904CB8F7AA43166CEEB2F18D470D@RTITMBSVM03.realtek.com.tw>



On 8/16/19 10:10 AM, Hayes Wang wrote:
> Eric Dumazet [mailto:eric.dumazet@gmail.com]
>> Sent: Friday, August 16, 2019 2:40 PM
> [...]
>> tasklet and NAPI are scheduled on the same core (the current
>> cpu calling napi_schedule() or tasklet_schedule())
>>
>> I would rather not add this dubious tasklet, and instead try to understand
>> what is wrong in this driver ;)
>>
>> The various napi_schedule() calls are suspect IMO.
> 
> The original method as following.
> 
> static int r8152_poll(struct napi_struct *napi, int budget)
> {
> 	struct r8152 *tp = container_of(napi, struct r8152, napi);
> 	int work_done;
> 
> 	work_done = rx_bottom(tp, budget); <-- RX
> 	bottom_half(tp); <-- Tx (tx_bottom)
> 	[...]
> 
> The rx_bottom and tx_bottom would only be called in r8152_poll.
> That is, tx_bottom wouldn't be run unless rx_bottom is finished.
> And, rx_bottom would be called if tx_bottom is running.
> 
> If the traffic is busy. rx_bottom or tx_bottom may take a lot
> of time to deal with the packets. And the one would increase
> the latency time for the other one.
> 
> Therefore, when I separate the tx_bottom and rx_bottom to
> different tasklet and napi, the callback functions of tx and
> rx may schedule the tasklet and napi to different cpu. Then,
> the rx_bottom and tx_bottom may be run at the same time.


Your patch makes absolutely no guarantee of doing what you
want, I am afraid.

> 
> Take our arm platform for example. There are five cpus to
> handle the interrupt of USB host controller. When the rx is
> completed, cpu #1 may handle the interrupt and napi would
> be scheduled. When the tx is finished, cpu #2 may handle
> the interrupt and the tasklet is scheduled. Then, napi is
> run on cpu #1, and tasklet is run on cpu #2.
> 
>> Also rtl8152_start_xmit() uses skb_queue_tail(&tp->tx_queue, skb);
>>
>> But I see nothing really kicking the transmit if tx_free is empty ?
> 
> Tx callback function "write_bulk_callback" would deal with it.
> The callback function would check if there are packets waiting
> to be sent.

Which callback ?

After an idle period (no activity, no prior packets being tx-completed ...),
a packet is sent by the upper stack, enters the ndo_start_xmit() of a network driver.

This driver ndo_start_xmit() simply adds an skb to a local list, and returns.

Where/how is scheduled this callback ?

Some kind of timer ?
An (unrelated) incoming packet ?

> 
> 
> Best Regards,
> Hayes
> 
> 

  reply index

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-08-14  8:30 Hayes Wang
2019-08-15 20:58 ` David Miller
2019-08-16  2:59   ` Hayes Wang
2019-08-16  5:17     ` David Miller
2019-08-16  6:39 ` Eric Dumazet
2019-08-16  8:10   ` Hayes Wang
2019-08-16  8:19     ` Eric Dumazet [this message]
2019-08-16  9:08       ` Hayes Wang
2019-08-16  9:27         ` Eric Dumazet
2019-08-16 10:04           ` Hayes Wang

Reply instructions:

You may reply publically 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=68015004-fb60-f6c6-05b0-610466223cf5@gmail.com \
    --to=eric.dumazet@gmail.com \
    --cc=hayeswang@realtek.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=nic_swsd@realtek.com \
    /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

Netdev Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/netdev/0 netdev/git/0.git
	git clone --mirror https://lore.kernel.org/netdev/1 netdev/git/1.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 netdev netdev/ https://lore.kernel.org/netdev \
		netdev@vger.kernel.org netdev@archiver.kernel.org
	public-inbox-index netdev


Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.netdev


AGPL code for this site: git clone https://public-inbox.org/ public-inbox