Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Toke =?utf-8?Q?H=C3=B8iland-J=C3=B8rgensen?= <toke@redhat.com>
To:     Johannes Berg <johannes@sipsolutions.net>,
        linux-wireless@vger.kernel.org
Cc:     Johannes Berg <johannes.berg@intel.com>
Subject: Re: [PATCH] mac80211: don't apply flow control on management frames
In-Reply-To: <20210319232800.0e876c800866.Id2b66eb5a17f3869b776c39b5ca713272ea09d5d@changeid>
References: <20210319232800.0e876c800866.Id2b66eb5a17f3869b776c39b5ca713272ea09d5d@changeid>
Date:   Sat, 20 Mar 2021 01:13:34 +0100
Message-ID: <87h7l6adht.fsf@toke.dk>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Johannes Berg <johannes@sipsolutions.net> writes:

> From: Johannes Berg <johannes.berg@intel.com>
>
> In some cases (depending on the driver, but it's true e.g. for
> iwlwifi) we're using an internal TXQ for management packets,
> mostly to simplify the code and to have a place to queue them.
> However, it appears that in certain cases we can confuse the
> code and management frames are dropped, which is certainly not
> what we want.
>
> Short-circuit the processing of management frames. To keep the
> impact minimal, only put them on the frags queue and check the
> tid == management only for doing that and to skip the airtime
> fairness checks, if applicable.
>
> Signed-off-by: Johannes Berg <johannes.berg@intel.com>
> ---
>  net/mac80211/tx.c | 18 +++++++++++++++---
>  1 file changed, 15 insertions(+), 3 deletions(-)
>
> diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
> index 5d06de61047a..b2d09acb9fb0 100644
> --- a/net/mac80211/tx.c
> +++ b/net/mac80211/tx.c
> @@ -5,7 +5,7 @@
>   * Copyright 2006-2007	Jiri Benc <jbenc@suse.cz>
>   * Copyright 2007	Johannes Berg <johannes@sipsolutions.net>
>   * Copyright 2013-2014  Intel Mobile Communications GmbH
> - * Copyright (C) 2018-2020 Intel Corporation
> + * Copyright (C) 2018-2021 Intel Corporation
>   *
>   * Transmit and frame generation functions.
>   */
> @@ -1388,8 +1388,17 @@ static void ieee80211_txq_enqueue(struct ieee80211_local *local,
>  	ieee80211_set_skb_enqueue_time(skb);
>  
>  	spin_lock_bh(&fq->lock);
> -	fq_tin_enqueue(fq, tin, flow_idx, skb,
> -		       fq_skb_free_func);
> +	/*
> +	 * For management frames, don't really apply codel etc.,
> +	 * we don't want to apply any shaping or anything we just
> +	 * want to simplify the driver API by having them on the
> +	 * txqi.
> +	 */
> +	if (unlikely(txqi->txq.tid == IEEE80211_NUM_TIDS))
> +		__skb_queue_tail(&txqi->frags, skb);
> +	else
> +		fq_tin_enqueue(fq, tin, flow_idx, skb,
> +			       fq_skb_free_func);

One consequence of this is that we create a strict priority queue for
management frames. With all the possibilities for badness (such as the
ability of starving all other queues) that carries with it. I guess
that's probably fine for management frames, though, right? As in, there
is some other mechanism that prevents abuse of this?

-Toke