netdev - Re: [PATCH net-next] liquidio: xmit

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-Id: <20171020.131836.933570982545888951.davem@davemloft.net>
Date:   Fri, 20 Oct 2017 13:18:36 +0100 (WEST)
From:   David Miller <davem@...emloft.net>
To:     felix.manlunas@...ium.com
Cc:     netdev@...r.kernel.org, raghu.vatsavayi@...ium.com,
        derek.chickles@...ium.com, satananda.burla@...ium.com
Subject: Re: [PATCH net-next] liquidio: xmit_more support

From: Felix Manlunas <felix.manlunas@...ium.com>
Date: Wed, 18 Oct 2017 10:36:02 -0700

> From: Intiyaz Basha <intiyaz.basha@...ium.com>
> 
> Do not write the Tx doorbell if skb->xmit_more is set unless the Tx
> queue is full or stopped.
> 
> Signed-off-by: Intiyaz Basha <intiyaz.basha@...ium.com>
> Signed-off-by: Derek Chickles <derek.chickles@...ium.com>
> Signed-off-by: Felix Manlunas <felix.manlunas@...ium.com>

You're going to need some kind of added heuristic for this.

If the TX queue is idle, and then you get a full TX queue's
worth of xmit_more frames, you do not want to defer the doorbell
until the entire queue has been filled.

But that is what you are doing.

You have to experment with different deferral limit values to
see where latency starts to be negatively impacted.