netdev - Re: Question about TPA/HDS feature of bnxt

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

Message-ID: <CAMArcTW2yvEJLr_55G7FDsGtzKjTa2zMndOrAOBCshsW7UUj5A@mail.gmail.com>
Date: Wed, 14 Aug 2024 16:51:14 +0900
From: Taehee Yoo <ap420073@...il.com>
To: Michael Chan <michael.chan@...adcom.com>
Cc: Jakub Kicinski <kuba@...nel.org>, David Wei <dw@...idwei.uk>, 
	Somnath Kotur <somnath.kotur@...adcom.com>, Mina Almasry <almasrymina@...gle.com>, 
	Netdev <netdev@...r.kernel.org>
Subject: Re: Question about TPA/HDS feature of bnxt_en

On Wed, Aug 14, 2024 at 11:08 AM Michael Chan <michael.chan@...adcom.com> wrote:
>

Hi Michael,

> On Tue, Aug 13, 2024 at 6:17 PM Jakub Kicinski <kuba@...nel.org> wrote:
> >
> > On Tue, 13 Aug 2024 19:42:51 +0900 Taehee Yoo wrote:
> > > Hi,
> > > I'm currently testing the device memory TCP feature with the bnxt_en
> > > driver because Broadcom NICs support TPA/HDS, which is a mandatory
> > > feature for the devmem TCP.
> > > But it doesn't work for short-sized packets(under 300?)
> > > So, the devmem TCP stops or errors out if it receives non-header-splitted skb.
> > >
> > > I hope the bnxt_en driver or firmware has options that force TPA to
> > > work for short-sized packets.
> > > So, Can I get any condition information on TPA?
> >
> > I don't have any non-public info but look around the driver for
> > rx_copy_thresh, it seems to be sent to FW.
> >
>
> Yes, the rx_copy_thresh is also the HDS threshold. The default value
> is 256, meaning that packet sizes below 256 will not be split. So a
> 300-byte packet should be split.
>
> TPA is related but is separate. There is a min_agg_len that is
> currently set to 512 in bnxt_hwrm_vnic_set_tpa(). I think it should
> be fine to reduce this value for TPA to work on smaller packets.

Thank you so much for confirming the hds_threshold variable.
I tested this variable, it worked as I expected.
For testing, I kept the rx_copy_threshold and tpa settings unchanged,
but modified the hds_threshold variable to 0.

BTW, how about separating rx_copy_threshold into rx-copy-break
and hds_threshold?
If so, we can implement `ethtool --set-tunable eth0 rx-copybreak N`
and `ethtool -G  eth0 tcp-data-split on`

Thank you so much again,
Taehee Yoo