netdev - Re: [PATCH net-next] net: gro: add a per device gro flush timer

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]

Date:	Thu, 06 Nov 2014 20:15:14 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	David Miller <davem@...emloft.net>
Cc:	netdev@...r.kernel.org, ogerlitz@...lanox.com, willemb@...gle.com
Subject: Re: [PATCH net-next] net: gro: add a per device gro flush timer

On Thu, 2014-11-06 at 22:36 -0500, David Miller wrote:
> From: Eric Dumazet <eric.dumazet@...il.com>
> Date: Thu, 06 Nov 2014 14:11:20 -0800
> 
> > My goal was to not change any driver, doing a generic change.
> > 
> > Drivers call napi_complete() in their rx napi handler without giving us
> > the 'work_done' value which tells us if a packet was processed.
> > 
> > So I added a counter that is increased for every packet given to GRO
> > engine (napi_rx_count), so that napi_complete() has a clue if a packet
> > was received in _this_ NAPI run.
> > 
> > If at least one packet was received (and we still have packets in
> > gro_list) -> We ream the 'timer'
> > If not, we flush packets in GRO engine.
> > 
> > In order to avoid this state, I would have to add a new method, like
> > napi_complete_done(napi, work_done), and change drivers. I am not sure
> > its worth the effort ?
> 
> I think for such a critical path in the kernel it's worth it to avoid
> these increments for every packet, just to compute a value that's
> sitting in a register already in the driver's poll routine.
> 
> Eric, you've been trimming cpu IRQ disables from these exact code
> paths, you should know better than me :-)
> 
> I'm willing to do some of the monkey work of converting as many
> drivers as can be trivially done if you want.  Almost all of the
> ones I looked at have the work_done variable right there at the
> napi_complete() call site.
> 
> The rest can stay unconverted and not get access to this new facility.

This is your call. I actually did this in my first implementation,
because I was not using a timer but rescheduling NAPI and not
re-enabling interrupts.

http://www.spinics.net/lists/netdev/msg302474.html

(This is why I had to cook "d75b1ade567 net: less interrupt masking in
NAPI")

Given the cost of GRO processing, setting a bool (this is all I need
actually) is really pure noise.



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html