netdev - RE: [PATCH net 4/5] be2net: wait longer to reap TX compls in be

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: <a9644286-d102-4645-96c7-4961379f8033@CMEXHTCAS2.ad.emulex.com>
Date:	Mon, 24 Feb 2014 12:08:17 +0000
From:	Sathya Perla <Sathya.Perla@...lex.Com>
To:	David Laight <David.Laight@...LAB.COM>,
	Somnath Kotur <Somnath.Kotur@...lex.Com>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>
CC:	"davem@...emloft.net" <davem@...emloft.net>,
	Vasundhara Volam <Vasundhara.Volam@...lex.Com>
Subject: RE: [PATCH net 4/5] be2net: wait longer to reap TX compls in
 be_close()

> -----Original Message-----
> From: David Laight [mailto:David.Laight@...LAB.COM]
> 
> > From: Sathya Perla
> > > From: David Laight [mailto:David.Laight@...LAB.COM]
> > >
> > > From: Of Somnath Kotur
> > > > be_close() currently waits for a max of 200ms to receive all pending
> > > > TX compls. This timeout value was roughly calcuated based on 10G
> > > > transmission speeds and the TX queue depth. This timeout may not be
> > > > enough when the link is operating at lower speeds or in
> > > > multi-channel/SR-IOV configs with TX-rate limiting setting.
> > > > Increase the timeout to 2000ms and also bail-out if there is
> > > > a HW-error.
> > >
> > > What about monitoring whether transmits are actually completing,
> > > and giving up if none are completed within a suitable time period?
> > >
> >
> > David, the code already does that. All the queues are checked for
> > any new TX compls each 1ms in the be_close()->be_tx_compl_clean()
> > path. This patch is not changing that logic.
> 
> Yes, but it might be better to modify the logic rather than just
> increase the timeout.
> 
> The code is waiting for any pending transmits to complete.
> The purpose of the timeout is to give up if the transmits aren't going
> to complete.
> So instead of assuming that a specific interval is long enough, abort
> if no transmits completed in (say) a 10ms period.
> Move the sleep() to the top of the loop and abort if nothing is found
> on any queue.
> Oh and move the zeroing of cmpl and num_wrbs to a sensible place.

So you're saying, keep polling on completions till the HW has been
completely silent for a particular (you mention 10ms) period of time.

This does sound like a nicer scheme. Thanks!
Will implement this and re-post the patch.

-Sathya


--
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