netdev - Re: Question on i40e PCIe relaxed ordering (RO)

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: <1467232362.3733.23.camel@intel.com>
Date:	Wed, 29 Jun 2016 13:32:42 -0700
From:	Jeff Kirsher <jeffrey.t.kirsher@...el.com>
To:	Greg <gvrose8192@...il.com>, tndave <tushar.n.dave@...cle.com>
Cc:	intel-wired-lan@...ts.osuosl.org, netdev <netdev@...r.kernel.org>,
	"e1000-devel@...ts.sf.net" <e1000-devel@...ts.sf.net>
Subject: Re: Question on i40e PCIe relaxed ordering (RO)

On Wed, 2016-06-29 at 13:18 -0700, Greg wrote:
> On Wed, 2016-06-29 at 13:05 -0700, tndave wrote:
> > Hi,
> > 
> > Running iperf tcp test on 2 sparc systems with i40e connected back to
> > back, I see huge number of 'port.rx_dropped' (on iperf server). Based
> on
> > past experience with ixgbe, this could very well because of PCIe RO
> > (relaxed ordering) not enabled.
> > 
> > I am trying to confirm RO is enabled. i40e datasheet mentioned RO
> > settings in 3 different sections:
> > 
> > 1. section 10.2.2.2.38 PCIe Global Config 2 - GLPCI_CNF2 register
> > contains global status fields of PCIe configuration. The bit 0 of the
> > register is "RO_DIS". If this bit is set to 1 RO is disabled.
> > 
> > RO_DIS in my setup is 0 imply RO is not disabled.
> > 
> > 2. section 12.3.5.5 Device Control Register (0xA8; RW) has bit 4
> > that enable/disable RO. This is pcie cap register.
> > 
> > In my i40e pcie config space value at offset 0xA8 is "2110". i.e 4th
> bit
> > set to 1 imply RO is enabled.
> > 
> > 3. section 3.1.2.7.2 mentions some relaxed ordering rules
> > e.g. "The GLLAN_RCTL.RXDESCRDROEN bit (loaded from NVM) enables relaxed
> > ordering for Rx descriptor reads"
> > 
> > However, GLLAN_RCTL register definition has not bit like RXDESCRDROEN.
> > Same goes for GLLAN_TCTL.TXDESCRDROEN.
> > 
> > Am I missing anything? please advise.
> 
> I would try posting this question to the e1000 developer list over at
> Source Forge.  The Intel customer support folks used to monitor that
> list closely when I was there, hopefully they still are.

We are supposed to be monitoring both lists, but just in case I have added
e1000-devel list...
Download attachment "signature.asc" of type "application/pgp-signature" (820 bytes)