[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <54db4c95-44a1-da74-a6d4-21d591926dbd@intel.com>
Date: Tue, 14 Jan 2020 08:47:13 -0800
From: Johnathan Mantey <johnathanx.mantey@...el.com>
To: samjonas <sam@...dozajonas.com>, netdev@...r.kernel.org
Cc: davem@...emloft.net
Subject: Re: [PATCH] Propagate NCSI channel carrier loss/gain events to the
kernel
Sam,
This code is working on a channel that is completely configured. This
code is covering a situation where the RJ45 cable is intentionally or
mistakenly removed from a system. In the event that the network cable is
removed/damaged/slips, the kernel needs to change state to show that
network interface no longer has a link. For my employers use case the
loss of carrier is then added to a log of system state changes. Thus for
each internal channel there needs to be a way to report that the channel
has/does not have a link carrier.
On 1/13/20 4:43 PM, samjonas wrote:
> On Fri, 2020-01-10 at 14:02 -0800, Johnathan Mantey wrote:
>> From 76d99782ec897b010ba507895d60d27dca8dca44 Mon Sep 17 00:00:00
>> 2001
>> From: Johnathan Mantey <johnathanx.mantey@...el.com>
>> Date: Fri, 10 Jan 2020 12:46:17 -0800
>> Subject: [PATCH] Propagate NCSI channel carrier loss/gain events to
>> the
>> kernel
>>
>> Problem statement:
>> Insertion or removal of a network cable attached to a NCSI controlled
>> network channel does not notify the kernel of the loss/gain of the
>> network link.
>>
>> The expectation is that /sys/class/net/eth(x)/carrier will change
>> state after a pull/insertion event. In addition the carrier_up_count
>> and carrier_down_count files should increment.
>>
>> Change statement:
>> Use the NCSI Asynchronous Event Notification handler to detect a
>> change in a NCSI link.
>> Add code to propagate carrier on/off state to the network interface.
>> The on/off state is only modified after the existing code identifies
>> if the network device HAD or HAS a link state change.
>
> If we set the carrier state off until we successfully configured a
> channel could we avoid this limitation?
>
> Cheers,
> Sam
>
>>
>> Test procedure:
>> Connected a L2 switch with only two ports connected.
>> One port was a DHCP corporate net, the other port attached to the
>> NCSI
>> controlled NIC.
>>
>> Starting with the L2 switch with DC on, check to make sure the NCSI
>> link is operating.
>> cat /sys/class/net/eth1/carrier
>> 1
>> cat /sys/class/net/eth1/carrier_up_count
>> 0
>> cat /sys/class/net/eth1/carrier_down_count
>> 0
>>
>> Remove DC from the L2 switch, and check link state
>> cat /sys/class/net/eth1/carrier
>> 0
>> cat /sys/class/net/eth1/carrier_up_count
>> 0
>> cat /sys/class/net/eth1/carrier_down_count
>> 1
>>
>> Restore DC to the L2 switch, and check link state
>> cat /sys/class/net/eth1/carrier
>> 1
>> cat /sys/class/net/eth1/carrier_up_count
>> 1
>> cat /sys/class/net/eth1/carrier_down_count
>> 1
>>
>> Signed-off-by: Johnathan Mantey <johnathanx.mantey@...el.com>
>> ---
>> net/ncsi/ncsi-aen.c | 6 ++++++
>> 1 file changed, 6 insertions(+)
>>
>> diff --git a/net/ncsi/ncsi-aen.c b/net/ncsi/ncsi-aen.c
>> index b635c194f0a8..274c415dcead 100644
>> --- a/net/ncsi/ncsi-aen.c
>> +++ b/net/ncsi/ncsi-aen.c
>> @@ -89,6 +89,12 @@ static int ncsi_aen_handler_lsc(struct
>> ncsi_dev_priv
>> *ndp,
>> if ((had_link == has_link) || chained)
>> return 0;
>>
>> + if (had_link) {
>> + netif_carrier_off(ndp->ndev.dev);
>> + } else {
>> + netif_carrier_on(ndp->ndev.dev);
>> + }
>> +
>> if (!ndp->multi_package && !nc->package->multi_channel) {
>> if (had_link) {
>> ndp->flags |= NCSI_DEV_RESHUFFLE;
>
--
Johnathan Mantey
Senior Software Engineer
*azad technology partners*
Contributing to Technology Innovation since 1992
Phone: (503) 712-6764
Email: johnathanx.mantey@...el.com <mailto:johnathanx.mantey@...el.com>
Powered by blists - more mailing lists