Thread (27 messages) flat view 27 messages, 5 authors, 2016-07-08

Re: [PATCH net-next 00/10] NCSI Support

From: Alexei Starovoitov <hidden>
Date: 2016-07-06 02:07:38

On Wed, Jul 06, 2016 at 07:42:39AM +1000, Benjamin Herrenschmidt wrote:
On Tue, 2016-07-05 at 10:44 -0700, Alexei Starovoitov wrote:
quoted
 .../...
quoted
quoted
quoted
The design for the patchset is highlighted as below:

   * The NCSI interface is abstracted with "struct ncsi_dev". It's registered
     when net_device is created, started to work by calling ncsi_start_dev()
     when net_device is opened (ndo_open()). For the first time, NCSI packets
     are sent and received to/from the far end (host in above figure) to probe
     available NCSI packages and channels. After that, one channel is chosen as
     active one to provide service.
   * The NCSI stack is driven by workqueue and state machine internally.
   * AEN (Asychronous Event Notification) might be received from the far end
     (host). The currently active NCSI channel fails over to another available
     one if possible. Otherwise, the NCSI channel is out of service.
   * NCSI stack should be configurable through netlink or another mechanism,
     but it's not implemented in this patchset. It's something TBD.
Gavin,
what configurations do you have in mind?
For ncsi itself or to control the nic on the host?
This set of patches is for BMC side, right?
What needs to be done on the host?
I'll respond for Gavin since I'm awake first ;-)

We use that stack today on OpenBMC on some OpenPOWER machines.

The configuration is thus for the above stack to run on the BMC in
order to control the host NIC.

NC-SI capable host NICs operate autonomously, so there is nothing to be
done on the host OS itself, at least not with the BCM NICs that we use
today, but of course the host NIC firmware needs to have the other side
of the stack.
Great! Thanks for clarifying.
So then future netlink api is mandatory to drive this kernel patches?
How one can use this set without it?
What is the main reason for this infra to be in the kernel instead of
userspace raw socket? Some interaction with the driver, right?
but it's not obvious from the patches.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help