IPMI Smbus

IPMI Smbus

[Patrick Venture]

Corey;

I saw you have a branch that is working on adding smbus IPMI support
(the ssif approach).

Can you provide details on the status of this work?

Patrick

Re: IPMI Smbus

[Corey Minyard]

On Tue, Apr 06, 2021 at 09:04:35AM -0700, Patrick Venture wrote:
> Corey;
> 
> I saw you have a branch that is working on adding smbus IPMI support
> (the ssif approach).
> 
> Can you provide details on the status of this work?

There is already an SMBus BMC device that can sit on the I2C bus.  It's
in mainstream qemu now.

It does not have smbus alert support, though.  I have patches for that,
but qemu doesn't have a flexibly general interrupt infrastructure, so I
hacked something in, and it's likely to be contentious.  The trouble is
that the interrupt is disconnected from any real device, and qemu
doesn't handle that very well.  And like the other things, I really only
created it for my testing.

For doing what you are trying to do, I don't think you need any special
support on either side for SSIF to work.

-corey

> 
> Patrick

Re: IPMI Smbus

[Patrick Venture]

On Tue, Apr 6, 2021 at 9:54 AM Corey Minyard <cminyard@mvista.com> wrote:
>
> On Tue, Apr 06, 2021 at 09:04:35AM -0700, Patrick Venture wrote:
> > Corey;
> >
> > I saw you have a branch that is working on adding smbus IPMI support
> > (the ssif approach).
> >
> > Can you provide details on the status of this work?
>
> There is already an SMBus BMC device that can sit on the I2C bus.  It's
> in mainstream qemu now.

This device:

github.com/qemu/qemu/blob/master/hw/ipmi/smbus_ipmi.c

So this device gets added to the BMC's i2c, and it's just that there's
no host-side support to then master this device?  in a multi-node
simulation.

I'm asking because my team is currently working on implementing ssif
support (between two nodes).

>
> It does not have smbus alert support, though.  I have patches for that,
> but qemu doesn't have a flexibly general interrupt infrastructure, so I
> hacked something in, and it's likely to be contentious.  The trouble is
> that the interrupt is disconnected from any real device, and qemu
> doesn't handle that very well.  And like the other things, I really only
> created it for my testing.
>
> For doing what you are trying to do, I don't think you need any special
> support on either side for SSIF to work.
>
> -corey
>
> >
> > Patrick

Re: IPMI Smbus

[Corey Minyard]

On Tue, Apr 06, 2021 at 10:06:09AM -0700, Patrick Venture wrote:
> On Tue, Apr 6, 2021 at 9:54 AM Corey Minyard <cminyard@mvista.com> wrote:
> >
> > On Tue, Apr 06, 2021 at 09:04:35AM -0700, Patrick Venture wrote:
> > > Corey;
> > >
> > > I saw you have a branch that is working on adding smbus IPMI support
> > > (the ssif approach).
> > >
> > > Can you provide details on the status of this work?
> >
> > There is already an SMBus BMC device that can sit on the I2C bus.  It's
> > in mainstream qemu now.
> 
> This device:
> 
> github.com/qemu/qemu/blob/master/hw/ipmi/smbus_ipmi.c
> 
> So this device gets added to the BMC's i2c, and it's just that there's
> no host-side support to then master this device?  in a multi-node
> simulation.

This works two ways, there is an internal, simple BMC simulator, or it
can make a connection to an external BMC.

> 
> I'm asking because my team is currently working on implementing ssif
> support (between two nodes).

Ah, there is already a protocol that I developed for this that others
are using already.  At Google I think, among others.  It's described in:
https://sourceforge.net/p/openipmi/code/ci/master/tree/lanserv/README.vm

The rest of that directory is a BMC simulator that can connect to qemu
and manage a qemu session, starting, stopping, monitoring, etc.  The
idea was to use IPMI to manage qemu, and some people are using it for
that.

There is obviously already a bridge on the host side to handle taking
the SSIF messages and transporting them over the wire to something else.
I know people were working on the BMC side, but I don't know the status.

-corey

> 
> >
> > It does not have smbus alert support, though.  I have patches for that,
> > but qemu doesn't have a flexibly general interrupt infrastructure, so I
> > hacked something in, and it's likely to be contentious.  The trouble is
> > that the interrupt is disconnected from any real device, and qemu
> > doesn't handle that very well.  And like the other things, I really only
> > created it for my testing.
> >
> > For doing what you are trying to do, I don't think you need any special
> > support on either side for SSIF to work.
> >
> > -corey
> >
> > >
> > > Patrick
>