VR control from the BMC

VR control from the BMC

[Ed Tanous]

In the near future, I'm going to have some needs for OpenBMC to be
able to control VRs.  These VRs might be on the baseboard itself, or
on detected PCIe add-in-cards, and controlled over PMBus.
Additionally, I will need a "hardware safe" way for the host to be
able to modify the behavior of these VRs (usually different voltage
settings), and to have that interface be constrained in such a way
that the host can never set a value that's outside of a predefined
range that's known to be safe for the hardware, which the BMC will own
the definition of for security purposes.

Does anyone else have similar needs?  I've been pointed to
phosphor-regulators which has some similar parallels;  As-is, that
solution won't work for me, because it's relying on fixed, platform
specific json scripting to accomplish its goals.  My hope would be for
a more generalized linux devicetree based solution, as well as a
representation on dbus that could be modified at runtime by
Redfish/IPMI in band.

Is there any other work I should look into that's similar?  Does
anyone have any strong opinions on how this should be organized or how
it could be built?

Thanks,

-Ed

Re: VR control from the BMC

[Vijay Khemka]

On 10/26/20, 4:04 PM, "openbmc on behalf of Ed Tanous" <openbmc-bounces+vijaykhemka=fb.com@lists.ozlabs.org on behalf of edtanous@google.com> wrote:

    In the near future, I'm going to have some needs for OpenBMC to be
    able to control VRs.  These VRs might be on the baseboard itself, or
    on detected PCIe add-in-cards, and controlled over PMBus.
    Additionally, I will need a "hardware safe" way for the host to be
    able to modify the behavior of these VRs (usually different voltage
    settings), and to have that interface be constrained in such a way
    that the host can never set a value that's outside of a predefined
    range that's known to be safe for the hardware, which the BMC will own
    the definition of for security purposes.

    Does anyone else have similar needs?  I've been pointed to
    phosphor-regulators which has some similar parallels;  As-is, that
    solution won't work for me, because it's relying on fixed, platform
    specific json scripting to accomplish its goals.  My hope would be for
    a more generalized linux devicetree based solution,
 
Voltage limits can certainly be passed via DTS file to limit user
application configuration setting.

    as well as a
    representation on dbus that could be modified at runtime by
    Redfish/IPMI in band.

    Is there any other work I should look into that's similar?  Does
    anyone have any strong opinions on how this should be organized or how
    it could be built?

I would also like to add VR firmware upgrades as well in this.

    Thanks,

    -Ed

Re: VR control from the BMC

[Ed Tanous]

On Tue, Oct 27, 2020 at 11:55 AM Vijay Khemka <vijaykhemka@fb.com> wrote:
>
> On 10/26/20, 4:04 PM, "openbmc on behalf of Ed Tanous" <openbmc-bounces+vijaykhemka=fb.com@lists.ozlabs.org on behalf of edtanous@google.com> wrote:
>
>     In the near future, I'm going to have some needs for OpenBMC to be
>     able to control VRs.  These VRs might be on the baseboard itself, or
>     on detected PCIe add-in-cards, and controlled over PMBus.
>     Additionally, I will need a "hardware safe" way for the host to be
>     able to modify the behavior of these VRs (usually different voltage
>     settings), and to have that interface be constrained in such a way
>     that the host can never set a value that's outside of a predefined
>     range that's known to be safe for the hardware, which the BMC will own
>     the definition of for security purposes.
>
>     Does anyone else have similar needs?  I've been pointed to
>     phosphor-regulators which has some similar parallels;  As-is, that
>     solution won't work for me, because it's relying on fixed, platform
>     specific json scripting to accomplish its goals.  My hope would be for
>     a more generalized linux devicetree based solution,
>
> Voltage limits can certainly be passed via DTS file to limit user
> application configuration setting.

Anything in the DTS would remove the possibility of controlling VRs
that weren't part of the baseboard, and would be a problem for me.
Also, these limits would likely need to be dynamic, and might rely on
certain calibration constants in the chip.  I don't think a hardcoded
DTS limit would meet my needs.

>
>     as well as a
>     representation on dbus that could be modified at runtime by
>     Redfish/IPMI in band.
>
>     Is there any other work I should look into that's similar?  Does
>     anyone have any strong opinions on how this should be organized or how
>     it could be built?
>
> I would also like to add VR firmware upgrades as well in this.

I don't have the need to upgrade VR firmware out of band today, but if
you're interested in helping define interfaces that would work for
firmware upgrade or make it easy to add in the future, I'd appreciate
the help.

Re: VR control from the BMC

[Vijay Khemka]

    I don't have the need to upgrade VR firmware out of band today, but if
    you're interested in helping define interfaces that would work for
    firmware upgrade or make it easy to add in the future, I'd appreciate
    the help.

I will certainly look into this as this is lower in my priority tasks.

Re: VR control from the BMC

[Patrick Williams]

[-- Attachment #1: Type: text/plain, Size: 1684 bytes --]

On Mon, Oct 26, 2020 at 04:02:22PM -0700, Ed Tanous wrote:
> In the near future, I'm going to have some needs for OpenBMC to be
> able to control VRs.  These VRs might be on the baseboard itself, or
> on detected PCIe add-in-cards, and controlled over PMBus.
> Additionally, I will need a "hardware safe" way for the host to be
> able to modify the behavior of these VRs (usually different voltage
> settings), and to have that interface be constrained in such a way
> that the host can never set a value that's outside of a predefined
> range that's known to be safe for the hardware, which the BMC will own
> the definition of for security purposes.
> 
> Does anyone else have similar needs?  I've been pointed to
> phosphor-regulators which has some similar parallels;  As-is, that
> solution won't work for me, because it's relying on fixed, platform
> specific json scripting to accomplish its goals.  My hope would be for
> a more generalized linux devicetree based solution, as well as a
> representation on dbus that could be modified at runtime by
> Redfish/IPMI in band.
> 
> Is there any other work I should look into that's similar?  Does
> anyone have any strong opinions on how this should be organized or how
> it could be built?
> 
> Thanks,
> 
> -Ed

Are these PMBus VRs?  I was able to program limits on a non-BMC project
I was working on using the PMBus interfaces to the VRs we used on that
project.

I did need to write this patch which I'm still suppose to clean up with
some additional clamp_val tweaks to avoid negative numbers.

https://lore.kernel.org/lkml/20191001160407.6265-1-alpawi@amazon.com/

-- 
Patrick Williams

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: VR control from the BMC

[Ed Tanous]

On Tue, Oct 27, 2020 at 1:26 PM Patrick Williams <patrick@stwcx.xyz> wrote:
>
> On Mon, Oct 26, 2020 at 04:02:22PM -0700, Ed Tanous wrote:
> > In the near future, I'm going to have some needs for OpenBMC to be
> > able to control VRs.  These VRs might be on the baseboard itself, or
> > on detected PCIe add-in-cards, and controlled over PMBus.
> > Additionally, I will need a "hardware safe" way for the host to be
> > able to modify the behavior of these VRs (usually different voltage
> > settings), and to have that interface be constrained in such a way
> > that the host can never set a value that's outside of a predefined
> > range that's known to be safe for the hardware, which the BMC will own
> > the definition of for security purposes.
> >
> > Does anyone else have similar needs?  I've been pointed to
> > phosphor-regulators which has some similar parallels;  As-is, that
> > solution won't work for me, because it's relying on fixed, platform
> > specific json scripting to accomplish its goals.  My hope would be for
> > a more generalized linux devicetree based solution, as well as a
> > representation on dbus that could be modified at runtime by
> > Redfish/IPMI in band.
> >
> > Is there any other work I should look into that's similar?  Does
> > anyone have any strong opinions on how this should be organized or how
> > it could be built?
> >
> > Thanks,
> >
> > -Ed
>
> Are these PMBus VRs?

Yes.

>  I was able to program limits on a non-BMC project
> I was working on using the PMBus interfaces to the VRs we used on that
> project.
>
> I did need to write this patch which I'm still suppose to clean up with
> some additional clamp_val tweaks to avoid negative numbers.
>
> https://lore.kernel.org/lkml/20191001160407.6265-1-alpawi@amazon.com/

That pointer helps.  Thanks.

>
> --
> Patrick Williams