Exposing BIOS attributes to a remote management console

Exposing BIOS attributes to a remote management console

[Deepak Kodihalli]

Hi,

To enable remote updates (via a remote management console that talks to 
the BMC) to the host firmware's BIOS settings table on IBM systems, 
we're thinking of the following flow:

a) The host firmware sends down a BIOS settings table to the BMC using 
PLDM [1].
b) The BMC uses phosphor-settingsd [2] as the backend to store BIOS 
attributes on D-Bus.
c) Bmcweb implements the Redfish BIOS schema [3] to enable remote 
reads/writes on the BIOS attributes.

However, c) is a problem because one needs to write YAML files [4] for 
the BMC to determine what D-Bus objects to make corresponding to the 
BIOS attributes. This requires a unique D-Bus interface *per* BIOS 
attribute. This means the BMC must have prior knowledge about the BIOS 
attributes.

I don't think that's the right way to go about this for two reasons. One 
- this creates a lockstep dependency on the host firmware when the BIOS 
settings table needs to be updated, and two - I think the OpenBMC 
implementation of this must be able to receive (via PLDM/IPMI/other 
standard in-band means) a set of BIOS attributes from different BIOS 
firmware stacks dynamically and expose them for out of band updates, 
without having prior/build-time knowledge of those attributes. So I 
think this calls for a different kind of D-Bus interface/infrastructure 
than what the phosphor-settingsd app relies on. Something that enables 
the BMC to add to D-Bus a BIOS attribute dynamically, knowing it's name, 
type and default value.

Any thoughts on this flow? I'm also curious to know if the Redfish BIOS 
schema/attribute registry model requires the BMC to have prior knowledge 
about the system BIOS attributes to update the registry. This wasn't 
obvious to me from a quick read of the schema.

Thanks,
Deepak

[1] 
https://www.dmtf.org/sites/default/files/standards/documents/DSP0247_1.0.0.pdf
[2] https://github.com/openbmc/phosphor-settingsd
[3] https://redfish.dmtf.org/schemas/v1/Bios.v1_1_0.json
[4] 
https://github.com/openbmc/openbmc/blob/master/meta-phosphor/recipes-phosphor/settings/phosphor-settings-defaults/defaults.yaml

Re: FW: Exposing BIOS attributes to a remote management console

[Sekar, Suryakanth]

[-- Attachment #1: Type: text/plain, Size: 3878 bytes --]

Hi Deepak,

OOB BIOS high level config design doc is in progress. I will send for 
review soon.

Yes, we need to support all interface -IPMI/ MCTP/ Redfish Host interface.

If BIOS is not booted at all. BMC wont have any attribute info. It will 
be empty in BMC side.

BIOS must provide BIOS OOB capability via KCS interface in early boot stage.
Like supported proprietary BIOS attribute file format or PLDM support 
via MCTP  or
Redfish BIOS attribute registry format to the BMC.

BIOS must send the master BIOS attributes file(Intel properitary - XML 
type 0)
via KCS interface or all attributes details via Bios configuration PLDM 
via MCTP or
Redfish host interface during first boot.

BIOS must collect the new attribute values from BMC and update the same 
in BIOS

BIOS must send the updated master attributes file to the BMC and once 
its updated in BIOS

and reset the system to reflect the BIOS configuration.

OOB daemon in BMC  must maintain and collect the  attribute registry 
file from MCTP/ IPMI/ Redfish interface.

Convert the proprietary XML format/ PLDM data  into BIOS attribute 
Registry format &
must support the below following dbus method.

https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-dbus-interfaces/+/18242

Validate the BIOS input data or User attribute value.

MCTP / Redfish host interface / IPMI must communicate with OOB daemon in 
BMC via D bus

Thanks

Suryakanth.S



> ---------- Forwarded message ----------
> From: Deepak Kodihalli <dkodihal@linux.vnet.ibm.com 
> <mailto:dkodihal@linux.vnet.ibm.com>>
> Date: Feb 8, 2020 1:59 PM
> Subject: Exposing BIOS attributes to a remote management console
> To: openbmc@lists.ozlabs.org <mailto:openbmc@lists.ozlabs.org>
> Cc:
>
> Hi,
>
> To enable remote updates (via a remote management console that talks to
> the BMC) to the host firmware's BIOS settings table on IBM systems,
> we're thinking of the following flow:
>
> a) The host firmware sends down a BIOS settings table to the BMC using
> PLDM [1].
> b) The BMC uses phosphor-settingsd [2] as the backend to store BIOS
> attributes on D-Bus.
> c) Bmcweb implements the Redfish BIOS schema [3] to enable remote
> reads/writes on the BIOS attributes.
>
> However, c) is a problem because one needs to write YAML files [4] for
> the BMC to determine what D-Bus objects to make corresponding to the
> BIOS attributes. This requires a unique D-Bus interface *per* BIOS
> attribute. This means the BMC must have prior knowledge about the BIOS
> attributes.
>
> I don't think that's the right way to go about this for two reasons. One
> - this creates a lockstep dependency on the host firmware when the BIOS
> settings table needs to be updated, and two - I think the OpenBMC
> implementation of this must be able to receive (via PLDM/IPMI/other
> standard in-band means) a set of BIOS attributes from different BIOS
> firmware stacks dynamically and expose them for out of band updates,
> without having prior/build-time knowledge of those attributes. So I
> think this calls for a different kind of D-Bus interface/infrastructure
> than what the phosphor-settingsd app relies on. Something that enables
> the BMC to add to D-Bus a BIOS attribute dynamically, knowing it's name,
> type and default value.
>
> Any thoughts on this flow? I'm also curious to know if the Redfish BIOS
> schema/attribute registry model requires the BMC to have prior knowledge
> about the system BIOS attributes to update the registry. This wasn't
> obvious to me from a quick read of the schema.
>
> Thanks,
> Deepak
>
> [1]
> https://www.dmtf.org/sites/default/files/standards/documents/DSP0247_1.0.0.pdf
> [2] https://github.com/openbmc/phosphor-settingsd
> [3] https://redfish.dmtf.org/schemas/v1/Bios.v1_1_0.json
> [4]
> https://github.com/openbmc/openbmc/blob/master/meta-phosphor/recipes-phosphor/settings/phosphor-settings-defaults/defaults.yaml
>

[-- Attachment #2: Type: text/html, Size: 6798 bytes --]

Re: FW: Exposing BIOS attributes to a remote management console

[Deepak Kodihalli]

On 10/02/20 4:19 pm, Sekar, Suryakanth wrote:
> Hi Deepak,
> 
> OOB BIOS high level config design doc is in progress. I will send for 
> review soon.


Hi Surya,

Thanks for responding. I'm glad that we have the same understanding on 
this topic.

I'll look out for the design doc from you. Can you share now what 
aspects that document would cover - redfish/in-band/D-Bus implementation 
- all of this? Is Intel planning on writing code for these areas soon? 
We're planning on implementing the PLDM BIOS spec for the same purpose. 
Some PLDM BIOS code's out there on openbmc/pldm already.

> Yes, we need to support all interface -IPMI/ MCTP/ Redfish Host interface.
> 
> If BIOS is not booted at all. BMC wont have any attribute info. It will 
> be empty in BMC side.
> 
> BIOS must provide BIOS OOB capability via KCS interface in early boot stage.
> Like supported proprietary BIOS attribute file format or PLDM support 
> via MCTP  or
> Redfish BIOS attribute registry format to the BMC.
> 
> BIOS must send the master BIOS attributes file(Intel properitary - XML 
> type 0)
> via KCS interface or all attributes details via Bios configuration PLDM 
> via MCTP or
> Redfish host interface during first boot.
> 
> BIOS must collect the new attribute values from BMC and update the same 
> in BIOS
> 
> BIOS must send the updated master attributes file to the BMC and once 
> its updated in BIOS
> 
> and reset the system to reflect the BIOS configuration.
> 
> OOB daemon in BMC  must maintain and collect the  attribute registry 
> file from MCTP/ IPMI/ Redfish interface.

All this sounds good to me.

> Convert the proprietary XML format/ PLDM data  into BIOS attribute 
> Registry format &
> must support the below following dbus method.
> 
> https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-dbus-interfaces/+/18242
> 
> Validate the BIOS input data or User attribute value.
> 
> MCTP / Redfish host interface / IPMI must communicate with OOB daemon in 
> BMC via D bus

Are you targeting specific implementation areas here? By knowing that, 
we can collaborate better.


> Thanks
> 
> Suryakanth.S

Regards,
Deepak

> 
> 
>> ---------- Forwarded message ----------
>> From: Deepak Kodihalli <dkodihal@linux.vnet.ibm.com 
>> <mailto:dkodihal@linux.vnet.ibm.com>>
>> Date: Feb 8, 2020 1:59 PM
>> Subject: Exposing BIOS attributes to a remote management console
>> To: openbmc@lists.ozlabs.org <mailto:openbmc@lists.ozlabs.org>
>> Cc:
>>
>> Hi,
>>
>> To enable remote updates (via a remote management console that talks to
>> the BMC) to the host firmware's BIOS settings table on IBM systems,
>> we're thinking of the following flow:
>>
>> a) The host firmware sends down a BIOS settings table to the BMC using
>> PLDM [1].
>> b) The BMC uses phosphor-settingsd [2] as the backend to store BIOS
>> attributes on D-Bus.
>> c) Bmcweb implements the Redfish BIOS schema [3] to enable remote
>> reads/writes on the BIOS attributes.
>>
>> However, c) is a problem because one needs to write YAML files [4] for
>> the BMC to determine what D-Bus objects to make corresponding to the
>> BIOS attributes. This requires a unique D-Bus interface *per* BIOS
>> attribute. This means the BMC must have prior knowledge about the BIOS
>> attributes.
>>
>> I don't think that's the right way to go about this for two reasons. One
>> - this creates a lockstep dependency on the host firmware when the BIOS
>> settings table needs to be updated, and two - I think the OpenBMC
>> implementation of this must be able to receive (via PLDM/IPMI/other
>> standard in-band means) a set of BIOS attributes from different BIOS
>> firmware stacks dynamically and expose them for out of band updates,
>> without having prior/build-time knowledge of those attributes. So I
>> think this calls for a different kind of D-Bus interface/infrastructure
>> than what the phosphor-settingsd app relies on. Something that enables
>> the BMC to add to D-Bus a BIOS attribute dynamically, knowing it's name,
>> type and default value.
>>
>> Any thoughts on this flow? I'm also curious to know if the Redfish BIOS
>> schema/attribute registry model requires the BMC to have prior knowledge
>> about the system BIOS attributes to update the registry. This wasn't
>> obvious to me from a quick read of the schema.
>>
>> Thanks,
>> Deepak
>>
>> [1]
>> https://www.dmtf.org/sites/default/files/standards/documents/DSP0247_1.0.0.pdf
>> [2] https://github.com/openbmc/phosphor-settingsd
>> [3] https://redfish.dmtf.org/schemas/v1/Bios.v1_1_0.json
>> [4]
>> https://github.com/openbmc/openbmc/blob/master/meta-phosphor/recipes-phosphor/settings/phosphor-settings-defaults/defaults.yaml
>>

Re: FW: Exposing BIOS attributes to a remote management console

[Sekar, Suryakanth]

Hi Deepak,

[Deepak]-->  I'll look out for the design doc from you. Can you share 
now what aspects that document would cover - redfish/in-band/D-Bus 
implementation - all of this? Is Intel planning on writing code for 
these areas soon? We're planning on implementing the PLDM BIOS spec for 
the same purpose. Some PLDM BIOS code's out there on openbmc/pldm already.

Yes, we need to support all interface -IPMI/ MCTP/ Redfish Host interface.

[surya]--> Cover IPMI based implementation. we defined some OEM 
commands  under review.

https://gerrit.openbmc-project.xyz/c/openbmc/intel-ipmi-oem/+/16246


[Deepak]-->  Are you targeting specific implementation areas here? By 
knowing that, we can collaborate better.

[surya] -->We are working on IPMI command based in OOB daemon and 
generating Attribute Registry for Redfish. you can implement the PLDM 
based in OOB daemon and BIOS configuration WebGUI.

Please review the design and interface. Once its done. we will request  
Brad to create separate recipe  repository for OOB daemon.

Please let me know any query or concern.

Thanks

Suryakanth.S


On 2/11/2020 2:22 PM, Deepak Kodihalli wrote:
> On 10/02/20 4:19 pm, Sekar, Suryakanth wrote:
>> Hi Deepak,
>>
>> OOB BIOS high level config design doc is in progress. I will send for 
>> review soon.
>
>
> Hi Surya,
>
> Thanks for responding. I'm glad that we have the same understanding on 
> this topic.
>
> I'll look out for the design doc from you. Can you share now what 
> aspects that document would cover - redfish/in-band/D-Bus 
> implementation - all of this? Is Intel planning on writing code for 
> these areas soon? We're planning on implementing the PLDM BIOS spec 
> for the same purpose. Some PLDM BIOS code's out there on openbmc/pldm 
> already.
>
>> Yes, we need to support all interface -IPMI/ MCTP/ Redfish Host 
>> interface.

[surya]--> Cover IPMI based implementation. we defined some OEM 
commands  under review.

https://gerrit.openbmc-project.xyz/c/openbmc/intel-ipmi-oem/+/16246

>>
>> If BIOS is not booted at all. BMC wont have any attribute info. It 
>> will be empty in BMC side.
>>
>> BIOS must provide BIOS OOB capability via KCS interface in early boot 
>> stage.
>> Like supported proprietary BIOS attribute file format or PLDM support 
>> via MCTP  or
>> Redfish BIOS attribute registry format to the BMC.
>>
>> BIOS must send the master BIOS attributes file(Intel properitary - 
>> XML type 0)
>> via KCS interface or all attributes details via Bios configuration 
>> PLDM via MCTP or
>> Redfish host interface during first boot.
>>
>> BIOS must collect the new attribute values from BMC and update the 
>> same in BIOS
>>
>> BIOS must send the updated master attributes file to the BMC and once 
>> its updated in BIOS
>>
>> and reset the system to reflect the BIOS configuration.
>>
>> OOB daemon in BMC  must maintain and collect the  attribute registry 
>> file from MCTP/ IPMI/ Redfish interface.
>
> All this sounds good to me.
>
>> Convert the proprietary XML format/ PLDM data  into BIOS attribute 
>> Registry format &
>> must support the below following dbus method.
>>
>> https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-dbus-interfaces/+/18242 
>>
>>
>> Validate the BIOS input data or User attribute value.
>>
>> MCTP / Redfish host interface / IPMI must communicate with OOB daemon 
>> in BMC via D bus
>
> Are you targeting specific implementation areas here? By knowing that, 
> we can collaborate better.

[surya] -->We are working on IPMI command based in OOB daemon and 
generating Attribute Registry for Redfish. you can implement the PLDM 
based in OOB daemon and BIOS configuration WebGUI.

Please review the design and interface. Once its done. we will request  
Brad to create separate recipe  repository for OOB daemon.

Please let me know any query or concern.


>
>
>> Thanks
>>
>> Suryakanth.S
>
> Regards,
> Deepak
>
>>
>>
>>> ---------- Forwarded message ----------
>>> From: Deepak Kodihalli <dkodihal@linux.vnet.ibm.com 
>>> <mailto:dkodihal@linux.vnet.ibm.com>>
>>> Date: Feb 8, 2020 1:59 PM
>>> Subject: Exposing BIOS attributes to a remote management console
>>> To: openbmc@lists.ozlabs.org <mailto:openbmc@lists.ozlabs.org>
>>> Cc:
>>>
>>> Hi,
>>>
>>> To enable remote updates (via a remote management console that talks to
>>> the BMC) to the host firmware's BIOS settings table on IBM systems,
>>> we're thinking of the following flow:
>>>
>>> a) The host firmware sends down a BIOS settings table to the BMC using
>>> PLDM [1].
>>> b) The BMC uses phosphor-settingsd [2] as the backend to store BIOS
>>> attributes on D-Bus.
>>> c) Bmcweb implements the Redfish BIOS schema [3] to enable remote
>>> reads/writes on the BIOS attributes.
>>>
>>> However, c) is a problem because one needs to write YAML files [4] for
>>> the BMC to determine what D-Bus objects to make corresponding to the
>>> BIOS attributes. This requires a unique D-Bus interface *per* BIOS
>>> attribute. This means the BMC must have prior knowledge about the BIOS
>>> attributes.
>>>
>>> I don't think that's the right way to go about this for two reasons. 
>>> One
>>> - this creates a lockstep dependency on the host firmware when the BIOS
>>> settings table needs to be updated, and two - I think the OpenBMC
>>> implementation of this must be able to receive (via PLDM/IPMI/other
>>> standard in-band means) a set of BIOS attributes from different BIOS
>>> firmware stacks dynamically and expose them for out of band updates,
>>> without having prior/build-time knowledge of those attributes. So I
>>> think this calls for a different kind of D-Bus interface/infrastructure
>>> than what the phosphor-settingsd app relies on. Something that enables
>>> the BMC to add to D-Bus a BIOS attribute dynamically, knowing it's 
>>> name,
>>> type and default value.
>>>
>>> Any thoughts on this flow? I'm also curious to know if the Redfish BIOS
>>> schema/attribute registry model requires the BMC to have prior 
>>> knowledge
>>> about the system BIOS attributes to update the registry. This wasn't
>>> obvious to me from a quick read of the schema.
>>>
>>> Thanks,
>>> Deepak
>>>
>>> [1]
>>> https://www.dmtf.org/sites/default/files/standards/documents/DSP0247_1.0.0.pdf 
>>>
>>> [2] https://github.com/openbmc/phosphor-settingsd
>>> [3] https://redfish.dmtf.org/schemas/v1/Bios.v1_1_0.json
>>> [4]
>>> https://github.com/openbmc/openbmc/blob/master/meta-phosphor/recipes-phosphor/settings/phosphor-settings-defaults/defaults.yaml 
>>>
>>>
>