eBMC LED Management :: Health and HealthRollup support

[vishwabmc]

Initially, redfish had Indicator LED property with enum Attention/Identify/Off. However, going forward, that would not be the case. The idea is to use "HealthRollUp" property of a FRU to indicate a fault on thh FRU.

What is needed:

when a client wants to read HealthRollUp, bmcweb will check if there is any association to led object for that FRU and if so, it reads the asserted state of the group and then populates the Rollup property

For example: each FRU instance will have "fault_led" and "identify_led".

Sequence of events: 1/. User reads HealthRollUp of a resource

2/. bmcweb checks if there is any led association object created for that Inventory Example:

• busctl get-property xyz.openbmc_project.ObjectMapper /xyz/openbmc_project/inventory/system/chassis/motherboard/dimm10/fault_led xyz.openbmc_project.Association endpoints

Example output: "/xyz/openbmc_project/led/groups/dimm10_fault"

3/. If there is an association existing, then read the "asserted" state.

• busctl get-property xyz.openbmc_project.LED.GroupManager /xyz/openbmc_project/led/groups/dimm10_fault xyz.openbmc_project.Led.Group Asserted

4/. Use the result from above step ^ and then set "HealthRollup" is "Critical" if it's "true"

[lxwinspur]

changed fault_led to fault_led_group changed identify_led to identify_led_group

[lxwinspur]

https://gerrit.openbmc-project.xyz/c/openbmc/bmcweb/+/35541

[vishwabmc]

After discussing this with Ed 'et al, I believe the better solution is not to derive health-rollup from the state of Fault_Led.

Instead, do 2 things as part of setting the fault LED.

• Set the LED
• Set the D-Bus property which is currently used by bmcweb to represent HealthRollup. ( Still need to figure that out )

Using this approach :

• No changes to bmcweb
• We need to add some code in bmc to do 2 things mentioned above

[vishwabmc]

Closing this...

[vishwabmc]

Reopening this since we need to implement Health and HealthRollup for all the resources.. Currently, it's there only for CPU and DIMM.

Today, the Health field is populated like this :

• Inventory D-Bus object need to have xyz.openbmc_project.State.Decorator.OperationalStatus implemented with Property Functional
• BMC firmware code will set that to false to indicate it's faulted.
• bmcweb looks at this property and if it's false, then Health is set to Critical..
• If the value is true, then Health is set to OK.

We need to implement the HealthRollup also.. We believe, the existing HealthRollup on [Chassis] is good enough for the GA. Here is how we see it works :

• bmcweb in chassis looks to see if xyz.openbmc_project.Inventory.Item.Global is implemented.
• if yes, then checks if the chassis D-Bus object has the association called critical
• If yes, then it sets the chassis HealthRollup to [Critical]

This github will be a epic and we can have sub tasks inside this to have health implemented on all the resources.. we have many new schemas introduced..

[gtmills]

For HealthRollup: Logging is creating the associations for us here: https://github.com/openbmc/phosphor-logging/blob/76198a2ea2e3deb14078e0d3dd3932b657ddd9b8/extensions/openpower-pels/service_indicators.cpp#L83

The only healthrollup we really care about is Chassis, at least for now. @santoshpuranik's app will implement the item.global interface.

For Health: We need all resources to use OperationalStatus. Need to modify https://github.com/openbmc/bmcweb/blob/master/redfish-core/lib/health.hpp#L44 (either removing Health from it or have it set the health)

Any changes for this can be downstream only.