Re: beep code manager for openbmc

[Patrick Williams <patrick@stwcx.xyz>]

[-- Attachment #1: Type: text/plain, Size: 2800 bytes --]

On Wed, Sep 08, 2021 at 03:29:48PM +0000, Ren, Zhikui wrote:
> >I see it creates dbus interfaces but I don't see them defined anywhere.  Can
> >we get that added to phosphor-dbus-interfaces?
> >
> [Ren, Zhikui] Yes, I will submit a review on that.
> >It isn't obvious to me from looking at what interfaces are there who would call
> >it.  I see various "beep code" patterns:
> >
> >    {beepVRWatchdogTimeout, "1-5-1-2"}, {beepPSUFailure, "1-5-1-4"},
> >    {beepCPUMIssing, "1-5-2-1"},        {beepCPUCatError, "1-5-2-2"},
> >    {beepCPUErr2, "1-5-2-3"},           {beepVoltageMismatch, "1-5-2-4"},
> >    {beepCPUConfigError, "1-5-2-5"},    {beepPowerFail, "1-5-4-2"},
> >    {beepPowerGoodTimeOut, "1-5-4-4"},
> >
> >How do these get triggered?  Do arbitrary programs call into the "BeepCode"
> >service when they hit their own event?  This doesn't seem very maintainable;
> >shouldn't the beepcode service react to events through some other method,
> >similar to what is going on for Redfish events?
> >
> [Ren, Zhikui] Currently, an application can call into the BeepCode by calling the dbus method
> One of improvement ideas is to make the beepcode service monitor the events it is interested in.

Fundamentally, I see this as a very poor design.  I don't really like us taking
code written in isolation without any design feedback and then expecting the
project to pick it up as-is.

We already have REDFISH_MESSAGE_ID and IPMI_SEL_* journal strings, a proposal
for audit, and now this beepcode proposal, which are all a type of event.  The
current design proposals for all of them is that every application needs to be
aware of every consumer and make N random function calls to each of those
consumers.  It is sloppy and doesn't scale.  I don't want to add to the mess by
adding beepcodes this way.

I'm slowly working on unraveling this by adding an Event concept to
phosphor-logging (distinct from the current Error concepts).  I'm hopeful that
this can be leveraged for any such event, including boot sequence states that
can be consumed by this beepcode application.  I should have the initial design
within a week or two and would certainly welcome any review (with respect to
beep code) and assistance in implementation.

> >How easily configurable would this be for someone who has a different
> >beepcode policy than Intel's?  Are you planning to update it with some JSON
> >config or leave that as an exercise for the next user?
> >
> [Ren, Zhikui] Good idea.  However I don't have a definitive plan on that currently and open to suggestions/collaborations.

Another reason why we shouldn't just take wholesale code that was written and
designed in isolation but isn't really fit for anyone else's purposes.

-- 
Patrick Williams

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