ipmi command implementation mismatch with the design document

[Vernon Mauery]

On 23-Nov-2020 06:39 AM, Tung Nguyen OS wrote:
>Hi everyone,
>I’m Tung Nguyen, developer for AmpereComputing - Altra system. While working with the IPMI commands I have a concern when comparing the design document and the implementation like the following:
>The state-management-and-external-interfaces.md
>The full mapping of Redfish and IPMI to xyz.openbmc_project.State.* is as follows:
>Redfish                 IPMI                xyz.openbmc_project.State.Transition
>
>ForceOff                power down          Chassis.Off
>ForceOn                 power up            Host.On
>ForceRestart            hard reset          Host.ForceWarmReboot
>GracefulRestart                             Host.GracefulWarmReboot
>GracefulShutdown        soft off            Host.Off
>On                      power up            Host.On
>PowerCycle (host on)    power cycle         Host.Reboot
>PowerCycle (host off)                       Chassis.PowerCycle
>
>the IPMI – chassishandler.cpp:
>
>ipmi::RspType<> ipmiChassisControl(uint8_t chassisControl)
>
>{
>
>    int rc = 0;
>
>    switch (chassisControl)
>
>    {
>
>        case CMD_POWER_ON:
>
>            rc = initiate_state_transition(State::Host::Transition::On);
>
>            break;
>
>        case CMD_POWER_OFF:
>
>            // This path would be hit in 2 conditions.
>
>            // 1: When user asks for power off using ipmi chassis command 0x04
>
>            // 2: Host asking for power off post shutting down.
>
>
>
>            // If it's a host requested power off, then need to nudge Softoff
>
>            // application that it needs to stop the watchdog timer if running.
>
>            // If it is a user requested power off, then this is not really
>
>            // needed. But then we need to differentiate between user and host
>
>            // calling this same command
>
>
>
>            // For now, we are going ahead with trying to nudge the soft off and
>
>            // interpret the failure to do so as a non softoff case
>
>            rc = stop_soft_off_timer();
>
>
>
>            // Only request the Off transition if the soft power off
>
>            // application is not running
>
>            if (rc < 0)
>
>            {
>
>                // First create a file to indicate to the soft off application
>
>                // that it should not run. Not doing this will result in State
>
>                // manager doing a default soft power off when asked for power
>
>                // off.
>
>                indicate_no_softoff_needed();
>
>
>
>                // Now request the shutdown
>
>                rc = initiate_state_transition(State::Host::Transition::Off);
>
>            }
>
>            else
>
>            {
>
>                log<level::INFO>("Soft off is running, so let shutdown target "
>
>                                 "stop the host");
>
>            }
>
>            break;
>
>
>
>The redfish – systems.hpp:
>        else if (resetType == "ForceOff")
>        {
>            command = "xyz.openbmc_project.State.Chassis.Transition.Off";
>            hostCommand = false;
>        }
>
>
>
>
>
>Although the indicate_no_softoff_needed() can prevent the host from soft off, but it seems like a mismatch b/w the design document and the IPMI implementation.
>
>So, my question: is it reasonable for IPMI command ?

This code has been in place for quite some time now, so I am not sure if 
the original authors have the context at this point in time. But nobody 
else has raised this question. git blame says that Andrew Geissler added 
this feature, but we would have to see if he still knows why. 

commit a6e3a3080d532536e02e304c819c1e17214e038a
Author: Andrew Geissler <andrewg at us.ibm.com>
Date:   Wed May 31 19:34:00 2017 -0500

    Create file to indicate host requested off/reboot

    Create a file to ensure the soft power off service is
    not run when the host is requesting a power off
    or reboot.  There's no need to notify the host (i.e.
    soft power off) when they are initiating it.

    Change-Id: Ic9f8e7110d30f477ceae38bba9d684559d9503d3
    Signed-off-by: Andrew Geissler <andrewg at us.ibm.com>


--Vernon

>Reference:
>https://github.com/openbmc/docs/blob/master/designs/state-management-and-external-interfaces.md
>
>Best regards,
>Tung