Re: [edk2-devel] [edk2-rfc] MdeModulePkg/StatusCodeHandler: Separate NULL class libraries for Memory and serial handlers from MdeModulePkg/Universal/StatusCodeHandler modules

["Brian J. Johnson" <brian.johnson@hpe.com>]

[-- Attachment #1: Type: text/plain, Size: 15250 bytes --]

Thanks for the link.  I agree that this change will make the 
StatusCodeHandler driver more modular, and is a step in the right direction.

But I think it could go further, with almost no additional work, and 
simplify the overall Status Code mechanism, not just the 
StatusCodeHandler driver.  Currently, the StatusCodeHandler driver entry 
routines run some initialization code, register callbacks (eg. for 
ExitBootServices and SetVirtualAddressMap), and call the RscHandler 
PPI/Protocol to register the worker routines.  If I'm understanding the 
proposal correctly, all that code will be moved to the individual NULL 
libraries, since any particular library may or may not need any of it.  
Then the StatusCodeHandler modules will be left with no code of their 
own at all:  they will only be wrappers for the NULL libraries. Their 
entry routines will do nothing except return EFI_SUCCESS! (1)

It seems strange and wasteful to keep around empty modules like this.  
So I'm suggesting adding the NULL libraries to the StatusCodeRouter 
modules instead.  They would need to export the protocol/PPI routines to 
the NULL libraries via a header file, so they could call them directly 
instead of looking up the protocol/PPI.  But that's a minor change.  
Then you could remove the empty StatusCodeHandler modules entirely.  The 
advantage would be that there would be fewer modules in the build, 
simplifying the .dsc and .fdf files slightly.  It would also reduce code 
size a bit by sharing common library routines, such as BaseLib, with the 
StatusCodeRouter modules.

If those don't seem like worthwhile advantages, that's OK with me.  I 
don't want to belabor the point, or impede progress.  If others are OK 
with the proposal as it stands, then I am too.

Thanks,

*Brian J. Johnson
*Enterprise X86 Lab

Hewlett Packard Enterprise


(1) The StatusCodeHandlerRuntimeDxe driver also handles 
PcdStatusCodeReplayIn as part of its entry code.  That code would 
probably have to stay in a separate module rather than being linked to 
StatusCodeRouter as a NULL library.  That way it could be dispatched 
after the ReportStatusCode protocol is available.

------------------------------------------------------------------------
*From:* Dong, Eric [mailto:eric.dong@intel.com]
*Sent:* Thursday, June 25, 2020, 10:41 AM
*To:* Brian J. Johnson <brian.johnson@hpe.com>, Bi, Dandan 
<dandan.bi@intel.com>, Andrew Fish <afish@apple.com>, 
edk2-devel-groups-io <devel@edk2.groups.io>
*Cc:* rfc@edk2.groups.io <rfc@edk2.groups.io>, Ni, Ray 
<ray.ni@intel.com>, Wang, Jian J <jian.j.wang@intel.com>, Wu, Hao A 
<hao.a.wu@intel.com>, Tan, Ming <ming.tan@intel.com>, Laszlo Ersek 
<lersek@redhat.com>
*Subject:* [edk2-devel] [edk2-rfc] MdeModulePkg/StatusCodeHandler: 
Separate NULL class libraries for Memory and serial handlers from 
MdeModulePkg/Universal/StatusCodeHandler modules

> Hi Brian,
>
> In this new design, we still use register status code handler 
> Protocol/Ppi to register the handler logic. We just want to change the 
> StatusCodeHandler driver. We try to split the register logic to NULL 
> library to make the code more modularity. We already created sample 
> library in Edk2-Platforms repo 
> https://github.com/tianocore/edk2-platforms/tree/master/Features/Intel/Debugging/PostCodeDebugFeaturePkg/Library/PostCodeStatusCodeHandlerLib. 
> You can check this code to understand more about what we want to do.
>
> Thanks,
>
> Eric
>
> *From:* Brian J. Johnson <brian.johnson@hpe.com>
> *Sent:* Thursday, June 25, 2020 4:25 AM
> *To:* Bi, Dandan <dandan.bi@intel.com>; Andrew Fish <afish@apple.com>; 
> edk2-devel-groups-io <devel@edk2.groups.io>
> *Cc:* rfc@edk2.groups.io; Dong, Eric <eric.dong@intel.com>; Ni, Ray 
> <ray.ni@intel.com>; Wang, Jian J <jian.j.wang@intel.com>; Wu, Hao A 
> <hao.a.wu@intel.com>; Tan, Ming <ming.tan@intel.com>; Laszlo Ersek 
> <lersek@redhat.com>
> *Subject:* Re: [edk2-devel] [edk2-rfc] MdeModulePkg/StatusCodeHandler: 
> Separate NULL class libraries for Memory and serial handlers from 
> MdeModulePkg/Universal/StatusCodeHandler modules
>
> Dandan,
>
> The Status Code Protocol/PPI is the high-level interface which is 
> being implemented.  The ReportStatusCodeRouter module implements this 
> in terms of the ReportStatusCodeHandler Protocol/PPI.  That allows 
> multiple ReportStatusCodeHandler modules to be used at once, so they 
> can each react to different types of status codes, or report them 
> through multiple channels.  That sort of multiplexing seems like a 
> useful feature.
>
> Now we're considering adding a mechanism which allows registering 
> status code handlers via NULL libraries, rather than via the 
> protocol/PPI. That sounds like exactly what ReportStatusCodeRouter is 
> intended for.  It wouldn't really change its scope, it would just make 
> it more flexible.  Adding this feature via a StatusCodeHandler module 
> wouldn't improve modularity, it would just add complexity.  As an OEM, 
> adding a custom handler would look the same to me either way:  I would 
> have to add the NULL class library to a MdeModulePkg driver's entry in 
> my .dsc file.  It doesn't matter to me whether it's the 
> ReportStatusCodeRouter or StatusCodeHandler module.  And if I can do 
> it in ReportStatusCodeRouter, then I don't need to include any 
> StatusCodeHandler modules in the build at all.  That saves code space 
> and reduces the number of modules in the APRIORI list, which are both 
> good things.
>
> ReportStatusCodeRouterPei already has to track registered handlers in 
> PEI when running from ROM (it uses a HOB.)  Tracking handlers from 
> NULL libraries wouldn't be any harder.  In fact, it looks like it 
> could just export the Register() function to the NULL libraries, and 
> they could call it in their constructors.
>
> I think using NULL libraries for status code handlers is a great 
> idea.  I'd just like to take that opportunity to reduce the complexity 
> of the overall status code stack while we're at it.
>
> Thanks,
>
> *Brian J. Johnson
> *Enterprise X86 Lab
>
> Hewlett Packard Enterprise
>
> ------------------------------------------------------------------------
>
> *From:* Bi, Dandan [mailto:dandan.bi@intel.com]
>
> *Sent:* Monday, June 22, 2020, 2:27 AM
>
> *To:* Andrew Fish <afish@apple.com> <mailto:afish@apple.com>, 
> edk2-devel-groups-io <devel@edk2.groups.io> 
> <mailto:devel@edk2.groups.io>, brian.johnson@hpe.com 
> <mailto:brian.johnson@hpe.com> <brian.johnson@hpe.com> 
> <mailto:brian.johnson@hpe.com>
>
> *Cc:* rfc@edk2.groups.io <mailto:rfc@edk2.groups.io> 
> <rfc@edk2.groups.io> <mailto:rfc@edk2.groups.io>, Dong, Eric 
> <eric.dong@intel.com> <mailto:eric.dong@intel.com>, Ni, Ray 
> <ray.ni@intel.com> <mailto:ray.ni@intel.com>, Wang, Jian J 
> <jian.j.wang@intel.com> <mailto:jian.j.wang@intel.com>, Wu, Hao A 
> <hao.a.wu@intel.com> <mailto:hao.a.wu@intel.com>, Tan, Ming 
> <ming.tan@intel.com> <mailto:ming.tan@intel.com>, Laszlo Ersek 
> <lersek@redhat.com> <mailto:lersek@redhat.com>, Bi, Dandan 
> <dandan.bi@intel.com> <mailto:dandan.bi@intel.com>
>
> *Subject:* [edk2-devel] [edk2-rfc] MdeModulePkg/StatusCodeHandler: 
> Separate NULL class libraries for Memory and serial handlers from 
> MdeModulePkg/Universal/StatusCodeHandler modules
>
>     Hi Brian,
>
>     Personally, I prefer to add the NULL class Library to
>     StatusCodeHandler modules.
>
>      1. I think we should make the functionality of each module clear
>         and separated. It may also be why we added
>         ReportStatusCodeRouter and StatusCodeHandler modules in edk2
>         repo before.
>
>     ReportStatusCodeRouter modules are responsible for producing
>     Status Code Protocol/PPI and Report Status Code Handler
>     Protocol/PPI, and StatusCodeHandler modules are responsible for
>     producing handlers (Handlers can be provided by NULL class
>     Libraries in this RFC).
>
>      So, that’s why I don’t want to add the NULL class Library to
>     ReportStatusCodeRouter modules directly, which change the
>     functionality scope of existing modules.
>
>      2. I agree that we have a lot of layers of indirection now, but
>         what we may gain is the good modularity. And you also
>         mentioned that one or more StatusCodeHandler Modules may be
>         used. We also want to achieve that only the StatusCodeHandler
>         modules in MdeModulePkg can be used after this separation,
>         platform can only add its own handler Libs to meet its
>         requirement.
>
>      3. As Andrew mentioned below, if add the libraries to
>         ReportStatusCodeRouter, there will be some issue we need to
>         fix, which seems also make the code logic a little tricky to me.
>
>     Thanks,
>
>     Dandan
>
>     *From:* Andrew Fish <afish@apple.com> <mailto:afish@apple.com>
>     *Sent:* Saturday, June 20, 2020 2:04 AM
>     *To:* edk2-devel-groups-io <devel@edk2.groups.io>
>     <mailto:devel@edk2.groups.io>; brian.johnson@hpe.com
>     <mailto:brian.johnson@hpe.com>
>     *Cc:* Bi, Dandan <dandan.bi@intel.com>
>     <mailto:dandan.bi@intel.com>; rfc@edk2.groups.io
>     <mailto:rfc@edk2.groups.io>; Dong, Eric <eric.dong@intel.com>
>     <mailto:eric.dong@intel.com>; Ni, Ray <ray.ni@intel.com>
>     <mailto:ray.ni@intel.com>; Wang, Jian J <jian.j.wang@intel.com>
>     <mailto:jian.j.wang@intel.com>; Wu, Hao A <hao.a.wu@intel.com>
>     <mailto:hao.a.wu@intel.com>; Tan, Ming <ming.tan@intel.com>
>     <mailto:ming.tan@intel.com>
>     *Subject:* Re: [edk2-devel] [edk2-rfc]
>     MdeModulePkg/StatusCodeHandler: Separate NULL class libraries for
>     Memory and serial handlers from
>     MdeModulePkg/Universal/StatusCodeHandler modules
>
>
>
>
>         On Jun 19, 2020, at 10:29 AM, Brian J. Johnson
>         <brian.johnson@hpe.com <mailto:brian.johnson@hpe.com>wrote:
>
>         On 6/18/20 2:01 AM, Dandan Bi wrote:
>
>             Hi All,
>
>             REF:https://bugzilla.tianocore.org/show_bug.cgi?id=2816
>             <https://bugzilla.tianocore.org/show_bug.cgi?id=2816>
>
>             We plan to separate two kinds of NULL class libraries for
>             Memory and serial handlers
>             from*MdeModulePkg/Universal/StatusCodeHandler/…/
>             StatusCodeHandlerPei/RuntimeDxe/Smm*modules.
>
>             The benefit we want to gain from this separation is to 1)
>             make the code clear and easy to maintain, 2) make platform
>             flexible to choose any handler library they need, and it
>             also can reduce image size since the unused handlers can
>             be excluded.
>
>             If you have any concern or comments for this separation,
>             please let me know.
>
>             We plan to add new separated NULL class
>             library*MemoryStausCodeHandlerLib*and***SerialStatusCodeHandlerLib*with
>             different phase implementation
>             into*MdeModulePkg\Library\*directory.
>
>             The main tree structure may like below:
>
>             MdeModulePkg\Library
>
>             |------*MemoryStausCodeHandlerLib*
>
>             |------|------ PeiMemoryStausCodeHandlerLib.inf
>
>             |------|------ RuntimeDxeMemoryStatusCodeHandlerLib.inf
>
>             |------|------ SmmMemoryStausCodeHandlerLib.inf
>
>             |------*SerialStatusCodeHandlerLib*
>
>             |------|------ PeiSerialStatusCodeHandlerLib.inf
>
>             |------|------ RuntimeDxeSerialStatusCodeHandlerLib.inf
>
>             |------|------ SmmSerialStatusCodeHandlerLib.inf
>
>             **
>
>             **
>
>             We will update existing platform use cases in edk2 and
>             edk2-platform repo to cover the new NULL class library to
>             make sure this change doesn’t impact any platform.
>
>             After this separation, StatusCodeHandler module usage will
>             like below, and it’s also very flexible for platform to
>             cover more handler libraries to meet their requirements.
>
>             MdeModulePkg/Universal/StatusCodeHandler/Pei/StatusCodeHandlerPei.inf
>             {
>
>             <LibraryClasses>
>
>             NULL|MdeModulePkg/Library/MemoryStausCodeHandlerLib/PeiMemoryStausCodeHandlerLib.inf
>
>             NULL|MdeModulePkg/Library/SerialStatusCodeHandlerLib/PeiSerialStatusCodeHandlerLib.inf
>
>             …
>
>             }
>
>             MdeModulePkg/Universal/StatusCodeHandler/RuntimeDxe/StatusCodeHandlerRuntimeDxe.inf
>             {
>
>             <LibraryClasses>
>
>             NULL|MdeModulePkg/Library/MemoryStausCodeHandlerLib/RuntimeDxeMemoryStausCodeHandlerLib.inf
>
>             NULL|MdeModulePkg/Library/SerialStatusCodeHandlerLib/RuntimeDxeSerialStatusCodeHandlerLib.inf
>
>             …
>
>             }
>
>             MdeModulePkg/Universal/StatusCodeHandler/Smm/StatusCodeHandlerSmm.inf
>             {
>
>             <LibraryClasses>
>
>              NULL|MdeModulePkg/Library/MemoryStausCodeHandlerLib/SmmMemoryStausCodeHandlerLib.inf
>
>             NULL|MdeModulePkg/Library/SerialStatusCodeHandlerLib/SmmSerialStatusCodeHandlerLib.inf
>
>             …
>
>             }
>
>             Thanks,
>
>             Dandan
>
>         Dandan,
>
>         We'll have a lot of layers of indirection....  The
>         ReportStatusCodeRouter modules will call one or more
>         StatusCodeHandlerModules, and the standard StatusCodeHandler
>         modules will call multiple StatusCodeHandlerLib libraries.
>
>         How about adding StatusCodeHandlerLib support directly to the
>         ReportStatusCodeRouter modules? Then platforms could omit the
>         StatusCodeHandler modules if they're only using the
>         open-source code.  That sounds like less overhead since fewer
>         modules would be needed.
>
>     I think the need to execute from ROM makes this tricky.
>
>     It looks to me that it is easy to move from PCD to libs for the
>     StatusCodeHandler since registration is basically
>     `RscHandlerPpi->Register (SerialStatusCodeReportWorker);`. The
>     issue I see is the ReportStatusCodeRouter publishes RscHandlerPpi
>     after the PEIMs constructor has been called and if the PEIM. Given
>     globals don’t work when running from ROM you would have to do
>     something like publish a HOB in the library constructor and then
>     have the GenericStatusCodePeiEntry() walk the HOBs and install the
>     handlers. So I guess it is a little easier than I 1st thought when
>     I started writing this mail, but it would require a new public API.
>
>     Thanks,
>
>     Andrew Fish
>
>
>         Thanks,
>
>         --
>
>         *Brian J. Johnson
>         *Enterprise X86 Lab
>
>         Hewlett Packard Enterprise
>
>         *hpe.com* <x-msg://64/3D%22hpe.com%22>
>
>         
>
> -
>



[-- Attachment #2: Type: text/html, Size: 45978 bytes --]