Re: [PATCH v2 2/5] OvmfPkg: add library to track boot option loading/starting on the console

[Laszlo Ersek <lersek@redhat.com>]

On 02/20/19 10:21, Ard Biesheuvel wrote:
> On Wed, 20 Feb 2019 at 09:16, Laszlo Ersek <lersek@redhat.com> wrote:
>>
>> Introduce the Platform Boot Manager Print Status Code Library (for short,
>> PlatformBmPrintScLib) class for catching and printing the LoadImage() /
>> StartImage() preparations, and return statuses, that are reported by
>> UefiBootManagerLib.
>>
>> In the primary library instance, catch only such status codes that
>> UefiBootManagerLib reports from the same module that contains
>> PlatformBmPrintScLib. The intent is to establish a reporting-printing
>> channel within BdsDxe, between UefiBootManagerLib and
>> PlatformBmPrintScLib. Ignore status codes originating elsewhence, e.g.
>> from UiApp's copy of UefiBootManagerLib.
>>
>> Cc: Anthony Perard <anthony.perard@citrix.com>
>> Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>> Cc: Jordan Justen <jordan.l.justen@intel.com>
>> Cc: Julien Grall <julien.grall@linaro.org>
>> Cc: Ray Ni <ray.ni@intel.com>
>> Ref: https://bugzilla.redhat.com/show_bug.cgi?id=1515418
>> Contributed-under: TianoCore Contribution Agreement 1.1
>> Signed-off-by: Laszlo Ersek <lersek@redhat.com>
>> ---
>>
>> Notes:
>>     v2:
>>
>>     - Split the status code handling to a separate library, so that it's
>>       easy to reuse in ArmVirtPkg.
>>
>>     - Rework the logic based on
>>       <https://bugzilla.tianocore.org/show_bug.cgi?id=1398> and
>>       <https://mantis.uefi.org/mantis/view.php?id=1885>, and follow Ray's
>>       advice in
>>       <http://mid.mail-archive.com/734D49CCEBEEF84792F5B80ED585239D5BACE29B@SHSMSX104.ccr.corp.intel.com>:
>>
>>       - The boot option details are fetched via BootCurrent.
>>
>>       - For reporting LoadImage() and StartImage() preparations, replace the
>>         originally proposed PcdDebugCodeOsLoaderDetail status code with the
>>         existent (edk2-specific) PcdProgressCodeOsLoaderLoad and
>>         PcdProgressCodeOsLoaderStart status codes.
>>
>>       - For reporting LoadImage() and StartImage() return values, replace
>>         the originally proposed PcdDebugCodeOsLoaderDetail status code with
>>         the standard EFI_SW_DXE_BS_EC_BOOT_OPTION_LOAD_ERROR and
>>         EFI_SW_DXE_BS_EC_BOOT_OPTION_FAILED status codes.
>>
>>       - For all four kinds of reports, replace the originally proposed "OS
>>         Loader Detail" structure (and GUID) with the recently standardized
>>         EFI_RETURN_STATUS_EXTENDED_DATA structure.
>>
>>  OvmfPkg/OvmfPkg.dec                                           |   5 +
>>  OvmfPkg/OvmfPkgIa32.dsc                                       |   1 +
>>  OvmfPkg/OvmfPkgIa32X64.dsc                                    |   1 +
>>  OvmfPkg/OvmfPkgX64.dsc                                        |   1 +
>>  OvmfPkg/Include/Library/PlatformBmPrintScLib.h                |  41 +++
>>  OvmfPkg/Library/PlatformBmPrintScLib/PlatformBmPrintScLib.inf |  66 +++++
>>  OvmfPkg/Library/PlatformBmPrintScLib/StatusCodeHandler.c      | 310 ++++++++++++++++++++
>>  7 files changed, 425 insertions(+)
>>

[...]

>> +  //
>> +  // Set the EFI_STATUS_CODE_VALUE convenience variables.
>> +  //
>> +  mLoadPrep  = PcdGet32 (PcdProgressCodeOsLoaderLoad);
>> +  mLoadFail  = (EFI_SOFTWARE_DXE_BS_DRIVER |
>> +                EFI_SW_DXE_BS_EC_BOOT_OPTION_LOAD_ERROR);
>> +  mStartPrep = PcdGet32 (PcdProgressCodeOsLoaderStart);
>> +  mStartFail = (EFI_SOFTWARE_DXE_BS_DRIVER |
>> +                EFI_SW_DXE_BS_EC_BOOT_OPTION_FAILED);
>> +
>
> This bit looks somewhat dodgy to me, but I suppose the asymmetry is
> 'prior art' from EDK2, no?

Yes, that's the case. All four status code values are taken verbatim
from EfiBootManagerBoot()
[MdeModulePkg/Library/UefiBootManagerLib/BmBoot.c], where they are
reported / produced.

I use module-global variables here because (a) I need no generality wrt.
status codes values in this module (I really only care for these four),
and (b) the original expressions are simply unbearably long; considering
the frequent use of these status code values in the patch.


Regarding the reporting in EfiBootManagerBoot(): the status code values

- (EFI_SOFTWARE_DXE_BS_DRIVER | EFI_SW_DXE_BS_EC_BOOT_OPTION_LOAD_ERROR)
- (EFI_SOFTWARE_DXE_BS_DRIVER | EFI_SW_DXE_BS_EC_BOOT_OPTION_FAILED)

are from the PI spec. If we expand the macros a bit, we get,
respectively:

- EFI_SOFTWARE | 0x00050000 | EFI_SUBCLASS_SPECIFIC | 0x00000002
- EFI_SOFTWARE | 0x00050000 | EFI_SUBCLASS_SPECIFIC | 0x00000003

So we are in the "software class", the "DXE Boot Service Driver"
subclass, and we report values 2 and 3, which are meant to be unique
only within that subclass.

Conversely, the "prep" status code values are edk2 extensions. The PCDs
allow a platform, in theory anyway, to tweak the exact values. But in
practice, that should never be necessary. Let's check their default
values, in "MdeModulePkg/MdeModulePkg.dec":

>   ## Progress Code for OS Loader LoadImage start.<BR><BR>
>   #  PROGRESS_CODE_OS_LOADER_LOAD   = (EFI_SOFTWARE_DXE_BS_DRIVER | (EFI_OEM_SPECIFIC | 0x00000000)) = 0x03058000<BR>
>   # @Prompt Progress Code for OS Loader LoadImage start.
>   # @ValidList  0x80000003 | 0x03058000
>   gEfiMdeModulePkgTokenSpaceGuid.PcdProgressCodeOsLoaderLoad|0x03058000|UINT32|0x30001030
>
>   ## Progress Code for OS Loader StartImage start.<BR><BR>
>   #  PROGRESS_CODE_OS_LOADER_START  = (EFI_SOFTWARE_DXE_BS_DRIVER | (EFI_OEM_SPECIFIC | 0x00000001)) = 0x03058001<BR>
>   # @Prompt Progress Code for OS Loader StartImage start.
>   # @ValidList  0x80000003 | 0x03058001
>   gEfiMdeModulePkgTokenSpaceGuid.PcdProgressCodeOsLoaderStart|0x03058001|UINT32|0x30001031

Meaning

- EFI_SOFTWARE | 0x00050000 | EFI_OEM_SPECIFIC | 0x00000000
- EFI_SOFTWARE | 0x00050000 | EFI_OEM_SPECIFIC | 0x00000001

We stay within the same class & subclass, but use OEM-specific values 0
and 1, rather than standard values 2 and 3 that are specific to the
subclass.

I'd prefer if these weren't even PCDs, and the
PROGRESS_CODE_OS_LOADER_LOAD and PROGRESS_CODE_OS_LOADER_START macros
actually existed in some header file. That would be similarly clear
about the values being edk2 extensions, but without muddying the picture
with (academic?) platform overrides.

> In any case, this looks good to me otherwise, so for the series
>
> Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

Thank you!
Laszlo