Re: how to load drivers from additional FV's?

[Andrew Fish <afish@apple.com>]

> On Mar 15, 2017, at 8:38 AM, Laszlo Ersek <lersek@redhat.com> wrote:
> 
> On 03/15/17 16:13, Andrew Fish wrote:
>> 
>>> On Mar 15, 2017, at 8:07 AM, Laszlo Ersek <lersek@redhat.com> wrote:
>>> 
>>> On 03/15/17 13:23, Michael Zimmermann wrote:
>>>> I'm trying to add another FV section FVMAIN_COMPACT so I can keep
>>>> Platform specific drivers in a separate, included fdf.
>>>> 
>>>> I did this:
>>>> FILE FV_IMAGE = 9E21FD93-9C72-4c15-8C4B-E77F1DB2D792 {
>>>>   SECTION GUIDED EE4E5898-3914-4259-9D6E-DC7BD79403CF
>>>> PROCESSING_REQUIRED = TRUE {
>>>>     SECTION FV_IMAGE = FVMAIN
>>>>     SECTION FV_IMAGE = FVMAINPLATFORM
>>>>   }
>>>> }
>>>> 
>>>> The image builds file and using uefitool I can verify that the new FV
>>>> is inside the compressed section.
>>>> But none of the drivers gets discovered/loaded and I get 'Protocol not
>>>> present!!' errors.
>>> 
>>> The FVs need to be exposed to the DXE core via FV HOBs. See
>>> - 9.8.5 "Firmware Volume HOBs" in Volume 2 of the Platform Init 1.5
>>> spec,
>>> - and more importantly, 5.7 "Firmware Volume HOB" in Volume 3 of the
>>> same.
>>> 
>>> You can use the BuildFvHob() function for this.
>>> 
>>> If the firmware volume contains PEIMs (... as well), then it has to be
>>> exposed to the PEI core too, I think. I think the
>>> PeiServicesInstallFvInfoPpi() function can be used for that. (See 3.3
>>> "PEI" in Volume 3 of the PI spec.)
>>> 
>>> ... I used the PeiFvInitialization() function in
>>> OvmfPkg/PlatformPei/Fv.c as a "cheat sheet" for the above.
>>> 
>> 
>> Laszlo,
>> 
>> I think this case is an FV that is compressed and nested in another FV that is discovered. I think the issues is multiple FV Sections in an FV file are not currently supported. 
> 
> In OVMF we do the same:
> 
> FILE FV_IMAGE = 9E21FD93-9C72-4c15-8C4B-E77F1DB2D792 {
>   SECTION GUIDED EE4E5898-3914-4259-9D6E-DC7BD79403CF PROCESSING_REQUIRED = TRUE {
>     #
>     # These firmware volumes will have files placed in them uncompressed,
>     # and then both firmware volumes will be compressed in a single
>     # compression operation in order to achieve better overall compression.
>     #
>     SECTION FV_IMAGE = PEIFV
>     SECTION FV_IMAGE = DXEFV
>   }
> }
> 
> "OvmfPkg/Sec/SecMain.c" decompresses the compressed FFS file into memory -- which is special to OVMF, i.e. the fact that there is writeable memory in SEC --, then jumps to the PEI entry point. PlatformPei then exposes DXEFV to the PEI core (for loading the DXE core AIUI) and to the DXE core as well (for loading DXE modules).
> 
> The multiple FV sections in the FFS file are not discovered automatically. First, the OVMF build process saves the final (= decompressed) in-memory locations of the nested FVs into PCDs. Second, in DecompressMemFvs(), the SEC module decompresses the outer, compressed, FFS file, then locates the embedded FVs manually with FindFfsSectionInstance(), then moves the decompressed FVs individually to their pre-recorded locations.
> 
> In Michael's case, I think the same should be possible -- since FVMAIN is running for him, the outer, compressed file must already be decompressed (in full) to some temporary buffer in memory. If he can find, manually, the second section (with the embedded FVMAINPLATFORM firmware volume in it) in that area, then he can also expose it to the DXE core, I believe.
> 

Laszlo,

The DXE Core will dispatch FVs with Depex "automagically" but only one FV_IMAGE Section per FV_IMAGE file. 

So one workaround would be to add a 2nd FV_IMAGE file that contains the 2nd FV in the 1st FV_IMAGE Section. 

Also I'm not sure if the outer FV is compressed, but compressing compressed data usually does not end well as the algorithms leverage entropy. It is better to just compress the outer FV and have everything else uncompressed and let it all get compressed together. This also helps as there will be common dictionary entries so compressing as much data as possible in one shot usually works better. So for example if the outer FV is compressed it may have a dictionary entry that represents CopyMem, and it can represent that pattern with a small number of bits. Doing compression multiple times causes duplication as each compressed image has one copy of CopyMem (massive over simplification). Also if you compress a nested FV then the CopyMem pattern does not exist in the compressed data, and since the compressed data is more random it is hard to find patterns to compress, so it hurts the compression of the outer FV. 

Thanks,

Andrew Fish

> Thanks
> Laszlo
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org <mailto:edk2-devel@lists.01.org>
> https://lists.01.org/mailman/listinfo/edk2-devel <https://lists.01.org/mailman/listinfo/edk2-devel>