Re: [PATCH v1 1/1] OvmfPkg/ResetVector: Depend on PCD values of the page tables.

["Marvin Häuser" <Marvin.Haeuser@outlook.com>]

Hey Laszlo,
Hey Jordan,

I sent a V2 of the patch. Changes:
* fixed the title to get rid of the numbering
* Introduced PT_ADDR() as proposed by Jordan
* introduced the size check proposed by Laszlo
* removed the 'ifndef' for the NASM ResetVector as only the PCD value should be allowed
* separated the 'ifndef' checks for the ASM ResetVector, so developers can change the base without needed to set a size

Thanks for your input!

Regards,
Marvin.

> -----Original Message-----
> From: Laszlo Ersek [mailto:lersek@redhat.com]
> Sent: Thursday, November 3, 2016 12:40 AM
> To: Jordan Justen <jordan.l.justen@intel.com>; Marvin Häuser
> <Marvin.Haeuser@outlook.com>
> Cc: edk2-devel@lists.01.org <edk2-devel@ml01.01.org>
> Subject: Re: [edk2] [PATCH v1 1/1] OvmfPkg/ResetVector: Depend on PCD
> values of the page tables.
> 
> On 11/03/16 00:26, Laszlo Ersek wrote:
> > On 11/02/16 23:31, Jordan Justen wrote:
> >> The "v1 1/1" isn't needed in the subject. For the first version of a
> >> single patch series, I would expect to just see [PATCH]. (Obviously
> >> this is not too important.)
> >
> > (I think the "v1" implies that Marvin expects to send a v2.)
> >
> >> Email headers seem to indicate that you aren't using git send-email.
> >> This will cause troubles if someday you have a multi-patch series.
> >>
> >> On 2016-11-02 11:00:34, Marvin Häuser wrote:
> >>>
> >>> diff --git a/OvmfPkg/ResetVector/Ia32/PageTables64.asm
> >>> b/OvmfPkg/ResetVector/Ia32/PageTables64.asm
> >>>
> >>>      ;
> >>>      ; Top level Page Directory Pointers (1 * 512GB entry)
> >>>      ;
> >>> -    mov     dword[0x800000], 0x801000 + PAGE_PDP_ATTR
> >>> +    mov     dword[OVMF_SEC_PAGE_TABLES_BASE],
> OVMF_SEC_PAGE_TABLES_BASE + 0x1000 + PAGE_PDP_ATTR
> >>>
> >>>      ;
> >>>      ; Next level Page Directory Pointers (4 * 1GB entries => 4GB)
> >>>      ;
> >>> -    mov     dword[0x801000], 0x802000 + PAGE_PDP_ATTR
> >>> -    mov     dword[0x801008], 0x803000 + PAGE_PDP_ATTR
> >>> -    mov     dword[0x801010], 0x804000 + PAGE_PDP_ATTR
> >>> -    mov     dword[0x801018], 0x805000 + PAGE_PDP_ATTR
> >>> +    mov     dword[OVMF_SEC_PAGE_TABLES_BASE + 0x1000],
> OVMF_SEC_PAGE_TABLES_BASE + 0x2000 + PAGE_PDP_ATTR
> >>> +    mov     dword[OVMF_SEC_PAGE_TABLES_BASE + 0x1008],
> OVMF_SEC_PAGE_TABLES_BASE + 0x3000 + PAGE_PDP_ATTR
> >>> +    mov     dword[OVMF_SEC_PAGE_TABLES_BASE + 0x1010],
> OVMF_SEC_PAGE_TABLES_BASE + 0x4000 + PAGE_PDP_ATTR
> >>> +    mov     dword[OVMF_SEC_PAGE_TABLES_BASE + 0x1018],
> OVMF_SEC_PAGE_TABLES_BASE + 0x5000 + PAGE_PDP_ATTR
> >>
> >> These line are too long. I guess you can use '\' at the end of a line
> >> to continue it, or maybe add a PT_ADDR() macro that adds in
> >> OVMF_SEC_PAGE_TABLES_BASE.
> >>
> >>> diff --git a/OvmfPkg/ResetVector/ResetVector.nasmb
> >>> b/OvmfPkg/ResetVector/ResetVector.nasmb
> >>> index 31ac06a..b47f647 100644
> >>> --- a/OvmfPkg/ResetVector/ResetVector.nasmb
> >>> +++ b/OvmfPkg/ResetVector/ResetVector.nasmb
> >>> @@ -53,6 +53,12 @@
> >>>  %include "Ia32/SearchForSecEntry.asm"
> >>>
> >>>  %ifdef ARCH_X64
> >>> +  %ifndef OVMF_SEC_PAGE_TABLES_BASE
> >>> +    #include <AutoGen.h>
> >>> +    %define OVMF_SEC_PAGE_TABLES_BASE FixedPcdGet32
> (PcdOvmfSecPageTablesBase)
> >>> +    %define OVMF_SEC_PAGE_TABLES_SIZE FixedPcdGet32
> >>> + (PcdOvmfSecPageTablesSize)  %endif
> >>> +
> >>>  %include "Ia32/Flat32ToFlat64.asm"
> >>>  %include "Ia32/PageTables64.asm"
> >>>  %endif
> >>> diff --git a/OvmfPkg/ResetVector/ResetVectorCode.asm
> >>> b/OvmfPkg/ResetVector/ResetVectorCode.asm
> >>> index 052c821..5b49387 100644
> >>> --- a/OvmfPkg/ResetVector/ResetVectorCode.asm
> >>> +++ b/OvmfPkg/ResetVector/ResetVectorCode.asm
> >>> @@ -40,6 +40,15 @@
> >>>  %include "Ia32/SearchForSecEntry.asm"
> >>>
> >>>  %ifdef ARCH_X64
> >>> +  %ifndef OVMF_SEC_PAGE_TABLES_BASE
> >>> +    ;
> >>> +    ; This range should match with PcdOvmfSecPageTablesBase and
> >>> +    ; PcdOvmfSecPageTablesSize which are declared in the FDF files.
> >>> +       ;
> >>> +    %define OVMF_SEC_PAGE_TABLES_BASE 0x800000
> >>> +    %define OVMF_SEC_PAGE_TABLES_SIZE 6 * 0x1000
> >>
> >> I thought we were using the PCDs?
> >
> > We were, yes, except where we couldn't. For the reset vector, our "root"
> > file is
> >
> >   OvmfPkg/ResetVector/ResetVector.nasmb
> >
> > (referenced by
> >
> >   OvmfPkg/ResetVector/ResetVector.inf
> >
> > )
> >
> > That nasmb file %include's a bunch of other files; among those,
> >
> >   OvmfPkg/ResetVector/Ia32/PageTables64.asm
> >
> > Notice that the %include and %define directives in these files are
> > NASM-native directives; they aren't C preprocessor directives. So we
> > couldn't use FixedPcdGetXX() in the source code.
> >
> > If I remember correctly, in commit b382ede3864e ("OvmfPkg X64
> > ResetVector: Move page tables from 512KB to 8MB"), you added the PCD
> > references in a source code comment because this way at least a "grep"
> > would find them, and then the programmer could update the constants
> > (if necessary).
> >
> > Note that in the era of commit b382ede3864e (2014-Jan-21), we used to
> > compile the reset vector manually, with the utility
> >
> >   OvmfPkg/ResetVector/Build.py
> >
> > On 2014-Aug-19 however, you added native NASMB support to BaseTools,
> > and adapted OvmfPkg/ResetVector:
> >
> >   abb158ded41f BaseTools: Add rules to build NASM source file into a binary
> >   5a1f324d946c UefiCpuPkg: Support building VTF0 ResetVector during the
> EDK II build
> >   eee1d2ca9078 UefiCpuPkg VTF0 X64: Build page tables in NASM code
> >   9b9fdbfa7059 OvmfPkg: Support building OVMF's ResetVector during the
> EDK II build
> >   497cbb530a58 OvmfPkg: Build OVMF ResetVector during EDK II build
> process
> >   70e46f44cd13 OvmfPkg/ResetVector: Remove pre-built binaries
> >   3449f56dac9c UefiCpuPkg: Add ResetVector/FixupVtf
> >
> > The BaseTools support would imply automatic preprocessing (with the C
> > preprocessor, $(PP)) for *.nasmb files.
> >
> > We both failed to notice :) that this enabled us to #include
> > <AutoGen.h> in the assembly source -- despite the fact that commit
> > 9b9fdbfa7059 #included <Base.h>!
> >
> > So, we forgot to take advantage of the preprocessor, and to replace
> > the open-coded constants with FixedPcdGetXX().
> >
> > I think the idea in Marvin's patch is solid: it converts the
> > FixedPcdGetXX() macros into NASM-level macros, with
> >
> > #include <AutoGen.h>
> > %define OVMF_SEC_PAGE_TABLES_BASE FixedPcdGet32
> > (PcdOvmfSecPageTablesBase) %define OVMF_SEC_PAGE_TABLES_SIZE
> > FixedPcdGet32 (PcdOvmfSecPageTablesSize)
> >
> > Namely, $(PP) will preprocess the %define directives themselves (see
> > commit abb158ded41f), and then NASM will %define the NASM-level
> macros
> > with the expanded constants already.
> >
> > At this point, I think we could simply use PT_ADDR(), as you suggest:
> > leave it all to the C preprocessor (i.e., $(PP)), and not bother with
> > NASM-level %define's at all. Something like:
> >
> > #include <AutoGen.h>
> > #define PT_ADDR(Offset) (FixedPcdGet32 (PcdOvmfSecPageTablesBase) +
> > (Offset))
> 
> Hold on: this won't work exactly like this. The reason is that CPP-level
> preprocessing happens *first*, and NASM-level %include directives are
> acted upon only *second*.
> 
> So, the above PT_ADDR() macro must be a %define (for NASM), not a
> #define (for the C preprocessor).
> 
> Thanks
> Laszlo
> 
> > I think NASM should be happy with the replacement text. And, this
> > should be done unconditionally.
> >
> > Regarding the size of the area: I think that's a harder question,
> > because the content is not just a plain array. The code inherently
> > open-codes the size constant. So, rather than introducing
> > OVMF_SEC_PAGE_TABLES_SIZE (purely for documentation purposes),
> how
> > about:
> >
> > #if (FixedPcdGet32 (PcdOvmfSecPageTablesSize) != 0x6000) #error "This
> > implementation inherently depends on PcdOvmfSecPageTablesSize"
> > #endif
> >
> > Consequences:
> >
> > - When PcdOvmfSecPageTablesBase is changed in the FDF files, the reset
> >   vector code follows it automatically.
> >
> > - When PcdOvmfSecPageTablesSize is changed in the FDF files, the reset
> >   vector code needs a manual rewrite -- and now the build will break
> >   loudly, accordingly.
> >
> > Thanks
> > Laszlo
> > _______________________________________________
> > edk2-devel mailing list
> > edk2-devel@lists.01.org
> > https://lists.01.org/mailman/listinfo/edk2-devel
> >