On Sun, 7 Mar 2021 06:02:16 +0100
Marek Vasut <ma...@denx.de> wrote:
> On 3/7/21 5:58 AM, Marek Behun wrote:
> > On Sun, 7 Mar 2021 05:46:24 +0100
> > Marek Vasut <ma...@denx.de> wrote:
> >
> >> On 3/7/21 5:25 AM, Marek Behún wrote:
> >>> When compiling with LTO, the compiler fails with an error saying that
> >>> `crc_table` causes a section type conflict with `efi_var_buf`.
> >>>
> >>> This is because both are declared to be in the same section (via macro
> >>> `__efi_runtime_data`), but one is const while the other is not.
> >>>
> >>> Make this variable non-const in order to fix this.
> >>
> >> This does not look right, the crc32 array is constant.
> >> Maybe what you want to do instead if create some __efi_constant_data
> >> section ?
> >
> > Yes, this was the easier solution, and maybe is not ideal.
> >
> > I thought it would not be much of a problem since this array can be
> > nonconstant (generated after boot) if CONFIG_DYNAMIC_CRC_TABLE is
> > enabled.
> >
> > Anyway I don't much understand the EFI code so I wanted to poke into it
> > as little as possible.
>
> Isn't the compiler capable of better optimization on constant stuff ?
> That's pretty much what prompted my suggestion to add separate section.
Yes, but
- for this case I don't think the compiler actually can do any
significat optimizations when declaring the table const, since it has
to access
tab[(crc ^ (x)) & 255]
I tried with arm compiler just now, with -O2 and -Os, and it
generated the same code either way
- when using LTO, the compiler theoretically should be able to deduce
that the table is never written to
But if people insist on declaring the table const, I will look into
this...
