Re: [PATCH] MIPS: support PCREL GOT access

[YunQiang Su <syq@gcc.gnu.org>]

Maciej W. Rozycki <macro@orcam.me.uk> 于2024年3月2日周六 01:38写道：
>
> On Mon, 26 Feb 2024, YunQiang Su wrote:
>
> > >  We can try and do our best to produce correctly linked executables and
> > > shared libraries and provide testsuite coverage for what we do, and then
> >
> > What my plan is (although I have no idea how to archive it with dejagnu):
> >    1. f.c
> >        int f() {return 0;}
> >        mipsel-linux-gnu-gcc -O2 -mabi=32 f.c -shared -o libf.so
> >    2. f1.c
> >        extern int f();
> >        int f1 () {return f();}
> >        mipsel-linux-gnu-gcc -O2 f1.c -shared -L. -lf -o libf1.so
>
>  But why do you want to write this code in C?  It's so trivial you can
> produce assembly for all the ABIs you want to cover (you can even start
> with compiler output if you don't want to handcode it from scratch) and
> use that as the test sources.  We'll be able to verify it reliably then,
> without a need for external components.
>

Sure, asm code is preferred by binutils testcase.
The above C examples are just to describe what we need to do.
In the real patch, I will use asm code.

> >    3. Let's inspect libf1.so with `objdump -DTC`
> >        3.1 We can see the symbol 'f' is in an address, 00000520.
> >              00000520      DF *UND*  00000000 f
> >        3.2 We can see a line "00000520 <_MIPS_STUBS_>"
> >        3.3 We can see the address of .got is "00010570 <__TMC_END__>"
> >        3.3 We can see " 50c:   8f998040        lw      t9,-32704(gp)"
> > in <f1>, and
> >              GOT_ADR   +   GP_MAGIC + LW_OFFSET  = GOT_SLOT_OF_F
> >              0x00010570 +  0x7ff0           + (-32704)          = 0x105a0
> >        3.4 The content of address GOT_SLOT_OF_F (0x105a0) is 0x00000520.
>
>  There are existing test cases in ld/testsuite/ld-mips-elf/mips-elf.exp
> that do various kinds of checks, including running multiple dump tools on
> single linker output.  Check the various `run_ld_link_tests' invocations.
> You can examine GOT entries in a cooked way with `readelf -A' BTW.
>

Thanks. I will have a try.

>   Maciej