Re: [PATCH] loongarch: Only select HAVE_OBJTOOL and allow ORC unwinder if the inline assembler supports R_LARCH_{32,64}_PCREL

[Huacai Chen]

On Sat, Jun 15, 2024 at 4:54 PM Xi Ruoyao <xry111@xxxxxxxxxxx> wrote:
>
> On Sat, 2024-06-15 at 16:45 +0800, Huacai Chen wrote:
> > Hi, Ruoyao and Jinyang,
> >
> > On Fri, Jun 7, 2024 at 4:29 PM Xi Ruoyao <xry111@xxxxxxxxxxx> wrote:
> > >
> > > On Fri, 2024-06-07 at 15:14 +0800, Jinyang He wrote:
> > > > >      Note: on RISC-V and LoongArch, the stack slot for the previous frame
> > > > >      pointer is stored at fp[-2] instead of fp[0]. See [Consider
> > > > >      standardising which stack slot fp points
> > > > >      to](https://github.com/riscv-non-isa/riscv-elf-psabi-doc/issues/18)
> > > > >      for the RISC-V discussion.
> > > >
> > > > In most cases the $fp is saved at cfa-16. But for va args, something
> > > > becomes different at LoongArch (I do not know the case of riscv), the
> > > > $fp isn't saved at cfa-16. (e.g. printk?)
> > >
> > > Oops indeed.  Even with a very simple case:
> > >
> > > int sum(int a, int b) {
> > >         return a + b;
> > > }
> > >
> > > with -fno-omit-frame-pointer we get:
> > >
> > > sum:
> > >         addi.d  $r3,$r3,-16
> > >         st.d    $r22,$r3,8
> > >         addi.d  $r22,$r3,16
> > >         ld.d    $r22,$r3,8
> > >         add.w   $r4,$r4,$r5
> > >         addi.d  $r3,$r3,16
> > >         jr      $r1
> > >
> > > So for leaf functions (where we don't save $ra) $fp is saved at cfa-8.
> > >
> > > > I feel that the update_cfi_state should be arch specific. I believe
> > > > that some logic can be reused, but each arch may have its own logic.
> > >
> > > I agree it now.
> > What is the conclusion about the clang part now? And for the original
> > -mno-thin-add-sub problem, do you have some way to fix it in the root?
> > I think we needn't rush, there are some weeks before 6.10 released.
>
> To me for now we should just make OBJTOOL and ORC depend on BROKEN and
> backport to stable...
But this patch allows clang to build objtool, which seems broken, too.

>
> Even if we can fix both the -mno-thin-add-sub problem and the frame
> pointer problem in these weeks, they'll be some nontrivial large change
> and improper to backport.  Thus we have to admit objtool doesn't really
> work for old releases and mark it broken.
I don't like to disable objtool, unless there is no better solution.
And it seems there has already been some "large fix" in objtool's
history.

Huacai

>
>
> --
> Xi Ruoyao <xry111@xxxxxxxxxxx>
> School of Aerospace Science and Technology, Xidian University