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

[Xi Ruoyao <xry111@xry111.site>]

On Wed, 2024-06-05 at 18:57 +0800, Xi Ruoyao wrote:
> On Tue, 2024-06-04 at 23:25 -0700, Nathan Chancellor wrote:
> > On Wed, Jun 05, 2024 at 01:54:24PM +0800, Xi Ruoyao wrote:
> > > On Tue, 2024-06-04 at 22:43 -0700, Nathan Chancellor wrote:
> > > > For what it's worth, I have noticed some warnings with clang that I
> > > > don't see with GCC but I only filed an issue on our GitHub and never
> > > > followed up on the mailing list, so sorry about that.
> > > > 
> > > > https://github.com/ClangBuiltLinux/linux/issues/2024
> > > > 
> > > > Might be tangential to this patch though but I felt it was worth
> > > > mentioning.
> > > 
> > > The warnings in GCC build is definitely the issue handled by this patch.
> > > But the warnings in Clang build should be a different issue.  Can you
> > > attach the kernel/events/core.o file from the Clang build for analysis?
> > > I guess we need to disable more optimization...
> > 
> > Sure thing. Let me know if there are any issues with the attachment.
> 
> Thanks!  I've simplified it and now even...
> 
> .global test
> .type test,@function
> test:
> 
> addi.d	$sp,$sp,-448
> st.d	$ra,$sp,440
> st.d	$fp,$sp,432
> addi.d	$fp,$sp,448
> 
> # do something
> 
> addi.d	$sp,$fp,-448
> ld.d	$fp,$sp,432
> ld.d	$ra,$sp,440
> addi.d	$sp,$sp,448
> ret
> 
> .size test,.-test
> 
> is enough to trigger a objtool warning:
> 
> /home/xry111/t1.o: warning: objtool: test+0x20: return with modified stack frame
> 
> And to me this warning is bogus?

Minimal C reproducer:

struct x { _Alignas(64) char buf[128]; };

void f(struct x *p);
void g()
{
	struct x x = { .buf = "1145141919810" };
	f(&x);
}

Then objtool is unhappy to the object file produced with "clang -c -O2"
from this translation unit:

/home/xry111/t2.o: warning: objtool: g+0x50: return with modified stack frame

It seems CFI_BP has a very specific semantic in objtool and Clang does
not operates $fp in the expected way.  I'm not sure about my conclusion
though.  Maybe Peter can explain it better.

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