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 13:21 +0800, Huacai Chen wrote:
> On Wed, Jun 5, 2024 at 12:38 PM Xi Ruoyao <xry111@xry111.site> wrote:
> > 
> > On Wed, 2024-06-05 at 09:52 +0800, Huacai Chen wrote:
> > > Hi, Ruoyao,
> > > 
> > > On Tue, Jun 4, 2024 at 11:08 PM Xi Ruoyao <xry111@xry111.site> wrote:
> > > > 
> > > > GAS <= 2.41 does not support generating R_LARCH_{32,64}_PCREL for
> > > > "label - ." and it generates R_LARCH_{ADD,SUB}{32,64} pairs instead.
> > > > objtool cannot handle R_LARCH_{ADD,SUB}{32,64} pair in __jump_table
> > > > (static key implementation) and etc. so it will produce some warnings.
> > > > This is causing the kernel CI systems to complain everywhere.
> > > > 
> > > > For GAS we can check if -mthin-add-sub option is available to know if
> > > > R_LARCH_{32,64}_PCREL are supported.
> > > I think we should give a chance to toolchains without the
> > > -mthin-add-sub option, so I hope Tiezhu can resubmit this patch to
> > > solve this problem.
> > > https://lore.kernel.org/loongarch/1690272910-11869-6-git-send-email-yangtiezhu@loongson.cn/
> > 
> > It only handles .discard.{un,}reachable but not __jump_table.
> > __jump_table is our main issue here.  And I don't see a quick way to
> > make -mno-thin-add-sub work for __jump_table because we really need to
> > resolve the relocation for __jump_table instead of simply skipping it.
>  __jump_table is disabled now, so we can only solve  -mno-thin-add-sub
> at present, and the next step is __jump_table.

No, -fno-jump-table does not disable __jump_table.  __jump_table is from
static key implementation in arch/loongarch/include/asm/jump_label.h,
not generated by the compiler.

And the problem of __jump_table is exact with -mno-thin-add-sub.  It
reads:

#define JUMP_TABLE_ENTRY                \
     ".pushsection  __jump_table, \"aw\"    \n\t"   \
     ".align    3           \n\t"   \
     ".long     1b - ., %l[l_yes] - .   \n\t"   \
     ".quad     %0 - .          \n\t"   \
     ".popsection               \n\t"

The "1b - ." and "%0 - ." things produce different relocs with -mno-
thin-add-sub and -mthin-add-sub.  The objtool can only handle the relocs
from -mthin-add-sub here.

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