Re: [RFC][PATCH 4/5] objtool: Enable compilation of objtool for all architectures

From: Matt Helsley
Date: Mon May 18 2020 - 15:50:49 EST


On Fri, May 15, 2020 at 03:56:10PM -0500, Josh Poimboeuf wrote:
> On Mon, May 11, 2020 at 10:35:12AM -0700, Matt Helsley wrote:
> > +struct insn_state {
> > + struct cfi_reg cfa;
> > + struct cfi_reg regs[CFI_NUM_REGS];
> > + int stack_size;
> > + unsigned char type;
> > + bool bp_scratch;
> > + bool drap, end, uaccess, df;
> > + bool noinstr;
> > + s8 instr;
> > + unsigned int uaccess_stack;
> > + int drap_reg, drap_offset;
> > + struct cfi_reg vals[CFI_NUM_REGS];
> > +};
> > +
> > +struct instruction {
> > + struct list_head list;
> > + struct hlist_node hash;
> > + struct section *sec;
> > + unsigned long offset;
> > + unsigned int len;
> > + enum insn_type type;
> > + unsigned long immediate;
> > + bool alt_group, dead_end, ignore, hint, save, restore, ignore_alts;
> > + bool retpoline_safe;
> > + s8 instr;
> > + u8 visited;
> > + struct symbol *call_dest;
> > + struct instruction *jump_dest;
> > + struct instruction *first_jump_src;
> > + struct rela *jump_table;
> > + struct list_head alts;
> > + struct symbol *func;
> > + struct stack_op stack_op;
> > + struct insn_state state;
> > + struct orc_entry orc;
> > +};
>
> Why were these moved to arch.h? They're not necessarily arch-specific,
> but rather "check"-specific, so I think they still belong in check.h, if
> possible.

Ah, found it. They are arch specific due to struct orc_entry, which is
presently not defined for any archs besides x86.

Prior to the patch (-> means "includes"):
check.h -> asm/orc_types.h (defines struct orc_entry)
orc_gen.c -> check,h

After patch:
check.c -> asm/orc_types.h
orc_gen.c -> asm/orc_types.h
orc_gen.c -> check.h
orc_gen.c -> arch.h
{ now weak.c } -> check.h

So this prevents the headers, which help us keep the weak definitions
consistent with the strong definitions, from breaking compiles on archs
that lack struct orc_entry.

I'm not sure what the best way to remove this dependency is without
a nasty void * for the orc entry, or some #ifdef games related to
checking for cpp defines from asm/orc_types.h. This approach neatly
avoids conditional preprocessor games and type casting though I do
agree it's surprising.

Do you have any advice here?

Cheers,
-Matt Helsley