[PATCH 3/3] powerpc/kprobes: Check return value of patch_instruction()
Steven Rostedt
rostedt at goodmis.org
Fri Apr 24 23:22:02 AEST 2020
On Thu, 23 Apr 2020 17:41:52 +0200
Christophe Leroy <christophe.leroy at c-s.fr> wrote:
> > diff --git a/arch/powerpc/kernel/optprobes.c b/arch/powerpc/kernel/optprobes.c
> > index 024f7aad1952..046485bb0a52 100644
> > --- a/arch/powerpc/kernel/optprobes.c
> > +++ b/arch/powerpc/kernel/optprobes.c
> > @@ -139,52 +139,67 @@ void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
> > }
> > }
> >
> > +#define PATCH_INSN(addr, instr) \
> > +do { \
> > + int rc = patch_instruction((unsigned int *)(addr), instr); \
> > + if (rc) { \
> > + pr_err("%s:%d Error patching instruction at 0x%pK (%pS): %d\n", \
> > + __func__, __LINE__, \
> > + (void *)(addr), (void *)(addr), rc); \
> > + return rc; \
> > + } \
> > +} while (0)
> > +
>
> I hate this kind of macro which hides the "return".
>
> What about keeping the return action in the caller ?
>
> Otherwise, what about implementing something based on the use of goto,
> on the same model as unsafe_put_user() for instance ?
#define PATCH_INSN(addr, instr) \
({
int rc = patch_instruction((unsigned int *)(addr), instr); \
if (rc) \
pr_err("%s:%d Error patching instruction at 0x%pK (%pS): %d\n", \
__func__, __LINE__, \
(void *)(addr), (void *)(addr), rc); \
rc; \
})
Then you can just do:
ret = PATCH_INSN(...);
if (ret)
return ret;
in the code.
-- Steve
More information about the Linuxppc-dev
mailing list