Re: [PATCH] arm64: kprobe: fix an error in single stepping support

[Will Deacon <will@kernel.org>]

On Fri, Jan 17, 2025 at 07:57:32AM +0000, Yeoreum Yun wrote:
> > It is obvious a conflict between the code and the comment.
> > The function aarch64_insn_is_steppable is used to check if a mrs
> > instruction can be safe in single-stepping environment, in the
> > comment it says only reading DAIF bits by mrs is safe in
> > single-stepping environment, and other mrs instructions are not. So
> > aarch64_insn_is_steppable should returen "TRUE" if the mrs instruction
> > being single stepped is reading DAIF bits.
> >
> > And have verified using a kprobe kernel module which reads the DAIF bits by
> > function arch_local_irq_save with offset setting to 0x4, confirmed that
> > without this modification, it encounters
> > "kprobe_init: register_kprobe failed, returned -22" error while inserting
> > the kprobe kernel module. and with this modification, it can read the DAIF
> > bits in single-stepping environment.
> >
> > Fixes: 2dd0e8d2d2a1 ("arm64: Kprobes with single stepping support")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Yiren Xie <1534428646@qq.com>
> > ---
> >  arch/arm64/kernel/probes/decode-insn.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/kernel/probes/decode-insn.c b/arch/arm64/kernel/probes/decode-insn.c
> > index 6438bf62e753..22383eb1c22c 100644
> > --- a/arch/arm64/kernel/probes/decode-insn.c
> > +++ b/arch/arm64/kernel/probes/decode-insn.c
> > @@ -40,7 +40,7 @@ static bool __kprobes aarch64_insn_is_steppable(u32 insn)
> >  		 */
> >  		if (aarch64_insn_is_mrs(insn))
> >  			return aarch64_insn_extract_system_reg(insn)
> > -			     != AARCH64_INSN_SPCLREG_DAIF;
> > +			     == AARCH64_INSN_SPCLREG_DAIF;
> >
> >  		/*
> >  		 * The HINT instruction is steppable only if it is in whitelist
> > --
> > 2.34.1
> >
> 
> Thanks to correct me. yes the comments seem conflict.
> 
> However, I couldn't agree to this change.
> As I mention in last, when single-step runs, all DAIF bits set,
> so, the result of reading DAIF is different between before install kprobe and after.
> Also, I think reading some sys_reg in single-step seems ok (i.e. SYS_MIDR_EL1).
> 
> Therefore, allowing only install kprobe on DAIF reading doesn't seem
> correct.

Right, the code seems ok. I think we should just remove the comment
instead.

Will