From:   =?utf-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn@kernel.org>
To:     Guo Ren <guoren@kernel.org>
Cc:     palmer@rivosinc.com, conor.dooley@microchip.com,
        liaochang1@huawei.com, linux-arch@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-riscv@lists.infradead.org,
        Guo Ren <guoren@linux.alibaba.com>,
        Bjorn Topel <bjorn.topel@gmail.com>
Subject: Re: [PATCH] riscv: kprobe: Fixup misaligned load text
In-Reply-To: <CAJF2gTSj12E9+peMF0rNY_psGDyEG5BbMY6V-s4dK0FpkCC9Yw@mail.gmail.com>
References: <20230201064608.3486136-1-guoren@kernel.org>
 <87tu05pvur.fsf@all.your.base.are.belong.to.us>
 <CAJF2gTSj12E9+peMF0rNY_psGDyEG5BbMY6V-s4dK0FpkCC9Yw@mail.gmail.com>
Date:   Thu, 02 Feb 2023 10:48:37 +0100
Message-ID: <87cz6s75ze.fsf@all.your.base.are.belong.to.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Guo Ren <guoren@kernel.org> writes:

> On Wed, Feb 1, 2023 at 5:40 PM Bj=C3=B6rn T=C3=B6pel <bjorn@kernel.org> w=
rote:
>>
>> guoren@kernel.org writes:
>>
>> > From: Guo Ren <guoren@linux.alibaba.com>
>> >
>> > The current kprobe would cause a misaligned load for the probe point.
>> > This patch fixup it with two half-word loads instead.
>> >
>> > Fixes: c22b0bcb1dd0 ("riscv: Add kprobes supported")
>> > Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
>> > Signed-off-by: Guo Ren <guoren@kernel.org>
>> > Link: https://lore.kernel.org/linux-riscv/878rhig9zj.fsf@all.your.base=
.are.belong.to.us/
>> > Reported-by: Bjorn Topel <bjorn.topel@gmail.com>
>> > ---
>> >  arch/riscv/kernel/probes/kprobes.c | 4 +++-
>> >  1 file changed, 3 insertions(+), 1 deletion(-)
>> >
>> > diff --git a/arch/riscv/kernel/probes/kprobes.c b/arch/riscv/kernel/pr=
obes/kprobes.c
>> > index 41c7481afde3..c1160629cef4 100644
>> > --- a/arch/riscv/kernel/probes/kprobes.c
>> > +++ b/arch/riscv/kernel/probes/kprobes.c
>> > @@ -74,7 +74,9 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
>> >               return -EILSEQ;
>> >
>> >       /* copy instruction */
>> > -     p->opcode =3D *p->addr;
>> > +     p->opcode =3D (kprobe_opcode_t)(*(u16 *)probe_addr);
>> > +     if (GET_INSN_LENGTH(p->opcode) =3D=3D 4)
>> > +             p->opcode |=3D (kprobe_opcode_t)(*(u16 *)(probe_addr + 2=
))
>> >       << 16;
>>
>> Ugh, those casts. :-( What about the memcpy variant you had in the other
>> thread?
> The memcpy version would force load probe_addr + 2. This one would
> save an lh operation. The code text guarantees half-word alignment. No
> misaligned load happened. Second, kprobe wouldn't write the last half
> of 32b instruction.

Ok, something more readable, like:

diff --git a/arch/riscv/kernel/probes/kprobes.c b/arch/riscv/kernel/probes/=
kprobes.c
index f21592d20306..3602352ba175 100644
--- a/arch/riscv/kernel/probes/kprobes.c
+++ b/arch/riscv/kernel/probes/kprobes.c
@@ -50,14 +50,16 @@ static void __kprobes arch_simulate_insn(struct kprobe =
*p, struct pt_regs *regs)
=20
 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 {
-	unsigned long probe_addr =3D (unsigned long)p->addr;
+	u16 *insn =3D (u16 *)p->addr;
=20
-	if (probe_addr & 0x1)
+	if ((uintptr_t)insn & 0x1)
 		return -EILSEQ;
=20
 	/* copy instruction */
-	p->opcode =3D *p->addr;
-
+	p->opcode =3D *insn++;
+	if (GET_INSN_LENGTH(p->opcode) =3D=3D 4)
+		p->opcode |=3D *insn << 16;
+=09
 	/* decode instruction */
 	switch (riscv_probe_decode_insn(p->addr, &p->ainsn.api)) {
 	case INSN_REJECTED:	/* insn not supported */


Bj=C3=B6rn