On Wed, Apr 09, 2025 at 03:11:16PM +0200, Oleg Nesterov wrote:
> On 04/09, Jiri Olsa wrote:
> >
> > On Wed, Apr 09, 2025 at 01:28:39PM +0200, Oleg Nesterov wrote:
> > > On 04/08, Jiri Olsa wrote:
> > > >
> > > > --- a/arch/x86/kernel/uprobes.c
> > > > +++ b/arch/x86/kernel/uprobes.c
> > > > @@ -608,6 +608,16 @@ static void riprel_post_xol(struct arch_uprobe
> > > > *auprobe, struct pt_regs *regs)
> > > > *sr = utask->autask.saved_scratch_register;
> > > > }
> > > > }
> > > > +
> > > > +static int is_nop5_insn(uprobe_opcode_t *insn)
> > > > +{
> > > > + return !memcmp(insn, x86_nops[5], 5);
> > > > +}
> > > > +
> > > > +static bool emulate_nop5_insn(struct arch_uprobe *auprobe)
> > > > +{
> > > > + return is_nop5_insn((uprobe_opcode_t *) &auprobe->insn);
> > > > +}
> > >
> > > Why do we need 2 functions? Can't branch_setup_xol_ops() just use
> > > is_nop5_insn(insn->kaddr) ?
> >
> > I need is_nop5_insn in other changes I have in queue, so did not want
> > to introduce extra changes
>
> But I didn't suggest to remove is_nop5_insn(), I meant that
> branch_setup_xol_ops() can do
>
> if (is_nop5_insn(insn->kaddr))
> goto setup;
> or
> if (is_nop5_insn(auprobe->insn))
> goto setup;
>
> this even looks more readable to me. but I won't insist.
>
> > > For the moment, lets forget about compat tasks on a 64-bit kernel, can't
> > > we simply do something like below?
> >
> > I sent similar change (CONFIG_X86_64 only) in this thread:
> >
> > https://lore.kernel.org/bpf/Z_O0Z1ON1YlRqyny@krava/T/#m59c430fb5a30cb9faeb9587fd672ea0adbf3ef4f
> >
> > uprobe won't attach on nop9/10/11 atm,
>
> Ah, OK, I didn't know. But this means that nop9/10/11 will be rejected
> by uprobe_init_insn() -> is_prefix_bad() before branch_setup_xol_ops() is
> called, right? So I guess it is safe to use ASM_NOP_MAX. Nevermind.
>
> > also I don't have practical justification
> > for doing that.. nop5 seems to have future, because of the optimization
>
> OK, I won't insist, up to you.
>
> Just it looks a bit strange to me. Even if we do not have a use-case
> for other nops, why we can't emulate them all just for consistency?
we can, I went with nop5 just for simplicity, if you think
having all nops support is better, let's do that
I checked and compact process executes 64bit nops just fine,
so we should be ok there
>
> And given that emulate_nop5_insn() compares the whole insn with
> x86_nops[5], I guess we don't even need to check OPCODE1(insn)...
right
> Nevermind.
>
> So, once again, I won't argue.
I'm happy to go with your version, wdyt?
thanks,
jirka
---
diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index 9194695662b2..63ecc5f6c235 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -840,12 +840,16 @@ static int branch_setup_xol_ops(struct arch_uprobe
*auprobe, struct insn *insn)
insn_byte_t p;
int i;
+ /* x86_nops[i]; same as jmp with .offs = 0 */
+ for (i = 1; i <= ASM_NOP_MAX; ++i) {
+ if (!memcmp(insn->kaddr, x86_nops[i], i))
+ goto setup;
+ }
+
switch (opc1) {
case 0xeb: /* jmp 8 */
case 0xe9: /* jmp 32 */
break;
- case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */
- goto setup;
case 0xe8: /* call relative */
branch_clear_offset(auprobe, insn);
