Re: [PATCH 2/4] arm64: implement support for static call trampolines

[Ard Biesheuvel]

On Mon, 25 Oct 2021 at 16:47, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Oct 25, 2021 at 04:19:16PM +0200, Peter Zijlstra wrote:
> > On Mon, Oct 25, 2021 at 04:08:37PM +0200, Ard Biesheuvel wrote:
>
> > > > Ooohh, but what if you go from !func to NOP.
> > > >
> > > > assuming:
> > > >
> > > >         .literal = 0
> > > >         BTI C
> > > >         RET
> > > >
> > > > Then
> > > >
> > > >         CPU0                    CPU1
> > > >
> > > >         [S] literal = func      [I] NOP
> > > >         [S] insn[1] = NOP       [L] x16 = literal (NULL)
> > > >                                 b x16
> > > >                                 *BANG*
> > > >
> > > > Is that possible? (total lack of memory ordering etc..)
> > > >
> > >
> > > The CBZ will branch to the RET instruction if x16 == 0x0, so this
> > > should not happen.
> >
> > Oooh, I missed that :/ I was about to suggest writing the address of a
> > bare 'ret' trampoline instead of NULL into the literal.
>
> Perhaps a little something like so.. Shaves 2 instructions off each
> trampoline.
>
> --- a/arch/arm64/include/asm/static_call.h
> +++ b/arch/arm64/include/asm/static_call.h
> @@ -11,9 +11,7 @@
>             "   hint    34      /* BTI C */                             \n" \
>                 insn "                                                  \n" \
>             "   ldr     x16, 0b                                         \n" \
> -           "   cbz     x16, 1f                                         \n" \
>             "   br      x16                                             \n" \
> -           "1: ret                                                     \n" \
>             "   .popsection                                             \n")
>
>  #define ARCH_DEFINE_STATIC_CALL_TRAMP(name, func)                      \
> --- a/arch/arm64/kernel/patching.c
> +++ b/arch/arm64/kernel/patching.c
> @@ -90,6 +90,11 @@ int __kprobes aarch64_insn_write(void *a
>         return __aarch64_insn_write(addr, &i, AARCH64_INSN_SIZE);
>  }
>
> +asm("__static_call_ret:                \n"
> +    "  ret                     \n")
> +

This breaks BTI as it lacks the landing pad, and it will be called indirectly.

> +extern void __static_call_ret(void);
> +

Better to have an ordinary C function here (with consistent linkage),
but we need to take the address in a way that works with Clang CFI.

As the two additional instructions are on an ice cold path anyway, I'm
not sure this is an obvious improvement tbh.

>  void arch_static_call_transform(void *site, void *tramp, void *func, bool tail)
>  {
>         /*
> @@ -97,9 +102,7 @@ void arch_static_call_transform(void *si
>          *  0x0 bti c           <--- trampoline entry point
>          *  0x4 <branch or nop>
>          *  0x8 ldr x16, <literal>
> -        *  0xc cbz x16, 20
> -        * 0x10 br x16
> -        * 0x14 ret
> +        *  0xc br x16
>          */
>         struct {
>                 u64     literal;
> @@ -113,6 +116,7 @@ void arch_static_call_transform(void *si
>         insns.insn[0] = cpu_to_le32(insn);
>
>         if (!func) {
> +               insns.literal = (unsigned long)&__static_call_ret;
>                 insn = aarch64_insn_gen_branch_reg(AARCH64_INSN_REG_LR,
>                                                    AARCH64_INSN_BRANCH_RETURN);
>         } else {