Re: [PATCH v2 10/11] x86/alternatives: Simplify ALTERNATIVE_n()

[Borislav Petkov]

On Sat, Sep 09, 2023 at 11:25:54AM +0200, Peter Zijlstra wrote:
> So what you end up with is:
> 
> 661:
>   "one byte orig insn"
>   "one nop because alt1 is 2 bytes"
>   "one nop because alt2 is 3 bytes"
> 
> right?

Right.

> This becomes more of a problem with your example above where the
> respective lengths are 0, 5, 16. In that case, when you patch 5, you'll
> leave 11 single nops in there.

Well, I know what you mean but the code handles that gracefully and it
works. Watch this:

I made it always apply the second one and not apply the third, the
longest one:

.macro UNTRAIN_RET
#if defined(CONFIG_CPU_UNRET_ENTRY) || defined(CONFIG_CPU_IBPB_ENTRY) || \
        defined(CONFIG_CALL_DEPTH_TRACKING) || defined(CONFIG_CPU_SRSO)
        VALIDATE_UNRET_END
        ALTERNATIVE_3 "",                                               \
                      CALL_UNTRAIN_RET, X86_FEATURE_UNRET,              \
                      "call entry_ibpb", X86_FEATURE_ALWAYS,            \
                      __stringify(RESET_CALL_DEPTH), X86_FEATURE_CALL_DEPTH
#endif
.endm

So it comes in and pokes in the padding for the first one:
X86_FEATURE_UNRET

[    0.903506] SMP alternatives: feat: 11*32+15, old: (entry_SYSCALL_64_after_hwframe+0x59/0xd8 (ffffffff81c000d1) len: 5), repl: (ffffffff833a362b, len: 5)
[    0.911256] SMP alternatives: ffffffff81c000d1: [0:5) optimized NOPs: 0f 1f 44 00 00

Then patches in the entry_ibpb call:

[    0.916849] SMP alternatives: feat: 3*32+21, old: (entry_SYSCALL_64_after_hwframe+0x59/0xd8 (ffffffff81c000d1) len: 5), repl: (ffffffff833a3630, len: 5)
[    0.924844] SMP alternatives: ffffffff81c000d1:   old_insn: 0f 1f 44 00 00
[    0.928842] SMP alternatives: ffffffff833a3630:   rpl_insn: e8 5b 9e 81 fe
[    0.932849] SMP alternatives: ffffffff81c000d1: final_insn: e8 ba d3 fb ff

and now it comes to the call depth thing which is of size 16:

[    0.936845] SMP alternatives: feat: 11*32+19, old: (entry_SYSCALL_64_after_hwframe+0x59/0xd8 (ffffffff81c000d1) len: 16), repl: (ffffffff833a3635, len: 16)
[    0.940844] SMP alternatives: __optimize_nops: next: 5, insn len: 5

and this is why it works: __optimize_nops() is cautious enough to do
insn_is_nop() and there is the CALL insn: e8 ba d3 fb ff, it skips over
it:

[    0.944852] SMP alternatives: __optimize_nops: next: 6, insn len: 1

next one is a NOP and it patches the rest of it. Resulting in an 11
bytes NOP:

[    0.950758] SMP alternatives: ffffffff81c000d1: [5:16) optimized NOPs: e8 ba d3 fb ff 66 66 2e 0f 1f 84 00 00 00 00 00

So we're good here without this max(repl_len) thing even if it is the
right thing to do.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette