On 03/31/2015 07:08 PM, Andy Lutomirski wrote:
> On Tue, Mar 31, 2015 at 9:43 AM, Ingo Molnar <mi...@kernel.org> wrote:
>>
>> * Denys Vlasenko <vda.li...@googlemail.com> wrote:
>>
>>>> I guess they could optimize it by adding a single "I am a modern
>>>> OS executing regular userspace" flag to the descriptor [or
>>>> expressing the same as a separate instruction], to avoid all that
>>>> legacy crap that won't trigger on like 99.999999% of systems ...
>>>
>>> Yes, that would be a useful addition. Interrupt servicing on x86
>>> takes a non-negligible hit because of IRET slowness.
>>
>> But ... to react to your other patch: detecting the common easy case
>> and doing a POPF+RET ourselves ought to be pretty good as well?
>>
>> But only if ptregs->rip != the magic RET itself, to avoid recursion.
>>
>> Even with all those extra checks it should still be much faster.
>>
>
> I have a smallish preference for doing sti;ret instead, because that
> keeps the funny special case entirely localized to the NMI code
> instead of putting it in the IRQ exit path. I suspect that the
> performance loss is at most a cycle or two (we're adding a branch, but
> sti itself is quite fast).
>
> That being said, I could easily be convinced otherwise.
Let me try to convince you. sti is 6 cycles.
The patch atop your code would be:
movq RIP-ARGOFFSET(%rsp), %rcx
+ cmp $magic_ret, %rcx
+ je real_iret
- btr $9, %rdi
movq %rdi, (%rsi)
movq %rcx, 8(%rsi)
movq %rsi, ORIG_RAX-ARGOFFSET(%rsp)
popq_cfi %r11
popq_cfi %r10
popq_cfi %r9
popq_cfi %r8
popq_cfi %rax
popq_cfi %rcx
popq_cfi %rdx
popq_cfi %rsi
popq_cfi %rdi
popq %rsp
- jc 1f
popfq_cfi
+magic_ret:
retq
-1:
- popfq_cfi
- sti
- retq
It's a clear (albeit small) win: the branch is almost never taken,
and we do not need sti.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
