RE: [PATCH 4/6] perf: Optimize get_recursion_context()

From: David Laight
Date: Sat Oct 31 2020 - 08:16:27 EST


From: Peter Zijlstra
> Sent: 30 October 2020 23:02
>
> On Fri, Oct 30, 2020 at 04:22:48PM -0400, Steven Rostedt wrote:
> > As this is something that ftrace recursion also does, perhaps we should
> > move this into interrupt.h so that anyone that needs a counter can get
> > it quickly, and not keep re-implementing it.
>
> Works for me, however:
>
> > /*
> > * Quickly find what context you are in.
> > * 0 - normal
> > * 1 - softirq
> > * 2 - hard interrupt
> > * 3 - NMI
> > */
> > static inline int irq_context()
> > {
> > unsigned int pc = preempt_count();
> > int rctx = 0;
>
> unsigned
>
> >
> > if (pc & (NMI_MASK))
> > rctx++;
> > if (pc & (NMI_MASK | HARDIRQ_MASK))
> > rctx++;
> > if (pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET))
> > rctx++;
> >
> > return rctx;
> > }
>
> otherwise you'll get an extra instruction to sign extend it, which is
> daft (yes, i've been staring at GCC output far too much).
>
> Also, gcc-9 does worse (like 1 byte iirc) with:
>
> rctx += !!(pc & (NMI_MASK));
> rctx += !!(pc & (NMI_MASK | HARDIRQ_MASK));
> rctx += !!(pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET));
>
> but gcc-10 doesn't seem to care.

You've made be look at some gcc output (it's raining).

The gcc 7.5.0 I have handy probably generates the best code for:

unsigned char q_2(unsigned int pc)
{
unsigned char rctx = 0;

rctx += !!(pc & (NMI_MASK));
rctx += !!(pc & (NMI_MASK | HARDIRQ_MASK));
rctx += !!(pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET));

return rctx;
}

0000000000000000 <q_2>:
0: f7 c7 00 00 f0 00 test $0xf00000,%edi # clock 0
6: 0f 95 c0 setne %al # clock 1
9: f7 c7 00 00 ff 00 test $0xff0000,%edi # clock 0
f: 0f 95 c2 setne %dl # clock 1
12: 01 c2 add %eax,%edx # clock 2
14: 81 e7 00 01 ff 00 and $0xff0100,%edi
1a: 0f 95 c0 setne %al
1d: 01 d0 add %edx,%eax # clock 3
1f: c3 retq

I doubt that is beatable.

I've annotated the register dependency chain.
Likely to be 3 (or maybe 4) clocks.
The other versions are a lot worse (7 or 8) without allowing
for 'sbb' taking 2 clocks on a lot of Intel cpus.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)