Re: [PATCH] clocksource/drivers/arm_arch_timer: Mark validate_timer_rate() init

From: Stephen Boyd
Date: Sat May 04 2024 - 02:00:06 EST


Quoting Mukesh Ojha (2024-05-03 04:27:52)
>
>
> On 5/3/2024 5:00 AM, Stephen Boyd wrote:
> > Add the __init marking to validate_timer_rate() so that it gets
> > discarded after init.
> >
> > $ ./scripts/bloat-o-meter vmlinux.before vmlinux.after
> > add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-8 (-8)
> > Function old new delta
> > arch_timer_register 1724 1716 -8
>
> Have not used this tool bloat-o-meter
> Just a question, why is it showing arch_timer_register is shrunk ?

Hmm good question. There shouldn't be any difference if it gets inlined.

It's an allmodconfig build. It looks like that causes noise in the build so it
isn't stable. :( I diffed the objdump and the function is missing this ror and
nop instruction. But then I built it again, and it became more different. If I
build an allnoconfig it has zero difference.

Before:

ffff800083021ed4: 974ba12d bl ffff80008030a388
<free_percpu_irq>
ffff800083021ed8: 93d3f273 ror x19, x19, #60
ffff800083021edc: d503201f nop
ffff800083021ee0: d29858e0 mov x0, #0xc2c7
// #49863
ffff800083021ee4: 910102b6 add x22, x21, #0x40
ffff800083021ee8: f2aeea20 movk x0, #0x7751, lsl #16
ffff800083021eec: f2ce7680 movk x0, #0x73b4, lsl #32
ffff800083021ef0: f2f09e20 movk x0, #0x84f1, lsl #48
ffff800083021ef4: ca000273 eor x19, x19, x0
ffff800083021ef8: 97527f02 bl ffff8000804c1b00
<__sanitizer_cov_trace_pc>

After:

ffff800083021ed4: 974ba12d bl ffff80008030a388
<free_percpu_irq>
ffff800083021ed8: d29f9f00 mov x0, #0xfcf8
// #64760
ffff800083021edc: 910102b6 add x22, x21, #0x40
ffff800083021ee0: f2b03940 movk x0, #0x81ca, lsl #16
ffff800083021ee4: f2d5e700 movk x0, #0xaf38, lsl #32
ffff800083021ee8: f2ee1340 movk x0, #0x709a, lsl #48
ffff800083021eec: ca000273 eor x19, x19, x0
ffff800083021ef0: 97527f04 bl ffff8000804c1b00
<__sanitizer_cov_trace_pc>

Maybe kcov is making things different?