Re: [PATCH 01/13] kernel/irq/proc: use seq_put_decimal_ull_width() for decimal values

From: Geert Uytterhoeven
Date: Tue Nov 19 2024 - 14:55:56 EST


Hi David,

On Sat, 9 Nov 2024, David Wang wrote:
seq_printf() is costy, on a system with m interrupts and n CPUs, there
would be m*n decimal values yield via seq_printf() when reading
/proc/interrupts, the cost parsing format strings grows with number of
CPU. Profiling on a x86 8-core system indicates seq_printf() takes ~47%
samples of show_interrupts(), and replace seq_printf() with
seq_put_decimal_ull_width() could have near 30% performance gain.

The improvement has pratical significance, considering many monitoring
tools would read /proc/interrupts periodically.

Signed-off-by: David Wang <00107082@xxxxxxx>

Thanks for your patch, which is now commit f9ed1f7c2e26fcd1
("genirq/proc: Use seq_put_decimal_ull_width() for decimal values")
in irqchip/irq/core.

This removes a space after the last CPU column, causing the values in
this column to be concatenated to the values in the next column.

E.g. on Koelsch (R-Car M-W), the output changes from:

CPU0 CPU1
27: 1871 2017 GIC-0 27 Level arch_timer
29: 646 0 GIC-0 205 Level e60b0000.i2c
30: 0 0 GIC-0 174 Level ffca0000.timer
31: 0 0 GIC-0 36 Level e6050000.gpio
32: 0 0 GIC-0 37 Level e6051000.gpio
[...]

to

CPU0 CPU1
27: 1966 1900GIC-0 27 Level arch_timer
29: 580 0GIC-0 205 Level e60b0000.i2c
30: 0 0GIC-0 174 Level ffca0000.timer
31: 0 0GIC-0 36 Level e6050000.gpio
32: 0 0GIC-0 37 Level e6051000.gpio
[...]

making the output hard to read, and probably breaking scripts that parse
its contents.

Reverting the commit fixes the issue for me.

--- a/kernel/irq/proc.c
+++ b/kernel/irq/proc.c
@@ -494,9 +494,11 @@ int show_interrupts(struct seq_file *p, void *v)
if (!desc->action || irq_desc_is_chained(desc) || !desc->kstat_irqs)
goto outsparse;

- seq_printf(p, "%*d: ", prec, i);
+ seq_printf(p, "%*d:", prec, i);
for_each_online_cpu(j)
- seq_printf(p, "%10u ", desc->kstat_irqs ? per_cpu(desc->kstat_irqs->cnt, j) : 0);
+ seq_put_decimal_ull_width(p, " ",
+ desc->kstat_irqs ? per_cpu(desc->kstat_irqs->cnt, j) : 0,
+ 10);

raw_spin_lock_irqsave(&desc->lock, flags);
if (desc->irq_data.chip) {

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds