sum of interrupts/s overflows

[marcobarlo]

The mpstat output regarding the sum of the interrupts per core overflows under certain conditions. On my PC I ran mpstat under stress-ng --udp and I have output similar to this:

11:20:57 CPU intr/s
11:21:00 all 1880,00
11:21:00 0 395027,00
11:21:00 1 399987,33
11:21:00 2 365727,00
11:21:00 3 303783,33

I guess there is no overflow control over the sum over the cores.
How to reproduce the issue:
run mpstat with periodic reports and generate a interrupt critical workload.

[stevekay]

Does mpstat get the 'all' value directly from the kernel, rather than totalling up the number itself ?

https://github.com/sysstat/sysstat/blob/master/mpstat.c#L1996-L2002 ?

	/*
	 * Read total number of interrupts received among all CPU.
	 * (this is the first value on the line "intr:" in the /proc/stat file).
	 */
	if (DISPLAY_IRQ_SUM(actflags)) {
		read_stat_total_irq(st_irq[0]);
	}

https://github.com/sysstat/sysstat/blob/master/mpstat.c#L1795-L1827

/*
 ***************************************************************************
 * Read total number of interrupts from /proc/stat.
 *
 * IN:
 * @st_irq	Structure where total number of interrupts will be saved.
 *
 * OUT:
 * @st_irq	Structure with total number of interrupts.
 ***************************************************************************
 */
void read_stat_total_irq(struct stats_global_irq *st_irq)
{
	FILE *fp;
	char line[1024];
	unsigned long long irq_nr;

	if ((fp = fopen(STAT, "r")) == NULL)
		return;

	while (fgets(line, sizeof(line), fp) != NULL) {

		if (!strncmp(line, "intr ", 5)) {
			/* Read total number of interrupts received since system boot */
			sscanf(line + 5, "%llu", &irq_nr);
			st_irq->irq_nr = (unsigned int) irq_nr;

			break;
		}
	}

	fclose(fp);
}

I wasn't able to generate a sufficiently high workload on my system, but observing awk '/^intr/ { print $2 }' /proc/stat during your high workload might help to get a better understanding of this issue.