Re: [PATCH 1/2] pidmap(2)

[Alexey Dobriyan]

On 9/6/17, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, 5 Sep 2017 22:05:00 +0300 Alexey Dobriyan <adobriyan@xxxxxxxxx>
> wrote:
>
>> Implement system call for bulk retrieveing of pids in binary form.
>>
>> Using /proc is slower than necessary: 3 syscalls + another 3 for each
>> thread +
>> converting with atoi().
>>
>> /proc may be not mounted especially in containers. Natural extension of
>> hidepid=2 efforts is to not mount /proc at all.
>>
>> It could be used by programs like ps, top or CRIU. Speed increase will
>> become more drastic once combined with bulk retrieval of process
>> statistics.
>
> The patches are performance optimizations, but their changelogs contain
> no performance measurements!
>
> Demonstration of some compelling real-world performance benefits would
> help things along a lot.

I forgot the sheet with numbers at work. :^)
They're very embarrassing for /proc.

pidmap:
N=1<<16 times
~130 processes (~250 task_structs) on a regular desktop system
opendir + readdir + closedir /proc + the same for every /proc/$PID/task
(roughly what htop(1) does) vs pidmap

/proc 16.80+-0.73%
pidmap 0.06+-0.31%

fdmap:
N=1<<22 times
4 opened descriptors (0, 1, 2, 3)
opendir+readdir+closedir /proc/self/fd (lsof(1)) vs fdmap

/proc 8.31+-0.37%
fdmap 0.32+-0.72%

Currently performance improvements may not be huge or even visible.
That's because programs like ps/top/lsof _have_ to use /proc to retrieve
other information. If combined with bulk taskstats-ish retrieval interfaces
they should run around /proc.
#include <sys/types.h>
#include <stdio.h>
#include <dirent.h>
#include <stdlib.h>
#include <unistd.h>

void f(void)
{
	DIR *d;
	struct dirent *de;

	d = opendir("/proc/");
	while ((de = readdir(d))) {
		if ('1' <= de->d_name[0] && de->d_name[0] <= '9') {
			int pid = atoi(de->d_name);
			char buf[32];
			DIR *dt;
			struct dirent *dte;

			snprintf(buf, sizeof(buf), "/proc/%d/task", pid);
			dt = opendir(buf);
			readdir(dt);
			readdir(dt);
			while ((dte = readdir(dt))) {
				int tid = atoi(dte->d_name);
				asm volatile ("" :: "g" (&tid) : "memory");
			}
			closedir(dt);
		}
	}
	closedir(d);
}

static inline long sys_pidmap(int *pid, unsigned int n, int start)
{
	register long r10 asm ("r10") = 0;
	long rv;
	asm volatile (
		"syscall"
		: "=a" (rv)
		: "0" (333), "D" (pid), "S" (n), "d" (start), "r" (r10)
		: "rcx", "r11", "cc", "memory"
	);
	return rv;
}

void g(void)
{
	int pid[1024];

	sys_pidmap(pid, sizeof(pid)/sizeof(pid[0]), 0);
}

int main(void)
{
	unsigned int i;

//	for (i = 0; i < (1<<16); i++)
//		f();

	for (i = 0; i < (1<<16); i++)
		g();

	return 0;
}
#include <sys/types.h>
#include <dirent.h>
#include <stdlib.h>
#include <unistd.h>

void f(void)
{
	DIR *d;
	struct dirent *de;

	d = opendir("/proc/self/fd");
	while ((de = readdir(d))) {
		int fd = atoi(de->d_name);
		asm volatile ("" :: "g" (&fd) : "memory");
	}
	closedir(d);
}

static inline long sys_fdmap(int pid, int *fd, unsigned int n, int start)
{
	register long r10 asm ("r10") = start;
	register long r8 asm ("r8") = 0;
	long rv;
	asm volatile (
		"syscall"
		: "=a" (rv)
		: "0" (334), "D" (pid), "S" (fd), "d" (n), "r" (r10), "r" (r8)
		: "rcx", "r11", "cc", "memory"
	);
	return rv;
}

void g(void)
{
	int fd[1024];

	sys_fdmap(0, fd, sizeof(fd)/sizeof(fd[0]), 0);
}

int main(void)
{
	unsigned int i;

//	for (i = 0; i < (1<<22); i++)
//		f();

	dup(0);
	for (i = 0; i < (1<<22); i++)
		g();

	return 0;
}