Date: Tue, 17 Jun 2008 09:10:36 -0400
From: Mathieu Desnoyers <compudj@krystal.dyndns.org>
To: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Cc: Jens Axboe <jens.axboe@oracle.com>, Pekka Enberg <penberg@cs.helsinki.fi>,
       Tom Zanussi <tzanussi@gmail.com>, akpm@linux-foundation.org,
       linux-kernel@vger.kernel.org, righi.andrea@gmail.com
Subject: Re: [PATCH 2/3] relay: Fix race condition which occurs when
	reading across CPUs.
Message-ID: <20080617131036.GB8696@Krystal>
References: <20080613040958.4f52ee29@linux360.ro> <1213417601.8237.37.camel@charm-linux> <20080614181103.17617db1@linux360.ro> <20080616122249.GB18561@Krystal> <20080616162212.27a8c119@linux360.ro> <20080616164609.GM20851@kernel.dk> <84144f020806161118n70a876aeyb5ccac7b1e21d842@mail.gmail.com> <20080616182843.GS20851@kernel.dk> <20080617153934.59a7c7ee@linux360.ro> <20080617154923.36195161@linux360.ro>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
In-Reply-To: <20080617154923.36195161@linux360.ro>
User-Agent: Mutt/1.5.16 (2007-06-11)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3300
Lines: 124

* Eduard - Gabriel Munteanu (eduard.munteanu@linux360.ro) wrote:
> On Tue, 17 Jun 2008 15:39:34 +0300
> Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro> wrote:
> 
> > kmemtrace will use the affine versions and set CPU affinity anyway,
> > but it would be nice to have a consistent behavior from relay's part.
> > 
> > 
> > 	Cheers,
> > 	Eduard
> 
> BTW, I wrote this stuff and tested affinity behavior. It seems like
> migration always occurs fast enough, even if under pressure. My machine
> has two CPUs, so there are a few hardcoded values.
> 

Do you get the same results when not using a buffered read to read
/proc/PID/stat ? Also, since the internal proc handle might buffer the
information, closing the file and re-opening it after the set affinity
would not be a bad idea.

Mathieu

> #define _GNU_SOURCE
> 
> #include <stdio.h>
> #include <stdlib.h>
> #include <sys/types.h>
> #include <unistd.h>
> #include <string.h>
> #include <sched.h>
> 
> unsigned long get_current_cpu(FILE *stat_fp)
> {
> 	char buf[256], *p;
> 	unsigned long n_delims = 0, ret;
> 	size_t count;
> 	
> 	fseek(stat_fp, 0, SEEK_SET);
> 	count = fread(buf, 1, 255, stat_fp);
> 	buf[count] = 0;
> 
> 	p = buf;
> 	while (n_delims < 38)
> 		if (*p++ == ' ')
> 			n_delims++;
> 	sscanf(p, "%lu", &ret);
> 
> 	return ret;
> }
> 
> int main(void)
> {
> 	FILE *stat_fp;
> 	char stat_filename[255];
> 	int err;
> 	unsigned long old_cpu, new_cpu, cpu;
> 	pid_t pid;
> 	cpu_set_t cpuset;
> 
> 	pid = getpid();
> 	sprintf(stat_filename, "/proc/%lu/stat", pid);
> 	stat_fp = fopen(stat_filename, "r");
> 	if (!stat_fp) {
> 		printf("Couldn't open %lu", pid);
> 		return 1;
> 	}
> 
> 	old_cpu = get_current_cpu(stat_fp);
> 	new_cpu = (old_cpu + 1) % 2;
> 
> 	printf("Started on CPU %lu, PID %lu, setting affinity to CPU %lu...\n",
> 	       old_cpu, pid, new_cpu);
> 	
> 	err = sched_getaffinity(pid, sizeof(cpu_set_t), &cpuset);
> 	if (err == -1){
> 		printf("Could not retrieve the affinity!\n");
> 		return 1;
> 	}
> 	printf("Previous affinity: %d %d\n",
> 	       CPU_ISSET(0, &cpuset), CPU_ISSET(1, &cpuset));
> 	CPU_ZERO(&cpuset);
> 	CPU_SET(new_cpu, &cpuset);
> 	err = sched_setaffinity(pid, sizeof(cpu_set_t), &cpuset);
> 	if (err == -1) {
> 		printf("Could not set affinity!\n");
> 		return 1;
> 	}
> 	err = sched_getaffinity(pid, sizeof(cpu_set_t), &cpuset);
> 	if (err == -1){
> 		printf("Could not retrieve the affinity!\n");
> 		return 1;
> 	}
> 	printf("Current affinity: %d %d\n",
> 	       CPU_ISSET(0, &cpuset), CPU_ISSET(1, &cpuset));
> 
> 	cpu = get_current_cpu(stat_fp);
> 	if (cpu != new_cpu) {
> 		printf("Process migration did not occur immediately!\n"
> 		       "--- we were supposed to be on %lu, but we're on %lu\n",
> 		       new_cpu, cpu);
> 		/* return 1; */
> 	}
> 
> 	for (;;) {
> 		cpu = get_current_cpu(stat_fp);
> 		if (cpu != new_cpu) {
> 			printf("Oh, we arrived on a different CPU!\n");
> 			/* return 1; */
> 		}
> 	}
> 
> 	return 0;
> }
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/