buggy mlock() behaviour in 2.6.9 on ppc64?

[Chris Friesen]

Olof Johansson wrote:
> On Wed, Feb 02, 2005 at 12:11:22PM -0600, Chris Friesen wrote:
>>I've got a simple test app that tries to mmap() and mlock() an amount of 
>>memory specified on the commandline.

>>With the ppc64 kernel, however, the system hangs, and the fans speed 
>>starts increasing.  It seems like it doesn't realize that there is 
>>simply no way that it will ever be able to succeed.
>>
>>Is this expected behaviour for ppc64?

> linuxppc64-dev at ozlabs.org is the ppc64 mailing list, you might want to
> Cc: there instead for the ppc64 questions.

Ah, right.  I'm copying the list if anyone else wants to take a look.

> Anyhow, can you send me the testcase? Sounds like you're running on a
> G5, it's easier to debug on a pSeries where we have debugger console.
> Unless you have a stealth card in the machine?

Yes, I am on a G5 and I don't have a stealth card.

Here's the testcase.  I've got 2gig of physical memory.

"chewtest 1 1000000" gives an error that it can't allocate memory
"chewtest 1 500000" gets killed by the oom killer
"chewtest 1000 1000" just sits there



Chris



#include <stdlib.h>
#include <stdio.h>
#include <sys/mman.h>
#include <sys/ioctl.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/mman.h>

int main(int argc, char **argv)
{
	if (argc != 3) {
		printf("usage: %s <mappings> <pages per mapping>\n", *argv);
		return -1;
	}
	
	int mappings = atoi(argv[1]);
	int ppm = atoi(argv[2]);
	
	int mapsize = sysconf(_SC_PAGESIZE)*ppm;
	
	int i;
	int rc;
	void *buf;
	for (i=0;i<mappings;i++) {
		buf = (char *) mmap(0, mapsize, PROT_READ|PROT_WRITE,
			MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
		if (buf == MAP_FAILED) {
			perror("unable to get memory");
			return -1;
		}

		rc = mlock(buf, mapsize);
		if (rc<0) {
			perror("unable to lock memory");
			return -1;
		}
	}
		
	return 0;
}