Problem with malloc(3)/sbrk(2)/brk(2)/something

Hi,

This is a follow-up to my footnote about not being able to run the 'config.guess' shell script with ash (and some other stuff) on our userland. Basically I've found a minimal testcase for the problem, which is that it's not possible to malloc() across a page boundary.

Beware that since we don't have a real page fault handler, this program will crash your machine. (Press TOC, and we'll take it from there :-)

If you turn on debugging assertions in the malloc code you'll see that it triggers this assertion:

/* We always land on a page boundary */
assert(((unsigned long)((char*)top(ar_ptr) + top_size) & (pagesz-1)) == 0);

I don't see sbrk() or brk() failing, so I'm not exactly sure why this happens. I suspect they are returning bogus values in this case. If someone knows that would be great although I expect to find the problem soon (I may try to get strace going, since it will be very useful for other similar problems).

#include <stdlib.h>
#include <stdio.h>

int main()

{
	/* get the first address */
	char *foo = malloc(16);

	/* now try to break a page boundary */
	char *target = (char *) ((unsigned long) (foo + 4095) & ~4095);

	size_t nbytes = target - foo;
	foo = malloc(nbytes);
	return 0;
}

[email protected], http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Oops, sorry, this did not seem to be the actual problem. It seems that I may have been running a kernel that was not %r8-clean, or something similar, because recompiling malloc.c with -ffixed-r8 made these test cases work :(

However config.guess still doesn't work :( Once I get a working kernel again I suppose I'll look at that.

[email protected], http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Okay, fixed. See parisc-linux-cvs and rebuild your kernels :-)

David Huggins-Daines - [email protected]