Re: mmap syscall problem

[Arnd Bergmann <arnd@arndb.de>]

On Monday 06 July 2009, Michal Simek wrote:
> 
> > Not necessarily, even on platforms that manage aliases in hardware
> > mappings that violate the aliasing constraints can still result in
> > undefined behaviour, this really depends more on your cache controller
> > and MMU than anything else. I notice that microblaze sets SHMLBA to
> > PAGE_SIZE, you may want to see if this test still breaks after bumping it
> > up to something like PAGE_SIZE * 4.
> >   
> Yes, test still break - behavior is the same. I don't have accurate
> information about MMU unit
> but I will ask a question about. We are able to turn off cache
> controller directly in HW.

There may still be a problem with data being queued in some write
buffers that don't get flushed before reading back from another
address.

What happens in a simple user space program that mmaps the same
page to two addresses? Something like

#include <sys/mman.h>
#include <fcntl.h>
int main(void)
{
	int fd = open("existing-4k-file", O_RDWR);
	char *p1 = mmap(0, 4096, PROT_READ | PROT_WRITE,
				MAP_SHARED, fd, 0);
	char *p2 = mmap(p1 + 4096, 4096, PROT_READ | PROT_WRITE,
				MAP_SHARED | MAP_FIXED, fd, 0);

	*p1 = 0xaa; *p2 = 0x55;

	return *p1; /* returns 0xaa if broken, 0x55 if correct */
}

	Arnd <><