On Wed, Jul 23, 2014 at 09:52:22AM -0400, Matthew Wilcox wrote:

On Wed, Jul 23, 2014 at 02:45:40PM +0300, Kirill A. Shutemov wrote:

quoted

On Tue, Jul 22, 2014 at 03:47:53PM -0400, Matthew Wilcox wrote:

quoted

From: Matthew Wilcox <redacted>

vm_insert_mixed() will fail if there is already a valid PTE at that
location.  The DAX code would rather replace the previous value with
the new PTE.

quoted

quoted

@@ -1492,8 +1492,12 @@ static int insert_page(struct vm_area_struct *vma, unsigned long addr,
 	if (!pte)
 		goto out;
 	retval = -EBUSY;
-	if (!pte_none(*pte))
-		goto out_unlock;
+	if (!pte_none(*pte)) {
+		if (!replace)
+			goto out_unlock;
+		VM_BUG_ON(!mutex_is_locked(&vma->vm_file->f_mapping->i_mmap_mutex));
+		zap_page_range_single(vma, addr, PAGE_SIZE, NULL);

zap_page_range_single() takes ptl by itself in zap_pte_range(). It's not
going to work.

I have a test program that exercises this path ... it seems to work!
Following the code, I don't understand why it does.  Maybe it's not
exercising this path after all?  I've attached the program (so that I
have an "oh, duh" moment about 5 seconds after sending the email).

See below.

quoted

And zap_page_range*() is pretty heavy weapon to shoot down one pte, which
we already have pointer to. Why?

I'd love to use a lighter-weight weapon!  What would you recommend using,
zap_pte_range()?

The most straight-forward way: extract body of pte cycle from
zap_pte_range() to separate function -- zap_pte() -- and use it.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>

int
main(int argc, char *argv[])
{
	int fd;
	void *addr;
	char buf[4096];

	if (argc != 2) {
		fprintf(stderr, "usage: %s filename\n", argv[0]);
		exit(1);
	}

	if ((fd = open(argv[1], O_CREAT|O_RDWR, 0666)) < 0) {
		perror(argv[1]);
		exit(1);
	}

	if (ftruncate(fd, 4096) < 0) {

Shouldn't this be ftruncate(fd, 0)? Otherwise the memcpy() below will
fault in page from backing storage, not hole and write will not replace
anything.

		perror("ftruncate");
		exit(1);
	}

	if ((addr = mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_SHARED,
					fd, 0)) == MAP_FAILED) {
		perror("mmap");
		exit(1);
	}

	close(fd);

	/* first read */
	memcpy(buf, addr, 4096);

	/* now write a bit */
	memcpy(addr, buf, 8);

	printf("%s: test passed.\n", argv[0]);
	exit(0);
}

-- 
 Kirill A. Shutemov

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>