Re: [RFC] page fault retry with NOPAGE_RETRY

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

> Perhaps we should concentrate on that for now.  Did we have a patch to look
> at?

Only a hand written proto-patch. Below is a real (but untested) one.
Note that there might be still issues when called from get_user_pages()
which of course won't go back to userland. For the two usage scenario I
have in mind, it should be ok though. One (a signal pending) will loop
back in until the resource is available, the other (no_page() inserts
the PTE itself) is just fine. For the former case, I've added a
cond_resched() to the loop, we might want to look into adding the info
of wether we are coming from get_user_pages() vs. do_page_fault() to
these new arguments you want to add to page fault handlers. That would
allow in our case to do a non-interruptibe sleep when caused by
get_user_pages().

---

Add a way for a no_page() handler to request a retry of the faulting
instruction. It goes back to userland on page faults and just tries
again in get_user_pages(). I added a cond_resched() in the loop in that
later case.

Signed-off-by: Benjamin Herrenchmidt <benh@kernel.crashing.org>

Index: linux-work/include/linux/mm.h
===================================================================
--- linux-work.orig/include/linux/mm.h	2006-08-30 08:51:21.000000000 +1000
+++ linux-work/include/linux/mm.h	2006-09-24 08:25:33.000000000 +1000
@@ -623,6 +623,7 @@
  */
 #define NOPAGE_SIGBUS	(NULL)
 #define NOPAGE_OOM	((struct page *) (-1))
+#define NOPAGE_RETRY	((struct page *) (-2))
 
 /*
  * Different kinds of faults, as returned by handle_mm_fault().
Index: linux-work/mm/memory.c
===================================================================
--- linux-work.orig/mm/memory.c	2006-08-17 16:16:06.000000000 +1000
+++ linux-work/mm/memory.c	2006-09-24 08:34:09.000000000 +1000
@@ -1081,6 +1081,7 @@
 				default:
 					BUG();
 				}
+				cond_resched();
 			}
 			if (pages) {
 				pages[i] = page;
@@ -2117,11 +2118,13 @@
 	 * after the next truncate_count read.
 	 */
 
-	/* no page was available -- either SIGBUS or OOM */
-	if (new_page == NOPAGE_SIGBUS)
+	/* no page was available -- either SIGBUS, OOM or RETRY */
+	if (unlikely(new_page == NOPAGE_SIGBUS))
 		return VM_FAULT_SIGBUS;
-	if (new_page == NOPAGE_OOM)
+	else if (unlikely(new_page == NOPAGE_OOM))
 		return VM_FAULT_OOM;
+	else if (unlikely(new_page == NOPAGE_RETRY))
+		return VM_FAULT_MINOR;
 
 	/*
 	 * Should we do an early C-O-W break?


--
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>