Message-Id: <200905142026.19862.rjw@sisk.pl>
Date:	Thu, 14 May 2009 20:26:18 +0200
From:	"Rafael J. Wysocki" <rjw@...k.pl>
To:	Andrew Morton <akpm@...ux-foundation.org>
Cc:	linux-pm@...ts.linux-foundation.org, fengguang.wu@...el.com,
	linux-kernel@...r.kernel.org, pavel@....cz, nigel@...onice.net,
	rientjes@...gle.com, linux-mm@...ck.org
Subject: Re: [PATCH 4/6] PM/Hibernate: Rework shrinking of memory

On Wednesday 13 May 2009, Rafael J. Wysocki wrote:
> On Wednesday 13 May 2009, Andrew Morton wrote:
> > On Wed, 13 May 2009 10:39:25 +0200
> > "Rafael J. Wysocki" <rjw@...k.pl> wrote:
> > 
> > > From: Rafael J. Wysocki <rjw@...k.pl>
> > > 
> > > Rework swsusp_shrink_memory() so that it calls shrink_all_memory()
> > > just once to make some room for the image and then allocates memory
> > > to apply more pressure to the memory management subsystem, if
> > > necessary.
> > > 
> > > Unfortunately, we don't seem to be able to drop shrink_all_memory()
> > > entirely just yet, because that would lead to huge performance
> > > regressions in some test cases.
> > > 
> > 
> > Isn't this a somewhat large problem?
> 
> Yes, it is.  The thing is 8 times slower (15 s vs 2 s) without the
> shrink_all_memory() in at least one test case.  100% reproducible.
> 
> > The main point (I thought) was to remove shrink_all_memory().  Instead,
> > we're retaining it and adding even more stuff?
> 
> The idea is that afterwards we can drop shrink_all_memory() once the
> performance problem has been resolved.  Also, we now allocate memory for the
> image using GFP_KERNEL instead of doing it with GFP_ATOMIC after freezing
> devices.  I'd think that's an improvement?
> 
> > > +/**
> > > + * compute_fraction - Compute approximate fraction x * (a/b)
> > > + * @x: Number to multiply.
> > > + * @numerator: Numerator of the fraction (a).
> > > + * @denominator: Denominator of the fraction (b).
> > >   *
> > > - *	Notice: all userland should be stopped before it is called, or
> > > - *	livelock is possible.
> > > + * Compute an approximate value of the expression x * (a/b), where a is less
> > > + * than b, all x, a, b are unsigned longs and x * a may be greater than the
> > > + * maximum unsigned long.
> > >   */
> > > +static unsigned long compute_fraction(
> > > +	unsigned long x, unsigned long numerator, unsigned long denominator)
> > 
> > I can't say I'm a great fan of the code layout here.
> > 
> > static unsigned long compute_fraction(unsigned long x, unsigned long numerator, unsigned long denominator)
> > 
> > or
> > 
> > static unsigned long compute_fraction(unsigned long x, unsigned long numerator,
> > 					unsigned long denominator)
> > 
> > would be more typical.
> 
> OK
>  
> > > +{
> > > +	unsigned long ratio = (numerator << FRACTION_SHIFT) / denominator;
> > >  
> > > -#define SHRINK_BITE	10000
> > > -static inline unsigned long __shrink_memory(long tmp)
> > > +	x *= ratio;
> > > +	return x >> FRACTION_SHIFT;
> > > +}
> > 
> > Strange function.  Would it not be simpler/clearer to do it with 64-bit
> > scalars, multiplication and do_div()?
> 
> Sure, I can do it this way too.  Is it fine to use u64 for this purpose?
> 
> > > +static unsigned long highmem_size(
> > > +	unsigned long size, unsigned long highmem, unsigned long count)
> > > +{
> > > +	return highmem > count / 2 ?
> > > +			compute_fraction(size, highmem, count) :
> > > +			size - compute_fraction(size, count - highmem, count);
> > > +}
> > 
> > This would be considerably easier to follow if we know what the three
> > arguments represent.  Amount of memory?  In what units?  `count' of
> > what?
> > 
> > The `count/2' thing there is quite mysterious.
> > 
> > <does some reverse-engineering>
> > 
> > OK, `count' is "the number of pageframes we can use".  (I don't think I
> > helped myself a lot there).  But what's up with that divde-by-two?
> > 
> > <considers poking at callers to work out what `size' is>
> > 
> > <gives up>
> > 
> > Is this code as clear as we can possibly make it??
> 
> Heh
> 
> OK, I'll do my best to clean it up.

Updated patch is appended.

---
From: Rafael J. Wysocki <rjw@...k.pl>
Subject: PM/Hibernate: Rework shrinking of memory (rev. 2)

Rework swsusp_shrink_memory() so that it calls shrink_all_memory()
just once to make some room for the image and then allocates memory
to apply more pressure to the memory management subsystem, if
necessary.

Unfortunately, we don't seem to be able to drop shrink_all_memory()
entirely just yet, because that would lead to huge performance
regressions in some test cases.

Signed-off-by: Rafael J. Wysocki <rjw@...k.pl>
---
 kernel/power/snapshot.c |  204 +++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 158 insertions(+), 46 deletions(-)

Index: linux-2.6/kernel/power/snapshot.c
===================================================================
--- linux-2.6.orig/kernel/power/snapshot.c
+++ linux-2.6/kernel/power/snapshot.c
@@ -1066,69 +1066,181 @@ void swsusp_free(void)
 	buffer = NULL;
 }
 
+/* Helper functions used for the shrinking of memory. */
+
+#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
+
 /**
- *	swsusp_shrink_memory -  Try to free as much memory as needed
+ * preallocate_image_pages - Allocate a number of pages for hibernation image
+ * @nr_pages: Number of page frames to allocate.
+ * @mask: GFP flags to use for the allocation.
  *
- *	... but do not OOM-kill anyone
- *
- *	Notice: all userland should be stopped before it is called, or
- *	livelock is possible.
+ * Return value: Number of page frames actually allocated
  */
+static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
+{
+	unsigned long nr_alloc = 0;
+
+	while (nr_pages > 0) {
+		if (!alloc_image_page(mask))
+			break;
+		nr_pages--;
+		nr_alloc++;
+	}
+
+	return nr_alloc;
+}
+
+static unsigned long preallocate_image_memory(unsigned long nr_pages)
+{
+	return preallocate_image_pages(nr_pages, GFP_IMAGE);
+}
 
-#define SHRINK_BITE	10000
-static inline unsigned long __shrink_memory(long tmp)
+#ifdef CONFIG_HIGHMEM
+static unsigned long preallocate_image_highmem(unsigned long nr_pages)
 {
-	if (tmp > SHRINK_BITE)
-		tmp = SHRINK_BITE;
-	return shrink_all_memory(tmp);
+	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
 }
 
+/**
+ *  __fraction - Compute (an approximation of) x * (multiplier / base)
+ */
+static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
+{
+	x *= multiplier;
+	do_div(x, base);
+	return (unsigned long)x;
+}
+
+static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
+						unsigned long highmem,
+						unsigned long total)
+{
+	unsigned long alloc = __fraction(nr_pages, highmem, total);
+
+	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
+}
+#else /* CONFIG_HIGHMEM */
+static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
+{
+	return 0;
+}
+
+static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
+						unsigned long highmem,
+						unsigned long total)
+{
+	return 0;
+}
+#endif /* CONFIG_HIGHMEM */
+
+/**
+ * swsusp_shrink_memory -  Make the kernel release as much memory as needed
+ *
+ * To create a hibernation image it is necessary to make a copy of every page
+ * frame in use.  We also need a number of page frames to be free during
+ * hibernation for allocations made while saving the image and for device
+ * drivers, in case they need to allocate memory from their hibernation
+ * callbacks (these two numbers are given by PAGES_FOR_IO and SPARE_PAGES,
+ * respectively, both of which are rough estimates).  To make this happen, we
+ * compute the total number of available page frames and allocate at least
+ *
+ * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 + 2 * SPARE_PAGES
+ *
+ * of them, which corresponds to the maximum size of a hibernation image.
+ *
+ * If image_size is set below the number following from the above formula,
+ * the preallocation of memory is continued until the total number of saveable
+ * pages in the system is below the requested image size or it is impossible to
+ * allocate more memory, whichever happens first.
+ */
 int swsusp_shrink_memory(void)
 {
-	long tmp;
 	struct zone *zone;
-	unsigned long pages = 0;
-	unsigned int i = 0;
-	char *p = "-\\|/";
+	unsigned long saveable, size, max_size, count, highmem, pages = 0;
+	unsigned long alloc, pages_highmem;
 	struct timeval start, stop;
+	int error = 0;
 
-	printk(KERN_INFO "PM: Shrinking memory...  ");
+	printk(KERN_INFO "PM: Shrinking memory... ");
 	do_gettimeofday(&start);
-	do {
-		long size, highmem_size;
 
-		highmem_size = count_highmem_pages();
-		size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES;
-		tmp = size;
-		size += highmem_size;
-		for_each_populated_zone(zone) {
-			tmp += snapshot_additional_pages(zone);
-			if (is_highmem(zone)) {
-				highmem_size -=
-					zone_page_state(zone, NR_FREE_PAGES);
-			} else {
-				tmp -= zone_page_state(zone, NR_FREE_PAGES);
-				tmp += zone->lowmem_reserve[ZONE_NORMAL];
-			}
-		}
+	/* Count the number of saveable data pages. */
+	highmem = count_highmem_pages();
+	saveable = count_data_pages();
+
+	/*
+	 * Compute the total number of page frames we can use (count) and the
+	 * number of pages needed for image metadata (size).
+	 */
+	count = saveable;
+	saveable += highmem;
+	size = 0;
+	for_each_populated_zone(zone) {
+		size += snapshot_additional_pages(zone);
+		if (is_highmem(zone))
+			highmem += zone_page_state(zone, NR_FREE_PAGES);
+		else
+			count += zone_page_state(zone, NR_FREE_PAGES);
+	}
+	count += highmem;
+	count -= totalreserve_pages;
+
+	/* Compute the maximum number of saveable pages to leave in memory. */
+	max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES;
+	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
+	if (size > max_size)
+		size = max_size;
+	/*
+	 * If the maximum is not less than the current number of saveable pages
+	 * in memory, we don't need to do anything more.
+	 */
+	if (size >= saveable)
+		goto out;
 
-		if (highmem_size < 0)
-			highmem_size = 0;
+	/*
+	 * Let the memory management subsystem know that we're going to need a
+	 * large number of page frames to allocate and make it free some memory.
+	 * NOTE: If this is not done, performance will be hurt badly in some
+	 * test cases.
+	 */
+	shrink_all_memory(saveable - size);
 
-		tmp += highmem_size;
-		if (tmp > 0) {
-			tmp = __shrink_memory(tmp);
-			if (!tmp)
-				return -ENOMEM;
-			pages += tmp;
-		} else if (size > image_size / PAGE_SIZE) {
-			tmp = __shrink_memory(size - (image_size / PAGE_SIZE));
-			pages += tmp;
-		}
-		printk("\b%c", p[i++%4]);
-	} while (tmp > 0);
+	/*
+	 * The number of saveable pages in memory was too high, so apply some
+	 * pressure to decrease it.  First, make room for the largest possible
+	 * image and fail if that doesn't work.  Next, try to decrease the size
+	 * of the image as much as indicated by image_size using allocations
+	 * from highmem and non-highmem zones separately.
+	 */
+	pages_highmem = preallocate_image_highmem(highmem / 2);
+	max_size += pages_highmem;
+	alloc = count - max_size;
+	pages = preallocate_image_memory(alloc);
+	if (pages < alloc) {
+		error = -ENOMEM;
+		goto free_out;
+	}
+	size = max_size - size;
+	alloc = size;
+	size = preallocate_highmem_fraction(size, highmem, count);
+	pages_highmem += size;
+	alloc -= size;
+	pages += preallocate_image_memory(alloc);
+	pages += pages_highmem;
+
+ free_out:
+	/* Release all of the preallocated page frames. */
+	swsusp_free();
+
+	if (error) {
+		printk(KERN_CONT "\n");
+		return error;
+	}
+
+ out:
 	do_gettimeofday(&stop);
-	printk("\bdone (%lu pages freed)\n", pages);
+	printk(KERN_CONT "done (preallocated %lu free pages)\n", pages);
 	swsusp_show_speed(&start, &stop, pages, "Freed");
 
 	return 0;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists