[Patch] New zone ZONE_EASY_RECLAIM take 4. (change build_zonelists)[3/8]

[Patch] New zone ZONE_EASY_RECLAIM take 4. (change build_zonelists)[3/8]

[Yasunori Goto]

This is changing build_zonelists for new zone.

__GFP_xxxs are flag for requires of page allocation which zone
is prefered. But, it is used as an index number for zonelists[] too.
But after my patch, __GFP_xxx might be set at same time. So,
last set bit number of __GFP is recognized for zonelists' index
by this patch.

take3->take4:
  take 3's modification was still wrong.
  __GFP_EASY_RECLAIM is 0x04 on i386, so fls(__GFP_EASY_RECLAIM)
  is 3. zone 3 is ZONE_HIGHMEM, not ZONE_EASY_RECLAIM.
  So, I rearranged __GFP_XXX flags (see: define gfp_easy_relcaim patch)
  and fls() is used at highest_zone() again.

take2 -> take 3:
 This patch is modified take 3 to avoid panic on i386.
 __GFP_DMA32 is 0 for i386. So, ZONE_DMA32 is selected 
 if zone_bits is 0 which means Zone_normal. 
 Zone_DMA32 is not allocated on i386, so kernel paniced 
 by no normal memory.
 In this patch, even if zone_bits is 0 adn __GFP_DMA32 is 0,
 Zone_Normal is selected.


Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: zone_reclaim/mm/page_alloc.c
===================================================================
--- zone_reclaim.orig/mm/page_alloc.c	2005-12-19 20:18:29.000000000 +0900
+++ zone_reclaim/mm/page_alloc.c	2005-12-19 20:19:56.000000000 +0900
@@ -1585,14 +1585,11 @@ static int __init build_zonelists_node(p
 {
 	struct zone *zone;
 
-	BUG_ON(zone_type > ZONE_HIGHMEM);
+	BUG_ON(zone_type > ZONE_EASY_RECLAIM);
 
 	do {
 		zone = pgdat->node_zones + zone_type;
 		if (populated_zone(zone)) {
-#ifndef CONFIG_HIGHMEM
-			BUG_ON(zone_type > ZONE_NORMAL);
-#endif
 			zonelist->zones[nr_zones++] = zone;
 			check_highest_zone(zone_type);
 		}
@@ -1605,12 +1602,17 @@ static int __init build_zonelists_node(p
 static inline int highest_zone(int zone_bits)
 {
 	int res = ZONE_NORMAL;
-	if (zone_bits & (__force int)__GFP_HIGHMEM)
-		res = ZONE_HIGHMEM;
-	if (zone_bits & (__force int)__GFP_DMA32)
-		res = ZONE_DMA32;
-	if (zone_bits & (__force int)__GFP_DMA)
+
+	if (zone_bits == fls((__force int)__GFP_DMA))
 		res = ZONE_DMA;
+	if (zone_bits == fls((__force int)__GFP_DMA32) &&
+	    (__force int)__GFP_DMA32 == 0x02)
+		res = ZONE_DMA32;
+	if (zone_bits == fls((__force int)__GFP_HIGHMEM))
+		res = ZONE_HIGHMEM;
+	if (zone_bits == fls((__force int)__GFP_EASY_RECLAIM))
+		res = ZONE_EASY_RECLAIM;
+
 	return res;
 }
 
Index: zone_reclaim/include/linux/gfp.h
===================================================================
--- zone_reclaim.orig/include/linux/gfp.h	2005-12-19 20:19:37.000000000 +0900
+++ zone_reclaim/include/linux/gfp.h	2005-12-19 20:19:56.000000000 +0900
@@ -81,7 +81,7 @@ struct vm_area_struct;
 
 static inline int gfp_zone(gfp_t gfp)
 {
-	int zone = GFP_ZONEMASK & (__force int) gfp;
+	int zone = fls(GFP_ZONEMASK & (__force int) gfp);
 	BUG_ON(zone >= GFP_ZONETYPES);
 	return zone;
 }

-- 
Yasunori Goto 


--
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>

Re: [Patch] New zone ZONE_EASY_RECLAIM take 4. (change build_zonelists)[3/8]

[Joel Schopp]

> -	BUG_ON(zone_type > ZONE_HIGHMEM);
> +	BUG_ON(zone_type > ZONE_EASY_RECLAIM);

It might be nice to check ifndef CONFIG_HIGHMEM that the zone isn't 
particularly ZONE_HIGHMEM.

>  	int res = ZONE_NORMAL;
> -	if (zone_bits & (__force int)__GFP_HIGHMEM)
> -		res = ZONE_HIGHMEM;
> -	if (zone_bits & (__force int)__GFP_DMA32)
> -		res = ZONE_DMA32;
> -	if (zone_bits & (__force int)__GFP_DMA)
> +
> +	if (zone_bits == fls((__force int)__GFP_DMA))
>  		res = ZONE_DMA;
> +	if (zone_bits == fls((__force int)__GFP_DMA32) &&
> +	    (__force int)__GFP_DMA32 == 0x02)
> +		res = ZONE_DMA32;
> +	if (zone_bits == fls((__force int)__GFP_HIGHMEM))
> +		res = ZONE_HIGHMEM;
> +	if (zone_bits == fls((__force int)__GFP_EASY_RECLAIM))
> +		res = ZONE_EASY_RECLAIM;
> +
>  	return res;
>  }

It is incredibly silly to check a constant for a value.  When it is zero 
instead of 2 the first part of the statement will be false anyway.

Which reminds me.  Why are we using fls again?  I don't see why we 
aren't just (zone_bits & value) the types.  It seems much easier to 
understand that way.

>  
> Index: zone_reclaim/include/linux/gfp.h
> ===================================================================
> --- zone_reclaim.orig/include/linux/gfp.h	2005-12-19 20:19:37.000000000 +0900
> +++ zone_reclaim/include/linux/gfp.h	2005-12-19 20:19:56.000000000 +0900
> @@ -81,7 +81,7 @@ struct vm_area_struct;
>  
>  static inline int gfp_zone(gfp_t gfp)
>  {
> -	int zone = GFP_ZONEMASK & (__force int) gfp;
> +	int zone = fls(GFP_ZONEMASK & (__force int) gfp);
>  	BUG_ON(zone >= GFP_ZONETYPES);
>  	return zone;
>  }
> 


--
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>

Re: [Patch] New zone ZONE_EASY_RECLAIM take 4. (change build_zonelists)[3/8]

[Yasunori Goto]

> > -	BUG_ON(zone_type > ZONE_HIGHMEM);
> > +	BUG_ON(zone_type > ZONE_EASY_RECLAIM);
> 
> It might be nice to check ifndef CONFIG_HIGHMEM that the zone isn't 
> particularly ZONE_HIGHMEM.

Hmm. I was a bit lazy :-(.

> >  	int res = ZONE_NORMAL;
> > -	if (zone_bits & (__force int)__GFP_HIGHMEM)
> > -		res = ZONE_HIGHMEM;
> > -	if (zone_bits & (__force int)__GFP_DMA32)
> > -		res = ZONE_DMA32;
> > -	if (zone_bits & (__force int)__GFP_DMA)
> > +
> > +	if (zone_bits == fls((__force int)__GFP_DMA))
> >  		res = ZONE_DMA;
> > +	if (zone_bits == fls((__force int)__GFP_DMA32) &&
> > +	    (__force int)__GFP_DMA32 == 0x02)
> > +		res = ZONE_DMA32;
> > +	if (zone_bits == fls((__force int)__GFP_HIGHMEM))
> > +		res = ZONE_HIGHMEM;
> > +	if (zone_bits == fls((__force int)__GFP_EASY_RECLAIM))
> > +		res = ZONE_EASY_RECLAIM;
> > +
> >  	return res;
> >  }
> 
> It is incredibly silly to check a constant for a value.  When it is zero 
> instead of 2 the first part of the statement will be false anyway.
> Which reminds me.  Why are we using fls again?  I don't see why we 
> aren't just (zone_bits & value) the types.  It seems much easier to 
> understand that way.

Ahhh. I might be still confused around it. :-(
Its cause came from the value of __GFP_DMA32 which might be 0, 1, or 2.
This was that I tried to solve making zonelists for all of them.
But, something looks redundant.....

Thanks.


> 
> >  
> > Index: zone_reclaim/include/linux/gfp.h
> > ===================================================================
> > --- zone_reclaim.orig/include/linux/gfp.h	2005-12-19 20:19:37.000000000 +0900
> > +++ zone_reclaim/include/linux/gfp.h	2005-12-19 20:19:56.000000000 +0900
> > @@ -81,7 +81,7 @@ struct vm_area_struct;
> >  
> >  static inline int gfp_zone(gfp_t gfp)
> >  {
> > -	int zone = GFP_ZONEMASK & (__force int) gfp;
> > +	int zone = fls(GFP_ZONEMASK & (__force int) gfp);
> >  	BUG_ON(zone >= GFP_ZONETYPES);
> >  	return zone;
> >  }
> > 
> 
> 

-- 
Yasunori Goto 


--
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>