[PATCH] Avoid !__GFP_IO allocations to eat from memory reservations

[PATCH] Avoid !__GFP_IO allocations to eat from memory reservations

[Marcelo Tosatti]

  
Message-ID: <Pine.LNX.4.21.0106140949550.8439-100000@freak.distro.conectiva>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


Linus, 

In pre3, GFP_BUFFER allocations can eat from the "emergency" memory
reservations in case try_to_free_pages() fails for those allocations in
__alloc_pages(). 


Here goes the (tested) patch to fix that: 


--- linux/mm/page_alloc.c.orig	Thu Jun 14 11:00:14 2001
+++ linux/mm/page_alloc.c	Thu Jun 14 11:32:56 2001
@@ -453,6 +453,12 @@
 				int progress = try_to_free_pages(gfp_mask);
 				if (progress || gfp_mask & __GFP_IO)
 					goto try_again;
+				/*
+				 * Fail in case no progress was made and the
+				 * allocation may not be able to block on IO.
+				 */
+				else
+					return NULL;
 			}
 		}
 	}

Re: [PATCH] Avoid !__GFP_IO allocations to eat from memory reservations

[Daniel Phillips]

On Thursday 14 June 2001 14:59, Marcelo Tosatti wrote:
> --- linux/mm/page_alloc.c.orig	Thu Jun 14 11:00:14 2001
> +++ linux/mm/page_alloc.c	Thu Jun 14 11:32:56 2001
> @@ -453,6 +453,12 @@
>  				int progress = try_to_free_pages(gfp_mask);
>  				if (progress || gfp_mask & __GFP_IO)
>  					goto try_again;
> +				/*
> +				 * Fail in case no progress was made and the
> +				 * allocation may not be able to block on IO.
> +				 */
> +				else
> +					return NULL;
>  			}
>  		}
>  	}

Nitpick dept: the 'else' is redundant.

--
Daniel

[PATCH] Avoid !__GFP_IO allocations to eat from memory reservations

[Marcelo Tosatti]

Linus,

I just read pre3<->pre4 diff and it seems you missed this patch... here it
goes again.

In pre3/4, GFP_BUFFER allocations can eat from the "emergency" memory
reservations in case try_to_free_pages() fails for those allocations in
__alloc_pages().


Here goes the (tested) patch to fix that:


--- linux/mm/page_alloc.c.orig	Thu Jun 14 11:00:14 2001
+++ linux/mm/page_alloc.c	Thu Jun 14 11:32:56 2001
@@ -453,6 +453,12 @@
 				int progress = try_to_free_pages(gfp_mask);
 				if (progress || gfp_mask & __GFP_IO)
 					goto try_again;
+				/*
+				 * Fail in case no progress was made and the
+				 * allocation may not be able to block on IO.
+				 */
+				else
+					return NULL;
 			}
 		}
 	}

Re: [PATCH] Avoid !__GFP_IO allocations to eat from memory reservations

[Chris Mason]

On Thursday, June 14, 2001 09:59:43 AM -0300 Marcelo Tosatti
<marcelo@conectiva.com.br> wrote:

> 
> In pre3, GFP_BUFFER allocations can eat from the "emergency" memory
> reservations in case try_to_free_pages() fails for those allocations in
> __alloc_pages(). 
> 
> 
> Here goes the (tested) patch to fix that: 

I started testing this because I expected problems under load with reiserfs
on it.  No deadlocks yet though...I owe Marcelo a beer ;-)

-chris