Skip to contentSkip to conversation

bounce: call flush_dcache_page after bounce_copy_vec

PATCH In review Gary King 4 participants 4 messages Sep 7, 2010 – Sep 9, 2010 Memory management

lore
Share & tools
lore

Gary King

author
#0 lore raw

PATCH 1/1 bounce: call flush_dcache_page after bounce_copy_vec

I have been seeing problems on Tegra 2 (ARMv7 SMP) systems with HIGHMEM enabled on 2.6.35 (plus some patches targetted at 2.6.36 to perform cache maintenance lazily), and the root cause appears to be that the mm bouncing code is calling flush_dcache_page before it copies the bounce buffer into the bio.

The patch below reorders these two operations, and eliminates numerous arbitrary application crashes on my dev system.

Gary

--

maintainer
#1 lore raw
Quoted 31 lines · On Tue, 7 Sep 2010 13:45:34 -0700 Gary King <[email protected]> wrote:

I have been seeing problems on Tegra 2 (ARMv7 SMP) systems with HIGHMEM enabled on 2.6.35 (plus some patches targetted at 2.6.36 to perform cache maintenance lazily), and the root cause appears to be that the mm bouncing code is calling flush_dcache_page before it copies the bounce buffer into the bio.

The patch below reorders these two operations, and eliminates numerous arbitrary application crashes on my dev system.

Gary

--

Quoted 1 line

From 678c9bca8d8a8f254f28af91e69fad3aa1be7593 Mon Sep 17 00:00:00 2001

From: Gary King <[email protected]>
Date: Mon, 6 Sep 2010 15:37:12 -0700
Subject: bounce: call flush_dcache_page after bounce_copy_vec

the bounced page needs to be flushed after data is copied into it, to ensure that architecture implementations can synchronize instruction and data caches if necessary.

---
 mm/bounce.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
mm/bounce.c +0 -0
index 13b6dad..1481de6 100644

mm/bounce.c static void copy_to_high_bio_irq(struct bio *to, struct bio *from) · line 122

Quoted 8 lines
mm/bounce.c +1 -1
@@ -116,8 +116,8 @@ static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
 		 */
 		vfrom = page_address(fromvec->bv_page) + tovec->bv_offset;
 
-		flush_dcache_page(tovec->bv_page);
 		bounce_copy_vec(tovec, vfrom);
+		flush_dcache_page(tovec->bv_page);
 	}
 }

Oh my, that was bad.

I queued your fix for 2.6.36 and tagged it for -stable backporting, thanks.

#2 lore raw
Quoted 37 lines · Andrew Morton
Quoted 34 lines · On Tue, 7 Sep 2010 13:45:34 -0700 Gary King <[email protected]> wrote:

I have been seeing problems on Tegra 2 (ARMv7 SMP) systems with HIGHMEM enabled on 2.6.35 (plus some patches targetted at 2.6.36 to perform cache maintenance lazily), and the root cause appears to be that the mm bouncing code is calling flush_dcache_page before it copies the bounce buffer into the bio.

The patch below reorders these two operations, and eliminates numerous arbitrary application crashes on my dev system.

Gary

--

Quoted 1 line

From 678c9bca8d8a8f254f28af91e69fad3aa1be7593 Mon Sep 17 00:00:00 2001

From: Gary King <[email protected]>
Date: Mon, 6 Sep 2010 15:37:12 -0700
Subject: bounce: call flush_dcache_page after bounce_copy_vec

the bounced page needs to be flushed after data is copied into it, to ensure that architecture implementations can synchronize instruction and data caches if necessary.

---
 mm/bounce.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
mm/bounce.c +0 -0
index 13b6dad..1481de6 100644
@@ -116,8 +116,8 @@ static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
 		 */
 		vfrom = page_address(fromvec->bv_page) + tovec->bv_offset;
 
Quoted 7 lines

mm/bounce.c static void copy_to_high_bio_irq(struct bio *to, struct bio *from) · line 122

Quoted 5 lines
mm/bounce.c +1 -1
-		flush_dcache_page(tovec->bv_page);
 		bounce_copy_vec(tovec, vfrom);
+		flush_dcache_page(tovec->bv_page);
 	}
 }

Oh my, that was bad.

Indeed, I wonder how that could have been wrong for so long.

Quoted 2 lines

I queued your fix for 2.6.36 and tagged it for -stable backporting, thanks.

Thanks, you can add my acked-by.

Signature · 1 line
Jens Axboe
#3 lore raw
In reply to Gary King
Quoted 9 lines · Gary King

I have been seeing problems on Tegra 2 (ARMv7 SMP) systems with HIGHMEM enabled on 2.6.35 (plus some patches targetted at 2.6.36 to perform cache maintenance lazily), and the root cause appears to be that the mm bouncing code is calling flush_dcache_page before it copies the bounce buffer into the bio.

The patch below reorders these two operations, and eliminates numerous arbitrary application crashes on my dev system.

We also experience the package building failure on OMAP4 SMP system with HIGHMEM enabled
on 2.6.35. Thanks a lot for this fixing, we will try it later soon.

-Bryan

Quoted 40 lines

Gary

--
From 678c9bca8d8a8f254f28af91e69fad3aa1be7593 Mon Sep 17 00:00:00 2001
From: Gary King <[email protected]>
Date: Mon, 6 Sep 2010 15:37:12 -0700
Subject: bounce: call flush_dcache_page after bounce_copy_vec

the bounced page needs to be flushed after data is copied into it, to ensure that architecture implementations can synchronize instruction and data caches if necessary.

---
?mm/bounce.c | ? ?2 +-
?1 files changed, 1 insertions(+), 1 deletions(-)

mm/bounce.c +0 -0
index 13b6dad..1481de6 100644
@@ -116,8 +116,8 @@ static void copy_to_high_bio_irq(struct bio *to, struct bio *from)

? ? ? ? ? ? ? ? */
? ? ? ? ? ? ? ?vfrom = page_address(fromvec->bv_page) + tovec->bv_offset;

- ? ? ? ? ? ? ? flush_dcache_page(tovec->bv_page); ? ? ? ? ? ? ? ?bounce_copy_vec(tovec, vfrom);
+ ? ? ? ? ? ? ? flush_dcache_page(tovec->bv_page); ? ? ? ?}
?}

--
1.7.0.4

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel