Re: 2.6.11.10 failure on Hypersparc sparc32

Re: 2.6.11.10 failure on Hypersparc sparc32

[David S. Miller]

From: Bob Breuer <breuerr@mc.net>
Date: Sat, 04 Jun 2005 11:34:25 -0500

> Look for hypersparc_copy_1page in arch/sparc/mm/hypersparc.S where it 
> uses ASI_M_BCOPY.  I think the src and dest need to be 32 byte aligned 
> in this case.

We fixed a similar problem in the past, by putting the
"aligned to 8 bytes" check into __constant_memcpy().

Let's just remove the __copy_1page() optimization altogether,
it causes more trouble than it's worth.  Similarly for
memset().

In fact, most of this inline stuff can be removed entirely.
In particular, we should just use a plain memcpy() symbol
and let GCC output the inline builtin copy if it wants
to automatically.

Re: 2.6.11.10 failure on Hypersparc sparc32

[William Lee Irwin III]

On Sun, Jun 05, 2005 at 02:30:21PM -0700, David S. Miller wrote:
> We fixed a similar problem in the past, by putting the
> "aligned to 8 bytes" check into __constant_memcpy().
> Let's just remove the __copy_1page() optimization altogether,
> it causes more trouble than it's worth.  Similarly for
> memset().
> In fact, most of this inline stuff can be removed entirely.
> In particular, we should just use a plain memcpy() symbol
> and let GCC output the inline builtin copy if it wants
> to automatically.

I've got something along these lines cooked up; does anyone who can
actively reproduce this bug care to take it for a spin?


-- wli

Re: 2.6.11.10 failure on Hypersparc sparc32

[David S. Miller]

From: William Lee Irwin III <wli@holomorphy.com>
Date: Mon, 6 Jun 2005 14:11:02 -0700

> I've got something along these lines cooked up; does anyone who can
> actively reproduce this bug care to take it for a spin?

Just post it when you have a patch ready :)