Xen devel - Feb 2007 - [PATCH] Use lib/swiotlb code for x86_64 SWIOTLB

As part of my efforts to enable GART/IOMMU on x86_64,
I switch from using arch/i386/swiotlb-xen.c for the
SWIOTLB implementation to lib/swiotlb-xen.c for x86_64
processors.

After this patch is accepted, I''ll submit the patch to
enable the GART/IOMMU.

Signed-off-by: Mark Langsdorf <mark.langsdorf@amd.com>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 15/2/07 22:59, "Langsdorf, Mark" <mark.langsdorf@amd.com>
wrote:
> As part of my efforts to enable GART/IOMMU on x86_64,
> I switch from using arch/i386/swiotlb-xen.c for the
> SWIOTLB implementation to lib/swiotlb-xen.c for x86_64
> processors.
> 
> After this patch is accepted, I''ll submit the patch to
> enable the GART/IOMMU.
> 
> Signed-off-by: Mark Langsdorf <mark.langsdorf@amd.com>
Jan has been through the swiotlb code recently so may have an opinion on
this patch. Using the common swiotlb code (as far as possible) for x86/64
may make sense -- many of the nasty differences in our own swiotlb
implementation are to deal with CONFIG_HIGHMEM which does not exist on
x86/64.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>>> Keir Fraser <Keir.Fraser@cl.cam.ac.uk> 16.02.07 01:00
>>>
>On 15/2/07 22:59, "Langsdorf, Mark" <mark.langsdorf@amd.com>
wrote:
>
>> As part of my efforts to enable GART/IOMMU on x86_64,
>> I switch from using arch/i386/swiotlb-xen.c for the
>> SWIOTLB implementation to lib/swiotlb-xen.c for x86_64
>> processors.
>> 
>> After this patch is accepted, I''ll submit the patch to
>> enable the GART/IOMMU.
>> 
>> Signed-off-by: Mark Langsdorf <mark.langsdorf@amd.com>
>
>Jan has been through the swiotlb code recently so may have an opinion on
>this patch. Using the common swiotlb code (as far as possible) for x86/64
>may make sense -- many of the nasty differences in our own swiotlb
>implementation are to deal with CONFIG_HIGHMEM which does not exist on
>x86/64.
I can''t see how this can work: First, the way the change is done i386
will also
use that file (and then arch/i386/kernel/swiotlb.c should also be deleted by
the patch). Second, there''s nothing Xen specific anymore in that file,
not
even the contiguous region creation (the sole difference to plain 2.6.20 is
the use of virt_to_bus/bus_to_virt). I think the first patch should be really
just a move of arch/i386/kernel/swiotlb.c to lib/swiotlb-xen.c, nothing else.

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> I can''t see how this can work: First, the way the change is 
> done i386 will also use that file (and then
> arch/i386/kernel/swiotlb.c should also be deleted by
> the patch).
That wasn''t intentional.  i386 can still use the 
arch/i386 code.
> Second, there''s nothing Xen specific anymore in that file, not
> even the contiguous region creation (the sole difference to 
> plain 2.6.20 is the use of virt_to_bus/bus_to_virt). I think
> the first patch should be really just a move of
> arch/i386/kernel/swiotlb.c to lib/swiotlb-xen.c, nothing else.
This should be an improved version of lib/swiotlb.c as 
lib/swiotlb-xen.c, with correct code to make the mappings
contiguous.

-Mark Langsdorf
AMD, Inc.


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>>> "Langsdorf, Mark" <mark.langsdorf@amd.com> 22.02.07
01:09 >>>
>> I can''t see how this can work: First, the way the change is 
>> done i386 will also use that file (and then
>> arch/i386/kernel/swiotlb.c should also be deleted by
>> the patch).
>
>That wasn''t intentional.  i386 can still use the 
>arch/i386 code.
I still don''t understand this - both x86 architectures should be able
to
use the same file.
>> Second, there''s nothing Xen specific anymore in that file, not
>> even the contiguous region creation (the sole difference to 
>> plain 2.6.20 is the use of virt_to_bus/bus_to_virt). I think
>> the first patch should be really just a move of
>> arch/i386/kernel/swiotlb.c to lib/swiotlb-xen.c, nothing else.
>
>This should be an improved version of lib/swiotlb.c as 
>lib/swiotlb-xen.c, with correct code to make the mappings
>contiguous.
I still don''t understand why this differs more than necessary from
arch/i386/swiotlb.c, and I continue to find it more desirable to move
arch/i386/swiotlb.c to lib/swiotlb-xen.c first (without any changes)
and then have a patch that changes what you really think needs to
be changed for your purpose.
I''m specifically asking this because I posted changes to lib/swiotlb.c
to lkml so that (if all of them get pushed into the kernel.org tree)
Xen wouldn''t even need its own version anymore, but would just
need a header providing certain hooks.

Jan

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel