Prev: Last ioctl bkl pushdown in the LogFS tree, but no pull request?
Next: dmaengine questions
From: Joerg Roedel on 12 Aug 2010 10:50
On Thu, Aug 05, 2010 at 07:14:14AM -0700, H. Peter Anvin wrote:
> On 08/05/2010 12:45 AM, Roedel, Joerg wrote:
> Correct, again. It's unclear to me if we can get away with the very
> simple 64-bit approach -- in particular, not including all the 1:1
> mappings in the kernel -- for all future users, though.

Yeah, We could probably use the same symbol names for the trampoline and
the swapper page-table on 32 and 64 bit and merge the code that handles
it between the architectures. Shouldn't be too difficult. We cook
something up :-)
But do you mind to take this patch first? It fixes the occurence of an
erratum on AMD hardware, it is a quite simple fix compared to the rework
suggested. In fact, I would like to have a simple patch to fix the
problem first (and backport it as necessary) and then go the way to
rework the current code differences between 32 and 64 bit.

Joerg

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Joerg Roedel on 12 Aug 2010 13:10
On Thu, Aug 12, 2010 at 08:34:37AM -0700, H. Peter Anvin wrote:

> Agreed. I do have a concern about the kernel page tables not being
> synchronized into trampoline_pg_dir (it's only an issue for 32-bit
> !PAE), so at the very least we need to keep an eye out for it...

Great.

The synchronization problem might be real. The cpu_init function uses
per-cpu variables which might not be present in the early synced
tramponline page table (iirc). The change to swapper_pg_dir and the
global tlb flush should probably be moved to the very beginning of the
start_secondary function.

Joerg

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Joerg Roedel on 12 Aug 2010 15:10
On Thu, Aug 12, 2010 at 12:01:27PM -0700, H. Peter Anvin wrote:
> On 08/12/2010 10:06 AM, Joerg Roedel wrote:

> > The synchronization problem might be real. The cpu_init function uses
> > per-cpu variables which might not be present in the early synced
> > tramponline page table (iirc). The change to swapper_pg_dir and the
> > global tlb flush should probably be moved to the very beginning of the
> > start_secondary function.
> >
>
> Are you going to submit an updated patch, then?

Yes, I will send it next week when I am back in office and gave it some
testing.

Joerg

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: Last ioctl bkl pushdown in the LogFS tree, but no pull request?
Next: dmaengine questions