Re: [GIT PULL] x86/uapi for 3.8

[Markus Trippelsdorf <markus@trippelsdorf.de>]

On 2012.12.15 at 11:58 -0800, Linus Torvalds wrote:
> On Sat, Dec 15, 2012 at 11:41 AM, H. Peter Anvin <hpa@zytor.com> wrote:
> >
> > Matt is on vacation, and I'm partly offline for the weekend, but that
> > definitely seems suspicious.  Do we have a memory map of the affected
> > machine(s)?
> 
> 
> but as mentioned, there's bound to be some particular kernel layout
> that triggers this, because I definitely ran a few kernels with that
> commit in it without problems (and clearly other people are too).
> Looking at my boot log, I had successful boots with both 6a57d104c8cb
> and c2714334b944, which contains that commit.
> 
> It might also be that it causes some massive corruption at boot time,
> but it then requires that that particular memory is actually used. So
> maybe it's not so much about the memory map except indirectly.
> 
> But that commit *does* look a lot more likely than the things I looked at.

The commit message says that only some broken implementations of EFI
firmware require the mapping for the physical I/O device addresses.

So I wonder if the following simple patch might be enough?
It fixes the issue for me at least.

diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
index e190f7b..402e4ca 100644
--- a/arch/x86/mm/ioremap.c
+++ b/arch/x86/mm/ioremap.c
@@ -50,7 +50,7 @@ int ioremap_change_attr(unsigned long vaddr, unsigned long size,
 	return err;
 }
 
-#ifdef CONFIG_X86_64
+#ifdef CONFIG_EFI
 static void ident_pte_range(unsigned long paddr, unsigned long vaddr,
 			    pmd_t *ppmd, pmd_t *vpmd, unsigned long end)
 {

-- 
Markus