Re: 2.6.28 64-bit domU not booting

[Chris Lalancette <clalance@redhat.com>]

Jeremy Fitzhardinge wrote:
> Chris Lalancette wrote:
>> Just to dig up this old thread: we've been seeing the same thing.  I *think* it
>> comes down to the NX bit (or the lack thereof); on machines without NX, the
>> early pagetable setup is trying to use the NX bit in the PTE flags, and the
>> hypervisor is telling it to go away.  The Red Hat bugzilla entry that I've filed
>> has more analysis:
>>
>> https://bugzilla.redhat.com/show_bug.cgi?id=492523
>>
>> Valtteri, can you give the output of "cat /proc/cpuinfo | grep flags", so we can
>> confirm that it's probably the same issue?
>>
>> And it looks like Ian Campbell actually posted a patch upstream
>> (http://lkml.indiana.edu/hypermail/linux/kernel/0901.3/02668.html) that should
>> fix this, but I don't see it in any of the upstream trees.  Ian, Jeremy, do you
>> guys know what the status is there?  It would be good to get into 2.6.30, and
>> probably backport to the stable trees as well.
>>   
> 
> Hm, yes, it does seem to have got lost.  But I wonder if a simpler fix 
> isn't just test for NX early and set __supported_pte_mask accordingly.  
> Does this work?
> 
>     J
> 
> From: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> Date: Fri, 27 Mar 2009 11:29:02 -0700
> Subject: [PATCH] xen: set _PAGE_NX in __supported_pte_mask before pagetable construction
> 
> Many 32-bit and some 64-bit machines don't support the NX flag in ptes.
> Check for NX before constructing the kernel pagetables.
> 
> Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> 
> diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
> index 992be7f..f8c9e49 100644
> --- a/arch/x86/xen/enlighten.c
> +++ b/arch/x86/xen/enlighten.c
> @@ -42,6 +42,7 @@
>  #include <asm/xen/hypervisor.h>
>  #include <asm/fixmap.h>
>  #include <asm/processor.h>
> +#include <asm/proto.h>
>  #include <asm/msr-index.h>
>  #include <asm/setup.h>
>  #include <asm/desc.h>
> @@ -1003,6 +1004,13 @@ asmlinkage void __init xen_start_kernel(void)
>  
>  	__supported_pte_mask |= _PAGE_IOMAP;
>  
> +	/* Work out if we support NX */
> +#ifdef CONFIG_X86_64
> +	check_efer();
> +#else
> +	set_nx();
> +#endif
> +
>  	/* Don't do the full vcpu_info placement stuff until we have a
>  	   possible map and a non-dummy shared_info. */
>  	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];

Yep, much happier this way (although I only tested out x86_64, not i386).

Tested-by: Chris Lalancette <clalance@redhat.com>