[PATCH] xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"

[PATCH] xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"

[Stefano Stabellini]

We only need to set max_pfn_mapped to the last pfn mapped on x86_64 to
make sure that cleanup_highmap doesn't remove important mappings at
_end.

We don't need to do this on x86_32 because cleanup_highmap is not called
on x86_32. Besides lowering max_pfn_mapped on x86_32 has the unwanted
side effect of limiting the amount of memory available for the 1:1
kernel pagetable allocation.

This patch reverts the x86_32 part of the original patch.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>

diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
index dc708dc..afe1d54 100644
--- a/arch/x86/xen/mmu.c
+++ b/arch/x86/xen/mmu.c
@@ -1599,6 +1599,11 @@ static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
 		for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
 			pte_t pte;
 
+#ifdef CONFIG_X86_32
+			if (pfn > max_pfn_mapped)
+				max_pfn_mapped = pfn;
+#endif
+
 			if (!pte_none(pte_page[pteidx]))
 				continue;
 
@@ -1766,7 +1771,9 @@ pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
 	initial_kernel_pmd =
 		extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
 
-	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
+	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
+				  xen_start_info->nr_pt_frames * PAGE_SIZE +
+				  512*1024);
 
 	kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
 	memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);

Re: [PATCH] xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"

[Konrad Rzeszutek Wilk]

On Tue, Jun 07, 2011 at 03:35:05PM +0100, Stefano Stabellini wrote:
> We only need to set max_pfn_mapped to the last pfn mapped on x86_64 to
> make sure that cleanup_highmap doesn't remove important mappings at
> _end.
> 
> We don't need to do this on x86_32 because cleanup_highmap is not called
> on x86_32. Besides lowering max_pfn_mapped on x86_32 has the unwanted
> side effect of limiting the amount of memory available for the 1:1
> kernel pagetable allocation.
> 
> This patch reverts the x86_32 part of the original patch.
> 
> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> 
> diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
> index dc708dc..afe1d54 100644
> --- a/arch/x86/xen/mmu.c
> +++ b/arch/x86/xen/mmu.c
> @@ -1599,6 +1599,11 @@ static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
>  		for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
>  			pte_t pte;
>  
> +#ifdef CONFIG_X86_32
> +			if (pfn > max_pfn_mapped)
> +				max_pfn_mapped = pfn;
> +#endif
> +
>  			if (!pte_none(pte_page[pteidx]))
>  				continue;
>  
> @@ -1766,7 +1771,9 @@ pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
>  	initial_kernel_pmd =
>  		extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
>  
> -	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
> +	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
> +				  xen_start_info->nr_pt_frames * PAGE_SIZE +
> +				  512*1024);

The x86_64 has about the same max_pfn_mapped value, but it does not have
that extra 512kbytes. What is that for? Perhaps we should provide a comment
to explain what that extra memory mapped is for?
>  
>  	kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
>  	memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);

Re: [PATCH] xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"

[Stefano Stabellini]

On Tue, 7 Jun 2011, Konrad Rzeszutek Wilk wrote:
> On Tue, Jun 07, 2011 at 03:35:05PM +0100, Stefano Stabellini wrote:
> > We only need to set max_pfn_mapped to the last pfn mapped on x86_64 to
> > make sure that cleanup_highmap doesn't remove important mappings at
> > _end.
> > 
> > We don't need to do this on x86_32 because cleanup_highmap is not called
> > on x86_32. Besides lowering max_pfn_mapped on x86_32 has the unwanted
> > side effect of limiting the amount of memory available for the 1:1
> > kernel pagetable allocation.
> > 
> > This patch reverts the x86_32 part of the original patch.
> > 
> > Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > 
> > diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
> > index dc708dc..afe1d54 100644
> > --- a/arch/x86/xen/mmu.c
> > +++ b/arch/x86/xen/mmu.c
> > @@ -1599,6 +1599,11 @@ static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
> >  		for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
> >  			pte_t pte;
> >  
> > +#ifdef CONFIG_X86_32
> > +			if (pfn > max_pfn_mapped)
> > +				max_pfn_mapped = pfn;
> > +#endif
> > +
> >  			if (!pte_none(pte_page[pteidx]))
> >  				continue;
> >  
> > @@ -1766,7 +1771,9 @@ pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
> >  	initial_kernel_pmd =
> >  		extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
> >  
> > -	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
> > +	max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
> > +				  xen_start_info->nr_pt_frames * PAGE_SIZE +
> > +				  512*1024);
> 
> The x86_64 has about the same max_pfn_mapped value, but it does not have
> that extra 512kbytes. What is that for? Perhaps we should provide a comment
> to explain what that extra memory mapped is for?

The main difference between x86_64 and x86_32 is not the initial value
of max_pfn_mapped here, but the fact that max_pfn_mapped is updated in
xen_map_identity_early. In fact I get about 4000 pages of difference
with/without this patch.