Re: [PATCH tip] x86: unsigned long pte_pfn

[Jeremy Fitzhardinge <jeremy@goop.org>]

Hugh Dickins wrote:
> I expected gcc to optimize away that difference, and often it does,
> but not always (I'm using 4.2.1 and CC_OPTIMIZE_FOR_SIZE here):
> pte_page() involved
>  228:	c1 e8 0c             	shr    $0xc,%eax
>  22b:	c1 e0 05             	shl    $0x5,%eax
> before the unification, but afterwards
>  228:	25 00 f0 ff ff       	and    $0xfffff000,%eax
>  22d:	c1 e8 07             	shr    $0x7,%eax
>
> So it's bloated that kernel by 0.001% (around 40 bytes).  Oh well,
> I think we may suppose that with a different version of gcc or
> different optimizations, it could just as well have gone the
> other way - I vote to go with your unification.
>   

Neither of those sequences make much sense to me in isolation, but I
guess it's setting up to index the struct page array.  But in general, I
think some CPUs are not very happy about shifting, so using the "and" is
more efficient anyway.

> [PATCH tip] x86: unsigned long pte_pfn
>
> pte_pfn() has always been of type unsigned long, even on 32-bit PAE;
> but in the current tip/next/mm tree it works out to be unsigned long
> long on 64-bit, which gives an irritating warning if you try to printk
> a pfn with the usual %lx.
>
> Now use the same pte_pfn() function, moved from pgtable-3level.h
> to pgtable.h, for all models: as suggested by Jeremy Fitzhardinge.
> And pte_page() can well move along with it (remaining a macro to
> avoid dependence on mm_types.h).
>
> Signed-off-by: Hugh Dickins <hugh@veritas.com>
>   
Acked-by: Jeremy Fitzhardinge <jeremy@goop.org>

> ---
>
>  include/asm-x86/pgtable-2level.h |    2 --
>  include/asm-x86/pgtable-3level.h |    7 -------
>  include/asm-x86/pgtable.h        |    7 +++++++
>  include/asm-x86/pgtable_64.h     |    2 --
>  4 files changed, 7 insertions(+), 11 deletions(-)
>
> --- 2.6.27-rc5-mm1/include/asm-x86/pgtable-2level.h	2008-09-05 10:05:51.000000000 +0100
> +++ linux/include/asm-x86/pgtable-2level.h	2008-09-09 13:53:34.000000000 +0100
> @@ -53,9 +53,7 @@ static inline pte_t native_ptep_get_and_
>  #define native_ptep_get_and_clear(xp) native_local_ptep_get_and_clear(xp)
>  #endif
>  
> -#define pte_page(x)		pfn_to_page(pte_pfn(x))
>  #define pte_none(x)		(!(x).pte_low)
> -#define pte_pfn(x)		(pte_val(x) >> PAGE_SHIFT)
>  
>  /*
>   * Bits 0, 6 and 7 are taken, split up the 29 bits of offset
> --- 2.6.27-rc5-mm1/include/asm-x86/pgtable-3level.h	2008-09-05 10:05:51.000000000 +0100
> +++ linux/include/asm-x86/pgtable-3level.h	2008-09-09 13:53:34.000000000 +0100
> @@ -151,18 +151,11 @@ static inline int pte_same(pte_t a, pte_
>  	return a.pte_low == b.pte_low && a.pte_high == b.pte_high;
>  }
>  
> -#define pte_page(x)	pfn_to_page(pte_pfn(x))
> -
>  static inline int pte_none(pte_t pte)
>  {
>  	return !pte.pte_low && !pte.pte_high;
>  }
>  
> -static inline unsigned long pte_pfn(pte_t pte)
> -{
> -	return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
> -}
> -
>  /*
>   * Bits 0, 6 and 7 are taken in the low part of the pte,
>   * put the 32 bits of offset into the high part.
> --- 2.6.27-rc5-mm1/include/asm-x86/pgtable.h	2008-09-05 10:05:51.000000000 +0100
> +++ linux/include/asm-x86/pgtable.h	2008-09-09 13:53:34.000000000 +0100
> @@ -186,6 +186,13 @@ static inline int pte_special(pte_t pte)
>  	return pte_val(pte) & _PAGE_SPECIAL;
>  }
>  
> +static inline unsigned long pte_pfn(pte_t pte)
> +{
> +	return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
> +}
> +
> +#define pte_page(pte)	pfn_to_page(pte_pfn(pte))
> +
>  static inline int pmd_large(pmd_t pte)
>  {
>  	return (pmd_val(pte) & (_PAGE_PSE | _PAGE_PRESENT)) ==
> --- 2.6.27-rc5-mm1/include/asm-x86/pgtable_64.h	2008-09-05 10:05:51.000000000 +0100
> +++ linux/include/asm-x86/pgtable_64.h	2008-09-09 13:53:34.000000000 +0100
> @@ -181,8 +181,6 @@ static inline int pmd_bad(pmd_t pmd)
>  #endif
>  
>  #define pages_to_mb(x)	((x) >> (20 - PAGE_SHIFT))   /* FIXME: is this right? */
> -#define pte_page(x)	pfn_to_page(pte_pfn((x)))
> -#define pte_pfn(x)	((pte_val((x)) & __PHYSICAL_MASK) >> PAGE_SHIFT)
>  
>  /*
>   * Macro to mark a page protection value as "uncacheable".
>