Re: SL*B: drop kmem cache argument from constructor

[Jon Tollefson <kniht@linux.vnet.ibm.com>]

Andrew Morton wrote:
> On Thu, 10 Jul 2008 11:44:16 -0500 Jon Tollefson <kniht@linux.vnet.ibm.com> wrote:
>
>   
>> Alexey Dobriyan wrote:
>>     
>>> Kmem cache passed to constructor is only needed for constructors that are
>>> themselves multiplexeres. Nobody uses this "feature", nor does anybody uses
>>> passed kmem cache in non-trivial way, so pass only pointer to object.
>>>
>>> Non-trivial places are:
>>> 	arch/powerpc/mm/init_64.c
>>> 	arch/powerpc/mm/hugetlbpage.c
>>>   
>>>       
>> ...<snip>...
>>     
>>> --- a/arch/powerpc/mm/hugetlbpage.c
>>> +++ b/arch/powerpc/mm/hugetlbpage.c
>>> @@ -595,9 +595,9 @@ static int __init hugepage_setup_sz(char *str)
>>>  }
>>>  __setup("hugepagesz=", hugepage_setup_sz);
>>>
>>> -static void zero_ctor(struct kmem_cache *cache, void *addr)
>>> +static void zero_ctor(void *addr)
>>>  {
>>> -	memset(addr, 0, kmem_cache_size(cache));
>>> +	memset(addr, 0, HUGEPTE_TABLE_SIZE);
>>>   
>>>       
>> This isn't going to work with the multiple huge page size support.  The
>> HUGEPTE_TABLE_SIZE macro now takes a parameter with of the mmu psize
>> index to indicate the size of page.
>>
>>     
>
> hrm.  I suppose we could hold our noses and use ksize(), assuming that
> we're ready to use ksize() at this stage in the object's lifetime.
>
> Better would be to just use kmem_cache_zalloc()?
>
> --- a/arch/powerpc/mm/hugetlbpage.c~slb-drop-kmem-cache-argument-from-constructor-fix
> +++ a/arch/powerpc/mm/hugetlbpage.c
> @@ -113,7 +113,7 @@ static inline pte_t *hugepte_offset(huge
>  static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
>  			   unsigned long address, unsigned int psize)
>  {
> -	pte_t *new = kmem_cache_alloc(huge_pgtable_cache(psize),
> +	pte_t *new = kmem_cache_zalloc(huge_pgtable_cache(psize),
>  				      GFP_KERNEL|__GFP_REPEAT);
>
>  	if (! new)
> @@ -730,11 +730,6 @@ static int __init hugepage_setup_sz(char
>  }
>  __setup("hugepagesz=", hugepage_setup_sz);
>
> -static void zero_ctor(void *addr)
> -{
> -	memset(addr, 0, HUGEPTE_TABLE_SIZE);
> -}
> -
>  static int __init hugetlbpage_init(void)
>  {
>  	unsigned int psize;
> @@ -756,7 +751,7 @@ static int __init hugetlbpage_init(void)
>  						HUGEPTE_TABLE_SIZE(psize),
>  						HUGEPTE_TABLE_SIZE(psize),
>  						0,
> -						zero_ctor);
> +						NULL);
>  			if (!huge_pgtable_cache(psize))
>  				panic("hugetlbpage_init(): could not create %s"\
>  				      "\n", HUGEPTE_CACHE_NAME(psize));
> _
>
>
> btw, Nick, what's with that dopey
>
> 	huge_pgtable_cache(psize) = kmem_cache_create(...
>
> trick?  The result of a function call is not an lvalue, and writing a
> macro which pretends to be a function and then using it in some manner
> in which a function cannot be used is seven ways silly :(
>   
That silliness came from me.
It came from my simplistic translation of the existing code to handle
multiple huge page sizes.  I would agree it would be easier to read and
more straight forward to just have the indexed array directly on the
left side instead of a macro.  I can send out a patch that makes that
change if desired.
Something such as

+#define HUGE_PGTABLE_INDEX(psize) (HUGEPTE_CACHE_NUM + psize - 1)

-huge_pgtable_cache(psize) = kmem_cache_create(...
+pgtable_cache[HUGE_PGTABLE_INDEX(psize)] = kmem_cache_create(...


or if there is a more accepted way of handling this situation I can
amend it differently.

Jon

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>