Paul Mackerras <paulus@xxxxxxxxx> writes:
> On Mon, Mar 04, 2013 at 04:32:24PM +0530, Aneesh Kumar K.V wrote:
>>
>> Now with table_size argument, the first arg is no more the shift value,
>> rather it is index into the array. Hence i changed the variable name. I
>> will split that patch to make it easy for review.
>
> OK, so you're saying that the simple relation between index and the
> size of the objects in PGT_CACHE(index) no longer holds. That worries
> me, because now, what guarantees that two callers won't use the same
> index value with two different sizes? And what guarantees that we
> won't have two callers using different index values but the same size
> (which wouldn't be a disaster but would be a waste of space)?
>
> I think it would be preferable to keep the relation between shift and
> the size of the objects and just arrange to use a different shift
> value for the pmd objects when you need to.
Most of the places we get the cache pointer by doing something like.
PGT_CACHE(PMD_INDEX_SIZE). What we need is that kmem_cache to return an
object twice the size of PMD_TABLE_SIZE. The relevant diff in the later
patch is below.
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+ /*
+ * we store the pgtable details in the second half of PMD
+ */
+ if (PGT_CACHE(PMD_INDEX_SIZE))
+ pr_err("PMD Page cache already initialized with different size\n");
+ __pgtable_cache_add(PMD_INDEX_SIZE, PMD_TABLE_SIZE * 2, pmd_ctor);
+#else
pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
+#endif
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>