RE: Hugetlb demanding paging for -mm tree

["Seth, Rohit" <rohit.seth@intel.com>]

William Lee Irwin III <> wrote on Monday, August 09, 2004 11:59 AM:

> Chen, Kenneth W <mailto:kenneth.w.chen@intel.com> wrote on Monday,
>>> I suppose this is fixable in update_mmu_cache() where it can check
>>> the type of pte and do appropriate sizing and other things.  ia64
>>> would have to check the address instead of looking at the pte.
> 
> On Mon, Aug 09, 2004 at 11:43:32AM -0700, Seth, Rohit wrote:
>> Why do we need update_mmu_cache for hugepages?
> 
> As things stand in mainline, it's not an obvious issue. Ken appears to
> be calling it for hugetlb in the ZFOD fault handling patches, which
> have the issue that it may behave badly in several respects when
> acting on large pages. The cache coherency bits in update_mmu_fault()
> are necessary in general, but mainline omits them. It should only
> result in intermittent failures on machines with sufficiently
> incoherent caches. 
> 

Will the flush_dcache_page for hugepages even on incoherent caches be
not enough.  And that flush_dcache_page should be done in alloc_hugepage
after clearing the page(or change the clear_highpage to
clear_user_high_page).