| Message ID | 20151108192739.9104.32105.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On Sun, Nov 08, 2015 at 02:27:39PM -0500, Dan Williams wrote: > As Dave points out when dealing with the contents of a page we use > PAGE_SIZE and PAGE_SHIFT, similary for huge pages use HPAGE_SIZE and > HPAGE_SHIFT. Hang on, no. We want to use PMD_SIZE here so that when we add PGD_SIZE support, it's all nicely symmetrical.
On 11/09/2015 11:52 AM, Matthew Wilcox wrote: > On Sun, Nov 08, 2015 at 02:27:39PM -0500, Dan Williams wrote: >> As Dave points out when dealing with the contents of a page we use >> PAGE_SIZE and PAGE_SHIFT, similary for huge pages use HPAGE_SIZE and >> HPAGE_SHIFT. > > Hang on, no. We want to use PMD_SIZE here so that when we add PGD_SIZE > support, it's all nicely symmetrical. This is splitting hairs a bit, but here goes. Do we care about the pagetables, or do we care about the page sizes? If we are doing pagetable things, we use P{TE,MD,UD,GD}_SIZE. If we are doing something related to page (contents) themselves, we use H?PAGE_SIZE. This makes it clear whether we're dealing with pagetables, or *contents*. In the end, they're the same thing on our architecture at the moment, so it matters very little.
On Mon, Nov 09, 2015 at 12:30:08PM -0800, Dave Hansen wrote: > On 11/09/2015 11:52 AM, Matthew Wilcox wrote: > > On Sun, Nov 08, 2015 at 02:27:39PM -0500, Dan Williams wrote: > >> As Dave points out when dealing with the contents of a page we use > >> PAGE_SIZE and PAGE_SHIFT, similary for huge pages use HPAGE_SIZE and > >> HPAGE_SHIFT. > > > > Hang on, no. We want to use PMD_SIZE here so that when we add PGD_SIZE > > support, it's all nicely symmetrical. > > This is splitting hairs a bit, but here goes. > > Do we care about the pagetables, or do we care about the page sizes? > > If we are doing pagetable things, we use P{TE,MD,UD,GD}_SIZE. If we are > doing something related to page (contents) themselves, we use > H?PAGE_SIZE. This makes it clear whether we're dealing with pagetables, > or *contents*. > > In the end, they're the same thing on our architecture at the moment, so > it matters very little. Agreed we're splitting hairs here but I'm not sure why the hair can be split :-) We come here from the page fault handler, where we've discovered that there is no page table for a PMD-shaped hole in the address space, and we'd like a PMD inserted by the fault handler. In that sense, we care about PMD_SIZE. Another problem is that the name 'HPAGE' doesn't obviously say "I cover a PMD-sized hole" -- hugetlbfs can use 1GB pages as well as 2MB pages. Then there's the problem that there *is* no equivalent to HPAGE for PUDs or PGDs -- if PMDs are "huge", then what are the names for pages pointed to by PUDs and PGDs? Why is it important to distinguish between "the size of something pointed to by a PMD" and "the size of the contents"? Actually, I'm not sure I even understand how to summarise the distinction.
diff --git a/fs/dax.c b/fs/dax.c index f8e543839e5c..149d6000d72a 100644 --- a/fs/dax.c +++ b/fs/dax.c @@ -511,7 +511,7 @@ EXPORT_SYMBOL_GPL(dax_fault); * The 'colour' (ie low bits) within a PMD of a page offset. This comes up * more often than one might expect in the below function. */ -#define PG_PMD_COLOUR ((PMD_SIZE >> PAGE_SHIFT) - 1) +#define PG_PMD_COLOUR ((HPAGE_SIZE >> PAGE_SHIFT) - 1) int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, pmd_t *pmd, unsigned int flags, get_block_t get_block, @@ -537,7 +537,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, /* If the PMD would extend outside the VMA */ if (pmd_addr < vma->vm_start) return VM_FAULT_FALLBACK; - if ((pmd_addr + PMD_SIZE) > vma->vm_end) + if ((pmd_addr + HPAGE_SIZE) > vma->vm_end) return VM_FAULT_FALLBACK; pgoff = linear_page_index(vma, pmd_addr); @@ -551,7 +551,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, memset(&bh, 0, sizeof(bh)); block = (sector_t)pgoff << (PAGE_SHIFT - blkbits); - bh.b_size = PMD_SIZE; + bh.b_size = HPAGE_SIZE; length = get_block(inode, block, &bh, write); if (length) return VM_FAULT_SIGBUS; @@ -562,7 +562,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, * just fall back to PTEs. Calling get_block 512 times in a loop * would be silly. */ - if (!buffer_size_valid(&bh) || bh.b_size < PMD_SIZE) + if (!buffer_size_valid(&bh) || bh.b_size < HPAGE_SIZE) goto fallback; /* @@ -571,7 +571,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, */ if (buffer_new(&bh)) { i_mmap_unlock_read(mapping); - unmap_mapping_range(mapping, pgoff << PAGE_SHIFT, PMD_SIZE, 0); + unmap_mapping_range(mapping, pgoff << PAGE_SHIFT, HPAGE_SIZE, 0); i_mmap_lock_read(mapping); } @@ -616,7 +616,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address, result = VM_FAULT_SIGBUS; goto out; } - if ((length < PMD_SIZE) || (pfn & PG_PMD_COLOUR)) + if ((length < HPAGE_SIZE) || (pfn & PG_PMD_COLOUR)) goto fallback; if (buffer_unwritten(&bh) || buffer_new(&bh)) {
As Dave points out when dealing with the contents of a page we use PAGE_SIZE and PAGE_SHIFT, similary for huge pages use HPAGE_SIZE and HPAGE_SHIFT. Reported-by: Dave Hansen <dave.hansen@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- fs/dax.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)