aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--include/linux/mm.h1
-rw-r--r--mm/memory.c79
2 files changed, 59 insertions, 21 deletions
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 526f810367d9..c657ea0bd6aa 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -107,6 +107,7 @@ extern unsigned int kobjsize(const void *objp);
107#define VM_ALWAYSDUMP 0x04000000 /* Always include in core dumps */ 107#define VM_ALWAYSDUMP 0x04000000 /* Always include in core dumps */
108 108
109#define VM_CAN_NONLINEAR 0x08000000 /* Has ->fault & does nonlinear pages */ 109#define VM_CAN_NONLINEAR 0x08000000 /* Has ->fault & does nonlinear pages */
110#define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */
110 111
111#ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */ 112#ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */
112#define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS 113#define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS
diff --git a/mm/memory.c b/mm/memory.c
index 46958fb97c2d..0da414c383e7 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -371,35 +371,65 @@ static inline int is_cow_mapping(unsigned int flags)
371} 371}
372 372
373/* 373/*
374 * This function gets the "struct page" associated with a pte. 374 * This function gets the "struct page" associated with a pte or returns
375 * NULL if no "struct page" is associated with the pte.
375 * 376 *
376 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping 377 * A raw VM_PFNMAP mapping (ie. one that is not COWed) may not have any "struct
377 * will have each page table entry just pointing to a raw page frame 378 * page" backing, and even if they do, they are not refcounted. COWed pages of
378 * number, and as far as the VM layer is concerned, those do not have 379 * a VM_PFNMAP do always have a struct page, and they are normally refcounted
379 * pages associated with them - even if the PFN might point to memory 380 * (they are _normal_ pages).
381 *
382 * So a raw PFNMAP mapping will have each page table entry just pointing
383 * to a page frame number, and as far as the VM layer is concerned, those do
384 * not have pages associated with them - even if the PFN might point to memory
380 * that otherwise is perfectly fine and has a "struct page". 385 * that otherwise is perfectly fine and has a "struct page".
381 * 386 *
382 * The way we recognize those mappings is through the rules set up 387 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
383 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set, 388 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
384 * and the vm_pgoff will point to the first PFN mapped: thus every 389 * set, and the vm_pgoff will point to the first PFN mapped: thus every
385 * page that is a raw mapping will always honor the rule 390 * page that is a raw mapping will always honor the rule
386 * 391 *
387 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 392 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
388 * 393 *
389 * and if that isn't true, the page has been COW'ed (in which case it 394 * A call to vm_normal_page() will return NULL for such a page.
390 * _does_ have a "struct page" associated with it even if it is in a 395 *
391 * VM_PFNMAP range). 396 * If the page doesn't follow the "remap_pfn_range()" rule in a VM_PFNMAP
397 * then the page has been COW'ed. A COW'ed page _does_ have a "struct page"
398 * associated with it even if it is in a VM_PFNMAP range. Calling
399 * vm_normal_page() on such a page will therefore return the "struct page".
400 *
401 *
402 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
403 * page" backing, however the difference is that _all_ pages with a struct
404 * page (that is, those where pfn_valid is true) are refcounted and considered
405 * normal pages by the VM. The disadvantage is that pages are refcounted
406 * (which can be slower and simply not an option for some PFNMAP users). The
407 * advantage is that we don't have to follow the strict linearity rule of
408 * PFNMAP mappings in order to support COWable mappings.
409 *
410 * A call to vm_normal_page() with a VM_MIXEDMAP mapping will return the
411 * associated "struct page" or NULL for memory not backed by a "struct page".
412 *
413 *
414 * All other mappings should have a valid struct page, which will be
415 * returned by a call to vm_normal_page().
392 */ 416 */
393struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 417struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
394{ 418{
395 unsigned long pfn = pte_pfn(pte); 419 unsigned long pfn = pte_pfn(pte);
396 420
397 if (unlikely(vma->vm_flags & VM_PFNMAP)) { 421 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
398 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT; 422 if (vma->vm_flags & VM_MIXEDMAP) {
399 if (pfn == vma->vm_pgoff + off) 423 if (!pfn_valid(pfn))
400 return NULL; 424 return NULL;
401 if (!is_cow_mapping(vma->vm_flags)) 425 goto out;
402 return NULL; 426 } else {
427 unsigned long off = (addr-vma->vm_start) >> PAGE_SHIFT;
428 if (pfn == vma->vm_pgoff + off)
429 return NULL;
430 if (!is_cow_mapping(vma->vm_flags))
431 return NULL;
432 }
403 } 433 }
404 434
405#ifdef CONFIG_DEBUG_VM 435#ifdef CONFIG_DEBUG_VM
@@ -422,6 +452,7 @@ struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_
422 * The PAGE_ZERO() pages and various VDSO mappings can 452 * The PAGE_ZERO() pages and various VDSO mappings can
423 * cause them to exist. 453 * cause them to exist.
424 */ 454 */
455out:
425 return pfn_to_page(pfn); 456 return pfn_to_page(pfn);
426} 457}
427 458
@@ -1232,8 +1263,11 @@ int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1232 pte_t *pte, entry; 1263 pte_t *pte, entry;
1233 spinlock_t *ptl; 1264 spinlock_t *ptl;
1234 1265
1235 BUG_ON(!(vma->vm_flags & VM_PFNMAP)); 1266 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1236 BUG_ON(is_cow_mapping(vma->vm_flags)); 1267 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1268 (VM_PFNMAP|VM_MIXEDMAP));
1269 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1270 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
1237 1271
1238 retval = -ENOMEM; 1272 retval = -ENOMEM;
1239 pte = get_locked_pte(mm, addr, &ptl); 1273 pte = get_locked_pte(mm, addr, &ptl);
@@ -2365,10 +2399,13 @@ static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2365 unsigned long pfn; 2399 unsigned long pfn;
2366 2400
2367 pte_unmap(page_table); 2401 pte_unmap(page_table);
2368 BUG_ON(!(vma->vm_flags & VM_PFNMAP)); 2402 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2369 BUG_ON(is_cow_mapping(vma->vm_flags)); 2403 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
2370 2404
2371 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK); 2405 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2406
2407 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2408
2372 if (unlikely(pfn == NOPFN_OOM)) 2409 if (unlikely(pfn == NOPFN_OOM))
2373 return VM_FAULT_OOM; 2410 return VM_FAULT_OOM;
2374 else if (unlikely(pfn == NOPFN_SIGBUS)) 2411 else if (unlikely(pfn == NOPFN_SIGBUS))