]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
mm, x86, pat: rework linear pfn-mmap tracking
authorKonstantin Khlebnikov <khlebnikov@openvz.org>
Mon, 8 Oct 2012 23:28:34 +0000 (16:28 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 9 Oct 2012 07:22:16 +0000 (16:22 +0900)
Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT.

We can toss mapping address from remap_pfn_range() into
track_pfn_vma_new(), and collect all PAT-related logic together in
arch/x86/.

This patch also restores orignal frustration-free is_cow_mapping() check
in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73
("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3")

is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c,
because it already handled by VM_PFNMAP in VM_NO_THP bit-mask.

[suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()]
Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Venkatesh Pallipadi <venki@google.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Carsten Otte <cotte@de.ibm.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Eric Paris <eparis@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Kentaro Takeda <takedakn@nttdata.co.jp>
Cc: Matt Helsley <matthltc@us.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Venkatesh Pallipadi <venki@google.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/x86/mm/pat.c
include/asm-generic/pgtable.h
include/linux/mm.h
mm/huge_memory.c
mm/memory.c

index 74a702674e866fd7346a348234ac9d8e06d0f3c6..0eb572eda4060338543bd0dfc97cf05bccaeb6cb 100644 (file)
@@ -677,7 +677,7 @@ int track_pfn_copy(struct vm_area_struct *vma)
        unsigned long vma_size = vma->vm_end - vma->vm_start;
        pgprot_t pgprot;
 
-       if (is_linear_pfn_mapping(vma)) {
+       if (vma->vm_flags & VM_PAT) {
                /*
                 * reserve the whole chunk covered by vma. We need the
                 * starting address and protection from pte.
@@ -699,14 +699,20 @@ int track_pfn_copy(struct vm_area_struct *vma)
  * single reserve_pfn_range call.
  */
 int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
-                   unsigned long pfn, unsigned long size)
+                   unsigned long pfn, unsigned long addr, unsigned long size)
 {
        resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
        unsigned long flags;
 
        /* reserve the whole chunk starting from paddr */
-       if (is_linear_pfn_mapping(vma))
-               return reserve_pfn_range(paddr, size, prot, 0);
+       if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
+               int ret;
+
+               ret = reserve_pfn_range(paddr, size, prot, 0);
+               if (!ret)
+                       vma->vm_flags |= VM_PAT;
+               return ret;
+       }
 
        if (!pat_enabled)
                return 0;
@@ -758,7 +764,7 @@ void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
        resource_size_t paddr;
        unsigned long prot;
 
-       if (!is_linear_pfn_mapping(vma))
+       if (!(vma->vm_flags & VM_PAT))
                return;
 
        /* free the chunk starting from pfn or the whole chunk */
@@ -772,6 +778,7 @@ void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
                size = vma->vm_end - vma->vm_start;
        }
        free_pfn_range(paddr, size);
+       vma->vm_flags &= ~VM_PAT;
 }
 
 pgprot_t pgprot_writecombine(pgprot_t prot)
index d4d4592c97fc8d7909617253568a961866a66158..c9a612069c8ebd2726ce0d819156c2ec7bc1297e 100644 (file)
@@ -391,7 +391,8 @@ static inline void ptep_modify_prot_commit(struct mm_struct *mm,
  * by remap_pfn_range() for physical range indicated by pfn and size.
  */
 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
-                                 unsigned long pfn, unsigned long size)
+                                 unsigned long pfn, unsigned long addr,
+                                 unsigned long size)
 {
        return 0;
 }
@@ -426,7 +427,8 @@ static inline void untrack_pfn(struct vm_area_struct *vma,
 }
 #else
 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
-                          unsigned long pfn, unsigned long size);
+                          unsigned long pfn, unsigned long addr,
+                          unsigned long size);
 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
                            unsigned long pfn);
 extern int track_pfn_copy(struct vm_area_struct *vma);
index 311be906b57d8498d5c3529d0e42de0996a0c1ec..75d1632d3477ee9ac792cb222f86aeff50f0ccf2 100644 (file)
@@ -117,7 +117,7 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_CAN_NONLINEAR 0x08000000    /* Has ->fault & does nonlinear pages */
 #define VM_MIXEDMAP    0x10000000      /* Can contain "struct page" and pure PFN pages */
 #define VM_SAO         0x20000000      /* Strong Access Ordering (powerpc) */
-#define VM_PFN_AT_MMAP 0x40000000      /* PFNMAP vma that is fully mapped at mmap time */
+#define VM_PAT         0x40000000      /* PAT reserves whole VMA at once (x86) */
 #define VM_MERGEABLE   0x80000000      /* KSM may merge identical pages */
 
 /* Bits set in the VMA until the stack is in its final location */
@@ -158,24 +158,6 @@ extern pgprot_t protection_map[16];
 #define FAULT_FLAG_RETRY_NOWAIT        0x10    /* Don't drop mmap_sem and wait when retrying */
 #define FAULT_FLAG_KILLABLE    0x20    /* The fault task is in SIGKILL killable region */
 
-/*
- * This interface is used by x86 PAT code to identify a pfn mapping that is
- * linear over entire vma. This is to optimize PAT code that deals with
- * marking the physical region with a particular prot. This is not for generic
- * mm use. Note also that this check will not work if the pfn mapping is
- * linear for a vma starting at physical address 0. In which case PAT code
- * falls back to slow path of reserving physical range page by page.
- */
-static inline int is_linear_pfn_mapping(struct vm_area_struct *vma)
-{
-       return !!(vma->vm_flags & VM_PFN_AT_MMAP);
-}
-
-static inline int is_pfn_mapping(struct vm_area_struct *vma)
-{
-       return !!(vma->vm_flags & VM_PFNMAP);
-}
-
 /*
  * vm_fault is filled by the the pagefault handler and passed to the vma's
  * ->fault function. The vma's ->fault is responsible for returning a bitmask
index 141dbb695097c1f0674b8978456eb1d0c98e3e67..73cb22ee96654c9c548d397f7b4f621e46721e22 100644 (file)
@@ -1655,11 +1655,7 @@ int khugepaged_enter_vma_merge(struct vm_area_struct *vma)
        if (vma->vm_ops)
                /* khugepaged not yet working on file or special mappings */
                return 0;
-       /*
-        * If is_pfn_mapping() is true is_learn_pfn_mapping() must be
-        * true too, verify it here.
-        */
-       VM_BUG_ON(is_linear_pfn_mapping(vma) || vma->vm_flags & VM_NO_THP);
+       VM_BUG_ON(vma->vm_flags & VM_NO_THP);
        hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
        hend = vma->vm_end & HPAGE_PMD_MASK;
        if (hstart < hend)
@@ -1912,11 +1908,7 @@ static void collapse_huge_page(struct mm_struct *mm,
                goto out;
        if (is_vma_temporary_stack(vma))
                goto out;
-       /*
-        * If is_pfn_mapping() is true is_learn_pfn_mapping() must be
-        * true too, verify it here.
-        */
-       VM_BUG_ON(is_linear_pfn_mapping(vma) || vma->vm_flags & VM_NO_THP);
+       VM_BUG_ON(vma->vm_flags & VM_NO_THP);
 
        pgd = pgd_offset(mm, address);
        if (!pgd_present(*pgd))
@@ -2154,12 +2146,7 @@ static unsigned int khugepaged_scan_mm_slot(unsigned int pages,
                        goto skip;
                if (is_vma_temporary_stack(vma))
                        goto skip;
-               /*
-                * If is_pfn_mapping() is true is_learn_pfn_mapping()
-                * must be true too, verify it here.
-                */
-               VM_BUG_ON(is_linear_pfn_mapping(vma) ||
-                         vma->vm_flags & VM_NO_THP);
+               VM_BUG_ON(vma->vm_flags & VM_NO_THP);
 
                hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
                hend = vma->vm_end & HPAGE_PMD_MASK;
index 6bef278ad3033868deebcb4256d64349e02aa380..655e1429388a364014899b4a84ea59e97500a5ea 100644 (file)
@@ -1055,7 +1055,7 @@ int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        if (is_vm_hugetlb_page(vma))
                return copy_hugetlb_page_range(dst_mm, src_mm, vma);
 
-       if (unlikely(is_pfn_mapping(vma))) {
+       if (unlikely(vma->vm_flags & VM_PFNMAP)) {
                /*
                 * We do not free on error cases below as remove_vma
                 * gets called on error from higher level routine
@@ -1327,7 +1327,7 @@ static void unmap_single_vma(struct mmu_gather *tlb,
        if (vma->vm_file)
                uprobe_munmap(vma, start, end);
 
-       if (unlikely(is_pfn_mapping(vma)))
+       if (unlikely(vma->vm_flags & VM_PFNMAP))
                untrack_pfn(vma, 0, 0);
 
        if (start != end) {
@@ -2299,26 +2299,20 @@ int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
         * There's a horrible special case to handle copy-on-write
         * behaviour that some programs depend on. We mark the "original"
         * un-COW'ed pages by matching them up with "vma->vm_pgoff".
+        * See vm_normal_page() for details.
         */
-       if (addr == vma->vm_start && end == vma->vm_end) {
+       if (is_cow_mapping(vma->vm_flags)) {
+               if (addr != vma->vm_start || end != vma->vm_end)
+                       return -EINVAL;
                vma->vm_pgoff = pfn;
-               vma->vm_flags |= VM_PFN_AT_MMAP;
-       } else if (is_cow_mapping(vma->vm_flags))
+       }
+
+       err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
+       if (err)
                return -EINVAL;
 
        vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
 
-       err = track_pfn_remap(vma, &prot, pfn, PAGE_ALIGN(size));
-       if (err) {
-               /*
-                * To indicate that track_pfn related cleanup is not
-                * needed from higher level routine calling unmap_vmas
-                */
-               vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
-               vma->vm_flags &= ~VM_PFN_AT_MMAP;
-               return -EINVAL;
-       }
-
        BUG_ON(addr >= end);
        pfn -= addr >> PAGE_SHIFT;
        pgd = pgd_offset(mm, addr);