mm: remove free_area_cache use in powerpc architecture
authorMichel Lespinasse <walken@google.com>
Mon, 29 Apr 2013 18:53:52 +0000 (11:53 -0700)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Tue, 30 Apr 2013 01:05:10 +0000 (11:05 +1000)
As all other architectures have been converted to use vm_unmapped_area(),
we are about to retire the free_area_cache.

This change simply removes the use of that cache in
slice_get_unmapped_area(), which will most certainly have a
performance cost. Next one will convert that function to use the
vm_unmapped_area() infrastructure and regain the performance.

Signed-off-by: Michel Lespinasse <walken@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/page_64.h
arch/powerpc/mm/hugetlbpage.c
arch/powerpc/mm/slice.c
arch/powerpc/platforms/cell/spufs/file.c

index cd915d6..88693ce 100644 (file)
@@ -99,8 +99,7 @@ extern unsigned long slice_get_unmapped_area(unsigned long addr,
                                             unsigned long len,
                                             unsigned long flags,
                                             unsigned int psize,
-                                            int topdown,
-                                            int use_cache);
+                                            int topdown);
 
 extern unsigned int get_slice_psize(struct mm_struct *mm,
                                    unsigned long addr);
index 1a6de0a..5dc52d8 100644 (file)
@@ -742,7 +742,7 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
        struct hstate *hstate = hstate_file(file);
        int mmu_psize = shift_to_mmu_psize(huge_page_shift(hstate));
 
-       return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1, 0);
+       return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1);
 }
 #endif
 
index cf9dada..999a74f 100644 (file)
@@ -240,23 +240,15 @@ static void slice_convert(struct mm_struct *mm, struct slice_mask mask, int psiz
 static unsigned long slice_find_area_bottomup(struct mm_struct *mm,
                                              unsigned long len,
                                              struct slice_mask available,
-                                             int psize, int use_cache)
+                                             int psize)
 {
        struct vm_area_struct *vma;
-       unsigned long start_addr, addr;
+       unsigned long addr;
        struct slice_mask mask;
        int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
 
-       if (use_cache) {
-               if (len <= mm->cached_hole_size) {
-                       start_addr = addr = TASK_UNMAPPED_BASE;
-                       mm->cached_hole_size = 0;
-               } else
-                       start_addr = addr = mm->free_area_cache;
-       } else
-               start_addr = addr = TASK_UNMAPPED_BASE;
+       addr = TASK_UNMAPPED_BASE;
 
-full_search:
        for (;;) {
                addr = _ALIGN_UP(addr, 1ul << pshift);
                if ((TASK_SIZE - len) < addr)
@@ -272,63 +264,24 @@ full_search:
                                addr = _ALIGN_UP(addr + 1,  1ul << SLICE_HIGH_SHIFT);
                        continue;
                }
-               if (!vma || addr + len <= vma->vm_start) {
-                       /*
-                        * Remember the place where we stopped the search:
-                        */
-                       if (use_cache)
-                               mm->free_area_cache = addr + len;
+               if (!vma || addr + len <= vma->vm_start)
                        return addr;
-               }
-               if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start)
-                       mm->cached_hole_size = vma->vm_start - addr;
                addr = vma->vm_end;
        }
 
-       /* Make sure we didn't miss any holes */
-       if (use_cache && start_addr != TASK_UNMAPPED_BASE) {
-               start_addr = addr = TASK_UNMAPPED_BASE;
-               mm->cached_hole_size = 0;
-               goto full_search;
-       }
        return -ENOMEM;
 }
 
 static unsigned long slice_find_area_topdown(struct mm_struct *mm,
                                             unsigned long len,
                                             struct slice_mask available,
-                                            int psize, int use_cache)
+                                            int psize)
 {
        struct vm_area_struct *vma;
        unsigned long addr;
        struct slice_mask mask;
        int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
 
-       /* check if free_area_cache is useful for us */
-       if (use_cache) {
-               if (len <= mm->cached_hole_size) {
-                       mm->cached_hole_size = 0;
-                       mm->free_area_cache = mm->mmap_base;
-               }
-
-               /* either no address requested or can't fit in requested
-                * address hole
-                */
-               addr = mm->free_area_cache;
-
-               /* make sure it can fit in the remaining address space */
-               if (addr > len) {
-                       addr = _ALIGN_DOWN(addr - len, 1ul << pshift);
-                       mask = slice_range_to_mask(addr, len);
-                       if (slice_check_fit(mask, available) &&
-                           slice_area_is_free(mm, addr, len))
-                                       /* remember the address as a hint for
-                                        * next time
-                                        */
-                                       return (mm->free_area_cache = addr);
-               }
-       }
-
        addr = mm->mmap_base;
        while (addr > len) {
                /* Go down by chunk size */
@@ -352,16 +305,8 @@ static unsigned long slice_find_area_topdown(struct mm_struct *mm,
                 * return with success:
                 */
                vma = find_vma(mm, addr);
-               if (!vma || (addr + len) <= vma->vm_start) {
-                       /* remember the address as a hint for next time */
-                       if (use_cache)
-                               mm->free_area_cache = addr;
+               if (!vma || (addr + len) <= vma->vm_start)
                        return addr;
-               }
-
-               /* remember the largest hole we saw so far */
-               if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start)
-                       mm->cached_hole_size = vma->vm_start - addr;
 
                /* try just below the current vma->vm_start */
                addr = vma->vm_start;
@@ -373,28 +318,18 @@ static unsigned long slice_find_area_topdown(struct mm_struct *mm,
         * can happen with large stack limits and large mmap()
         * allocations.
         */
-       addr = slice_find_area_bottomup(mm, len, available, psize, 0);
-
-       /*
-        * Restore the topdown base:
-        */
-       if (use_cache) {
-               mm->free_area_cache = mm->mmap_base;
-               mm->cached_hole_size = ~0UL;
-       }
-
-       return addr;
+       return slice_find_area_bottomup(mm, len, available, psize);
 }
 
 
 static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
                                     struct slice_mask mask, int psize,
-                                    int topdown, int use_cache)
+                                    int topdown)
 {
        if (topdown)
-               return slice_find_area_topdown(mm, len, mask, psize, use_cache);
+               return slice_find_area_topdown(mm, len, mask, psize);
        else
-               return slice_find_area_bottomup(mm, len, mask, psize, use_cache);
+               return slice_find_area_bottomup(mm, len, mask, psize);
 }
 
 #define or_mask(dst, src)      do {                    \
@@ -415,7 +350,7 @@ static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
 
 unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
                                      unsigned long flags, unsigned int psize,
-                                     int topdown, int use_cache)
+                                     int topdown)
 {
        struct slice_mask mask = {0, 0};
        struct slice_mask good_mask;
@@ -430,8 +365,8 @@ unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
        BUG_ON(mm->task_size == 0);
 
        slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize);
-       slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d, use_cache=%d\n",
-                 addr, len, flags, topdown, use_cache);
+       slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n",
+                 addr, len, flags, topdown);
 
        if (len > mm->task_size)
                return -ENOMEM;
@@ -503,8 +438,7 @@ unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
                /* Now let's see if we can find something in the existing
                 * slices for that size
                 */
-               newaddr = slice_find_area(mm, len, good_mask, psize, topdown,
-                                         use_cache);
+               newaddr = slice_find_area(mm, len, good_mask, psize, topdown);
                if (newaddr != -ENOMEM) {
                        /* Found within the good mask, we don't have to setup,
                         * we thus return directly
@@ -536,8 +470,7 @@ unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
         * anywhere in the good area.
         */
        if (addr) {
-               addr = slice_find_area(mm, len, good_mask, psize, topdown,
-                                      use_cache);
+               addr = slice_find_area(mm, len, good_mask, psize, topdown);
                if (addr != -ENOMEM) {
                        slice_dbg(" found area at 0x%lx\n", addr);
                        return addr;
@@ -547,15 +480,14 @@ unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
        /* Now let's see if we can find something in the existing slices
         * for that size plus free slices
         */
-       addr = slice_find_area(mm, len, potential_mask, psize, topdown,
-                              use_cache);
+       addr = slice_find_area(mm, len, potential_mask, psize, topdown);
 
 #ifdef CONFIG_PPC_64K_PAGES
        if (addr == -ENOMEM && psize == MMU_PAGE_64K) {
                /* retry the search with 4k-page slices included */
                or_mask(potential_mask, compat_mask);
                addr = slice_find_area(mm, len, potential_mask, psize,
-                                      topdown, use_cache);
+                                      topdown);
        }
 #endif
 
@@ -586,8 +518,7 @@ unsigned long arch_get_unmapped_area(struct file *filp,
                                     unsigned long flags)
 {
        return slice_get_unmapped_area(addr, len, flags,
-                                      current->mm->context.user_psize,
-                                      0, 1);
+                                      current->mm->context.user_psize, 0);
 }
 
 unsigned long arch_get_unmapped_area_topdown(struct file *filp,
@@ -597,8 +528,7 @@ unsigned long arch_get_unmapped_area_topdown(struct file *filp,
                                             const unsigned long flags)
 {
        return slice_get_unmapped_area(addr0, len, flags,
-                                      current->mm->context.user_psize,
-                                      1, 1);
+                                      current->mm->context.user_psize, 1);
 }
 
 unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr)
index 68c57d3..0026a37 100644 (file)
@@ -352,7 +352,7 @@ static unsigned long spufs_get_unmapped_area(struct file *file,
 
        /* Else, try to obtain a 64K pages slice */
        return slice_get_unmapped_area(addr, len, flags,
-                                      MMU_PAGE_64K, 1, 0);
+                                      MMU_PAGE_64K, 1);
 }
 #endif /* CONFIG_SPU_FS_64K_LS */