mm/cma: fix potential memory loss on cma_declare_contiguous_nid
authorLevi Yun <ppbuk5246@gmail.com>
Wed, 18 Jan 2023 08:05:23 +0000 (17:05 +0900)
committerAndrew Morton <akpm@linux-foundation.org>
Fri, 3 Feb 2023 06:33:24 +0000 (22:33 -0800)
Suppose memblock_alloc_range_nid() with highmem_start succeeds when
cma_declare_contiguous_nid is called with !fixed on a 32-bit system with
PHYS_ADDR_T_64BIT enabled with memblock.bottom_up == false.

But the next trial to memblock_alloc_range_nid() to allocate in [SIZE_4G,
limits) nullifies former successfully allocated addr and it retries
memblock_alloc_ragne_nid().

In this situation, the first successfully allocated address area is lost.

Change the order of allocation (SIZE_4G, high_memory and base) and check
whether the allocated succeeded to prevent potential memory loss.

Link: https://lkml.kernel.org/r/20230118080523.44522-1-ppbuk5246@gmail.com
Signed-off-by: Levi Yun <ppbuk5246@gmail.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/cma.c

index a75b17b..a7263aa 100644 (file)
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -322,18 +322,6 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
                phys_addr_t addr = 0;
 
                /*
-                * All pages in the reserved area must come from the same zone.
-                * If the requested region crosses the low/high memory boundary,
-                * try allocating from high memory first and fall back to low
-                * memory in case of failure.
-                */
-               if (base < highmem_start && limit > highmem_start) {
-                       addr = memblock_alloc_range_nid(size, alignment,
-                                       highmem_start, limit, nid, true);
-                       limit = highmem_start;
-               }
-
-               /*
                 * If there is enough memory, try a bottom-up allocation first.
                 * It will place the new cma area close to the start of the node
                 * and guarantee that the compaction is moving pages out of the
@@ -350,6 +338,18 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
                }
 #endif
 
+               /*
+                * All pages in the reserved area must come from the same zone.
+                * If the requested region crosses the low/high memory boundary,
+                * try allocating from high memory first and fall back to low
+                * memory in case of failure.
+                */
+               if (!addr && base < highmem_start && limit > highmem_start) {
+                       addr = memblock_alloc_range_nid(size, alignment,
+                                       highmem_start, limit, nid, true);
+                       limit = highmem_start;
+               }
+
                if (!addr) {
                        addr = memblock_alloc_range_nid(size, alignment, base,
                                        limit, nid, true);