nvdimm: use 'u64' for pfn flags
authorArnd Bergmann <arnd@arndb.de>
Mon, 22 Feb 2016 21:58:34 +0000 (22:58 +0100)
committerDan Williams <dan.j.williams@intel.com>
Wed, 24 Feb 2016 01:17:20 +0000 (17:17 -0800)
A recent bugfix changed pfn_t to always be 64-bit wide, but did not
change the code in pmem.c, which is now broken on 32-bit architectures
as reported by gcc:

In file included from ../drivers/nvdimm/pmem.c:28:0:
drivers/nvdimm/pmem.c: In function 'pmem_alloc':
include/linux/pfn_t.h:15:17: error: large integer implicitly truncated to unsigned type [-Werror=overflow]
 #define PFN_DEV (1ULL << (BITS_PER_LONG_LONG - 3))

This changes the intermediate pfn_flags in struct pmem_device to
be 64 bit wide as well, so they can store the flags correctly.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: db78c22230d0 ("mm: fix pfn_t vs highmem")
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/nvdimm/pmem.c

index 7edf31671dabed8f8d193806f54f3d1b30401e7c..8d0b546701848eb2b57d87b7207490133e5c32a9 100644 (file)
@@ -41,7 +41,7 @@ struct pmem_device {
        phys_addr_t             phys_addr;
        /* when non-zero this device is hosting a 'pfn' instance */
        phys_addr_t             data_offset;
-       unsigned long           pfn_flags;
+       u64                     pfn_flags;
        void __pmem             *virt_addr;
        size_t                  size;
        struct badblocks        bb;