NOMMU: Add support for direct mapping through mtdconcat if possible
authorDavid Howells <dhowells@redhat.com>
Thu, 12 Feb 2009 10:40:05 +0000 (10:40 +0000)
committerDavid Woodhouse <David.Woodhouse@intel.com>
Tue, 24 Mar 2009 09:01:08 +0000 (09:01 +0000)
Add support for direct mapping through mtdconcat, if possible, by attaching the
samebacking_dev_info structure to the master.

It has some restrictions:

 (1) It won't permit direct mapping of concatenated devices that have differing
     BDIs.

 (2) It doesn't support maps that span the 'gap' between devices, although it
     possibly could if the devices spanned across return compatible
     (ie. contiguous) addresses from their get_unmapped_area() ops.

Signed-off-by: Gavin Lambert <gavinl@compacsort.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Bernd Schmidt <bernd.schmidt@analog.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
drivers/mtd/mtdconcat.c

index 3dbb1b38db666cf4294047ec8cac5f8610c19525..792b547786b8fdf6c4122f28240b1458c0d80f66 100644 (file)
@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/sched.h>
 #include <linux/types.h>
+#include <linux/backing-dev.h>
 
 #include <linux/mtd/mtd.h>
 #include <linux/mtd/concat.h>
@@ -683,6 +684,40 @@ static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
        return err;
 }
 
+/*
+ * try to support NOMMU mmaps on concatenated devices
+ * - we don't support subdev spanning as we can't guarantee it'll work
+ */
+static unsigned long concat_get_unmapped_area(struct mtd_info *mtd,
+                                             unsigned long len,
+                                             unsigned long offset,
+                                             unsigned long flags)
+{
+       struct mtd_concat *concat = CONCAT(mtd);
+       int i;
+
+       for (i = 0; i < concat->num_subdev; i++) {
+               struct mtd_info *subdev = concat->subdev[i];
+
+               if (offset >= subdev->size) {
+                       offset -= subdev->size;
+                       continue;
+               }
+
+               /* we've found the subdev over which the mapping will reside */
+               if (offset + len > subdev->size)
+                       return (unsigned long) -EINVAL;
+
+               if (subdev->get_unmapped_area)
+                       return subdev->get_unmapped_area(subdev, len, offset,
+                                                        flags);
+
+               break;
+       }
+
+       return (unsigned long) -ENOSYS;
+}
+
 /*
  * This function constructs a virtual MTD device by concatenating
  * num_devs MTD devices. A pointer to the new device object is
@@ -740,6 +775,8 @@ struct mtd_info *mtd_concat_create(struct mtd_info *subdev[],       /* subdevices to c
 
        concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
 
+       concat->mtd.backing_dev_info = subdev[0]->backing_dev_info;
+
        concat->subdev[0] = subdev[0];
 
        for (i = 1; i < num_devs; i++) {
@@ -766,6 +803,15 @@ struct mtd_info *mtd_concat_create(struct mtd_info *subdev[],      /* subdevices to c
                                concat->mtd.flags |=
                                    subdev[i]->flags & MTD_WRITEABLE;
                }
+
+               /* only permit direct mapping if the BDIs are all the same
+                * - copy-mapping is still permitted
+                */
+               if (concat->mtd.backing_dev_info !=
+                   subdev[i]->backing_dev_info)
+                       concat->mtd.backing_dev_info =
+                               &default_backing_dev_info;
+
                concat->mtd.size += subdev[i]->size;
                concat->mtd.ecc_stats.badblocks +=
                        subdev[i]->ecc_stats.badblocks;
@@ -796,6 +842,7 @@ struct mtd_info *mtd_concat_create(struct mtd_info *subdev[],       /* subdevices to c
        concat->mtd.unlock = concat_unlock;
        concat->mtd.suspend = concat_suspend;
        concat->mtd.resume = concat_resume;
+       concat->mtd.get_unmapped_area = concat_get_unmapped_area;
 
        /*
         * Combine the erase block size info of the subdevices: