Re: [RFC PATCH 21/23] powerpc64 is not necessarily BigEndian anymore! :)

[Andrey Borzenkov]

В Fri, 04 Apr 2014 20:24:58 +0200
Dinar Valeev <address@hidden> пишет:

> 
> Right, my mistake. I recall a message message with 64bit LE patches. But
> seems that came from somewhere else.
> 
> Long story short. With 32Bit BE stage one I had several issues like
> accessing btrfs and booting from media. I gave up on my hack, and now
> use proposed patches (64Bit LE).
>

Well, then this is the real bug that has to be fixed. btrfs driver is
supposed to be endian-clean.

Did you try to disable SUSE btrfs patch to verify? There is at least one
suspicious place

+  tree = grub_le_to_cpu64(data->sblock.root_tree);
+  err = get_fs_root(data, tree, grub_cpu_to_le64(GRUB_BTRFS_FS_TREE_OBJECTID),
+                    0, &fs_root);

get_fs_root expects "tree" in on-disk format, not in CPU format.

+get_fs_root(struct grub_btrfs_data *data, grub_uint64_t tree,
+            grub_uint64_t objectid, grub_uint64_t offset,
+            grub_uint64_t *fs_root)
...

+  err = lower_bound(data, &key_in, &key_out, tree,
+                    &elemaddr, &elemsize, &desc, 0);

and lower_bound converts fourth argument again

lower_bound (struct grub_btrfs_data *data,
             const struct grub_btrfs_key *key_in,
             struct grub_btrfs_key *key_out,
             grub_uint64_t root
...
  grub_disk_addr_t addr = grub_le_to_cpu64 (root);


> I'll submit it to Base:System.
>