mm: hotplug: Don't release twice the resource on error

The function add_memory_resource take in parameter a resource allocated
by the caller. On error, both add_memory_resource and the caller will
release the resource via release_memory_source.

This will result to Linux crashing when the caller is trying to release
the resource:

CPU: 1 PID: 45 Comm: xenwatch Not tainted 4.3.0-rc6-00043-g5e1d6ca-dirty torvalds#170
Hardware name: XENVM-4.7 (DT)
task: ffffffc1fb2421c0 ti: ffffffc1fb270000 task.ti:
ffffffc1fb270000
PC is at __release_resource+0x28/0x8c
LR is at __release_resource+0x24/0x8c

[...]

Call trace:
[<ffffffc0000b711c>] __release_resource+0x28/0x8c
[<ffffffc0000b71a4>] release_resource+0x24/0x44
[<ffffffc00033509c>] reserve_additional_memory+0x114/0x128
[<ffffffc0003358c8>] alloc_xenballooned_pages+0x98/0x16c
[<ffffffc0003a75f0>] blkfront_gather_backend_features+0x14c/0xd68
[<ffffffc0003aa4dc>] blkback_changed+0x678/0x150c
[<ffffffc00033c538>] xenbus_otherend_changed+0x9c/0xa4
[<ffffffc00033e518>] backend_changed+0xc/0x18
[<ffffffc00033bc68>] xenwatch_thread+0xa0/0x13c
[<ffffffc0000cc51c>] kthread+0xdc/0xf4

As the caller is allocating the resource, let him handle the release.
This has been introduced by commit b75351f "mm: memory hotplug with
an existing resource".

Signed-off-by: Julien Grall <[email protected]>

mm/memory_hotplug.c

@@ -1300,7 +1300,6 @@ int __ref add_memory(int nid, u64 start, u64 size)
 	/* rollback pgdat allocation and others */
 	if (new_pgdat)
 		rollback_node_hotadd(nid, pgdat);
-	release_memory_resource(res);
 	memblock_remove(start, size);
 
 out: