xen/balloon: Fix crash when ballooning on x86 32 bit PAE

commit dfd74a1edfaba5864276a2859190a8d242d18952 upstream.

Commit 55b3da98a40dbb3776f7454daf0d95dde25c33d2 (xen/balloon: find
non-conflicting regions to place hotplugged memory) caused a
regression in 4.4.

When ballooning on an x86 32 bit PAE system with close to 64 GiB of
memory, the address returned by allocate_resource may be above 64 GiB.
When using CONFIG_SPARSEMEM, this setup is limited to using physical
addresses < 64 GiB.  When adding memory at this address, it runs off
the end of the mem_section array and causes a crash.  Instead, fail
the ballooning request.

Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
index 12eab50..364bc44 100644 (file)
--- a/drivers/xen/balloon.c
+++ b/drivers/xen/balloon.c
@@ -152,6 +152,8 @@ static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
 static void balloon_process(struct work_struct *work);
 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
 
+static void release_memory_resource(struct resource *resource);
+
 /* When ballooning out (allocating memory to return to Xen) we don't really
    want the kernel to try too hard since that can trigger the oom killer. */
 #define GFP_BALLOON \
@@ -268,6 +270,20 @@ static struct resource *additional_memory_resource(phys_addr_t size)
                return NULL;
        }
 
+#ifdef CONFIG_SPARSEMEM
+       {
+               unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
+               unsigned long pfn = res->start >> PAGE_SHIFT;
+
+               if (pfn > limit) {
+                       pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
+                              pfn, limit);
+                       release_memory_resource(res);
+                       return NULL;
+               }
+       }
+#endif
+
        return res;
 }