x86/pv-shim: correct ballooning up for compat guests

The compat layer for multi-extent memory ops may need to split incoming
requests. Since the guest handles in the interface structures may not be
altered, it does so by leveraging do_memory_op()'s continuation
handling: It hands on non-initial requests with a non-zero start extent,
with the (native) handle suitably adjusted down. As a result
do_memory_op() sees only the first of potentially several requests with
start extent being zero. It's only that case when the function would
issue a call to pv_shim_online_memory(), yet the range then covers only
the first sub-range that results from the split.

Address that breakage by making a complementary call to
pv_shim_online_memory() in compat layer.

Fixes: b2245acc60 ("xen/pvshim: memory hotplug")
Signed-off-by: Igor Druzhinin <igor.druzhinin@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
master commit: a0bfdd201e
master date: 2022-10-28 15:48:50 +0200

xen/common/compat/memory.c

@@ -7,6 +7,7 @@ EMIT_FILE;
 #include <xen/event.h>
 #include <xen/mem_access.h>
 #include <asm/current.h>
+#include <asm/guest.h>
 #include <compat/memory.h>
 
 #define xen_domid_t domid_t
@@ -146,7 +147,10 @@ int compat_memory_op(unsigned int cmd, XEN_GUEST_HANDLE_PARAM(void) compat)
                 nat.rsrv->nr_extents = end_extent;
                 ++split;
             }
-
+           /* Avoid calling pv_shim_online_memory() when in a continuation. */
+           if ( pv_shim && op != XENMEM_decrease_reservation && !start_extent )
+               pv_shim_online_memory(cmp.rsrv.nr_extents - nat.rsrv->nr_extents,
+                                     cmp.rsrv.extent_order);
             break;
 
         case XENMEM_exchange: