hv_balloon: don't check for memhp_auto_online manually
authorDavid Hildenbrand <david@redhat.com>
Tue, 7 Apr 2020 03:07:32 +0000 (20:07 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 7 Apr 2020 17:43:40 +0000 (10:43 -0700)
We get the MEM_ONLINE notifier call if memory is added right from the
kernel via add_memory() or later from user space.

Let's get rid of the "ha_waiting" flag - the wait event has an inbuilt
mechanism (->done) for that.  Initialize the wait event only once and
reinitialize before adding memory.  Unconditionally call complete() and
wait_for_completion_timeout().

If there are no waiters, complete() will only increment ->done - which
will be reset by reinit_completion().  If complete() has already been
called, wait_for_completion_timeout() will not wait.

There is still the chance for a small race between concurrent
reinit_completion() and complete().  If complete() wins, we would not wait
- which is tolerable (and the race exists in current code as well).

Note: We only wait for "some" memory to get onlined, which seems to be
      good enough for now.

[akpm@linux-foundation.org: register_memory_notifier() after init_completion(), per David]
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Wei Liu <wei.liu@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Yumei Huang <yuhuang@redhat.com>
Link: http://lkml.kernel.org/r/20200317104942.11178-6-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/hv/hv_balloon.c

index a02ce43d778da22dd5e6cee60ce38dfd4fedc38b..32e3bc0aa665a1ba6e0c45377d94d3ad9f7e2f30 100644 (file)
@@ -533,7 +533,6 @@ struct hv_dynmem_device {
         * State to synchronize hot-add.
         */
        struct completion  ol_waitevent;
-       bool ha_waiting;
        /*
         * This thread handles hot-add
         * requests from the host as well as notifying
@@ -634,10 +633,7 @@ static int hv_memory_notifier(struct notifier_block *nb, unsigned long val,
        switch (val) {
        case MEM_ONLINE:
        case MEM_CANCEL_ONLINE:
-               if (dm_device.ha_waiting) {
-                       dm_device.ha_waiting = false;
-                       complete(&dm_device.ol_waitevent);
-               }
+               complete(&dm_device.ol_waitevent);
                break;
 
        case MEM_OFFLINE:
@@ -726,8 +722,7 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
                has->covered_end_pfn +=  processed_pfn;
                spin_unlock_irqrestore(&dm_device.ha_lock, flags);
 
-               init_completion(&dm_device.ol_waitevent);
-               dm_device.ha_waiting = !memhp_auto_online;
+               reinit_completion(&dm_device.ol_waitevent);
 
                nid = memory_add_physaddr_to_nid(PFN_PHYS(start_pfn));
                ret = add_memory(nid, PFN_PHYS((start_pfn)),
@@ -753,15 +748,14 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
                }
 
                /*
-                * Wait for the memory block to be onlined when memory onlining
-                * is done outside of kernel (memhp_auto_online). Since the hot
-                * add has succeeded, it is ok to proceed even if the pages in
-                * the hot added region have not been "onlined" within the
-                * allowed time.
+                * Wait for memory to get onlined. If the kernel onlined the
+                * memory when adding it, this will return directly. Otherwise,
+                * it will wait for user space to online the memory. This helps
+                * to avoid adding memory faster than it is getting onlined. As
+                * adding succeeded, it is ok to proceed even if the memory was
+                * not onlined in time.
                 */
-               if (dm_device.ha_waiting)
-                       wait_for_completion_timeout(&dm_device.ol_waitevent,
-                                                   5*HZ);
+               wait_for_completion_timeout(&dm_device.ol_waitevent, 5 * HZ);
                post_status(&dm_device);
        }
 }
@@ -1706,6 +1700,7 @@ static int balloon_probe(struct hv_device *dev,
 
 #ifdef CONFIG_MEMORY_HOTPLUG
        set_online_page_callback(&hv_online_page);
+       init_completion(&dm_device.ol_waitevent);
        register_memory_notifier(&hv_memory_nb);
 #endif