If we hot-remove memory only and leave the cpus alive, the corresponding
node will not be removed.  But the node_start_pfn and node_spanned_pages
in pg_data will be reset to 0.  In this case, when we hot-add the memory
back next time, the node_start_pfn will always be 0 because no pfn is less
than 0.  After that, if we hot-remove the memory again, it will cause
kernel panic in function find_biggest_section_pfn() when it tries to scan
all the pfns.
The zone will also have the same problem.
This patch sets start_pfn to the start_pfn of the section being added when
spanned_pages of the zone or pg_data is 0.
  ---How to reproduce---
1. hot-add a container with some memory and cpus;
2. hot-remove the container's memory, and leave cpus there;
3. hot-add these memory again;
4. hot-remove them again;
then, the kernel will panic.
  ---Call trace---
  BUG: unable to handle kernel paging request at 
00000fff82a8cc38
  IP: [<
ffffffff811c0d55>] find_biggest_section_pfn+0xe5/0x180
  ......
  Call Trace:
   [<
ffffffff811c1124>] __remove_zone+0x184/0x1b0
   [<
ffffffff811c11dc>] __remove_section+0x8c/0xb0
   [<
ffffffff811c12e7>] __remove_pages+0xe7/0x120
   [<
ffffffff81654f7c>] arch_remove_memory+0x2c/0x80
   [<
ffffffff81655bb6>] remove_memory+0x56/0x90
   [<
ffffffff813da0c8>] acpi_memory_device_remove_memory+0x48/0x73
   [<
ffffffff813da55a>] acpi_memory_device_notify+0x153/0x274
   [<
ffffffff813b6786>] acpi_ev_notify_dispatch+0x41/0x5f
   [<
ffffffff813a3867>] acpi_os_execute_deferred+0x27/0x34
   [<
ffffffff81090589>] process_one_work+0x219/0x680
   [<
ffffffff810923be>] worker_thread+0x12e/0x320
   [<
ffffffff81098396>] kthread+0xc6/0xd0
   [<
ffffffff8167c7c4>] kernel_thread_helper+0x4/0x10
  ......
  ---[ end trace 
96d845dbf33fee11 ]---
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
 
        zone_span_writelock(zone);
 
        old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
-       if (start_pfn < zone->zone_start_pfn)
+       if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
                zone->zone_start_pfn = start_pfn;
 
        zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
        unsigned long old_pgdat_end_pfn =
                pgdat->node_start_pfn + pgdat->node_spanned_pages;
 
-       if (start_pfn < pgdat->node_start_pfn)
+       if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
                pgdat->node_start_pfn = start_pfn;
 
        pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -