mm: memcg/slab: call flush_memcg_workqueue() only if memcg workqueue is valid
authorAdrian Huang <ahuang12@lenovo.com>
Tue, 14 Jan 2020 00:29:32 +0000 (16:29 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 Jan 2020 02:19:02 +0000 (18:19 -0800)
When booting with amd_iommu=off, the following WARNING message
appears:

  AMD-Vi: AMD IOMMU disabled on kernel command-line
  ------------[ cut here ]------------
  WARNING: CPU: 0 PID: 0 at kernel/workqueue.c:2772 flush_workqueue+0x42e/0x450
  Modules linked in:
  CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.5.0-rc3-amd-iommu #6
  Hardware name: Lenovo ThinkSystem SR655-2S/7D2WRCZ000, BIOS D8E101L-1.00 12/05/2019
  RIP: 0010:flush_workqueue+0x42e/0x450
  Code: ff 0f 0b e9 7a fd ff ff 4d 89 ef e9 33 fe ff ff 0f 0b e9 7f fd ff ff 0f 0b e9 bc fd ff ff 0f 0b e9 a8 fd ff ff e8 52 2c fe ff <0f> 0b 31 d2 48 c7 c6 e0 88 c5 95 48 c7 c7 d8 ad f0 95 e8 19 f5 04
  Call Trace:
   kmem_cache_destroy+0x69/0x260
   iommu_go_to_state+0x40c/0x5ab
   amd_iommu_prepare+0x16/0x2a
   irq_remapping_prepare+0x36/0x5f
   enable_IR_x2apic+0x21/0x172
   default_setup_apic_routing+0x12/0x6f
   apic_intr_mode_init+0x1a1/0x1f1
   x86_late_time_init+0x17/0x1c
   start_kernel+0x480/0x53f
   secondary_startup_64+0xb6/0xc0
  ---[ end trace 30894107c3749449 ]---
  x2apic: IRQ remapping doesn't support X2APIC mode
  x2apic disabled

The warning is caused by the calling of 'kmem_cache_destroy()'
in free_iommu_resources(). Here is the call path:

  free_iommu_resources
    kmem_cache_destroy
      flush_memcg_workqueue
        flush_workqueue

The root cause is that the IOMMU subsystem runs before the workqueue
subsystem, which the variable 'wq_online' is still 'false'.  This leads
to the statement 'if (WARN_ON(!wq_online))' in flush_workqueue() is
'true'.

Since the variable 'memcg_kmem_cache_wq' is not allocated during the
time, it is unnecessary to call flush_memcg_workqueue().  This prevents
the WARNING message triggered by flush_workqueue().

Link: http://lkml.kernel.org/r/20200103085503.1665-1-ahuang12@lenovo.com
Fixes: 92ee383f6daab ("mm: fix race between kmem_cache destroy, create and deactivate")
Signed-off-by: Adrian Huang <ahuang12@lenovo.com>
Reported-by: Xiaochun Lee <lixc17@lenovo.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slab_common.c

index f0ab6d4..0d95dde 100644 (file)
@@ -903,7 +903,8 @@ static void flush_memcg_workqueue(struct kmem_cache *s)
         * deactivates the memcg kmem_caches through workqueue. Make sure all
         * previous workitems on workqueue are processed.
         */
-       flush_workqueue(memcg_kmem_cache_wq);
+       if (likely(memcg_kmem_cache_wq))
+               flush_workqueue(memcg_kmem_cache_wq);
 
        /*
         * If we're racing with children kmem_cache deactivation, it might