git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From de35994ecd2dd6148ab5a6c5050a1670a04dec77 Mon Sep 17 00:00:00 2001
   2 From: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
   3 Date: Thu, 19 Dec 2024 09:30:30 +0000
   4 Subject: workqueue: Do not warn when cancelling WQ_MEM_RECLAIM work from !WQ_MEM_RECLAIM worker
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 From: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
  10
  11 commit de35994ecd2dd6148ab5a6c5050a1670a04dec77 upstream.
  12
  13 After commit
  14 746ae46c1113 ("drm/sched: Mark scheduler work queues with WQ_MEM_RECLAIM")
  15 amdgpu started seeing the following warning:
  16
  17  [ ] workqueue: WQ_MEM_RECLAIM sdma0:drm_sched_run_job_work [gpu_sched] is flushing !WQ_MEM_RECLAIM events:amdgpu_device_delay_enable_gfx_off [amdgpu]
  18 ...
  19  [ ] Workqueue: sdma0 drm_sched_run_job_work [gpu_sched]
  20 ...
  21  [ ] Call Trace:
  22  [ ]  <TASK>
  23 ...
  24  [ ]  ? check_flush_dependency+0xf5/0x110
  25 ...
  26  [ ]  cancel_delayed_work_sync+0x6e/0x80
  27  [ ]  amdgpu_gfx_off_ctrl+0xab/0x140 [amdgpu]
  28  [ ]  amdgpu_ring_alloc+0x40/0x50 [amdgpu]
  29  [ ]  amdgpu_ib_schedule+0xf4/0x810 [amdgpu]
  30  [ ]  ? drm_sched_run_job_work+0x22c/0x430 [gpu_sched]
  31  [ ]  amdgpu_job_run+0xaa/0x1f0 [amdgpu]
  32  [ ]  drm_sched_run_job_work+0x257/0x430 [gpu_sched]
  33  [ ]  process_one_work+0x217/0x720
  34 ...
  35  [ ]  </TASK>
  36
  37 The intent of the verifcation done in check_flush_depedency is to ensure
  38 forward progress during memory reclaim, by flagging cases when either a
  39 memory reclaim process, or a memory reclaim work item is flushed from a
  40 context not marked as memory reclaim safe.
  41
  42 This is correct when flushing, but when called from the
  43 cancel(_delayed)_work_sync() paths it is a false positive because work is
  44 either already running, or will not be running at all. Therefore
  45 cancelling it is safe and we can relax the warning criteria by letting the
  46 helper know of the calling context.
  47
  48 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
  49 Fixes: fca839c00a12 ("workqueue: warn if memory reclaim tries to flush !WQ_MEM_RECLAIM workqueue")
  50 References: 746ae46c1113 ("drm/sched: Mark scheduler work queues with WQ_MEM_RECLAIM")
  51 Cc: Tejun Heo <tj@kernel.org>
  52 Cc: Peter Zijlstra <peterz@infradead.org>
  53 Cc: Lai Jiangshan <jiangshanlai@gmail.com>
  54 Cc: Alex Deucher <alexander.deucher@amd.com>
  55 Cc: Christian König <christian.koenig@amd.com
  56 Cc: Matthew Brost <matthew.brost@intel.com>
  57 Cc: <stable@vger.kernel.org> # v4.5+
  58 Signed-off-by: Tejun Heo <tj@kernel.org>
  59 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  60 ---
  61  kernel/workqueue.c |   22 +++++++++++++---------
  62  1 file changed, 13 insertions(+), 9 deletions(-)
  63
  64 --- a/kernel/workqueue.c
  65 +++ b/kernel/workqueue.c
  66 @@ -3680,23 +3680,27 @@ void workqueue_softirq_dead(unsigned int
  67   * check_flush_dependency - check for flush dependency sanity
  68   * @target_wq: workqueue being flushed
  69   * @target_work: work item being flushed (NULL for workqueue flushes)
  70 + * @from_cancel: are we called from the work cancel path
  71   *
  72   * %current is trying to flush the whole @target_wq or @target_work on it.
  73 - * If @target_wq doesn't have %WQ_MEM_RECLAIM, verify that %current is not
  74 - * reclaiming memory or running on a workqueue which doesn't have
  75 - * %WQ_MEM_RECLAIM as that can break forward-progress guarantee leading to
  76 - * a deadlock.
  77 + * If this is not the cancel path (which implies work being flushed is either
  78 + * already running, or will not be at all), check if @target_wq doesn't have
  79 + * %WQ_MEM_RECLAIM and verify that %current is not reclaiming memory or running
  80 + * on a workqueue which doesn't have %WQ_MEM_RECLAIM as that can break forward-
  81 + * progress guarantee leading to a deadlock.
  82   */
  83  static void check_flush_dependency(struct workqueue_struct *target_wq,
  84 -                                  struct work_struct *target_work)
  85 +                                  struct work_struct *target_work,
  86 +                                  bool from_cancel)
  87  {
  88 -       work_func_t target_func = target_work ? target_work->func : NULL;
  89 +       work_func_t target_func;
  90         struct worker *worker;
  91
  92 -       if (target_wq->flags & WQ_MEM_RECLAIM)
  93 +       if (from_cancel || target_wq->flags & WQ_MEM_RECLAIM)
  94                 return;
  95
  96         worker = current_wq_worker();
  97 +       target_func = target_work ? target_work->func : NULL;
  98
  99         WARN_ONCE(current->flags & PF_MEMALLOC,
 100                   "workqueue: PF_MEMALLOC task %d(%s) is flushing !WQ_MEM_RECLAIM %s:%ps",
 101 @@ -3966,7 +3970,7 @@ void __flush_workqueue(struct workqueue_
 102                 list_add_tail(&this_flusher.list, &wq->flusher_overflow);
 103         }
 104
 105 -       check_flush_dependency(wq, NULL);
 106 +       check_flush_dependency(wq, NULL, false);
 107
 108         mutex_unlock(&wq->mutex);
 109
 110 @@ -4141,7 +4145,7 @@ static bool start_flush_work(struct work
 111         }
 112
 113         wq = pwq->wq;
 114 -       check_flush_dependency(wq, work);
 115 +       check_flush_dependency(wq, work, from_cancel);
 116
 117         insert_wq_barrier(pwq, barr, work, worker);
 118         raw_spin_unlock_irq(&pool->lock);