releases/4.4.73/pm-runtime-avoid-false-positive-warnings-from-might_sleep_if.patch

   1 From foo@baz Thu Jun 15 11:39:37 CEST 2017
   2 From: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
   3 Date: Tue, 23 May 2017 21:53:54 -0400
   4 Subject: PM / runtime: Avoid false-positive warnings from might_sleep_if()
   5
   6 From: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
   7
   8
   9 [ Upstream commit a9306a63631493afc75893a4ac405d4e1cbae6aa ]
  10
  11 The might_sleep_if() assertions in __pm_runtime_idle(),
  12 __pm_runtime_suspend() and __pm_runtime_resume() may generate
  13 false-positive warnings in some situations.  For example, that
  14 happens if a nested pm_runtime_get_sync()/pm_runtime_put() pair
  15 is executed with disabled interrupts within an outer
  16 pm_runtime_get_sync()/pm_runtime_put() section for the same device.
  17 [Generally, pm_runtime_get_sync() may sleep, so it should not be
  18 called with disabled interrupts, but in this particular case the
  19 previous pm_runtime_get_sync() guarantees that the device will not
  20 be suspended, so the inner pm_runtime_get_sync() will return
  21 immediately after incrementing the device's usage counter.]
  22
  23 That started to happen in the i915 driver in 4.10-rc, leading to
  24 the following splat:
  25
  26  BUG: sleeping function called from invalid context at drivers/base/power/runtime.c:1032
  27  in_atomic(): 1, irqs_disabled(): 0, pid: 1500, name: Xorg
  28  1 lock held by Xorg/1500:
  29   #0:  (&dev->struct_mutex){+.+.+.}, at:
  30   [<ffffffffa0680c13>] i915_mutex_lock_interruptible+0x43/0x140 [i915]
  31  CPU: 0 PID: 1500 Comm: Xorg Not tainted
  32  Call Trace:
  33   dump_stack+0x85/0xc2
  34   ___might_sleep+0x196/0x260
  35   __might_sleep+0x53/0xb0
  36   __pm_runtime_resume+0x7a/0x90
  37   intel_runtime_pm_get+0x25/0x90 [i915]
  38   aliasing_gtt_bind_vma+0xaa/0xf0 [i915]
  39   i915_vma_bind+0xaf/0x1e0 [i915]
  40   i915_gem_execbuffer_relocate_entry+0x513/0x6f0 [i915]
  41   i915_gem_execbuffer_relocate_vma.isra.34+0x188/0x250 [i915]
  42   ? trace_hardirqs_on+0xd/0x10
  43   ? i915_gem_execbuffer_reserve_vma.isra.31+0x152/0x1f0 [i915]
  44   ? i915_gem_execbuffer_reserve.isra.32+0x372/0x3a0 [i915]
  45   i915_gem_do_execbuffer.isra.38+0xa70/0x1a40 [i915]
  46   ? __might_fault+0x4e/0xb0
  47   i915_gem_execbuffer2+0xc5/0x260 [i915]
  48   ? __might_fault+0x4e/0xb0
  49   drm_ioctl+0x206/0x450 [drm]
  50   ? i915_gem_execbuffer+0x340/0x340 [i915]
  51   ? __fget+0x5/0x200
  52   do_vfs_ioctl+0x91/0x6f0
  53   ? __fget+0x111/0x200
  54   ? __fget+0x5/0x200
  55   SyS_ioctl+0x79/0x90
  56   entry_SYSCALL_64_fastpath+0x23/0xc6
  57
  58 even though the code triggering it is correct.
  59
  60 Unfortunately, the might_sleep_if() assertions in question are
  61 too coarse-grained to cover such cases correctly, so make them
  62 a bit less sensitive in order to avoid the false-positives.
  63
  64 Reported-and-tested-by: Sedat Dilek <sedat.dilek@gmail.com>
  65 Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
  66 Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
  67 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  68 ---
  69  drivers/base/power/runtime.c |   11 ++++++-----
  70  1 file changed, 6 insertions(+), 5 deletions(-)
  71
  72 --- a/drivers/base/power/runtime.c
  73 +++ b/drivers/base/power/runtime.c
  74 @@ -889,13 +889,13 @@ int __pm_runtime_idle(struct device *dev
  75         unsigned long flags;
  76         int retval;
  77
  78 -       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
  79 -
  80         if (rpmflags & RPM_GET_PUT) {
  81                 if (!atomic_dec_and_test(&dev->power.usage_count))
  82                         return 0;
  83         }
  84
  85 +       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
  86 +
  87         spin_lock_irqsave(&dev->power.lock, flags);
  88         retval = rpm_idle(dev, rpmflags);
  89         spin_unlock_irqrestore(&dev->power.lock, flags);
  90 @@ -921,13 +921,13 @@ int __pm_runtime_suspend(struct device *
  91         unsigned long flags;
  92         int retval;
  93
  94 -       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
  95 -
  96         if (rpmflags & RPM_GET_PUT) {
  97                 if (!atomic_dec_and_test(&dev->power.usage_count))
  98                         return 0;
  99         }
 100
 101 +       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
 102 +
 103         spin_lock_irqsave(&dev->power.lock, flags);
 104         retval = rpm_suspend(dev, rpmflags);
 105         spin_unlock_irqrestore(&dev->power.lock, flags);
 106 @@ -952,7 +952,8 @@ int __pm_runtime_resume(struct device *d
 107         unsigned long flags;
 108         int retval;
 109
 110 -       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
 111 +       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
 112 +                       dev->power.runtime_status != RPM_ACTIVE);
 113
 114         if (rpmflags & RPM_GET_PUT)
 115                 atomic_inc(&dev->power.usage_count);