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

   1 From f4359cedfb43b934f38c50d1604db21333abe57b Mon Sep 17 00:00:00 2001
   2 From: Hans de Goede <j.w.r.degoede@gmail.com>
   3 Date: Fri, 10 Nov 2017 16:03:01 +0100
   4 Subject: drm/i915: Fix false-positive assert_rpm_wakelock_held in i915_pmic_bus_access_notifier v2
   5
   6 From: Hans de Goede <j.w.r.degoede@gmail.com>
   7
   8 commit f4359cedfb43b934f38c50d1604db21333abe57b upstream.
   9
  10 assert_rpm_wakelock_held is triggered from i915_pmic_bus_access_notifier
  11 even though it gets unregistered on (runtime) suspend, this is caused
  12 by a race happening under the following circumstances:
  13
  14 intel_runtime_pm_put does:
  15
  16    atomic_dec(&dev_priv->pm.wakeref_count);
  17
  18    pm_runtime_mark_last_busy(kdev);
  19    pm_runtime_put_autosuspend(kdev);
  20
  21 And pm_runtime_put_autosuspend calls intel_runtime_suspend from
  22 a workqueue, so there is ample of time between the atomic_dec() and
  23 intel_runtime_suspend() unregistering the notifier. If the notifier
  24 gets called in this windowd assert_rpm_wakelock_held falsely triggers
  25 (at this point we're not runtime-suspended yet).
  26
  27 This commit adds disable_rpm_wakeref_asserts and
  28 enable_rpm_wakeref_asserts calls around the
  29 intel_uncore_forcewake_get(FORCEWAKE_ALL) call in
  30 i915_pmic_bus_access_notifier fixing the false-positive WARN_ON.
  31
  32 Changes in v2:
  33 -Reword comment explaining why disabling the wakeref asserts is
  34  ok and necessary
  35
  36 Reported-by: FKr <bugs-freedesktop@ubermail.me>
  37 Reviewed-by: Imre Deak <imre.deak@intel.com>
  38 Signed-off-by: Hans de Goede <hdegoede@redhat.com>
  39 Link: https://patchwork.freedesktop.org/patch/msgid/20171110150301.9601-2-hdegoede@redhat.com
  40 (cherry picked from commit ce30560c80dead91e98a03d90fb8791e57a9b69d)
  41 Signed-off-by: Jani Nikula <jani.nikula@intel.com>
  42 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  43
  44 ---
  45  drivers/gpu/drm/i915/intel_uncore.c |    7 +++++++
  46  1 file changed, 7 insertions(+)
  47
  48 --- a/drivers/gpu/drm/i915/intel_uncore.c
  49 +++ b/drivers/gpu/drm/i915/intel_uncore.c
  50 @@ -1171,8 +1171,15 @@ static int i915_pmic_bus_access_notifier
  51                  * bus, which will be busy after this notification, leading to:
  52                  * "render: timed out waiting for forcewake ack request."
  53                  * errors.
  54 +                *
  55 +                * The notifier is unregistered during intel_runtime_suspend(),
  56 +                * so it's ok to access the HW here without holding a RPM
  57 +                * wake reference -> disable wakeref asserts for the time of
  58 +                * the access.
  59                  */
  60 +               disable_rpm_wakeref_asserts(dev_priv);
  61                 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  62 +               enable_rpm_wakeref_asserts(dev_priv);
  63                 break;
  64         case MBI_PMIC_BUS_ACCESS_END:
  65                 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);