releases/4.19.51/vfio-fix-warning-do-not-call-blocking-ops-when-task_.patch

   1 From fbfc05061c443e73e5ad8d08c6c5e7416f94ff04 Mon Sep 17 00:00:00 2001
   2 From: Farhan Ali <alifm@linux.ibm.com>
   3 Date: Wed, 3 Apr 2019 14:22:27 -0400
   4 Subject: vfio: Fix WARNING "do not call blocking ops when !TASK_RUNNING"
   5
   6 [ Upstream commit 41be3e2618174fdf3361e49e64f2bf530f40c6b0 ]
   7
   8 vfio_dev_present() which is the condition to
   9 wait_event_interruptible_timeout(), will call vfio_group_get_device
  10 and try to acquire the mutex group->device_lock.
  11
  12 wait_event_interruptible_timeout() will set the state of the current
  13 task to TASK_INTERRUPTIBLE, before doing the condition check. This
  14 means that we will try to acquire the mutex while already in a
  15 sleeping state. The scheduler warns us by giving the following
  16 warning:
  17
  18 [ 4050.264464] ------------[ cut here ]------------
  19 [ 4050.264508] do not call blocking ops when !TASK_RUNNING; state=1 set at [<00000000b33c00e2>] prepare_to_wait_event+0x14a/0x188
  20 [ 4050.264529] WARNING: CPU: 12 PID: 35924 at kernel/sched/core.c:6112 __might_sleep+0x76/0x90
  21 ....
  22
  23  4050.264756] Call Trace:
  24 [ 4050.264765] ([<000000000017bbaa>] __might_sleep+0x72/0x90)
  25 [ 4050.264774]  [<0000000000b97edc>] __mutex_lock+0x44/0x8c0
  26 [ 4050.264782]  [<0000000000b9878a>] mutex_lock_nested+0x32/0x40
  27 [ 4050.264793]  [<000003ff800d7abe>] vfio_group_get_device+0x36/0xa8 [vfio]
  28 [ 4050.264803]  [<000003ff800d87c0>] vfio_del_group_dev+0x238/0x378 [vfio]
  29 [ 4050.264813]  [<000003ff8015f67c>] mdev_remove+0x3c/0x68 [mdev]
  30 [ 4050.264825]  [<00000000008e01b0>] device_release_driver_internal+0x168/0x268
  31 [ 4050.264834]  [<00000000008de692>] bus_remove_device+0x162/0x190
  32 [ 4050.264843]  [<00000000008daf42>] device_del+0x1e2/0x368
  33 [ 4050.264851]  [<00000000008db12c>] device_unregister+0x64/0x88
  34 [ 4050.264862]  [<000003ff8015ed84>] mdev_device_remove+0xec/0x130 [mdev]
  35 [ 4050.264872]  [<000003ff8015f074>] remove_store+0x6c/0xa8 [mdev]
  36 [ 4050.264881]  [<000000000046f494>] kernfs_fop_write+0x14c/0x1f8
  37 [ 4050.264890]  [<00000000003c1530>] __vfs_write+0x38/0x1a8
  38 [ 4050.264899]  [<00000000003c187c>] vfs_write+0xb4/0x198
  39 [ 4050.264908]  [<00000000003c1af2>] ksys_write+0x5a/0xb0
  40 [ 4050.264916]  [<0000000000b9e270>] system_call+0xdc/0x2d8
  41 [ 4050.264925] 4 locks held by sh/35924:
  42 [ 4050.264933]  #0: 000000001ef90325 (sb_writers#4){.+.+}, at: vfs_write+0x9e/0x198
  43 [ 4050.264948]  #1: 000000005c1ab0b3 (&of->mutex){+.+.}, at: kernfs_fop_write+0x1cc/0x1f8
  44 [ 4050.264963]  #2: 0000000034831ab8 (kn->count#297){++++}, at: kernfs_remove_self+0x12e/0x150
  45 [ 4050.264979]  #3: 00000000e152484f (&dev->mutex){....}, at: device_release_driver_internal+0x5c/0x268
  46 [ 4050.264993] Last Breaking-Event-Address:
  47 [ 4050.265002]  [<000000000017bbaa>] __might_sleep+0x72/0x90
  48 [ 4050.265010] irq event stamp: 7039
  49 [ 4050.265020] hardirqs last  enabled at (7047): [<00000000001cee7a>] console_unlock+0x6d2/0x740
  50 [ 4050.265029] hardirqs last disabled at (7054): [<00000000001ce87e>] console_unlock+0xd6/0x740
  51 [ 4050.265040] softirqs last  enabled at (6416): [<0000000000b8fe26>] __udelay+0xb6/0x100
  52 [ 4050.265049] softirqs last disabled at (6415): [<0000000000b8fe06>] __udelay+0x96/0x100
  53 [ 4050.265057] ---[ end trace d04a07d39d99a9f9 ]---
  54
  55 Let's fix this as described in the article
  56 https://lwn.net/Articles/628628/.
  57
  58 Signed-off-by: Farhan Ali <alifm@linux.ibm.com>
  59 [remove now redundant vfio_dev_present()]
  60 Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
  61 Signed-off-by: Sasha Levin <sashal@kernel.org>
  62 ---
  63  drivers/vfio/vfio.c | 30 ++++++++++--------------------
  64  1 file changed, 10 insertions(+), 20 deletions(-)
  65
  66 diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
  67 index 64833879f75d..7a386fb30bf1 100644
  68 --- a/drivers/vfio/vfio.c
  69 +++ b/drivers/vfio/vfio.c
  70 @@ -34,6 +34,7 @@
  71  #include <linux/uaccess.h>
  72  #include <linux/vfio.h>
  73  #include <linux/wait.h>
  74 +#include <linux/sched/signal.h>
  75
  76  #define DRIVER_VERSION "0.3"
  77  #define DRIVER_AUTHOR  "Alex Williamson <alex.williamson@redhat.com>"
  78 @@ -904,30 +905,17 @@ void *vfio_device_data(struct vfio_device *device)
  79  }
  80  EXPORT_SYMBOL_GPL(vfio_device_data);
  81
  82 -/* Given a referenced group, check if it contains the device */
  83 -static bool vfio_dev_present(struct vfio_group *group, struct device *dev)
  84 -{
  85 -       struct vfio_device *device;
  86 -
  87 -       device = vfio_group_get_device(group, dev);
  88 -       if (!device)
  89 -               return false;
  90 -
  91 -       vfio_device_put(device);
  92 -       return true;
  93 -}
  94 -
  95  /*
  96   * Decrement the device reference count and wait for the device to be
  97   * removed.  Open file descriptors for the device... */
  98  void *vfio_del_group_dev(struct device *dev)
  99  {
 100 +       DEFINE_WAIT_FUNC(wait, woken_wake_function);
 101         struct vfio_device *device = dev_get_drvdata(dev);
 102         struct vfio_group *group = device->group;
 103         void *device_data = device->device_data;
 104         struct vfio_unbound_dev *unbound;
 105         unsigned int i = 0;
 106 -       long ret;
 107         bool interrupted = false;
 108
 109         /*
 110 @@ -964,6 +952,8 @@ void *vfio_del_group_dev(struct device *dev)
 111          * interval with counter to allow the driver to take escalating
 112          * measures to release the device if it has the ability to do so.
 113          */
 114 +       add_wait_queue(&vfio.release_q, &wait);
 115 +
 116         do {
 117                 device = vfio_group_get_device(group, dev);
 118                 if (!device)
 119 @@ -975,12 +965,10 @@ void *vfio_del_group_dev(struct device *dev)
 120                 vfio_device_put(device);
 121
 122                 if (interrupted) {
 123 -                       ret = wait_event_timeout(vfio.release_q,
 124 -                                       !vfio_dev_present(group, dev), HZ * 10);
 125 +                       wait_woken(&wait, TASK_UNINTERRUPTIBLE, HZ * 10);
 126                 } else {
 127 -                       ret = wait_event_interruptible_timeout(vfio.release_q,
 128 -                                       !vfio_dev_present(group, dev), HZ * 10);
 129 -                       if (ret == -ERESTARTSYS) {
 130 +                       wait_woken(&wait, TASK_INTERRUPTIBLE, HZ * 10);
 131 +                       if (signal_pending(current)) {
 132                                 interrupted = true;
 133                                 dev_warn(dev,
 134                                          "Device is currently in use, task"
 135 @@ -989,8 +977,10 @@ void *vfio_del_group_dev(struct device *dev)
 136                                          current->comm, task_pid_nr(current));
 137                         }
 138                 }
 139 -       } while (ret <= 0);
 140
 141 +       } while (1);
 142 +
 143 +       remove_wait_queue(&vfio.release_q, &wait);
 144         /*
 145          * In order to support multiple devices per group, devices can be
 146          * plucked from the group while other devices in the group are still
 147 --
 148 2.20.1
 149