drm/mgag200: fix kernel hang in cursor code.
authorWang, Rui Y <rui.y.wang@intel.com>
Wed, 18 Nov 2015 15:00:53 +0000 (23:00 +0800)
committerDave Airlie <airlied@redhat.com>
Thu, 19 Nov 2015 03:20:01 +0000 (13:20 +1000)
The machine hang completely with the following message on the console:

[  487.777538] BUG: unable to handle kernel NULL pointer dereference at 0000000000000060
[  487.777554] IP: [<ffffffff8158aaee>] _raw_spin_lock+0xe/0x30
[  487.777557] PGD 42e9f7067 PUD 42f2fa067 PMD 0
[  487.777560] Oops: 0002 [#1] SMP
...
[  487.777618] CPU: 21 PID: 3190 Comm: Xorg Tainted: G            E   4.4.0-rc1-3-default+ #6
[  487.777620] Hardware name: Intel Corporation BRICKLAND/BRICKLAND, BIOS BRHSXSD1.86B.0059.R00.1501081238 01/08/2015
[  487.777621] task: ffff880853ae4680 ti: ffff8808696d4000 task.ti: ffff8808696d4000
[  487.777625] RIP: 0010:[<ffffffff8158aaee>]  [<ffffffff8158aaee>] _raw_spin_lock+0xe/0x30
[  487.777627] RSP: 0018:ffff8808696d79c0  EFLAGS: 00010246
[  487.777628] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
[  487.777629] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000000000000060
[  487.777630] RBP: ffff8808696d79e0 R08: 0000000000000000 R09: ffff88086924a780
[  487.777631] R10: 000000000001bb40 R11: 0000000000003246 R12: 0000000000000000
[  487.777632] R13: ffff880463a27360 R14: ffff88046ca50218 R15: 0000000000000080
[  487.777634] FS:  00007f3f81c5a8c0(0000) GS:ffff88086f060000(0000) knlGS:0000000000000000
[  487.777635] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  487.777636] CR2: 0000000000000060 CR3: 000000042e678000 CR4: 00000000001406e0
[  487.777638] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  487.777639] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  487.777639] Stack:
[  487.777642]  ffffffffa00eb5fa ffff8808696d7b60 ffff88086b87d800 0000000000000000
[  487.777644]  ffff8808696d7ac8 ffffffffa01694b6 ffff8808696d7ae8 ffffffff8109c8d5
[  487.777647]  ffff880469158740 ffff880463a27000 ffff88086b87d800 ffff88086b87d800
[  487.777647] Call Trace:
[  487.777674]  [<ffffffffa00eb5fa>] ? drm_gem_object_lookup+0x1a/0xa0 [drm]
[  487.777681]  [<ffffffffa01694b6>] mga_crtc_cursor_set+0xc6/0xb60 [mgag200]
[  487.777691]  [<ffffffff8109c8d5>] ? find_busiest_group+0x35/0x4a0
[  487.777696]  [<ffffffff81086294>] ? __might_sleep+0x44/0x80
[  487.777699]  [<ffffffff815888c2>] ? __ww_mutex_lock+0x22/0x9c
[  487.777722]  [<ffffffffa0104f64>] ? drm_modeset_lock+0x34/0xf0 [drm]
[  487.777733]  [<ffffffffa0148d9e>] restore_fbdev_mode+0xee/0x2a0 [drm_kms_helper]
[  487.777742]  [<ffffffffa014afce>] drm_fb_helper_restore_fbdev_mode_unlocked+0x2e/0x70 [drm_kms_helper]
[  487.777748]  [<ffffffffa014b037>] drm_fb_helper_set_par+0x27/0x50 [drm_kms_helper]
[  487.777752]  [<ffffffff8134560c>] fb_set_var+0x18c/0x3f0
[  487.777777]  [<ffffffffa02a9b0a>] ? __ext4_handle_dirty_metadata+0x8a/0x210 [ext4]
[  487.777783]  [<ffffffff8133cb97>] fbcon_blank+0x1b7/0x2b0
[  487.777790]  [<ffffffff813be2a3>] do_unblank_screen+0xb3/0x1c0
[  487.777795]  [<ffffffff813b5aba>] vt_ioctl+0x118a/0x1210
[  487.777801]  [<ffffffff813a8fe0>] tty_ioctl+0x3f0/0xc90
[  487.777808]  [<ffffffff81172018>] ? kzfree+0x28/0x30
[  487.777813]  [<ffffffff811e053f>] ? mntput+0x1f/0x30
[  487.777817]  [<ffffffff811d3f5d>] do_vfs_ioctl+0x30d/0x570
[  487.777822]  [<ffffffff8107ed3a>] ? task_work_run+0x8a/0xa0
[  487.777825]  [<ffffffff811d4234>] SyS_ioctl+0x74/0x80
[  487.777829]  [<ffffffff8158aeae>] entry_SYSCALL_64_fastpath+0x12/0x71
[  487.777851] Code: 65 ff 0d ce 02 a8 7e 5d c3 ba 01 00 00 00 f0 0f b1 17 85 c0 75 e8 b0 01 5d c3 0f 1f 00 65 ff 05 b1 02 a8 7e 31 c0 ba 01 00 00 00 <f0> 0f b1 17 85 c0 75 01 c3 55 89 c6 48 89 e5 e8 4e f5 b1 ff 5d
[  487.777854] RIP  [<ffffffff8158aaee>] _raw_spin_lock+0xe/0x30
[  487.777855]  RSP <ffff8808696d79c0>
[  487.777856] CR2: 0000000000000060
[  487.777860] ---[ end trace 672a2cd555e0ebd3 ]---

The cursor code may be entered with file_priv == NULL && handle == NULL.
The problem was introduced by:

"bf89209 drm/mga200g: Hold a proper reference for cursor_set"

which calls drm_gem_object_lookup(dev, file_priv...). Previously this wasn't
a problem because we checked the handle. Move the check early in the function
can fix the problem.

Signed-off-by: Rui Wang <rui.y.wang@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/mgag200/mgag200_cursor.c

index 4f2068fe5d885cdba438a44745c27ccade5df9d0..a7bf6a90eae5a30e40ae3c1c5dbada0ea893bc8b 100644 (file)
@@ -70,6 +70,11 @@ int mga_crtc_cursor_set(struct drm_crtc *crtc,
        BUG_ON(pixels_2 != pixels_current && pixels_2 != pixels_prev);
        BUG_ON(pixels_current == pixels_prev);
 
+       if (!handle || !file_priv) {
+               mga_hide_cursor(mdev);
+               return 0;
+       }
+
        obj = drm_gem_object_lookup(dev, file_priv, handle);
        if (!obj)
                return -ENOENT;
@@ -88,12 +93,6 @@ int mga_crtc_cursor_set(struct drm_crtc *crtc,
                goto out_unreserve1;
        }
 
-       if (!handle) {
-               mga_hide_cursor(mdev);
-               ret = 0;
-               goto out1;
-       }
-
        /* Move cursor buffers into VRAM if they aren't already */
        if (!pixels_1->pin_count) {
                ret = mgag200_bo_pin(pixels_1, TTM_PL_FLAG_VRAM,