drm/i915: Hold the table lock whilst walking the file's idr and counting the objects...
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 17 Jun 2014 08:56:24 +0000 (09:56 +0100)
committerJani Nikula <jani.nikula@intel.com>
Tue, 24 Jun 2014 12:00:48 +0000 (15:00 +0300)
Fixes an issue whereby we may race with the table updates (before the
core takes the struct_mutex) and so risk dereferencing a stale pointer in
the iterator for /debugfs/.../i915_gem_objects. For example,

[ 1524.757545] BUG: unable to handle kernel paging request at f53af748
[ 1524.757572] IP: [<c1406982>] per_file_stats+0x12/0x100
[ 1524.757599] *pdpt = 0000000001b13001 *pde = 00000000379fb067 *pte = 80000000353af060
[ 1524.757621] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
[ 1524.757637] Modules linked in: ctr ccm arc4 ath9k ath9k_common ath9k_hw ath snd_hda_codec_conexant mac80211 snd_hda_codec_generic snd_hda_intel snd_hda_controller snd_hda_codec bnep snd_hwdep rfcomm snd_pcm gpio_ich dell_wmi sparse_keymap snd_seq_midi hid_multitouch uvcvideo snd_seq_midi_event dell_laptop snd_rawmidi dcdbas snd_seq videobuf2_vmalloc videobuf2_memops videobuf2_core usbhid videodev snd_seq_device coretemp snd_timer hid joydev kvm_intel cfg80211 ath3k kvm btusb bluetooth serio_raw snd microcode soundcore lpc_ich wmi mac_hid parport_pc ppdev lp parport psmouse ahci libahci
[ 1524.757825] CPU: 3 PID: 1911 Comm: intel-gpu-overl Tainted: G        W  OE 3.15.0-rc3+ #96
[ 1524.757840] Hardware name: Dell Inc. Inspiron 1090/Inspiron 1090, BIOS A06 08/23/2011
[ 1524.757855] task: f52f36c0 ti: f4cbc000 task.ti: f4cbc000
[ 1524.757869] EIP: 0060:[<c1406982>] EFLAGS: 00210202 CPU: 3
[ 1524.757884] EIP is at per_file_stats+0x12/0x100
[ 1524.757896] EAX: 0000002d EBX: 00000000 ECX: f4cbdefc EDX: f53af700
[ 1524.757909] ESI: c1406970 EDI: f53af700 EBP: f4cbde6c ESP: f4cbde5c
[ 1524.757922]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
[ 1524.757934] CR0: 80050033 CR2: f53af748 CR3: 356af000 CR4: 000007f0
[ 1524.757945] Stack:
[ 1524.757957]  f4cbdefc 00000000 c1406970 f53af700 f4cbdea8 c12e5f15 f4cbdefc c1406970
[ 1524.757993]  0000ffff f4cbde90 0000002d f5dc5cd0 e4e80438 c1181d59 f4cbded8 f4d89900
[ 1524.758027]  f5631b40 e5131074 c1903f37 f4cbdf28 c14068e6 f52648a0 c1927748 c1903f37
[ 1524.758062] Call Trace:
[ 1524.758084]  [<c1406970>] ? i915_gem_object_info+0x510/0x510
[ 1524.758106]  [<c12e5f15>] idr_for_each+0xa5/0x100
[ 1524.758126]  [<c1406970>] ? i915_gem_object_info+0x510/0x510
[ 1524.758148]  [<c1181d59>] ? seq_vprintf+0x29/0x50
[ 1524.758168]  [<c14068e6>] i915_gem_object_info+0x486/0x510
[ 1524.758189]  [<c11823a6>] seq_read+0xd6/0x380
[ 1524.758208]  [<c116d11d>] ? final_putname+0x1d/0x40
[ 1524.758227]  [<c11822d0>] ? seq_hlist_next_percpu+0x90/0x90
[ 1524.758246]  [<c1163e52>] vfs_read+0x82/0x150
[ 1524.758265]  [<c11645d6>] SyS_read+0x46/0x90
[ 1524.758285]  [<c16b8d8c>] sysenter_do_call+0x12/0x22
[ 1524.758298] Code: f5 8f 2a 00 83 c4 6c 31 c0 5b 5e 5f 5d c3 8d 74 26 00 8d bc 27 00 00 00 00 55 89 e5 57 56 53 83 ec 04 3e 8d 74 26 00 83 41 04 01 <8b> 42 48 01 41 08 8b 42 4c 89 d7 85 c0 75 07 8b 42 60 85 c0 74
[ 1524.758461] EIP: [<c1406982>] per_file_stats+0x12/0x100 SS:ESP 0068:f4cbde5c
[ 1524.758485] CR2: 00000000f53af748

Reported-by: Sam Jansen <sam.jansen@starleaf.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Sam Jansen <sam.jansen@starleaf.com>
Cc: stable@vger.kernel.org
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
drivers/gpu/drm/i915/i915_debugfs.c

index 601caa88c0928cb818c6af27cbb9f25c0a4bebf2..b8c689202c4041c4e22dc63d7415ffb2232e94a9 100644 (file)
@@ -446,7 +446,9 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 
                memset(&stats, 0, sizeof(stats));
                stats.file_priv = file->driver_priv;
+               spin_lock(&file->table_lock);
                idr_for_each(&file->object_idr, per_file_stats, &stats);
+               spin_unlock(&file->table_lock);
                /*
                 * Although we have a valid reference on file->pid, that does
                 * not guarantee that the task_struct who called get_pid() is