drm/nouveau: Avoid looping through fake MST connectors
authorLyude Paul <lyude@redhat.com>
Fri, 13 Jul 2018 17:06:33 +0000 (13:06 -0400)
committerBen Skeggs <bskeggs@redhat.com>
Mon, 16 Jul 2018 07:59:59 +0000 (17:59 +1000)
commit37afe55b4ae0600deafe7c0e0e658593c4754f1b
tree88645004a61e493769ef15aa39d3840faaeca6e9
parent22b76bbe089cd901f5260ecb9a3dc41f9edb97a0
drm/nouveau: Avoid looping through fake MST connectors

When MST and atomic were introduced to nouveau, another structure that
could contain a drm_connector embedded within it was introduced; struct
nv50_mstc. This meant that we no longer would be able to simply loop
through our connector list and assume that nouveau_connector() would
return a proper pointer for each connector, since the assertion that
all connectors coming from nouveau have a full nouveau_connector struct
became invalid.

Unfortunately, none of the actual code that looped through connectors
ever got updated, which means that we've been causing invalid memory
accesses for quite a while now.

An example that was caught by KASAN:

[  201.038698] ==================================================================
[  201.038792] BUG: KASAN: slab-out-of-bounds in nvif_notify_get+0x190/0x1a0 [nouveau]
[  201.038797] Read of size 4 at addr ffff88076738c650 by task kworker/0:3/718
[  201.038800]
[  201.038822] CPU: 0 PID: 718 Comm: kworker/0:3 Tainted: G           O      4.18.0-rc4Lyude-Test+ #1
[  201.038825] Hardware name: LENOVO 20EQS64N0B/20EQS64N0B, BIOS N1EET78W (1.51 ) 05/18/2018
[  201.038882] Workqueue: events nouveau_display_hpd_work [nouveau]
[  201.038887] Call Trace:
[  201.038894]  dump_stack+0xa4/0xfd
[  201.038900]  print_address_description+0x71/0x239
[  201.038929]  ? nvif_notify_get+0x190/0x1a0 [nouveau]
[  201.038935]  kasan_report.cold.6+0x242/0x2fe
[  201.038942]  __asan_report_load4_noabort+0x19/0x20
[  201.038970]  nvif_notify_get+0x190/0x1a0 [nouveau]
[  201.038998]  ? nvif_notify_put+0x1f0/0x1f0 [nouveau]
[  201.039003]  ? kmsg_dump_rewind_nolock+0xe4/0xe4
[  201.039049]  nouveau_display_init.cold.12+0x34/0x39 [nouveau]
[  201.039089]  ? nouveau_user_framebuffer_create+0x120/0x120 [nouveau]
[  201.039133]  nouveau_display_resume+0x5c0/0x810 [nouveau]
[  201.039173]  ? nvkm_client_ioctl+0x20/0x20 [nouveau]
[  201.039215]  nouveau_do_resume+0x19f/0x570 [nouveau]
[  201.039256]  nouveau_pmops_runtime_resume+0xd8/0x2a0 [nouveau]
[  201.039264]  pci_pm_runtime_resume+0x130/0x250
[  201.039269]  ? pci_restore_standard_config+0x70/0x70
[  201.039275]  __rpm_callback+0x1f2/0x5d0
[  201.039279]  ? rpm_resume+0x560/0x18a0
[  201.039283]  ? pci_restore_standard_config+0x70/0x70
[  201.039287]  ? pci_restore_standard_config+0x70/0x70
[  201.039291]  ? pci_restore_standard_config+0x70/0x70
[  201.039296]  rpm_callback+0x175/0x210
[  201.039300]  ? pci_restore_standard_config+0x70/0x70
[  201.039305]  rpm_resume+0xcc3/0x18a0
[  201.039312]  ? rpm_callback+0x210/0x210
[  201.039317]  ? __pm_runtime_resume+0x9e/0x100
[  201.039322]  ? kasan_check_write+0x14/0x20
[  201.039326]  ? do_raw_spin_lock+0xc2/0x1c0
[  201.039333]  __pm_runtime_resume+0xac/0x100
[  201.039374]  nouveau_display_hpd_work+0x67/0x1f0 [nouveau]
[  201.039380]  process_one_work+0x7a0/0x14d0
[  201.039388]  ? cancel_delayed_work_sync+0x20/0x20
[  201.039392]  ? lock_acquire+0x113/0x310
[  201.039398]  ? kasan_check_write+0x14/0x20
[  201.039402]  ? do_raw_spin_lock+0xc2/0x1c0
[  201.039409]  worker_thread+0x86/0xb50
[  201.039418]  kthread+0x2e9/0x3a0
[  201.039422]  ? process_one_work+0x14d0/0x14d0
[  201.039426]  ? kthread_create_worker_on_cpu+0xc0/0xc0
[  201.039431]  ret_from_fork+0x3a/0x50
[  201.039441]
[  201.039444] Allocated by task 79:
[  201.039449]  save_stack+0x43/0xd0
[  201.039452]  kasan_kmalloc+0xc4/0xe0
[  201.039456]  kmem_cache_alloc_trace+0x10a/0x260
[  201.039494]  nv50_mstm_add_connector+0x9a/0x340 [nouveau]
[  201.039504]  drm_dp_add_port+0xff5/0x1fc0 [drm_kms_helper]
[  201.039511]  drm_dp_send_link_address+0x4a7/0x740 [drm_kms_helper]
[  201.039518]  drm_dp_check_and_send_link_address+0x1a7/0x210 [drm_kms_helper]
[  201.039525]  drm_dp_mst_link_probe_work+0x71/0xb0 [drm_kms_helper]
[  201.039529]  process_one_work+0x7a0/0x14d0
[  201.039533]  worker_thread+0x86/0xb50
[  201.039537]  kthread+0x2e9/0x3a0
[  201.039541]  ret_from_fork+0x3a/0x50
[  201.039543]
[  201.039546] Freed by task 0:
[  201.039549] (stack is not available)
[  201.039551]
[  201.039555] The buggy address belongs to the object at ffff88076738c1a8
                                 which belongs to the cache kmalloc-2048 of size 2048
[  201.039559] The buggy address is located 1192 bytes inside of
                                 2048-byte region [ffff88076738c1a8ffff88076738c9a8)
[  201.039563] The buggy address belongs to the page:
[  201.039567] page:ffffea001d9ce200 count:1 mapcount:0 mapping:ffff88084000d0c0 index:0x0 compound_mapcount: 0
[  201.039573] flags: 0x8000000000008100(slab|head)
[  201.039578] raw: 8000000000008100 ffffea001da3be08 ffffea001da25a08 ffff88084000d0c0
[  201.039582] raw: 0000000000000000 00000000000d000d 00000001ffffffff 0000000000000000
[  201.039585] page dumped because: kasan: bad access detected
[  201.039588]
[  201.039591] Memory state around the buggy address:
[  201.039594]  ffff88076738c500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[  201.039598]  ffff88076738c580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[  201.039601] >ffff88076738c600: 00 00 00 00 00 00 00 00 00 00 fc fc fc fc fc fc
[  201.039604]                                                  ^
[  201.039607]  ffff88076738c680: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[  201.039611]  ffff88076738c700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[  201.039613] ==================================================================

Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: stable@vger.kernel.org
Cc: Karol Herbst <karolherbst@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/nouveau_connector.c
drivers/gpu/drm/nouveau/nouveau_connector.h
drivers/gpu/drm/nouveau/nouveau_display.c