drm/i915: Fix NULL deref when re-enabling HPD IRQs on systems with MST
authorLyude Paul <lyude@redhat.com>
Tue, 6 Nov 2018 21:30:13 +0000 (16:30 -0500)
committerLyude Paul <lyude@redhat.com>
Wed, 7 Nov 2018 20:10:58 +0000 (15:10 -0500)
commitfee61deecb1d850bf34f682a6a452e5ee51b7572
tree83acd1833f7bfce5e3a08fc0d5fd192fe1caca60
parent66a5ab1034be801630816d1fa6cfc30db1a2f0b0
drm/i915: Fix NULL deref when re-enabling HPD IRQs on systems with MST

Turns out that if you trigger an HPD storm on a system that has an MST
topology connected to it, you'll end up causing the kernel to eventually
hit a NULL deref:

[  332.339041] BUG: unable to handle kernel NULL pointer dereference at 00000000000000ec
[  332.340906] PGD 0 P4D 0
[  332.342750] Oops: 0000 [#1] SMP PTI
[  332.344579] CPU: 2 PID: 25 Comm: kworker/2:0 Kdump: loaded Tainted: G           O      4.18.0-rc3short-hpd-storm+ #2
[  332.346453] Hardware name: LENOVO 20BWS1KY00/20BWS1KY00, BIOS JBET71WW (1.35 ) 09/14/2018
[  332.348361] Workqueue: events intel_hpd_irq_storm_reenable_work [i915]
[  332.350301] RIP: 0010:intel_hpd_irq_storm_reenable_work.cold.3+0x2f/0x86 [i915]
[  332.352213] Code: 00 00 ba e8 00 00 00 48 c7 c6 c0 aa 5f a0 48 c7 c7 d0 73 62 a0 4c 89 c1 4c 89 04 24 e8 7f f5 af e0 4c 8b 04 24 44 89 f8 29 e8 <41> 39 80 ec 00 00 00 0f 85 43 13 fc ff 41 0f b6 86 b8 04 00 00 41
[  332.354286] RSP: 0018:ffffc90000147e48 EFLAGS: 00010006
[  332.356344] RAX: 0000000000000005 RBX: ffff8802c226c9d4 RCX: 0000000000000006
[  332.358404] RDX: 0000000000000000 RSI: 0000000000000082 RDI: ffff88032dc95570
[  332.360466] RBP: 0000000000000005 R08: 0000000000000000 R09: ffff88031b3dc840
[  332.362528] R10: 0000000000000000 R11: 000000031a069602 R12: ffff8802c226ca20
[  332.364575] R13: ffff8802c2268000 R14: ffff880310661000 R15: 000000000000000a
[  332.366615] FS:  0000000000000000(0000) GS:ffff88032dc80000(0000) knlGS:0000000000000000
[  332.368658] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  332.370690] CR2: 00000000000000ec CR3: 000000000200a003 CR4: 00000000003606e0
[  332.372724] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  332.374773] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  332.376798] Call Trace:
[  332.378809]  process_one_work+0x1a1/0x350
[  332.380806]  worker_thread+0x30/0x380
[  332.382777]  ? wq_update_unbound_numa+0x10/0x10
[  332.384772]  kthread+0x112/0x130
[  332.386740]  ? kthread_create_worker_on_cpu+0x70/0x70
[  332.388706]  ret_from_fork+0x35/0x40
[  332.390651] Modules linked in: i915(O) vfat fat joydev btusb btrtl btbcm btintel bluetooth ecdh_generic iTCO_wdt wmi_bmof i2c_algo_bit drm_kms_helper intel_rapl syscopyarea sysfillrect x86_pkg_temp_thermal sysimgblt coretemp fb_sys_fops crc32_pclmul drm psmouse pcspkr mei_me mei i2c_i801 lpc_ich mfd_core i2c_core tpm_tis tpm_tis_core thinkpad_acpi wmi tpm rfkill video crc32c_intel serio_raw ehci_pci xhci_pci ehci_hcd xhci_hcd [last unloaded: i915]
[  332.394963] CR2: 00000000000000ec

This appears to be due to the fact that with an MST topology, not all
intel_connector structs will have ->encoder set. So, fix this by
skipping connectors without encoders in
intel_hpd_irq_storm_reenable_work().

For those wondering, this bug was found on accident while simulating HPD
storms using a Chamelium connected to a ThinkPad T450s (Broadwell).

Changes since v1:
- Check intel_connector->mst_port instead of intel_connector->encoder

Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: stable@vger.kernel.org
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181106213017.14563-3-lyude@redhat.com
drivers/gpu/drm/i915/intel_hotplug.c