[PATCH] drm/i915/gvt: Fix unsafe locking caused by spin_unlock_bh

[changbin.du at intel.com]

From: Changbin Du <changbin.du at intel.com>

The caller of shadow_context_status_change may disable irqs. So it is not
safe to use spin_unlock_bh in such context. Let's switch to irqsave version
for safety.

------------[ cut here ]------------
WARNING: CPU: 2 PID: 4504 at kernel/softirq.c:161 __local_bh_enable_ip+0x46/0x60
[  168.797710] Hardware name: Dell Inc. OptiPlex 7040/0Y7WYT, BIOS 1.2.8 01/26/2016
[  168.797712] task: ffff8c693d22db80 task.stack: ffffb51b482bc000
[  168.797718] RIP: 0010:__local_bh_enable_ip+0x46/0x60
[  168.797721] RSP: 0018:ffffb51b482bfa10 EFLAGS: 00010046
[  168.797724] RAX: 0000000000000046 RBX: ffff8c6900278000 RCX: 00000000ffffffff
[  168.797726] RDX: 0000000000000001 RSI: 0000000000000200 RDI: ffffffffc06a0330
[  168.797728] RBP: ffffb51b482bfa10 R08: 0000000000000000 R09: ffff8c690027cb90
[  168.797730] R10: ffffb51b482bfa40 R11: 00000004072f0001 R12: 0000000000000000
[  168.797732] R13: 0000000000000000 R14: ffff8c690027ca9c R15: 0000000000000000
[  168.797735] FS:  00007ff187c56700(0000) GS:ffff8c6959d00000(0000) knlGS:0000000000000000
[  168.797738] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  168.797740] CR2: 0000562bc0c3991f CR3: 0000000430614006 CR4: 00000000003606e0
[  168.797742] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  168.797744] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  168.797745] Call Trace:
[  168.797755]  _raw_spin_unlock_bh+0x1e/0x20
[  168.797826]  shadow_context_status_change+0x120/0x1e0 [i915]
[  168.797831]  notifier_call_chain+0x4a/0x70
[  168.797834]  atomic_notifier_call_chain+0x1a/0x20
[  168.797896]  execlists_cancel_port_requests+0x4f/0x80 [i915]
[  168.797956]  reset_common_ring+0x30/0x100 [i915]
[  168.798007]  i915_gem_reset_engine+0x114/0x330 [i915]
[  168.798060]  ? i915_gem_retire_requests+0x75/0x180 [i915]
[  168.798111]  i915_gem_reset+0x3e/0xb0 [i915]
[  168.798149]  i915_reset+0x10b/0x1c0 [i915]
[  168.798187]  i915_reset_device+0x209/0x220 [i915]
[  168.798225]  ? gen8_gt_irq_ack+0x170/0x170 [i915]
[  168.798229]  ? __queue_work+0x430/0x430
[  168.798270]  i915_handle_error+0x285/0x420 [i915]
[  168.798275]  ? mntput+0x24/0x40
[  168.798281]  ? terminate_walk+0x8e/0xf0
[  168.798328]  i915_wedged_set+0x84/0xc0 [i915]
[  168.798333]  simple_attr_write+0xab/0xc0
[  168.798337]  full_proxy_write+0x54/0x90
[  168.798343]  __vfs_write+0x37/0x170
[  168.798349]  ? common_file_perm+0x4c/0x100
[  168.798355]  ? apparmor_file_permission+0x1a/0x20
[  168.798361]  ? security_file_permission+0x3b/0xc0
[  168.798365]  vfs_write+0xb8/0x1b0
[  168.798370]  SyS_write+0x55/0xc0
[  168.798376]  entry_SYSCALL_64_fastpath+0x1e/0xa9

Fixes: 0e86cc9 ("drm/i915/gvt: implement per-vm mmio switching optimization")
Signed-off-by: Changbin Du <changbin.du at intel.com>
---
 drivers/gpu/drm/i915/gvt/scheduler.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 7b80e70..576d073 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -154,9 +154,10 @@ static int shadow_context_status_change(struct notifier_block *nb,
 	struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
 	enum intel_engine_id ring_id = req->engine->id;
 	struct intel_vgpu_workload *workload;
+	unsigned long flags;
 
 	if (!is_gvt_request(req)) {
-		spin_lock_bh(&scheduler->mmio_context_lock);
+		spin_lock_irqsave(&scheduler->mmio_context_lock, flags);
 		if (action == INTEL_CONTEXT_SCHEDULE_IN &&
 		    scheduler->engine_owner[ring_id]) {
 			/* Switch ring from vGPU to host. */
@@ -164,7 +165,7 @@ static int shadow_context_status_change(struct notifier_block *nb,
 					      NULL, ring_id);
 			scheduler->engine_owner[ring_id] = NULL;
 		}
-		spin_unlock_bh(&scheduler->mmio_context_lock);
+		spin_unlock_irqrestore(&scheduler->mmio_context_lock, flags);
 
 		return NOTIFY_OK;
 	}
@@ -175,7 +176,7 @@ static int shadow_context_status_change(struct notifier_block *nb,
 
 	switch (action) {
 	case INTEL_CONTEXT_SCHEDULE_IN:
-		spin_lock_bh(&scheduler->mmio_context_lock);
+		spin_lock_irqsave(&scheduler->mmio_context_lock, flags);
 		if (workload->vgpu != scheduler->engine_owner[ring_id]) {
 			/* Switch ring from host to vGPU or vGPU to vGPU. */
 			intel_gvt_switch_mmio(scheduler->engine_owner[ring_id],
@@ -184,7 +185,7 @@ static int shadow_context_status_change(struct notifier_block *nb,
 		} else
 			gvt_dbg_sched("skip ring %d mmio switch for vgpu%d\n",
 				      ring_id, workload->vgpu->id);
-		spin_unlock_bh(&scheduler->mmio_context_lock);
+		spin_unlock_irqrestore(&scheduler->mmio_context_lock, flags);
 		atomic_set(&workload->shadow_ctx_active, 1);
 		break;
 	case INTEL_CONTEXT_SCHEDULE_OUT:
-- 
2.7.4