[PATCH 2/2] drm/amdgpu/virt: schedule work to handle vm fault for VF

[Pixel.Ding@xxxxxxx (Pixel Ding)]

VF uses KIQ to access registers that invoking fence_wait to get the
accessing completed. When VM fault occurs, the driver can't sleep in
interrupt context.

For some test cases, VM fault is 'legal' and shouldn't cause driver soft
lockup.

Signed-off-by: Pixel Ding <Pixel.Ding at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 46 ++++++++++++++++++++++++++++++++---
 1 file changed, 43 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index 7669b32..75c913f 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -1231,9 +1231,9 @@ static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
 	return 0;
 }
 
-static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
-				      struct amdgpu_irq_src *source,
-				      struct amdgpu_iv_entry *entry)
+static int gmc_v8_0_process_vm_fault(struct amdgpu_device *adev,
+				     struct amdgpu_irq_src *source,
+				     struct amdgpu_iv_entry *entry)
 {
 	u32 addr, status, mc_client;
 
@@ -1262,6 +1262,46 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
 	return 0;
 }
 
+struct gmc_vm_fault_work {
+	struct work_struct      base;
+	struct amdgpu_device    *adev;
+	struct amdgpu_irq_src   *source;
+	struct amdgpu_iv_entry  *entry;
+};
+
+static void gmc_v8_0_vm_fault_sched(struct work_struct *work)
+{
+	struct gmc_vm_fault_work *vm_work =
+		container_of(work, struct gmc_vm_fault_work, base);
+	struct amdgpu_device *adev = vm_work->adev;
+	struct amdgpu_irq_src *source = vm_work->source;
+	struct amdgpu_iv_entry *entry = vm_work->entry;
+
+	gmc_v8_0_process_vm_fault(adev, source, entry);
+
+	kfree(vm_work);
+}
+
+static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
+				      struct amdgpu_irq_src *source,
+				      struct amdgpu_iv_entry *entry)
+{
+	struct gmc_vm_fault_work *work = NULL;
+
+	if (amdgpu_sriov_vf(adev)) {
+		work = kmalloc(sizeof(struct gmc_vm_fault_work), GFP_ATOMIC);
+		if (!work)
+			return -ENOMEM;
+		INIT_WORK(&work->base, gmc_v8_0_vm_fault_sched);
+		work->adev = adev;
+		work->source = source;
+		work->entry = entry;
+		return schedule_work(&work->base);
+	}
+
+	return gmc_v8_0_process_vm_fault(adev, source, entry);
+}
+
 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
 						     bool enable)
 {
-- 
2.7.4