drm/amdgpu: add codes to capture invalid hardware access when recovery

When recovery thread has begun GPU reset, there should be not other
threads to access hardware, otherwise system randomly hang.

v2 (chk): rewritten from scratch, use trylock and lockdep instead of
hand wiring the logic.

v3: add in_irq check

v4: change to check in_task
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -1390,6 +1390,8 @@ void amdgpu_pci_resume(struct pci_dev *pdev);
 bool amdgpu_device_cache_pci_state(struct pci_dev *pdev);
 bool amdgpu_device_load_pci_state(struct pci_dev *pdev);
 
+bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev);
+
 #include "amdgpu_object.h"
 
 static inline bool amdgpu_is_tmz(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -326,6 +326,35 @@ void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos,
 /*
  * register access helper functions.
  */
+
+/* Check if hw access should be skipped because of hotplug or device error */
+bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev)
+{
+	if (adev->in_pci_err_recovery)
+		return true;
+
+#ifdef CONFIG_LOCKDEP
+	/*
+	 * This is a bit complicated to understand, so worth a comment. What we assert
+	 * here is that the GPU reset is not running on another thread in parallel.
+	 *
+	 * For this we trylock the read side of the reset semaphore, if that succeeds
+	 * we know that the reset is not running in paralell.
+	 *
+	 * If the trylock fails we assert that we are either already holding the read
+	 * side of the lock or are the reset thread itself and hold the write side of
+	 * the lock.
+	 */
+	if (in_task()) {
+		if (down_read_trylock(&adev->reset_sem))
+			up_read(&adev->reset_sem);
+		else
+			lockdep_assert_held(&adev->reset_sem);
+	}
+#endif
+	return false;
+}
+
 /**
  * amdgpu_device_rreg - read a memory mapped IO or indirect register
  *
@@ -340,7 +369,7 @@ uint32_t amdgpu_device_rreg(struct amdgpu_device *adev,
 {
 	uint32_t ret;
 
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return 0;
 
 	if ((reg * 4) < adev->rmmio_size) {
@@ -377,7 +406,7 @@ uint32_t amdgpu_device_rreg(struct amdgpu_device *adev,
  */
 uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return 0;
 
 	if (offset < adev->rmmio_size)
@@ -402,7 +431,7 @@ uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset)
  */
 void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	if (offset < adev->rmmio_size)
@@ -425,7 +454,7 @@ void amdgpu_device_wreg(struct amdgpu_device *adev,
 			uint32_t reg, uint32_t v,
 			uint32_t acc_flags)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	if ((reg * 4) < adev->rmmio_size) {
@@ -452,7 +481,7 @@ void amdgpu_device_wreg(struct amdgpu_device *adev,
 void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev,
 			     uint32_t reg, uint32_t v)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	if (amdgpu_sriov_fullaccess(adev) &&
@@ -476,7 +505,7 @@ void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev,
  */
 u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return 0;
 
 	if (index < adev->doorbell.num_doorbells) {
@@ -499,7 +528,7 @@ u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
  */
 void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	if (index < adev->doorbell.num_doorbells) {
@@ -520,7 +549,7 @@ void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
  */
 u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return 0;
 
 	if (index < adev->doorbell.num_doorbells) {
@@ -543,7 +572,7 @@ u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
  */
 void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
 {
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	if (index < adev->doorbell.num_doorbells) {

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c

@@ -705,7 +705,7 @@ uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg)
 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
 	struct amdgpu_ring *ring = &kiq->ring;
 
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return 0;
 
 	BUG_ON(!ring->funcs->emit_rreg);
@@ -772,7 +772,7 @@ void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v)
 
 	BUG_ON(!ring->funcs->emit_wreg);
 
-	if (adev->in_pci_err_recovery)
+	if (amdgpu_device_skip_hw_access(adev))
 		return;
 
 	spin_lock_irqsave(&kiq->ring_lock, flags);