Skip to contentSkip to conversation

drm/amdgpu: check if vram is lost

PATCH In review Chunming Zhou 2 participants 6 messages May 16, 2017 3 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/3 drm/amdgpu: check if vram is lost

bakup first 64 byte of gart table as reset magic, check if magic is same after gpu hw reset.

Change-Id I9a73720da4084ea8677c3031dfb62e8157ee5704
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 26 +++++++++++++++++++++++++-
 2 files changed, 27 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index de08ff0..f9da215 100644
@@ -1502,6 +1502,7 @@ struct amdgpu_ssg {
 #endif
 };
 
+#define AMDGPU_RESET_MAGIC_NUM 64
 struct amdgpu_device {
 	struct device			*dev;
 	struct drm_device		*ddev;
@@ -1705,6 +1706,7 @@ struct amdgpu_device {
 
 	/* record hw reset is performed */
 	bool has_hw_reset;
+	u8				reset_magic[AMDGPU_RESET_MAGIC_NUM];
 
 };
 
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +25 -1
index 0a31fb1..58af9ea 100644
@@ -1685,6 +1685,23 @@ static int amdgpu_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+static void amdgpu_fill_reset_magic(struct amdgpu_device *adev)
+{
+	memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);
+}
+
+static bool amdgpu_check_vram_lost(struct amdgpu_device *adev)
+{
+	int i;
+
+	for (i = 0; i < AMDGPU_RESET_MAGIC_NUM; i++) {
+		if (*((u8 *)(adev->gart.ptr + i)) != adev->reset_magic[i])
+			return true;
+	}
+
+	return false;
+}
+
 static int amdgpu_late_init(struct amdgpu_device *adev)
 {
 	int i = 0, r;
@@ -1715,6 +1732,8 @@ static int amdgpu_late_init(struct amdgpu_device *adev)
 		}
 	}
 
+	amdgpu_fill_reset_magic(adev);
+
 	return 0;
 }
 
@@ -2830,7 +2849,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 	struct drm_atomic_state *state = NULL;
 	int i, r;
 	int resched;
-	bool need_full_reset;
+	bool need_full_reset, vram_lost = false;
 
 	if (amdgpu_sriov_vf(adev))
 		return amdgpu_sriov_gpu_reset(adev, true);
@@ -2899,12 +2918,17 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 			r = amdgpu_resume_phase1(adev);
 			if (r)
 				goto out;
+			vram_lost = amdgpu_check_vram_lost(adev);
+			if (vram_lost)
+				DRM_ERROR("VRAM is lost!\n");
 			r = amdgpu_ttm_recover_gart(adev);
 			if (r)
 				goto out;
 			r = amdgpu_resume_phase2(adev);
 			if (r)
 				goto out;
+			if (vram_lost)
+				amdgpu_fill_reset_magic(adev);
 		}
 	}
 out:
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1494917069-29054-1-git-send-email-David1.Zhou-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 2/3 drm/amdgpu: return -ENODEV to user space when vram is lost

below ioctl will return -ENODEV:
amdgpu_cs_ioctl
amdgpu_cs_wait_ioctl
amdgpu_cs_wait_fences_ioctl
amdgpu_gem_va_ioctl
amdgpu_info_ioctl

Change-Id I8970cde3301b7cfeb4263cc0f0e54aece215c98e
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |  9 +++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    | 10 ++++++++++
 5 files changed, 28 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +4 -0
index f9da215..dcd6203 100644
@@ -855,6 +855,7 @@ struct amdgpu_fpriv {
 	struct amdgpu_ctx_mgr	ctx_mgr;
 	spinlock_t		sem_handles_lock;
 	struct idr		sem_handles;
+	u32			vram_lost_counter;
 };
 
 /*
@@ -1607,6 +1608,7 @@ struct amdgpu_device {
 	atomic64_t			num_bytes_moved;
 	atomic64_t			num_evictions;
 	atomic_t			gpu_reset_counter;
+	atomic_t			vram_lost_counter;
 
 	/* data for buffer migration throttling */
 	struct {
@@ -2005,6 +2007,8 @@ static inline void amdgpu_unregister_atpx_handler(void) {}
 extern const struct drm_ioctl_desc amdgpu_ioctls_kms[];
 extern const int amdgpu_max_kms_ioctl;
 
+bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+			  struct amdgpu_fpriv *fpriv);
 int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags);
 int amdgpu_driver_unload_kms(struct drm_device *dev);
 void amdgpu_driver_lastclose_kms(struct drm_device *dev);
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +9 -0
index b803412..911aa02 100644
@@ -1097,6 +1097,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	union drm_amdgpu_cs *cs = data;
 	struct amdgpu_cs_parser parser = {};
 	bool reserved_buffers = false;
@@ -1104,6 +1105,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
 	if (!adev->accel_working)
 		return -EBUSY;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 
 	parser.adev = adev;
 	parser.filp = filp;
@@ -1165,12 +1168,15 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
 {
 	union drm_amdgpu_wait_cs *wait = data;
 	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
 	struct amdgpu_ring *ring = NULL;
 	struct amdgpu_ctx *ctx;
 	struct fence *fence;
 	long r;
 
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 	r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
 			       wait->in.ring, &ring);
 	if (r)
@@ -1344,12 +1350,15 @@ int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	union drm_amdgpu_wait_fences *wait = data;
 	uint32_t fence_count = wait->in.fence_count;
 	struct drm_amdgpu_fence *fences_user;
 	struct drm_amdgpu_fence *fences;
 	int r;
 
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 	/* Get the fences from userspace */
 	fences = kmalloc_array(fence_count, sizeof(struct drm_amdgpu_fence),
 			GFP_KERNEL);
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -1
index 58af9ea..417b8f8 100644
@@ -2919,8 +2919,10 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 			if (r)
 				goto out;
 			vram_lost = amdgpu_check_vram_lost(adev);
-			if (vram_lost)
+			if (vram_lost) {
 				DRM_ERROR("VRAM is lost!\n");
+				atomic_inc(&adev->vram_lost_counter);
+			}
 			r = amdgpu_ttm_recover_gart(adev);
 			if (r)
 				goto out;
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +2 -0
index d8275ef..0f0b736 100644
@@ -776,6 +776,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 
 	if (!adev->vm_manager.enabled)
 		return -ENOTTY;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 
 	if (args->va_address < AMDGPU_VA_RESERVED_SIZE) {
 		dev_err(&dev->pdev->dev,
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +10 -0
index 368829a..a231aa1 100644
@@ -235,6 +235,7 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	struct drm_amdgpu_info *info = data;
 	struct amdgpu_mode_info *minfo = &adev->mode_info;
 	void __user *out = (void __user *)(uintptr_t)info->return_pointer;
@@ -247,6 +248,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 
 	if (!info->return_size || !info->return_pointer)
 		return -EINVAL;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 
 	switch (info->query) {
 	case AMDGPU_INFO_VIRTUAL_RANGE: {
@@ -779,6 +782,12 @@ void amdgpu_driver_lastclose_kms(struct drm_device *dev)
 	vga_switcheroo_process_delayed_switch();
 }
 
+bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+			  struct amdgpu_fpriv *fpriv)
+{
+	return fpriv->vram_lost_counter != atomic_read(&adev->vram_lost_counter);
+}
+
 /**
  * amdgpu_driver_open_kms - drm callback for open
  *
@@ -833,6 +842,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 
 	amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
 
+	fpriv->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 	file_priv->driver_priv = fpriv;
 
 out_suspend:
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1494917069-29054-2-git-send-email-David1.Zhou-5C7GfCeVMHo@public.gmane.org

#4 lore raw

Re: [PATCH 2/3] drm/amdgpu: return -ENODEV to user space when vram is lost

Am 16.05.2017 um 08:44 schrieb Chunming Zhou:

Quoted 120 lines

below ioctl will return -ENODEV:
amdgpu_cs_ioctl
amdgpu_cs_wait_ioctl
amdgpu_cs_wait_fences_ioctl
amdgpu_gem_va_ioctl
amdgpu_info_ioctl

Change-Id I8970cde3301b7cfeb4263cc0f0e54aece215c98e
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  4 ++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |  9 +++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++-
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c    |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    | 10 ++++++++++
  5 files changed, 28 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +4 -0
index f9da215..dcd6203 100644
@@ -855,6 +855,7 @@ struct amdgpu_fpriv {
  	struct amdgpu_ctx_mgr	ctx_mgr;
  	spinlock_t		sem_handles_lock;
  	struct idr		sem_handles;
+	u32			vram_lost_counter;
  };
 
  /*
@@ -1607,6 +1608,7 @@ struct amdgpu_device {
  	atomic64_t			num_bytes_moved;
  	atomic64_t			num_evictions;
  	atomic_t			gpu_reset_counter;
+	atomic_t			vram_lost_counter;
 
  	/* data for buffer migration throttling */
  	struct {
@@ -2005,6 +2007,8 @@ static inline void amdgpu_unregister_atpx_handler(void) {}
  extern const struct drm_ioctl_desc amdgpu_ioctls_kms[];
  extern const int amdgpu_max_kms_ioctl;
 
+bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+			  struct amdgpu_fpriv *fpriv);
  int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags);
  int amdgpu_driver_unload_kms(struct drm_device *dev);
  void amdgpu_driver_lastclose_kms(struct drm_device *dev);
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +9 -0
index b803412..911aa02 100644
@@ -1097,6 +1097,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
  int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  {
  	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
  	union drm_amdgpu_cs *cs = data;
  	struct amdgpu_cs_parser parser = {};
  	bool reserved_buffers = false;
@@ -1104,6 +1105,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
  	if (!adev->accel_working)
  		return -EBUSY;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 
  	parser.adev = adev;
  	parser.filp = filp;
@@ -1165,12 +1168,15 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
  {
  	union drm_amdgpu_wait_cs *wait = data;
  	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
  	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
  	struct amdgpu_ring *ring = NULL;
  	struct amdgpu_ctx *ctx;
  	struct fence *fence;
  	long r;
 
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
  	r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
  			       wait->in.ring, &ring);
  	if (r)
@@ -1344,12 +1350,15 @@ int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data,
  				struct drm_file *filp)
  {
  	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
  	union drm_amdgpu_wait_fences *wait = data;
  	uint32_t fence_count = wait->in.fence_count;
  	struct drm_amdgpu_fence *fences_user;
  	struct drm_amdgpu_fence *fences;
  	int r;
 
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
  	/* Get the fences from userspace */
  	fences = kmalloc_array(fence_count, sizeof(struct drm_amdgpu_fence),
  			GFP_KERNEL);
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -1
index 58af9ea..417b8f8 100644
@@ -2919,8 +2919,10 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  			if (r)
  				goto out;
  			vram_lost = amdgpu_check_vram_lost(adev);
-			if (vram_lost)
+			if (vram_lost) {
  				DRM_ERROR("VRAM is lost!\n");
+				atomic_inc(&adev->vram_lost_counter);
+			}
  			r = amdgpu_ttm_recover_gart(adev);
  			if (r)
  				goto out;
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +0 -0
index d8275ef..0f0b736 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, · line 780

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +2 -0
@@ -776,6 +776,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 
  	if (!adev->vm_manager.enabled)
  		return -ENOTTY;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;

We should only block AMDGPU_VA_OP_MAP and AMDGPU_VA_OP_REPLACE here and still allow AMDGPU_VA_OP_UNMAP and AMDGPU_VA_OP_CLEAR.

BTW: How should the UMD recover from that situation? Completely close the fd and recreate it?

That might be tricky for processes like X or the Compositor. Should we have an IOCTL to reset the vram_lost counter for an fd?

Christian.

Quoted 38 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +0 -0
  	if (args->va_address < AMDGPU_VA_RESERVED_SIZE) {
  		dev_err(&dev->pdev->dev,
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +9 -0
index 368829a..a231aa1 100644
@@ -235,6 +235,7 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
  static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  {
  	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
  	struct drm_amdgpu_info *info = data;
  	struct amdgpu_mode_info *minfo = &adev->mode_info;
  	void __user *out = (void __user *)(uintptr_t)info->return_pointer;
@@ -247,6 +248,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 
  	if (!info->return_size || !info->return_pointer)
  		return -EINVAL;
+	if (amdgpu_kms_vram_lost(adev, fpriv))
+		return -ENODEV;
 
  	switch (info->query) {
  	case AMDGPU_INFO_VIRTUAL_RANGE: {
@@ -779,6 +782,12 @@ void amdgpu_driver_lastclose_kms(struct drm_device *dev)
  	vga_switcheroo_process_delayed_switch();
  }
 
+bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+			  struct amdgpu_fpriv *fpriv)
+{
+	return fpriv->vram_lost_counter != atomic_read(&adev->vram_lost_counter);
+}
+
  /**
   * amdgpu_driver_open_kms - drm callback for open
   *

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv) · line 848

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +1 -0
@@ -833,6 +842,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 
  	amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
 
+	fpriv->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
  	file_priv->driver_priv = fpriv;
 
  out_suspend:

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 0383ac20-28f1-0ec3-97f9-7268328d619b-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org

author
#6 lore raw

Re: [PATCH 2/3] drm/amdgpu: return -ENODEV to user space when vram is lost

Quoted 131 lines · Christian König

Am 16.05.2017 um 08:44 schrieb Chunming Zhou:

Quoted 122 lines

below ioctl will return -ENODEV:
amdgpu_cs_ioctl
amdgpu_cs_wait_ioctl
amdgpu_cs_wait_fences_ioctl
amdgpu_gem_va_ioctl
amdgpu_info_ioctl

Change-Id I8970cde3301b7cfeb4263cc0f0e54aece215c98e
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  4 ++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |  9 +++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++-
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c    |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    | 10 ++++++++++
  5 files changed, 28 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +3 -0
index f9da215..dcd6203 100644
@@ -855,6 +855,7 @@ struct amdgpu_fpriv {
      struct amdgpu_ctx_mgr    ctx_mgr;
      spinlock_t        sem_handles_lock;
      struct idr        sem_handles;
+    u32            vram_lost_counter;
  };
    /*
@@ -1607,6 +1608,7 @@ struct amdgpu_device {
      atomic64_t            num_bytes_moved;
      atomic64_t            num_evictions;
      atomic_t            gpu_reset_counter;
+    atomic_t            vram_lost_counter;
        /* data for buffer migration throttling */
      struct {
@@ -2005,6 +2007,8 @@ static inline void 
 amdgpu_unregister_atpx_handler(void) {}
  extern const struct drm_ioctl_desc amdgpu_ioctls_kms[];
  extern const int amdgpu_max_kms_ioctl;
  +bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+              struct amdgpu_fpriv *fpriv);
  int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long 
 flags);
  int amdgpu_driver_unload_kms(struct drm_device *dev);
  void amdgpu_driver_lastclose_kms(struct drm_device *dev);
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +1 -0
index b803412..911aa02 100644
@@ -1097,6 +1097,7 @@ static int amdgpu_cs_submit(struct 
 amdgpu_cs_parser *p,
  int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct 
 drm_file *filp)
  {
      struct amdgpu_device *adev = dev->dev_private;
+    struct amdgpu_fpriv *fpriv = filp->driver_priv;
      union drm_amdgpu_cs *cs = data;
      struct amdgpu_cs_parser parser = {};
      bool reserved_buffers = false;
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +6 -0
@@ -1104,6 +1105,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, 
 void *data, struct drm_file *filp)
        if (!adev->accel_working)
          return -EBUSY;
+    if (amdgpu_kms_vram_lost(adev, fpriv))
+        return -ENODEV;
        parser.adev = adev;
      parser.filp = filp;
@@ -1165,12 +1168,15 @@ int amdgpu_cs_wait_ioctl(struct drm_device 
 *dev, void *data,
  {
      union drm_amdgpu_wait_cs *wait = data;
      struct amdgpu_device *adev = dev->dev_private;
+    struct amdgpu_fpriv *fpriv = filp->driver_priv;
      unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
      struct amdgpu_ring *ring = NULL;
      struct amdgpu_ctx *ctx;
      struct fence *fence;
      long r;
  +    if (amdgpu_kms_vram_lost(adev, fpriv))
+        return -ENODEV;
      r = amdgpu_cs_get_ring(adev, wait->in.ip_type, 
 wait->in.ip_instance,
                     wait->in.ring, &ring);
      if (r)
@@ -1344,12 +1350,15 @@ int amdgpu_cs_wait_fences_ioctl(struct 
 drm_device *dev, void *data,
                  struct drm_file *filp)
  {
      struct amdgpu_device *adev = dev->dev_private;
+    struct amdgpu_fpriv *fpriv = filp->driver_priv;
      union drm_amdgpu_wait_fences *wait = data;
      uint32_t fence_count = wait->in.fence_count;
      struct drm_amdgpu_fence *fences_user;
      struct drm_amdgpu_fence *fences;
      int r;
  +    if (amdgpu_kms_vram_lost(adev, fpriv))
+        return -ENODEV;
      /* Get the fences from userspace */
      fences = kmalloc_array(fence_count, sizeof(struct 

drm_amdgpu_fence),

              GFP_KERNEL);
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -1
index 58af9ea..417b8f8 100644
@@ -2919,8 +2919,10 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
              if (r)
                  goto out;
              vram_lost = amdgpu_check_vram_lost(adev);
-            if (vram_lost)
+            if (vram_lost) {
                  DRM_ERROR("VRAM is lost!\n");
+                atomic_inc(&adev->vram_lost_counter);
+            }
              r = amdgpu_ttm_recover_gart(adev);
              if (r)
                  goto out;
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +0 -0
index d8275ef..0f0b736 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_va_ioctl(struct drm_device *dev, · line 780

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +2 -0
@@ -776,6 +776,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, 
 void *data,
        if (!adev->vm_manager.enabled)
          return -ENOTTY;
+    if (amdgpu_kms_vram_lost(adev, fpriv))
+        return -ENODEV;

We should only block AMDGPU_VA_OP_MAP and AMDGPU_VA_OP_REPLACE here and still allow AMDGPU_VA_OP_UNMAP and AMDGPU_VA_OP_CLEAR.

Quoted 3 lines

BTW: How should the UMD recover from that situation? Completely close the fd and recreate it?

That might be tricky for processes like X or the Compositor. Should we have an IOCTL to reset the vram_lost counter for an fd?

if no NAK, I can add it.

Regards,
David Zhou

Quoted 39 lines

Christian.

Quoted 36 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +0 -0
        if (args->va_address < AMDGPU_VA_RESERVED_SIZE) {
          dev_err(&dev->pdev->dev,
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +1 -0
index 368829a..a231aa1 100644
@@ -235,6 +235,7 @@ static int amdgpu_firmware_info(struct 
 drm_amdgpu_info_firmware *fw_info,
  static int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
 struct drm_file *filp)
  {
      struct amdgpu_device *adev = dev->dev_private;
+    struct amdgpu_fpriv *fpriv = filp->driver_priv;
      struct drm_amdgpu_info *info = data;
      struct amdgpu_mode_info *minfo = &adev->mode_info;
      void __user *out = (void __user *)(uintptr_t)info->return_pointer;
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +7 -0
@@ -247,6 +248,8 @@ static int amdgpu_info_ioctl(struct drm_device 
 *dev, void *data, struct drm_file
        if (!info->return_size || !info->return_pointer)
          return -EINVAL;
+    if (amdgpu_kms_vram_lost(adev, fpriv))
+        return -ENODEV;
        switch (info->query) {
      case AMDGPU_INFO_VIRTUAL_RANGE: {
@@ -779,6 +782,12 @@ void amdgpu_driver_lastclose_kms(struct 
 drm_device *dev)
      vga_switcheroo_process_delayed_switch();
  }
  +bool amdgpu_kms_vram_lost(struct amdgpu_device *adev,
+              struct amdgpu_fpriv *fpriv)
+{
+    return fpriv->vram_lost_counter != 
 atomic_read(&adev->vram_lost_counter);
+}
+
  /**
Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c int amdgpu_driver_open_kms(struct drm_device · line 846

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +0 -0
   * amdgpu_driver_open_kms - drm callback for open
   *
@@ -833,6 +842,7 @@ int amdgpu_driver_open_kms(struct drm_device 
 *dev, struct drm_file *file_priv)
        amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
  +    fpriv->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
      file_priv->driver_priv = fpriv;
    out_suspend:

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#7 lore raw

PATCH 3/3 drm/amdgpu: skip all jobs of guilty vm

In reply to a message not in the archive
Change-Id Ib305282a21d37d6872afe4c1ce63d53b6517f338
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +11 -4
index 4a5ffa5..da2f87e1 100644
@@ -167,6 +167,7 @@ static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
 {
 	struct fence *fence = NULL;
 	struct amdgpu_job *job;
+	struct amdgpu_fpriv *fpriv = NULL;
 	int r;
 
 	if (!sched_job) {
@@ -178,10 +179,16 @@ static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
 	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
 
 	trace_amdgpu_sched_run_job(job);
-	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
-	if (r)
-		DRM_ERROR("Error scheduling IBs (%d)\n", r);
-
+	if (job->vm)
+		fpriv = container_of(job->vm, struct amdgpu_fpriv, vm);
+	/* skip ib schedule when vram is lost */
+	if (fpriv && amdgpu_kms_vram_lost(job->adev, fpriv))
+		DRM_ERROR("Skip scheduling IBs!\n");
+	else {
+		r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
+		if (r)
+			DRM_ERROR("Error scheduling IBs (%d)\n", r);
+	}
 	/* if gpu reset, hw fence will be replaced here */
 	fence_put(job->fence);
 	job->fence = fence_get(fence);
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#8 lore raw
In reply to a message not in the archive

Am 16.05.2017 um 08:44 schrieb Chunming Zhou:

Quoted 41 lines

bakup first 64 byte of gart table as reset magic, check if magic is same after gpu hw reset.

Change-Id I9a73720da4084ea8677c3031dfb62e8157ee5704
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 26 +++++++++++++++++++++++++-
  2 files changed, 27 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index de08ff0..f9da215 100644
@@ -1502,6 +1502,7 @@ struct amdgpu_ssg {
  #endif
  };
 
+#define AMDGPU_RESET_MAGIC_NUM 64
  struct amdgpu_device {
  	struct device			*dev;
  	struct drm_device		*ddev;
@@ -1705,6 +1706,7 @@ struct amdgpu_device {
 
  	/* record hw reset is performed */
  	bool has_hw_reset;
+	u8				reset_magic[AMDGPU_RESET_MAGIC_NUM];
 
  };
 
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -0
index 0a31fb1..58af9ea 100644
@@ -1685,6 +1685,23 @@ static int amdgpu_init(struct amdgpu_device *adev)
  	return 0;
  }
 
+static void amdgpu_fill_reset_magic(struct amdgpu_device *adev)
+{
+	memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c static int amdgpu_init(struct amdgpu_device *adev) · line 1700

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +10 -0
+}
+
+static bool amdgpu_check_vram_lost(struct amdgpu_device *adev)
+{
+	int i;
+
+	for (i = 0; i < AMDGPU_RESET_MAGIC_NUM; i++) {
+		if (*((u8 *)(adev->gart.ptr + i)) != adev->reset_magic[i])
+			return true;
+	}

How about using memcmp() here?

Apart from that the patch looks good to me.

Christian.

Quoted 33 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +10 -1
+
+	return false;
+}
+
  static int amdgpu_late_init(struct amdgpu_device *adev)
  {
  	int i = 0, r;
@@ -1715,6 +1732,8 @@ static int amdgpu_late_init(struct amdgpu_device *adev)
  		}
  	}
 
+	amdgpu_fill_reset_magic(adev);
+
  	return 0;
  }
 
@@ -2830,7 +2849,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  	struct drm_atomic_state *state = NULL;
  	int i, r;
  	int resched;
-	bool need_full_reset;
+	bool need_full_reset, vram_lost = false;
 
  	if (amdgpu_sriov_vf(adev))
  		return amdgpu_sriov_gpu_reset(adev, true);
@@ -2899,12 +2918,17 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  			r = amdgpu_resume_phase1(adev);
  			if (r)
  				goto out;
+			vram_lost = amdgpu_check_vram_lost(adev);
+			if (vram_lost)
+				DRM_ERROR("VRAM is lost!\n");
  			r = amdgpu_ttm_recover_gart(adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c int amdgpu_gpu_reset(struct amdgpu_device *adev) · line 2934

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
  			if (r)
  				goto out;
  			r = amdgpu_resume_phase2(adev);
  			if (r)
  				goto out;
+			if (vram_lost)
+				amdgpu_fill_reset_magic(adev);
  		}
  	}
  out:

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx