From b56c22853eabde22a93d77bbd5250c1655405d74 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 29 Oct 2015 17:01:19 +0100
Subject: drm/amdgpu: use fence_is_later() for vm_flush as well v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

v2: remove superfluous check

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 633a32a48560..8c80381b1ea9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -207,24 +207,21 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
 	uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory);
 	struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
 	struct fence *flushed_updates = vm_id->flushed_updates;
-	bool is_earlier = false;
+	bool is_later;
 
-	if (flushed_updates && updates) {
-		BUG_ON(flushed_updates->context != updates->context);
-		is_earlier = (updates->seqno - flushed_updates->seqno <=
-			      INT_MAX) ? true : false;
-	}
-
-	if (pd_addr != vm_id->pd_gpu_addr || !flushed_updates ||
-	    is_earlier) {
+	if (!flushed_updates)
+		is_later = true;
+	else if (!updates)
+		is_later = false;
+	else
+		is_later = fence_is_later(updates, flushed_updates);
 
+	if (pd_addr != vm_id->pd_gpu_addr || is_later) {
 		trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id);
-		if (is_earlier) {
+		if (is_later) {
 			vm_id->flushed_updates = fence_get(updates);
 			fence_put(flushed_updates);
 		}
-		if (!flushed_updates)
-			vm_id->flushed_updates = fence_get(updates);
 		vm_id->pd_gpu_addr = pd_addr;
 		amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr);
 	}
-- 
cgit v1.2.3


From 16ae42feb0ce6cc80473d35dd82a1cd07e2e7178 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 3 Nov 2015 14:53:28 +0100
Subject: drm/amdgpu: use common fence for amdgpu_vm_fence
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Just cleanup the function parameters.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 12 ++++++------
 3 files changed, 8 insertions(+), 8 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 3630cb545d0d..b801b6702812 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -2332,7 +2332,7 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
 		     struct fence *updates);
 void amdgpu_vm_fence(struct amdgpu_device *adev,
 		     struct amdgpu_vm *vm,
-		     struct amdgpu_fence *fence);
+		     struct fence *fence);
 uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr);
 int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 				    struct amdgpu_vm *vm);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index e65987743871..3902e7a9f8cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -216,7 +216,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 	}
 
 	if (ib->vm)
-		amdgpu_vm_fence(adev, ib->vm, ib->fence);
+		amdgpu_vm_fence(adev, ib->vm, &ib->fence->base);
 
 	amdgpu_ring_unlock_commit(ring);
 	return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 8c80381b1ea9..a9fcc995d27e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -241,16 +241,16 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
  */
 void amdgpu_vm_fence(struct amdgpu_device *adev,
 		     struct amdgpu_vm *vm,
-		     struct amdgpu_fence *fence)
+		     struct fence *fence)
 {
-	unsigned ridx = fence->ring->idx;
-	unsigned vm_id = vm->ids[ridx].id;
+	struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence);
+	unsigned vm_id = vm->ids[ring->idx].id;
 
 	fence_put(adev->vm_manager.active[vm_id]);
-	adev->vm_manager.active[vm_id] = fence_get(&fence->base);
+	adev->vm_manager.active[vm_id] = fence_get(fence);
 
-	fence_put(vm->ids[ridx].last_id_use);
-	vm->ids[ridx].last_id_use = fence_get(&fence->base);
+	fence_put(vm->ids[ring->idx].last_id_use);
+	vm->ids[ring->idx].last_id_use = fence_get(fence);
 }
 
 /**
-- 
cgit v1.2.3


From ea89f8c9e8ba8a7b75446eef36917da50d2186d9 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sun, 15 Nov 2015 20:52:06 +0100
Subject: drm/amdgpu: move VM manager clean into the VM code again
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

It's not a good idea to duplicate that code.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <davdi1.zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  8 ++------
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  8 ++------
 4 files changed, 20 insertions(+), 12 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index d1e16726e225..beb74854a8a3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -972,6 +972,7 @@ struct amdgpu_vm_manager {
 	struct amdgpu_ring                      *vm_pte_funcs_ring;
 };
 
+void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
 int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a9fcc995d27e..6bb209bc0d36 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1318,3 +1318,18 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 
 	mutex_destroy(&vm->mutex);
 }
+
+/**
+ * amdgpu_vm_manager_fini - cleanup VM manager
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Cleanup the VM manager and free resources.
+ */
+void amdgpu_vm_manager_fini(struct amdgpu_device *adev)
+{
+	unsigned i;
+
+	for (i = 0; i < AMDGPU_NUM_VM; ++i)
+		fence_put(adev->vm_manager.active[i]);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
index 98ee39001c45..7427d8cd4c43 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
@@ -961,12 +961,10 @@ static int gmc_v7_0_sw_init(void *handle)
 
 static int gmc_v7_0_sw_fini(void *handle)
 {
-	int i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	if (adev->vm_manager.enabled) {
-		for (i = 0; i < AMDGPU_NUM_VM; ++i)
-			fence_put(adev->vm_manager.active[i]);
+		amdgpu_vm_manager_fini(adev);
 		gmc_v7_0_vm_fini(adev);
 		adev->vm_manager.enabled = false;
 	}
@@ -1011,12 +1009,10 @@ static int gmc_v7_0_hw_fini(void *handle)
 
 static int gmc_v7_0_suspend(void *handle)
 {
-	int i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	if (adev->vm_manager.enabled) {
-		for (i = 0; i < AMDGPU_NUM_VM; ++i)
-			fence_put(adev->vm_manager.active[i]);
+		amdgpu_vm_manager_fini(adev);
 		gmc_v7_0_vm_fini(adev);
 		adev->vm_manager.enabled = false;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index c9209b45d901..cb0e50ebb528 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -980,12 +980,10 @@ static int gmc_v8_0_sw_init(void *handle)
 
 static int gmc_v8_0_sw_fini(void *handle)
 {
-	int i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	if (adev->vm_manager.enabled) {
-		for (i = 0; i < AMDGPU_NUM_VM; ++i)
-			fence_put(adev->vm_manager.active[i]);
+		amdgpu_vm_manager_fini(adev);
 		gmc_v8_0_vm_fini(adev);
 		adev->vm_manager.enabled = false;
 	}
@@ -1032,12 +1030,10 @@ static int gmc_v8_0_hw_fini(void *handle)
 
 static int gmc_v8_0_suspend(void *handle)
 {
-	int i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	if (adev->vm_manager.enabled) {
-		for (i = 0; i < AMDGPU_NUM_VM; ++i)
-			fence_put(adev->vm_manager.active[i]);
+		amdgpu_vm_manager_fini(adev);
 		gmc_v8_0_vm_fini(adev);
 		adev->vm_manager.enabled = false;
 	}
-- 
cgit v1.2.3


From 1c16c0a7b26c6c905dc79c4194135ca2f360f0f5 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sat, 14 Nov 2015 21:31:40 +0100
Subject: drm/amdgpu: keep the owner for VMIDs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

We don't need the last VM use any more, keep the owner directly.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <davdi1.zhou@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  8 +++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 33 +++++++++++++++++++--------------
 2 files changed, 24 insertions(+), 17 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index beb74854a8a3..a5692624070a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -925,8 +925,6 @@ struct amdgpu_vm_id {
 	uint64_t		pd_gpu_addr;
 	/* last flushed PD/PT update */
 	struct fence	        *flushed_updates;
-	/* last use of vmid */
-	struct fence		*last_id_use;
 };
 
 struct amdgpu_vm {
@@ -959,7 +957,11 @@ struct amdgpu_vm {
 };
 
 struct amdgpu_vm_manager {
-	struct fence				*active[AMDGPU_NUM_VM];
+	struct {
+		struct fence	*active;
+		atomic_long_t	owner;
+	} ids[AMDGPU_NUM_VM];
+
 	uint32_t				max_pfn;
 	/* number of VMIDs */
 	unsigned				nvm;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 6bb209bc0d36..0bdbb2480f9b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -143,10 +143,15 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 	unsigned i;
 
 	/* check if the id is still valid */
-	if (vm_id->id && vm_id->last_id_use &&
-	    vm_id->last_id_use == adev->vm_manager.active[vm_id->id]) {
-		trace_amdgpu_vm_grab_id(vm_id->id, ring->idx);
-		return 0;
+	if (vm_id->id) {
+		unsigned id = vm_id->id;
+		long owner;
+
+		owner = atomic_long_read(&adev->vm_manager.ids[id].owner);
+		if (owner == (long)vm) {
+			trace_amdgpu_vm_grab_id(vm_id->id, ring->idx);
+			return 0;
+		}
 	}
 
 	/* we definately need to flush */
@@ -154,7 +159,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 
 	/* skip over VMID 0, since it is the system VM */
 	for (i = 1; i < adev->vm_manager.nvm; ++i) {
-		struct fence *fence = adev->vm_manager.active[i];
+		struct fence *fence = adev->vm_manager.ids[i].active;
 		struct amdgpu_ring *fring;
 
 		if (fence == NULL) {
@@ -176,7 +181,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		if (choices[i]) {
 			struct fence *fence;
 
-			fence  = adev->vm_manager.active[choices[i]];
+			fence  = adev->vm_manager.ids[choices[i]].active;
 			vm_id->id = choices[i];
 
 			trace_amdgpu_vm_grab_id(choices[i], ring->idx);
@@ -246,11 +251,9 @@ void amdgpu_vm_fence(struct amdgpu_device *adev,
 	struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence);
 	unsigned vm_id = vm->ids[ring->idx].id;
 
-	fence_put(adev->vm_manager.active[vm_id]);
-	adev->vm_manager.active[vm_id] = fence_get(fence);
-
-	fence_put(vm->ids[ring->idx].last_id_use);
-	vm->ids[ring->idx].last_id_use = fence_get(fence);
+	fence_put(adev->vm_manager.ids[vm_id].active);
+	adev->vm_manager.ids[vm_id].active = fence_get(fence);
+	atomic_long_set(&adev->vm_manager.ids[vm_id].owner, (long)vm);
 }
 
 /**
@@ -1238,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
 		vm->ids[i].id = 0;
 		vm->ids[i].flushed_updates = NULL;
-		vm->ids[i].last_id_use = NULL;
 	}
 	mutex_init(&vm->mutex);
 	vm->va = RB_ROOT;
@@ -1312,8 +1314,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	fence_put(vm->page_directory_fence);
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+		unsigned id = vm->ids[i].id;
+
+		atomic_long_cmpxchg(&adev->vm_manager.ids[id].owner,
+				    (long)vm, 0);
 		fence_put(vm->ids[i].flushed_updates);
-		fence_put(vm->ids[i].last_id_use);
 	}
 
 	mutex_destroy(&vm->mutex);
@@ -1331,5 +1336,5 @@ void amdgpu_vm_manager_fini(struct amdgpu_device *adev)
 	unsigned i;
 
 	for (i = 0; i < AMDGPU_NUM_VM; ++i)
-		fence_put(adev->vm_manager.active[i]);
+		fence_put(adev->vm_manager.ids[i].active);
 }
-- 
cgit v1.2.3


From c25867dfabf045a8148fd179fa759bb17f670e42 Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Fri, 13 Nov 2015 13:32:01 +0800
Subject: drm/amdgpu: add lock for interval tree in vm
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Change-Id: I62b892a22af37b32e6b4aefca80a25cf45426ed2
Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++--
 2 files changed, 15 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index a5692624070a..306f75700bf8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -954,6 +954,8 @@ struct amdgpu_vm {
 
 	/* for id and flush management per ring */
 	struct amdgpu_vm_id	ids[AMDGPU_MAX_RINGS];
+	/* for interval tree */
+	spinlock_t		it_lock;
 };
 
 struct amdgpu_vm_manager {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 0bdbb2480f9b..0513f3fed2c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1028,7 +1028,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
 	eaddr /= AMDGPU_GPU_PAGE_SIZE;
 
+	spin_lock(&vm->it_lock);
 	it = interval_tree_iter_first(&vm->va, saddr, eaddr - 1);
+	spin_unlock(&vm->it_lock);
 	if (it) {
 		struct amdgpu_bo_va_mapping *tmp;
 		tmp = container_of(it, struct amdgpu_bo_va_mapping, it);
@@ -1055,7 +1057,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	mapping->flags = flags;
 
 	list_add(&mapping->list, &bo_va->invalids);
+	spin_lock(&vm->it_lock);
 	interval_tree_insert(&mapping->it, &vm->va);
+	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_map(bo_va, mapping);
 
 	/* Make sure the page tables are allocated */
@@ -1101,7 +1105,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 
 error_free:
 	list_del(&mapping->list);
+	spin_lock(&vm->it_lock);
 	interval_tree_remove(&mapping->it, &vm->va);
+	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 	kfree(mapping);
 
@@ -1151,7 +1157,9 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	}
 
 	list_del(&mapping->list);
+	spin_lock(&vm->it_lock);
 	interval_tree_remove(&mapping->it, &vm->va);
+	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 
 	if (valid)
@@ -1187,13 +1195,17 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 
 	list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
 		list_del(&mapping->list);
+		spin_lock(&vm->it_lock);
 		interval_tree_remove(&mapping->it, &vm->va);
+		spin_unlock(&vm->it_lock);
 		trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 		list_add(&mapping->list, &vm->freed);
 	}
 	list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
 		list_del(&mapping->list);
+		spin_lock(&vm->it_lock);
 		interval_tree_remove(&mapping->it, &vm->va);
+		spin_unlock(&vm->it_lock);
 		kfree(mapping);
 	}
 
@@ -1248,7 +1260,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	INIT_LIST_HEAD(&vm->invalidated);
 	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
-
+	spin_lock_init(&vm->it_lock);
 	pd_size = amdgpu_vm_directory_size(adev);
 	pd_entries = amdgpu_vm_num_pdes(adev);
 
@@ -1312,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 
 	amdgpu_bo_unref(&vm->page_directory);
 	fence_put(vm->page_directory_fence);
-
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
 		unsigned id = vm->ids[i].id;
 
-- 
cgit v1.2.3


From ef9f0a83d68cecca241ee5d61b797d598722772e Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Fri, 13 Nov 2015 13:43:22 +0800
Subject: drm/amdgpu: move bo_reserve out of amdgpu_vm_clear_bo
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Change-Id: Ifbb0c06680494bfa04d0be5e5941d31ae2e5ef28
Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 33 ++++++++++++++++++---------------
 1 file changed, 18 insertions(+), 15 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 0513f3fed2c2..e6dc19bc2dd8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -332,6 +332,8 @@ int amdgpu_vm_free_job(struct amdgpu_job *job)
  *
  * @adev: amdgpu_device pointer
  * @bo: bo to clear
+ *
+ * need to reserve bo first before calling it.
  */
 static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 			      struct amdgpu_bo *bo)
@@ -343,24 +345,20 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	uint64_t addr;
 	int r;
 
-	r = amdgpu_bo_reserve(bo, false);
-	if (r)
-		return r;
-
 	r = reservation_object_reserve_shared(bo->tbo.resv);
 	if (r)
 		return r;
 
 	r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
 	if (r)
-		goto error_unreserve;
+		goto error;
 
 	addr = amdgpu_bo_gpu_offset(bo);
 	entries = amdgpu_bo_size(bo) / 8;
 
 	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
 	if (!ib)
-		goto error_unreserve;
+		goto error;
 
 	r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, ib);
 	if (r)
@@ -378,16 +376,14 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	if (!r)
 		amdgpu_bo_fence(bo, fence, true);
 	fence_put(fence);
-	if (amdgpu_enable_scheduler) {
-		amdgpu_bo_unreserve(bo);
+	if (amdgpu_enable_scheduler)
 		return 0;
-	}
+
 error_free:
 	amdgpu_ib_free(adev, ib);
 	kfree(ib);
 
-error_unreserve:
-	amdgpu_bo_unreserve(bo);
+error:
 	return r;
 }
 
@@ -1087,11 +1083,12 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 				     AMDGPU_GEM_DOMAIN_VRAM,
 				     AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
 				     NULL, resv, &pt);
-		ww_mutex_unlock(&resv->lock);
-		if (r)
+		if (r) {
+			ww_mutex_unlock(&resv->lock);
 			goto error_free;
-
+		}
 		r = amdgpu_vm_clear_bo(adev, pt);
+		ww_mutex_unlock(&resv->lock);
 		if (r) {
 			amdgpu_bo_unref(&pt);
 			goto error_free;
@@ -1280,8 +1277,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 			     NULL, NULL, &vm->page_directory);
 	if (r)
 		return r;
-
+	r = amdgpu_bo_reserve(vm->page_directory, false);
+	if (r) {
+		amdgpu_bo_unref(&vm->page_directory);
+		vm->page_directory = NULL;
+		return r;
+	}
 	r = amdgpu_vm_clear_bo(adev, vm->page_directory);
+	amdgpu_bo_unreserve(vm->page_directory);
 	if (r) {
 		amdgpu_bo_unref(&vm->page_directory);
 		vm->page_directory = NULL;
-- 
cgit v1.2.3


From 49b02b180a541d6fb31031aaffe75496e9238942 Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Fri, 13 Nov 2015 14:18:38 +0800
Subject: drm/amdgpu: reserve/unreserve objects out of map/unmap operations
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Change-Id: Id6514f2fb6e002437fdbe99353d5d35f4ac736c7
Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 21 ++++++++++++++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 29 +++++++----------------------
 2 files changed, 25 insertions(+), 25 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 16dca46314bd..00c5b580f56c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -515,6 +515,9 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	struct amdgpu_bo *rbo;
 	struct amdgpu_bo_va *bo_va;
+	struct ttm_validate_buffer tv, tv_pd;
+	struct ww_acquire_ctx ticket;
+	struct list_head list, duplicates;
 	uint32_t invalid_flags, va_flags = 0;
 	int r = 0;
 
@@ -552,7 +555,18 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 		return -ENOENT;
 	mutex_lock(&fpriv->vm.mutex);
 	rbo = gem_to_amdgpu_bo(gobj);
-	r = amdgpu_bo_reserve(rbo, false);
+	INIT_LIST_HEAD(&list);
+	INIT_LIST_HEAD(&duplicates);
+	tv.bo = &rbo->tbo;
+	tv.shared = true;
+	list_add(&tv.head, &list);
+
+	if (args->operation == AMDGPU_VA_OP_MAP) {
+		tv_pd.bo = &fpriv->vm.page_directory->tbo;
+		tv_pd.shared = true;
+		list_add(&tv_pd.head, &list);
+	}
+	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r) {
 		mutex_unlock(&fpriv->vm.mutex);
 		drm_gem_object_unreference_unlocked(gobj);
@@ -561,7 +575,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 
 	bo_va = amdgpu_vm_bo_find(&fpriv->vm, rbo);
 	if (!bo_va) {
-		amdgpu_bo_unreserve(rbo);
+		ttm_eu_backoff_reservation(&ticket, &list);
+		drm_gem_object_unreference_unlocked(gobj);
 		mutex_unlock(&fpriv->vm.mutex);
 		return -ENOENT;
 	}
@@ -584,7 +599,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	default:
 		break;
 	}
-
+	ttm_eu_backoff_reservation(&ticket, &list);
 	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE))
 		amdgpu_gem_va_update_vm(adev, bo_va, args->operation);
 	mutex_unlock(&fpriv->vm.mutex);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index e6dc19bc2dd8..159ce54bbd8d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -985,7 +985,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
  * Add a mapping of the BO at the specefied addr into the VM.
  * Returns 0 for success, error for failure.
  *
- * Object has to be reserved and gets unreserved by this function!
+ * Object has to be reserved and unreserved outside!
  */
 int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		     struct amdgpu_bo_va *bo_va,
@@ -1001,23 +1001,18 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 
 	/* validate the parameters */
 	if (saddr & AMDGPU_GPU_PAGE_MASK || offset & AMDGPU_GPU_PAGE_MASK ||
-	    size == 0 || size & AMDGPU_GPU_PAGE_MASK) {
-		amdgpu_bo_unreserve(bo_va->bo);
+	    size == 0 || size & AMDGPU_GPU_PAGE_MASK)
 		return -EINVAL;
-	}
 
 	/* make sure object fit at this offset */
 	eaddr = saddr + size;
-	if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo))) {
-		amdgpu_bo_unreserve(bo_va->bo);
+	if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo)))
 		return -EINVAL;
-	}
 
 	last_pfn = eaddr / AMDGPU_GPU_PAGE_SIZE;
 	if (last_pfn > adev->vm_manager.max_pfn) {
 		dev_err(adev->dev, "va above limit (0x%08X > 0x%08X)\n",
 			last_pfn, adev->vm_manager.max_pfn);
-		amdgpu_bo_unreserve(bo_va->bo);
 		return -EINVAL;
 	}
 
@@ -1034,14 +1029,12 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		dev_err(adev->dev, "bo %p va 0x%010Lx-0x%010Lx conflict with "
 			"0x%010lx-0x%010lx\n", bo_va->bo, saddr, eaddr,
 			tmp->it.start, tmp->it.last + 1);
-		amdgpu_bo_unreserve(bo_va->bo);
 		r = -EINVAL;
 		goto error;
 	}
 
 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
 	if (!mapping) {
-		amdgpu_bo_unreserve(bo_va->bo);
 		r = -ENOMEM;
 		goto error;
 	}
@@ -1067,8 +1060,6 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	if (eaddr > vm->max_pde_used)
 		vm->max_pde_used = eaddr;
 
-	amdgpu_bo_unreserve(bo_va->bo);
-
 	/* walk over the address space and allocate the page tables */
 	for (pt_idx = saddr; pt_idx <= eaddr; ++pt_idx) {
 		struct reservation_object *resv = vm->page_directory->tbo.resv;
@@ -1077,18 +1068,15 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		if (vm->page_tables[pt_idx].bo)
 			continue;
 
-		ww_mutex_lock(&resv->lock, NULL);
 		r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8,
 				     AMDGPU_GPU_PAGE_SIZE, true,
 				     AMDGPU_GEM_DOMAIN_VRAM,
 				     AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
 				     NULL, resv, &pt);
-		if (r) {
-			ww_mutex_unlock(&resv->lock);
+		if (r)
 			goto error_free;
-		}
+
 		r = amdgpu_vm_clear_bo(adev, pt);
-		ww_mutex_unlock(&resv->lock);
 		if (r) {
 			amdgpu_bo_unref(&pt);
 			goto error_free;
@@ -1122,7 +1110,7 @@ error:
  * Remove a mapping of the BO at the specefied addr from the VM.
  * Returns 0 for success, error for failure.
  *
- * Object has to be reserved and gets unreserved by this function!
+ * Object has to be reserved and unreserved outside!
  */
 int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 		       struct amdgpu_bo_va *bo_va,
@@ -1147,10 +1135,8 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 				break;
 		}
 
-		if (&mapping->list == &bo_va->invalids) {
-			amdgpu_bo_unreserve(bo_va->bo);
+		if (&mapping->list == &bo_va->invalids)
 			return -ENOENT;
-		}
 	}
 
 	list_del(&mapping->list);
@@ -1163,7 +1149,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 		list_add(&mapping->list, &vm->freed);
 	else
 		kfree(mapping);
-	amdgpu_bo_unreserve(bo_va->bo);
 
 	return 0;
 }
-- 
cgit v1.2.3


From 69b576a1bc8b466ae7bff0208f1c139dbaaf802c Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Wed, 18 Nov 2015 11:17:39 +0800
Subject: drm/amdgpu: add mutex for ba_va->valids/invalids
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 17 +++++++++++------
 2 files changed, 12 insertions(+), 6 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 306f75700bf8..50672bb19798 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -496,6 +496,7 @@ struct amdgpu_bo_va_mapping {
 
 /* bo virtual addresses in a specific vm */
 struct amdgpu_bo_va {
+	struct mutex		        mutex;
 	/* protected by bo being reserved */
 	struct list_head		bo_list;
 	struct fence		        *last_pt_update;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 159ce54bbd8d..d6904ef742f0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -922,8 +922,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
 		bo_va = list_first_entry(&vm->invalidated,
 			struct amdgpu_bo_va, vm_status);
 		spin_unlock(&vm->status_lock);
-
+		mutex_lock(&bo_va->mutex);
 		r = amdgpu_vm_bo_update(adev, bo_va, NULL);
+		mutex_unlock(&bo_va->mutex);
 		if (r)
 			return r;
 
@@ -967,7 +968,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 	INIT_LIST_HEAD(&bo_va->valids);
 	INIT_LIST_HEAD(&bo_va->invalids);
 	INIT_LIST_HEAD(&bo_va->vm_status);
-
+	mutex_init(&bo_va->mutex);
 	list_add_tail(&bo_va->bo_list, &bo->va);
 
 	return bo_va;
@@ -1045,7 +1046,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	mapping->offset = offset;
 	mapping->flags = flags;
 
+	mutex_lock(&bo_va->mutex);
 	list_add(&mapping->list, &bo_va->invalids);
+	mutex_unlock(&bo_va->mutex);
 	spin_lock(&vm->it_lock);
 	interval_tree_insert(&mapping->it, &vm->va);
 	spin_unlock(&vm->it_lock);
@@ -1121,7 +1124,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	bool valid = true;
 
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
-
+	mutex_lock(&bo_va->mutex);
 	list_for_each_entry(mapping, &bo_va->valids, list) {
 		if (mapping->it.start == saddr)
 			break;
@@ -1135,10 +1138,12 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 				break;
 		}
 
-		if (&mapping->list == &bo_va->invalids)
+		if (&mapping->list == &bo_va->invalids) {
+			mutex_unlock(&bo_va->mutex);
 			return -ENOENT;
+		}
 	}
-
+	mutex_unlock(&bo_va->mutex);
 	list_del(&mapping->list);
 	spin_lock(&vm->it_lock);
 	interval_tree_remove(&mapping->it, &vm->va);
@@ -1190,8 +1195,8 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		spin_unlock(&vm->it_lock);
 		kfree(mapping);
 	}
-
 	fence_put(bo_va->last_pt_update);
+	mutex_destroy(&bo_va->mutex);
 	kfree(bo_va);
 }
 
-- 
cgit v1.2.3


From e98c1b0de6fe73f488df62d83d83f377b1b6e2b8 Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Fri, 13 Nov 2015 15:22:04 +0800
Subject: drm/amdgpu: remove vm->mutex
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  |  4 ----
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 14 ++------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  |  2 --
 4 files changed, 2 insertions(+), 20 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 50672bb19798..251b14736de9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -929,8 +929,6 @@ struct amdgpu_vm_id {
 };
 
 struct amdgpu_vm {
-	struct mutex		mutex;
-
 	struct rb_root		va;
 
 	/* protecting invalidated */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 3afcf0237c25..1d44d508d4d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -784,8 +784,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
 	union drm_amdgpu_cs *cs = data;
-	struct amdgpu_fpriv *fpriv = filp->driver_priv;
-	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_cs_parser parser = {};
 	bool reserved_buffers = false;
 	int i, r;
@@ -803,7 +801,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		r = amdgpu_cs_handle_lockup(adev, r);
 		return r;
 	}
-	mutex_lock(&vm->mutex);
 	r = amdgpu_cs_parser_relocs(&parser);
 	if (r == -ENOMEM)
 		DRM_ERROR("Not enough memory for command submission!\n");
@@ -888,7 +885,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
 out:
 	amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-	mutex_unlock(&vm->mutex);
 	r = amdgpu_cs_handle_lockup(adev, r);
 	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 00c5b580f56c..fc32fc01a64b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -115,12 +115,9 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
 	int r;
-	mutex_lock(&vm->mutex);
 	r = amdgpu_bo_reserve(rbo, false);
-	if (r) {
-		mutex_unlock(&vm->mutex);
+	if (r)
 		return r;
-	}
 
 	bo_va = amdgpu_vm_bo_find(vm, rbo);
 	if (!bo_va) {
@@ -129,7 +126,6 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
 		++bo_va->ref_count;
 	}
 	amdgpu_bo_unreserve(rbo);
-	mutex_unlock(&vm->mutex);
 	return 0;
 }
 
@@ -142,10 +138,8 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
 	int r;
-	mutex_lock(&vm->mutex);
 	r = amdgpu_bo_reserve(rbo, true);
 	if (r) {
-		mutex_unlock(&vm->mutex);
 		dev_err(adev->dev, "leaking bo va because "
 			"we fail to reserve bo (%d)\n", r);
 		return;
@@ -157,7 +151,6 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
 		}
 	}
 	amdgpu_bo_unreserve(rbo);
-	mutex_unlock(&vm->mutex);
 }
 
 static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r)
@@ -553,7 +546,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	gobj = drm_gem_object_lookup(dev, filp, args->handle);
 	if (gobj == NULL)
 		return -ENOENT;
-	mutex_lock(&fpriv->vm.mutex);
 	rbo = gem_to_amdgpu_bo(gobj);
 	INIT_LIST_HEAD(&list);
 	INIT_LIST_HEAD(&duplicates);
@@ -568,7 +560,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	}
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r) {
-		mutex_unlock(&fpriv->vm.mutex);
 		drm_gem_object_unreference_unlocked(gobj);
 		return r;
 	}
@@ -577,7 +568,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	if (!bo_va) {
 		ttm_eu_backoff_reservation(&ticket, &list);
 		drm_gem_object_unreference_unlocked(gobj);
-		mutex_unlock(&fpriv->vm.mutex);
 		return -ENOENT;
 	}
 
@@ -602,7 +592,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	ttm_eu_backoff_reservation(&ticket, &list);
 	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE))
 		amdgpu_gem_va_update_vm(adev, bo_va, args->operation);
-	mutex_unlock(&fpriv->vm.mutex);
+
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index d6904ef742f0..ae037e5b6ad0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1241,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		vm->ids[i].id = 0;
 		vm->ids[i].flushed_updates = NULL;
 	}
-	mutex_init(&vm->mutex);
 	vm->va = RB_ROOT;
 	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->invalidated);
@@ -1325,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		fence_put(vm->ids[i].flushed_updates);
 	}
 
-	mutex_destroy(&vm->mutex);
 }
 
 /**
-- 
cgit v1.2.3


From 82b9c55b1edfcb87f5568add56bc7313f5893b60 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 27 Nov 2015 16:49:00 +0100
Subject: drm/amdgpu: fix VM page table reference counting
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

We use the reservation object of the page directory for the page tables as
well, because of this the page directory should be freed last. Ensure that
by keeping a reference from the page tables to the directory.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c     | 5 +++++
 3 files changed, 7 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 251b14736de9..670fefb56945 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -539,6 +539,7 @@ struct amdgpu_bo {
 	/* Constant after initialization */
 	struct amdgpu_device		*adev;
 	struct drm_gem_object		gem_base;
+	struct amdgpu_bo		*parent;
 
 	struct ttm_bo_kmap_obj		dma_buf_vmap;
 	pid_t				pid;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 0d524384ff79..c3ce103b6a33 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -100,6 +100,7 @@ static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 	list_del_init(&bo->list);
 	mutex_unlock(&bo->adev->gem.mutex);
 	drm_gem_object_release(&bo->gem_base);
+	amdgpu_bo_unref(&bo->parent);
 	kfree(bo->metadata);
 	kfree(bo);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index ae037e5b6ad0..a582ef553499 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1079,6 +1079,11 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		if (r)
 			goto error_free;
 
+		/* Keep a reference to the page table to avoid freeing
+		 * them up in the wrong order.
+		 */
+		pt->parent = amdgpu_bo_ref(vm->page_directory);
+
 		r = amdgpu_vm_clear_bo(adev, pt);
 		if (r) {
 			amdgpu_bo_unref(&pt);
-- 
cgit v1.2.3


From 9c4153b1eef9bc8da6a624252a3a25790b705136 Mon Sep 17 00:00:00 2001
From: jimqu <Jim.Qu@amd.com>
Date: Fri, 4 Dec 2015 17:17:00 +0800
Subject: drm/amdgpu: add spin lock to protect freed list in vm (v2)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

there is a protection fault about freed list when OCL test.
add a spin lock to protect it.

v2: drop changes in vm_fini

Signed-off-by: JimQu <jim.qu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 16 +++++++++++++---
 2 files changed, 15 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 5f975030bb72..3b5d3706f0cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -955,6 +955,8 @@ struct amdgpu_vm {
 	struct amdgpu_vm_id	ids[AMDGPU_MAX_RINGS];
 	/* for interval tree */
 	spinlock_t		it_lock;
+	/* protecting freed */
+	spinlock_t		freed_lock;
 };
 
 struct amdgpu_vm_manager {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index ae037e5b6ad0..fce4c6d952c8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -885,17 +885,21 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 	struct amdgpu_bo_va_mapping *mapping;
 	int r;
 
+	spin_lock(&vm->freed_lock);
 	while (!list_empty(&vm->freed)) {
 		mapping = list_first_entry(&vm->freed,
 			struct amdgpu_bo_va_mapping, list);
 		list_del(&mapping->list);
-
+		spin_unlock(&vm->freed_lock);
 		r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, 0, 0, NULL);
 		kfree(mapping);
 		if (r)
 			return r;
 
+		spin_lock(&vm->freed_lock);
 	}
+	spin_unlock(&vm->freed_lock);
+
 	return 0;
 
 }
@@ -1150,10 +1154,13 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 
-	if (valid)
+	if (valid) {
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
-	else
+		spin_unlock(&vm->freed_lock);
+	} else {
 		kfree(mapping);
+	}
 
 	return 0;
 }
@@ -1186,7 +1193,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		interval_tree_remove(&mapping->it, &vm->va);
 		spin_unlock(&vm->it_lock);
 		trace_amdgpu_vm_bo_unmap(bo_va, mapping);
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
+		spin_unlock(&vm->freed_lock);
 	}
 	list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
 		list_del(&mapping->list);
@@ -1247,6 +1256,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
 	spin_lock_init(&vm->it_lock);
+	spin_lock_init(&vm->freed_lock);
 	pd_size = amdgpu_vm_directory_size(adev);
 	pd_entries = amdgpu_vm_num_pdes(adev);
 
-- 
cgit v1.2.3


From 81d75a30c6ed006a314f5c760196d04758660ca6 Mon Sep 17 00:00:00 2001
From: jimqu <Jim.Qu@amd.com>
Date: Fri, 4 Dec 2015 17:17:00 +0800
Subject: drm/amdgpu: add spin lock to protect freed list in vm (v2)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

there is a protection fault about freed list when OCL test.
add a spin lock to protect it.

v2: drop changes in vm_fini

Signed-off-by: JimQu <jim.qu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 16 +++++++++++++---
 2 files changed, 15 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 670fefb56945..5a5f04d0902d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -956,6 +956,8 @@ struct amdgpu_vm {
 	struct amdgpu_vm_id	ids[AMDGPU_MAX_RINGS];
 	/* for interval tree */
 	spinlock_t		it_lock;
+	/* protecting freed */
+	spinlock_t		freed_lock;
 };
 
 struct amdgpu_vm_manager {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a582ef553499..b53d273eb7a1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -885,17 +885,21 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 	struct amdgpu_bo_va_mapping *mapping;
 	int r;
 
+	spin_lock(&vm->freed_lock);
 	while (!list_empty(&vm->freed)) {
 		mapping = list_first_entry(&vm->freed,
 			struct amdgpu_bo_va_mapping, list);
 		list_del(&mapping->list);
-
+		spin_unlock(&vm->freed_lock);
 		r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, 0, 0, NULL);
 		kfree(mapping);
 		if (r)
 			return r;
 
+		spin_lock(&vm->freed_lock);
 	}
+	spin_unlock(&vm->freed_lock);
+
 	return 0;
 
 }
@@ -1155,10 +1159,13 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 
-	if (valid)
+	if (valid) {
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
-	else
+		spin_unlock(&vm->freed_lock);
+	} else {
 		kfree(mapping);
+	}
 
 	return 0;
 }
@@ -1191,7 +1198,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		interval_tree_remove(&mapping->it, &vm->va);
 		spin_unlock(&vm->it_lock);
 		trace_amdgpu_vm_bo_unmap(bo_va, mapping);
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
+		spin_unlock(&vm->freed_lock);
 	}
 	list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
 		list_del(&mapping->list);
@@ -1252,6 +1261,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
 	spin_lock_init(&vm->it_lock);
+	spin_lock_init(&vm->freed_lock);
 	pd_size = amdgpu_vm_directory_size(adev);
 	pd_entries = amdgpu_vm_num_pdes(adev);
 
-- 
cgit v1.2.3


From 3c0eea6c35d932c4d25070868067dc9cd9ceab91 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 11 Dec 2015 14:39:05 +0100
Subject: drm/amdgpu: put VM page tables directly into duplicates list
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

They share the reservation object with the page directory anyway.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  3 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 12 +++++++-----
 4 files changed, 12 insertions(+), 9 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index c3996e0e2e7e..dc3dab539e4c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -982,7 +982,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
 					       struct amdgpu_vm *vm,
-					       struct list_head *head);
+					       struct list_head *validated,
+					       struct list_head *duplicates);
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 1d44d508d4d4..9591c13781bd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -386,13 +386,13 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 		amdgpu_cs_buckets_get_list(&buckets, &p->validated);
 	}
 
+	INIT_LIST_HEAD(&duplicates);
 	p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm,
-				      &p->validated);
+				      &p->validated, &duplicates);
 
 	if (need_mmap_lock)
 		down_read(&current->mm->mmap_sem);
 
-	INIT_LIST_HEAD(&duplicates);
 	r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, &duplicates);
 	if (unlikely(r != 0))
 		goto error_reserve;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index fc32fc01a64b..7fe7f8afa5ff 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -460,7 +460,7 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 	tv.shared = true;
 	list_add(&tv.head, &list);
 
-	vm_bos = amdgpu_vm_get_bos(adev, bo_va->vm, &list);
+	vm_bos = amdgpu_vm_get_bos(adev, bo_va->vm, &list, &duplicates);
 	if (!vm_bos)
 		return;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index fce4c6d952c8..f6c1d6f0bf37 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -78,14 +78,16 @@ static unsigned amdgpu_vm_directory_size(struct amdgpu_device *adev)
  * amdgpu_vm_get_bos - add the vm BOs to a validation list
  *
  * @vm: vm providing the BOs
- * @head: head of validation list
+ * @validated: head of validation list
+ * @duplicates: head of duplicates list
  *
  * Add the page directory to the list of BOs to
  * validate for command submission (cayman+).
  */
 struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
-					  struct amdgpu_vm *vm,
-					  struct list_head *head)
+					       struct amdgpu_vm *vm,
+					       struct list_head *validated,
+					       struct list_head *duplicates)
 {
 	struct amdgpu_bo_list_entry *list;
 	unsigned i, idx;
@@ -103,7 +105,7 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
 	list[0].priority = 0;
 	list[0].tv.bo = &vm->page_directory->tbo;
 	list[0].tv.shared = true;
-	list_add(&list[0].tv.head, head);
+	list_add(&list[0].tv.head, validated);
 
 	for (i = 0, idx = 1; i <= vm->max_pde_used; i++) {
 		if (!vm->page_tables[i].bo)
@@ -115,7 +117,7 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
 		list[idx].priority = 0;
 		list[idx].tv.bo = &list[idx].robj->tbo;
 		list[idx].tv.shared = true;
-		list_add(&list[idx++].tv.head, head);
+		list_add(&list[idx++].tv.head, duplicates);
 	}
 
 	return list;
-- 
cgit v1.2.3


From 56467ebfb254836dc30eb45d4ac8a46a400bfad6 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 11 Dec 2015 15:16:32 +0100
Subject: drm/amdgpu: split VM PD and PT handling during CS
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This way we avoid the extra allocation for the page directory entry.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     | 10 +++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  |  9 +++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 16 +++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 45 ++++++++++++++++++++-------------
 4 files changed, 51 insertions(+), 29 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index dc3dab539e4c..40850afa763f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -980,10 +980,11 @@ struct amdgpu_vm_manager {
 void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
 int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
-struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
-					       struct amdgpu_vm *vm,
-					       struct list_head *validated,
-					       struct list_head *duplicates);
+void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
+			 struct list_head *validated,
+			 struct amdgpu_bo_list_entry *entry);
+struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
+						  struct list_head *duplicates);
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
@@ -1253,6 +1254,7 @@ struct amdgpu_cs_parser {
 	unsigned		nchunks;
 	struct amdgpu_cs_chunk	*chunks;
 	/* relocations */
+	struct amdgpu_bo_list_entry	vm_pd;
 	struct amdgpu_bo_list_entry	*vm_bos;
 	struct list_head	validated;
 	struct fence		*fence;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 9591c13781bd..3fb21ecd29e0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -387,8 +387,7 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	}
 
 	INIT_LIST_HEAD(&duplicates);
-	p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm,
-				      &p->validated, &duplicates);
+	amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
 
 	if (need_mmap_lock)
 		down_read(&current->mm->mmap_sem);
@@ -397,6 +396,12 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	if (unlikely(r != 0))
 		goto error_reserve;
 
+	p->vm_bos = amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
+	if (!p->vm_bos) {
+		r = -ENOMEM;
+		goto error_validate;
+	}
+
 	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &p->validated);
 	if (r)
 		goto error_validate;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 7fe7f8afa5ff..ea0fe94e4b54 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -448,6 +448,7 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 {
 	struct ttm_validate_buffer tv, *entry;
 	struct amdgpu_bo_list_entry *vm_bos;
+	struct amdgpu_bo_list_entry vm_pd;
 	struct ww_acquire_ctx ticket;
 	struct list_head list, duplicates;
 	unsigned domain;
@@ -460,14 +461,18 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 	tv.shared = true;
 	list_add(&tv.head, &list);
 
-	vm_bos = amdgpu_vm_get_bos(adev, bo_va->vm, &list, &duplicates);
-	if (!vm_bos)
-		return;
+	amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
 
 	/* Provide duplicates to avoid -EALREADY */
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r)
-		goto error_free;
+		goto error_print;
+
+	vm_bos = amdgpu_vm_get_pt_bos(bo_va->vm, &duplicates);
+	if (!vm_bos) {
+		r = -ENOMEM;
+		goto error_unreserve;
+	}
 
 	list_for_each_entry(entry, &list, head) {
 		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
@@ -489,10 +494,9 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 
 error_unreserve:
 	ttm_eu_backoff_reservation(&ticket, &list);
-
-error_free:
 	drm_free_large(vm_bos);
 
+error_print:
 	if (r && r != -ERESTARTSYS)
 		DRM_ERROR("Couldn't update BO_VA (%d)\n", r);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index f6c1d6f0bf37..592be6438a6c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -75,39 +75,50 @@ static unsigned amdgpu_vm_directory_size(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_vm_get_bos - add the vm BOs to a validation list
+ * amdgpu_vm_get_pd_bo - add the VM PD to a validation list
  *
  * @vm: vm providing the BOs
  * @validated: head of validation list
+ * @entry: entry to add
+ *
+ * Add the page directory to the list of BOs to
+ * validate for command submission.
+ */
+void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
+			 struct list_head *validated,
+			 struct amdgpu_bo_list_entry *entry)
+{
+	entry->robj = vm->page_directory;
+	entry->prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
+	entry->allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
+	entry->priority = 0;
+	entry->tv.bo = &vm->page_directory->tbo;
+	entry->tv.shared = true;
+	list_add(&entry->tv.head, validated);
+}
+
+/**
+ * amdgpu_vm_get_bos - add the vm BOs to a validation list
+ *
+ * @vm: vm providing the BOs
  * @duplicates: head of duplicates list
  *
  * Add the page directory to the list of BOs to
  * validate for command submission (cayman+).
  */
-struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
-					       struct amdgpu_vm *vm,
-					       struct list_head *validated,
-					       struct list_head *duplicates)
+struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
+						  struct list_head *duplicates)
 {
 	struct amdgpu_bo_list_entry *list;
 	unsigned i, idx;
 
-	list = drm_malloc_ab(vm->max_pde_used + 2,
+	list = drm_malloc_ab(vm->max_pde_used + 1,
 			     sizeof(struct amdgpu_bo_list_entry));
-	if (!list) {
+	if (!list)
 		return NULL;
-	}
 
 	/* add the vm page table to the list */
-	list[0].robj = vm->page_directory;
-	list[0].prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
-	list[0].allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
-	list[0].priority = 0;
-	list[0].tv.bo = &vm->page_directory->tbo;
-	list[0].tv.shared = true;
-	list_add(&list[0].tv.head, validated);
-
-	for (i = 0, idx = 1; i <= vm->max_pde_used; i++) {
+	for (i = 0, idx = 0; i <= vm->max_pde_used; i++) {
 		if (!vm->page_tables[i].bo)
 			continue;
 
-- 
cgit v1.2.3


From ee1782c3f27fec5462363af48f27811b049155ab Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 11 Dec 2015 21:01:23 +0100
Subject: drm/amdgpu: keep the PTs validation list in the VM v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This avoids allocating it on the fly.

v2: fix grammar in comment

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  8 ++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  |  7 +----
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c |  9 +-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 50 +++++++++++++++------------------
 4 files changed, 27 insertions(+), 47 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 40850afa763f..d4e9272b60e8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -917,8 +917,8 @@ struct amdgpu_ring {
 #define AMDGPU_VM_FAULT_STOP_ALWAYS	2
 
 struct amdgpu_vm_pt {
-	struct amdgpu_bo	*bo;
-	uint64_t		addr;
+	struct amdgpu_bo_list_entry	entry;
+	uint64_t			addr;
 };
 
 struct amdgpu_vm_id {
@@ -983,8 +983,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 			 struct list_head *validated,
 			 struct amdgpu_bo_list_entry *entry);
-struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
-						  struct list_head *duplicates);
+void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates);
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
@@ -1255,7 +1254,6 @@ struct amdgpu_cs_parser {
 	struct amdgpu_cs_chunk	*chunks;
 	/* relocations */
 	struct amdgpu_bo_list_entry	vm_pd;
-	struct amdgpu_bo_list_entry	*vm_bos;
 	struct list_head	validated;
 	struct fence		*fence;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 3fb21ecd29e0..6ce595ff1aff 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -396,11 +396,7 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	if (unlikely(r != 0))
 		goto error_reserve;
 
-	p->vm_bos = amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
-	if (!p->vm_bos) {
-		r = -ENOMEM;
-		goto error_validate;
-	}
+	amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
 
 	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &p->validated);
 	if (r)
@@ -483,7 +479,6 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 	if (parser->bo_list)
 		amdgpu_bo_list_put(parser->bo_list);
 
-	drm_free_large(parser->vm_bos);
 	for (i = 0; i < parser->nchunks; i++)
 		drm_free_large(parser->chunks[i].kdata);
 	kfree(parser->chunks);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index ea0fe94e4b54..8c5687e4a6d1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -447,7 +447,6 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 				    struct amdgpu_bo_va *bo_va, uint32_t operation)
 {
 	struct ttm_validate_buffer tv, *entry;
-	struct amdgpu_bo_list_entry *vm_bos;
 	struct amdgpu_bo_list_entry vm_pd;
 	struct ww_acquire_ctx ticket;
 	struct list_head list, duplicates;
@@ -468,12 +467,7 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 	if (r)
 		goto error_print;
 
-	vm_bos = amdgpu_vm_get_pt_bos(bo_va->vm, &duplicates);
-	if (!vm_bos) {
-		r = -ENOMEM;
-		goto error_unreserve;
-	}
-
+	amdgpu_vm_get_pt_bos(bo_va->vm, &duplicates);
 	list_for_each_entry(entry, &list, head) {
 		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
 		/* if anything is swapped out don't swap it in here,
@@ -494,7 +488,6 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 
 error_unreserve:
 	ttm_eu_backoff_reservation(&ticket, &list);
-	drm_free_large(vm_bos);
 
 error_print:
 	if (r && r != -ERESTARTSYS)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 592be6438a6c..e0fa9d9ff5c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -98,40 +98,27 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 }
 
 /**
- * amdgpu_vm_get_bos - add the vm BOs to a validation list
+ * amdgpu_vm_get_bos - add the vm BOs to a duplicates list
  *
  * @vm: vm providing the BOs
  * @duplicates: head of duplicates list
  *
- * Add the page directory to the list of BOs to
- * validate for command submission (cayman+).
+ * Add the page directory to the BO duplicates list
+ * for command submission.
  */
-struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
-						  struct list_head *duplicates)
+void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates)
 {
-	struct amdgpu_bo_list_entry *list;
-	unsigned i, idx;
-
-	list = drm_malloc_ab(vm->max_pde_used + 1,
-			     sizeof(struct amdgpu_bo_list_entry));
-	if (!list)
-		return NULL;
+	unsigned i;
 
 	/* add the vm page table to the list */
-	for (i = 0, idx = 0; i <= vm->max_pde_used; i++) {
-		if (!vm->page_tables[i].bo)
+	for (i = 0; i <= vm->max_pde_used; ++i) {
+		struct amdgpu_bo_list_entry *entry = &vm->page_tables[i].entry;
+
+		if (!entry->robj)
 			continue;
 
-		list[idx].robj = vm->page_tables[i].bo;
-		list[idx].prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
-		list[idx].allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
-		list[idx].priority = 0;
-		list[idx].tv.bo = &list[idx].robj->tbo;
-		list[idx].tv.shared = true;
-		list_add(&list[idx++].tv.head, duplicates);
+		list_add(&entry->tv.head, duplicates);
 	}
-
-	return list;
 }
 
 /**
@@ -474,7 +461,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 
 	/* walk over the address space and update the page directory */
 	for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) {
-		struct amdgpu_bo *bo = vm->page_tables[pt_idx].bo;
+		struct amdgpu_bo *bo = vm->page_tables[pt_idx].entry.robj;
 		uint64_t pde, pt;
 
 		if (bo == NULL)
@@ -651,7 +638,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
 	/* walk over the address space and update the page tables */
 	for (addr = start; addr < end; ) {
 		uint64_t pt_idx = addr >> amdgpu_vm_block_size;
-		struct amdgpu_bo *pt = vm->page_tables[pt_idx].bo;
+		struct amdgpu_bo *pt = vm->page_tables[pt_idx].entry.robj;
 		unsigned nptes;
 		uint64_t pte;
 		int r;
@@ -1083,9 +1070,11 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	/* walk over the address space and allocate the page tables */
 	for (pt_idx = saddr; pt_idx <= eaddr; ++pt_idx) {
 		struct reservation_object *resv = vm->page_directory->tbo.resv;
+		struct amdgpu_bo_list_entry *entry;
 		struct amdgpu_bo *pt;
 
-		if (vm->page_tables[pt_idx].bo)
+		entry = &vm->page_tables[pt_idx].entry;
+		if (entry->robj)
 			continue;
 
 		r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8,
@@ -1102,8 +1091,13 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 			goto error_free;
 		}
 
+		entry->robj = pt;
+		entry->prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
+		entry->allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
+		entry->priority = 0;
+		entry->tv.bo = &entry->robj->tbo;
+		entry->tv.shared = true;
 		vm->page_tables[pt_idx].addr = 0;
-		vm->page_tables[pt_idx].bo = pt;
 	}
 
 	return 0;
@@ -1334,7 +1328,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	}
 
 	for (i = 0; i < amdgpu_vm_num_pdes(adev); i++)
-		amdgpu_bo_unref(&vm->page_tables[i].bo);
+		amdgpu_bo_unref(&vm->page_tables[i].entry.robj);
 	kfree(vm->page_tables);
 
 	amdgpu_bo_unref(&vm->page_directory);
-- 
cgit v1.2.3


From 005ae95e6ec119c64e2d16eb65a94c49e1dcf9f0 Mon Sep 17 00:00:00 2001
From: Felix Kuehling <Felix.Kuehling@amd.com>
Date: Mon, 23 Nov 2015 17:43:48 -0500
Subject: drm/amdgpu: Fix off-by-one errors in amdgpu_vm_bo_map
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

eaddr is sometimes treated as the last address inside the address
range, and sometimes as the first address outside the range. This
was resulting in errors when a test filled up the entire address
space. Make it consistent to always be the last address within the
range.

Signed-off-by: Felix.Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index e0fa9d9ff5c2..d6ff5dad98f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1010,13 +1010,13 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		return -EINVAL;
 
 	/* make sure object fit at this offset */
-	eaddr = saddr + size;
+	eaddr = saddr + size - 1;
 	if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo)))
 		return -EINVAL;
 
 	last_pfn = eaddr / AMDGPU_GPU_PAGE_SIZE;
-	if (last_pfn > adev->vm_manager.max_pfn) {
-		dev_err(adev->dev, "va above limit (0x%08X > 0x%08X)\n",
+	if (last_pfn >= adev->vm_manager.max_pfn) {
+		dev_err(adev->dev, "va above limit (0x%08X >= 0x%08X)\n",
 			last_pfn, adev->vm_manager.max_pfn);
 		return -EINVAL;
 	}
@@ -1025,7 +1025,7 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	eaddr /= AMDGPU_GPU_PAGE_SIZE;
 
 	spin_lock(&vm->it_lock);
-	it = interval_tree_iter_first(&vm->va, saddr, eaddr - 1);
+	it = interval_tree_iter_first(&vm->va, saddr, eaddr);
 	spin_unlock(&vm->it_lock);
 	if (it) {
 		struct amdgpu_bo_va_mapping *tmp;
@@ -1046,7 +1046,7 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 
 	INIT_LIST_HEAD(&mapping->list);
 	mapping->it.start = saddr;
-	mapping->it.last = eaddr - 1;
+	mapping->it.last = eaddr;
 	mapping->offset = offset;
 	mapping->flags = flags;
 
-- 
cgit v1.2.3


From eceb8a1562db6fc3e1c21cd54a4be0189aa1d0e3 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 11 Jan 2016 15:35:21 +0100
Subject: drm/amdgpu: move VM page tables to the LRU end on CS v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This makes it less likely to run into an ENOMEM because
VM page tables are evicted last.

v2: move the BOs in the LRU tail after validation

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c |  7 ++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 27 +++++++++++++++++++++++++++
 3 files changed, 35 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 003959f99251..313b0cc8d676 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -987,6 +987,8 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 			 struct list_head *validated,
 			 struct amdgpu_bo_list_entry *entry);
 void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates);
+void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
+				  struct amdgpu_vm *vm);
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index ce0254d4dcd7..1fffc338f69d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -428,8 +428,10 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &duplicates);
 
 error_validate:
-	if (r)
+	if (r) {
+		amdgpu_vm_move_pt_bos_in_lru(p->adev, &fpriv->vm);
 		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
+	}
 
 error_reserve:
 	if (need_mmap_lock)
@@ -473,8 +475,11 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a,
  **/
 static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
 {
+	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
 	unsigned i;
 
+	amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
+
 	if (!error) {
 		/* Sort the buffer list from the smallest to largest buffer,
 		 * which affects the order of buffers in the LRU list.
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 8f7688e598a1..aefc668e6b5d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -119,6 +119,33 @@ void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates)
 
 		list_add(&entry->tv.head, duplicates);
 	}
+
+}
+
+/**
+ * amdgpu_vm_move_pt_bos_in_lru - move the PT BOs to the LRU tail
+ *
+ * @adev: amdgpu device instance
+ * @vm: vm providing the BOs
+ *
+ * Move the PT BOs to the tail of the LRU.
+ */
+void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
+				  struct amdgpu_vm *vm)
+{
+	struct ttm_bo_global *glob = adev->mman.bdev.glob;
+	unsigned i;
+
+	spin_lock(&glob->lru_lock);
+	for (i = 0; i <= vm->max_pde_used; ++i) {
+		struct amdgpu_bo_list_entry *entry = &vm->page_tables[i].entry;
+
+		if (!entry->robj)
+			continue;
+
+		ttm_bo_move_to_lru_tail(&entry->robj->tbo);
+	}
+	spin_unlock(&glob->lru_lock);
 }
 
 /**
-- 
cgit v1.2.3