Merge branch 'drm-next-4.6' of git://people.freedesktop.org/~agd5f/li…

…nux into drm-next

Some more radeon and amdgpu stuff for drm-next.  Mostly just bug fixes
for new features and cleanups.

* 'drm-next-4.6' of git://people.freedesktop.org/~agd5f/linux:
  drm/amdgpu: fix rb bitmap & cu bitmap calculation
  drm/amdgpu: trace the pd_addr in vm_grab_id as well
  drm/amdgpu: fix VM faults caused by vm_grab_id() v4
  drm/amdgpu: update radeon acpi header
  drm/radeon: update radeon acpi header
  drm/amd: cleanup get_mfd_cell_dev()
  drm/amdgpu: fix error handling in amdgpu_bo_list_set
  drm/amd/powerplay: fix code style warning.
  drm/amd: Do not make DRM_AMD_ACP default to y
  drm/amdgpu/gfx: fix off by one in rb rework (v2)

drivers/gpu/drm/amd/acp/Kconfig

@@ -2,7 +2,6 @@ menu "ACP Configuration"
 
 config DRM_AMD_ACP
        bool "Enable ACP IP support"
-       default y
        select MFD_CORE
        select PM_GENERIC_DOMAINS if PM
        help

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

@@ -769,8 +769,9 @@ struct amdgpu_ib {
 	uint32_t			*ptr;
 	struct amdgpu_fence		*fence;
 	struct amdgpu_user_fence        *user;
-	bool				grabbed_vmid;
 	struct amdgpu_vm		*vm;
+	unsigned			vm_id;
+	uint64_t			vm_pd_addr;
 	struct amdgpu_ctx		*ctx;
 	uint32_t			gds_base, gds_size;
 	uint32_t			gws_base, gws_size;
@@ -877,10 +878,10 @@ struct amdgpu_vm_pt {
 };
 
 struct amdgpu_vm_id {
-	unsigned		id;
-	uint64_t		pd_gpu_addr;
+	struct amdgpu_vm_manager_id	*mgr_id;
+	uint64_t			pd_gpu_addr;
 	/* last flushed PD/PT update */
-	struct fence	        *flushed_updates;
+	struct fence			*flushed_updates;
 };
 
 struct amdgpu_vm {
@@ -954,10 +955,11 @@ 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, struct fence *fence);
+		      struct amdgpu_sync *sync, struct fence *fence,
+		      unsigned *vm_id, uint64_t *vm_pd_addr);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
-		     struct amdgpu_vm *vm,
-		     struct fence *updates);
+		     unsigned vmid,
+		     uint64_t pd_addr);
 uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr);
 int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 				    struct amdgpu_vm *vm);

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

@@ -240,12 +240,10 @@ static int acp_poweron(struct generic_pm_domain *genpd)
 static struct device *get_mfd_cell_dev(const char *device_name, int r)
 {
 	char auto_dev_name[25];
-	char buf[8];
 	struct device *dev;
 
-	sprintf(buf, ".%d.auto", r);
-	strcpy(auto_dev_name, device_name);
-	strcat(auto_dev_name, buf);
+	snprintf(auto_dev_name, sizeof(auto_dev_name),
+		 "%s.%d.auto", device_name, r);
 	dev = bus_find_device_by_name(&platform_bus_type, NULL, auto_dev_name);
 	dev_info(dev, "device %s added to pm domain\n", auto_dev_name);
 

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

@@ -118,6 +118,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 		usermm = amdgpu_ttm_tt_get_usermm(entry->robj->tbo.ttm);
 		if (usermm) {
 			if (usermm != current->mm) {
+				amdgpu_bo_unref(&entry->robj);
 				r = -EPERM;
 				goto error_free;
 			}
@@ -151,6 +152,8 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 	return 0;
 
 error_free:
+	while (i--)
+		amdgpu_bo_unref(&array[i].robj);
 	drm_free_large(array);
 	return r;
 }

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

@@ -75,6 +75,7 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	}
 
 	ib->vm = vm;
+	ib->vm_id = 0;
 
 	return 0;
 }
@@ -139,7 +140,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		return -EINVAL;
 	}
 
-	if (vm && !ibs->grabbed_vmid) {
+	if (vm && !ibs->vm_id) {
 		dev_err(adev->dev, "VM IB without ID\n");
 		return -EINVAL;
 	}
@@ -152,10 +153,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 
 	if (vm) {
 		/* do context switch */
-		amdgpu_vm_flush(ring, vm, last_vm_update);
+		amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr);
 
 		if (ring->funcs->emit_gds_switch)
-			amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id,
+			amdgpu_ring_emit_gds_switch(ring, ib->vm_id,
 						    ib->gds_base, ib->gds_size,
 						    ib->gws_base, ib->gws_size,
 						    ib->oa_base, ib->oa_size);

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

@@ -105,16 +105,23 @@ static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
 
 	struct fence *fence = amdgpu_sync_get_fence(&job->sync);
 
-	if (fence == NULL && vm && !job->ibs->grabbed_vmid) {
+	if (fence == NULL && vm && !job->ibs->vm_id) {
 		struct amdgpu_ring *ring = job->ring;
+		unsigned i, vm_id;
+		uint64_t vm_pd_addr;
 		int r;
 
 		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
-				      &job->base.s_fence->base);
+				      &job->base.s_fence->base,
+				      &vm_id, &vm_pd_addr);
 		if (r)
 			DRM_ERROR("Error getting VM ID (%d)\n", r);
-		else
-			job->ibs->grabbed_vmid = true;
+		else {
+			for (i = 0; i < job->num_ibs; ++i) {
+				job->ibs[i].vm_id = vm_id;
+				job->ibs[i].vm_pd_addr = vm_pd_addr;
+			}
+		}
 
 		fence = amdgpu_sync_get_fence(&job->sync);
 	}

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

@@ -100,21 +100,24 @@ TRACE_EVENT(amdgpu_sched_run_job,
 
 
 TRACE_EVENT(amdgpu_vm_grab_id,
-	    TP_PROTO(struct amdgpu_vm *vm, unsigned vmid, int ring),
-	    TP_ARGS(vm, vmid, ring),
+	    TP_PROTO(struct amdgpu_vm *vm, int ring, unsigned vmid,
+		     uint64_t pd_addr),
+	    TP_ARGS(vm, ring, vmid, pd_addr),
 	    TP_STRUCT__entry(
 			     __field(struct amdgpu_vm *, vm)
-			     __field(u32, vmid)
 			     __field(u32, ring)
+			     __field(u32, vmid)
+			     __field(u64, pd_addr)
 			     ),
 
 	    TP_fast_assign(
 			   __entry->vm = vm;
-			   __entry->vmid = vmid;
 			   __entry->ring = ring;
+			   __entry->vmid = vmid;
+			   __entry->pd_addr = pd_addr;
 			   ),
-	    TP_printk("vm=%p, id=%u, ring=%u", __entry->vm, __entry->vmid,
-		      __entry->ring)
+	    TP_printk("vm=%p, ring=%u, id=%u, pd_addr=%010Lx", __entry->vm,
+		      __entry->ring, __entry->vmid, __entry->pd_addr)
 );
 
 TRACE_EVENT(amdgpu_vm_bo_map,
@@ -231,8 +234,8 @@ TRACE_EVENT(amdgpu_vm_flush,
 			   __entry->ring = ring;
 			   __entry->id = id;
 			   ),
-	    TP_printk("pd_addr=%010Lx, ring=%u, id=%u",
-		      __entry->pd_addr, __entry->ring, __entry->id)
+	    TP_printk("ring=%u, id=%u, pd_addr=%010Lx",
+		      __entry->ring, __entry->id, __entry->pd_addr)
 );
 
 TRACE_EVENT(amdgpu_bo_list_set,

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

@@ -50,6 +50,9 @@
  * SI supports 16.
  */
 
+/* Special value that no flush is necessary */
+#define AMDGPU_VM_NO_FLUSH (~0ll)
+
 /**
  * amdgpu_vm_num_pde - return the number of page directory entries
  *
@@ -157,50 +160,70 @@ void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
  * Allocate an id for the vm, adding fences to the sync obj as necessary.
  */
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
-		      struct amdgpu_sync *sync, struct fence *fence)
+		      struct amdgpu_sync *sync, struct fence *fence,
+		      unsigned *vm_id, uint64_t *vm_pd_addr)
 {
-	struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
+	uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory);
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_vm_manager_id *id;
+	struct amdgpu_vm_id *id = &vm->ids[ring->idx];
+	struct fence *updates = sync->last_vm_update;
 	int r;
 
 	mutex_lock(&adev->vm_manager.lock);
 
 	/* check if the id is still valid */
-	if (vm_id->id) {
+	if (id->mgr_id) {
+		struct fence *flushed = id->flushed_updates;
+		bool is_later;
 		long owner;
 
-		id = &adev->vm_manager.ids[vm_id->id];
-		owner = atomic_long_read(&id->owner);
-		if (owner == (long)vm) {
-			list_move_tail(&id->list, &adev->vm_manager.ids_lru);
-			trace_amdgpu_vm_grab_id(vm, vm_id->id, ring->idx);
+		if (!flushed)
+			is_later = true;
+		else if (!updates)
+			is_later = false;
+		else
+			is_later = fence_is_later(updates, flushed);
+
+		owner = atomic_long_read(&id->mgr_id->owner);
+		if (!is_later && owner == (long)id &&
+		    pd_addr == id->pd_gpu_addr) {
+
+			fence_put(id->mgr_id->active);
+			id->mgr_id->active = fence_get(fence);
+
+			list_move_tail(&id->mgr_id->list,
+				       &adev->vm_manager.ids_lru);
 
-			fence_put(id->active);
-			id->active = fence_get(fence);
+			*vm_id = id->mgr_id - adev->vm_manager.ids;
+			*vm_pd_addr = AMDGPU_VM_NO_FLUSH;
+			trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id,
+						*vm_pd_addr);
 
 			mutex_unlock(&adev->vm_manager.lock);
 			return 0;
 		}
 	}
 
-	/* we definately need to flush */
-	vm_id->pd_gpu_addr = ~0ll;
+	id->mgr_id = list_first_entry(&adev->vm_manager.ids_lru,
+				      struct amdgpu_vm_manager_id,
+				      list);
 
-	id = list_first_entry(&adev->vm_manager.ids_lru,
-			      struct amdgpu_vm_manager_id,
-			      list);
-	list_move_tail(&id->list, &adev->vm_manager.ids_lru);
-	atomic_long_set(&id->owner, (long)vm);
+	r = amdgpu_sync_fence(ring->adev, sync, id->mgr_id->active);
+	if (!r) {
+		fence_put(id->mgr_id->active);
+		id->mgr_id->active = fence_get(fence);
 
-	vm_id->id = id - adev->vm_manager.ids;
-	trace_amdgpu_vm_grab_id(vm, vm_id->id, ring->idx);
+		fence_put(id->flushed_updates);
+		id->flushed_updates = fence_get(updates);
 
-	r = amdgpu_sync_fence(ring->adev, sync, id->active);
+		id->pd_gpu_addr = pd_addr;
 
-	if (!r) {
-		fence_put(id->active);
-		id->active = fence_get(fence);
+		list_move_tail(&id->mgr_id->list, &adev->vm_manager.ids_lru);
+		atomic_long_set(&id->mgr_id->owner, (long)id);
+
+		*vm_id = id->mgr_id - adev->vm_manager.ids;
+		*vm_pd_addr = pd_addr;
+		trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, *vm_pd_addr);
 	}
 
 	mutex_unlock(&adev->vm_manager.lock);
@@ -211,35 +234,18 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
  * amdgpu_vm_flush - hardware flush the vm
  *
  * @ring: ring to use for flush
- * @vm: vm we want to flush
- * @updates: last vm update that we waited for
+ * @vmid: vmid number to use
+ * @pd_addr: address of the page directory
  *
- * Flush the vm.
+ * Emit a VM flush when it is necessary.
  */
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
-		     struct amdgpu_vm *vm,
-		     struct fence *updates)
+		     unsigned vmid,
+		     uint64_t pd_addr)
 {
-	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_later;
-
-	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_later) {
-			vm_id->flushed_updates = fence_get(updates);
-			fence_put(flushed_updates);
-		}
-		vm_id->pd_gpu_addr = pd_addr;
-		amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr);
+	if (pd_addr != AMDGPU_VM_NO_FLUSH) {
+		trace_amdgpu_vm_flush(pd_addr, ring->idx, vmid);
+		amdgpu_ring_emit_vm_flush(ring, vmid, pd_addr);
 	}
 }
 
@@ -1284,7 +1290,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	int i, r;
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		vm->ids[i].id = 0;
+		vm->ids[i].mgr_id = NULL;
 		vm->ids[i].flushed_updates = NULL;
 	}
 	vm->va = RB_ROOT;
@@ -1381,13 +1387,13 @@ 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;
+		struct amdgpu_vm_id *id = &vm->ids[i];
 
-		atomic_long_cmpxchg(&adev->vm_manager.ids[id].owner,
-				    (long)vm, 0);
-		fence_put(vm->ids[i].flushed_updates);
+		if (id->mgr_id)
+			atomic_long_cmpxchg(&id->mgr_id->owner,
+					    (long)id, 0);
+		fence_put(id->flushed_updates);
 	}
-
 }
 
 /**

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

@@ -212,7 +212,7 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
 static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 			   struct amdgpu_ib *ib)
 {
-	u32 extra_bits = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf;
+	u32 extra_bits = ib->vm_id & 0xf;
 	u32 next_rptr = ring->wptr + 5;
 
 	while ((next_rptr & 7) != 4)

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

@@ -46,9 +46,6 @@
 #define BONAIRE_GB_ADDR_CONFIG_GOLDEN        0x12010001
 #define HAWAII_GB_ADDR_CONFIG_GOLDEN         0x12011003
 
-#define CIK_RB_BITMAP_WIDTH_PER_SH     2
-#define HAWAII_RB_BITMAP_WIDTH_PER_SH  4
-
 #define AMDGPU_NUM_OF_VMIDS	8
 
 #define		PIPEID(x)					((x) << 0)