Merge tag 'drm-amdkfd-next-2015-01-09' of git://people.freedesktop.org/~gabbayo/linux into drm-next

- Add support for SDMA usermode queues
- Replace logic of sub-allocating from GART buffer in amdkfd. Instead
  of using radeon_sa module, use a new module that is more suited for
  this purpose
- Add the number of watch points to amdkfd topology
- Split a function that did two things into two seperate functions.

* tag 'drm-amdkfd-next-2015-01-09' of git://people.freedesktop.org/~gabbayo/linux:
  drm/amd: Remove old radeon_sa funcs from kfd-->kgd interface
  drm/radeon: Remove old radeon_sa usage from kfd-->kgd interface
  drm/amdkfd: Using new gtt sa in amdkfd
  drm/amdkfd: Allocate gart memory using new interface
  drm/amdkfd: Fixed calculation of gart buffer size
  drm/amdkfd: Add kfd gtt sub-allocator functions
  drm/amdkfd: Add gtt sa related data to kfd_dev struct
  drm/radeon: Impl. new gtt allocate/free functions
  drm/amd: Add new kfd-->kgd interface for gart usage
  drm/radeon: Enable sdma preemption
  drm/amdkfd: Pass queue type to pqm_create_queue()
  drm/amdkfd: Identify SDMA queue in create queue ioctl
  drm/amdkfd: Add SDMA user-mode queues support to QCM
  drm/amdkfd: Add SDMA mqd support
  drm/radeon: Implement SDMA interface functions
  drm/amd: Add SDMA functions to kfd-->kgd interface
  drm/amdkfd: Process-device data creation and lookup split
  drm/amdkfd: Add number of watch points to topology

17 files changed, 1078 insertions, 278 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 7d4974b83af7..4c0b1e42e405 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -192,6 +192,8 @@ static int set_queue_properties_from_user(struct queue_properties *q_properties,
 	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE ||
 		args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
 		q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
+	else if (args->queue_type == KFD_IOC_QUEUE_TYPE_SDMA)
+		q_properties->type = KFD_QUEUE_TYPE_SDMA;
 	else
 		return -ENOTSUPP;
 
@@ -258,8 +260,8 @@ static long kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 			p->pasid,
 			dev->id);
 
-	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, 0,
-				KFD_QUEUE_TYPE_COMPUTE, &queue_id);
+	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties,
+				0, q_properties.type, &queue_id);
 	if (err != 0)
 		goto err_create_queue;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 43884ebd4303..994a9c1bdd04 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -26,12 +26,14 @@
 #include <linux/slab.h>
 #include "kfd_priv.h"
 #include "kfd_device_queue_manager.h"
+#include "kfd_pm4_headers.h"
 
 #define MQD_SIZE_ALIGNED 768
 
 static const struct kfd_device_info kaveri_device_info = {
 	.max_pasid_bits = 16,
 	.ih_ring_entry_size = 4 * sizeof(uint32_t),
+	.num_of_watch_points = 4,
 	.mqd_size_aligned = MQD_SIZE_ALIGNED
 };
 
@@ -66,6 +68,10 @@ static const struct kfd_deviceid supported_devices[] = {
 	{ 0x131D, &kaveri_device_info },	/* Kaveri */
 };
 
+static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
+				unsigned int chunk_size);
+static void kfd_gtt_sa_fini(struct kfd_dev *kfd);
+
 static const struct kfd_device_info *lookup_device_info(unsigned short did)
 {
 	size_t i;
@@ -173,16 +179,39 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		max_num_of_queues_per_process *
 		kfd->device_info->mqd_size_aligned;
 
-	/* add another 512KB for all other allocations on gart */
+	/*
+	 * calculate max size of runlist packet.
+	 * There can be only 2 packets at once
+	 */
+	size += (max_num_of_processes * sizeof(struct pm4_map_process) +
+		max_num_of_processes * max_num_of_queues_per_process *
+		sizeof(struct pm4_map_queues) + sizeof(struct pm4_runlist)) * 2;
+
+	/* Add size of HIQ & DIQ */
+	size += KFD_KERNEL_QUEUE_SIZE * 2;
+
+	/* add another 512KB for all other allocations on gart (HPD, fences) */
 	size += 512 * 1024;
 
-	if (kfd2kgd->init_sa_manager(kfd->kgd, size)) {
+	if (kfd2kgd->init_gtt_mem_allocation(kfd->kgd, size, &kfd->gtt_mem,
+			&kfd->gtt_start_gpu_addr, &kfd->gtt_start_cpu_ptr)) {
 		dev_err(kfd_device,
-			"Error initializing sa manager for device (%x:%x)\n",
-			kfd->pdev->vendor, kfd->pdev->device);
+			"Could not allocate %d bytes for device (%x:%x)\n",
+			size, kfd->pdev->vendor, kfd->pdev->device);
 		goto out;
 	}
 
+	dev_info(kfd_device,
+		"Allocated %d bytes on gart for device(%x:%x)\n",
+		size, kfd->pdev->vendor, kfd->pdev->device);
+
+	/* Initialize GTT sa with 512 byte chunk size */
+	if (kfd_gtt_sa_init(kfd, size, 512) != 0) {
+		dev_err(kfd_device,
+			"Error initializing gtt sub-allocator\n");
+		goto kfd_gtt_sa_init_error;
+	}
+
 	kfd_doorbell_init(kfd);
 
 	if (kfd_topology_add_device(kfd) != 0) {
@@ -241,7 +270,9 @@ device_iommu_pasid_error:
 kfd_interrupt_error:
 	kfd_topology_remove_device(kfd);
 kfd_topology_add_device_error:
-	kfd2kgd->fini_sa_manager(kfd->kgd);
+	kfd_gtt_sa_fini(kfd);
+kfd_gtt_sa_init_error:
+	kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);
 	dev_err(kfd_device,
 		"device (%x:%x) NOT added due to errors\n",
 		kfd->pdev->vendor, kfd->pdev->device);
@@ -256,6 +287,8 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd)
 		amd_iommu_free_device(kfd->pdev);
 		kfd_interrupt_exit(kfd);
 		kfd_topology_remove_device(kfd);
+		kfd_gtt_sa_fini(kfd);
+		kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);
 	}
 
 	kfree(kfd);
@@ -306,3 +339,185 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 		spin_unlock(&kfd->interrupt_lock);
 	}
 }
+
+static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
+				unsigned int chunk_size)
+{
+	unsigned int num_of_bits;
+
+	BUG_ON(!kfd);
+	BUG_ON(!kfd->gtt_mem);
+	BUG_ON(buf_size < chunk_size);
+	BUG_ON(buf_size == 0);
+	BUG_ON(chunk_size == 0);
+
+	kfd->gtt_sa_chunk_size = chunk_size;
+	kfd->gtt_sa_num_of_chunks = buf_size / chunk_size;
+
+	num_of_bits = kfd->gtt_sa_num_of_chunks / BITS_PER_BYTE;
+	BUG_ON(num_of_bits == 0);
+
+	kfd->gtt_sa_bitmap = kzalloc(num_of_bits, GFP_KERNEL);
+
+	if (!kfd->gtt_sa_bitmap)
+		return -ENOMEM;
+
+	pr_debug("kfd: gtt_sa_num_of_chunks = %d, gtt_sa_bitmap = %p\n",
+			kfd->gtt_sa_num_of_chunks, kfd->gtt_sa_bitmap);
+
+	mutex_init(&kfd->gtt_sa_lock);
+
+	return 0;
+
+}
+
+static void kfd_gtt_sa_fini(struct kfd_dev *kfd)
+{
+	mutex_destroy(&kfd->gtt_sa_lock);
+	kfree(kfd->gtt_sa_bitmap);
+}
+
+static inline uint64_t kfd_gtt_sa_calc_gpu_addr(uint64_t start_addr,
+						unsigned int bit_num,
+						unsigned int chunk_size)
+{
+	return start_addr + bit_num * chunk_size;
+}
+
+static inline uint32_t *kfd_gtt_sa_calc_cpu_addr(void *start_addr,
+						unsigned int bit_num,
+						unsigned int chunk_size)
+{
+	return (uint32_t *) ((uint64_t) start_addr + bit_num * chunk_size);
+}
+
+int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size,
+			struct kfd_mem_obj **mem_obj)
+{
+	unsigned int found, start_search, cur_size;
+
+	BUG_ON(!kfd);
+
+	if (size == 0)
+		return -EINVAL;
+
+	if (size > kfd->gtt_sa_num_of_chunks * kfd->gtt_sa_chunk_size)
+		return -ENOMEM;
+
+	*mem_obj = kmalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL);
+	if ((*mem_obj) == NULL)
+		return -ENOMEM;
+
+	pr_debug("kfd: allocated mem_obj = %p for size = %d\n", *mem_obj, size);
+
+	start_search = 0;
+
+	mutex_lock(&kfd->gtt_sa_lock);
+
+kfd_gtt_restart_search:
+	/* Find the first chunk that is free */
+	found = find_next_zero_bit(kfd->gtt_sa_bitmap,
+					kfd->gtt_sa_num_of_chunks,
+					start_search);
+
+	pr_debug("kfd: found = %d\n", found);
+
+	/* If there wasn't any free chunk, bail out */
+	if (found == kfd->gtt_sa_num_of_chunks)
+		goto kfd_gtt_no_free_chunk;
+
+	/* Update fields of mem_obj */
+	(*mem_obj)->range_start = found;
+	(*mem_obj)->range_end = found;
+	(*mem_obj)->gpu_addr = kfd_gtt_sa_calc_gpu_addr(
+					kfd->gtt_start_gpu_addr,
+					found,
+					kfd->gtt_sa_chunk_size);
+	(*mem_obj)->cpu_ptr = kfd_gtt_sa_calc_cpu_addr(
+					kfd->gtt_start_cpu_ptr,
+					found,
+					kfd->gtt_sa_chunk_size);
+
+	pr_debug("kfd: gpu_addr = %p, cpu_addr = %p\n",
+			(uint64_t *) (*mem_obj)->gpu_addr, (*mem_obj)->cpu_ptr);
+
+	/* If we need only one chunk, mark it as allocated and get out */
+	if (size <= kfd->gtt_sa_chunk_size) {
+		pr_debug("kfd: single bit\n");
+		set_bit(found, kfd->gtt_sa_bitmap);
+		goto kfd_gtt_out;
+	}
+
+	/* Otherwise, try to see if we have enough contiguous chunks */
+	cur_size = size - kfd->gtt_sa_chunk_size;
+	do {
+		(*mem_obj)->range_end =
+			find_next_zero_bit(kfd->gtt_sa_bitmap,
+					kfd->gtt_sa_num_of_chunks, ++found);
+		/*
+		 * If next free chunk is not contiguous than we need to
+		 * restart our search from the last free chunk we found (which
+		 * wasn't contiguous to the previous ones
+		 */
+		if ((*mem_obj)->range_end != found) {
+			start_search = found;
+			goto kfd_gtt_restart_search;
+		}
+
+		/*
+		 * If we reached end of buffer, bail out with error
+		 */
+		if (found == kfd->gtt_sa_num_of_chunks)
+			goto kfd_gtt_no_free_chunk;
+
+		/* Check if we don't need another chunk */
+		if (cur_size <= kfd->gtt_sa_chunk_size)
+			cur_size = 0;
+		else
+			cur_size -= kfd->gtt_sa_chunk_size;
+
+	} while (cur_size > 0);
+
+	pr_debug("kfd: range_start = %d, range_end = %d\n",
+		(*mem_obj)->range_start, (*mem_obj)->range_end);
+
+	/* Mark the chunks as allocated */
+	for (found = (*mem_obj)->range_start;
+		found <= (*mem_obj)->range_end;
+		found++)
+		set_bit(found, kfd->gtt_sa_bitmap);
+
+kfd_gtt_out:
+	mutex_unlock(&kfd->gtt_sa_lock);
+	return 0;
+
+kfd_gtt_no_free_chunk:
+	pr_debug("kfd: allocation failed with mem_obj = %p\n", mem_obj);
+	mutex_unlock(&kfd->gtt_sa_lock);
+	kfree(mem_obj);
+	return -ENOMEM;
+}
+
+int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
+{
+	unsigned int bit;
+
+	BUG_ON(!kfd);
+	BUG_ON(!mem_obj);
+
+	pr_debug("kfd: free mem_obj = %p, range_start = %d, range_end = %d\n",
+			mem_obj, mem_obj->range_start, mem_obj->range_end);
+
+	mutex_lock(&kfd->gtt_sa_lock);
+
+	/* Mark the chunks as free */
+	for (bit = mem_obj->range_start;
+		bit <= mem_obj->range_end;
+		bit++)
+		clear_bit(bit, kfd->gtt_sa_bitmap);
+
+	mutex_unlock(&kfd->gtt_sa_lock);
+
+	kfree(mem_obj);
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 924e90c072e5..6806e64c5ffd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -46,9 +46,24 @@ static int set_pasid_vmid_mapping(struct device_queue_manager *dqm,
 static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 					struct queue *q,
 					struct qcm_process_device *qpd);
+
 static int execute_queues_cpsch(struct device_queue_manager *dqm, bool lock);
 static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock);
 
+static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
+					struct queue *q,
+					struct qcm_process_device *qpd);
+
+static void deallocate_sdma_queue(struct device_queue_manager *dqm,
+				unsigned int sdma_queue_id);
+
+static inline
+enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type)
+{
+	if (type == KFD_QUEUE_TYPE_SDMA)
+		return KFD_MQD_TYPE_CIK_SDMA;
+	return KFD_MQD_TYPE_CIK_CP;
+}
 
 static inline unsigned int get_pipes_num(struct device_queue_manager *dqm)
 {
@@ -75,7 +90,6 @@ get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd)
 	nybble = (pdd->lds_base >> 60) & 0x0E;
 
 	return nybble;
-
 }
 
 static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
@@ -190,7 +204,10 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 	*allocated_vmid = qpd->vmid;
 	q->properties.vmid = qpd->vmid;
 
-	retval = create_compute_queue_nocpsch(dqm, q, qpd);
+	if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
+		retval = create_compute_queue_nocpsch(dqm, q, qpd);
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+		retval = create_sdma_queue_nocpsch(dqm, q, qpd);
 
 	if (retval != 0) {
 		if (list_empty(&qpd->queues_list)) {
@@ -203,7 +220,8 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 
 	list_add(&q->list, &qpd->queues_list);
 	dqm->queue_count++;
-
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+		dqm->sdma_queue_count++;
 	mutex_unlock(&dqm->lock);
 	return 0;
 }
@@ -280,8 +298,7 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm,
 				struct queue *q)
 {
 	int retval;
-	struct mqd_manager *mqd;
-
+	struct mqd_manager *mqd, *mqd_sdma;
 	BUG_ON(!dqm || !q || !q->mqd || !qpd);
 
 	retval = 0;
@@ -295,6 +312,12 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm,
 		goto out;
 	}
 
+	mqd_sdma = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_SDMA);
+	if (mqd_sdma == NULL) {
+		mutex_unlock(&dqm->lock);
+		return -ENOMEM;
+	}
+
 	retval = mqd->destroy_mqd(mqd, q->mqd,
 				KFD_PREEMPT_TYPE_WAVEFRONT,
 				QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
@@ -303,7 +326,12 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm,
 	if (retval != 0)
 		goto out;
 
-	deallocate_hqd(dqm, q);
+	if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
+		deallocate_hqd(dqm, q);
+	else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) {
+		dqm->sdma_queue_count--;
+		deallocate_sdma_queue(dqm, q->sdma_id);
+	}
 
 	mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj);
 
@@ -324,7 +352,7 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q)
 	BUG_ON(!dqm || !q || !q->mqd);
 
 	mutex_lock(&dqm->lock);
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE);
+	mqd = dqm->get_mqd_manager(dqm, q->properties.type);
 	if (mqd == NULL) {
 		mutex_unlock(&dqm->lock);
 		return -ENOMEM;
@@ -491,11 +519,8 @@ static int init_pipelines(struct device_queue_manager *dqm,
 	 * because it contains no data when there are no active queues.
 	 */
 
-	err = kfd2kgd->allocate_mem(dqm->dev->kgd,
-				CIK_HPD_EOP_BYTES * pipes_num,
-				PAGE_SIZE,
-				KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-				(struct kgd_mem **) &dqm->pipeline_mem);
+	err = kfd_gtt_sa_allocate(dqm->dev, CIK_HPD_EOP_BYTES * pipes_num,
+					&dqm->pipeline_mem);
 
 	if (err) {
 		pr_err("kfd: error allocate vidmem num pipes: %d\n",
@@ -510,8 +535,7 @@ static int init_pipelines(struct device_queue_manager *dqm,
 
 	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE);
 	if (mqd == NULL) {
-		kfd2kgd->free_mem(dqm->dev->kgd,
-				(struct kgd_mem *) dqm->pipeline_mem);
+		kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem);
 		return -ENOMEM;
 	}
 
@@ -527,7 +551,6 @@ static int init_pipelines(struct device_queue_manager *dqm,
 	return 0;
 }
 
-
 static int init_scheduler(struct device_queue_manager *dqm)
 {
 	int retval;
@@ -557,6 +580,7 @@ static int initialize_nocpsch(struct device_queue_manager *dqm)
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->next_pipe_to_allocate = 0;
+	dqm->sdma_queue_count = 0;
 	dqm->allocated_queues = kcalloc(get_pipes_num(dqm),
 					sizeof(unsigned int), GFP_KERNEL);
 	if (!dqm->allocated_queues) {
@@ -568,6 +592,7 @@ static int initialize_nocpsch(struct device_queue_manager *dqm)
 		dqm->allocated_queues[i] = (1 << QUEUES_PER_PIPE) - 1;
 
 	dqm->vmid_bitmap = (1 << VMID_PER_DEVICE) - 1;
+	dqm->sdma_bitmap = (1 << CIK_SDMA_QUEUES) - 1;
 
 	init_scheduler(dqm);
 	return 0;
@@ -585,8 +610,7 @@ static void uninitialize_nocpsch(struct device_queue_manager *dqm)
 	for (i = 0 ; i < KFD_MQD_TYPE_MAX ; i++)
 		kfree(dqm->mqds[i]);
 	mutex_destroy(&dqm->lock);
-	kfd2kgd->free_mem(dqm->dev->kgd,
-			(struct kgd_mem *) dqm->pipeline_mem);
+	kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem);
 }
 
 static int start_nocpsch(struct device_queue_manager *dqm)
@@ -599,6 +623,77 @@ static int stop_nocpsch(struct device_queue_manager *dqm)
 	return 0;
 }
 
+static int allocate_sdma_queue(struct device_queue_manager *dqm,
+				unsigned int *sdma_queue_id)
+{
+	int bit;
+
+	if (dqm->sdma_bitmap == 0)
+		return -ENOMEM;
+
+	bit = find_first_bit((unsigned long *)&dqm->sdma_bitmap,
+				CIK_SDMA_QUEUES);
+
+	clear_bit(bit, (unsigned long *)&dqm->sdma_bitmap);
+	*sdma_queue_id = bit;
+
+	return 0;
+}
+
+static void deallocate_sdma_queue(struct device_queue_manager *dqm,
+				unsigned int sdma_queue_id)
+{
+	if (sdma_queue_id < 0 || sdma_queue_id >= CIK_SDMA_QUEUES)
+		return;
+	set_bit(sdma_queue_id, (unsigned long *)&dqm->sdma_bitmap);
+}
+
+static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
+				struct qcm_process_device *qpd)
+{
+	uint32_t value = SDMA_ATC;
+
+	if (q->process->is_32bit_user_mode)
+		value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd));
+	else
+		value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64(
+							qpd_to_pdd(qpd)));
+	q->properties.sdma_vm_addr = value;
+}
+
+static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
+					struct queue *q,
+					struct qcm_process_device *qpd)
+{
+	struct mqd_manager *mqd;
+	int retval;
+
+	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_SDMA);
+	if (!mqd)
+		return -ENOMEM;
+
+	retval = allocate_sdma_queue(dqm, &q->sdma_id);
+	if (retval != 0)
+		return retval;
+
+	q->properties.sdma_queue_id = q->sdma_id % CIK_SDMA_QUEUES_PER_ENGINE;
+	q->properties.sdma_engine_id = q->sdma_id / CIK_SDMA_ENGINE_NUM;
+
+	pr_debug("kfd: sdma id is:    %d\n", q->sdma_id);
+	pr_debug("     sdma queue id: %d\n", q->properties.sdma_queue_id);
+	pr_debug("     sdma engine id: %d\n", q->properties.sdma_engine_id);
+
+	retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj,
+				&q->gart_mqd_addr, &q->properties);
+	if (retval != 0) {
+		deallocate_sdma_queue(dqm, q->sdma_id);
+		return retval;
+	}
+
+	init_sdma_vm(dqm, q, qpd);
+	return 0;
+}
+
 /*
  * Device Queue Manager implementation for cp scheduler
  */
@@ -640,6 +735,7 @@ static int initialize_cpsch(struct device_queue_manager *dqm)
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->processes_count = 0;
+	dqm->sdma_queue_count = 0;
 	dqm->active_runlist = false;
 	retval = init_pipelines(dqm, get_pipes_num(dqm), 0);
 	if (retval != 0)
@@ -672,18 +768,14 @@ static int start_cpsch(struct device_queue_manager *dqm)
 	pr_debug("kfd: allocating fence memory\n");
 
 	/* allocate fence memory on the gart */
-	retval = kfd2kgd->allocate_mem(dqm->dev->kgd,
-					sizeof(*dqm->fence_addr),
-					32,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &dqm->fence_mem);
+	retval = kfd_gtt_sa_allocate(dqm->dev, sizeof(*dqm->fence_addr),
+					&dqm->fence_mem);
 
 	if (retval != 0)
 		goto fail_allocate_vidmem;
 
 	dqm->fence_addr = dqm->fence_mem->cpu_ptr;
 	dqm->fence_gpu_addr = dqm->fence_mem->gpu_addr;
-
 	list_for_each_entry(node, &dqm->queues, list)
 		if (node->qpd->pqm->process && dqm->dev)
 			kfd_bind_process_to_device(dqm->dev,
@@ -712,8 +804,7 @@ static int stop_cpsch(struct device_queue_manager *dqm)
 		pdd = qpd_to_pdd(node->qpd);
 		pdd->bound = false;
 	}
-	kfd2kgd->free_mem(dqm->dev->kgd,
-			(struct kgd_mem *) dqm->fence_mem);
+	kfd_gtt_sa_free(dqm->dev, dqm->fence_mem);
 	pm_uninit(&dqm->packets);
 
 	return 0;
@@ -754,6 +845,14 @@ static void destroy_kernel_queue_cpsch(struct device_queue_manager *dqm,
 	mutex_unlock(&dqm->lock);
 }
 
+static void select_sdma_engine_id(struct queue *q)
+{
+	static int sdma_id;
+
+	q->sdma_id = sdma_id;
+	sdma_id = (sdma_id + 1) % 2;
+}
+
 static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 			struct qcm_process_device *qpd, int *allocate_vmid)
 {
@@ -769,7 +868,12 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 
 	mutex_lock(&dqm->lock);
 
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_CP);
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+		select_sdma_engine_id(q);
+
+	mqd = dqm->get_mqd_manager(dqm,
+			get_mqd_type_from_queue_type(q->properties.type));
+
 	if (mqd == NULL) {
 		mutex_unlock(&dqm->lock);
 		return -ENOMEM;
@@ -786,6 +890,9 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 		retval = execute_queues_cpsch(dqm, false);
 	}
 
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+			dqm->sdma_queue_count++;
+
 out:
 	mutex_unlock(&dqm->lock);
 	return retval;
@@ -809,6 +916,14 @@ static int fence_wait_timeout(unsigned int *fence_addr,
 	return 0;
 }
 
+static int destroy_sdma_queues(struct device_queue_manager *dqm,
+				unsigned int sdma_engine)
+{
+	return pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_SDMA,
+			KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false,
+			sdma_engine);
+}
+
 static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock)
 {
 	int retval;
@@ -821,6 +936,15 @@ static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock)
 		mutex_lock(&dqm->lock);
 	if (dqm->active_runlist == false)
 		goto out;
+
+	pr_debug("kfd: Before destroying queues, sdma queue count is : %u\n",
+		dqm->sdma_queue_count);
+
+	if (dqm->sdma_queue_count > 0) {
+		destroy_sdma_queues(dqm, 0);
+		destroy_sdma_queues(dqm, 1);
+	}
+
 	retval = pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_COMPUTE,
 			KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false, 0);
 	if (retval != 0)
@@ -892,13 +1016,16 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
 
 	/* remove queue from list to prevent rescheduling after preemption */
 	mutex_lock(&dqm->lock);
-
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_CP);
+	mqd = dqm->get_mqd_manager(dqm,
+			get_mqd_type_from_queue_type(q->properties.type));
 	if (!mqd) {
 		retval = -ENOMEM;
 		goto failed;
 	}
 
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+		dqm->sdma_queue_count--;
+
 	list_del(&q->list);
 	dqm->queue_count--;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index c3f189e8ae35..554c06ee8892 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -36,6 +36,9 @@
 #define KFD_VMID_START_OFFSET			(8)
 #define VMID_PER_DEVICE				CIK_VMID_NUM
 #define KFD_DQM_FIRST_PIPE			(0)
+#define CIK_SDMA_QUEUES				(4)
+#define CIK_SDMA_QUEUES_PER_ENGINE		(2)
+#define CIK_SDMA_ENGINE_NUM			(2)
 
 struct device_process_node {
 	struct qcm_process_device *qpd;
@@ -130,8 +133,10 @@ struct device_queue_manager {
 	struct list_head	queues;
 	unsigned int		processes_count;
 	unsigned int		queue_count;
+	unsigned int		sdma_queue_count;
 	unsigned int		next_pipe_to_allocate;
 	unsigned int		*allocated_queues;
+	unsigned int		sdma_bitmap;
 	unsigned int		vmid_bitmap;
 	uint64_t		pipelines_addr;
 	struct kfd_mem_obj	*pipeline_mem;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c
index b5791a5c7c06..1a9b355dd114 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c
@@ -137,10 +137,6 @@ int kfd_doorbell_mmap(struct kfd_process *process, struct vm_area_struct *vma)
 	if (dev == NULL)
 		return -EINVAL;
 
-	/* Find if pdd exists for combination of process and gpu id */
-	if (!kfd_get_process_device_data(dev, process, 0))
-		return -EINVAL;
-
 	/* Calculate physical address of doorbell */
 	address = kfd_get_process_doorbells(dev, process);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
index e64aa99e5e41..35b987574633 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
@@ -303,10 +303,11 @@ int kfd_init_apertures(struct kfd_process *process)
 	while ((dev = kfd_topology_enum_kfd_devices(id)) != NULL &&
 		id < NUM_OF_SUPPORTED_GPUS) {
 
-		pdd = kfd_get_process_device_data(dev, process, 1);
-		if (!pdd)
+		pdd = kfd_create_process_device_data(dev, process);
+		if (pdd == NULL) {
+			pr_err("Failed to create process device data\n");
 			return -1;
-
+		}
 		/*
 		 * For 64 bit process aperture will be statically reserved in
 		 * the x86_64 non canonical process address space
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
index 935071410724..0fd8bb7c863e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
@@ -72,11 +72,7 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	if (prop.doorbell_ptr == NULL)
 		goto err_get_kernel_doorbell;
 
-	retval = kfd2kgd->allocate_mem(dev->kgd,
-					queue_size,
-					PAGE_SIZE,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &kq->pq);
+	retval = kfd_gtt_sa_allocate(dev, queue_size, &kq->pq);
 
 	if (retval != 0)
 		goto err_pq_allocate_vidmem;
@@ -84,11 +80,8 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	kq->pq_kernel_addr = kq->pq->cpu_ptr;
 	kq->pq_gpu_addr = kq->pq->gpu_addr;
 
-	retval = kfd2kgd->allocate_mem(dev->kgd,
-					sizeof(*kq->rptr_kernel),
-					32,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &kq->rptr_mem);
+	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->rptr_kernel),
+					&kq->rptr_mem);
 
 	if (retval != 0)
 		goto err_rptr_allocate_vidmem;
@@ -96,11 +89,8 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	kq->rptr_kernel = kq->rptr_mem->cpu_ptr;
 	kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr;
 
-	retval = kfd2kgd->allocate_mem(dev->kgd,
-					sizeof(*kq->wptr_kernel),
-					32,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &kq->wptr_mem);
+	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->wptr_kernel),
+					&kq->wptr_mem);
 
 	if (retval != 0)
 		goto err_wptr_allocate_vidmem;
@@ -145,11 +135,8 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	} else {
 		/* allocate fence for DIQ */
 
-		retval = kfd2kgd->allocate_mem(dev->kgd,
-					sizeof(uint32_t),
-					32,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &kq->fence_mem_obj);
+		retval = kfd_gtt_sa_allocate(dev, sizeof(uint32_t),
+						&kq->fence_mem_obj);
 
 		if (retval != 0)
 			goto err_alloc_fence;
@@ -165,11 +152,11 @@ err_alloc_fence:
 err_init_mqd:
 	uninit_queue(kq->queue);
 err_init_queue:
-	kfd2kgd->free_mem(dev->kgd, (struct kgd_mem *) kq->wptr_mem);
+	kfd_gtt_sa_free(dev, kq->wptr_mem);
 err_wptr_allocate_vidmem:
-	kfd2kgd->free_mem(dev->kgd, (struct kgd_mem *) kq->rptr_mem);
+	kfd_gtt_sa_free(dev, kq->rptr_mem);
 err_rptr_allocate_vidmem:
-	kfd2kgd->free_mem(dev->kgd, (struct kgd_mem *) kq->pq);
+	kfd_gtt_sa_free(dev, kq->pq);
 err_pq_allocate_vidmem:
 	pr_err("kfd: error init pq\n");
 	kfd_release_kernel_doorbell(dev, prop.doorbell_ptr);
@@ -190,10 +177,12 @@ static void uninitialize(struct kernel_queue *kq)
 					QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
 					kq->queue->pipe,
 					kq->queue->queue);
+	else if (kq->queue->properties.type == KFD_QUEUE_TYPE_DIQ)
+		kfd_gtt_sa_free(kq->dev, kq->fence_mem_obj);
 
-	kfd2kgd->free_mem(kq->dev->kgd, (struct kgd_mem *) kq->rptr_mem);
-	kfd2kgd->free_mem(kq->dev->kgd, (struct kgd_mem *) kq->wptr_mem);
-	kfd2kgd->free_mem(kq->dev->kgd, (struct kgd_mem *) kq->pq);
+	kfd_gtt_sa_free(kq->dev, kq->rptr_mem);
+	kfd_gtt_sa_free(kq->dev, kq->wptr_mem);
+	kfd_gtt_sa_free(kq->dev, kq->pq);
 	kfd_release_kernel_doorbell(kq->dev,
 					kq->queue->properties.doorbell_ptr);
 	uninit_queue(kq->queue);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
index adc31474e786..678c33f0a1b8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
@@ -26,6 +26,7 @@
 #include "kfd_priv.h"
 #include "kfd_mqd_manager.h"
 #include "cik_regs.h"
+#include "../../radeon/cikd.h"
 #include "../../radeon/cik_reg.h"
 
 inline void busy_wait(unsigned long ms)
@@ -51,11 +52,8 @@ static int init_mqd(struct mqd_manager *mm, void **mqd,
 
 	pr_debug("kfd: In func %s\n", __func__);
 
-	retval = kfd2kgd->allocate_mem(mm->dev->kgd,
-					sizeof(struct cik_mqd),
-					256,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) mqd_mem_obj);
+	retval = kfd_gtt_sa_allocate(mm->dev, sizeof(struct cik_mqd),
+					mqd_mem_obj);
 
 	if (retval != 0)
 		return -ENOMEM;
@@ -111,18 +109,60 @@ static int init_mqd(struct mqd_manager *mm, void **mqd,
 	return retval;
 }
 
+static int init_mqd_sdma(struct mqd_manager *mm, void **mqd,
+			struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+			struct queue_properties *q)
+{
+	int retval;
+	struct cik_sdma_rlc_registers *m;
+
+	BUG_ON(!mm || !mqd || !mqd_mem_obj);
+
+	retval = kfd_gtt_sa_allocate(mm->dev,
+					sizeof(struct cik_sdma_rlc_registers),
+					mqd_mem_obj);
+
+	if (retval != 0)
+		return -ENOMEM;
+
+	m = (struct cik_sdma_rlc_registers *) (*mqd_mem_obj)->cpu_ptr;
+
+	memset(m, 0, sizeof(struct cik_sdma_rlc_registers));
+
+	*mqd = m;
+	if (gart_addr != NULL)
+		*gart_addr = (*mqd_mem_obj)->gpu_addr;
+
+	retval = mm->update_mqd(mm, m, q);
+
+	return retval;
+}
+
 static void uninit_mqd(struct mqd_manager *mm, void *mqd,
 			struct kfd_mem_obj *mqd_mem_obj)
 {
 	BUG_ON(!mm || !mqd);
-	kfd2kgd->free_mem(mm->dev->kgd, (struct kgd_mem *) mqd_mem_obj);
+	kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
+}
+
+static void uninit_mqd_sdma(struct mqd_manager *mm, void *mqd,
+				struct kfd_mem_obj *mqd_mem_obj)
+{
+	BUG_ON(!mm || !mqd);
+	kfd_gtt_sa_free(mm->