Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "This is the main drm pull, it has a shared branch with some alsa
  crossover but everything should be acked by relevant people.

  New drivers:
     - ATMEL HLCDC driver
     - designware HDMI core support (used in multiple SoCs).

  core:
     - lots more atomic modesetting work, properties and atomic ioctl
       (hidden under option)
     - bridge rework allows support for Samsung exynos chromebooks to
       work finally.
     - some more panels supported

  i915:
     - atomic plane update support
     - DSI uses shared DSI infrastructure
     - Skylake basic support is all merged now
     - component framework used for i915/snd-hda interactions
     - write-combine cpu memory mappings
     - engine init code refactored
     - full ppgtt enabled where execlists are enabled.
     - cherryview rps/gpu turbo and pipe CRC support.

  radeon:
     - indirect draw support for evergreen/cayman
     - SMC and manual fan control for SI/CI
     - Displayport audio support

  amdkfd:
     - SDMA usermode queue support
     - replace suballocator usage with more suitable one
     - rework for allowing interfacing to more than radeon

  nouveau:
     - major renaming in prep for later splitting work
     - merge arm platform driver into nouveau
     - GK20A reclocking support

  msm:
     - conversion to atomic modesetting
     - YUV support for mdp4/5
     - eDP support
     - hw cursor for mdp5

  tegra:
     - conversion to atomic modesetting
     - better suspend/resume support for child devices

  rcar-du:
     - interlaced support

  imx:
     - move to using dw_hdmi shared support
     - mode_fixup support

  sti:
     - DVO support
     - HDMI infoframe support

  exynos:
     - refactoring and cleanup, removed lots of internal unnecessary
       abstraction
     - exynos7 DECON display controller support

  Along with the usual bunch of fixes, cleanups etc"

* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (724 commits)
  drm/radeon: fix voltage setup on hawaii
  drm/radeon/dp: Set EDP_CONFIGURATION_SET for bridge chips if necessary
  drm/radeon: only enable kv/kb dpm interrupts once v3
  drm/radeon: workaround for CP HW bug on CIK
  drm/radeon: Don't try to enable write-combining without PAT
  drm/radeon: use 0-255 rather than 0-100 for pwm fan range
  drm/i915: Clamp efficient frequency to valid range
  drm/i915: Really ignore long HPD pulses on eDP
  drm/exynos: Add DECON driver
  drm/i915: Correct the base value while updating LP_OUTPUT_HOLD in MIPI_PORT_CTRL
  drm/i915: Insert a command barrier on BLT/BSD cache flushes
  drm/i915: Drop vblank wait from intel_dp_link_down
  drm/exynos: fix NULL pointer reference
  drm/exynos: remove exynos_plane_dpms
  drm/exynos: remove mode property of exynos crtc
  drm/exynos: Remove exynos_plane_dpms() call with no effect
  drm/i915: Squelch overzealous uncore reset WARN_ON
  drm/i915: Take runtime pm reference on hangcheck_info
  drm/i915: Correct the IOSF Dev_FN field for IOSF transfers
  drm/exynos: fix DMA_ATTR_NO_KERNEL_MAPPING usage
  ...

23 files changed, 1632 insertions, 740 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile
index 307a309110e6..0f4960148126 100644
--- a/drivers/gpu/drm/amd/amdkfd/Makefile
+++ b/drivers/gpu/drm/amd/amdkfd/Makefile
@@ -7,7 +7,10 @@ ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/
 amdkfd-y	:= kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
 		kfd_pasid.o kfd_doorbell.o kfd_flat_memory.o \
 		kfd_process.o kfd_queue.o kfd_mqd_manager.o \
-		kfd_kernel_queue.o kfd_packet_manager.o \
-		kfd_process_queue_manager.o kfd_device_queue_manager.o
+		kfd_mqd_manager_cik.o kfd_mqd_manager_vi.o \
+		kfd_kernel_queue.o kfd_kernel_queue_cik.o \
+		kfd_kernel_queue_vi.o kfd_packet_manager.o \
+		kfd_process_queue_manager.o kfd_device_queue_manager.o \
+		kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \
 
 obj-$(CONFIG_HSA_AMD)	+= amdkfd.o
diff --git a/drivers/gpu/drm/amd/amdkfd/cik_regs.h b/drivers/gpu/drm/amd/amdkfd/cik_regs.h
index 607fc5ceadbe..01ff332fabd4 100644
--- a/drivers/gpu/drm/amd/amdkfd/cik_regs.h
+++ b/drivers/gpu/drm/amd/amdkfd/cik_regs.h
@@ -168,6 +168,8 @@
 #define	IB_ATC_EN					(1U << 23)
 #define	DEFAULT_MIN_IB_AVAIL_SIZE			(3U << 20)
 
+#define	AQL_ENABLE					1
+
 #define CP_HQD_DEQUEUE_REQUEST				0xC974
 #define	DEQUEUE_REQUEST_DRAIN				1
 #define DEQUEUE_REQUEST_RESET				2
@@ -188,6 +190,17 @@
 #define	MQD_VMID_MASK					(0xf << 0)
 #define	MQD_CONTROL_PRIV_STATE_EN			(1U << 8)
 
+#define	SDMA_RB_VMID(x)					(x << 24)
+#define	SDMA_RB_ENABLE					(1 << 0)
+#define	SDMA_RB_SIZE(x)					((x) << 1) /* log2 */
+#define	SDMA_RPTR_WRITEBACK_ENABLE			(1 << 12)
+#define	SDMA_RPTR_WRITEBACK_TIMER(x)			((x) << 16) /* log2 */
+#define	SDMA_OFFSET(x)					(x << 0)
+#define	SDMA_DB_ENABLE					(1 << 28)
+#define	SDMA_ATC					(1 << 0)
+#define	SDMA_VA_PTR32					(1 << 4)
+#define	SDMA_VA_SHARED_BASE(x)				(x << 8)
+
 #define GRBM_GFX_INDEX					0x30800
 #define	INSTANCE_INDEX(x)				((x) << 0)
 #define	SH_INDEX(x)					((x) << 8)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index fcfdf23e1913..5c50aa8a8908 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -178,6 +178,22 @@ static int set_queue_properties_from_user(struct queue_properties *q_properties,
 		return -EFAULT;
 	}
 
+	if (args->eop_buffer_address &&
+		!access_ok(VERIFY_WRITE,
+			(const void __user *) args->eop_buffer_address,
+			sizeof(uint32_t))) {
+		pr_debug("kfd: can't access eop buffer");
+		return -EFAULT;
+	}
+
+	if (args->ctx_save_restore_address &&
+		!access_ok(VERIFY_WRITE,
+			(const void __user *) args->ctx_save_restore_address,
+			sizeof(uint32_t))) {
+		pr_debug("kfd: can't access ctx save restore buffer");
+		return -EFAULT;
+	}
+
 	q_properties->is_interop = false;
 	q_properties->queue_percent = args->queue_percentage;
 	q_properties->priority = args->queue_priority;
@@ -185,9 +201,16 @@ static int set_queue_properties_from_user(struct queue_properties *q_properties,
 	q_properties->queue_size = args->ring_size;
 	q_properties->read_ptr = (uint32_t *) args->read_pointer_address;
 	q_properties->write_ptr = (uint32_t *) args->write_pointer_address;
+	q_properties->eop_ring_buffer_address = args->eop_buffer_address;
+	q_properties->eop_ring_buffer_size = args->eop_buffer_size;
+	q_properties->ctx_save_restore_area_address =
+			args->ctx_save_restore_address;
+	q_properties->ctx_save_restore_area_size = args->ctx_save_restore_size;
 	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;
 
@@ -214,6 +237,11 @@ static int set_queue_properties_from_user(struct queue_properties *q_properties,
 
 	pr_debug("Queue Format (%d)\n", q_properties->format);
 
+	pr_debug("Queue EOP (0x%llX)\n", q_properties->eop_ring_buffer_address);
+
+	pr_debug("Queue CTX save arex (0x%llX)\n",
+			q_properties->ctx_save_restore_area_address);
+
 	return 0;
 }
 
@@ -235,9 +263,12 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 	if (err)
 		return err;
 
+	pr_debug("kfd: looking for gpu id 0x%x\n", args->gpu_id);
 	dev = kfd_device_by_id(args->gpu_id);
-	if (dev == NULL)
+	if (dev == NULL) {
+		pr_debug("kfd: gpu id 0x%x was not found\n", args->gpu_id);
 		return -EINVAL;
+	}
 
 	mutex_lock(&p->mutex);
 
@@ -251,8 +282,8 @@ static int 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;
 
@@ -385,7 +416,7 @@ static int kfd_ioctl_set_memory_policy(struct file *filep,
 		(args->alternate_policy == KFD_IOC_CACHE_POLICY_COHERENT)
 		   ? cache_policy_coherent : cache_policy_noncoherent;
 
-	if (!dev->dqm->set_cache_memory_policy(dev->dqm,
+	if (!dev->dqm->ops.set_cache_memory_policy(dev->dqm,
 				&pdd->qpd,
 				default_policy,
 				alternate_policy,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 25bc47f3c1cf..5bc32c26b989 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -31,11 +31,20 @@
 #define MQD_SIZE_ALIGNED 768
 
 static const struct kfd_device_info kaveri_device_info = {
+	.asic_family = CHIP_KAVERI,
 	.max_pasid_bits = 16,
 	.ih_ring_entry_size = 4 * sizeof(uint32_t),
 	.mqd_size_aligned = MQD_SIZE_ALIGNED
 };
 
+static const struct kfd_device_info carrizo_device_info = {
+	.asic_family = CHIP_CARRIZO,
+	.max_pasid_bits = 16,
+	.ih_ring_entry_size = 4 * sizeof(uint32_t),
+	.num_of_watch_points = 4,
+	.mqd_size_aligned = MQD_SIZE_ALIGNED
+};
+
 struct kfd_deviceid {
 	unsigned short did;
 	const struct kfd_device_info *device_info;
@@ -64,9 +73,13 @@ static const struct kfd_deviceid supported_devices[] = {
 	{ 0x1318, &kaveri_device_info },	/* Kaveri */
 	{ 0x131B, &kaveri_device_info },	/* Kaveri */
 	{ 0x131C, &kaveri_device_info },	/* Kaveri */
-	{ 0x131D, &kaveri_device_info },	/* Kaveri */
+	{ 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 +186,39 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	size = max_num_of_queues_per_device *
 			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 += (KFD_MAX_NUM_OF_PROCESSES * sizeof(struct pm4_map_process) +
+		max_num_of_queues_per_device *
+		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) {
@@ -209,7 +245,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		goto device_queue_manager_error;
 	}
 
-	if (kfd->dqm->start(kfd->dqm) != 0) {
+	if (kfd->dqm->ops.start(kfd->dqm) != 0) {
 		dev_err(kfd_device,
 			"Error starting queuen manager for device (%x:%x)\n",
 			kfd->pdev->vendor, kfd->pdev->device);
@@ -232,7 +268,9 @@ device_queue_manager_error:
 device_iommu_pasid_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);
@@ -246,6 +284,8 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd)
 		device_queue_manager_uninit(kfd->dqm);
 		amd_iommu_free_device(kfd->pdev);
 		kfd_topology_remove_device(kfd);
+		kfd_gtt_sa_fini(kfd);
+		kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);
 	}
 
 	kfree(kfd);
@@ -256,7 +296,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd)
 	BUG_ON(kfd == NULL);
 
 	if (kfd->init_complete) {
-		kfd->dqm->stop(kfd->dqm);
+		kfd->dqm->ops.stop(kfd->dqm);
 		amd_iommu_set_invalidate_ctx_cb(kfd->pdev, NULL);
 		amd_iommu_free_device(kfd->pdev);
 	}
@@ -277,7 +317,7 @@ int kgd2kfd_resume(struct kfd_dev *kfd)
 			return -ENXIO;
 		amd_iommu_set_invalidate_ctx_cb(kfd->pdev,
 						iommu_pasid_shutdown_callback);
-		kfd->dqm->start(kfd->dqm);
+		kfd->dqm->ops.start(kfd->dqm);
 	}
 
 	return 0;
@@ -288,3 +328,188 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 {
 	/* Process interrupts / schedule work as necessary */
 }
+
+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);
+
+	/* Act like kfree when trying to free a NULL object */
+	if (!mem_obj)
+		return 0;
+
+	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 0fd592799d58..b3589d0e39b9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -26,34 +26,40 @@
 #include <linux/types.h>
 #include <linux/printk.h>
 #include <linux/bitops.h>
+#include <linux/sched.h>
 #include "kfd_priv.h"
 #include "kfd_device_queue_manager.h"
 #include "kfd_mqd_manager.h"
 #include "cik_regs.h"
 #include "kfd_kernel_queue.h"
-#include "../../radeon/cik_reg.h"
 
 /* Size of the per-pipe EOP queue */
 #define CIK_HPD_EOP_BYTES_LOG2 11
 #define CIK_HPD_EOP_BYTES (1U << CIK_HPD_EOP_BYTES_LOG2)
 
-static bool is_mem_initialized;
-
-static int init_memory(struct device_queue_manager *dqm);
 static int set_pasid_vmid_mapping(struct device_queue_manager *dqm,
 					unsigned int pasid, unsigned int vmid);
 
 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 unsigned int get_pipes_num(struct device_queue_manager *dqm)
+static inline
+enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type)
 {
-	BUG_ON(!dqm || !dqm->dev);
-	return dqm->dev->shared_resources.compute_pipe_count;
+	if (type == KFD_QUEUE_TYPE_SDMA)
+		return KFD_MQD_TYPE_SDMA;
+	return KFD_MQD_TYPE_CP;
 }
 
 static inline unsigned int get_first_pipe(struct device_queue_manager *dqm)
@@ -67,61 +73,7 @@ static inline unsigned int get_pipes_num_cpsch(void)
 	return PIPE_PER_ME_CP_SCHEDULING;
 }
 
-static inline unsigned int
-get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd)
-{
-	uint32_t nybble;
-
-	nybble = (pdd->lds_base >> 60) & 0x0E;
-
-	return nybble;
-
-}
-
-static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
-{
-	unsigned int shared_base;
-
-	shared_base = (pdd->lds_base >> 16) & 0xFF;
-
-	return shared_base;
-}
-
-static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble);
-static void init_process_memory(struct device_queue_manager *dqm,
-				struct qcm_process_device *qpd)
-{
-	struct kfd_process_device *pdd;
-	unsigned int temp;
-
-	BUG_ON(!dqm || !qpd);
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-			ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) |
-			DEFAULT_MTYPE(MTYPE_NONCACHED) |
-			APE1_MTYPE(MTYPE_NONCACHED);
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
-
-	if (qpd->pqm->process->is_32bit_user_mode) {
-		temp = get_sh_mem_bases_32(pdd);
-		qpd->sh_mem_bases = SHARED_BASE(temp);
-		qpd->sh_mem_config |= PTR32;
-	} else {
-		temp = get_sh_mem_bases_nybble_64(pdd);
-		qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
-	}
-
-	pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n",
-		qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases);
-}
-
-static void program_sh_mem_settings(struct device_queue_manager *dqm,
+void program_sh_mem_settings(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd)
 {
 	return kfd2kgd->program_sh_mem_settings(dqm->dev->kgd, qpd->vmid,
@@ -200,7 +152,10 @@ 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)) {
@@ -212,7 +167,11 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 	}
 
 	list_add(&q->list, &qpd->queues_list);
-	dqm->queue_count++;
+	if (q->properties.is_active)
+		dqm->queue_count++;
+
+	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
+		dqm->sdma_queue_count++;
 
 	/*
 	 * Unconditionally increment this counter, regardless of the queue's
@@ -229,12 +188,12 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 static int allocate_hqd(struct device_queue_manager *dqm, struct queue *q)
 {
 	bool set;
-	int pipe, bit;
+	int pipe, bit, i;
 
 	set = false;
 
-	for (pipe = dqm->next_pipe_to_allocate; pipe < get_pipes_num(dqm);
-			pipe = (pipe + 1) % get_pipes_num(dqm)) {
+	for (pipe = dqm->next_pipe_to_allocate, i = 0; i < get_pipes_num(dqm);
+			pipe = ((pipe + 1) % get_pipes_num(dqm)), ++i) {
 		if (dqm->allocated_queues[pipe] != 0) {
 			bit = find_first_bit(
 				(unsigned long *)&dqm->allocated_queues[pipe],
@@ -275,7 +234,7 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 
 	BUG_ON(!dqm || !q || !qpd);
 
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE);
+	mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE);
 	if (mqd == NULL)
 		return -ENOMEM;
 
@@ -319,28 +278,44 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm,
 	pr_debug("kfd: In Func %s\n", __func__);
 
 	mutex_lock(&dqm->lock);
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE);
-	if (mqd == NULL) {
-		retval = -ENOMEM;
+
+	if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) {
+		mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE);
+		if (mqd == NULL) {
+			retval = -ENOMEM;
+			goto out;
+		}
+		deallocate_hqd(dqm, q);
+	} else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) {
+		mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA);
+		if (mqd == NULL) {
+			retval = -ENOMEM;
+			goto out;
+		}
+		dqm->sdma_queue_count--;
+		deallocate_sdma_queue(dqm, q->sdma_id);
+	} else {
+		pr_debug("q->properties.type is invalid (%d)\n",
+				q->properties.type);
+		retval = -EINVAL;
 		goto out;
 	}
 
 	retval = mqd->destroy_mqd(mqd, q->mqd,
-				KFD_PREEMPT_TYPE_WAVEFRONT,
+				KFD_PREEMPT_TYPE_WAVEFRONT_RESET,
 				QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
 				q->pipe, q->queue);
 
 	if (retval != 0)
 		goto out;
 
-	deallocate_hqd(dqm, q);
-
 	mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj);
 
 	list_del(&q->list);
 	if (list_empty(&qpd->queues_list))
 		deallocate_vmid(dqm, qpd, q);
-	dqm->queue_count--;
+	if (q->properties.is_active)
+		dqm->queue_count--;
 
 	/*
 	 * Unconditionally decrement this counter, regardless of the queue's
@@ -364,7 +339,8 @@ 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->ops.get_mqd_manager(dqm,
+			get_mqd_type_from_queue_type(q->properties.type));
 	if (mqd == NULL) {
 		mutex_unlock(&dqm->lock);
 		return -ENOMEM;
@@ -415,6 +391,7 @@ static int register_process_nocpsch(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd)
 {
 	struct device_process_node *n;
+	int retval;
 
 	BUG_ON(!dqm || !qpd);
 
@@ -429,12 +406,13 @@ static int register_process_nocpsch(struct device_queue_manager *dqm,
 	mutex_lock(&dqm->lock);
 	list_add(&n->list, &dqm->queues);
 
-	init_process_memory(dqm, qpd);
+	retval = dqm->ops_asic_specific.register_process(dqm, qpd);
+
 	dqm->processes_count++;
 
 	mutex_unlock(&dqm->lock);
 
-	return 0;
+	return retval;
 }
 
 static int unregister_process_nocpsch(struct device_queue_manager *dqm,
@@ -479,48 +457,7 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid,
 						vmid);
 }
 
-static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble)
-{
-	/* In 64-bit mode, we can only control the top 3 bits of the LDS,
-	 * scratch and GPUVM apertures.
-	 * The hardware fills in the remaining 59 bits according to the
-	 * following pattern:
-	 * LDS:		X0000000'00000000 - X0000001'00000000 (4GB)
-	 * Scratch:	X0000001'00000000 - X0000002'00000000 (4GB)
-	 * GPUVM:	Y0010000'00000000 - Y0020000'00000000 (1TB)
-	 *
-	 * (where X/Y is the configurable nybble with the low-bit 0)
-	 *
-	 * LDS and scratch will have the same top nybble programmed in the
-	 * top 3 bits of SH_MEM_BASES.PRIVATE_BASE.
-	 * GPUVM can have a different top nybble programmed in the
-	 * top 3 bits of SH_MEM_BASES.SHARED_BASE.
-	 * We don't bother to support different top nybbles
-	 * for LDS/Scratch and GPUVM.
-	 */
-
-	BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE ||
-		top_address_nybble == 0);
-
-	return PRIVATE_BASE(top_address_nybble << 12) |
-			SHARED_BASE(top_address_nybble << 12);
-}
-
-static int init_memory(struct device_queue_manager *dqm)
-{
-	int i, retval;
-
-	for (i = 8; i < 16; i++)
-		set_pasid_vmid_mapping(dqm, 0, i);
-
-	retval = kfd2kgd->init_memory(dqm->dev->kgd);
-	if (retval == 0)
-		is_mem_initialized = true;
-	return retval;
-}
-
-
-static int init_pipelines(struct device_queue_manager *dqm,
+int init_pipelines(struct device_queue_manager *dqm,
 			unsigned int pipes_num, unsigned int first_pipe)
 {
 	void *hpdptr;
@@ -539,11 +476,8 @@ 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",
@@ -556,10 +490,9 @@ static int init_pipelines(struct device_queue_manager *dqm,
 
 	memset(hpdptr, 0, CIK_HPD_EOP_BYTES * pipes_num);
 
-	mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE);
+	mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_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;
 	}
 
@@ -579,7 +512,6 @@ static int init_pipelines(struct device_queue_manager *dqm,
 	return 0;
 }
 
-
 static int init_scheduler(struct device_queue_manager *dqm)
 {
 	int retval;
@@ -589,11 +521,6 @@ static int init_scheduler(struct device_queue_manager *dqm)
 	pr_debug("kfd: In %s\n", __func__);
 
 	retval = init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm));
-	if (retval != 0)
-		return retval;
-
-	retval = init_memory(dqm);
-
 	return retval;
 }
 
@@ -609,6 +536,7 @@ 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) {
@@ -620,6 +548,7 @@ 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;
@@ -637,8 +566,7 @@ 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)
@@ -651,6 +579,77 @@ 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 >= 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->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_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
  */
@@ -692,8 +691,9 @@ 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);
+	retval = dqm->ops_asic_specific.initialize(dqm);
 	if (retval != 0)
 		goto fail_init_pipelines;
 
@@ -724,18 +724,14 @@ 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,
@@ -764,8 +760,7 @@ 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;
@@ -828,6 +823,14 @@ 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)
 {
@@ -850,7 +853,12 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 		goto out;
 	}
 
-	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->ops.get_mqd_manager(dqm,
+			get_mqd_type_from_queue_type(q->properties.type));
+
 	if (mqd == NULL) {
 		mutex_unlock(&dqm->lock);
 		return -ENOMEM;
@@ -867,6 +875,8 @@ 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++;
 	/*
 	 * Unconditionally increment this counter, regardless of the queue's
 	 * type or whether the queue is active.
@@ -893,12 +903,20 @@ static int fence_wait_timeout(unsigned int *fence_addr,
 			pr_err("kfd: qcm fence wait loop timeout expired\n");
 			return -ETIME;
 		}
-		cpu_relax();
+		schedule();
 	}
 
 	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;
@@ -911,6 +929,15 @@ 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)
@@ -982,15 +1009,19 @@ 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->ops.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--;
+	if (q->properties.is_active)
+		dqm->queue_count--;
 
 	execute_queues_cpsch(dqm, false);
 
@@ -1028,8 +1059,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 				   void __user *alternate_aperture_base,
 				   uint64_t alternate_aperture_size)
 {
-	uint32_t default_mtype;
-	uint32_t ape1_mtype;
+	bool retval;
 
 	pr_debug("kfd: In func %s\n", __func__);
 
@@ -1066,18 +1096,13 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 		qpd->sh_mem_ape1_limit = limit >> 16;
 	}
 
-	default_mtype = (default_policy == cache_policy_coherent) ?
-			MTYPE_NONCACHED :
-			MTYPE_CACHED;
-
-	ape1_mtype = (alternate_policy == cache_policy_coherent) ?
-			MTYPE_NONCACHED :
-			MTYPE_CACHED;
-
-	qpd->sh_mem_config = (qpd->sh_mem_config & PTR32)
-			| ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED)
-			| DEFAULT_MTYPE(default_mtype)
-			| APE1_MTYPE(ape1_mtype);
+	retval = dqm->ops_asic_specific.set_cache_memory_policy(
+			dqm,
+			qpd,
+			default_policy,
+			alternate_policy,
+			alternate_aperture_base,
+			alternate_aperture_size);
 
 	if ((sched_policy == KFD_SCHED_POLICY_NO_HWS) && (qpd->vmid != 0))
 		program_sh_mem_settings(dqm, qpd);
@@ -1087,7 +1112,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 		qpd->sh_mem_ape1_limit);
 
 	mutex_unlock(&dqm->lock);
-	return true;
+	return retval;
 
 out:
 	mutex_unlock(&dqm->lock);
@@ -1100,6 +1125,8 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 
 	BUG_ON(!dev);
 
+	pr_debug("kfd: loading device queue manager\n");
+
 	dqm = kzalloc(sizeof(struct device_queue_manager), GFP_KERNEL);
 	if (!dqm)
 		return NULL;
@@ -1109,40 +1136,50 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 	case KFD_SCHED_POLICY_HWS:
 	case KFD_SCHED_POLICY_HWS_NO_OVERSUBSCRIPTION:
 		/* initialize dqm for cp scheduling */
-		dqm->create_queue = create_queue_cpsch;
-		dqm->initialize = initialize_cpsch;
-		dqm->start = start_cpsch;
-		dqm->stop = stop_cpsch;
-		dqm->destroy_queue = destroy_queue_cpsch;
-		dqm->update_queue = update_queue;
-		dqm->get_mqd_manager = get_mqd_manager_nocpsch;
-		dqm->register_process = register_process_nocpsch;
-		dqm->unregister_process = unregister_process_nocpsch;
-		dqm->uninitialize = uninitialize_nocpsch;
-		dqm->create_kernel_queue = create_kernel_queue_cpsch;
-		dqm->destroy_kernel_queue = destroy_kernel_queue_cpsch;
-		dqm->set_cache_memory_policy = set_cache_memory_policy;
+		dqm->ops.create_queue = create_queue_cpsch;
+		dqm->ops.initialize = initialize_cpsch;
+		dqm->ops.start = start_cpsch;
+		dqm->ops.stop = stop_cpsch;
+		dqm->ops.destroy_queue = destroy_queue_cpsch;
+		dqm->ops.update_queue = update_queue;
+		dqm->ops.get_mqd_manager = get_mqd_manager_nocpsch;
+		dqm->ops.register_process = register_process_nocpsch;
+		dqm->ops.unregister_process = unregister_process_nocpsch;
+		dqm->ops.uninitialize = uninitialize_nocpsch;
+		dqm->ops.create_kernel_queue = create_kernel_queue_cpsch;
+		dqm->ops.destroy_kernel_queue = destroy_kernel_queue_cpsch;
+		dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
 		break;
 	case KFD_SCHED_POLICY_NO_HWS:
 		/* initialize dqm for no cp scheduling */
-		dqm->start = start_nocpsch;
-		dqm->stop = stop_nocpsch;
-		dqm->create_queue = create_queue_nocpsch;
-		dqm->destroy_queue = destroy_queue_nocpsch;
-		dqm->update_queue = update_queue;
-		dqm->get_mqd_manager = get_mqd_manager_nocpsch;
-		dqm->register_process = register_process_nocpsch;
-		dqm->unregister_process = unregister_process_nocpsch;
-		dqm->initialize = initialize_nocpsch;
-		dqm->uninitialize = uninitialize_nocpsch;
-		dqm->set_cache_memory_policy = set_cache_memory_policy;
+		dqm->ops.start = start_nocpsch;
+		dqm->ops.stop = stop_nocpsch;
+		dqm->ops.create_queue = create_queue_nocpsch;
+		dqm->ops.destroy_queue = destroy_queue_nocpsch;
+		dqm->ops.update_queue = update_queue;
+		dqm->ops.get_mqd_manager = get_mqd_manager_nocpsch;
+		dqm->ops.register_process = register_process_nocpsch;
+		dqm->ops.unregister_process = unregister_process_nocpsch;
+		dqm->ops.initialize = initialize_nocpsch;
+		dqm->ops.uninitialize = uninitialize_nocpsch;
+		dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
 		break;
 	default:
 		BUG();
 		break;
 	}
 
-	if (dqm->initialize(dqm) != 0) {
+	switch (dev->device_info->asic_family) {
+	case CHIP_CARRIZO:
+		device_queue_manager_init_vi(&dqm->ops_asic_specific);
+		break;
+
+	case CHIP_KAVERI:
+		device_queue_manager_init_cik(&dqm->ops_asic_specific);
+		break;
+	}
+
+	if (dqm->ops.initialize(dqm) != 0) {
 		kfree(dqm);
 		return NULL;
 	}
@@ -1154,7 +1191,6 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm)
 {
 	BUG_ON(!dqm);
 
-	dqm->uninitialize(dqm);
+	dqm->ops.uninitialize(dqm);
 	kfree(dqm);
 }
-
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 52035bf0c1cb..d64f86cda34f 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;
@@ -43,7 +46,7 @@ struct device_process_node {
 };
 
 /**
- * struct device_queue_manager
+ * struct device_queue_manager_ops
  *
  * @create_queue: Queue creation routine.
  *
@@ -78,15 +81,9 @@ struct device_process_node {
  * @set_cache_memory_policy: Sets memory policy (cached/ non cached) for the
  * memory apertures.
  *
- * This struct is a base class for the kfd queues scheduler in the
- * device level. The device base class should expose the basic operations
- * for queue creation and queue destruction. This base class hides the
- * scheduling mode of the driver and the specific implementation of the
- * concrete device. This class is the only class in the queues scheduler
- * that configures the H/W.
  */
 
-struct device_queue_manager {
+struct device_queue_manager_ops {
 	int	(*create_queue)(struct device_queue_manager *dqm,
 				struct queue *q,
 				struct qcm_process_device *qpd,
@@ -121,7 +118,23 @@ struct device_queue_manager {
 					   enum cache_policy alternate_policy,
 					   void __user *alternate_aperture_base,
 					   uint64_t alternate_aperture_size);
+};
 
+/**
+ * struct device_queue_manager
+ *
+ * This struct is a base class for the kfd queues scheduler in the
+ * device level. The device base class should expose the basic operations
+ * for queue creation and queue destruction. This base class hides the
+ * scheduling mode of the driver and the specific implementation of the
+ * concrete device. This class is the only class in the queues scheduler
+ * that configures the H/W.
+ *
+ */
+
+struct device_queue_manager {
+	struct device_queue_manager_ops ops;
+	struct device_queue_manager_ops ops_asic_specific;
 
 	struct mqd_manager	*mqds[KFD_MQD_TYPE_MAX];
 	struct packet_manager	packets;
@@ -130,9 +143,11 @@ struct device_queue_manager {
 	struct list_head	queues;
 	unsigned int		processes_count;
 	unsigned int		queue_count;
+	unsigned int		sdma_queue_count;
 	unsigned int		total_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;
@@ -142,6 +157,28 @@ struct device_queue_manager {
 	bool			active_runlist;
 };
 
+void device_queue_manager_init_cik(struct device_queue_manager_ops *ops);
+void device_queue_manager_init_vi(struct device_queue_manager_ops *ops);
+void program_sh_mem_settings(struct device_queue_manager *dqm,
+					struct qcm_process_device *qpd);
+int init_pipelines(struct device_queue_manager *dqm,
+		unsigned int pipes_num, unsigned int first_pipe);
+
+extern inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
+{
+	return (pdd->lds_base >> 16) & 0xFF;
+}
+
+extern inline unsigned int
+get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd)
+{
+	return (pdd->lds_base >> 60) & 0x0E;
+}
 
+extern inline unsigned int get_pipes_num(struct device_queue_manager *dqm)
+{
+	BUG_ON(!dqm || !dqm->dev);
+	return dqm->dev->shared_resources.compute_pipe_count;
+}
 
 #endif /* KFD_DEVICE_QUEUE_MANAGER_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
new file mode 100644
index 000000000000..6b072466e2a6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_device_queue_manager.h"
+#include "cik_regs.h"
+
+static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
+static int register_process_cik(struct device_queue_manager *dqm,
+					struct qcm_process_device *qpd);
+static int initialize_cpsch_cik(struct device_queue_manager *dqm);
+
+void device_queue_manager_init_cik(struct device_queue_manager_ops *ops)
+{
+	ops->set_cache_memory_policy = set_cache_memory_policy_cik;
+	ops->register_process = register_process_cik;
+	ops->initialize = initialize_cpsch_cik;
+}
+
+static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble)
+{
+	/* In 64-bit mode, we can only control the top 3 bits of the LDS,
+	 * scratch and GPUVM apertures.
+	 * The hardware fills in the remaining 59 bits according to the
+	 * following pattern:
+	 * LDS:		X0000000'00000000 - X0000001'00000000 (4GB)
+	 * Scratch:	X0000001'00000000 - X0000002'00000000 (4GB)
+	 * GPUVM:	Y0010000'00000000 - Y0020000'00000000 (1TB)
+	 *
+	 * (where X/Y is the configurable nybble with the low-bit 0)
+	 *
+	 * LDS and scratch will have the same top nybble programmed in the
+	 * top 3 bits of SH_MEM_BASES.PRIVATE_BASE.
+	 * GPUVM can have a different top nybble programmed in the
+	 * top 3 bits of SH_MEM_BASES.SHARED_BASE.
+	 * We don't bother to support different top nybbles
+	 * for LDS/Scratch and GPUVM.
+	 */
+
+	BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE ||
+		top_address_nybble == 0);
+
+	return PRIVATE_BASE(top_address_nybble << 12) |
+			SHARED_BASE(top_address_nybble << 12);
+}
+
+static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
+{
+	uint32_t default_mtype;
+	uint32_t ape1_mtype;
+
+	default_mtype = (default_policy == cache_policy_coherent) ?
+			MTYPE_NONCACHED :
+			MTYPE_CACHED;
+
+	ape1_mtype = (alternate_policy == cache_policy_coherent) ?
+			MTYPE_NONCACHED :
+			MTYPE_CACHED;
+
+	qpd->sh_mem_config = (qpd->sh_mem_config & PTR32)
+			| ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED)
+			| DEFAULT_MTYPE(default_mtype)
+			| APE1_MTYPE(ape1_mtype);
+
+	return true;
+}
+
+static int register_process_cik(struct device_queue_manager *dqm,
+		struct qcm_process_device *qpd)
+{
+	struct kfd_process_device *pdd;
+	unsigned int temp;
+
+	BUG_ON(!dqm || !qpd);
+
+	pdd = qpd_to_pdd(qpd);
+
+	/* check if sh_mem_config register already configured */
+	if (qpd->sh_mem_config == 0) {
+		qpd->sh_mem_config =
+			ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) |
+			DEFAULT_MTYPE(MTYPE_NONCACHED) |
+			APE1_MTYPE(MTYPE_NONCACHED);
+		qpd->sh_mem_ape1_limit = 0;
+		qpd->sh_mem_ape1_base = 0;
+	}
+
+	if (qpd->pqm->process->is_32bit_user_mode) {
+		temp = get_sh_mem_bases_32(pdd);
+		qpd->sh_mem_bases = SHARED_BASE(temp);
+		qpd->sh_mem_config |= PTR32;
+	} else {
+		temp = get_sh_mem_bases_nybble_64(pdd);
+		qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
+	}
+
+	pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n",
+		qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases);
+
+	return 0;
+}
+
+static int initialize_cpsch_cik(struct device_queue_manager *dqm)
+{
+	return init_pipelines(dqm, get_pipes_num(dqm), 0);
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
new file mode 100644
index 000000000000..20553dcd257d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
@@ -0,0 +1,64 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_device_queue_manager.h"
+
+static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
+static int register_process_vi(struct device_queue_manager *dqm,
+					struct qcm_process_device *qpd);
+static int initialize_cpsch_vi(struct device_queue_manager *dqm);
+
+void device_queue_manager_init_vi(struct device_queue_manager_ops *ops)
+{
+	pr_warn("amdkfd: VI DQM is not currently supported\n");
+
+	ops->set_cache_memory_policy = set_cache_memory_policy_vi;
+	ops->register_process = register_process_vi;
+	ops->initialize = initialize_cpsch_vi;
+}
+
+static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
+{
+	return false;
+}
+
+static int register_process_vi(struct device_queue_manager *dqm,
+					struct qcm_process_device *qpd)
+{
+	return -1;
+}
+
+static int initialize_cpsch_vi(struct device_queue_manager *dqm)
+{
+	return 0;
+}
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..e415a2a9207e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
@@ -56,8 +56,8 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	switch (type) {
 	case KFD_QUEUE_TYPE_DIQ:
 	case KFD_QUEUE_TYPE_HIQ:
-		kq->mqd = dev->dqm->get_mqd_manager(dev->dqm,
-						KFD_MQD_TYPE_CIK_HIQ);
+		kq->mqd = dev->dqm->ops.get_mqd_manager(dev->dqm,
+						KFD_MQD_TYPE_HIQ);
 		break;
 	default:
 		BUG();
@@ -72,23 +72,19 @@ 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;
 
 	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 = kq->ops_asic_specific.initialize(kq, dev, type, queue_size);
+	if (retval == false)
+		goto err_eop_allocate_vidmem;
+
+	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->rptr_kernel),
+					&kq->rptr_mem);
 
 	if (retval != 0)
 		goto err_rptr_allocate_vidmem;
@@ -96,11 +92,8 @@ 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;
@@ -121,6 +114,8 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	prop.queue_address = kq->pq_gpu_addr;
 	prop.read_ptr = (uint32_t *) kq->rptr_gpu_addr;
 	prop.write_ptr = (uint32_t *) kq->wptr_gpu_addr;
+	prop.eop_ring_buffer_address = kq->eop_gpu_addr;
+	prop.eop_ring_buffer_size = PAGE_SIZE;
 
 	if (init_queue(&kq->queue, prop) != 0)
 		goto err_init_queue;
@@ -145,11 +140,8 @@ 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 +157,13 @@ 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->eop_mem);
+err_eop_allocate_vidmem:
+	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 +184,13 @@ 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);
+	kq->ops_asic_specific.uninitialize(kq);
+	kfd_gtt_sa_free(kq->dev, kq->pq);
 	kfd_release_kernel_doorbell(kq->dev,
 					kq->queue->properties.doorbell_ptr);
 	uninit_queue(kq->queue);
@@ -265,28 +262,6 @@ static void submit_packet(struct kernel_queue *kq)
 				kq->pending_wptr);
 }
 
-static int sync_with_hw(struct kernel_queue *kq, unsigned long timeout_ms)
-{
-	unsigned long org_timeout_ms;
-
-	BUG_ON(!kq);
-
-	org_timeout_ms = timeout_ms;
-	timeout_ms += jiffies * 1000 / HZ;
-	while (*kq->wptr_kernel != *kq->rptr_kernel) {
-		if (time_after(jiffies * 1000 / HZ, timeout_ms)) {
-			pr_err("kfd: kernel_queue %s timeout expired %lu\n",
-				__func__, org_timeout_ms);
-			pr_err("kfd: wptr: %d rptr: %d\n",
-				*kq->wptr_kernel, *kq->rptr_kernel);
-			return -ETIME;
-		}
-		schedule();
-	}
-
-	return 0;
-}
-
 static void rollback_packet(struct kernel_queue *kq)
 {
 	BUG_ON(!kq);
@@ -304,14 +279,23 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 	if (!kq)
 		return NULL;
 
-	kq->initialize = initialize;
-	kq->uninitialize = uninitialize;
-	kq->acquire_packet_buffer = acquire_packet_buffer;
-	kq->submit_packet = submit_packet;
-	kq->sync_with_hw = sync_with_hw;
-	kq->rollback_packet = rollback_packet;
+	kq->ops.initialize = initialize;
+	kq->ops.uninitialize = uninitialize;
+	kq->ops.acquire_packet_buffer = acquire_packet_buffer;
+	kq->ops.submit_packet = submit_packet;
+	kq->ops.rollback_packet = rollback_packet;
+
+	switch (dev->device_info->asic_family) {
+	case CHIP_CARRIZO:
+		kernel_queue_init_vi(&kq->ops_asic_specific);
+		break;
+
+	case CHIP_KAVERI:
+		kernel_queue_init_cik(&kq->ops_asic_specific);
+		break;
+	}
 
-	if (kq->initialize(kq, dev, type, KFD_KERNEL_QUEUE_SIZE) == false) {
+	if (kq->ops.initialize(kq, dev, type, KFD_KERNEL_QUEUE_SIZE) == false) {
 		pr_err("kfd: failed to init kernel queue\n");
 		kfree(kq);
 		return NULL;
@@ -323,7 +307,7 @@ void kernel_queue_uninit(struct kernel_queue *kq)
 {
 	BUG_ON(!kq);
 
-	kq->uninitialize(kq);
+	kq->ops.uninitialize(kq);
 	kfree(kq);
 }
 
@@ -335,19 +319,18 @@ static __attribute__((unused)) void test_kq(struct kfd_dev *dev)
 
 	BUG_ON(!dev);
 
-	pr_debug("kfd: starting kernel queue test\n");
+	pr_err("kfd: starting kernel queue test\n");
 
 	kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_HIQ);
 	BUG_ON(!kq);
 
-	retval = kq->acquire_packet_buffer(kq, 5, &buffer);
+	retval = kq->ops.acquire_packet_buffer(kq, 5, &buffer);
 	BUG_ON(retval != 0);
 	for (i = 0; i < 5; i++)
 		buffer[i] = kq->nop_packet;
-	kq->submit_packet(kq);
-	kq->sync_with_hw(kq, 1000);
+	kq->ops.submit_packet(kq);
 
-	pr_debug("kfd: ending kernel queue test\n");
+	pr_err("kfd: ending kernel queue test\n");
 }
 
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
index dcd2bdb68d44..594053136ee4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
@@ -28,8 +28,31 @@
 #include <linux/types.h>
 #include "kfd_priv.h"
 
-struct kernel_queue {
-	/* interface */
+/**
+ * struct kernel_queue_ops
+ *
+ * @initialize: Initialize a kernel queue, including allocations of GART memory
+ * needed for the queue.
+ *
+ * @uninitialize: Uninitialize a kernel queue and free all its memory usages.
+ *
+ * @acquire_packet_buffer: Returns a pointer to the location in the kernel
+ * queue ring buffer where the calling function can write its packet. It is
+ * Guaranteed that there is enough space for that packet. It also updates the
+ * pending write pointer to that location so subsequent calls to
+ * acquire_packet_buffer will get a correct write pointer
+ *
+ * @submit_packet: Update the write pointer and doorbell of a kernel queue.
+ *
+ * @sync_with_hw: Wait until the write pointer and the read pointer of a kernel
+ * queue are equal, which means the CP has read all the submitted packets.
+ *
+ * @rollback_packet: This routine is called if we failed to build an acquired
+ * packet for some reason. It just overwrites the pending wptr with the current
+ * one
+ *
+ */
+struct kernel_queue_ops {
 	bool	(*initialize)(struct kernel_queue *kq, struct kfd_dev *dev,
 			enum kfd_queue_type type, unsigned int queue_size);
 	void	(*uninitialize)(struct kernel_queue *kq);
@@ -38,9 +61,12 @@ struct kernel_queue {
 					unsigned int **buffer_ptr);
 
 	void	(*submit_packet)(struct kernel_queue *kq);
-	int	(*sync_with_hw)(struct kernel_queue *kq,
-				unsigned long timeout_ms);
 	void	(*rollback_packet)(struct kernel_queue *kq);
+};
+
+struct kernel_queue {
+	struct kernel_queue_ops ops;
+	struct kernel_queue_ops ops_asic_specific;
 
 	/* data */
 	struct kfd_dev		*dev;
@@ -58,6 +84,9 @@ struct kernel_queue {
 	struct kfd_mem_obj	*pq;
 	uint64_t		pq_gpu_addr;
 	uint32_t		*pq_kernel_addr;
+	struct kfd_mem_obj	*eop_mem;
+	uint64_t		eop_gpu_addr;
+	uint32_t		*eop_kernel_addr;
 
 	struct kfd_mem_obj	*fence_mem_obj;
 	uint64_t		fence_gpu_addr;
@@ -66,4 +95,7 @@ struct kernel_queue {
 	struct list_head	list;
 };
 
+void kernel_queue_init_cik(struct kernel_queue_ops *ops);
+void kernel_queue_init_vi(struct kernel_queue_ops *ops);
+
 #endif /* KFD_KERNEL_QUEUE_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_cik.c
new file mode 100644
index 000000000000..a90eb440b1fb
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_cik.c
@@ -0,0 +1,44 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_kernel_queue.h"
+
+static bool initialize_cik(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size);
+static void uninitialize_cik(struct kernel_queue *kq);
+
+void kernel_queue_init_cik(struct kernel_queue_ops *ops)
+{
+	ops->initialize = initialize_cik;
+	ops->uninitialize = uninitialize_cik;
+}
+
+static bool initialize_cik(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size)
+{
+	return true;
+}
+
+static void uninitialize_cik(struct kernel_queue *kq)
+{
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_vi.c
new file mode 100644
index 000000000000..f1d48281e322
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_vi.c
@@ -0,0 +1,56 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_kernel_queue.h"
+
+static bool initialize_vi(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size);
+static void uninitialize_vi(struct kernel_queue *kq);
+
+void kernel_queue_init_vi(struct kernel_queue_ops *ops)
+{
+	ops->initialize = initialize_vi;
+	ops->uninitialize = uninitialize_vi;
+}
+
+static bool initialize_vi(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size)
+{
+	int retval;
+
+	retval = kfd_gtt_sa_allocate(dev, PAGE_SIZE, &kq->eop_mem);
+	if (retval != 0)
+		return false;
+
+	kq->eop_gpu_addr = kq->eop_mem->gpu_addr;
+	kq->eop_kernel_addr = kq->eop_mem->cpu_ptr;
+
+	memset(kq->eop_kernel_addr, 0, PAGE_SIZE);
+
+	return true;
+}
+
+static void uninitialize_vi(struct kernel_queue *kq)
+{
+	kfd_gtt_sa_free(kq->dev, kq->eop_mem);
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_module.c b/drivers/gpu/drm/amd/amdkfd/kfd_module.c
index 1c385c23dd0b..3f34ae16f075 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_module.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_module.c
@@ -29,10 +29,10 @@
 #define KFD_DRIVER_AUTHOR	"AMD Inc. and others"
 
 #define KFD_DRIVER_DESC		"Standalone HSA driver for AMD's GPUs"
-#define KFD_DRIVER_DATE		"20141113"
+#define KFD_DRIVER_DATE		"20150122"
 #define KFD_DRIVER_MAJOR	0
 #define KFD_DRIVER_MINOR	7
-#define KFD_DRIVER_PATCHLEVEL	0
+#define KFD_DRIVER_PATCHLEVEL	1
 
 const struct kfd2kgd_calls *kfd2kgd;
 static const struct kgd2kfd_calls kgd2kfd = {
@@ -48,7 +48,7 @@ static const struct kgd2kfd_calls kgd2kfd = {
 int sched_policy = KFD_SCHED_POLICY_HWS;
 module_param(sched_policy, int, 0444);
 MODULE_PARM_DESC(sched_policy,
-	"Kernel cmdline parameter that defines the amdkfd scheduling policy");
+	"Scheduling policy (0 = HWS (Default), 1 = HWS without over-subscription, 2 = Non-HWS (Used for debugging only)");
 
 int max_num_of_queues_per_device = KFD_MAX_NUM_OF_QUEUES_PER_DEVICE_DEFAULT;
 module_param(max_num_of_queues_per_device, int, 0444);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
index 4c3828cf45bf..b1ef1368c3bb 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
@@ -21,326 +21,17 @@
  *
  */
 
-#include <linux/printk.h>
-#include <linux/slab.h>
 #include "kfd_priv.h"
-#include "kfd_mqd_manager.h"
-#include "cik_regs.h"
-#include "../../radeon/cik_reg.h"
-
-inline void busy_wait(unsigned long ms)
-{
-	while (time_before(jiffies, ms))
-		cpu_relax();
-}
-
-static inline struct cik_mqd *get_mqd(void *mqd)
-{
-	return (struct cik_mqd *)mqd;
-}
-
-static int init_mqd(struct mqd_manager *mm, void **mqd,
-		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
-		struct queue_properties *q)
-{
-	uint64_t addr;
-	struct cik_mqd *m;
-	int retval;
-
-	BUG_ON(!mm || !q || !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);
-
-	if (retval != 0)
-		return -ENOMEM;
-
-	m = (struct cik_mqd *) (*mqd_mem_obj)->cpu_ptr;
-	addr = (*mqd_mem_obj)->gpu_addr;
-
-	memset(m, 0, ALIGN(sizeof(struct cik_mqd), 256));
-
-	m->header = 0xC0310800;
-	m->compute_pipelinestat_enable = 1;
-	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
-
-	/*
-	 * Make sure to use the last queue state saved on mqd when the cp
-	 * reassigns the queue, so when queue is switched on/off (e.g over
-	 * subscription or quantum timeout) the context will be consistent
-	 */
-	m->cp_hqd_persistent_state =
-				DEFAULT_CP_HQD_PERSISTENT_STATE | PRELOAD_REQ;
-
-	m->cp_mqd_control             = MQD_CONTROL_PRIV_STATE_EN;
-	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
-	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
-
-	m->cp_hqd_ib_control = DEFAULT_MIN_IB_AVAIL_SIZE | IB_ATC_EN;
-	/* Although WinKFD writes this, I suspect it should not be necessary */
-	m->cp_hqd_ib_control = IB_ATC_EN | DEFAULT_MIN_IB_AVAIL_SIZE;
-
-	m->cp_hqd_quantum = QUANTUM_EN | QUANTUM_SCALE_1MS |
-				QUANTUM_DURATION(10);
-
-	/*
-	 * Pipe Priority
-	 * Identifies the pipe relative priority when this queue is connected
-	 * to the pipeline. The pipe priority is against the GFX pipe and HP3D.
-	 * In KFD we are using a fixed pipe priority set to CS_MEDIUM.
-	 * 0 = CS_LOW (typically below GFX)
-	 * 1 = CS_MEDIUM (typically between HP3D and GFX
-	 * 2 = CS_HIGH (typically above HP3D)
-	 */
-	m->cp_hqd_pipe_priority = 1;
-	m->cp_hqd_queue_priority = 15;
-
-	*mqd = m;
-	if (gart_addr != NULL)
-		*gart_addr = 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);
-}
-
-static int load_mqd(struct mqd_manager *mm, void *mqd, uint32_t pipe_id,
-			uint32_t queue_id, uint32_t __user *wptr)
-{
-	return kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, wptr);
-
-}
-
-static int update_mqd(struct mqd_manager *mm, void *mqd,
-			struct queue_properties *q)
-{
-	struct cik_mqd *m;
-
-	BUG_ON(!mm || !q || !mqd);
-
-	pr_debug("kfd: In func %s\n", __func__);
-
-	m = get_mqd(mqd);
-	m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
-				DEFAULT_MIN_AVAIL_SIZE | PQ_ATC_EN;
-
-	/*
-	 * Calculating queue size which is log base 2 of actual queue size -1
-	 * dwords and another -1 for ffs
-	 */
-	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
-								- 1 - 1;
-	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
-	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
-	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
-	m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
-	m->cp_hqd_pq_doorbell_control = DOORBELL_EN |
-					DOORBELL_OFFSET(q->doorbell_off);
-
-	m->cp_hqd_vmid = q->vmid;
-
-	if (q->format == KFD_QUEUE_FORMAT_AQL) {
-		m->cp_hqd_iq_rptr = AQL_ENABLE;
-		m->cp_hqd_pq_control |= NO_UPDATE_RPTR;
-	}
-
-	m->cp_hqd_active = 0;
-	q->is_active = false;
-	if (q->queue_size > 0 &&
-			q->queue_address != 0 &&
-			q->queue_percent > 0) {
-		m->cp_hqd_active = 1;
-		q->is_active = true;
-	}
-
-	return 0;
-}
-
-static int destroy_mqd(struct mqd_manager *mm, void *mqd,
-			enum kfd_preempt_type type,
-			unsigned int timeout, uint32_t pipe_id,
-			uint32_t queue_id)
-{
-	return kfd2kgd->hqd_destroy(mm->dev->kgd, type, timeout,
-					pipe_id, queue_id);
-}
-
-static bool is_occupied(struct mqd_manager *mm, void *mqd,
-			uint64_t queue_address,	uint32_t pipe_id,
-			uint32_t queue_id)
-{
-
-	return kfd2kgd->hqd_is_occupied(mm->dev->kgd, queue_address,
-					pipe_id, queue_id);
-
-}
-
-/*
- * HIQ MQD Implementation, concrete implementation for HIQ MQD implementation.
- * The HIQ queue in Kaveri is using the same MQD structure as all the user mode
- * queues but with different initial values.
- */
-
-static int init_mqd_hiq(struct mqd_manager *mm, void **mqd,
-		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
-		struct queue_properties *q)
-{
-	uint64_t addr;
-	struct cik_mqd *m;
-	int retval;
-
-	BUG_ON(!mm || !q || !mqd || !mqd_mem_obj);
-
-	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);
-
-	if (retval != 0)
-		return -ENOMEM;
-
-	m = (struct cik_mqd *) (*mqd_mem_obj)->cpu_ptr;
-	addr = (*mqd_mem_obj)->gpu_addr;
-
-	memset(m, 0, ALIGN(sizeof(struct cik_mqd), 256));
-
-	m->header = 0xC0310800;
-	m->compute_pipelinestat_enable = 1;
-	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
-
-	m->cp_hqd_persistent_state = DEFAULT_CP_HQD_PERSISTENT_STATE |
-					PRELOAD_REQ;
-	m->cp_hqd_quantum = QUANTUM_EN | QUANTUM_SCALE_1MS |
-				QUANTUM_DURATION(10);
-
-	m->cp_mqd_control             = MQD_CONTROL_PRIV_STATE_EN;
-	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
-	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
-
-	m->cp_hqd_ib_control = DEFAULT_MIN_IB_AVAIL_SIZE;
-
-	/*
-	 * Pipe Priority
-	 * Identifies the pipe relative priority when this queue is connected
-	 * to the pipeline. The pipe priority is against the GFX pipe and HP3D.
-	 * In KFD we are using a fixed pipe priority set to CS_MEDIUM.
-	 * 0 = CS_LOW (typically below GFX)
-	 * 1 = CS_MEDIUM (typically between HP3D and GFX
-	 * 2 = CS_HIGH (typically above HP3D)
-	 */
-	m->cp_hqd_pipe_priority = 1;
-	m->cp_hqd_queue_priority = 15;
-
-	*mqd = m;
-	if (gart_addr)
-		*gart_addr = addr;
-	retval = mm->update_mqd(mm, m, q);
-
-	return retval;
-}
-
-static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
-				struct queue_properties *q)
-{
-	struct cik_mqd *m;
-
-	BUG_ON(!mm || !q || !mqd);
-
-	pr_debug("kfd: In func %s\n", __func__);
-
-	m = get_mqd(mqd);
-	m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
-				DEFAULT_MIN_AVAIL_SIZE |
-				PRIV_STATE |
-				KMD_QUEUE;
-
-	/*
-	 * Calculating queue size which is log base 2 of actual queue
-	 * size -1 dwords
-	 */
-	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
-								- 1 - 1;
-	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
-	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
-	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
-	m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
-	m->cp_hqd_pq_doorbell_control = DOORBELL_EN |
-					DOORBELL_OFFSET(q->doorbell_off);
-
-	m->cp_hqd_vmid = q->vmid;
-
-	m->cp_hqd_active = 0;
-	q->is_active = false;
-	if (q->queue_size > 0 &&
-			q->queue_address != 0 &&
-			q->queue_percent > 0) {
-		m->cp_hqd_active = 1;
-		q->is_active = true;
-	}
-
-	return 0;
-}
 
 struct mqd_manager *mqd_manager_init(enum KFD_MQD_TYPE type,
 					struct kfd_dev *dev)
 {
-	struct mqd_manager *mqd;
-
-	BUG_ON(!dev);
-	BUG_ON(type >= KFD_MQD_TYPE_MAX);
-
-	pr_debug("kfd: In func %s\n", __func__);
-
-	mqd = kzalloc(sizeof(struct mqd_manager), GFP_KERNEL);
-	if (!mqd)
-		return NULL;
-
-	mqd->dev = dev;
-
-	switch (type) {
-	case KFD_MQD_TYPE_CIK_CP:
-	case KFD_MQD_TYPE_CIK_COMPUTE:
-		mqd->init_mqd = init_mqd;
-		mqd->uninit_mqd = uninit_mqd;
-		mqd->load_mqd = load_mqd;
-		mqd->update_mqd = update_mqd;
-		mqd->destroy_mqd = destroy_mqd;
-		mqd->is_occupied = is_occupied;
-		break;
-	case KFD_MQD_TYPE_CIK_HIQ:
-		mqd->init_mqd = init_mqd_hiq;
-		mqd->uninit_mqd = uninit_mqd;
-		mqd->load_mqd = load_mqd;
-		mqd->update_mqd = update_mqd_hiq;
-		mqd->destroy_mqd = destroy_mqd;
-		mqd->is_occupied = is_occupied;
-		break;
-	default:
-		kfree(mqd);
-		return NULL;
+	switch (dev->device_info->asic_family) {
+	case CHIP_KAVERI:
+		return mqd_manager_init_cik(type, dev);
+	case CHIP_CARRIZO:
+		return mqd_manager_init_vi(type, dev);
 	}
 
-	return mqd;
+	return NULL;
 }
-
-/* SDMA queues should be implemented here when the cp will supports them */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
new file mode 100644
index 000000000000..a09e18a339f3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
@@ -0,0 +1,450 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/printk.h>
+#include <linux/slab.h>
+#include "kfd_priv.h"
+#include "kfd_mqd_manager.h"
+#include "cik_regs.h"
+#include "cik_structs.h"
+
+static inline struct cik_mqd *get_mqd(void *mqd)
+{
+	return (struct cik_mqd *)mqd;
+}
+
+static int init_mqd(struct mqd_manager *mm, void **mqd,
+		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+		struct queue_properties *q)
+{
+	uint64_t addr;
+	struct cik_mqd *m;
+	int retval;
+
+	BUG_ON(!mm || !q || !mqd);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	retval = kfd_gtt_sa_allocate(mm->dev, sizeof(struct cik_mqd),
+					mqd_mem_obj);
+
+	if (retval != 0)
+		return -ENOMEM;
+
+	m = (struct cik_mqd *) (*mqd_mem_obj)->cpu_ptr;
+	addr = (*mqd_mem_obj)->gpu_addr;
+
+	memset(m, 0, ALIGN(sizeof(struct cik_mqd), 256));
+
+	m->header = 0xC0310800;
+	m->compute_pipelinestat_enable = 1;
+	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
+
+	/*
+	 * Make sure to use the last queue state saved on mqd when the cp
+	 * reassigns the queue, so when queue is switched on/off (e.g over
+	 * subscription or quantum timeout) the context will be consistent
+	 */
+	m->cp_hqd_persistent_state =
+				DEFAULT_CP_HQD_PERSISTENT_STATE | PRELOAD_REQ;
+
+	m->cp_mqd_control             = MQD_CONTROL_PRIV_STATE_EN;
+	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
+	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
+
+	m->cp_hqd_ib_control = DEFAULT_MIN_IB_AVAIL_SIZE | IB_ATC_EN;
+	/* Although WinKFD writes this, I suspect it should not be necessary */
+	m->cp_hqd_ib_control = IB_ATC_EN | DEFAULT_MIN_IB_AVAIL_SIZE;
+
+	m->cp_hqd_quantum = QUANTUM_EN | QUANTUM_SCALE_1MS |
+				QUANTUM_DURATION(10);
+
+	/*
+	 * Pipe Priority
+	 * Identifies the pipe relative priority when this queue is connected
+	 * to the pipeline. The pipe priority is against the GFX pipe and HP3D.
+	 * In KFD we are using a fixed pipe priority set to CS_MEDIUM.
+	 * 0 = CS_LOW (typically below GFX)
+	 * 1 = CS_MEDIUM (typically between HP3D and GFX
+	 * 2 = CS_HIGH (typically above HP3D)
+	 */
+	m->cp_hqd_pipe_priority = 1;
+	m->cp_hqd_queue_priority = 15;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL)
+		m->cp_hqd_iq_rptr = AQL_ENABLE;
+
+	*mqd = m;
+	if (gart_addr != NULL)
+		*gart_addr = addr;
+	retval = mm->update_mqd(mm, m, q);
+
+	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);
+	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->dev, mqd_mem_obj);
+}
+
+static int load_mqd(struct mqd_manager *mm, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	return kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, wptr);
+}
+
+static int load_mqd_sdma(struct mqd_manager *mm, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr)
+{
+	return kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd);
+}
+
+static int update_mqd(struct mqd_manager *mm, void *mqd,
+			struct queue_properties *q)
+{
+	struct cik_mqd *m;
+
+	BUG_ON(!mm || !q || !mqd);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	m = get_mqd(mqd);
+	m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
+				DEFAULT_MIN_AVAIL_SIZE | PQ_ATC_EN;
+
+	/*
+	 * Calculating queue size which is log base 2 of actual queue size -1
+	 * dwords and another -1 for ffs
+	 */
+	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
+								- 1 - 1;
+	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
+	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
+	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_doorbell_control = DOORBELL_EN |
+					DOORBELL_OFFSET(q->doorbell_off);
+
+	m->cp_hqd_vmid = q->vmid;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL) {
+		m->cp_hqd_pq_control |= NO_UPDATE_RPTR;
+	}
+
+	m->cp_hqd_active = 0;
+	q->is_active = false;
+	if (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0) {
+		m->cp_hqd_active = 1;
+		q->is_active = true;
+	}
+
+	return 0;
+}
+
+static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
+				struct queue_properties *q)
+{
+	struct cik_sdma_rlc_registers *m;
+
+	BUG_ON(!mm || !mqd || !q);
+
+	m = get_sdma_mqd(mqd);
+	m->sdma_rlc_rb_cntl =
+		SDMA_RB_SIZE((ffs(q->queue_size / sizeof(unsigned int)))) |
+		SDMA_RB_VMID(q->vmid) |
+		SDMA_RPTR_WRITEBACK_ENABLE |
+		SDMA_RPTR_WRITEBACK_TIMER(6);
+
+	m->sdma_rlc_rb_base = lower_32_bits(q->queue_address >> 8);
+	m->sdma_rlc_rb_base_hi = upper_32_bits(q->queue_address >> 8);
+	m->sdma_rlc_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->sdma_rlc_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->sdma_rlc_doorbell = SDMA_OFFSET(q->doorbell_off) | SDMA_DB_ENABLE;
+	m->sdma_rlc_virtual_addr = q->sdma_vm_addr;
+
+	m->sdma_engine_id = q->sdma_engine_id;
+	m->sdma_queue_id = q->sdma_queue_id;
+
+	q->is_active = false;
+	if (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0) {
+		m->sdma_rlc_rb_cntl |= SDMA_RB_ENABLE;
+		q->is_active = true;
+	}
+
+	return 0;
+}
+
+static int destroy_mqd(struct mqd_manager *mm, void *mqd,
+			enum kfd_preempt_type type,
+			unsigned int timeout, uint32_t pipe_id,
+			uint32_t queue_id)
+{
+	return kfd2kgd->hqd_destroy(mm->dev->kgd, type, timeout,
+					pipe_id, queue_id);
+}
+
+/*
+ * preempt type here is ignored because there is only one way
+ * to preempt sdma queue
+ */
+static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd,
+				enum kfd_preempt_type type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	return kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout);
+}
+
+static bool is_occupied(struct mqd_manager *mm, void *mqd,
+			uint64_t queue_address,	uint32_t pipe_id,
+			uint32_t queue_id)
+{
+
+	return kfd2kgd->hqd_is_occupied(mm->dev->kgd, queue_address,
+					pipe_id, queue_id);
+
+}
+
+static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd,
+			uint64_t queue_address,	uint32_t pipe_id,
+			uint32_t queue_id)
+{
+	return kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd);
+}
+
+/*
+ * HIQ MQD Implementation, concrete implementation for HIQ MQD implementation.
+ * The HIQ queue in Kaveri is using the same MQD structure as all the user mode
+ * queues but with different initial values.
+ */
+
+static int init_mqd_hiq(struct mqd_manager *mm, void **mqd,
+		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+		struct queue_properties *q)
+{
+	uint64_t addr;
+	struct cik_mqd *m;
+	int retval;
+
+	BUG_ON(!mm || !q || !mqd || !mqd_mem_obj);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	retval = kfd_gtt_sa_allocate(mm->dev, sizeof(struct cik_mqd),
+					mqd_mem_obj);
+
+	if (retval != 0)
+		return -ENOMEM;
+
+	m = (struct cik_mqd *) (*mqd_mem_obj)->cpu_ptr;
+	addr = (*mqd_mem_obj)->gpu_addr;
+
+	memset(m, 0, ALIGN(sizeof(struct cik_mqd), 256));
+
+	m->header = 0xC0310800;
+	m->compute_pipelinestat_enable = 1;
+	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
+
+	m->cp_hqd_persistent_state = DEFAULT_CP_HQD_PERSISTENT_STATE |
+					PRELOAD_REQ;
+	m->cp_hqd_quantum = QUANTUM_EN | QUANTUM_SCALE_1MS |
+				QUANTUM_DURATION(10);
+
+	m->cp_mqd_control             = MQD_CONTROL_PRIV_STATE_EN;
+	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
+	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
+
+	m->cp_hqd_ib_control = DEFAULT_MIN_IB_AVAIL_SIZE;
+
+	/*
+	 * Pipe Priority
+	 * Identifies the pipe relative priority when this queue is connected
+	 * to the pipeline. The pipe priority is against the GFX pipe and HP3D.
+	 * In KFD we are using a fixed pipe priority set to CS_MEDIUM.
+	 * 0 = CS_LOW (typically below GFX)
+	 * 1 = CS_MEDIUM (typically between HP3D and GFX
+	 * 2 = CS_HIGH (typically above HP3D)
+	 */
+	m->cp_hqd_pipe_priority = 1;
+	m->cp_hqd_queue_priority = 15;
+
+	*mqd = m;
+	if (gart_addr)
+		*gart_addr = addr;
+	retval = mm->update_mqd(mm, m, q);
+
+	return retval;
+}
+
+static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
+				struct queue_properties *q)
+{
+	struct cik_mqd *m;
+
+	BUG_ON(!mm || !q || !mqd);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	m = get_mqd(mqd);
+	m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
+				DEFAULT_MIN_AVAIL_SIZE |
+				PRIV_STATE |
+				KMD_QUEUE;
+
+	/*
+	 * Calculating queue size which is log base 2 of actual queue
+	 * size -1 dwords
+	 */
+	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
+								- 1 - 1;
+	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
+	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
+	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_doorbell_control = DOORBELL_EN |
+					DOORBELL_OFFSET(q->doorbell_off);
+
+	m->cp_hqd_vmid = q->vmid;
+
+	m->cp_hqd_active = 0;
+	q->is_active = false;
+	if (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0) {
+		m->cp_hqd_active = 1;
+		q->is_active = true;
+	}
+
+	return 0;
+}
+
+struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	struct cik_sdma_rlc_registers *m;
+
+	BUG_ON(!mqd);
+
+	m = (struct cik_sdma_rlc_registers *)mqd;
+
+	return m;
+}
+
+struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
+		struct kfd_dev *dev)
+{
+	struct mqd_manager *mqd;
+
+	BUG_ON(!dev);
+	BUG_ON(type >= KFD_MQD_TYPE_MAX);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	mqd = kzalloc(sizeof(struct mqd_manager), GFP_KERNEL);
+	if (!mqd)
+		return NULL;
+
+	mqd->dev = dev;
+
+	switch (type) {
+	case KFD_MQD_TYPE_CP:
+	case KFD_MQD_TYPE_COMPUTE:
+		mqd->init_mqd = init_mqd;
+		mqd->uninit_mqd = uninit_mqd;
+		mqd->load_mqd = load_mqd;
+		mqd->update_mqd = update_mqd;
+		mqd->destroy_mqd = destroy_mqd;
+		mqd->is_occupied = is_occupied;
+		break;
+	case KFD_MQD_TYPE_HIQ:
+		mqd->init_mqd = init_mqd_hiq;
+		mqd->uninit_mqd = uninit_mqd;
+		mqd->load_mqd = load_mqd;
+		mqd->update_mqd = update_mqd_hiq;
+		mqd->destroy_mqd = destroy_mqd;
+		mqd->is_occupied = is_occupied;
+		break;
+	case KFD_MQD_TYPE_SDMA:
+		mqd->init_mqd = init_mqd_sdma;
+		mqd->uninit_mqd = uninit_mqd_sdma;
+		mqd->load_mqd = load_mqd_sdma;
+		mqd->update_mqd = update_mqd_sdma;
+		mqd->destroy_mqd = destroy_mqd_sdma;
+		mqd->is_occupied = is_occupied_sdma;
+		break;
+	default:
+		kfree(mqd);
+		return NULL;
+	}
+
+	return mqd;
+}
+
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
new file mode 100644
index 000000000000..b3a7e3ba1e38
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
@@ -0,0 +1,33 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/printk.h>
+#include "kfd_priv.h"
+#include "kfd_mqd_manager.h"
+
+struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
+					struct kfd_dev *dev)
+{
+	pr_warn("amdkfd: VI MQD is not currently supported\n");
+	return NULL;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index 5ce9233d2004..e2533d875f43 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -97,11 +97,8 @@ static int pm_allocate_runlist_ib(struct packet_manager *pm,
 
 	pm_calc_rlib_size(pm, rl_buffer_size, is_over_subscription);
 
-	retval = kfd2kgd->allocate_mem(pm->dqm->dev->kgd,
-					*rl_buffer_size,
-					PAGE_SIZE,
-					KFD_MEMPOOL_SYSTEM_WRITECOMBINE,
-					(struct kgd_mem **) &pm->ib_buffer_obj);
+	retval = kfd_gtt_sa_allocate(pm->dqm->dev, *rl_buffer_size,
+					&pm->ib_buffer_obj);
 
 	if (retval != 0) {
 		pr_err("kfd: failed to allocate runlist IB\n");
@@ -351,7 +348,7 @@ int pm_send_set_resources(struct packet_manager *pm,
 	pr_debug("kfd: In func %s\n", __func__);
 
 	mutex_lock(&pm->lock);
-	pm->priv_queue->acquire_packet_buffer(pm->priv_queue,
+	pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
 					sizeof(*packet) / sizeof(uint32_t),
 			(unsigned int **)&packet);
 	if (packet == NULL) {
@@ -378,8 +375,7 @@ int pm_send_set_resources(struct packet_manager *pm,
 	packet->queue_mask_lo = lower_32_bits(res->queue_mask);
 	packet->queue_mask_hi = upper_32_bits(res->queue_mask);
 
-	pm->priv_queue->submit_packet(pm->priv_queue);
-	pm->priv_queue->sync_with_hw(pm->priv_queue, KFD_HIQ_TIMEOUT);
+	pm->priv_queue->ops.submit_packet(pm->priv_queue);
 
 	mutex_unlock(&pm->lock);
 
@@ -405,7 +401,7 @@ int pm_send_runlist(struct packet_manager *pm, struct list_head *dqm_queues)
 	packet_size_dwords = sizeof(struct pm4_runlist) / sizeof(uint32_t);
 	mutex_lock(&pm->lock);
 
-	retval = pm->priv_queue->acquire_packet_buffer(pm->priv_queue,
+	retval = pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
 					packet_size_dwords, &rl_buffer);
 	if (retval != 0)
 		goto fail_acquire_packet_buffer;
@@ -415,15 +411,14 @@ int pm_send_runlist(struct packet_manager *pm, struct list_head *dqm_queues)
 	if (retval != 0)
 		goto fail_create_runlist;
 
-	pm->priv_queue->submit_packet(pm->priv_queue);
-	pm->priv_queue->sync_with_hw(pm->priv_queue, KFD_HIQ_TIMEOUT);
+	pm->priv_queue->ops.submit_packet(pm->priv_queue);
 
 	mutex_unlock(&pm->lock);
 
 	return retval;
 
 fail_create_runlist:
-	pm->priv_queue->rollback_packet(pm->priv_queue);
+	pm->priv_queue->ops.rollback_packet(pm->priv_queue);
 fail_acquire_packet_buffer:
 	mutex_unlock(&pm->lock);
 fail_create_runlist_ib:
@@ -441,7 +436,7 @@ int pm_send_query_status(struct packet_manager *pm, uint64_t fence_address,
 	BUG_ON(!pm || !fence_address);
 
 	mutex_lock(&pm->lock);
-	retval = pm->priv_queue->acquire_packet_buffer(
+	retval = pm->priv_queue->ops.acquire_packet_buffer(
 			pm->priv_queue,
 			sizeof(struct pm4_query_status) / sizeof(uint32_t),
 			(unsigned int **)&packet);
@@ -462,8 +457,7 @@ int pm_send_query_status(struct packet_manager *pm, uint64_t fence_address,
 	packet->data_hi = upper_32_bits((uint64_t)fence_value);
 	packet->data_lo = lower_32_bits((uint64_t)fence_value);
 
-	pm->priv_queue->submit_packet(pm->priv_queue);
-	pm->priv_queue->sync_with_hw(pm->priv_queue, KFD_HIQ_TIMEOUT);
+	pm->priv_queue->ops.submit_packet(pm->priv_queue);
 	mutex_unlock(&pm->lock);
 
 	return 0;
@@ -485,7 +479,7 @@ int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type,
 	BUG_ON(!pm);
 
 	mutex_lock(&pm->lock);
-	retval = pm->priv_queue->acquire_packet_buffer(
+	retval = pm->priv_queue->ops.acquire_packet_buffer(
 			pm->priv_queue,
 			sizeof(struct pm4_unmap_queues) / sizeof(uint32_t),
 			&buffer);
@@ -540,8 +534,7 @@ int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type,
 		break;
 	};
 
-	pm->priv_queue->submit_packet(pm->priv_queue);
-	pm->priv_queue->sync_with_hw(pm->priv_queue, KFD_HIQ_TIMEOUT);
+	pm->priv_queue->ops.submit_packet(pm->priv_queue);
 
 	mutex_unlock(&pm->lock);
 	return 0;
@@ -557,8 +550,7 @@ void pm_release_ib(struct packet_manager *pm)
 
 	mutex_lock(&pm->lock);
 	if (pm->allocated) {
-		kfd2kgd->free_mem(pm->dqm->dev->kgd,
-				(struct kgd_mem *) pm->ib_buffer_obj);
+		kfd_gtt_sa_free(pm->dqm->dev, pm->ib_buffer_obj);
 		pm->allocated = false;
 	}
 	mutex_unlock(&pm->lock);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 96dc10e8904a..5a44f2fecf38 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -103,12 +103,26 @@ enum cache_policy {
 	cache_policy_noncoherent
 };
 
+enum asic_family_type {
+	CHIP_KAVERI = 0,
+	CHIP_CARRIZO
+};
+
 struct kfd_device_info {
+	unsigned int asic_family;
 	unsigned int max_pasid_bits;
 	size_t ih_ring_entry_size;
+	uint8_t num_of_watch_points;
 	uint16_t mqd_size_aligned;
 };
 
+struct kfd_mem_obj {
+	uint32_t range_start;
+	uint32_t range_end;
+	uint64_t gpu_addr;
+	uint32_t *cpu_ptr;
+};
+
 struct kfd_dev {
 	struct kgd_dev *kgd;
 
@@ -134,6 +148,14 @@ struct kfd_dev {
 
 	struct kgd2kfd_shared_resources shared_resources;
 
+	void *gtt_mem;
+	uint64_t gtt_start_gpu_addr;
+	void *gtt_start_cpu_ptr;
+	void *gtt_sa_bitmap;
+	struct mutex gtt_sa_lock;
+	unsigned int gtt_sa_chunk_size;
+	unsigned int gtt_sa_num_of_chunks;
+
 	/* QCM Device instance */
 	struct device_queue_manager *dqm;
 
@@ -149,12 +171,6 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd);
 
 extern const struct kfd2kgd_calls *kfd2kgd;
 
-struct kfd_mem_obj {
-	void *bo;
-	uint64_t gpu_addr;
-	uint32_t *cpu_ptr;
-};
-
 enum kfd_mempool {
 	KFD_MEMPOOL_SYSTEM_CACHEABLE = 1,
 	KFD_MEMPOOL_SYSTEM_WRITECOMBINE = 2,
@@ -272,6 +288,15 @@ struct queue_properties {
 	bool is_active;
 	/* Not relevant for user mode queues in cp scheduling */
 	unsigned int vmid;
+	/* Relevant only for sdma queues*/
+	uint32_t sdma_engine_id;
+	uint32_t sdma_queue_id;
+	uint32_t sdma_vm_addr;
+	/* Relevant only for VI */
+	uint64_t eop_ring_buffer_address;
+	uint32_t eop_ring_buffer_size;
+	uint64_t ctx_save_restore_area_address;
+	uint32_t ctx_save_restore_area_size;
 };
 
 /**
@@ -314,6 +339,8 @@ struct queue {
 	uint32_t pipe;
 	uint32_t queue;
 
+	unsigned int sdma_id;
+
 	struct kfd_process	*process;
 	struct kfd_dev		*device;
 };
@@ -322,10 +349,10 @@ struct queue {
  * Please read the kfd_mqd_manager.h description.
  */
 enum KFD_MQD_TYPE {
-	KFD_MQD_TYPE_CIK_COMPUTE = 0, /* for no cp scheduling */
-	KFD_MQD_TYPE_CIK_HIQ, /* for hiq */
-	KFD_MQD_TYPE_CIK_CP, /* for cp queues and diq */
-	KFD_MQD_TYPE_CIK_SDMA, /* for sdma queues */
+	KFD_MQD_TYPE_COMPUTE = 0,	/* for no cp scheduling */
+	KFD_MQD_TYPE_HIQ,		/* for hiq */
+	KFD_MQD_TYPE_CP,		/* for cp queues and diq */
+	KFD_MQD_TYPE_SDMA,		/* for sdma queues */
 	KFD_MQD_TYPE_MAX
 };
 
@@ -477,8 +504,9 @@ struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
 							struct kfd_process *p);
 void kfd_unbind_process_from_device(struct kfd_dev *dev, unsigned int pasid);
 struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
-							struct kfd_process *p,
-							int create_pdd);
+							struct kfd_process *p);
+struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
+							struct kfd_process *p);
 
 /* Process device data iterator */
 struct kfd_process_device *kfd_get_first_process_device_data(struct kfd_process *p);
@@ -506,6 +534,13 @@ unsigned int kfd_queue_id_to_doorbell(struct kfd_dev *kfd,
 					struct kfd_process *process,
 					unsigned int queue_id);
 
+/* GTT Sub-Allocator */
+
+int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size,
+			struct kfd_mem_obj **mem_obj);
+
+int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj);
+
 extern struct device *kfd_device;
 
 /* Topology */
@@ -530,6 +565,8 @@ int kfd_init_apertures(struct kfd_process *process);
 /* Queue Context Management */
 inline uint32_t lower_32(uint64_t x);
 inline uint32_t upper_32(uint64_t x);
+struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd);
+inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m);
 
 int init_queue(struct queue **q, struct queue_properties properties);
 void uninit_queue(struct queue *q);
@@ -538,6 +575,10 @@ void print_queue(struct queue *q);
 
 struct mqd_manager *mqd_manager_init(enum KFD_MQD_TYPE type,
 					struct kfd_dev *dev);
+struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
+		struct kfd_dev *dev);
+struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
+		struct kfd_dev *dev);
 struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev);
 void device_queue_manager_uninit(struct device_queue_manager *dqm);
 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 3c76ef05cbcf..a369c149d172 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -311,24 +311,29 @@ err_alloc_process:
 }
 
 struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
-							struct kfd_process *p,
-							int create_pdd)
+							struct kfd_process *p)
 {
 	struct kfd_process_device *pdd = NULL;
 
 	list_for_each_entry(pdd, &p->per_device_data, per_device_list)
 		if (pdd->dev == dev)
-			return pdd;
-
-	if (create_pdd) {
-		pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
-		if (pdd != NULL) {
-			pdd->dev = dev;
-			INIT_LIST_HEAD(&pdd->qpd.queues_list);
-			INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
-			pdd->qpd.dqm = dev->dqm;
-			list_add(&pdd->per_device_list, &p->per_device_data);
-		}
+			break;
+
+	return pdd;
+}
+
+struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
+							struct kfd_process *p)
+{
+	struct kfd_process_device *pdd = NULL;
+
+	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
+	if (pdd != NULL) {
+		pdd->dev = dev;
+		INIT_LIST_HEAD(&pdd->qpd.queues_list);
+		INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
+		pdd->qpd.dqm = dev->dqm;
+		list_add(&pdd->per_device_list, &p->per_device_data);
 	}
 
 	return pdd;
@@ -344,11 +349,14 @@ struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
 struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
 							struct kfd_process *p)
 {
-	struct kfd_process_device *pdd = kfd_get_process_device_data(dev, p, 1);
+	struct kfd_process_device *pdd;
 	int err;
 
-	if (pdd == NULL)
+	pdd = kfd_get_process_device_data(dev, p);
+	if (!pdd) {
+		pr_err("Process device data doesn't exist\n");
 		return ERR_PTR(-ENOMEM);
+	}
 
 	if (pdd->bound)
 		return pdd;
@@ -384,7 +392,7 @@ void kfd_unbind_process_from_device(struct kfd_dev *dev, unsigned int pasid)
 
 	pqm_uninit(&p->pqm);
 
-	pdd = kfd_get_process_device_data(dev, p, 0);
+	pdd = kfd_get_process_device_data(dev, p);
 
 	/*
 	 * Just mark pdd as unbound, because we still need it to call
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 2fda1927bff7..530b82c4e78b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -128,7 +128,6 @@ static int create_cp_queue(struct process_queue_manager *pqm,
 	/* let DQM handle it*/
 	q_properties->vmid = 0;
 	q_properties->queue_id = qid;
-	q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
 
 	retval = init_queue(q, *q_properties);
 	if (retval != 0)
@@ -167,8 +166,11 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	q = NULL;
 	kq = NULL;
 
-	pdd = kfd_get_process_device_data(dev, pqm->process, 1);
-	BUG_ON(!pdd);
+	pdd = kfd_get_process_device_data(dev, pqm->process);
+	if (!pdd) {
+		pr_err("Process device data doesn't exist\n");
+		return -1;
+	}
 
 	retval = find_available_queue_slot(pqm, qid);
 	if (retval != 0)
@@ -176,7 +178,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 
 	if (list_empty(&pqm->queues)) {
 		pdd->qpd.pqm = pqm;
-		dev->dqm->register_process(dev->dqm, &pdd->qpd);
+		dev->dqm->ops.register_process(dev->dqm, &pdd->qpd);
 	}
 
 	pqn = kzalloc(sizeof(struct process_queue_node), GFP_KERNEL);
@@ -186,6 +188,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	}
 
 	switch (type) {
+	case KFD_QUEUE_TYPE_SDMA:
 	case KFD_QUEUE_TYPE_COMPUTE:
 		/* check if there is over subscription */
 		if ((sched_policy == KFD_SCHED_POLICY_HWS_NO_OVERSUBSCRIPTION) &&
@@ -201,7 +204,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 			goto err_create_queue;
 		pqn->q = q;
 		pqn->kq = NULL;
-		retval = dev->dqm->create_queue(dev->dqm, q, &pdd->qpd,
+		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd,
 						&q->properties.vmid);
 		pr_debug("DQM returned %d for create_queue\n", retval);
 		print_queue(q);
@@ -215,7 +218,8 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 		kq->queue->properties.queue_id = *qid;
 		pqn->kq = kq;
 		pqn->q = NULL;
-		retval = dev->dqm->create_kernel_queue(dev->dqm, kq, &pdd->qpd);
+		retval = dev->dqm->ops.create_kernel_queue(dev->dqm,
+							kq, &pdd->qpd);
 		break;
 	default:
 		BUG();
@@ -245,7 +249,7 @@ err_allocate_pqn:
 	/* check if queues list is empty unregister process from device */
 	clear_bit(*qid, pqm->queue_slot_bitmap);
 	if (list_empty(&pqm->queues))
-		dev->dqm->unregister_process(dev->dqm, &pdd->qpd);
+		dev->dqm->ops.unregister_process(dev->dqm, &pdd->qpd);
 	return retval;
 }
 
@@ -277,19 +281,22 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 		dev = pqn->q->device;
 	BUG_ON(!dev);
 
-	pdd = kfd_get_process_device_data(dev, pqm->process, 1);
-	BUG_ON(!pdd);
+	pdd = kfd_get_process_device_data(dev, pqm->process);
+	if (!pdd) {
+		pr_err("Process device data doesn't exist\n");
+		return -1;
+	}
 
 	if (pqn->kq) {
 		/* destroy kernel queue (DIQ) */
 		dqm = pqn->kq->dev->dqm;
-		dqm->destroy_kernel_queue(dqm, pqn->kq, &pdd->qpd);
+		dqm->ops.destroy_kernel_queue(dqm, pqn->kq, &pdd->qpd);
 		kernel_queue_uninit(pqn->kq);
 	}
 
 	if (pqn->q) {
 		dqm = pqn->q->device->dqm;
-		retval = dqm->destroy_queue(dqm, &pdd->qpd, pqn->q);
+		retval = dqm->ops.destroy_queue(dqm, &pdd->qpd, pqn->q);
 		if (retval != 0)
 			return retval;
 
@@ -301,7 +308,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 	clear_bit(qid, pqm->queue_slot_bitmap);
 
 	if (list_empty(&pqm->queues))
-		dqm->unregister_process(dqm, &pdd->qpd);
+		dqm->ops.unregister_process(dqm, &pdd->qpd);
 
 	return retval;
 }
@@ -326,7 +333,8 @@ int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid,
 	pqn->q->properties.queue_percent = p->queue_percent;
 	pqn->q->properties.priority = p->priority;
 
-	retval = pqn->q->device->dqm->update_queue(pqn->q->device->dqm, pqn->q);
+	retval = pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm,
+							pqn->q);
 	if (retval != 0)
 		return retval;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index cca1708fd811..498399323a8c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -27,6 +27,7 @@
 #include <linux/acpi.h>
 #include <linux/hash.h>
 #include <linux/cpufreq.h>
+#include <linux/log2.h>
 
 #include "kfd_priv.h"
 #include "kfd_crat.h"
@@ -630,10 +631,10 @@ static struct kobj_type cache_type = {
 static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 		char *buffer)
 {
-	ssize_t ret;
 	struct kfd_topology_device *dev;
 	char public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
 	uint32_t i;
+	uint32_t log_max_watch_addr;
 
 	/* Making sure that the buffer is an empty string */
 	buffer[0] = 0;
@@ -641,8 +642,10 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 	if (strcmp(attr->name, "gpu_id") == 0) {
 		dev = container_of(attr, struct kfd_topology_device,
 				attr_gpuid);
-		ret = sysfs_show_32bit_val(buffer, dev->gpu_id);
-	} else if (strcmp(attr->name, "name") == 0) {
+		return sysfs_show_32bit_val(buffer, dev->gpu_id);
+	}
+
+	if (strcmp(attr->name, "name") == 0) {
 		dev = container_of(attr, struct kfd_topology_device,
 				attr_name);
 		for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE; i++) {
@@ -652,80 +655,90 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 				break;
 		}
 		public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1] = 0x0;
-		ret = sysfs_show_str_val(buffer, public_name);
-	} else {
-		dev = container_of(attr, struct kfd_topology_device,
-				attr_props);
-		sysfs_show_32bit_prop(buffer, "cpu_cores_count",
-				dev->node_props.cpu_cores_count);
-		sysfs_show_32bit_prop(buffer, "simd_count",
-				dev->node_props.simd_count);
-
-		if (dev->mem_bank_count < dev->node_props.mem_banks_count) {
-			pr_warn("kfd: mem_banks_count truncated from %d to %d\n",
-					dev->node_props.mem_banks_count,
-					dev->mem_bank_count);
-			sysfs_show_32bit_prop(buffer, "mem_banks_count",
-					dev->mem_bank_count);
-		} else {
-			sysfs_show_32bit_prop(buffer, "mem_banks_count",
-					dev->node_props.mem_banks_count);
-		}
+		return sysfs_show_str_val(buffer, public_name);
+	}
 
-		sysfs_show_32bit_prop(buffer, "caches_count",
-				dev->node_props.caches_count);
-		sysfs_show_32bit_prop(buffer, "io_links_count",
-				dev->node_props.io_links_count);
-		sysfs_show_32bit_prop(buffer, "cpu_core_id_base",
-				dev->node_props.cpu_core_id_base);
-		sysfs_show_32bit_prop(buffer, "simd_id_base",
-				dev->node_props.simd_id_base);
-		sysfs_show_32bit_prop(buffer, "capability",
-				dev->node_props.capability);
-		sysfs_show_32bit_prop(buffer, "max_waves_per_simd",
-				dev->node_props.max_waves_per_simd);
-		sysfs_show_32bit_prop(buffer, "lds_size_in_kb",
-				dev->node_props.lds_size_in_kb);
-		sysfs_show_32bit_prop(buffer, "gds_size_in_kb",
-				dev->node_props.gds_size_in_kb);
-		sysfs_show_32bit_prop(buffer, "wave_front_size",
-				dev->node_props.wave_front_size);
-		sysfs_show_32bit_prop(buffer, "array_count",
-				dev->node_props.array_count);
-		sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine",
-				dev->node_props.simd_arrays_per_engine);
-		sysfs_show_32bit_prop(buffer, "cu_per_simd_array",
-				dev->node_props.cu_per_simd_array);
-		sysfs_show_32bit_prop(buffer, "simd_per_cu",
-				dev->node_props.simd_per_cu);
-		sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu",
-				dev->node_props.max_slots_scratch_cu);
-		sysfs_show_32bit_prop(buffer, "vendor_id",
-				dev->node_props.vendor_id);
-		sysfs_show_32bit_prop(buffer, "device_id",
-				dev->node_props.device_id);
-		sysfs_show_32bit_prop(buffer, "location_id",
-				dev->node_props.location_id);
-
-		if (dev->gpu) {
-			sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute",
-					kfd2kgd->get_max_engine_clock_in_mhz(
-						dev->gpu->kgd));
-			sysfs_show_64bit_prop(buffer, "local_mem_size",
-					kfd2kgd->get_vmem_size(dev->gpu->kgd));
-
-			sysfs_show_32bit_prop(buffer, "fw_version",
-					kfd2kgd->get_fw_version(
-							dev->gpu->kgd,
-							KGD_ENGINE_MEC1));
+	dev = container_of(attr, struct kfd_topology_device,
+			attr_props);
+	sysfs_show_32bit_prop(buffer, "cpu_cores_count",
+			dev->node_props.cpu_cores_count);
+	sysfs_show_32bit_prop(buffer, "simd_count",
+			dev->node_props.simd_count);
+
+	if (dev->mem_bank_count < dev->node_props.mem_banks_count) {
+		pr_warn("kfd: mem_banks_count truncated from %d to %d\n",
+				dev->node_props.mem_banks_count,
+				dev->mem_bank_count);
+		sysfs_show_32bit_prop(buffer, "mem_banks_count",
+				dev->mem_bank_count);
+	} else {
+		sysfs_show_32bit_prop(buffer, "mem_banks_count",
+				dev->node_props.mem_banks_count);
+	}
 
+	sysfs_show_32bit_prop(buffer, "caches_count",
+			dev->node_props.caches_count);
+	sysfs_show_32bit_prop(buffer, "io_links_count",
+			dev->node_props.io_links_count);
+	sysfs_show_32bit_prop(buffer, "cpu_core_id_base",
+			dev->node_props.cpu_core_id_base);
+	sysfs_show_32bit_prop(buffer, "simd_id_base",
+			dev->node_props.simd_id_base);
+	sysfs_show_32bit_prop(buffer, "capability",
+			dev->node_props.capability);
+	sysfs_show_32bit_prop(buffer, "max_waves_per_simd",
+			dev->node_props.max_waves_per_simd);
+	sysfs_show_32bit_prop(buffer, "lds_size_in_kb",
+			dev->node_props.lds_size_in_kb);
+	sysfs_show_32bit_prop(buffer, "gds_size_in_kb",
+			dev->node_props.gds_size_in_kb);
+	sysfs_show_32bit_prop(buffer, "wave_front_size",
+			dev->node_props.wave_front_size);
+	sysfs_show_32bit_prop(buffer, "array_count",
+			dev->node_props.array_count);
+	sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine",
+			dev->node_props.simd_arrays_per_engine);
+	sysfs_show_32bit_prop(buffer, "cu_per_simd_array",
+			dev->node_props.cu_per_simd_array);
+	sysfs_show_32bit_prop(buffer, "simd_per_cu",
+			dev->node_props.simd_per_cu);
+	sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu",
+			dev->node_props.max_slots_scratch_cu);
+	sysfs_show_32bit_prop(buffer, "vendor_id",
+			dev->node_props.vendor_id);
+	sysfs_show_32bit_prop(buffer, "device_id",
+			dev->node_props.device_id);
+	sysfs_show_32bit_prop(buffer, "location_id",
+			dev->node_props.location_id);
+
+	if (dev->gpu) {
+		log_max_watch_addr =
+			__ilog2_u32(dev->gpu->device_info->num_of_watch_points);
+
+		if (log_max_watch_addr) {
+			dev->node_props.capability |=
+					HSA_CAP_WATCH_POINTS_SUPPORTED;
+
+			dev->node_props.capability |=
+				((log_max_watch_addr <<
+					HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT) &
+				HSA_CAP_WATCH_POINTS_TOTALBITS_MASK);
 		}
 
-		ret = sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute",
-				cpufreq_quick_get_max(0)/1000);
+		sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute",
+				kfd2kgd->get_max_engine_clock_in_mhz(
+					dev->gpu->kgd));
+		sysfs_show_64bit_prop(buffer, "local_mem_size",
+				kfd2kgd->get_vmem_size(dev->gpu->kgd));
+
+		sysfs_show_32bit_prop(buffer, "fw_version",
+				kfd2kgd->get_fw_version(
+						dev->gpu->kgd,
+						KGD_ENGINE_MEC1));
 	}
 
-	return ret;
+	return sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute",
+					cpufreq_quick_get_max(0)/1000);
 }
 
 static const struct sysfs_ops node_ops = {