Merge tag 'drm-amdkfd-next-2015-07-20' of git://people.freedesktop.org/~gabbayo/linux into drm-next

- Add Carrizo support for amdkfd, using the new amdgpu driver as the relevant
  kgd. The support includes interfaces with amdgpu both for gfx7 (Kaveri) and
  gfx8 (Carrizo). However, gfx7 interface is used for debugging purposes only,
  so amdkfd defaults to using radeon when Kaveri is installed.

I would like to note that no new IOCTLs are being introduced, and there is no
change in the current IOCTLs, as they are suited both for gfx7 and gfx8.

* tag 'drm-amdkfd-next-2015-07-20' of git://people.freedesktop.org/~gabbayo/linux:
  drm/amdkfd: Set correct doorbell packet type for Carrizo
  drm/amdkfd: Use generic defines in new amd headers
  drm/amdkfd: Implement create_map_queues() for Carrizo
  drm/amdkfd: fix runlist length calculation
  drm/amdkfd: Add support for VI in DQM
  drm/amdkfd: add support for VI in MQD manager
  drm/amdkfd: add CP HWS packet headers for VI
  drm/amdkfd: add supported CZ devices PCI IDs to amdkfd
  drm/amdkfd: Add dependency of DRM_AMDGPU to Kconfig
  drm/amdgpu: Add amdgpu <--> amdkfd gfx8 interface
  drm/amdgpu: add amdgpu <--> amdkfd gfx7 interface
  drm/amdgpu: Add H/W agnostic amdgpu <--> amdkfd interface
  drm/radeon: Modify kgd_engine_type enum to match CZ

12 files changed, 1594 insertions, 1 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 616dfd4a1398..908360584e4d 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -71,6 +71,12 @@ amdgpu-y += \
 	amdgpu_vce.o \
 	vce_v3_0.o
 
+# add amdkfd interfaces
+amdgpu-y += \
+	 amdgpu_amdkfd.o \
+	 amdgpu_amdkfd_gfx_v7.o \
+	 amdgpu_amdkfd_gfx_v8.o
+
 amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o
 amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o
 amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 01657830b470..f3791e0d27d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -2011,6 +2011,9 @@ struct amdgpu_device {
 	/* tracking pinned memory */
 	u64 vram_pin_size;
 	u64 gart_pin_size;
+
+	/* amdkfd interface */
+	struct kfd_dev          *kfd;
 };
 
 bool amdgpu_device_is_px(struct drm_device *dev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
new file mode 100644
index 000000000000..bc763e0c8f4c
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -0,0 +1,267 @@
+/*
+ * 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 "amdgpu_amdkfd.h"
+#include "amdgpu_family.h"
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include <linux/module.h>
+
+const struct kfd2kgd_calls *kfd2kgd;
+const struct kgd2kfd_calls *kgd2kfd;
+bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+
+bool amdgpu_amdkfd_init(void)
+{
+#if defined(CONFIG_HSA_AMD_MODULE)
+	bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+
+	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+
+	if (kgd2kfd_init_p == NULL)
+		return false;
+#endif
+	return true;
+}
+
+bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev)
+{
+#if defined(CONFIG_HSA_AMD_MODULE)
+	bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+#endif
+
+	switch (rdev->asic_type) {
+	case CHIP_KAVERI:
+		kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions();
+		break;
+	case CHIP_CARRIZO:
+		kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
+		break;
+	default:
+		return false;
+	}
+
+#if defined(CONFIG_HSA_AMD_MODULE)
+	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+
+	if (kgd2kfd_init_p == NULL) {
+		kfd2kgd = NULL;
+		return false;
+	}
+
+	if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) {
+		symbol_put(kgd2kfd_init);
+		kfd2kgd = NULL;
+		kgd2kfd = NULL;
+
+		return false;
+	}
+
+	return true;
+#elif defined(CONFIG_HSA_AMD)
+	if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) {
+		kfd2kgd = NULL;
+		kgd2kfd = NULL;
+		return false;
+	}
+
+	return true;
+#else
+	kfd2kgd = NULL;
+	return false;
+#endif
+}
+
+void amdgpu_amdkfd_fini(void)
+{
+	if (kgd2kfd) {
+		kgd2kfd->exit();
+		symbol_put(kgd2kfd_init);
+	}
+}
+
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev)
+{
+	if (kgd2kfd)
+		rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev,
+					rdev->pdev, kfd2kgd);
+}
+
+void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd) {
+		struct kgd2kfd_shared_resources gpu_resources = {
+			.compute_vmid_bitmap = 0xFF00,
+
+			.first_compute_pipe = 1,
+			.compute_pipe_count = 4 - 1,
+		};
+
+		amdgpu_doorbell_get_kfd_info(rdev,
+				&gpu_resources.doorbell_physical_address,
+				&gpu_resources.doorbell_aperture_size,
+				&gpu_resources.doorbell_start_offset);
+
+		kgd2kfd->device_init(rdev->kfd, &gpu_resources);
+	}
+}
+
+void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd) {
+		kgd2kfd->device_exit(rdev->kfd);
+		rdev->kfd = NULL;
+	}
+}
+
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+		const void *ih_ring_entry)
+{
+	if (rdev->kfd)
+		kgd2kfd->interrupt(rdev->kfd, ih_ring_entry);
+}
+
+void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd)
+		kgd2kfd->suspend(rdev->kfd);
+}
+
+int amdgpu_amdkfd_resume(struct amdgpu_device *rdev)
+{
+	int r = 0;
+
+	if (rdev->kfd)
+		r = kgd2kfd->resume(rdev->kfd);
+
+	return r;
+}
+
+u32 pool_to_domain(enum kgd_memory_pool p)
+{
+	switch (p) {
+	case KGD_POOL_FRAMEBUFFER: return AMDGPU_GEM_DOMAIN_VRAM;
+	default: return AMDGPU_GEM_DOMAIN_GTT;
+	}
+}
+
+int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
+			void **mem_obj, uint64_t *gpu_addr,
+			void **cpu_ptr)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+	struct kgd_mem **mem = (struct kgd_mem **) mem_obj;
+	int r;
+
+	BUG_ON(kgd == NULL);
+	BUG_ON(gpu_addr == NULL);
+	BUG_ON(cpu_ptr == NULL);
+
+	*mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if ((*mem) == NULL)
+		return -ENOMEM;
+
+	r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT,
+			AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, &(*mem)->bo);
+	if (r) {
+		dev_err(rdev->dev,
+			"failed to allocate BO for amdkfd (%d)\n", r);
+		return r;
+	}
+
+	/* map the buffer */
+	r = amdgpu_bo_reserve((*mem)->bo, true);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r);
+		goto allocate_mem_reserve_bo_failed;
+	}
+
+	r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT,
+				&(*mem)->gpu_addr);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r);
+		goto allocate_mem_pin_bo_failed;
+	}
+	*gpu_addr = (*mem)->gpu_addr;
+
+	r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr);
+	if (r) {
+		dev_err(rdev->dev,
+			"(%d) failed to map bo to kernel for amdkfd\n", r);
+		goto allocate_mem_kmap_bo_failed;
+	}
+	*cpu_ptr = (*mem)->cpu_ptr;
+
+	amdgpu_bo_unreserve((*mem)->bo);
+
+	return 0;
+
+allocate_mem_kmap_bo_failed:
+	amdgpu_bo_unpin((*mem)->bo);
+allocate_mem_pin_bo_failed:
+	amdgpu_bo_unreserve((*mem)->bo);
+allocate_mem_reserve_bo_failed:
+	amdgpu_bo_unref(&(*mem)->bo);
+
+	return r;
+}
+
+void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj)
+{
+	struct kgd_mem *mem = (struct kgd_mem *) mem_obj;
+
+	BUG_ON(mem == NULL);
+
+	amdgpu_bo_reserve(mem->bo, true);
+	amdgpu_bo_kunmap(mem->bo);
+	amdgpu_bo_unpin(mem->bo);
+	amdgpu_bo_unreserve(mem->bo);
+	amdgpu_bo_unref(&(mem->bo));
+	kfree(mem);
+}
+
+uint64_t get_vmem_size(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev =
+		(struct amdgpu_device *)kgd;
+
+	BUG_ON(kgd == NULL);
+
+	return rdev->mc.real_vram_size;
+}
+
+uint64_t get_gpu_clock_counter(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+
+	if (rdev->asic_funcs->get_gpu_clock_counter)
+		return rdev->asic_funcs->get_gpu_clock_counter(rdev);
+	return 0;
+}
+
+uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+
+	/* The sclk is in quantas of 10kHz */
+	return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
new file mode 100644
index 000000000000..a8be765542e6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -0,0 +1,65 @@
+/*
+ * 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.
+ */
+
+/* amdgpu_amdkfd.h defines the private interface between amdgpu and amdkfd. */
+
+#ifndef AMDGPU_AMDKFD_H_INCLUDED
+#define AMDGPU_AMDKFD_H_INCLUDED
+
+#include <linux/types.h>
+#include <kgd_kfd_interface.h>
+
+struct amdgpu_device;
+
+struct kgd_mem {
+	struct amdgpu_bo *bo;
+	uint64_t gpu_addr;
+	void *cpu_ptr;
+};
+
+bool amdgpu_amdkfd_init(void);
+void amdgpu_amdkfd_fini(void);
+
+bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev);
+
+void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev);
+int amdgpu_amdkfd_resume(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+			const void *ih_ring_entry);
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev);
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
+
+/* Shared API */
+int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
+			void **mem_obj, uint64_t *gpu_addr,
+			void **cpu_ptr);
+void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj);
+uint64_t get_vmem_size(struct kgd_dev *kgd);
+uint64_t get_gpu_clock_counter(struct kgd_dev *kgd);
+
+uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd);
+
+#endif /* AMDGPU_AMDKFD_H_INCLUDED */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
new file mode 100644
index 000000000000..2daad335b809
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -0,0 +1,670 @@
+/*
+ * 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/fdtable.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "cikd.h"
+#include "cik_sdma.h"
+#include "amdgpu_ucode.h"
+#include "gca/gfx_7_2_d.h"
+#include "gca/gfx_7_2_enum.h"
+#include "gca/gfx_7_2_sh_mask.h"
+#include "oss/oss_2_0_d.h"
+#include "oss/oss_2_0_sh_mask.h"
+#include "gmc/gmc_7_1_d.h"
+#include "gmc/gmc_7_1_sh_mask.h"
+#include "cik_structs.h"
+
+#define CIK_PIPE_PER_MEC	(4)
+
+enum {
+	MAX_TRAPID = 8,		/* 3 bits in the bitfield. */
+	MAX_WATCH_ADDRESSES = 4
+};
+
+enum {
+	ADDRESS_WATCH_REG_ADDR_HI = 0,
+	ADDRESS_WATCH_REG_ADDR_LO,
+	ADDRESS_WATCH_REG_CNTL,
+	ADDRESS_WATCH_REG_MAX
+};
+
+/*  not defined in the CI/KV reg file  */
+enum {
+	ADDRESS_WATCH_REG_CNTL_ATC_BIT = 0x10000000UL,
+	ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK = 0x00FFFFFF,
+	ADDRESS_WATCH_REG_ADDLOW_MASK_EXTENSION = 0x03000000,
+	/* extend the mask to 26 bits to match the low address field */
+	ADDRESS_WATCH_REG_ADDLOW_SHIFT = 6,
+	ADDRESS_WATCH_REG_ADDHIGH_MASK = 0xFFFF
+};
+
+static const uint32_t watchRegs[MAX_WATCH_ADDRESSES * ADDRESS_WATCH_REG_MAX] = {
+	mmTCP_WATCH0_ADDR_H, mmTCP_WATCH0_ADDR_L, mmTCP_WATCH0_CNTL,
+	mmTCP_WATCH1_ADDR_H, mmTCP_WATCH1_ADDR_L, mmTCP_WATCH1_CNTL,
+	mmTCP_WATCH2_ADDR_H, mmTCP_WATCH2_ADDR_L, mmTCP_WATCH2_CNTL,
+	mmTCP_WATCH3_ADDR_H, mmTCP_WATCH3_ADDR_L, mmTCP_WATCH3_CNTL
+};
+
+union TCP_WATCH_CNTL_BITS {
+	struct {
+		uint32_t mask:24;
+		uint32_t vmid:4;
+		uint32_t atc:1;
+		uint32_t mode:2;
+		uint32_t valid:1;
+	} bitfields, bits;
+	uint32_t u32All;
+	signed int i32All;
+	float f32All;
+};
+
+/*
+ * Register access functions
+ */
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t sh_mem_config,	uint32_t sh_mem_ape1_base,
+		uint32_t sh_mem_ape1_limit, uint32_t sh_mem_bases);
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid);
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr);
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id);
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id);
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout);
+static int kgd_address_watch_disable(struct kgd_dev *kgd);
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo);
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd);
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset);
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, uint8_t vmid);
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+							uint8_t vmid);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_gtt_mem_allocation = alloc_gtt_mem,
+	.free_gtt_mem = free_gtt_mem,
+	.get_vmem_size = get_vmem_size,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_pipeline = kgd_init_pipeline,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.address_watch_disable = kgd_address_watch_disable,
+	.address_watch_execute = kgd_address_watch_execute,
+	.wave_control_execute = kgd_wave_control_execute,
+	.address_watch_get_offset = kgd_address_watch_get_offset,
+	.get_atc_vmid_pasid_mapping_pasid = get_atc_vmid_pasid_mapping_pasid,
+	.get_atc_vmid_pasid_mapping_valid = get_atc_vmid_pasid_mapping_valid,
+	.write_vmid_invalidate_request = write_vmid_invalidate_request,
+	.get_fw_version = get_fw_version
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions()
+{
+	return (struct kfd2kgd_calls *)&kfd2kgd;
+}
+
+static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
+{
+	return (struct amdgpu_device *)kgd;
+}
+
+static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+	mutex_lock(&adev->srbm_mutex);
+	WREG32(mmSRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	WREG32(mmSRBM_GFX_CNTL, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+static void release_queue(struct kgd_dev *kgd)
+{
+	unlock_srbm(kgd);
+}
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	lock_srbm(kgd, 0, 0, 0, vmid);
+
+	WREG32(mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
+	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+	WREG32(mmSH_MEM_BASES, sh_mem_bases);
+
+	unlock_srbm(kgd);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished and the
+	 * SW cleared it. So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping);
+
+	while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid)))
+		cpu_relax();
+	WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping);
+
+	return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+	WREG32(mmCP_HPD_EOP_BASE_ADDR, lower_32_bits(hpd_gpu_addr >> 8));
+	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(hpd_gpu_addr >> 8));
+	WREG32(mmCP_HPD_EOP_VMID, 0);
+	WREG32(mmCP_HPD_EOP_CONTROL, hpd_size);
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / CIK_PIPE_PER_MEC) + 1;
+	pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+
+	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+			CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+{
+	uint32_t retval;
+
+	retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET +
+			m->sdma_queue_id * KFD_CIK_SDMA_QUEUE_OFFSET;
+
+	pr_debug("kfd: sdma base address: 0x%x\n", retval);
+
+	return retval;
+}
+
+static inline struct cik_mqd *get_mqd(void *mqd)
+{
+	return (struct cik_mqd *)mqd;
+}
+
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	return (struct cik_sdma_rlc_registers *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t wptr_shadow, is_wptr_shadow_valid;
+	struct cik_mqd *m;
+
+	m = get_mqd(mqd);
+
+	is_wptr_shadow_valid = !get_user(wptr_shadow, wptr);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+
+	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+
+	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+	WREG32(mmCP_HQD_IB_BASE_ADDR, m->cp_hqd_ib_base_addr_lo);
+	WREG32(mmCP_HQD_IB_BASE_ADDR_HI, m->cp_hqd_ib_base_addr_hi);
+
+	WREG32(mmCP_HQD_IB_RPTR, m->cp_hqd_ib_rptr);
+
+	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
+	WREG32(mmCP_HQD_SEMA_CMD, m->cp_hqd_sema_cmd);
+	WREG32(mmCP_HQD_MSG_TYPE, m->cp_hqd_msg_type);
+
+	WREG32(mmCP_HQD_ATOMIC0_PREOP_LO, m->cp_hqd_atomic0_preop_lo);
+	WREG32(mmCP_HQD_ATOMIC0_PREOP_HI, m->cp_hqd_atomic0_preop_hi);
+	WREG32(mmCP_HQD_ATOMIC1_PREOP_LO, m->cp_hqd_atomic1_preop_lo);
+	WREG32(mmCP_HQD_ATOMIC1_PREOP_HI, m->cp_hqd_atomic1_preop_hi);
+
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+			m->cp_hqd_pq_rptr_report_addr_hi);
+
+	WREG32(mmCP_HQD_PQ_RPTR, m->cp_hqd_pq_rptr);
+
+	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, m->cp_hqd_pq_wptr_poll_addr_lo);
+	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, m->cp_hqd_pq_wptr_poll_addr_hi);
+
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+
+	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+
+	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+
+	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+
+	WREG32(mmCP_HQD_IQ_RPTR, m->cp_hqd_iq_rptr);
+
+	if (is_wptr_shadow_valid)
+		WREG32(mmCP_HQD_PQ_WPTR, wptr_shadow);
+
+	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+	release_queue(kgd);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
+			m->sdma_rlc_virtual_addr);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE,
+			m->sdma_rlc_rb_base);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdma_rlc_rb_base_hi);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdma_rlc_rb_rptr_addr_lo);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdma_rlc_rb_rptr_addr_hi);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
+			m->sdma_rlc_doorbell);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+			m->sdma_rlc_rb_cntl);
+
+	return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	act = RREG32(mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(mmCP_HQD_PQ_BASE) &&
+				high == RREG32(mmCP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	release_queue(kgd);
+	return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t temp;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+
+	while (true) {
+		temp = RREG32(mmCP_HQD_ACTIVE);
+		if (temp & CP_HQD_ACTIVE__ACTIVE__SHIFT)
+			break;
+		if (timeout == 0) {
+			pr_err("kfd: cp queue preemption time out (%dms)\n",
+				temp);
+			release_queue(kgd);
+			return -ETIME;
+		}
+		msleep(20);
+		timeout -= 20;
+	}
+
+	release_queue(kgd);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t temp;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+			break;
+		if (timeout == 0)
+			return -ETIME;
+		msleep(20);
+		timeout -= 20;
+	}
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+
+	return 0;
+}
+
+static int kgd_address_watch_disable(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	union TCP_WATCH_CNTL_BITS cntl;
+	unsigned int i;
+
+	cntl.u32All = 0;
+
+	cntl.bitfields.valid = 0;
+	cntl.bitfields.mask = ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK;
+	cntl.bitfields.atc = 1;
+
+	/* Turning off this address until we set all the registers */
+	for (i = 0; i < MAX_WATCH_ADDRESSES; i++)
+		WREG32(watchRegs[i * ADDRESS_WATCH_REG_MAX +
+			ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	return 0;
+}
+
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	union TCP_WATCH_CNTL_BITS cntl;
+
+	cntl.u32All = cntl_val;
+
+	/* Turning off this watch point until we set all the registers */
+	cntl.bitfields.valid = 0;
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_ADDR_HI], addr_hi);
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_ADDR_LO], addr_lo);
+
+	/* Enable the watch point */
+	cntl.bitfields.valid = 1;
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	return 0;
+}
+
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t data;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32(mmSQ_CMD, sq_cmd);
+
+	/*  Restore the GRBM_GFX_INDEX register  */
+
+	data = GRBM_GFX_INDEX__INSTANCE_BROADCAST_WRITES_MASK |
+		GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
+		GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
+
+	WREG32(mmGRBM_GFX_INDEX, data);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset)
+{
+	return watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + reg_offset];
+}
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+							uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+								uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+}
+
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	const union amdgpu_firmware_header *hdr;
+
+	BUG_ON(kgd == NULL);
+
+	switch (type) {
+	case KGD_ENGINE_PFP:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.pfp_fw->data;
+		break;
+
+	case KGD_ENGINE_ME:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.me_fw->data;
+		break;
+
+	case KGD_ENGINE_CE:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.ce_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC1:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.mec_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.mec2_fw->data;
+		break;
+
+	case KGD_ENGINE_RLC:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.rlc_fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA1:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[0].fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[1].fw->data;
+		break;
+
+	default:
+		return 0;
+	}
+
+	if (hdr == NULL)
+		return 0;
+
+	/* Only 12 bit in use*/
+	return hdr->common.ucode_version;
+}
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
new file mode 100644
index 000000000000..dfd1d503bccf
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -0,0 +1,543 @@
+/*
+ * 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/module.h>
+#include <linux/fdtable.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_ucode.h"
+#include "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_d.h"
+#include "gca/gfx_8_0_enum.h"
+#include "oss/oss_3_0_sh_mask.h"
+#include "oss/oss_3_0_d.h"
+#include "gmc/gmc_8_1_sh_mask.h"
+#include "gmc/gmc_8_1_d.h"
+#include "vi_structs.h"
+#include "vid.h"
+
+#define VI_PIPE_PER_MEC	(4)
+
+struct cik_sdma_rlc_registers;
+
+/*
+ * Register access functions
+ */
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t sh_mem_config,
+		uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
+		uint32_t sh_mem_bases);
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+		unsigned int vmid);
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+		uint32_t hpd_size, uint64_t hpd_gpu_addr);
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+		uint32_t queue_id, uint32_t __user *wptr);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+		uint32_t pipe_id, uint32_t queue_id);
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id);
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+static int kgd_address_watch_disable(struct kgd_dev *kgd);
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo);
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd);
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset);
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+		uint8_t vmid);
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+		uint8_t vmid);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_gtt_mem_allocation = alloc_gtt_mem,
+	.free_gtt_mem = free_gtt_mem,
+	.get_vmem_size = get_vmem_size,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_pipeline = kgd_init_pipeline,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.address_watch_disable = kgd_address_watch_disable,
+	.address_watch_execute = kgd_address_watch_execute,
+	.wave_control_execute = kgd_wave_control_execute,
+	.address_watch_get_offset = kgd_address_watch_get_offset,
+	.get_atc_vmid_pasid_mapping_pasid =
+			get_atc_vmid_pasid_mapping_pasid,
+	.get_atc_vmid_pasid_mapping_valid =
+			get_atc_vmid_pasid_mapping_valid,
+	.write_vmid_invalidate_request = write_vmid_invalidate_request,
+	.get_fw_version = get_fw_version
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions()
+{
+	return (struct kfd2kgd_calls *)&kfd2kgd;
+}
+
+static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
+{
+	return (struct amdgpu_device *)kgd;
+}
+
+static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+	mutex_lock(&adev->srbm_mutex);
+	WREG32(mmSRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	WREG32(mmSRBM_GFX_CNTL, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+static void release_queue(struct kgd_dev *kgd)
+{
+	unlock_srbm(kgd);
+}
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	lock_srbm(kgd, 0, 0, 0, vmid);
+
+	WREG32(mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
+	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+	WREG32(mmSH_MEM_BASES, sh_mem_bases);
+
+	unlock_srbm(kgd);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished
+	 * and the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping);
+
+	while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid)))
+		cpu_relax();
+	WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping);
+
+	return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+	return 0;
+}
+
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
+	pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+
+	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK);
+
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+{
+	return 0;
+}
+
+static inline struct vi_mqd *get_mqd(void *mqd)
+{
+	return (struct vi_mqd *)mqd;
+}
+
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	return (struct cik_sdma_rlc_registers *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	struct vi_mqd *m;
+	uint32_t shadow_wptr, valid_wptr;
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	m = get_mqd(mqd);
+
+	valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr));
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+
+	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
+	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+			m->cp_hqd_pq_rptr_report_addr_hi);
+
+	if (valid_wptr > 0)
+		WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr);
+
+	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+
+	WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo);
+	WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi);
+	WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control);
+	WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
+	WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
+	WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events);
+
+	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo);
+	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi);
+	WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control);
+	WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset);
+	WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size);
+	WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset);
+	WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size);
+
+	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request);
+	WREG32(mmCP_HQD_ERROR, m->cp_hqd_error);
+	WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
+	WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones);
+
+	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+
+	release_queue(kgd);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+{
+	return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	act = RREG32(mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(mmCP_HQD_PQ_BASE) &&
+				high == RREG32(mmCP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	release_queue(kgd);
+	return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t temp;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+
+	while (true) {
+		temp = RREG32(mmCP_HQD_ACTIVE);
+		if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+			break;
+		if (timeout == 0) {
+			pr_err("kfd: cp queue preemption time out (%dms)\n",
+				temp);
+			release_queue(kgd);
+			return -ETIME;
+		}
+		msleep(20);
+		timeout -= 20;
+	}
+
+	release_queue(kgd);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t temp;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+			break;
+		if (timeout == 0)
+			return -ETIME;
+		msleep(20);
+		timeout -= 20;
+	}
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+
+	return 0;
+}
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+							uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+								uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+}
+
+static int kgd_address_watch_disable(struct kgd_dev *kgd)
+{
+	return 0;
+}
+
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo)
+{
+	return 0;
+}
+
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32(mmSQ_CMD, sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SH_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32(mmGRBM_GFX_INDEX, data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset)
+{
+	return 0;
+}
+
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	const union amdgpu_firmware_header *hdr;
+
+	BUG_ON(kgd == NULL);
+
+	switch (type) {
+	case KGD_ENGINE_PFP:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.pfp_fw->data;
+		break;
+
+	case KGD_ENGINE_ME:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.me_fw->data;
+		break;
+
+	case KGD_ENGINE_CE:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.ce_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC1:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.mec_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.mec2_fw->data;
+		break;
+
+	case KGD_ENGINE_RLC:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->gfx.rlc_fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA1:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[0].fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[1].fw->data;
+		break;
+
+	default:
+		return 0;
+	}
+
+	if (hdr == NULL)
+		return 0;
+
+	/* Only 12 bit in use*/
+	return hdr->common.ucode_version;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 56da962231fc..115906f5fda0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -44,6 +44,8 @@
 #include "amdgpu.h"
 #include "amdgpu_irq.h"
 
+#include "amdgpu_amdkfd.h"
+
 /*
  * KMS wrapper.
  * - 3.0.0 - initial driver
@@ -527,12 +529,15 @@ static int __init amdgpu_init(void)
 	driver->num_ioctls = amdgpu_max_kms_ioctl;
 	amdgpu_register_atpx_handler();
 
+	amdgpu_amdkfd_init();
+
 	/* let modprobe override vga console setting */
 	return drm_pci_init(driver, pdriver);
 }
 
 static void __exit amdgpu_exit(void)
 {
+	amdgpu_amdkfd_fini();
 	drm_pci_exit(driver, pdriver);
 	amdgpu_unregister_atpx_handler();
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
index db5422e65ec5..fb44dd2231b1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
@@ -24,6 +24,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_ih.h"
+#include "amdgpu_amdkfd.h"
 
 /**
  * amdgpu_ih_ring_alloc - allocate memory for the IH ring
@@ -199,6 +200,12 @@ restart_ih:
 	rmb();
 
 	while (adev->irq.ih.rptr != wptr) {
+		u32 ring_index = adev->irq.ih.rptr >> 2;
+
+		/* Before dispatching irq to IP blocks, send it to amdkfd */
+		amdgpu_amdkfd_interrupt(adev,
+				(const void *) &adev->irq.ih.ring[ring_index]);
+
 		amdgpu_ih_decode_iv(adev, &entry);
 		adev->irq.ih.rptr &= adev->irq.ih.ptr_mask;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 5533434c7a8f..8c40a9671b9f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -34,6 +34,7 @@
 #include <linux/vga_switcheroo.h>
 #include <linux/slab.h>
 #include <linux/pm_runtime.h>
+#include "amdgpu_amdkfd.h"
 
 #if defined(CONFIG_VGA_SWITCHEROO)
 bool amdgpu_has_atpx(void);
@@ -61,6 +62,8 @@ int amdgpu_driver_unload_kms(struct drm_device *dev)
 
 	pm_runtime_get_sync(dev->dev);
 
+	amdgpu_amdkfd_device_fini(adev);
+
 	amdgpu_acpi_fini(adev);
 
 	amdgpu_device_fini(adev);
@@ -118,6 +121,10 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
 				"Error during ACPI methods call\n");
 	}
 
+	amdgpu_amdkfd_load_interface(adev);
+	amdgpu_amdkfd_device_probe(adev);
+	amdgpu_amdkfd_device_init(adev);
+
 	if (amdgpu_device_is_px(dev)) {
 		pm_runtime_use_autosuspend(dev->dev);
 		pm_runtime_set_autosuspend_delay(dev->dev, 5000);
diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c
index 341c56681841..b3b66a0d5ff7 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik.c
@@ -64,6 +64,8 @@
 #include "oss/oss_2_0_d.h"
 #include "oss/oss_2_0_sh_mask.h"
 
+#include "amdgpu_amdkfd.h"
+
 /*
  * Indirect registers accessor
  */
@@ -2448,14 +2450,21 @@ static int cik_common_suspend(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_amdkfd_suspend(adev);
+
 	return cik_common_hw_fini(adev);
 }
 
 static int cik_common_resume(void *handle)
 {
+	int r;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	return cik_common_hw_init(adev);
+	r = cik_common_hw_init(adev);
+	if (r)
+		return r;
+
+	return amdgpu_amdkfd_resume(adev);
 }
 
 static bool cik_common_is_idle(void *handle)
diff --git a/drivers/gpu/drm/amd/amdgpu/cikd.h b/drivers/gpu/drm/amd/amdgpu/cikd.h
index d19085a97064..a3e3dfaa01a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/cikd.h
+++ b/drivers/gpu/drm/amd/amdgpu/cikd.h
@@ -552,6 +552,12 @@
 #define VCE_CMD_IB_AUTO		0x00000005
 #define VCE_CMD_SEMAPHORE	0x00000006
 
+/* if PTR32, these are the bases for scratch and lds */
+#define	PRIVATE_BASE(x)	((x) << 0) /* scratch */
+#define	SHARED_BASE(x)	((x) << 16) /* LDS */
+
+#define KFD_CIK_SDMA_QUEUE_OFFSET	0x200
+
 /* valid for both DEFAULT_MTYPE and APE1_MTYPE */
 enum {
 	MTYPE_CACHED = 0,
diff --git a/drivers/gpu/drm/amd/amdgpu/vid.h b/drivers/gpu/drm/amd/amdgpu/vid.h
index 31bb89452e12..d98aa9d82fa1 100644
--- a/drivers/gpu/drm/amd/amdgpu/vid.h
+++ b/drivers/gpu/drm/amd/amdgpu/vid.h
@@ -66,6 +66,11 @@
 
 #define AMDGPU_NUM_OF_VMIDS			8
 
+#define		PIPEID(x)					((x) << 0)
+#define		MEID(x)						((x) << 2)
+#define		VMID(x)						((x) << 4)
+#define		QUEUEID(x)					((x) << 8)
+
 #define RB_BITMAP_WIDTH_PER_SH     2
 
 #define MC_SEQ_MISC0__MT__MASK	0xf0000000