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

27 files changed, 2888 insertions, 40 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
diff --git a/drivers/gpu/drm/amd/amdkfd/Kconfig b/drivers/gpu/drm/amd/amdkfd/Kconfig
index 8dfac37ff327..e13c67c8d2c0 100644
--- a/drivers/gpu/drm/amd/amdkfd/Kconfig
+++ b/drivers/gpu/drm/amd/amdkfd/Kconfig
@@ -4,6 +4,6 @@
 
 config HSA_AMD
 	tristate "HSA kernel driver for AMD GPU devices"
-	depends on DRM_RADEON && AMD_IOMMU_V2 && X86_64
+	depends on (DRM_RADEON || DRM_AMDGPU) && AMD_IOMMU_V2 && X86_64
 	help
 	  Enable this if you want to use HSA features on AMD GPU devices.
diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile
index 28551153ec6d..7fc9b0f444cb 100644
--- a/drivers/gpu/drm/amd/amdkfd/Makefile
+++ b/drivers/gpu/drm/amd/amdkfd/Makefile
@@ -2,7 +2,8 @@
 # Makefile for Heterogenous System Architecture support for AMD GPU devices
 #
 
-ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/
+ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/  \
+		-Idrivers/gpu/drm/amd/include/asic_reg
 
 amdkfd-y	:= kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
 		kfd_pasid.o kfd_doorbell.o kfd_flat_memory.o \
diff --git a/drivers/gpu/drm/amd/amdkfd/cik_regs.h b/drivers/gpu/drm/amd/amdkfd/cik_regs.h
index 183be5b8414f..48769d12dd7b 100644
--- a/drivers/gpu/drm/amd/amdkfd/cik_regs.h
+++ b/drivers/gpu/drm/amd/amdkfd/cik_regs.h
@@ -65,17 +65,6 @@
 
 #define	AQL_ENABLE					1
 
-#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	ATC_VMID_PASID_MAPPING_VALID			(1U << 31)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 75312c82969f..3f95f7cb4019 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -80,7 +80,12 @@ 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 */
+	{ 0x9870, &carrizo_device_info },	/* Carrizo */
+	{ 0x9874, &carrizo_device_info },	/* Carrizo */
+	{ 0x9875, &carrizo_device_info },	/* Carrizo */
+	{ 0x9876, &carrizo_device_info },	/* Carrizo */
+	{ 0x9877, &carrizo_device_info }	/* Carrizo */
 };
 
 static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
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
index 9ce8a20a7aff..23ce774ff09d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -23,6 +23,7 @@
 
 #include "kfd_device_queue_manager.h"
 #include "cik_regs.h"
+#include "oss/oss_2_4_sh_mask.h"
 
 static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
@@ -135,13 +136,16 @@ static int register_process_cik(struct device_queue_manager *dqm,
 static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 				struct qcm_process_device *qpd)
 {
-	uint32_t value = SDMA_ATC;
+	uint32_t value = (1 << SDMA0_RLC0_VIRTUAL_ADDR__ATC__SHIFT);
 
 	if (q->process->is_32bit_user_mode)
-		value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd));
+		value |= (1 << SDMA0_RLC0_VIRTUAL_ADDR__PTR32__SHIFT) |
+				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)));
+		value |= ((get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd))) <<
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE__SHIFT) &&
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE_MASK;
+
 	q->properties.sdma_vm_addr = value;
 }
 
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
index 4c15212a3899..44c38e8e54d3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
@@ -22,6 +22,10 @@
  */
 
 #include "kfd_device_queue_manager.h"
+#include "gca/gfx_8_0_enum.h"
+#include "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_enum.h"
+#include "oss/oss_3_0_sh_mask.h"
 
 static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
@@ -37,14 +41,40 @@ static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 
 void device_queue_manager_init_vi(struct device_queue_manager_asic_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;
 	ops->init_sdma_vm = init_sdma_vm;
 }
 
+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 top_address_nybble << 12 |
+			(top_address_nybble << 12) <<
+			SH_MEM_BASES__SHARED_BASE__SHIFT;
+}
+
 static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
 				   enum cache_policy default_policy,
@@ -52,18 +82,83 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   void __user *alternate_aperture_base,
 				   uint64_t alternate_aperture_size)
 {
-	return false;
+	uint32_t default_mtype;
+	uint32_t ape1_mtype;
+
+	default_mtype = (default_policy == cache_policy_coherent) ?
+			MTYPE_CC :
+			MTYPE_NC;
+
+	ape1_mtype = (alternate_policy == cache_policy_coherent) ?
+			MTYPE_CC :
+			MTYPE_NC;
+
+	qpd->sh_mem_config = (qpd->sh_mem_config &
+			SH_MEM_CONFIG__ADDRESS_MODE_MASK) |
+		SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
+		default_mtype << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
+		ape1_mtype << SH_MEM_CONFIG__APE1_MTYPE__SHIFT |
+		SH_MEM_CONFIG__PRIVATE_ATC_MASK;
+
+	return true;
 }
 
 static int register_process_vi(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd)
 {
-	return -1;
+	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 =
+			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
+			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
+			MTYPE_CC << SH_MEM_CONFIG__APE1_MTYPE__SHIFT |
+			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
+
+		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 = temp << SH_MEM_BASES__SHARED_BASE__SHIFT;
+		qpd->sh_mem_config |= SH_MEM_ADDRESS_MODE_HSA32 <<
+					SH_MEM_CONFIG__ADDRESS_MODE__SHIFT;
+	} else {
+		temp = get_sh_mem_bases_nybble_64(pdd);
+		qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
+		qpd->sh_mem_config |= SH_MEM_ADDRESS_MODE_HSA64 <<
+			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT;
+	}
+
+	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 void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 				struct qcm_process_device *qpd)
 {
+	uint32_t value = (1 << SDMA0_RLC0_VIRTUAL_ADDR__ATC__SHIFT);
+
+	if (q->process->is_32bit_user_mode)
+		value |= (1 << SDMA0_RLC0_VIRTUAL_ADDR__PTR32__SHIFT) |
+				get_sh_mem_bases_32(qpd_to_pdd(qpd));
+	else
+		value |= ((get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd))) <<
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE__SHIFT) &&
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE_MASK;
+
+	q->properties.sdma_vm_addr = value;
 }
 
 static int initialize_cpsch_vi(struct device_queue_manager *dqm)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
index 434979428fc0..d83de985e88c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
@@ -27,6 +27,7 @@
 #include "kfd_mqd_manager.h"
 #include "cik_regs.h"
 #include "cik_structs.h"
+#include "oss/oss_2_4_sh_mask.h"
 
 static inline struct cik_mqd *get_mqd(void *mqd)
 {
@@ -214,17 +215,20 @@ static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
 	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_cntl = ffs(q->queue_size / sizeof(unsigned int)) <<
+			SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
+			q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
+			1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
+			6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
 
 	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_doorbell = q->doorbell_off <<
+			SDMA0_RLC0_DOORBELL__OFFSET__SHIFT |
+			1 << SDMA0_RLC0_DOORBELL__ENABLE__SHIFT;
+
 	m->sdma_rlc_virtual_addr = q->sdma_vm_addr;
 
 	m->sdma_engine_id = q->sdma_engine_id;
@@ -234,7 +238,9 @@ static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
 	if (q->queue_size > 0 &&
 			q->queue_address != 0 &&
 			q->queue_percent > 0) {
-		m->sdma_rlc_rb_cntl |= SDMA_RB_ENABLE;
+		m->sdma_rlc_rb_cntl |=
+				1 << SDMA0_RLC0_RB_CNTL__RB_ENABLE__SHIFT;
+
 		q->is_active = true;
 	}
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
index b3a7e3ba1e38..fa32c32fa1c2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
@@ -22,12 +22,255 @@
  */
 
 #include <linux/printk.h>
+#include <linux/slab.h>
 #include "kfd_priv.h"
 #include "kfd_mqd_manager.h"
+#include "vi_structs.h"
+#include "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_enum.h"
+
+#define CP_MQD_CONTROL__PRIV_STATE__SHIFT 0x8
+
+static inline struct vi_mqd *get_mqd(void *mqd)
+{
+	return (struct vi_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)
+{
+	int retval;
+	uint64_t addr;
+	struct vi_mqd *m;
+
+	retval = kfd_gtt_sa_allocate(mm->dev, sizeof(struct vi_mqd),
+			mqd_mem_obj);
+	if (retval != 0)
+		return -ENOMEM;
+
+	m = (struct vi_mqd *) (*mqd_mem_obj)->cpu_ptr;
+	addr = (*mqd_mem_obj)->gpu_addr;
+
+	memset(m, 0, sizeof(struct vi_mqd));
+
+	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 = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
+			0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
+
+	m->cp_mqd_control = 1 << CP_MQD_CONTROL__PRIV_STATE__SHIFT |
+			MTYPE_UC << CP_MQD_CONTROL__MTYPE__SHIFT;
+
+	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
+	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
+
+	m->cp_hqd_quantum = 1 << CP_HQD_QUANTUM__QUANTUM_EN__SHIFT |
+			1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
+			10 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
+
+	m->cp_hqd_pipe_priority = 1;
+	m->cp_hqd_queue_priority = 15;
+
+	m->cp_hqd_eop_rptr = 1 << CP_HQD_EOP_RPTR__INIT_FETCHER__SHIFT;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL)
+		m->cp_hqd_iq_rptr = 1;
+
+	*mqd = m;
+	if (gart_addr != NULL)
+		*gart_addr = addr;
+	retval = mm->update_mqd(mm, m, q);
+
+	return retval;
+}
+
+static int load_mqd(struct mqd_manager *mm, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr)
+{
+	return mm->dev->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, unsigned int mtype,
+			unsigned int atc_bit)
+{
+	struct vi_mqd *m;
+
+	BUG_ON(!mm || !q || !mqd);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	m = get_mqd(mqd);
+
+	m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT |
+			atc_bit << CP_HQD_PQ_CONTROL__PQ_ATC__SHIFT |
+			mtype << CP_HQD_PQ_CONTROL__MTYPE__SHIFT;
+	m->cp_hqd_pq_control |=
+			ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1;
+	pr_debug("kfd: cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
+
+	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 =
+		1 << CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN__SHIFT |
+		q->doorbell_off <<
+			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
+	pr_debug("kfd: cp_hqd_pq_doorbell_control 0x%x\n",
+			m->cp_hqd_pq_doorbell_control);
+
+	m->cp_hqd_eop_control = atc_bit << CP_HQD_EOP_CONTROL__EOP_ATC__SHIFT |
+			mtype << CP_HQD_EOP_CONTROL__MTYPE__SHIFT;
+
+	m->cp_hqd_ib_control = atc_bit << CP_HQD_IB_CONTROL__IB_ATC__SHIFT |
+			3 << CP_HQD_IB_CONTROL__MIN_IB_AVAIL_SIZE__SHIFT |
+			mtype << CP_HQD_IB_CONTROL__MTYPE__SHIFT;
+
+	m->cp_hqd_eop_control |=
+		ffs(q->eop_ring_buffer_size / sizeof(unsigned int)) - 1 - 1;
+	m->cp_hqd_eop_base_addr_lo =
+			lower_32_bits(q->eop_ring_buffer_address >> 8);
+	m->cp_hqd_eop_base_addr_hi =
+			upper_32_bits(q->eop_ring_buffer_address >> 8);
+
+	m->cp_hqd_iq_timer = atc_bit << CP_HQD_IQ_TIMER__IQ_ATC__SHIFT |
+			mtype << CP_HQD_IQ_TIMER__MTYPE__SHIFT;
+
+	m->cp_hqd_vmid = q->vmid;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL) {
+		m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
+				2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT;
+	}
+
+	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(struct mqd_manager *mm, void *mqd,
+			struct queue_properties *q)
+{
+	return __update_mqd(mm, mqd, q, MTYPE_CC, 1);
+}
+
+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 mm->dev->kfd2kgd->hqd_destroy
+		(mm->dev->kgd, type, timeout,
+		pipe_id, queue_id);
+}
+
+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 bool is_occupied(struct mqd_manager *mm, void *mqd,
+			uint64_t queue_address,	uint32_t pipe_id,
+			uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_is_occupied(
+		mm->dev->kgd, queue_address,
+		pipe_id, queue_id);
+}
+
+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)
+{
+	struct vi_mqd *m;
+	int retval = init_mqd(mm, mqd, mqd_mem_obj, gart_addr, q);
+
+	if (retval != 0)
+		return retval;
+
+	m = get_mqd(*mqd);
+
+	m->cp_hqd_pq_control |= 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
+			1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
+
+	return retval;
+}
+
+static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
+			struct queue_properties *q)
+{
+	struct vi_mqd *m;
+	int retval = __update_mqd(mm, mqd, q, MTYPE_UC, 0);
+
+	if (retval != 0)
+		return retval;
+
+	m = get_mqd(mqd);
+	m->cp_hqd_vmid = q->vmid;
+	return retval;
+}
 
 struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
-					struct kfd_dev *dev)
+		struct kfd_dev *dev)
 {
-	pr_warn("amdkfd: VI MQD is not currently supported\n");
-	return NULL;
+	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:
+		break;
+	default:
+		kfree(mqd);
+		return NULL;
+	}
+
+	return mqd;
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index 99b6d28a11c3..90f391434fa3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -27,6 +27,7 @@
 #include "kfd_kernel_queue.h"
 #include "kfd_priv.h"
 #include "kfd_pm4_headers.h"
+#include "kfd_pm4_headers_vi.h"
 #include "kfd_pm4_opcodes.h"
 
 static inline void inc_wptr(unsigned int *wptr, unsigned int increment_bytes,
@@ -55,6 +56,7 @@ static void pm_calc_rlib_size(struct packet_manager *pm,
 				bool *over_subscription)
 {
 	unsigned int process_count, queue_count;
+	unsigned int map_queue_size;
 
 	BUG_ON(!pm || !rlib_size || !over_subscription);
 
@@ -69,9 +71,13 @@ static void pm_calc_rlib_size(struct packet_manager *pm,
 		pr_debug("kfd: over subscribed runlist\n");
 	}
 
+	map_queue_size =
+		(pm->dqm->dev->device_info->asic_family == CHIP_CARRIZO) ?
+		sizeof(struct pm4_mes_map_queues) :
+		sizeof(struct pm4_map_queues);
 	/* calculate run list ib allocation size */
 	*rlib_size = process_count * sizeof(struct pm4_map_process) +
-		     queue_count * sizeof(struct pm4_map_queues);
+		     queue_count * map_queue_size;
 
 	/*
 	 * Increase the allocation size in case we need a chained run list
@@ -176,6 +182,71 @@ static int pm_create_map_process(struct packet_manager *pm, uint32_t *buffer,
 	return 0;
 }
 
+static int pm_create_map_queue_vi(struct packet_manager *pm, uint32_t *buffer,
+		struct queue *q, bool is_static)
+{
+	struct pm4_mes_map_queues *packet;
+	bool use_static = is_static;
+
+	BUG_ON(!pm || !buffer || !q);
+
+	pr_debug("kfd: In func %s\n", __func__);
+
+	packet = (struct pm4_mes_map_queues *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_map_queues));
+
+	packet->header.u32all = build_pm4_header(IT_MAP_QUEUES,
+						sizeof(struct pm4_map_queues));
+	packet->bitfields2.alloc_format =
+		alloc_format__mes_map_queues__one_per_pipe_vi;
+	packet->bitfields2.num_queues = 1;
+	packet->bitfields2.queue_sel =
+		queue_sel__mes_map_queues__map_to_hws_determined_queue_slots_vi;
+
+	packet->bitfields2.engine_sel =
+		engine_sel__mes_map_queues__compute_vi;
+	packet->bitfields2.queue_type =
+		queue_type__mes_map_queues__normal_compute_vi;
+
+	switch (q->properties.type) {
+	case KFD_QUEUE_TYPE_COMPUTE:
+		if (use_static)
+			packet->bitfields2.queue_type =
+		queue_type__mes_map_queues__normal_latency_static_queue_vi;
+		break;
+	case KFD_QUEUE_TYPE_DIQ:
+		packet->bitfields2.queue_type =
+			queue_type__mes_map_queues__debug_interface_queue_vi;
+		break;
+	case KFD_QUEUE_TYPE_SDMA:
+		packet->bitfields2.engine_sel =
+				engine_sel__mes_map_queues__sdma0_vi;
+		use_static = false; /* no static queues under SDMA */
+		break;
+	default:
+		pr_err("kfd: in %s queue type %d\n", __func__,
+				q->properties.type);
+		BUG();
+		break;
+	}
+	packet->bitfields3.doorbell_offset =
+			q->properties.doorbell_off;
+
+	packet->mqd_addr_lo =
+			lower_32_bits(q->gart_mqd_addr);
+
+	packet->mqd_addr_hi =
+			upper_32_bits(q->gart_mqd_addr);
+
+	packet->wptr_addr_lo =
+			lower_32_bits((uint64_t)q->properties.write_ptr);
+
+	packet->wptr_addr_hi =
+			upper_32_bits((uint64_t)q->properties.write_ptr);
+
+	return 0;
+}
+
 static int pm_create_map_queue(struct packet_manager *pm, uint32_t *buffer,
 				struct queue *q, bool is_static)
 {
@@ -292,8 +363,17 @@ static int pm_create_runlist_ib(struct packet_manager *pm,
 			pr_debug("kfd: static_queue, mapping kernel q %d, is debug status %d\n",
 				kq->queue->queue, qpd->is_debug);
 
-			retval = pm_create_map_queue(pm, &rl_buffer[rl_wptr],
-						kq->queue, qpd->is_debug);
+			if (pm->dqm->dev->device_info->asic_family ==
+					CHIP_CARRIZO)
+				retval = pm_create_map_queue_vi(pm,
+						&rl_buffer[rl_wptr],
+						kq->queue,
+						qpd->is_debug);
+			else
+				retval = pm_create_map_queue(pm,
+						&rl_buffer[rl_wptr],
+						kq->queue,
+						qpd->is_debug);
 			if (retval != 0)
 				return retval;
 
@@ -309,8 +389,17 @@ static int pm_create_runlist_ib(struct packet_manager *pm,
 			pr_debug("kfd: static_queue, mapping user queue %d, is debug status %d\n",
 				q->queue, qpd->is_debug);
 
-			retval = pm_create_map_queue(pm, &rl_buffer[rl_wptr],
-						q,  qpd->is_debug);
+			if (pm->dqm->dev->device_info->asic_family ==
+					CHIP_CARRIZO)
+				retval = pm_create_map_queue_vi(pm,
+						&rl_buffer[rl_wptr],
+						q,
+						qpd->is_debug);
+			else
+				retval = pm_create_map_queue(pm,
+						&rl_buffer[rl_wptr],
+						q,
+						qpd->is_debug);
 
 			if (retval != 0)
 				return retval;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_vi.h b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_vi.h
new file mode 100644
index 000000000000..08c721922812
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_vi.h
@@ -0,0 +1,398 @@
+/*
+ * 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.
+ *
+ */
+
+#ifndef F32_MES_PM4_PACKETS_H
+#define F32_MES_PM4_PACKETS_H
+
+#ifndef PM4_MES_HEADER_DEFINED
+#define PM4_MES_HEADER_DEFINED
+union PM4_MES_TYPE_3_HEADER {
+	struct {
+		uint32_t reserved1 : 8; /* < reserved */
+		uint32_t opcode    : 8; /* < IT opcode */
+		uint32_t count     : 14;/* < number of DWORDs - 1 in the
+		information body. */
+		uint32_t type      : 2; /* < packet identifier.
+					It should be 3 for type 3 packets */
+	};
+	uint32_t u32All;
+};
+#endif /* PM4_MES_HEADER_DEFINED */
+
+/*--------------------MES_SET_RESOURCES--------------------*/
+
+#ifndef PM4_MES_SET_RESOURCES_DEFINED
+#define PM4_MES_SET_RESOURCES_DEFINED
+enum mes_set_resources_queue_type_enum {
+	queue_type__mes_set_resources__kernel_interface_queue_kiq = 0,
+	queue_type__mes_set_resources__hsa_interface_queue_hiq = 1,
+	queue_type__mes_set_resources__hsa_debug_interface_queue = 4
+};
+
+
+struct pm4_mes_set_resources {
+	union {
+		union PM4_MES_TYPE_3_HEADER	header;		/* header */
+		uint32_t			ordinal1;
+	};
+
+	union {
+		struct {
+			uint32_t vmid_mask:16;
+			uint32_t unmap_latency:8;
+			uint32_t reserved1:5;
+			enum mes_set_resources_queue_type_enum queue_type:3;
+		} bitfields2;
+		uint32_t ordinal2;
+	};
+
+	uint32_t queue_mask_lo;
+	uint32_t queue_mask_hi;
+	uint32_t gws_mask_lo;
+	uint32_t gws_mask_hi;
+
+	union {
+		struct {
+			uint32_t oac_mask:16;
+			uint32_t reserved2:16;
+		} bitfields7;
+		uint32_t ordinal7;
+	};
+
+	union {
+		struct {
+		uint32_t gds_heap_base:6;
+		uint32_t reserved3:5;
+		uint32_t gds_heap_size:6;
+		uint32_t reserved4:15;
+		} bitfields8;
+		uint32_t ordinal8;
+	};
+
+};
+#endif
+
+/*--------------------MES_RUN_LIST--------------------*/
+
+#ifndef PM4_MES_RUN_LIST_DEFINED
+#define PM4_MES_RUN_LIST_DEFINED
+
+struct pm4_mes_runlist {
+	union {
+	    union PM4_MES_TYPE_3_HEADER   header;            /* header */
+	    uint32_t            ordinal1;
+	};
+
+	union {
+		struct {
+			uint32_t reserved1:2;
+			uint32_t ib_base_lo:30;
+		} bitfields2;
+		uint32_t ordinal2;
+	};
+
+	union {
+		struct {
+			uint32_t ib_base_hi:16;
+			uint32_t reserved2:16;
+		} bitfields3;
+		uint32_t ordinal3;
+	};
+
+	union {
+		struct {
+			uint32_t ib_size:20;
+			uint32_t chain:1;
+			uint32_t offload_polling:1;
+			uint32_t reserved3:1;
+			uint32_t valid:1;
+			uint32_t reserved4:8;
+		} bitfields4;
+		uint32_t ordinal4;
+	};
+
+};
+#endif
+
+/*--------------------MES_MAP_PROCESS--------------------*/
+
+#ifndef PM4_MES_MAP_PROCESS_DEFINED
+#define PM4_MES_MAP_PROCESS_DEFINED
+
+struct pm4_mes_map_process {
+	union {
+		union PM4_MES_TYPE_3_HEADER   header;            /* header */
+		uint32_t            ordinal1;
+	};
+
+	union {
+		struct {
+			uint32_t pasid:16;
+			uint32_t reserved1:8;
+			uint32_t diq_enable:1;
+			uint32_t process_quantum:7;
+		} bitfields2;
+		uint32_t ordinal2;
+};
+
+	union {
+		struct {
+			uint32_t page_table_base:28;
+			uint32_t reserved2:4;
+		} bitfields3;
+		uint32_t ordinal3;
+	};
+
+	uint32_t sh_mem_bases;
+	uint32_t sh_mem_ape1_base;
+	uint32_t sh_mem_ape1_limit;
+	uint32_t sh_mem_config;
+	uint32_t gds_addr_lo;
+	uint32_t gds_addr_hi;
+
+	union {
+		struct {
+			uint32_t num_gws:6;
+			uint32_t reserved3:2;
+			uint32_t num_oac:4;
+			uint32_t reserved4:4;
+			uint32_t gds_size:6;
+			uint32_t num_queues:10;
+		} bitfields10;
+		uint32_t ordinal10;
+	};
+
+};
+#endif
+
+/*--------------------MES_MAP_QUEUES--------------------*/
+
+#ifndef PM4_MES_MAP_QUEUES_VI_DEFINED
+#define PM4_MES_MAP_QUEUES_VI_DEFINED
+enum mes_map_queues_queue_sel_vi_enum {
+	queue_sel__mes_map_queues__map_to_specified_queue_slots_vi = 0,
+queue_sel__mes_map_queues__map_to_hws_determined_queue_slots_vi = 1
+};
+
+enum mes_map_queues_queue_type_vi_enum {
+	queue_type__mes_map_queues__normal_compute_vi = 0,
+	queue_type__mes_map_queues__debug_interface_queue_vi = 1,
+	queue_type__mes_map_queues__normal_latency_static_queue_vi = 2,
+queue_type__mes_map_queues__low_latency_static_queue_vi = 3
+};
+
+enum mes_map_queues_alloc_format_vi_enum {
+	alloc_format__mes_map_queues__one_per_pipe_vi = 0,
+alloc_format__mes_map_queues__all_on_one_pipe_vi = 1
+};
+
+enum mes_map_queues_engine_sel_vi_enum {
+	engine_sel__mes_map_queues__compute_vi = 0,
+	engine_sel__mes_map_queues__sdma0_vi = 2,
+	engine_sel__mes_map_queues__sdma1_vi = 3
+};
+
+
+struct pm4_mes_map_queues {
+	union {
+		union PM4_MES_TYPE_3_HEADER   header;            /* header */
+		uint32_t            ordinal1;
+	};
+
+	union {
+		struct {
+			uint32_t reserved1:4;
+			enum mes_map_queues_queue_sel_vi_enum queue_sel:2;
+			uint32_t reserved2:15;
+			enum mes_map_queues_queue_type_vi_enum queue_type:3;
+			enum mes_map_queues_alloc_format_vi_enum alloc_format:2;
+			enum mes_map_queues_engine_sel_vi_enum engine_sel:3;
+			uint32_t num_queues:3;
+		} bitfields2;
+		uint32_t ordinal2;
+	};
+
+	union {
+		struct {
+			uint32_t reserved3:1;
+			uint32_t check_disable:1;
+			uint32_t doorbell_offset:21;
+			uint32_t reserved4:3;
+			uint32_t queue:6;
+		} bitfields3;
+		uint32_t ordinal3;
+	};
+
+	uint32_t mqd_addr_lo;
+	uint32_t mqd_addr_hi;
+	uint32_t wptr_addr_lo;
+	uint32_t wptr_addr_hi;
+};
+#endif
+
+/*--------------------MES_QUERY_STATUS--------------------*/
+
+#ifndef PM4_MES_QUERY_STATUS_DEFINED
+#define PM4_MES_QUERY_STATUS_DEFINED
+enum mes_query_status_interrupt_sel_enum {
+	interrupt_sel__mes_query_status__completion_status = 0,
+	interrupt_sel__mes_query_status__process_status = 1,
+	interrupt_sel__mes_query_status__queue_status = 2
+};
+
+enum mes_query_status_command_enum {
+	command__mes_query_status__interrupt_only = 0,
+	command__mes_query_status__fence_only_immediate = 1,
+	command__mes_query_status__fence_only_after_write_ack = 2,
+	command__mes_query_status__fence_wait_for_write_ack_send_interrupt = 3
+};
+
+enum mes_query_status_engine_sel_enum {
+	engine_sel__mes_query_status__compute = 0,
+	engine_sel__mes_query_status__sdma0_queue = 2,
+	engine_sel__mes_query_status__sdma1_queue = 3
+};
+
+struct pm4_mes_query_status {
+	union {
+		union PM4_MES_TYPE_3_HEADER   header;            /* header */
+		uint32_t            ordinal1;
+	};
+
+	union {
+		struct {
+			uint32_t context_id:28;
+			enum mes_query_status_interrupt_sel_enum
+				interrupt_sel:2;
+			enum mes_query_status_command_enum command:2;
+		} bitfields2;
+		uint32_t ordinal2;
+	};
+
+	union {
+		struct {
+			uint32_t pasid:16;
+			uint32_t reserved1:16;
+		} bitfields3a;
+		struct {
+			uint32_t reserved2:2;
+			uint32_t doorbell_offset:21;
+			uint32_t reserved3:2;
+			enum mes_query_status_engine_sel_enum engine_sel:3;
+			uint32_t reserved4:4;
+		} bitfields3b;
+		uint32_t ordinal3;
+	};
+
+	uint32_t addr_lo;
+	uint32_t addr_hi;
+	uint32_t data_lo;
+	uint32_t data_hi;
+};
+#endif
+
+/*--------------------MES_UNMAP_QUEUES--------------------*/
+
+#ifndef PM4_MES_UNMAP_QUEUES_DEFINED
+#define PM4_MES_UNMAP_QUEUES_DEFINED
+enum mes_unmap_queues_action_enum {
+	action__mes_unmap_queues__preempt_queues = 0,
+	action__mes_unmap_queues__reset_queues = 1,
+	action__mes_unmap_queues__disable_process_queues = 2,
+	action__mes_unmap_queues__reserved = 3
+};
+
+enum mes_unmap_queues_queue_sel_enum {
+	queue_sel__mes_unmap_queues__perform_request_on_specified_queues = 0,
+	queue_sel__mes_unmap_queues__perform_request_on_pasid_queues = 1,
+	queue_sel__mes_unmap_queues__unmap_all_queues = 2,
+	queue_sel__mes_unmap_queues__unmap_all_non_static_queues = 3
+};
+
+enum mes_unmap_queues_engine_sel_enum {
+	engine_sel__mes_unmap_queues__compute = 0,
+	engine_sel__mes_unmap_queues__sdma0 = 2,
+	engine_sel__mes_unmap_queues__sdmal = 3
+};
+
+struct PM4_MES_UNMAP_QUEUES {
+	union {
+		union PM4_MES_TYPE_3_HEADER   header;            /* header */
+		uint32_t            ordinal1;
+	};
+
+	union {
+		struct {
+			enum mes_unmap_queues_action_enum action:2;
+			uint32_t reserved1:2;
+			enum mes_unmap_queues_queue_sel_enum queue_sel:2;
+			uint32_t reserved2:20;
+			enum mes_unmap_queues_engine_sel_enum engine_sel:3;
+			uint32_t num_queues:3;
+		} bitfields2;
+		uint32_t ordinal2;
+	};
+
+	union {
+		struct {
+			uint32_t pasid:16;
+			uint32_t reserved3:16;
+		} bitfields3a;
+		struct {
+			uint32_t reserved4:2;
+			uint32_t doorbell_offset0:21;
+			uint32_t reserved5:9;
+		} bitfields3b;
+		uint32_t ordinal3;
+	};
+
+	union {
+	struct {
+			uint32_t reserved6:2;
+			uint32_t doorbell_offset1:21;
+			uint32_t reserved7:9;
+		} bitfields4;
+		uint32_t ordinal4;
+	};
+
+	union {
+		struct {
+			uint32_t reserved8:2;
+			uint32_t doorbell_offset2:21;
+			uint32_t reserved9:9;
+		} bitfields5;
+		uint32_t ordinal5;
+	};
+
+	union {
+		struct {
+			uint32_t reserved10:2;
+			uint32_t doorbell_offset3:21;
+			uint32_t reserved11:9;
+		} bitfields6;
+		uint32_t ordinal6;
+	};
+};
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index c25728bc388a..74909e72a009 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1186,6 +1186,11 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	 * TODO: Retrieve max engine clock values from KGD
 	 */
 
+	if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) {
+		dev->node_props.capability |= HSA_CAP_DOORBELL_PACKET_TYPE;
+		pr_info("amdkfd: adding doorbell packet type capability\n");
+	}
+
 	res = 0;
 
 err:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
index 989624b3cd14..c3ddb9b95ff8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
@@ -40,6 +40,7 @@
 #define HSA_CAP_WATCH_POINTS_TOTALBITS_MASK	0x00000f00
 #define HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT	8
 #define HSA_CAP_RESERVED			0xfffff000
+#define HSA_CAP_DOORBELL_PACKET_TYPE		0x00001000
 
 struct kfd_node_properties {
 	uint32_t cpu_cores_count;
diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
index 9080daa116b6..888250b33ea8 100644
--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
@@ -52,7 +52,8 @@ enum kgd_engine_type {
 	KGD_ENGINE_MEC1,
 	KGD_ENGINE_MEC2,
 	KGD_ENGINE_RLC,
-	KGD_ENGINE_SDMA,
+	KGD_ENGINE_SDMA1,
+	KGD_ENGINE_SDMA2,
 	KGD_ENGINE_MAX
 };
 
diff --git a/drivers/gpu/drm/amd/include/vi_structs.h b/drivers/gpu/drm/amd/include/vi_structs.h
new file mode 100644
index 000000000000..65cfacd7a66c
--- /dev/null
+++ b/drivers/gpu/drm/amd/include/vi_structs.h
@@ -0,0 +1,417 @@
+/*
+ * Copyright 2012 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.
+ *
+ */
+
+#ifndef VI_STRUCTS_H_
+#define VI_STRUCTS_H_
+
+struct vi_sdma_mqd {
+	uint32_t sdmax_rlcx_rb_cntl;
+	uint32_t sdmax_rlcx_rb_base;
+	uint32_t sdmax_rlcx_rb_base_hi;
+	uint32_t sdmax_rlcx_rb_rptr;
+	uint32_t sdmax_rlcx_rb_wptr;
+	uint32_t sdmax_rlcx_rb_wptr_poll_cntl;
+	uint32_t sdmax_rlcx_rb_wptr_poll_addr_hi;
+	uint32_t sdmax_rlcx_rb_wptr_poll_addr_lo;
+	uint32_t sdmax_rlcx_rb_rptr_addr_hi;
+	uint32_t sdmax_rlcx_rb_rptr_addr_lo;
+	uint32_t sdmax_rlcx_ib_cntl;
+	uint32_t sdmax_rlcx_ib_rptr;
+	uint32_t sdmax_rlcx_ib_offset;
+	uint32_t sdmax_rlcx_ib_base_lo;
+	uint32_t sdmax_rlcx_ib_base_hi;
+	uint32_t sdmax_rlcx_ib_size;
+	uint32_t sdmax_rlcx_skip_cntl;
+	uint32_t sdmax_rlcx_context_status;
+	uint32_t sdmax_rlcx_doorbell;
+	uint32_t sdmax_rlcx_virtual_addr;
+	uint32_t sdmax_rlcx_ape1_cntl;
+	uint32_t sdmax_rlcx_doorbell_log;
+	uint32_t reserved_22;
+	uint32_t reserved_23;
+	uint32_t reserved_24;
+	uint32_t reserved_25;
+	uint32_t reserved_26;
+	uint32_t reserved_27;
+	uint32_t reserved_28;
+	uint32_t reserved_29;
+	uint32_t reserved_30;
+	uint32_t reserved_31;
+	uint32_t reserved_32;
+	uint32_t reserved_33;
+	uint32_t reserved_34;
+	uint32_t reserved_35;
+	uint32_t reserved_36;
+	uint32_t reserved_37;
+	uint32_t reserved_38;
+	uint32_t reserved_39;
+	uint32_t reserved_40;
+	uint32_t reserved_41;
+	uint32_t reserved_42;
+	uint32_t reserved_43;
+	uint32_t reserved_44;
+	uint32_t reserved_45;
+	uint32_t reserved_46;
+	uint32_t reserved_47;
+	uint32_t reserved_48;
+	uint32_t reserved_49;
+	uint32_t reserved_50;
+	uint32_t reserved_51;
+	uint32_t reserved_52;
+	uint32_t reserved_53;
+	uint32_t reserved_54;
+	uint32_t reserved_55;
+	uint32_t reserved_56;
+	uint32_t reserved_57;
+	uint32_t reserved_58;
+	uint32_t reserved_59;
+	uint32_t reserved_60;
+	uint32_t reserved_61;
+	uint32_t reserved_62;
+	uint32_t reserved_63;
+	uint32_t reserved_64;
+	uint32_t reserved_65;
+	uint32_t reserved_66;
+	uint32_t reserved_67;
+	uint32_t reserved_68;
+	uint32_t reserved_69;
+	uint32_t reserved_70;
+	uint32_t reserved_71;
+	uint32_t reserved_72;
+	uint32_t reserved_73;
+	uint32_t reserved_74;
+	uint32_t reserved_75;
+	uint32_t reserved_76;
+	uint32_t reserved_77;
+	uint32_t reserved_78;
+	uint32_t reserved_79;
+	uint32_t reserved_80;
+	uint32_t reserved_81;
+	uint32_t reserved_82;
+	uint32_t reserved_83;
+	uint32_t reserved_84;
+	uint32_t reserved_85;
+	uint32_t reserved_86;
+	uint32_t reserved_87;
+	uint32_t reserved_88;
+	uint32_t reserved_89;
+	uint32_t reserved_90;
+	uint32_t reserved_91;
+	uint32_t reserved_92;
+	uint32_t reserved_93;
+	uint32_t reserved_94;
+	uint32_t reserved_95;
+	uint32_t reserved_96;
+	uint32_t reserved_97;
+	uint32_t reserved_98;
+	uint32_t reserved_99;
+	uint32_t reserved_100;
+	uint32_t reserved_101;
+	uint32_t reserved_102;
+	uint32_t reserved_103;
+	uint32_t reserved_104;
+	uint32_t reserved_105;
+	uint32_t reserved_106;
+	uint32_t reserved_107;
+	uint32_t reserved_108;
+	uint32_t reserved_109;
+	uint32_t reserved_110;
+	uint32_t reserved_111;
+	uint32_t reserved_112;
+	uint32_t reserved_113;
+	uint32_t reserved_114;
+	uint32_t reserved_115;
+	uint32_t reserved_116;
+	uint32_t reserved_117;
+	uint32_t reserved_118;
+	uint32_t reserved_119;
+	uint32_t reserved_120;
+	uint32_t reserved_121;
+	uint32_t reserved_122;
+	uint32_t reserved_123;
+	uint32_t reserved_124;
+	uint32_t reserved_125;
+	uint32_t reserved_126;
+	uint32_t reserved_127;
+};
+
+struct vi_mqd {
+	uint32_t header;
+	uint32_t compute_dispatch_initiator;
+	uint32_t compute_dim_x;
+	uint32_t compute_dim_y;
+	uint32_t compute_dim_z;
+	uint32_t compute_start_x;
+	uint32_t compute_start_y;
+	uint32_t compute_start_z;
+	uint32_t compute_num_thread_x;
+	uint32_t compute_num_thread_y;
+	uint32_t compute_num_thread_z;
+	uint32_t compute_pipelinestat_enable;
+	uint32_t compute_perfcount_enable;
+	uint32_t compute_pgm_lo;
+	uint32_t compute_pgm_hi;
+	uint32_t compute_tba_lo;
+	uint32_t compute_tba_hi;
+	uint32_t compute_tma_lo;
+	uint32_t compute_tma_hi;
+	uint32_t compute_pgm_rsrc1;
+	uint32_t compute_pgm_rsrc2;
+	uint32_t compute_vmid;
+	uint32_t compute_resource_limits;
+	uint32_t compute_static_thread_mgmt_se0;
+	uint32_t compute_static_thread_mgmt_se1;
+	uint32_t compute_tmpring_size;
+	uint32_t compute_static_thread_mgmt_se2;
+	uint32_t compute_static_thread_mgmt_se3;
+	uint32_t compute_restart_x;
+	uint32_t compute_restart_y;
+	uint32_t compute_restart_z;
+	uint32_t compute_thread_trace_enable;
+	uint32_t compute_misc_reserved;
+	uint32_t compute_dispatch_id;
+	uint32_t compute_threadgroup_id;
+	uint32_t compute_relaunch;
+	uint32_t compute_wave_restore_addr_lo;
+	uint32_t compute_wave_restore_addr_hi;
+	uint32_t compute_wave_restore_control;
+	uint32_t reserved_39;
+	uint32_t reserved_40;
+	uint32_t reserved_41;
+	uint32_t reserved_42;
+	uint32_t reserved_43;
+	uint32_t reserved_44;
+	uint32_t reserved_45;
+	uint32_t reserved_46;
+	uint32_t reserved_47;
+	uint32_t reserved_48;
+	uint32_t reserved_49;
+	uint32_t reserved_50;
+	uint32_t reserved_51;
+	uint32_t reserved_52;
+	uint32_t reserved_53;
+	uint32_t reserved_54;
+	uint32_t reserved_55;
+	uint32_t reserved_56;
+	uint32_t reserved_57;
+	uint32_t reserved_58;
+	uint32_t reserved_59;
+	uint32_t reserved_60;
+	uint32_t reserved_61;
+	uint32_t reserved_62;
+	uint32_t reserved_63;
+	uint32_t reserved_64;
+	uint32_t compute_user_data_0;
+	uint32_t compute_user_data_1;
+	uint32_t compute_user_data_2;
+	uint32_t compute_user_data_3;
+	uint32_t compute_user_data_4;
+	uint32_t compute_user_data_5;
+	uint32_t compute_user_data_6;
+	uint32_t compute_user_data_7;
+	uint32_t compute_user_data_8;
+	uint32_t compute_user_data_9;
+	uint32_t compute_user_data_10;
+	uint32_t compute_user_data_11;
+	uint32_t compute_user_data_12;
+	uint32_t compute_user_data_13;
+	uint32_t compute_user_data_14;
+	uint32_t compute_user_data_15;
+	uint32_t cp_compute_csinvoc_count_lo;
+	uint32_t cp_compute_csinvoc_count_hi;
+	uint32_t reserved_83;
+	uint32_t reserved_84;
+	uint32_t reserved_85;
+	uint32_t cp_mqd_query_time_lo;
+	uint32_t cp_mqd_query_time_hi;
+	uint32_t cp_mqd_connect_start_time_lo;
+	uint32_t cp_mqd_connect_start_time_hi;
+	uint32_t cp_mqd_connect_end_time_lo;
+	uint32_t cp_mqd_connect_end_time_hi;
+	uint32_t cp_mqd_connect_end_wf_count;
+	uint32_t cp_mqd_connect_end_pq_rptr;
+	uint32_t cp_mqd_connect_end_pq_wptr;
+	uint32_t cp_mqd_connect_end_ib_rptr;
+	uint32_t reserved_96;
+	uint32_t reserved_97;
+	uint32_t cp_mqd_save_start_time_lo;
+	uint32_t cp_mqd_save_start_time_hi;
+	uint32_t cp_mqd_save_end_time_lo;
+	uint32_t cp_mqd_save_end_time_hi;
+	uint32_t cp_mqd_restore_start_time_lo;
+	uint32_t cp_mqd_restore_start_time_hi;
+	uint32_t cp_mqd_restore_end_time_lo;
+	uint32_t cp_mqd_restore_end_time_hi;
+	uint32_t reserved_106;
+	uint32_t reserved_107;
+	uint32_t gds_cs_ctxsw_cnt0;
+	uint32_t gds_cs_ctxsw_cnt1;
+	uint32_t gds_cs_ctxsw_cnt2;
+	uint32_t gds_cs_ctxsw_cnt3;
+	uint32_t reserved_112;
+	uint32_t reserved_113;
+	uint32_t cp_pq_exe_status_lo;
+	uint32_t cp_pq_exe_status_hi;
+	uint32_t cp_packet_id_lo;
+	uint32_t cp_packet_id_hi;
+	uint32_t cp_packet_exe_status_lo;
+	uint32_t cp_packet_exe_status_hi;
+	uint32_t gds_save_base_addr_lo;
+	uint32_t gds_save_base_addr_hi;
+	uint32_t gds_save_mask_lo;
+	uint32_t gds_save_mask_hi;
+	uint32_t ctx_save_base_addr_lo;
+	uint32_t ctx_save_base_addr_hi;
+	uint32_t reserved_126;
+	uint32_t reserved_127;
+	uint32_t cp_mqd_base_addr_lo;
+	uint32_t cp_mqd_base_addr_hi;
+	uint32_t cp_hqd_active;
+	uint32_t cp_hqd_vmid;
+	uint32_t cp_hqd_persistent_state;
+	uint32_t cp_hqd_pipe_priority;
+	uint32_t cp_hqd_queue_priority;
+	uint32_t cp_hqd_quantum;
+	uint32_t cp_hqd_pq_base_lo;
+	uint32_t cp_hqd_pq_base_hi;
+	uint32_t cp_hqd_pq_rptr;
+	uint32_t cp_hqd_pq_rptr_report_addr_lo;
+	uint32_t cp_hqd_pq_rptr_report_addr_hi;
+	uint32_t cp_hqd_pq_wptr_poll_addr_lo;
+	uint32_t cp_hqd_pq_wptr_poll_addr_hi;
+	uint32_t cp_hqd_pq_doorbell_control;
+	uint32_t cp_hqd_pq_wptr;
+	uint32_t cp_hqd_pq_control;
+	uint32_t cp_hqd_ib_base_addr_lo;
+	uint32_t cp_hqd_ib_base_addr_hi;
+	uint32_t cp_hqd_ib_rptr;
+	uint32_t cp_hqd_ib_control;
+	uint32_t cp_hqd_iq_timer;
+	uint32_t cp_hqd_iq_rptr;
+	uint32_t cp_hqd_dequeue_request;
+	uint32_t cp_hqd_dma_offload;
+	uint32_t cp_hqd_sema_cmd;
+	uint32_t cp_hqd_msg_type;
+	uint32_t cp_hqd_atomic0_preop_lo;
+	uint32_t cp_hqd_atomic0_preop_hi;
+	uint32_t cp_hqd_atomic1_preop_lo;
+	uint32_t cp_hqd_atomic1_preop_hi;
+	uint32_t cp_hqd_hq_status0;
+	uint32_t cp_hqd_hq_control0;
+	uint32_t cp_mqd_control;
+	uint32_t cp_hqd_hq_status1;
+	uint32_t cp_hqd_hq_control1;
+	uint32_t cp_hqd_eop_base_addr_lo;
+	uint32_t cp_hqd_eop_base_addr_hi;
+	uint32_t cp_hqd_eop_control;
+	uint32_t cp_hqd_eop_rptr;
+	uint32_t cp_hqd_eop_wptr;
+	uint32_t cp_hqd_eop_done_events;
+	uint32_t cp_hqd_ctx_save_base_addr_lo;
+	uint32_t cp_hqd_ctx_save_base_addr_hi;
+	uint32_t cp_hqd_ctx_save_control;
+	uint32_t cp_hqd_cntl_stack_offset;
+	uint32_t cp_hqd_cntl_stack_size;
+	uint32_t cp_hqd_wg_state_offset;
+	uint32_t cp_hqd_ctx_save_size;
+	uint32_t cp_hqd_gds_resource_state;
+	uint32_t cp_hqd_error;
+	uint32_t cp_hqd_eop_wptr_mem;
+	uint32_t cp_hqd_eop_dones;
+	uint32_t reserved_182;
+	uint32_t reserved_183;
+	uint32_t reserved_184;
+	uint32_t reserved_185;
+	uint32_t reserved_186;
+	uint32_t reserved_187;
+	uint32_t reserved_188;
+	uint32_t reserved_189;
+	uint32_t reserved_190;
+	uint32_t reserved_191;
+	uint32_t iqtimer_pkt_header;
+	uint32_t iqtimer_pkt_dw0;
+	uint32_t iqtimer_pkt_dw1;
+	uint32_t iqtimer_pkt_dw2;
+	uint32_t iqtimer_pkt_dw3;
+	uint32_t iqtimer_pkt_dw4;
+	uint32_t iqtimer_pkt_dw5;
+	uint32_t iqtimer_pkt_dw6;
+	uint32_t iqtimer_pkt_dw7;
+	uint32_t iqtimer_pkt_dw8;
+	uint32_t iqtimer_pkt_dw9;
+	uint32_t iqtimer_pkt_dw10;
+	uint32_t iqtimer_pkt_dw11;
+	uint32_t iqtimer_pkt_dw12;
+	uint32_t iqtimer_pkt_dw13;
+	uint32_t iqtimer_pkt_dw14;
+	uint32_t iqtimer_pkt_dw15;
+	uint32_t iqtimer_pkt_dw16;
+	uint32_t iqtimer_pkt_dw17;
+	uint32_t iqtimer_pkt_dw18;
+	uint32_t iqtimer_pkt_dw19;
+	uint32_t iqtimer_pkt_dw20;
+	uint32_t iqtimer_pkt_dw21;
+	uint32_t iqtimer_pkt_dw22;
+	uint32_t iqtimer_pkt_dw23;
+	uint32_t iqtimer_pkt_dw24;
+	uint32_t iqtimer_pkt_dw25;
+	uint32_t iqtimer_pkt_dw26;
+	uint32_t iqtimer_pkt_dw27;
+	uint32_t iqtimer_pkt_dw28;
+	uint32_t iqtimer_pkt_dw29;
+	uint32_t iqtimer_pkt_dw30;
+	uint32_t iqtimer_pkt_dw31;
+	uint32_t reserved_225;
+	uint32_t reserved_226;
+	uint32_t reserved_227;
+	uint32_t set_resources_header;
+	uint32_t set_resources_dw1;
+	uint32_t set_resources_dw2;
+	uint32_t set_resources_dw3;
+	uint32_t set_resources_dw4;
+	uint32_t set_resources_dw5;
+	uint32_t set_resources_dw6;
+	uint32_t set_resources_dw7;
+	uint32_t reserved_236;
+	uint32_t reserved_237;
+	uint32_t reserved_238;
+	uint32_t reserved_239;
+	uint32_t queue_doorbell_id0;
+	uint32_t queue_doorbell_id1;
+	uint32_t queue_doorbell_id2;
+	uint32_t queue_doorbell_id3;
+	uint32_t queue_doorbell_id4;
+	uint32_t queue_doorbell_id5;
+	uint32_t queue_doorbell_id6;
+	uint32_t queue_doorbell_id7;
+	uint32_t queue_doorbell_id8;
+	uint32_t queue_doorbell_id9;
+	uint32_t queue_doorbell_id10;
+	uint32_t queue_doorbell_id11;
+	uint32_t queue_doorbell_id12;
+	uint32_t queue_doorbell_id13;
+	uint32_t queue_doorbell_id14;
+	uint32_t queue_doorbell_id15;
+};
+
+#endif /* VI_STRUCTS_H_ */
diff --git a/drivers/gpu/drm/radeon/radeon_kfd.c b/drivers/gpu/drm/radeon/radeon_kfd.c
index e476c331f3fa..9a4d69e59401 100644
--- a/drivers/gpu/drm/radeon/radeon_kfd.c
+++ b/drivers/gpu/drm/radeon/radeon_kfd.c
@@ -845,7 +845,8 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
 		hdr = (const union radeon_firmware_header *) rdev->rlc_fw->data;
 		break;
 
-	case KGD_ENGINE_SDMA:
+	case KGD_ENGINE_SDMA1:
+	case KGD_ENGINE_SDMA2:
 		hdr = (const union radeon_firmware_header *)
 							rdev->sdma_fw->data;
 		break;