xref: /linux/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c (revision 92c4c9fdc838d3b41a996bb700ea64b9e78fc7ea)

/*
 * Copyright 2025 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/firmware.h>
#include <linux/module.h>
#include "amdgpu.h"
#include "soc15_common.h"
#include "soc_v1_0.h"
#include "gc/gc_12_1_0_offset.h"
#include "gc/gc_12_1_0_sh_mask.h"
#include "gc/gc_11_0_0_default.h"
#include "v12_structs.h"
#include "mes_v12_api_def.h"
#include "gfx_v12_1_pkt.h"
#include "sdma_v7_1_0_pkt_open.h"

MODULE_FIRMWARE("amdgpu/gc_12_1_0_mes.bin");
MODULE_FIRMWARE("amdgpu/gc_12_1_0_mes1.bin");
MODULE_FIRMWARE("amdgpu/gc_12_1_0_uni_mes.bin");

static int mes_v12_1_hw_init(struct amdgpu_ip_block *ip_block);
static int mes_v12_1_xcc_hw_init(struct amdgpu_ip_block *ip_block, int xcc_id);
static int mes_v12_1_hw_fini(struct amdgpu_ip_block *ip_block);
static int mes_v12_1_kiq_hw_init(struct amdgpu_device *adev, uint32_t xcc_id);
static int mes_v12_1_kiq_hw_fini(struct amdgpu_device *adev, uint32_t xcc_id);
static int mes_v12_1_self_test(struct amdgpu_device *adev, int xcc_id);

#define MES_EOP_SIZE   2048

#define regCP_HQD_IB_CONTROL_MES_12_1_DEFAULT 0x100000
#define XCC_MID_MASK 0x41000000

static void mes_v12_1_ring_set_wptr(struct amdgpu_ring *ring)
{
	struct amdgpu_device *adev = ring->adev;

	if (ring->use_doorbell) {
		atomic64_set((atomic64_t *)ring->wptr_cpu_addr,
			     ring->wptr);
		WDOORBELL64(ring->doorbell_index, ring->wptr);
	} else {
		BUG();
	}
}

static u64 mes_v12_1_ring_get_rptr(struct amdgpu_ring *ring)
{
	return *ring->rptr_cpu_addr;
}

static u64 mes_v12_1_ring_get_wptr(struct amdgpu_ring *ring)
{
	u64 wptr;

	if (ring->use_doorbell)
		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
	else
		BUG();
	return wptr;
}

static const struct amdgpu_ring_funcs mes_v12_1_ring_funcs = {
	.type = AMDGPU_RING_TYPE_MES,
	.align_mask = 1,
	.nop = 0,
	.support_64bit_ptrs = true,
	.get_rptr = mes_v12_1_ring_get_rptr,
	.get_wptr = mes_v12_1_ring_get_wptr,
	.set_wptr = mes_v12_1_ring_set_wptr,
	.insert_nop = amdgpu_ring_insert_nop,
};

static const char *mes_v12_1_opcodes[] = {
	"SET_HW_RSRC",
	"SET_SCHEDULING_CONFIG",
	"ADD_QUEUE",
	"REMOVE_QUEUE",
	"PERFORM_YIELD",
	"SET_GANG_PRIORITY_LEVEL",
	"SUSPEND",
	"RESUME",
	"RESET",
	"SET_LOG_BUFFER",
	"CHANGE_GANG_PRORITY",
	"QUERY_SCHEDULER_STATUS",
	"unused",
	"SET_DEBUG_VMID",
	"MISC",
	"UPDATE_ROOT_PAGE_TABLE",
	"AMD_LOG",
	"SET_SE_MODE",
	"SET_GANG_SUBMIT",
	"SET_HW_RSRC_1",
	"INVALIDATE_TLBS",
};

static const char *mes_v12_1_misc_opcodes[] = {
	"WRITE_REG",
	"INV_GART",
	"QUERY_STATUS",
	"READ_REG",
	"WAIT_REG_MEM",
	"SET_SHADER_DEBUGGER",
	"NOTIFY_WORK_ON_UNMAPPED_QUEUE",
	"NOTIFY_TO_UNMAP_PROCESSES",
};

static const char *mes_v12_1_get_op_string(union MESAPI__MISC *x_pkt)
{
	const char *op_str = NULL;

	if (x_pkt->header.opcode < ARRAY_SIZE(mes_v12_1_opcodes))
		op_str = mes_v12_1_opcodes[x_pkt->header.opcode];

	return op_str;
}

static const char *mes_v12_1_get_misc_op_string(union MESAPI__MISC *x_pkt)
{
	const char *op_str = NULL;

	if ((x_pkt->header.opcode == MES_SCH_API_MISC) &&
	    (x_pkt->opcode < ARRAY_SIZE(mes_v12_1_misc_opcodes)))
		op_str = mes_v12_1_misc_opcodes[x_pkt->opcode];

	return op_str;
}

static int mes_v12_1_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
					    int xcc_id, int pipe, void *pkt,
					    int size, int api_status_off)
{
	union MESAPI__QUERY_MES_STATUS mes_status_pkt;
	signed long timeout = 2100000; /* 2100 ms */
	struct amdgpu_device *adev = mes->adev;
	struct amdgpu_ring *ring = &mes->ring[MES_PIPE_INST(xcc_id, pipe)];
	spinlock_t *ring_lock = &mes->ring_lock[MES_PIPE_INST(xcc_id, pipe)];
	struct MES_API_STATUS *api_status;
	union MESAPI__MISC *x_pkt = pkt;
	const char *op_str, *misc_op_str;
	unsigned long flags;
	u64 status_gpu_addr;
	u32 seq, status_offset;
	u64 *status_ptr;
	signed long r;
	int ret;

	if (x_pkt->header.opcode >= MES_SCH_API_MAX)
		return -EINVAL;

	if (amdgpu_emu_mode) {
		timeout *= 1000;
	} else if (amdgpu_sriov_vf(adev)) {
		/* Worst case in sriov where all other 15 VF timeout, each VF needs about 600ms */
		timeout = 15 * 600 * 1000;
	}

	ret = amdgpu_device_wb_get(adev, &status_offset);
	if (ret)
		return ret;

	status_gpu_addr = adev->wb.gpu_addr + (status_offset * 4);
	status_ptr = (u64 *)&adev->wb.wb[status_offset];
	*status_ptr = 0;

	spin_lock_irqsave(ring_lock, flags);
	r = amdgpu_ring_alloc(ring, (size + sizeof(mes_status_pkt)) / 4);
	if (r)
		goto error_unlock_free;

	seq = ++ring->fence_drv.sync_seq;
	r = amdgpu_fence_wait_polling(ring,
				      seq - ring->fence_drv.num_fences_mask,
				      timeout);
	if (r < 1)
		goto error_undo;

	api_status = (struct MES_API_STATUS *)((char *)pkt + api_status_off);
	api_status->api_completion_fence_addr = status_gpu_addr;
	api_status->api_completion_fence_value = 1;

	amdgpu_ring_write_multiple(ring, pkt, size / 4);

	memset(&mes_status_pkt, 0, sizeof(mes_status_pkt));
	mes_status_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_status_pkt.header.opcode = MES_SCH_API_QUERY_SCHEDULER_STATUS;
	mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
	mes_status_pkt.api_status.api_completion_fence_addr =
		ring->fence_drv.gpu_addr;
	mes_status_pkt.api_status.api_completion_fence_value = seq;

	amdgpu_ring_write_multiple(ring, &mes_status_pkt,
				   sizeof(mes_status_pkt) / 4);

	amdgpu_ring_commit(ring);
	spin_unlock_irqrestore(ring_lock, flags);

	op_str = mes_v12_1_get_op_string(x_pkt);
	misc_op_str = mes_v12_1_get_misc_op_string(x_pkt);

	if (misc_op_str)
		dev_dbg(adev->dev, "MES(%d, %d) msg=%s (%s) was emitted\n",
			xcc_id, pipe, op_str, misc_op_str);
	else if (op_str)
		dev_dbg(adev->dev, "MES(%d, %d) msg=%s was emitted\n",
			xcc_id, pipe, op_str);
	else
		dev_dbg(adev->dev, "MES(%d, %d) msg=%d was emitted\n",
			xcc_id, pipe, x_pkt->header.opcode);

	r = amdgpu_fence_wait_polling(ring, seq, timeout);
	if (r < 1 || !*status_ptr) {
		if (misc_op_str)
			dev_err(adev->dev,
				"MES(%d, %d) failed to respond to msg=%s (%s)\n",
				xcc_id, pipe, op_str, misc_op_str);
		else if (op_str)
			dev_err(adev->dev,
				"MES(%d, %d) failed to respond to msg=%s\n",
				xcc_id, pipe, op_str);
		else
			dev_err(adev->dev,
				"MES(%d, %d) failed to respond to msg=%d\n",
				xcc_id, pipe, x_pkt->header.opcode);

		while (halt_if_hws_hang)
			schedule();

		r = -ETIMEDOUT;
		goto error_wb_free;
	}

	amdgpu_device_wb_free(adev, status_offset);
	return 0;

error_undo:
	dev_err(adev->dev, "MES(%d, %d) ring buffer is full.\n", xcc_id, pipe);
	amdgpu_ring_undo(ring);

error_unlock_free:
	spin_unlock_irqrestore(ring_lock, flags);

error_wb_free:
	amdgpu_device_wb_free(adev, status_offset);
	return r;
}

static int convert_to_mes_queue_type(int queue_type)
{
	if (queue_type == AMDGPU_RING_TYPE_GFX)
		return MES_QUEUE_TYPE_GFX;
	else if (queue_type == AMDGPU_RING_TYPE_COMPUTE)
		return MES_QUEUE_TYPE_COMPUTE;
	else if (queue_type == AMDGPU_RING_TYPE_SDMA)
		return MES_QUEUE_TYPE_SDMA;
	else if (queue_type == AMDGPU_RING_TYPE_MES)
		return MES_QUEUE_TYPE_SCHQ;
	else
		BUG();
	return -1;
}

static int mes_v12_1_add_hw_queue(struct amdgpu_mes *mes,
				  struct mes_add_queue_input *input)
{
	union MESAPI__ADD_QUEUE mes_add_queue_pkt;
	int xcc_id = input->xcc_id;
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_SCHED_PIPE);

	if (mes->enable_coop_mode)
		xcc_id = mes->master_xcc_ids[inst];

	memset(&mes_add_queue_pkt, 0, sizeof(mes_add_queue_pkt));

	mes_add_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_add_queue_pkt.header.opcode = MES_SCH_API_ADD_QUEUE;
	mes_add_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_add_queue_pkt.process_id = input->process_id;
	mes_add_queue_pkt.page_table_base_addr = input->page_table_base_addr;
	mes_add_queue_pkt.process_va_start = input->process_va_start;
	mes_add_queue_pkt.process_va_end = input->process_va_end;
	mes_add_queue_pkt.process_quantum = input->process_quantum;
	mes_add_queue_pkt.process_context_addr = input->process_context_addr;
	mes_add_queue_pkt.gang_quantum = input->gang_quantum;
	mes_add_queue_pkt.gang_context_addr = input->gang_context_addr;
	mes_add_queue_pkt.inprocess_gang_priority =
		input->inprocess_gang_priority;
	mes_add_queue_pkt.gang_global_priority_level =
		input->gang_global_priority_level;
	mes_add_queue_pkt.doorbell_offset = input->doorbell_offset;
	mes_add_queue_pkt.mqd_addr = input->mqd_addr;

	mes_add_queue_pkt.wptr_addr = input->wptr_mc_addr;

	mes_add_queue_pkt.queue_type =
		convert_to_mes_queue_type(input->queue_type);
	mes_add_queue_pkt.paging = input->paging;
	mes_add_queue_pkt.vm_context_cntl = input->vm_cntx_cntl;
	mes_add_queue_pkt.gws_base = input->gws_base;
	mes_add_queue_pkt.gws_size = input->gws_size;
	mes_add_queue_pkt.trap_handler_addr = input->tba_addr;
	mes_add_queue_pkt.tma_addr = input->tma_addr;
	mes_add_queue_pkt.trap_en = input->trap_en;
	mes_add_queue_pkt.skip_process_ctx_clear = input->skip_process_ctx_clear;
	mes_add_queue_pkt.is_kfd_process = input->is_kfd_process;

	/* For KFD, gds_size is re-used for queue size (needed in MES for AQL queues) */
	mes_add_queue_pkt.is_aql_queue = input->is_aql_queue;
	mes_add_queue_pkt.gds_size = input->queue_size;

	/* For KFD, gds_size is re-used for queue size (needed in MES for AQL queues) */
	mes_add_queue_pkt.is_aql_queue = input->is_aql_queue;
	mes_add_queue_pkt.gds_size = input->queue_size;

	mes_add_queue_pkt.full_sh_mem_config_data = input->sh_mem_config_data;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			xcc_id, AMDGPU_MES_SCHED_PIPE,
			&mes_add_queue_pkt, sizeof(mes_add_queue_pkt),
			offsetof(union MESAPI__ADD_QUEUE, api_status));
}

static int mes_v12_1_remove_hw_queue(struct amdgpu_mes *mes,
				     struct mes_remove_queue_input *input)
{
	union MESAPI__REMOVE_QUEUE mes_remove_queue_pkt;
	int xcc_id = input->xcc_id;
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_SCHED_PIPE);

	if (mes->enable_coop_mode)
		xcc_id = mes->master_xcc_ids[inst];

	memset(&mes_remove_queue_pkt, 0, sizeof(mes_remove_queue_pkt));

	mes_remove_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_remove_queue_pkt.header.opcode = MES_SCH_API_REMOVE_QUEUE;
	mes_remove_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_remove_queue_pkt.doorbell_offset = input->doorbell_offset;
	mes_remove_queue_pkt.gang_context_addr = input->gang_context_addr;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			xcc_id, AMDGPU_MES_SCHED_PIPE,
			&mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt),
			offsetof(union MESAPI__REMOVE_QUEUE, api_status));
}

static int mes_v12_1_reset_hw_queue(struct amdgpu_mes *mes,
				    struct mes_reset_queue_input *input)
{
	union MESAPI__RESET mes_reset_queue_pkt;
	int pipe;

	memset(&mes_reset_queue_pkt, 0, sizeof(mes_reset_queue_pkt));

	mes_reset_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_reset_queue_pkt.header.opcode = MES_SCH_API_RESET;
	mes_reset_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_reset_queue_pkt.doorbell_offset = input->doorbell_offset;
	/* mes_reset_queue_pkt.gang_context_addr = input->gang_context_addr; */
	/*mes_reset_queue_pkt.reset_queue_only = 1;*/

	if (mes->adev->enable_uni_mes)
		pipe = AMDGPU_MES_KIQ_PIPE;
	else
		pipe = AMDGPU_MES_SCHED_PIPE;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			input->xcc_id, pipe,
			&mes_reset_queue_pkt, sizeof(mes_reset_queue_pkt),
			offsetof(union MESAPI__REMOVE_QUEUE, api_status));
}

static int mes_v12_1_map_legacy_queue(struct amdgpu_mes *mes,
				      struct mes_map_legacy_queue_input *input)
{
	union MESAPI__ADD_QUEUE mes_add_queue_pkt;
	int pipe;

	memset(&mes_add_queue_pkt, 0, sizeof(mes_add_queue_pkt));

	mes_add_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_add_queue_pkt.header.opcode = MES_SCH_API_ADD_QUEUE;
	mes_add_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_add_queue_pkt.pipe_id = input->pipe_id;
	mes_add_queue_pkt.queue_id = input->queue_id;
	mes_add_queue_pkt.doorbell_offset = input->doorbell_offset;
	mes_add_queue_pkt.mqd_addr = input->mqd_addr;
	mes_add_queue_pkt.wptr_addr = input->wptr_addr;
	mes_add_queue_pkt.queue_type =
		convert_to_mes_queue_type(input->queue_type);
	mes_add_queue_pkt.map_legacy_kq = 1;

	if (mes->adev->enable_uni_mes)
		pipe = AMDGPU_MES_KIQ_PIPE;
	else
		pipe = AMDGPU_MES_SCHED_PIPE;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			input->xcc_id, pipe,
			&mes_add_queue_pkt, sizeof(mes_add_queue_pkt),
			offsetof(union MESAPI__ADD_QUEUE, api_status));
}

static int mes_v12_1_unmap_legacy_queue(struct amdgpu_mes *mes,
			struct mes_unmap_legacy_queue_input *input)
{
	union MESAPI__REMOVE_QUEUE mes_remove_queue_pkt;
	int pipe;

	memset(&mes_remove_queue_pkt, 0, sizeof(mes_remove_queue_pkt));

	mes_remove_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_remove_queue_pkt.header.opcode = MES_SCH_API_REMOVE_QUEUE;
	mes_remove_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_remove_queue_pkt.doorbell_offset = input->doorbell_offset;
	mes_remove_queue_pkt.gang_context_addr = 0;

	mes_remove_queue_pkt.pipe_id = input->pipe_id;
	mes_remove_queue_pkt.queue_id = input->queue_id;

	if (input->action == PREEMPT_QUEUES_NO_UNMAP) {
		mes_remove_queue_pkt.preempt_legacy_gfx_queue = 1;
		mes_remove_queue_pkt.tf_addr = input->trail_fence_addr;
		mes_remove_queue_pkt.tf_data =
			lower_32_bits(input->trail_fence_data);
	} else {
		mes_remove_queue_pkt.unmap_legacy_queue = 1;
		mes_remove_queue_pkt.queue_type =
			convert_to_mes_queue_type(input->queue_type);
	}

	if (mes->adev->enable_uni_mes)
		pipe = AMDGPU_MES_KIQ_PIPE;
	else
		pipe = AMDGPU_MES_SCHED_PIPE;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			input->xcc_id, pipe,
			&mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt),
			offsetof(union MESAPI__REMOVE_QUEUE, api_status));
}

static int mes_v12_1_suspend_gang(struct amdgpu_mes *mes,
				  struct mes_suspend_gang_input *input)
{
	return 0;
}

static int mes_v12_1_resume_gang(struct amdgpu_mes *mes,
				 struct mes_resume_gang_input *input)
{
	return 0;
}

static int mes_v12_1_query_sched_status(struct amdgpu_mes *mes,
					  int pipe, int xcc_id)
{
	union MESAPI__QUERY_MES_STATUS mes_status_pkt;

	memset(&mes_status_pkt, 0, sizeof(mes_status_pkt));

	mes_status_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_status_pkt.header.opcode = MES_SCH_API_QUERY_SCHEDULER_STATUS;
	mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
			&mes_status_pkt, sizeof(mes_status_pkt),
			offsetof(union MESAPI__QUERY_MES_STATUS, api_status));
}
static uint32_t mes_v12_1_get_xcc_from_reg(uint32_t reg_offset)
{
	return ((reg_offset >> 16) & 0x7);
}

static void mes_v12_1_get_rrmt(uint32_t reg, uint32_t xcc_id,
			       struct RRMT_OPTION *rrmt_opt,
			       uint32_t *out_reg)
{
	uint32_t normalized_reg = soc_v1_0_normalize_xcc_reg_offset(reg);

	if (soc_v1_0_normalize_xcc_reg_range(normalized_reg)) {
		rrmt_opt->xcd_die_id = mes_v12_1_get_xcc_from_reg(reg);
		rrmt_opt->mode = (xcc_id == rrmt_opt->xcd_die_id) ?
			 MES_RRMT_MODE_LOCAL_XCD : MES_RRMT_MODE_REMOTE_XCD;
	} else {
		rrmt_opt->mode = MES_RRMT_MODE_REMOTE_MID;
		if (soc_v1_0_mid1_reg_range(reg))
			rrmt_opt->mid_die_id = 1;
	}

	*out_reg = soc_v1_0_normalize_reg_offset(reg);
}

static int mes_v12_1_misc_op(struct amdgpu_mes *mes,
			     struct mes_misc_op_input *input)
{
	struct amdgpu_device *adev = mes->adev;
	union MESAPI__MISC misc_pkt;
	int pipe;

	if (mes->adev->enable_uni_mes)
		pipe = AMDGPU_MES_KIQ_PIPE;
	else
		pipe = AMDGPU_MES_SCHED_PIPE;

	memset(&misc_pkt, 0, sizeof(misc_pkt));

	misc_pkt.header.type = MES_API_TYPE_SCHEDULER;
	misc_pkt.header.opcode = MES_SCH_API_MISC;
	misc_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	switch (input->op) {
	case MES_MISC_OP_READ_REG:
		misc_pkt.opcode = MESAPI_MISC__READ_REG;
		misc_pkt.read_reg.buffer_addr = input->read_reg.buffer_addr;
		mes_v12_1_get_rrmt(input->read_reg.reg_offset,
				   GET_INST(GC, input->xcc_id),
				   &misc_pkt.read_reg.rrmt_opt,
				   &misc_pkt.read_reg.reg_offset);
		break;
	case MES_MISC_OP_WRITE_REG:
		misc_pkt.opcode = MESAPI_MISC__WRITE_REG;
		misc_pkt.write_reg.reg_value = input->write_reg.reg_value;
		mes_v12_1_get_rrmt(input->write_reg.reg_offset,
				   GET_INST(GC, input->xcc_id),
				   &misc_pkt.write_reg.rrmt_opt,
				   &misc_pkt.write_reg.reg_offset);
		break;
	case MES_MISC_OP_WRM_REG_WAIT:
		misc_pkt.opcode = MESAPI_MISC__WAIT_REG_MEM;
		misc_pkt.wait_reg_mem.op = WRM_OPERATION__WAIT_REG_MEM;
		misc_pkt.wait_reg_mem.reference = input->wrm_reg.ref;
		misc_pkt.wait_reg_mem.mask = input->wrm_reg.mask;
		misc_pkt.wait_reg_mem.reg_offset2 = 0;
		mes_v12_1_get_rrmt(input->wrm_reg.reg0,
				   GET_INST(GC, input->xcc_id),
				   &misc_pkt.wait_reg_mem.rrmt_opt1,
				   &misc_pkt.wait_reg_mem.reg_offset1);
		break;
	case MES_MISC_OP_WRM_REG_WR_WAIT:
		misc_pkt.opcode = MESAPI_MISC__WAIT_REG_MEM;
		misc_pkt.wait_reg_mem.op = WRM_OPERATION__WR_WAIT_WR_REG;
		misc_pkt.wait_reg_mem.reference = input->wrm_reg.ref;
		misc_pkt.wait_reg_mem.mask = input->wrm_reg.mask;
		mes_v12_1_get_rrmt(input->wrm_reg.reg0,
				   GET_INST(GC, input->xcc_id),
				   &misc_pkt.wait_reg_mem.rrmt_opt1,
				   &misc_pkt.wait_reg_mem.reg_offset1);
		mes_v12_1_get_rrmt(input->wrm_reg.reg1,
				   GET_INST(GC, input->xcc_id),
				   &misc_pkt.wait_reg_mem.rrmt_opt2,
				   &misc_pkt.wait_reg_mem.reg_offset2);
		break;
	case MES_MISC_OP_SET_SHADER_DEBUGGER:
		pipe = AMDGPU_MES_SCHED_PIPE;
		misc_pkt.opcode = MESAPI_MISC__SET_SHADER_DEBUGGER;
		misc_pkt.set_shader_debugger.process_context_addr =
				input->set_shader_debugger.process_context_addr;
		misc_pkt.set_shader_debugger.flags.u32all =
				input->set_shader_debugger.flags.u32all;
		misc_pkt.set_shader_debugger.spi_gdbg_per_vmid_cntl =
				input->set_shader_debugger.spi_gdbg_per_vmid_cntl;
		memcpy(misc_pkt.set_shader_debugger.tcp_watch_cntl,
				input->set_shader_debugger.tcp_watch_cntl,
				sizeof(misc_pkt.set_shader_debugger.tcp_watch_cntl));
		misc_pkt.set_shader_debugger.trap_en = input->set_shader_debugger.trap_en;
		break;
	case MES_MISC_OP_CHANGE_CONFIG:
		misc_pkt.opcode = MESAPI_MISC__CHANGE_CONFIG;
		misc_pkt.change_config.opcode =
			MESAPI_MISC__CHANGE_CONFIG_OPTION_LIMIT_SINGLE_PROCESS;
		misc_pkt.change_config.option.bits.limit_single_process =
			input->change_config.option.limit_single_process;
		break;
	default:
		DRM_ERROR("unsupported misc op (%d) \n", input->op);
		return -EINVAL;
	}

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			input->xcc_id, pipe,
			&misc_pkt, sizeof(misc_pkt),
			offsetof(union MESAPI__MISC, api_status));
}

static int mes_v12_1_set_hw_resources_1(struct amdgpu_mes *mes,
					  int pipe, int xcc_id)
{
	union MESAPI_SET_HW_RESOURCES_1 mes_set_hw_res_1_pkt;
	int master_xcc_id, inst = MES_PIPE_INST(xcc_id, pipe);

	memset(&mes_set_hw_res_1_pkt, 0, sizeof(mes_set_hw_res_1_pkt));

	mes_set_hw_res_1_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_set_hw_res_1_pkt.header.opcode = MES_SCH_API_SET_HW_RSRC_1;
	mes_set_hw_res_1_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
	mes_set_hw_res_1_pkt.mes_kiq_unmap_timeout = 100;

	if (mes->enable_coop_mode && pipe == AMDGPU_MES_SCHED_PIPE) {
		master_xcc_id = mes->master_xcc_ids[inst];
		mes_set_hw_res_1_pkt.mes_coop_mode = 1;
		mes_set_hw_res_1_pkt.coop_sch_shared_mc_addr =
			mes->shared_cmd_buf_gpu_addr[master_xcc_id];
	}

	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
			&mes_set_hw_res_1_pkt, sizeof(mes_set_hw_res_1_pkt),
			offsetof(union MESAPI_SET_HW_RESOURCES_1, api_status));
}

static void mes_v12_1_set_gfx_hqd_mask(union MESAPI_SET_HW_RESOURCES *pkt)
{
	/*
	 * GFX V12 has only one GFX pipe, but 8 queues in it.
	 * GFX pipe 0 queue 0 is being used by Kernel queue.
	 * Set GFX pipe 0 queue 1-7 for MES scheduling
	 * mask = 1111 1110b
	 */
	pkt->gfx_hqd_mask[0] = 0xFE;
}

static int mes_v12_1_set_hw_resources(struct amdgpu_mes *mes,
					int pipe, int xcc_id)
{
	int i;
	struct amdgpu_device *adev = mes->adev;
	union MESAPI_SET_HW_RESOURCES mes_set_hw_res_pkt;

	memset(&mes_set_hw_res_pkt, 0, sizeof(mes_set_hw_res_pkt));

	mes_set_hw_res_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_set_hw_res_pkt.header.opcode = MES_SCH_API_SET_HW_RSRC;
	mes_set_hw_res_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	if (pipe == AMDGPU_MES_SCHED_PIPE) {
		mes_set_hw_res_pkt.vmid_mask_mmhub = mes->vmid_mask_mmhub;
		mes_set_hw_res_pkt.vmid_mask_gfxhub = mes->vmid_mask_gfxhub;
		mes_set_hw_res_pkt.gds_size = adev->gds.gds_size;
		mes_set_hw_res_pkt.paging_vmid = 0;

		for (i = 0; i < MAX_COMPUTE_PIPES; i++)
			mes_set_hw_res_pkt.compute_hqd_mask[i] =
				mes->compute_hqd_mask[i];

		mes_v12_1_set_gfx_hqd_mask(&mes_set_hw_res_pkt);

		for (i = 0; i < MAX_SDMA_PIPES; i++)
			mes_set_hw_res_pkt.sdma_hqd_mask[i] =
				mes->sdma_hqd_mask[i];

		for (i = 0; i < AMD_PRIORITY_NUM_LEVELS; i++)
			mes_set_hw_res_pkt.aggregated_doorbells[i] =
				mes->aggregated_doorbells[i];
	}

	mes_set_hw_res_pkt.g_sch_ctx_gpu_mc_ptr =
		mes->sch_ctx_gpu_addr[pipe];
	mes_set_hw_res_pkt.query_status_fence_gpu_mc_ptr =
		mes->query_status_fence_gpu_addr[pipe];

	for (i = 0; i < 5; i++) {
		mes_set_hw_res_pkt.gc_base[i] =
			adev->reg_offset[GC_HWIP][0][i];
		mes_set_hw_res_pkt.mmhub_base[i] =
				adev->reg_offset[MMHUB_HWIP][0][i];
		mes_set_hw_res_pkt.osssys_base[i] =
		adev->reg_offset[OSSSYS_HWIP][0][i];
	}

	mes_set_hw_res_pkt.disable_reset = 1;
	mes_set_hw_res_pkt.disable_mes_log = 1;
	mes_set_hw_res_pkt.use_different_vmid_compute = 1;
	mes_set_hw_res_pkt.enable_reg_active_poll = 1;
	mes_set_hw_res_pkt.enable_level_process_quantum_check = 1;

	/*
	 * Keep oversubscribe timer for sdma . When we have unmapped doorbell
	 * handling support, other queue will not use the oversubscribe timer.
	 * handling  mode - 0: disabled; 1: basic version; 2: basic+ version
	 */
	mes_set_hw_res_pkt.oversubscription_timer = 50;
	mes_set_hw_res_pkt.unmapped_doorbell_handling = 1;

	if (amdgpu_mes_log_enable) {
		mes_set_hw_res_pkt.enable_mes_event_int_logging = 1;
		mes_set_hw_res_pkt.event_intr_history_gpu_mc_ptr =
			mes->event_log_gpu_addr + MES_PIPE_INST(xcc_id, pipe) * AMDGPU_MES_LOG_BUFFER_SIZE;
	}

	if (adev->enforce_isolation[0] == AMDGPU_ENFORCE_ISOLATION_ENABLE)
		mes_set_hw_res_pkt.limit_single_process = 1;

	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
			&mes_set_hw_res_pkt, sizeof(mes_set_hw_res_pkt),
			offsetof(union MESAPI_SET_HW_RESOURCES, api_status));
}

static void mes_v12_1_init_aggregated_doorbell(struct amdgpu_mes *mes,
						 int xcc_id)
{
	struct amdgpu_device *adev = mes->adev;
	uint32_t data;

	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL1);
	data &= ~(CP_MES_DOORBELL_CONTROL1__DOORBELL_OFFSET_MASK |
		  CP_MES_DOORBELL_CONTROL1__DOORBELL_EN_MASK |
		  CP_MES_DOORBELL_CONTROL1__DOORBELL_HIT_MASK);
	data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_LOW] <<
		CP_MES_DOORBELL_CONTROL1__DOORBELL_OFFSET__SHIFT;
	data |= 1 << CP_MES_DOORBELL_CONTROL1__DOORBELL_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL1, data);

	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL2);
	data &= ~(CP_MES_DOORBELL_CONTROL2__DOORBELL_OFFSET_MASK |
		  CP_MES_DOORBELL_CONTROL2__DOORBELL_EN_MASK |
		  CP_MES_DOORBELL_CONTROL2__DOORBELL_HIT_MASK);
	data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_NORMAL] <<
		CP_MES_DOORBELL_CONTROL2__DOORBELL_OFFSET__SHIFT;
	data |= 1 << CP_MES_DOORBELL_CONTROL2__DOORBELL_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL2, data);

	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL3);
	data &= ~(CP_MES_DOORBELL_CONTROL3__DOORBELL_OFFSET_MASK |
		  CP_MES_DOORBELL_CONTROL3__DOORBELL_EN_MASK |
		  CP_MES_DOORBELL_CONTROL3__DOORBELL_HIT_MASK);
	data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_MEDIUM] <<
		CP_MES_DOORBELL_CONTROL3__DOORBELL_OFFSET__SHIFT;
	data |= 1 << CP_MES_DOORBELL_CONTROL3__DOORBELL_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL3, data);

	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL4);
	data &= ~(CP_MES_DOORBELL_CONTROL4__DOORBELL_OFFSET_MASK |
		  CP_MES_DOORBELL_CONTROL4__DOORBELL_EN_MASK |
		  CP_MES_DOORBELL_CONTROL4__DOORBELL_HIT_MASK);
	data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_HIGH] <<
		CP_MES_DOORBELL_CONTROL4__DOORBELL_OFFSET__SHIFT;
	data |= 1 << CP_MES_DOORBELL_CONTROL4__DOORBELL_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL4, data);

	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL5);
	data &= ~(CP_MES_DOORBELL_CONTROL5__DOORBELL_OFFSET_MASK |
		  CP_MES_DOORBELL_CONTROL5__DOORBELL_EN_MASK |
		  CP_MES_DOORBELL_CONTROL5__DOORBELL_HIT_MASK);
	data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_REALTIME] <<
		CP_MES_DOORBELL_CONTROL5__DOORBELL_OFFSET__SHIFT;
	data |= 1 << CP_MES_DOORBELL_CONTROL5__DOORBELL_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_DOORBELL_CONTROL5, data);

	data = 1 << CP_HQD_GFX_CONTROL__DB_UPDATED_MSG_EN__SHIFT;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_GFX_CONTROL, data);
}


static void mes_v12_1_enable_unmapped_doorbell_handling(
	struct amdgpu_mes *mes, bool enable, int xcc_id)
{
	struct amdgpu_device *adev = mes->adev;
	uint32_t data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_UNMAPPED_DOORBELL);

	/*
	 * The default PROC_LSB settng is 0xc which means doorbell
	 * addr[16:12] gives the doorbell page number. For kfd, each
	 * process will use 2 pages of doorbell, we need to change the
	 * setting to 0xd
	 */
	data &= ~CP_UNMAPPED_DOORBELL__PROC_LSB_MASK;
	data |= 0xd <<  CP_UNMAPPED_DOORBELL__PROC_LSB__SHIFT;

	data |= (enable ? 1 : 0) << CP_UNMAPPED_DOORBELL__ENABLE__SHIFT;

	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_UNMAPPED_DOORBELL, data);
}

#if 0
static int mes_v12_1_reset_legacy_queue(struct amdgpu_mes *mes,
					struct mes_reset_legacy_queue_input *input)
{
	union MESAPI__RESET mes_reset_queue_pkt;
	int pipe;

	memset(&mes_reset_queue_pkt, 0, sizeof(mes_reset_queue_pkt));

	mes_reset_queue_pkt.header.type = MES_API_TYPE_SCHEDULER;
	mes_reset_queue_pkt.header.opcode = MES_SCH_API_RESET;
	mes_reset_queue_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_reset_queue_pkt.queue_type =
		convert_to_mes_queue_type(input->queue_type);

	if (mes_reset_queue_pkt.queue_type == MES_QUEUE_TYPE_GFX) {
		mes_reset_queue_pkt.reset_legacy_gfx = 1;
		mes_reset_queue_pkt.pipe_id_lp = input->pipe_id;
		mes_reset_queue_pkt.queue_id_lp = input->queue_id;
		mes_reset_queue_pkt.mqd_mc_addr_lp = input->mqd_addr;
		mes_reset_queue_pkt.doorbell_offset_lp = input->doorbell_offset;
		mes_reset_queue_pkt.wptr_addr_lp = input->wptr_addr;
		mes_reset_queue_pkt.vmid_id_lp = input->vmid;
	} else {
		mes_reset_queue_pkt.reset_queue_only = 1;
		mes_reset_queue_pkt.doorbell_offset = input->doorbell_offset;
	}

	if (mes->adev->enable_uni_mes)
		pipe = AMDGPU_MES_KIQ_PIPE;
	else
		pipe = AMDGPU_MES_SCHED_PIPE;

	return mes_v12_1_submit_pkt_and_poll_completion(mes,
			input->xcc_id, pipe,
			&mes_reset_queue_pkt, sizeof(mes_reset_queue_pkt),
			offsetof(union MESAPI__RESET, api_status));
}
#endif

static int mes_v12_inv_tlb_convert_hub_id(uint8_t id)
{
	/*
	 * MES doesn't support invalidate gc_hub on slave xcc individually
	 * master xcc will invalidate all gc_hub for the partition
	 */
	if (AMDGPU_IS_GFXHUB(id))
		return 0;
	else if (AMDGPU_IS_MMHUB0(id))
		return 1;
	else if (AMDGPU_IS_MMHUB1(id))
		return 2;
	return -EINVAL;

}

static int mes_v12_1_inv_tlbs_pasid(struct amdgpu_mes *mes,
				    struct mes_inv_tlbs_pasid_input *input)
{
	union MESAPI__INV_TLBS mes_inv_tlbs;
	int xcc_id = input->xcc_id;
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_SCHED_PIPE);
	int ret;

	if (mes->enable_coop_mode)
		xcc_id = mes->master_xcc_ids[inst];

	memset(&mes_inv_tlbs, 0, sizeof(mes_inv_tlbs));

	mes_inv_tlbs.header.type = MES_API_TYPE_SCHEDULER;
	mes_inv_tlbs.header.opcode = MES_SCH_API_INV_TLBS;
	mes_inv_tlbs.header.dwsize = API_FRAME_SIZE_IN_DWORDS;

	mes_inv_tlbs.invalidate_tlbs.inv_sel = 0;
	mes_inv_tlbs.invalidate_tlbs.flush_type = input->flush_type;
	mes_inv_tlbs.invalidate_tlbs.inv_sel_id = input->pasid;

	/*convert amdgpu_mes_hub_id to mes expected hub_id */
	ret = mes_v12_inv_tlb_convert_hub_id(input->hub_id);
	if (ret < 0)
		return -EINVAL;
	mes_inv_tlbs.invalidate_tlbs.hub_id = ret;
	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, AMDGPU_MES_KIQ_PIPE,
			&mes_inv_tlbs, sizeof(mes_inv_tlbs),
			offsetof(union MESAPI__INV_TLBS, api_status));

}

static const struct amdgpu_mes_funcs mes_v12_1_funcs = {
	.add_hw_queue = mes_v12_1_add_hw_queue,
	.remove_hw_queue = mes_v12_1_remove_hw_queue,
	.map_legacy_queue = mes_v12_1_map_legacy_queue,
	.unmap_legacy_queue = mes_v12_1_unmap_legacy_queue,
	.suspend_gang = mes_v12_1_suspend_gang,
	.resume_gang = mes_v12_1_resume_gang,
	.misc_op = mes_v12_1_misc_op,
	.reset_hw_queue = mes_v12_1_reset_hw_queue,
	.invalidate_tlbs_pasid = mes_v12_1_inv_tlbs_pasid,
};

static int mes_v12_1_allocate_ucode_buffer(struct amdgpu_device *adev,
					     enum amdgpu_mes_pipe pipe,
					     int xcc_id)
{
	int r, inst = MES_PIPE_INST(xcc_id, pipe);
	const struct mes_firmware_header_v1_0 *mes_hdr;
	const __le32 *fw_data;
	unsigned fw_size;

	mes_hdr = (const struct mes_firmware_header_v1_0 *)
		adev->mes.fw[pipe]->data;

	fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
		   le32_to_cpu(mes_hdr->mes_ucode_offset_bytes));
	fw_size = le32_to_cpu(mes_hdr->mes_ucode_size_bytes);

	r = amdgpu_bo_create_reserved(adev, fw_size,
				      PAGE_SIZE,
				      AMDGPU_GEM_DOMAIN_VRAM,
				      &adev->mes.ucode_fw_obj[inst],
				      &adev->mes.ucode_fw_gpu_addr[inst],
				      (void **)&adev->mes.ucode_fw_ptr[inst]);
	if (r) {
		dev_err(adev->dev, "(%d) failed to create mes fw bo\n", r);
		return r;
	}

	memcpy(adev->mes.ucode_fw_ptr[inst], fw_data, fw_size);

	amdgpu_bo_kunmap(adev->mes.ucode_fw_obj[inst]);
	amdgpu_bo_unreserve(adev->mes.ucode_fw_obj[inst]);

	return 0;
}

static int mes_v12_1_allocate_ucode_data_buffer(struct amdgpu_device *adev,
						  enum amdgpu_mes_pipe pipe,
						  int xcc_id)
{
	int r, inst = MES_PIPE_INST(xcc_id, pipe);
	const struct mes_firmware_header_v1_0 *mes_hdr;
	const __le32 *fw_data;
	unsigned fw_size;

	mes_hdr = (const struct mes_firmware_header_v1_0 *)
		adev->mes.fw[pipe]->data;

	fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
		   le32_to_cpu(mes_hdr->mes_ucode_data_offset_bytes));
	fw_size = le32_to_cpu(mes_hdr->mes_ucode_data_size_bytes);

	r = amdgpu_bo_create_reserved(adev, fw_size,
				      64 * 1024,
				      AMDGPU_GEM_DOMAIN_VRAM,
				      &adev->mes.data_fw_obj[inst],
				      &adev->mes.data_fw_gpu_addr[inst],
				      (void **)&adev->mes.data_fw_ptr[inst]);
	if (r) {
		dev_err(adev->dev, "(%d) failed to create mes data fw bo\n", r);
		return r;
	}

	memcpy(adev->mes.data_fw_ptr[inst], fw_data, fw_size);

	amdgpu_bo_kunmap(adev->mes.data_fw_obj[inst]);
	amdgpu_bo_unreserve(adev->mes.data_fw_obj[inst]);

	return 0;
}

static void mes_v12_1_free_ucode_buffers(struct amdgpu_device *adev,
					   enum amdgpu_mes_pipe pipe,
					   int xcc_id)
{
	int inst = MES_PIPE_INST(xcc_id, pipe);

	amdgpu_bo_free_kernel(&adev->mes.data_fw_obj[inst],
			      &adev->mes.data_fw_gpu_addr[inst],
			      (void **)&adev->mes.data_fw_ptr[inst]);

	amdgpu_bo_free_kernel(&adev->mes.ucode_fw_obj[inst],
			      &adev->mes.ucode_fw_gpu_addr[inst],
			      (void **)&adev->mes.ucode_fw_ptr[inst]);
}

static void mes_v12_1_enable(struct amdgpu_device *adev,
			       bool enable, int xcc_id)
{
	uint64_t ucode_addr;
	uint32_t pipe, data = 0;

	if (enable) {
		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_CNTL);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE0_RESET, 1);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE1_RESET, 1);
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_CNTL, data);

		mutex_lock(&adev->srbm_mutex);
		for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) {
			soc_v1_0_grbm_select(adev, 3, pipe, 0, 0,
					     GET_INST(GC, xcc_id));

			ucode_addr = adev->mes.uc_start_addr[pipe] >> 2;
			WREG32_SOC15(GC, GET_INST(GC, xcc_id),
				     regCP_MES_PRGRM_CNTR_START,
				     lower_32_bits(ucode_addr));
			WREG32_SOC15(GC, GET_INST(GC, xcc_id),
				     regCP_MES_PRGRM_CNTR_START_HI,
				     upper_32_bits(ucode_addr));
		}
		soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
		mutex_unlock(&adev->srbm_mutex);

		/* unhalt MES and activate pipe0 */
		data = REG_SET_FIELD(0, CP_MES_CNTL, MES_PIPE0_ACTIVE, 1);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE1_ACTIVE, 1);
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_CNTL, data);

		if (amdgpu_emu_mode)
			msleep(500);
		else if (adev->enable_uni_mes)
			udelay(500);
		else
			udelay(50);
	} else {
		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_CNTL);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE0_ACTIVE, 0);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE1_ACTIVE, 0);
		data = REG_SET_FIELD(data, CP_MES_CNTL,
				     MES_INVALIDATE_ICACHE, 1);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE0_RESET, 1);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE1_RESET, 1);
		data = REG_SET_FIELD(data, CP_MES_CNTL, MES_HALT, 1);
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_CNTL, data);
	}
}

static void mes_v12_1_set_ucode_start_addr(struct amdgpu_device *adev,
					     int xcc_id)
{
	uint64_t ucode_addr;
	int pipe;

	mes_v12_1_enable(adev, false, xcc_id);

	mutex_lock(&adev->srbm_mutex);
	for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) {
		/* me=3, queue=0 */
		soc_v1_0_grbm_select(adev, 3, pipe, 0, 0, GET_INST(GC, xcc_id));

		/* set ucode start address */
		ucode_addr = adev->mes.uc_start_addr[pipe] >> 2;
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_PRGRM_CNTR_START,
				lower_32_bits(ucode_addr));
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_PRGRM_CNTR_START_HI,
				upper_32_bits(ucode_addr));

		soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
	}
	mutex_unlock(&adev->srbm_mutex);
}

/* This function is for backdoor MES firmware */
static int mes_v12_1_load_microcode(struct amdgpu_device *adev,
				      enum amdgpu_mes_pipe pipe,
				      bool prime_icache, int xcc_id)
{
	int r, inst = MES_PIPE_INST(xcc_id, pipe);
	uint32_t data;

	mes_v12_1_enable(adev, false, xcc_id);

	if (!adev->mes.fw[pipe])
		return -EINVAL;

	r = mes_v12_1_allocate_ucode_buffer(adev, pipe, xcc_id);
	if (r)
		return r;

	r = mes_v12_1_allocate_ucode_data_buffer(adev, pipe, xcc_id);
	if (r) {
		mes_v12_1_free_ucode_buffers(adev, pipe, xcc_id);
		return r;
	}

	mutex_lock(&adev->srbm_mutex);
	/* me=3, pipe=0, queue=0 */
	soc_v1_0_grbm_select(adev, 3, pipe, 0, 0, GET_INST(GC, xcc_id));

	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_BASE_CNTL, 0);

	/* set ucode fimrware address */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_BASE_LO,
		     lower_32_bits(adev->mes.ucode_fw_gpu_addr[inst]));
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_BASE_HI,
		     upper_32_bits(adev->mes.ucode_fw_gpu_addr[inst]));

	/* set ucode instruction cache boundary to 2M-1 */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_MIBOUND_LO, 0x1FFFFF);

	/* set ucode data firmware address */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_MDBASE_LO,
		     lower_32_bits(adev->mes.data_fw_gpu_addr[inst]));
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_MDBASE_HI,
		     upper_32_bits(adev->mes.data_fw_gpu_addr[inst]));

	/* Set data cache boundary CP_MES_MDBOUND_LO */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_MDBOUND_LO, 0x7FFFF);

	if (prime_icache) {
		/* invalidate ICACHE */
		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_OP_CNTL);
		data = REG_SET_FIELD(data, CP_MES_IC_OP_CNTL, PRIME_ICACHE, 0);
		data = REG_SET_FIELD(data, CP_MES_IC_OP_CNTL, INVALIDATE_CACHE, 1);
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_OP_CNTL, data);

		/* prime the ICACHE. */
		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_OP_CNTL);
		data = REG_SET_FIELD(data, CP_MES_IC_OP_CNTL, PRIME_ICACHE, 1);
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_IC_OP_CNTL, data);
	}

	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
	mutex_unlock(&adev->srbm_mutex);

	return 0;
}

static int mes_v12_1_allocate_eop_buf(struct amdgpu_device *adev,
					enum amdgpu_mes_pipe pipe,
					int xcc_id)
{
	int r, inst = MES_PIPE_INST(xcc_id, pipe);
	u32 *eop;

	r = amdgpu_bo_create_reserved(adev, MES_EOP_SIZE, PAGE_SIZE,
			      AMDGPU_GEM_DOMAIN_GTT,
			      &adev->mes.eop_gpu_obj[inst],
			      &adev->mes.eop_gpu_addr[inst],
			      (void **)&eop);
	if (r) {
		dev_warn(adev->dev, "(%d) create EOP bo failed\n", r);
		return r;
	}

	memset(eop, 0,
	       adev->mes.eop_gpu_obj[inst]->tbo.base.size);

	amdgpu_bo_kunmap(adev->mes.eop_gpu_obj[inst]);
	amdgpu_bo_unreserve(adev->mes.eop_gpu_obj[inst]);

	return 0;
}

static int mes_v12_1_allocate_shared_cmd_buf(struct amdgpu_device *adev,
					     enum amdgpu_mes_pipe pipe,
					     int xcc_id)
{
	int r, inst = MES_PIPE_INST(xcc_id, pipe);

	if (pipe == AMDGPU_MES_KIQ_PIPE)
		return 0;

	r = amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE,
				    AMDGPU_GEM_DOMAIN_VRAM,
				    &adev->mes.shared_cmd_buf_obj[inst],
				    &adev->mes.shared_cmd_buf_gpu_addr[inst],
				    NULL);
	if (r) {
		dev_err(adev->dev,
			"(%d) failed to create shared cmd buf bo\n", r);
		return r;
	}

	return 0;
}

static int mes_v12_1_mqd_init(struct amdgpu_ring *ring)
{
	struct v12_1_mes_mqd *mqd = ring->mqd_ptr;
	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
	uint32_t tmp;

	mqd->header = 0xC0310800;
	mqd->compute_pipelinestat_enable = 0x00000001;
	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
	mqd->compute_misc_reserved = 0x00000007;

	eop_base_addr = ring->eop_gpu_addr >> 8;

	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
	tmp = regCP_HQD_EOP_CONTROL_DEFAULT;
	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
			(order_base_2(MES_EOP_SIZE / 4) - 1));

	mqd->cp_hqd_eop_base_addr_lo = lower_32_bits(eop_base_addr);
	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
	mqd->cp_hqd_eop_control = tmp;

	/* disable the queue if it's active */
	ring->wptr = 0;
	mqd->cp_hqd_pq_rptr = 0;
	mqd->cp_hqd_pq_wptr_lo = 0;
	mqd->cp_hqd_pq_wptr_hi = 0;

	/* set the pointer to the MQD */
	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);

	/* set MQD vmid to 0 */
	tmp = regCP_MQD_CONTROL_DEFAULT;
	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
	mqd->cp_mqd_control = tmp;

	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
	hqd_gpu_addr = ring->gpu_addr >> 8;
	mqd->cp_hqd_pq_base_lo = lower_32_bits(hqd_gpu_addr);
	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);

	/* set the wb address whether it's enabled or not */
	wb_gpu_addr = ring->rptr_gpu_addr;
	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
	mqd->cp_hqd_pq_rptr_report_addr_hi =
		upper_32_bits(wb_gpu_addr) & 0xffff;

	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
	wb_gpu_addr = ring->wptr_gpu_addr;
	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffff8;
	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;

	/* set up the HQD, this is similar to CP_RB0_CNTL */
	tmp = regCP_HQD_PQ_CONTROL_DEFAULT;
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
			    (order_base_2(ring->ring_size / 4) - 1));
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
			    ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 1);
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, NO_UPDATE_RPTR, 1);
	mqd->cp_hqd_pq_control = tmp;

	/* enable doorbell */
	tmp = 0;
	if (ring->use_doorbell) {
		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
				    DOORBELL_OFFSET, ring->doorbell_index);
		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
				    DOORBELL_EN, 1);
		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
				    DOORBELL_SOURCE, 0);
		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
				    DOORBELL_HIT, 0);
	} else {
		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
				    DOORBELL_EN, 0);
	}
	mqd->cp_hqd_pq_doorbell_control = tmp;

	mqd->cp_hqd_vmid = 0;
	/* activate the queue */
	mqd->cp_hqd_active = 1;

	tmp = regCP_HQD_PERSISTENT_STATE_DEFAULT;
	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE,
			    PRELOAD_SIZE, 0x63);
	mqd->cp_hqd_persistent_state = tmp;

	mqd->cp_hqd_ib_control = regCP_HQD_IB_CONTROL_MES_12_1_DEFAULT;
	mqd->cp_hqd_iq_timer = regCP_HQD_IQ_TIMER_DEFAULT;
	mqd->cp_hqd_quantum = regCP_HQD_QUANTUM_DEFAULT;

	/*
	 * Set CP_HQD_GFX_CONTROL.DB_UPDATED_MSG_EN[15] to enable unmapped
	 * doorbell handling. This is a reserved CP internal register can
	 * not be accesss by others
	 */
	mqd->cp_hqd_gfx_control = BIT(15);

	return 0;
}

static void mes_v12_1_queue_init_register(struct amdgpu_ring *ring,
					    int xcc_id)
{
	struct v12_1_mes_mqd *mqd = ring->mqd_ptr;
	struct amdgpu_device *adev = ring->adev;
	uint32_t data = 0;

	mutex_lock(&adev->srbm_mutex);
	soc_v1_0_grbm_select(adev, 3, ring->pipe, 0, 0, GET_INST(GC, xcc_id));

	/* set CP_HQD_VMID.VMID = 0. */
	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_VMID);
	data = REG_SET_FIELD(data, CP_HQD_VMID, VMID, 0);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_VMID, data);

	/* set CP_HQD_PQ_DOORBELL_CONTROL.DOORBELL_EN=0 */
	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL);
	data = REG_SET_FIELD(data, CP_HQD_PQ_DOORBELL_CONTROL,
			     DOORBELL_EN, 0);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, data);

	/* set CP_MQD_BASE_ADDR/HI with the MQD base address */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR, mqd->cp_mqd_base_addr_lo);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR_HI, mqd->cp_mqd_base_addr_hi);

	/* set CP_MQD_CONTROL.VMID=0 */
	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL);
	data = REG_SET_FIELD(data, CP_MQD_CONTROL, VMID, 0);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL, 0);

	/* set CP_HQD_PQ_BASE/HI with the ring buffer base address */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE, mqd->cp_hqd_pq_base_lo);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE_HI, mqd->cp_hqd_pq_base_hi);

	/* set CP_HQD_PQ_RPTR_REPORT_ADDR/HI */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR,
		     mqd->cp_hqd_pq_rptr_report_addr_lo);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
		     mqd->cp_hqd_pq_rptr_report_addr_hi);

	/* set CP_HQD_PQ_CONTROL */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_CONTROL, mqd->cp_hqd_pq_control);

	/* set CP_HQD_PQ_WPTR_POLL_ADDR/HI */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR,
		     mqd->cp_hqd_pq_wptr_poll_addr_lo);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
		     mqd->cp_hqd_pq_wptr_poll_addr_hi);

	/* set CP_HQD_PQ_DOORBELL_CONTROL */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL,
		     mqd->cp_hqd_pq_doorbell_control);

	/* set CP_HQD_PERSISTENT_STATE.PRELOAD_SIZE=0x53 */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE, mqd->cp_hqd_persistent_state);

	/* set CP_HQD_ACTIVE.ACTIVE=1 */
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE, mqd->cp_hqd_active);

	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
	mutex_unlock(&adev->srbm_mutex);
}

static int mes_v12_1_kiq_enable_queue(struct amdgpu_device *adev, int xcc_id)
{
	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[xcc_id].ring;
	int r, inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_SCHED_PIPE);

	if (!kiq->pmf || !kiq->pmf->kiq_map_queues)
		return -EINVAL;

	r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size);
	if (r) {
		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
		return r;
	}

	kiq->pmf->kiq_map_queues(kiq_ring, &adev->mes.ring[inst]);

	r = amdgpu_ring_test_ring(kiq_ring);
	if (r) {
		DRM_ERROR("kfq enable failed\n");
		kiq_ring->sched.ready = false;
	}
	return r;
}

static int mes_v12_1_queue_init(struct amdgpu_device *adev,
				  enum amdgpu_mes_pipe pipe,
				  int xcc_id)
{
	struct amdgpu_ring *ring;
	int r;

	if (!adev->enable_uni_mes && pipe == AMDGPU_MES_KIQ_PIPE)
		ring = &adev->gfx.kiq[xcc_id].ring;
	else
		ring = &adev->mes.ring[MES_PIPE_INST(xcc_id, pipe)];

	if ((adev->enable_uni_mes || pipe == AMDGPU_MES_SCHED_PIPE) &&
	    (amdgpu_in_reset(adev) || adev->in_suspend)) {
		*(ring->wptr_cpu_addr) = 0;
		*(ring->rptr_cpu_addr) = 0;
		amdgpu_ring_clear_ring(ring);
	}

	r = mes_v12_1_mqd_init(ring);
	if (r)
		return r;

	if (pipe == AMDGPU_MES_SCHED_PIPE) {
		if (adev->enable_uni_mes)
			r = amdgpu_mes_map_legacy_queue(adev, ring, xcc_id);
		else
			r = mes_v12_1_kiq_enable_queue(adev, xcc_id);
		if (r)
			return r;
	} else {
		mes_v12_1_queue_init_register(ring, xcc_id);
	}

	/* get MES scheduler/KIQ versions */
	mutex_lock(&adev->srbm_mutex);
	soc_v1_0_grbm_select(adev, 3, pipe, 0, 0, GET_INST(GC, xcc_id));

	if (pipe == AMDGPU_MES_SCHED_PIPE)
		adev->mes.sched_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);
	else if (pipe == AMDGPU_MES_KIQ_PIPE && adev->enable_mes_kiq)
		adev->mes.kiq_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);

	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
	mutex_unlock(&adev->srbm_mutex);

	return 0;
}

static int mes_v12_1_ring_init(struct amdgpu_device *adev,
				 int xcc_id, int pipe)
{
	struct amdgpu_ring *ring;
	int inst = MES_PIPE_INST(xcc_id, pipe);

	ring = &adev->mes.ring[inst];

	ring->funcs = &mes_v12_1_ring_funcs;

	ring->me = 3;
	ring->pipe = pipe;
	ring->queue = 0;
	ring->xcc_id = xcc_id;
	ring->vm_hub = AMDGPU_GFXHUB(xcc_id);

	ring->ring_obj = NULL;
	ring->use_doorbell = true;
	ring->eop_gpu_addr = adev->mes.eop_gpu_addr[inst];
	ring->no_scheduler = true;
	snprintf(ring->name, sizeof(ring->name), "mes_%hhu.%hhu.%hhu.%hhu",
		 (unsigned char)xcc_id, (unsigned char)ring->me,
		 (unsigned char)ring->pipe, (unsigned char)ring->queue);

	if (pipe == AMDGPU_MES_SCHED_PIPE)
		ring->doorbell_index =
			(adev->doorbell_index.mes_ring0 +
			 xcc_id * adev->doorbell_index.xcc_doorbell_range)
			<< 1;
	else
		ring->doorbell_index =
			(adev->doorbell_index.mes_ring1 +
			 xcc_id * adev->doorbell_index.xcc_doorbell_range)
			<< 1;

	return amdgpu_ring_init(adev, ring, 1024, NULL, 0,
				AMDGPU_RING_PRIO_DEFAULT, NULL);
}

static int mes_v12_1_kiq_ring_init(struct amdgpu_device *adev, int xcc_id)
{
	struct amdgpu_ring *ring;
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_KIQ_PIPE);

	spin_lock_init(&adev->gfx.kiq[xcc_id].ring_lock);

	ring = &adev->gfx.kiq[xcc_id].ring;

	ring->me = 3;
	ring->pipe = 1;
	ring->queue = 0;
	ring->xcc_id = xcc_id;
	ring->vm_hub = AMDGPU_GFXHUB(xcc_id);

	ring->adev = NULL;
	ring->ring_obj = NULL;
	ring->use_doorbell = true;
	ring->eop_gpu_addr = adev->mes.eop_gpu_addr[inst];
	ring->no_scheduler = true;
	ring->doorbell_index =
		(adev->doorbell_index.mes_ring1 +
		 xcc_id * adev->doorbell_index.xcc_doorbell_range)
		<< 1;

	snprintf(ring->name, sizeof(ring->name), "mes_kiq_%hhu.%hhu.%hhu.%hhu",
		 (unsigned char)xcc_id, (unsigned char)ring->me,
		 (unsigned char)ring->pipe, (unsigned char)ring->queue);

	return amdgpu_ring_init(adev, ring, 1024, NULL, 0,
				AMDGPU_RING_PRIO_DEFAULT, NULL);
}

static int mes_v12_1_mqd_sw_init(struct amdgpu_device *adev,
				   enum amdgpu_mes_pipe pipe,
				   int xcc_id)
{
	int r, mqd_size = sizeof(struct v12_1_mes_mqd);
	struct amdgpu_ring *ring;
	int inst = MES_PIPE_INST(xcc_id, pipe);

	if (!adev->enable_uni_mes && pipe == AMDGPU_MES_KIQ_PIPE)
		ring = &adev->gfx.kiq[xcc_id].ring;
	else
		ring = &adev->mes.ring[inst];

	if (ring->mqd_obj)
		return 0;

	r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
				    AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
				    &ring->mqd_gpu_addr, &ring->mqd_ptr);
	if (r) {
		dev_warn(adev->dev, "failed to create ring mqd bo (%d)", r);
		return r;
	}

	memset(ring->mqd_ptr, 0, mqd_size);

	/* prepare MQD backup */
	adev->mes.mqd_backup[inst] = kmalloc(mqd_size, GFP_KERNEL);
	if (!adev->mes.mqd_backup[inst])
		dev_warn(adev->dev,
			 "no memory to create MQD backup for ring %s\n",
			 ring->name);

	return 0;
}

static int mes_v12_1_sw_init(struct amdgpu_ip_block *ip_block)
{
	struct amdgpu_device *adev = ip_block->adev;
	int pipe, r, xcc_id, num_xcc = NUM_XCC(adev->gfx.xcc_mask);

	adev->mes.funcs = &mes_v12_1_funcs;
	adev->mes.kiq_hw_init = &mes_v12_1_kiq_hw_init;
	adev->mes.kiq_hw_fini = &mes_v12_1_kiq_hw_fini;
	adev->mes.enable_legacy_queue_map = true;

	adev->mes.event_log_size =
		adev->enable_uni_mes ? (AMDGPU_MAX_MES_PIPES * AMDGPU_MES_LOG_BUFFER_SIZE * num_xcc) : AMDGPU_MES_LOG_BUFFER_SIZE;

	r = amdgpu_mes_init(adev);
	if (r)
		return r;

	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
		for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) {
			r = mes_v12_1_allocate_eop_buf(adev, pipe, xcc_id);
			if (r)
				return r;

			r = mes_v12_1_mqd_sw_init(adev, pipe, xcc_id);
			if (r)
				return r;

			if (!adev->enable_uni_mes && pipe ==
			    AMDGPU_MES_KIQ_PIPE)
				r = mes_v12_1_kiq_ring_init(adev, xcc_id);
			else
				r = mes_v12_1_ring_init(adev, xcc_id, pipe);
			if (r)
				return r;

			if (adev->enable_uni_mes && num_xcc > 1) {
				r = mes_v12_1_allocate_shared_cmd_buf(adev,
							      pipe, xcc_id);
				if (r)
					return r;
			}
		}
	}

	return 0;
}

static int mes_v12_1_sw_fini(struct amdgpu_ip_block *ip_block)
{
	struct amdgpu_device *adev = ip_block->adev;
	int pipe, inst, xcc_id, num_xcc = NUM_XCC(adev->gfx.xcc_mask);

	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
		for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) {
			inst = MES_PIPE_INST(xcc_id, pipe);

			amdgpu_bo_free_kernel(&adev->mes.shared_cmd_buf_obj[inst],
					      &adev->mes.shared_cmd_buf_gpu_addr[inst],
					      NULL);

			kfree(adev->mes.mqd_backup[inst]);

			amdgpu_bo_free_kernel(&adev->mes.eop_gpu_obj[inst],
					      &adev->mes.eop_gpu_addr[inst],
					      NULL);

			if (adev->enable_uni_mes || pipe == AMDGPU_MES_SCHED_PIPE) {
				amdgpu_bo_free_kernel(&adev->mes.ring[inst].mqd_obj,
						      &adev->mes.ring[inst].mqd_gpu_addr,
						      &adev->mes.ring[inst].mqd_ptr);
				amdgpu_ring_fini(&adev->mes.ring[inst]);
			}
		}
	}

	for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++)
		amdgpu_ucode_release(&adev->mes.fw[pipe]);

	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
		if (!adev->enable_uni_mes) {
			amdgpu_bo_free_kernel(&adev->gfx.kiq[xcc_id].ring.mqd_obj,
				      &adev->gfx.kiq[xcc_id].ring.mqd_gpu_addr,
				      &adev->gfx.kiq[xcc_id].ring.mqd_ptr);
			amdgpu_ring_fini(&adev->gfx.kiq[xcc_id].ring);
		}

		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
			mes_v12_1_free_ucode_buffers(adev,
				       AMDGPU_MES_KIQ_PIPE, xcc_id);
			mes_v12_1_free_ucode_buffers(adev,
				       AMDGPU_MES_SCHED_PIPE, xcc_id);
		}
	}

	amdgpu_mes_fini(adev);
	return 0;
}

static void mes_v12_1_kiq_dequeue_sched(struct amdgpu_device *adev,
					  int xcc_id)
{
	uint32_t data;
	int i;

	mutex_lock(&adev->srbm_mutex);
	soc_v1_0_grbm_select(adev, 3, AMDGPU_MES_SCHED_PIPE, 0, 0,
			     GET_INST(GC, xcc_id));

	/* disable the queue if it's active */
	if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1) {
		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1);
		for (i = 0; i < adev->usec_timeout; i++) {
			if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1))
				break;
			udelay(1);
		}
	}
	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL);
	data = REG_SET_FIELD(data, CP_HQD_PQ_DOORBELL_CONTROL,
				DOORBELL_EN, 0);
	data = REG_SET_FIELD(data, CP_HQD_PQ_DOORBELL_CONTROL,
				DOORBELL_HIT, 1);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, data);

	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 0);

	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO, 0);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI, 0);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR, 0);

	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
	mutex_unlock(&adev->srbm_mutex);

	adev->mes.ring[MES_PIPE_INST(xcc_id, 0)].sched.ready = false;
}

static void mes_v12_1_kiq_setting(struct amdgpu_ring *ring, int xcc_id)
{
	uint32_t tmp;
	struct amdgpu_device *adev = ring->adev;

	/* tell RLC which is KIQ queue */
	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS);
	tmp &= 0xffffff00;
	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp);
	tmp |= 0x80;
	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp);
}

static int mes_v12_1_kiq_hw_init(struct amdgpu_device *adev, uint32_t xcc_id)
{
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_KIQ_PIPE);
	int r = 0;
	struct amdgpu_ip_block *ip_block;

	if (adev->enable_uni_mes)
		mes_v12_1_kiq_setting(&adev->mes.ring[inst], xcc_id);
	else
		mes_v12_1_kiq_setting(&adev->gfx.kiq[xcc_id].ring, xcc_id);

	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {

		r = mes_v12_1_load_microcode(adev, AMDGPU_MES_SCHED_PIPE,
					       false, xcc_id);
		if (r) {
			DRM_ERROR("failed to load MES fw, r=%d\n", r);
			return r;
		}

		r = mes_v12_1_load_microcode(adev, AMDGPU_MES_KIQ_PIPE,
					       true, xcc_id);
		if (r) {
			DRM_ERROR("failed to load MES kiq fw, r=%d\n", r);
			return r;
		}

		mes_v12_1_set_ucode_start_addr(adev, xcc_id);

	} else if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
		mes_v12_1_set_ucode_start_addr(adev, xcc_id);

	mes_v12_1_enable(adev, true, xcc_id);

	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_MES);
	if (unlikely(!ip_block)) {
		dev_err(adev->dev, "Failed to get MES handle\n");
		return -EINVAL;
	}

	r = mes_v12_1_queue_init(adev, AMDGPU_MES_KIQ_PIPE, xcc_id);
	if (r)
		goto failure;

	if (adev->enable_uni_mes) {
		r = mes_v12_1_set_hw_resources(&adev->mes,
						 AMDGPU_MES_KIQ_PIPE, xcc_id);
		if (r)
			goto failure;

		mes_v12_1_set_hw_resources_1(&adev->mes,
					       AMDGPU_MES_KIQ_PIPE, xcc_id);
	}

	if (adev->mes.enable_legacy_queue_map) {
		r = mes_v12_1_xcc_hw_init(ip_block, xcc_id);
		if (r)
			goto failure;
	}

	return r;

failure:
	mes_v12_1_hw_fini(ip_block);
	return r;
}

static int mes_v12_1_kiq_hw_fini(struct amdgpu_device *adev, uint32_t xcc_id)
{
	int inst = MES_PIPE_INST(xcc_id, AMDGPU_MES_SCHED_PIPE);

	if (adev->mes.ring[inst].sched.ready) {
		if (adev->enable_uni_mes)
			amdgpu_mes_unmap_legacy_queue(adev,
				      &adev->mes.ring[inst],
				      RESET_QUEUES, 0, 0, xcc_id);
		else
			mes_v12_1_kiq_dequeue_sched(adev, xcc_id);

		adev->mes.ring[inst].sched.ready = false;
	}

	mes_v12_1_enable(adev, false, xcc_id);

	return 0;
}

static int mes_v12_1_setup_coop_mode(struct amdgpu_device *adev, int xcc_id)
{
	u32 num_xcc_per_xcp, num_xcc = NUM_XCC(adev->gfx.xcc_mask);
	int r = 0;

	if (num_xcc == 1)
		return r;

	if (adev->gfx.funcs &&
	    adev->gfx.funcs->get_xccs_per_xcp)
		num_xcc_per_xcp = adev->gfx.funcs->get_xccs_per_xcp(adev);
	else
		return -EINVAL;

	switch (adev->xcp_mgr->mode) {
	case AMDGPU_SPX_PARTITION_MODE:
		adev->mes.enable_coop_mode = 1;
		adev->mes.master_xcc_ids[xcc_id] = 0;
		break;
	case AMDGPU_DPX_PARTITION_MODE:
		adev->mes.enable_coop_mode = 1;
		adev->mes.master_xcc_ids[xcc_id] =
			(xcc_id/num_xcc_per_xcp) * (num_xcc / 2);
		break;
	case AMDGPU_QPX_PARTITION_MODE:
		adev->mes.enable_coop_mode = 1;
		adev->mes.master_xcc_ids[xcc_id] =
			(xcc_id/num_xcc_per_xcp) * (num_xcc / 4);
		break;
	case AMDGPU_CPX_PARTITION_MODE:
		adev->mes.enable_coop_mode = 0;
		break;
	default:
		r = -EINVAL;
		break;
	}
	return r;
}

static int mes_v12_1_xcc_hw_init(struct amdgpu_ip_block *ip_block, int xcc_id)
{
	int r;
	struct amdgpu_device *adev = ip_block->adev;

	if (adev->mes.ring[MES_PIPE_INST(xcc_id, 0)].sched.ready)
		goto out;

	if (!adev->enable_mes_kiq) {
		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
			r = mes_v12_1_load_microcode(adev,
				       AMDGPU_MES_SCHED_PIPE, true, xcc_id);
			if (r) {
				DRM_ERROR("failed to MES fw, r=%d\n", r);
				return r;
			}

			mes_v12_1_set_ucode_start_addr(adev, xcc_id);

		} else if (adev->firmware.load_type ==
			   AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {

			mes_v12_1_set_ucode_start_addr(adev, xcc_id);
		}

		mes_v12_1_enable(adev, true, xcc_id);
	}

	/* Enable the MES to handle doorbell ring on unmapped queue */
	mes_v12_1_enable_unmapped_doorbell_handling(&adev->mes, true, xcc_id);

	r = mes_v12_1_queue_init(adev, AMDGPU_MES_SCHED_PIPE, xcc_id);
	if (r)
		goto failure;

	r = mes_v12_1_set_hw_resources(&adev->mes,
					 AMDGPU_MES_SCHED_PIPE, xcc_id);
	if (r)
		goto failure;

	if (adev->enable_uni_mes) {
		r = mes_v12_1_setup_coop_mode(adev, xcc_id);
		if (r)
			goto failure;
		mes_v12_1_set_hw_resources_1(&adev->mes,
					       AMDGPU_MES_SCHED_PIPE, xcc_id);
	}
	mes_v12_1_init_aggregated_doorbell(&adev->mes, xcc_id);

	r = mes_v12_1_query_sched_status(&adev->mes,
					   AMDGPU_MES_SCHED_PIPE, xcc_id);
	if (r) {
		DRM_ERROR("MES is busy\n");
		goto failure;
	}

out:
	/*
	 * Disable KIQ ring usage from the driver once MES is enabled.
	 * MES uses KIQ ring exclusively so driver cannot access KIQ ring
	 * with MES enabled.
	 */
	adev->gfx.kiq[xcc_id].ring.sched.ready = false;
	adev->mes.ring[MES_PIPE_INST(xcc_id, 0)].sched.ready = true;

	return 0;

failure:
	mes_v12_1_hw_fini(ip_block);
	return r;
}

static int mes_v12_1_hw_init(struct amdgpu_ip_block *ip_block)
{
	struct amdgpu_device *adev = ip_block->adev;
	int r, xcc_id, num_xcc = NUM_XCC(adev->gfx.xcc_mask);

	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
		r = mes_v12_1_xcc_hw_init(ip_block, xcc_id);
		if (r)
			return r;
	}

	return 0;
}

static int mes_v12_1_hw_fini(struct amdgpu_ip_block *ip_block)
{
	return 0;
}

static int mes_v12_1_suspend(struct amdgpu_ip_block *ip_block)
{
	int r;

	r = amdgpu_mes_suspend(ip_block->adev);
	if (r)
		return r;

	return mes_v12_1_hw_fini(ip_block);
}

static int mes_v12_1_resume(struct amdgpu_ip_block *ip_block)
{
	int r;

	r = mes_v12_1_hw_init(ip_block);
	if (r)
		return r;

	return amdgpu_mes_resume(ip_block->adev);
}

static int mes_v12_1_early_init(struct amdgpu_ip_block *ip_block)
{
	struct amdgpu_device *adev = ip_block->adev;
	int pipe, r;

	for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) {
		r = amdgpu_mes_init_microcode(adev, pipe);
		if (r)
			return r;
	}

	return 0;
}

static int mes_v12_1_late_init(struct amdgpu_ip_block *ip_block)
{
	struct amdgpu_device *adev = ip_block->adev;
	int xcc_id, num_xcc = NUM_XCC(adev->gfx.xcc_mask);

	/* TODO: remove it if issue fixed. */
	if (adev->mes.enable_coop_mode)
		return 0;

	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
		/* for COOP mode, only test master xcc. */
		if (adev->mes.enable_coop_mode &&
		    adev->mes.master_xcc_ids[xcc_id] != xcc_id)
			continue;

		mes_v12_1_self_test(adev, xcc_id);
	}

	return 0;
}

static const struct amd_ip_funcs mes_v12_1_ip_funcs = {
	.name = "mes_v12_1",
	.early_init = mes_v12_1_early_init,
	.late_init = mes_v12_1_late_init,
	.sw_init = mes_v12_1_sw_init,
	.sw_fini = mes_v12_1_sw_fini,
	.hw_init = mes_v12_1_hw_init,
	.hw_fini = mes_v12_1_hw_fini,
	.suspend = mes_v12_1_suspend,
	.resume = mes_v12_1_resume,
};

const struct amdgpu_ip_block_version mes_v12_1_ip_block = {
	.type = AMD_IP_BLOCK_TYPE_MES,
	.major = 12,
	.minor = 1,
	.rev = 0,
	.funcs = &mes_v12_1_ip_funcs,
};

static int mes_v12_1_alloc_test_buf(struct amdgpu_device *adev,
				    struct amdgpu_bo **bo, uint64_t *addr,
				    void **ptr, int size)
{
	amdgpu_bo_create_kernel(adev, size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
				bo, addr, ptr);
	if (!*bo) {
		dev_err(adev->dev, "failed to allocate test buffer bo\n");
		return -ENOMEM;
	}
	memset(*ptr, 0, size);
	return 0;
}

static int mes_v12_1_map_test_bo(struct amdgpu_device *adev,
				 struct amdgpu_bo *bo, struct amdgpu_vm *vm,
				 struct amdgpu_bo_va **bo_va, u64 va, int size)
{
	struct amdgpu_sync sync;
	int r;

	r = amdgpu_map_static_csa(adev, vm, bo, bo_va, va, size);
	if (r)
		return r;

	amdgpu_sync_create(&sync);

	r = amdgpu_vm_bo_update(adev, *bo_va, false);
	if (r) {
		dev_err(adev->dev, "failed to do vm_bo_update on meta data\n");
		goto error;
	}
	amdgpu_sync_fence(&sync, (*bo_va)->last_pt_update, GFP_KERNEL);

	r = amdgpu_vm_update_pdes(adev, vm, false);
	if (r) {
		dev_err(adev->dev, "failed to update pdes on meta data\n");
		goto error;
	}
	amdgpu_sync_fence(&sync, vm->last_update, GFP_KERNEL);
	amdgpu_sync_wait(&sync, false);

error:
	amdgpu_sync_free(&sync);
	return 0;
}

static int mes_v12_1_test_ring(struct amdgpu_device *adev, int xcc_id,
			       u32 *queue_ptr, u64 fence_gpu_addr,
			       void *fence_cpu_ptr, void *wptr_cpu_addr,
			       u64 doorbell_idx, int queue_type)
{
	volatile uint32_t *cpu_ptr = fence_cpu_ptr;
	int num_xcc = NUM_XCC(adev->gfx.xcc_mask);
	int sdma_ring_align = 0x10, compute_ring_align = 0x100;
	uint32_t tmp, xcc_offset;
	int r = 0, i, j, wptr = 0;

	if (queue_type == AMDGPU_RING_TYPE_COMPUTE) {
		if (!adev->mes.enable_coop_mode) {
			WREG32_SOC15(GC, GET_INST(GC, xcc_id),
				     regSCRATCH_REG0, 0xCAFEDEAD);
		} else {
			for (i = 0; i < num_xcc; i++) {
				if (adev->mes.master_xcc_ids[i] == xcc_id)
					WREG32_SOC15(GC, GET_INST(GC, i),
					       regSCRATCH_REG0, 0xCAFEDEAD);
			}
		}

		xcc_offset = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
		queue_ptr[wptr++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
		queue_ptr[wptr++] = xcc_offset - PACKET3_SET_UCONFIG_REG_START;
		queue_ptr[wptr++] = 0xDEADBEEF;

		for (i = wptr; i < compute_ring_align; i++)
			queue_ptr[wptr++] = PACKET3(PACKET3_NOP, 0x3FFF);

	}  else if (queue_type == AMDGPU_RING_TYPE_SDMA) {
		*cpu_ptr = 0xCAFEDEAD;

		queue_ptr[wptr++] = SDMA_PKT_COPY_LINEAR_HEADER_OP(SDMA_OP_WRITE) |
			SDMA_PKT_COPY_LINEAR_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
		queue_ptr[wptr++] = lower_32_bits(fence_gpu_addr);
		queue_ptr[wptr++] = upper_32_bits(fence_gpu_addr);
		queue_ptr[wptr++] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
		queue_ptr[wptr++] = 0xDEADBEEF;

		for (i = wptr; i < sdma_ring_align; i++)
			queue_ptr[wptr++] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);

		wptr <<= 2;
	}

	atomic64_set((atomic64_t *)wptr_cpu_addr, wptr);
	WDOORBELL64(doorbell_idx, wptr);

	for (i = 0; i < adev->usec_timeout; i++) {
		if (queue_type == AMDGPU_RING_TYPE_SDMA) {
			tmp = le32_to_cpu(*cpu_ptr);
		} else {
			if (!adev->mes.enable_coop_mode) {
				tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
						   regSCRATCH_REG0);
			} else {
				for (j = 0; j < num_xcc; j++) {
					if (xcc_id != adev->mes.master_xcc_ids[j])
						continue;

					tmp = RREG32_SOC15(GC, GET_INST(GC, j),
							   regSCRATCH_REG0);
					if (tmp != 0xDEADBEEF)
						break;
				}
			}
		}

		if (tmp == 0xDEADBEEF)
			break;

		if (amdgpu_emu_mode == 1)
			msleep(1);
		else
			udelay(1);
	}

	if (i >= adev->usec_timeout) {
		dev_err(adev->dev, "xcc%d: mes self test (%s) failed\n", xcc_id,
		      queue_type == AMDGPU_RING_TYPE_SDMA ? "sdma" : "compute");

		while (halt_if_hws_hang)
			schedule();

		r = -ETIMEDOUT;
	} else {
		dev_info(adev->dev, "xcc%d: mes self test (%s) pass\n", xcc_id,
		      queue_type == AMDGPU_RING_TYPE_SDMA ? "sdma" : "compute");
	}

	return r;
}

#define USER_CTX_SIZE (PAGE_SIZE * 2)
#define USER_CTX_VA AMDGPU_VA_RESERVED_BOTTOM
#define RING_OFFSET(addr) ((addr))
#define EOP_OFFSET(addr)  ((addr) + PAGE_SIZE)
#define WPTR_OFFSET(addr) ((addr) + USER_CTX_SIZE - sizeof(u64))
#define RPTR_OFFSET(addr) ((addr) + USER_CTX_SIZE - sizeof(u64) * 2)
#define FENCE_OFFSET(addr) ((addr) + USER_CTX_SIZE - sizeof(u64) * 3)

static int mes_v12_1_test_queue(struct amdgpu_device *adev, int xcc_id,
		     int pasid, struct amdgpu_vm *vm, u64 meta_gpu_addr,
		     u64 queue_gpu_addr, void *ctx_ptr, int queue_type)
{
	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];
	struct amdgpu_mqd *mqd_mgr = &adev->mqds[queue_type];
	struct amdgpu_mqd_prop mqd_prop = {0};
	struct mes_add_queue_input add_queue = {0};
	struct mes_remove_queue_input remove_queue = {0};
	struct amdgpu_bo *mqd_bo = NULL;
	int num_xcc = NUM_XCC(adev->gfx.xcc_mask);
	int i, r, off, mqd_size, mqd_count = 1;
	void *mqd_ptr = NULL;
	u64 mqd_gpu_addr, doorbell_idx;

	/* extra one page size padding for mes fw */
	mqd_size = mqd_mgr->mqd_size + PAGE_SIZE;

	if (queue_type == AMDGPU_RING_TYPE_SDMA) {
		doorbell_idx = adev->mes.db_start_dw_offset +	\
			adev->doorbell_index.sdma_engine[0];
	} else {
		doorbell_idx = adev->mes.db_start_dw_offset + \
			adev->doorbell_index.userqueue_start;
	}

	if (adev->mes.enable_coop_mode &&
	    queue_type == AMDGPU_RING_TYPE_COMPUTE) {
		for (i = 0, mqd_count = 0; i < num_xcc; i++) {
			if (adev->mes.master_xcc_ids[i] == xcc_id)
				mqd_count++;
		}
		mqd_size *= mqd_count;
	}

	r = mes_v12_1_alloc_test_buf(adev, &mqd_bo, &mqd_gpu_addr,
				     &mqd_ptr, mqd_size * mqd_count);
	if (r < 0)
		return r;

	mqd_prop.mqd_gpu_addr = mqd_gpu_addr;
	mqd_prop.hqd_base_gpu_addr = RING_OFFSET(USER_CTX_VA);
	mqd_prop.eop_gpu_addr = EOP_OFFSET(USER_CTX_VA);
	mqd_prop.wptr_gpu_addr = WPTR_OFFSET(USER_CTX_VA);
	mqd_prop.rptr_gpu_addr = RPTR_OFFSET(USER_CTX_VA);
	mqd_prop.doorbell_index = doorbell_idx;
	mqd_prop.queue_size = PAGE_SIZE;
	mqd_prop.mqd_stride_size = mqd_size;
	mqd_prop.use_doorbell = true;
	mqd_prop.hqd_active = false;

	mqd_mgr->init_mqd(adev, mqd_ptr, &mqd_prop);
	if (mqd_count > 1) {
		for (i = 1; i < mqd_count; i++) {
			off = mqd_size * i;
			mqd_prop.mqd_gpu_addr = mqd_gpu_addr + off;
			mqd_mgr->init_mqd(adev, (char *)mqd_ptr + off,
					  &mqd_prop);
		}
	}

	add_queue.xcc_id = xcc_id;
	add_queue.process_id = pasid;
	add_queue.page_table_base_addr = adev->vm_manager.vram_base_offset +
		amdgpu_bo_gpu_offset(vm->root.bo) - adev->gmc.vram_start;
	add_queue.process_va_start = 0;
	add_queue.process_va_end = adev->vm_manager.max_pfn - 1;
	add_queue.process_context_addr = meta_gpu_addr;
	add_queue.gang_context_addr = meta_gpu_addr + AMDGPU_MES_PROC_CTX_SIZE;
	add_queue.doorbell_offset = doorbell_idx;
	add_queue.mqd_addr = mqd_gpu_addr;
	add_queue.wptr_addr = mqd_prop.wptr_gpu_addr;
	add_queue.wptr_mc_addr = WPTR_OFFSET(queue_gpu_addr);
	add_queue.queue_type = queue_type;
	add_queue.vm_cntx_cntl = hub->vm_cntx_cntl;

	r = mes_v12_1_add_hw_queue(&adev->mes, &add_queue);
	if (r)
		goto error;

	mes_v12_1_test_ring(adev, xcc_id, (u32 *)RING_OFFSET((char *)ctx_ptr),
			    FENCE_OFFSET(USER_CTX_VA),
			    FENCE_OFFSET((char *)ctx_ptr),
			    WPTR_OFFSET((char *)ctx_ptr),
			    doorbell_idx, queue_type);

	remove_queue.xcc_id = xcc_id;
	remove_queue.doorbell_offset = doorbell_idx;
	remove_queue.gang_context_addr = add_queue.gang_context_addr;
	r = mes_v12_1_remove_hw_queue(&adev->mes, &remove_queue);

error:
	amdgpu_bo_free_kernel(&mqd_bo, &mqd_gpu_addr, &mqd_ptr);
	return r;
}

static int mes_v12_1_self_test(struct amdgpu_device *adev, int xcc_id)
{
	int queue_types[] = { AMDGPU_RING_TYPE_COMPUTE,
		              /* AMDGPU_RING_TYPE_SDMA */ };
	struct amdgpu_bo_va *bo_va = NULL;
	struct amdgpu_vm *vm = NULL;
	struct amdgpu_bo *meta_bo = NULL, *ctx_bo = NULL;
	void *meta_ptr = NULL, *ctx_ptr = NULL;
	u64 meta_gpu_addr, ctx_gpu_addr;
	int size, i, r, pasid;

	pasid = amdgpu_pasid_alloc(16);
	if (pasid < 0)
		pasid = 0;

	size = AMDGPU_MES_PROC_CTX_SIZE + AMDGPU_MES_GANG_CTX_SIZE;
	r = mes_v12_1_alloc_test_buf(adev, &meta_bo, &meta_gpu_addr,
				     &meta_ptr, size);
	if (r < 0)
		goto err2;

	r = mes_v12_1_alloc_test_buf(adev, &ctx_bo, &ctx_gpu_addr,
				     &ctx_ptr, USER_CTX_SIZE);
	if (r < 0)
		goto err2;

	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
	if (!vm) {
		r = -ENOMEM;
		goto err2;
	}

	r = amdgpu_vm_init(adev, vm, -1, pasid);
	if (r)
		goto err1;

	r = mes_v12_1_map_test_bo(adev, ctx_bo, vm, &bo_va,
				  USER_CTX_VA, USER_CTX_SIZE);
	if (r)
		goto err0;

	for (i = 0; i < ARRAY_SIZE(queue_types); i++) {
		memset(ctx_ptr, 0, USER_CTX_SIZE);

		r = mes_v12_1_test_queue(adev, xcc_id, pasid, vm, meta_gpu_addr,
					 ctx_gpu_addr, ctx_ptr, queue_types[i]);
		if (r)
			break;
	}

	amdgpu_unmap_static_csa(adev, vm, ctx_bo, bo_va, USER_CTX_VA);
err0:
	amdgpu_vm_fini(adev, vm);
err1:
	kfree(vm);
err2:
	amdgpu_bo_free_kernel(&meta_bo, &meta_gpu_addr, &meta_ptr);
	amdgpu_bo_free_kernel(&ctx_bo, &ctx_gpu_addr, &ctx_ptr);
	amdgpu_pasid_free(pasid);
	return r;
}