drm/amdgpu/mes_v12_1: initial support for mes_v12_1

Duplicated and rename mes ip version name to v12_1_0.
Fix to access correct ring pipe by xcc_id.
Fix to access correct instance registers by xcc_id.
Fix to access correct index registers by grbm/xcc_id.

v2: rebase (Alex)
v3: fix sw_fini (Alex)

Signed-off-by: Jack Xiao <Jack.Xiao@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 1792 insertions, 0 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 139ceafa164a..367f2da5f510 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -179,6 +179,7 @@ amdgpu-y += \
 	amdgpu_mes.o \
 	mes_v11_0.o \
 	mes_v12_0.o \
+	mes_v12_1.o
 
 # add GFX userqueue support
 amdgpu-y += mes_userqueue.o
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
new file mode 100644
index 000000000000..ee5a2317e414
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
@@ -0,0 +1,1762 @@
+/*
+ * 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"
+
+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_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);
+
+#define MES_EOP_SIZE   2048
+
+#define regCP_HQD_IB_CONTROL_MES_12_1_DEFAULT 0x100000
+
+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",
+};
+
+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)
+{
+	struct amdgpu_device *adev = mes->adev;
+	union MESAPI__ADD_QUEUE mes_add_queue_pkt;
+	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];
+	uint32_t vm_cntx_cntl = hub->vm_cntx_cntl;
+
+	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 = 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;
+
+	return mes_v12_1_submit_pkt_and_poll_completion(mes,
+			input->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;
+
+	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,
+			input->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 int mes_v12_1_misc_op(struct amdgpu_mes *mes,
+			     struct mes_misc_op_input *input)
+{
+	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.reg_offset = input->read_reg.reg_offset;
+		misc_pkt.read_reg.buffer_addr = input->read_reg.buffer_addr;
+		break;
+	case MES_MISC_OP_WRITE_REG:
+		misc_pkt.opcode = MESAPI_MISC__WRITE_REG;
+		misc_pkt.write_reg.reg_offset = input->write_reg.reg_offset;
+		misc_pkt.write_reg.reg_value = input->write_reg.reg_value;
+		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_offset1 = input->wrm_reg.reg0;
+		misc_pkt.wait_reg_mem.reg_offset2 = 0;
+		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;
+		misc_pkt.wait_reg_mem.reg_offset1 = input->wrm_reg.reg0;
+		misc_pkt.wait_reg_mem.reg_offset2 = input->wrm_reg.reg1;
+		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;
+
+	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;
+
+	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 + 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 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,
+};
+
+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) {