drm/amdkfd: Add navi10 support to amdkfd. (v3)

KFD (kernel fusion driver) is the kernel driver
for the compute backend for usermode compute
stack.

v2: squash in updates (Alex)
v3: squash in rebase fixes (Alex)
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Signed-off-by: Philip Cox <Philip.Cox@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -693,6 +693,14 @@ MODULE_PARM_DESC(halt_if_hws_hang, "Halt if HWS hang is detected (0 = off (defau
 bool hws_gws_support;
 module_param(hws_gws_support, bool, 0444);
 MODULE_PARM_DESC(hws_gws_support, "MEC FW support gws barriers (false = not supported (Default), true = supported)");
+
+/**
+  * DOC: queue_preemption_timeout_ms (int)
+  * queue preemption timeout in ms (1 = Minimum, 9000 = default)
+  */
+int queue_preemption_timeout_ms;
+module_param(queue_preemption_timeout_ms, int, 0644);
+MODULE_PARM_DESC(queue_preemption_timeout_ms, "queue preemption timeout in ms (1 = Minimum, 9000 = default)");
 #endif
 
 /**

drivers/gpu/drm/amd/amdkfd/Makefile

@@ -36,16 +36,19 @@ AMDKFD_FILES	:= $(AMDKFD_PATH)/kfd_module.o \
 		$(AMDKFD_PATH)/kfd_mqd_manager_cik.o \
 		$(AMDKFD_PATH)/kfd_mqd_manager_vi.o \
 		$(AMDKFD_PATH)/kfd_mqd_manager_v9.o \
+		$(AMDKFD_PATH)/kfd_mqd_manager_v10.o \
 		$(AMDKFD_PATH)/kfd_kernel_queue.o \
 		$(AMDKFD_PATH)/kfd_kernel_queue_cik.o \
 		$(AMDKFD_PATH)/kfd_kernel_queue_vi.o \
 		$(AMDKFD_PATH)/kfd_kernel_queue_v9.o \
+		$(AMDKFD_PATH)/kfd_kernel_queue_v10.o \
 		$(AMDKFD_PATH)/kfd_packet_manager.o \
 		$(AMDKFD_PATH)/kfd_process_queue_manager.o \
 		$(AMDKFD_PATH)/kfd_device_queue_manager.o \
 		$(AMDKFD_PATH)/kfd_device_queue_manager_cik.o \
 		$(AMDKFD_PATH)/kfd_device_queue_manager_vi.o \
 		$(AMDKFD_PATH)/kfd_device_queue_manager_v9.o \
+		$(AMDKFD_PATH)/kfd_device_queue_manager_v10.o \
 		$(AMDKFD_PATH)/kfd_interrupt.o \
 		$(AMDKFD_PATH)/kfd_events.o \
 		$(AMDKFD_PATH)/cik_event_interrupt.o \

drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h

@@ -561,3 +561,302 @@ static const uint32_t cwsr_trap_gfx9_hex[] = {
 	0xbf8a0000, 0x95806f6c,
 	0xbf810000, 0x00000000,
 };
+
+static const uint32_t cwsr_trap_gfx10_hex[] = {
+	0xbf820001, 0xbf82012e,
+	0xb0804004, 0xb970f802,
+	0x8a708670, 0xb971f803,
+	0x8771ff71, 0x00000400,
+	0xbf850008, 0xb971f803,
+	0x8771ff71, 0x000001ff,
+	0xbf850001, 0x806c846c,
+	0x876dff6d, 0x0000ffff,
+	0xbe80226c, 0xb971f803,
+	0x8771ff71, 0x00000100,
+	0xbf840006, 0xbef60380,
+	0xb9f60203, 0x876dff6d,
+	0x0000ffff, 0x80ec886c,
+	0x82ed806d, 0xbef60380,
+	0xb9f60283, 0xb973f816,
+	0xb9762c07, 0x8f769c76,
+	0x886d766d, 0xb97603c7,
+	0x8f769b76, 0x886d766d,
+	0xb976f807, 0x8776ff76,
+	0x00007fff, 0xb9f6f807,
+	0xbeee037e, 0xbeef037f,
+	0xbefe0480, 0xbf900004,
+	0xbf8e0002, 0xbf88fffe,
+	0xbef4037e, 0x8775ff7f,
+	0x0000ffff, 0x8875ff75,
+	0x00040000, 0xbef60380,
+	0xbef703ff, 0x00807fac,
+	0x8776ff7f, 0x08000000,
+	0x90768376, 0x88777677,
+	0x8776ff7f, 0x70000000,
+	0x90768176, 0x88777677,
+	0xbefb037c, 0xbefa0380,
+	0xb97202dc, 0x8872727f,
+	0xbefe03c1, 0x877c8172,
+	0xbf06817c, 0xbf850002,
+	0xbeff0380, 0xbf820001,
+	0xbeff03c1, 0xb9712a05,
+	0x80718171, 0x8f718271,
+	0x877c8172, 0xbf06817c,
+	0xbf85000d, 0x8f768771,
+	0xbef603ff, 0x01000000,
+	0xbefc0380, 0x7e008700,
+	0xe0704000, 0x7a5d0000,
+	0x807c817c, 0x807aff7a,
+	0x00000080, 0xbf0a717c,
+	0xbf85fff8, 0xbf82001b,
+	0x8f768871, 0xbef603ff,
+	0x01000000, 0xbefc0380,
+	0x7e008700, 0xe0704000,
+	0x7a5d0000, 0x807c817c,
+	0x807aff7a, 0x00000100,
+	0xbf0a717c, 0xbf85fff8,
+	0xb9711e06, 0x8771c171,
+	0xbf84000c, 0x8f718371,
+	0x80717c71, 0xbefe03c1,
+	0xbeff0380, 0x7e008700,
+	0xe0704000, 0x7a5d0000,
+	0x807c817c, 0x807aff7a,
+	0x00000080, 0xbf0a717c,
+	0xbf85fff8, 0xbf8a0000,
+	0x8776ff72, 0x04000000,
+	0xbf84002b, 0xbefe03c1,
+	0x877c8172, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820001, 0xbeff03c1,
+	0xb9714306, 0x8771c171,
+	0xbf840021, 0x8f718671,
+	0x8f718271, 0xbef60371,
+	0xbef603ff, 0x01000000,
+	0xd7650000, 0x000100c1,
+	0xd7660000, 0x000200c1,
+	0x16000084, 0x877c8172,
+	0xbf06817c, 0xbefc0380,
+	0xbf85000a, 0x807cff7c,
+	0x00000080, 0x807aff7a,
+	0x00000080, 0xd5250000,
+	0x0001ff00, 0x00000080,
+	0xbf0a717c, 0xbf85fff7,
+	0xbf820009, 0x807cff7c,
+	0x00000100, 0x807aff7a,
+	0x00000100, 0xd5250000,
+	0x0001ff00, 0x00000100,
+	0xbf0a717c, 0xbf85fff7,
+	0x877c8172, 0xbf06817c,
+	0xbf850003, 0x8f7687ff,
+	0x0000006a, 0xbf820002,
+	0x8f7688ff, 0x0000006a,
+	0xbef603ff, 0x01000000,
+	0x877c8172, 0xbf06817c,
+	0xbefc0380, 0xbf800000,
+	0xbf85000b, 0xbe802e00,
+	0x7e000200, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0x807c817c,
+	0xbf0aff7c, 0x0000006a,
+	0xbf85fff6, 0xbf82000a,
+	0xbe802e00, 0x7e000200,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000100,
+	0x807c817c, 0xbf0aff7c,
+	0x0000006a, 0xbf85fff6,
+	0xbef60384, 0xbef603ff,
+	0x01000000, 0x877c8172,
+	0xbf06817c, 0xbf850030,
+	0x7e00027b, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0x7e00026c,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000080,
+	0x7e00026d, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0x7e00026e,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000080,
+	0x7e00026f, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0x7e000270,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000080,
+	0xb971f803, 0x7e000271,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000080,
+	0x7e000273, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0xb97bf801,
+	0x7e00027b, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000080, 0xbf82002f,
+	0x7e00027b, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000100, 0x7e00026c,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000100,
+	0x7e00026d, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000100, 0x7e00026e,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000100,
+	0x7e00026f, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000100, 0x7e000270,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000100,
+	0xb971f803, 0x7e000271,
+	0xe0704000, 0x7a5d0000,
+	0x807aff7a, 0x00000100,
+	0x7e000273, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000100, 0xb97bf801,
+	0x7e00027b, 0xe0704000,
+	0x7a5d0000, 0x807aff7a,
+	0x00000100, 0xbf820119,
+	0xbef4037e, 0x8775ff7f,
+	0x0000ffff, 0x8875ff75,
+	0x00040000, 0xbef60380,
+	0xbef703ff, 0x00807fac,
+	0x8772ff7f, 0x08000000,
+	0x90728372, 0x88777277,
+	0x8772ff7f, 0x70000000,
+	0x90728172, 0x88777277,
+	0xb97902dc, 0x8879797f,
+	0xbef80380, 0xbefe03c1,
+	0x877c8179, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820001, 0xbeff03c1,
+	0xb96f2a05, 0x806f816f,
+	0x8f6f826f, 0x877c8179,
+	0xbf06817c, 0xbf850013,
+	0x8f76876f, 0xbef603ff,
+	0x01000000, 0xbef20378,
+	0x8078ff78, 0x00000080,
+	0xbefc0381, 0xe0304000,
+	0x785d0000, 0xbf8c3f70,
+	0x7e008500, 0x807c817c,
+	0x8078ff78, 0x00000080,
+	0xbf0a6f7c, 0xbf85fff7,
+	0xe0304000, 0x725d0000,
+	0xbf820023, 0x8f76886f,
+	0xbef603ff, 0x01000000,
+	0xbef20378, 0x8078ff78,
+	0x00000100, 0xbefc0381,
+	0xe0304000, 0x785d0000,
+	0xbf8c3f70, 0x7e008500,
+	0x807c817c, 0x8078ff78,
+	0x00000100, 0xbf0a6f7c,
+	0xbf85fff7, 0xb96f1e06,
+	0x876fc16f, 0xbf84000e,
+	0x8f6f836f, 0x806f7c6f,
+	0xbefe03c1, 0xbeff0380,
+	0xe0304000, 0x785d0000,
+	0xbf8c3f70, 0x7e008500,
+	0x807c817c, 0x8078ff78,
+	0x00000080, 0xbf0a6f7c,
+	0xbf85fff7, 0xbeff03c1,
+	0xe0304000, 0x725d0000,
+	0x8772ff79, 0x04000000,
+	0xbf840020, 0xbefe03c1,
+	0x877c8179, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820001, 0xbeff03c1,
+	0xb96f4306, 0x876fc16f,
+	0xbf840016, 0x8f6f866f,
+	0x8f6f826f, 0xbef6036f,
+	0xbef603ff, 0x01000000,
+	0x877c8172, 0xbf06817c,
+	0xbefc0380, 0xbf850007,
+	0x807cff7c, 0x00000080,
+	0x8078ff78, 0x00000080,
+	0xbf0a6f7c, 0xbf85fffa,
+	0xbf820006, 0x807cff7c,
+	0x00000100, 0x8078ff78,
+	0x00000100, 0xbf0a6f7c,
+	0xbf85fffa, 0x877c8179,
+	0xbf06817c, 0xbf850003,
+	0x8f7687ff, 0x0000006a,
+	0xbf820002, 0x8f7688ff,
+	0x0000006a, 0xbef603ff,
+	0x01000000, 0x877c8179,
+	0xbf06817c, 0xbf850012,
+	0xf4211cba, 0xf0000000,
+	0x8078ff78, 0x00000080,
+	0xbefc0381, 0xf421003a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xbf8cc07f,
+	0xbe803000, 0xbf800000,
+	0x807c817c, 0xbf0aff7c,
+	0x0000006a, 0xbf85fff5,
+	0xbe800372, 0xbf820011,
+	0xf4211cba, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xbefc0381, 0xf421003a,
+	0xf0000000, 0x8078ff78,
+	0x00000100, 0xbf8cc07f,
+	0xbe803000, 0xbf800000,
+	0x807c817c, 0xbf0aff7c,
+	0x0000006a, 0xbf85fff5,
+	0xbe800372, 0xbef60384,
+	0xbef603ff, 0x01000000,
+	0x877c8179, 0xbf06817c,
+	0xbf850025, 0xf4211bfa,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211b3a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211b7a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211eba,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211efa,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211c3a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211c7a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211cfa,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xf4211e7a,
+	0xf0000000, 0x8078ff78,
+	0x00000080, 0xbf820024,
+	0xf4211bfa, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211b3a, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211b7a, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211eba, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211efa, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211c3a, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211c7a, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211cfa, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xf4211e7a, 0xf0000000,
+	0x8078ff78, 0x00000100,
+	0xbf8cc07f, 0x876dff6d,
+	0x0000ffff, 0xbefc036f,
+	0xbefe037a, 0xbeff037b,
+	0x876f71ff, 0x000003ff,
+	0xb9ef4803, 0xb9f3f816,
+	0x876f71ff, 0xfffff800,
+	0x906f8b6f, 0xb9efa2c3,
+	0xb9f9f801, 0x876fff6d,
+	0xf0000000, 0x906f9c6f,
+	0x8f6f906f, 0xbef20380,
+	0x88726f72, 0x876fff6d,
+	0x08000000, 0x906f9b6f,
+	0x8f6f8f6f, 0x88726f72,
+	0x876fff70, 0x00800000,
+	0x906f976f, 0xb9f2f807,
+	0xb9f0f802, 0xbf8a0000,
+	0xbe80226c, 0xbf810000,
+	0xbf9f0000, 0xbf9f0000,
+	0xbf9f0000, 0xbf9f0000,
+	0xbf9f0000, 0x00000000,
+};

drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm

+/*
+ * Copyright 2018 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.
+ */
+
+
+shader main
+
+asic(DEFAULT)
+
+type(CS)
+
+wave_size(32)
+/*************************************************************************/
+/*					control on how to run the shader					 */
+/*************************************************************************/
+//any hack that needs to be made to run this code in EMU (either becasue various EMU code are not ready or no compute save & restore in EMU run)
+var EMU_RUN_HACK					=	0
+var EMU_RUN_HACK_RESTORE_NORMAL		=	0
+var EMU_RUN_HACK_SAVE_NORMAL_EXIT	=	0
+var	EMU_RUN_HACK_SAVE_SINGLE_WAVE	=	0
+var EMU_RUN_HACK_SAVE_FIRST_TIME	= 	0					//for interrupted restore in which the first save is through EMU_RUN_HACK
+var SAVE_LDS						= 	0
+var WG_BASE_ADDR_LO					=   0x9000a000
+var WG_BASE_ADDR_HI					=	0x0
+var WAVE_SPACE						=	0x9000				//memory size that each wave occupies in workgroup state mem, increase from 5000 to 9000 for more SGPR need to be saved
+var CTX_SAVE_CONTROL				=	0x0
+var CTX_RESTORE_CONTROL				=	CTX_SAVE_CONTROL
+var SIM_RUN_HACK					=	0					//any hack that needs to be made to run this code in SIM (either becasue various RTL code are not ready or no compute save & restore in RTL run)
+var	SGPR_SAVE_USE_SQC				=	0					//use SQC D$ to do the write
+var USE_MTBUF_INSTEAD_OF_MUBUF		=	0					//need to change BUF_DATA_FORMAT in S_SAVE_BUF_RSRC_WORD3_MISC from 0 to BUF_DATA_FORMAT_32 if set to 1 (i.e. 0x00827FAC)
+var SWIZZLE_EN						=	0					//whether we use swizzled buffer addressing
+var SAVE_RESTORE_HWID_DDID          =   0
+var RESTORE_DDID_IN_SGPR18          =   0
+/**************************************************************************/
+/*                     	variables							              */
+/**************************************************************************/
+var SQ_WAVE_STATUS_INST_ATC_SHIFT  = 23
+var SQ_WAVE_STATUS_INST_ATC_MASK   = 0x00800000
+var SQ_WAVE_STATUS_SPI_PRIO_MASK   = 0x00000006
+
+var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT	= 12
+var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE		= 9
+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT	= 8
+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE	= 6
+var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT	= 24
+var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE	= 4						//FIXME	 sq.blk still has 4 bits at this time while SQ programming guide has 3 bits
+var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT    = 24
+var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE     = 4
+var SQ_WAVE_IB_STS2_WAVE64_SHIFT        = 11
+var SQ_WAVE_IB_STS2_WAVE64_SIZE         = 1
+
+var	SQ_WAVE_TRAPSTS_SAVECTX_MASK	=	0x400
+var SQ_WAVE_TRAPSTS_EXCE_MASK       =   0x1FF          			// Exception mask
+var	SQ_WAVE_TRAPSTS_SAVECTX_SHIFT	=	10					
+var	SQ_WAVE_TRAPSTS_MEM_VIOL_MASK	=	0x100					
+var	SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT	=	8		
+var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK 	=	0x3FF
+var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT 	=	0x0
+var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE 	=	10
+var	SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK 	=	0xFFFFF800	
+var	SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT 	=	11
+var	SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE 	=	21	
+
+var SQ_WAVE_IB_STS_RCNT_SHIFT			=	16					//FIXME
+var SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT	=	15					//FIXME
+var SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE    =   1                   //FIXME
+var SQ_WAVE_IB_STS_RCNT_SIZE            =   6                   //FIXME
+var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF	//FIXME
+ 
+var	SQ_BUF_RSRC_WORD1_ATC_SHIFT		=	24
+var	SQ_BUF_RSRC_WORD3_MTYPE_SHIFT	=	27
+
+
+/*      Save        */
+var	S_SAVE_BUF_RSRC_WORD1_STRIDE		=	0x00040000  		//stride is 4 bytes 
+var	S_SAVE_BUF_RSRC_WORD3_MISC			= 	0x00807FAC			//SQ_SEL_X/Y/Z/W, BUF_NUM_FORMAT_FLOAT, (0 for MUBUF stride[17:14] when ADD_TID_ENABLE and BUF_DATA_FORMAT_32 for MTBUF), ADD_TID_ENABLE			
+
+var	S_SAVE_SPI_INIT_ATC_MASK			=	0x08000000			//bit[27]: ATC bit
+var	S_SAVE_SPI_INIT_ATC_SHIFT			=	27
+var	S_SAVE_SPI_INIT_MTYPE_MASK			=	0x70000000			//bit[30:28]: Mtype
+var	S_SAVE_SPI_INIT_MTYPE_SHIFT			=	28
+var	S_SAVE_SPI_INIT_FIRST_WAVE_MASK		=	0x04000000			//bit[26]: FirstWaveInTG
+var	S_SAVE_SPI_INIT_FIRST_WAVE_SHIFT	=	26
+
+var S_SAVE_PC_HI_RCNT_SHIFT				=	28					//FIXME	 check with Brian to ensure all fields other than PC[47:0] can be used
+var S_SAVE_PC_HI_RCNT_MASK				=   0xF0000000			//FIXME
+var S_SAVE_PC_HI_FIRST_REPLAY_SHIFT		=	27					//FIXME
+var S_SAVE_PC_HI_FIRST_REPLAY_MASK		=	0x08000000			//FIXME
+
+var	s_save_spi_init_lo				=	exec_lo
+var s_save_spi_init_hi				=	exec_hi
+
+var	s_save_pc_lo			=	ttmp0			//{TTMP1, TTMP0} = {3h0,pc_rewind[3:0], HT[0],trapID[7:0], PC[47:0]}
+var	s_save_pc_hi			=	ttmp1			
+var s_save_exec_lo			=	ttmp2
+var s_save_exec_hi			= 	ttmp3			
+var	s_save_status			=	ttmp4			
+var	s_save_trapsts			=	ttmp5			//not really used until the end of the SAVE routine
+var s_wave_size         	=	ttmp6           //ttmp6 is not needed now, since it's only 32bit xnack mask, now use it to determine wave32 or wave64 in EMU_HACK
+var s_save_xnack_mask	    =	ttmp7
+var	s_save_buf_rsrc0		=	ttmp8
+var	s_save_buf_rsrc1		=	ttmp9
+var	s_save_buf_rsrc2		=	ttmp10
+var	s_save_buf_rsrc3		=	ttmp11
+
+var s_save_mem_offset		= 	ttmp14
+var s_sgpr_save_num         =   106                     //in gfx10, all sgpr must be saved
+var s_save_alloc_size		=	s_save_trapsts			//conflict
+var s_save_tmp              =   s_save_buf_rsrc2       	//shared with s_save_buf_rsrc2  (conflict: should not use mem access with s_save_tmp at the same time)
+var s_save_m0				=	ttmp15					
+
+/*      Restore     */
+var	S_RESTORE_BUF_RSRC_WORD1_STRIDE			=	S_SAVE_BUF_RSRC_WORD1_STRIDE 
+var	S_RESTORE_BUF_RSRC_WORD3_MISC			= 	S_SAVE_BUF_RSRC_WORD3_MISC		 
+
+var	S_RESTORE_SPI_INIT_ATC_MASK			    =	0x08000000			//bit[27]: ATC bit
+var	S_RESTORE_SPI_INIT_ATC_SHIFT			=	27
+var	S_RESTORE_SPI_INIT_MTYPE_MASK			=	0x70000000			//bit[30:28]: Mtype
+var	S_RESTORE_SPI_INIT_MTYPE_SHIFT			=	28
+var	S_RESTORE_SPI_INIT_FIRST_WAVE_MASK		=	0x04000000			//bit[26]: FirstWaveInTG
+var	S_RESTORE_SPI_INIT_FIRST_WAVE_SHIFT	    =	26
+
+var S_RESTORE_PC_HI_RCNT_SHIFT				=	S_SAVE_PC_HI_RCNT_SHIFT
+var S_RESTORE_PC_HI_RCNT_MASK				=   S_SAVE_PC_HI_RCNT_MASK
+var S_RESTORE_PC_HI_FIRST_REPLAY_SHIFT		=	S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+var S_RESTORE_PC_HI_FIRST_REPLAY_MASK		=	S_SAVE_PC_HI_FIRST_REPLAY_MASK
+
+var s_restore_spi_init_lo                   =   exec_lo
+var s_restore_spi_init_hi                   =   exec_hi
+
+var s_restore_mem_offset		= 	ttmp12
+var s_restore_alloc_size		=	ttmp3
+var s_restore_tmp           	=   ttmp6
+var s_restore_mem_offset_save	= 	s_restore_tmp 		//no conflict
+
+var s_restore_m0			=	s_restore_alloc_size	//no conflict			
+
+var s_restore_mode			=  	ttmp13
+var s_restore_hwid1         =  ttmp2
+var s_restore_ddid          =  s_restore_hwid1
+var	s_restore_pc_lo		    =	ttmp0			
+var	s_restore_pc_hi		    =	ttmp1
+var s_restore_exec_lo		=	ttmp14
+var s_restore_exec_hi		= 	ttmp15
+var	s_restore_status	    =	ttmp4			
+var	s_restore_trapsts	    =	ttmp5
+//var s_restore_xnack_mask_lo	=	xnack_mask_lo
+//var s_restore_xnack_mask_hi	=	xnack_mask_hi
+var s_restore_xnack_mask    =   ttmp7
+var	s_restore_buf_rsrc0		=	ttmp8
+var	s_restore_buf_rsrc1		=	ttmp9
+var	s_restore_buf_rsrc2		=	ttmp10
+var	s_restore_buf_rsrc3		=	ttmp11
+var s_restore_size         	=	ttmp13                  //ttmp13 has no conflict
+
+/**************************************************************************/
+/*                     	trap handler entry points			              */
+/**************************************************************************/
+    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL)) 					//hack to use trap_id for determining save/restore
+		//FIXME VCCZ un-init assertion s_getreg_b32  	s_save_status, hwreg(HW_REG_STATUS)			//save STATUS since we will change SCC
+		s_and_b32 s_save_tmp, s_save_pc_hi, 0xffff0000 				//change SCC
+    	s_cmp_eq_u32 s_save_tmp, 0x007e0000  						//Save: trap_id = 0x7e. Restore: trap_id = 0x7f.  
+    	s_cbranch_scc0 L_JUMP_TO_RESTORE							//do not need to recover STATUS here  since we are going to RESTORE
+		//FIXME  s_setreg_b32 	hwreg(HW_REG_STATUS), 	s_save_status		//need to recover STATUS since we are going to SAVE	
+		s_branch L_SKIP_RESTORE 									//NOT restore, SAVE actually
+	else	
+		s_branch L_SKIP_RESTORE 									//NOT restore. might be a regular trap or save
+    end
+
+L_JUMP_TO_RESTORE:
+    s_branch L_RESTORE												//restore
+
+L_SKIP_RESTORE:
+	
+	s_getreg_b32  	s_save_status, hwreg(HW_REG_STATUS)								//save STATUS since we will change SCC
+    s_andn2_b32		s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK      //check whether this is for save
+	s_getreg_b32  	s_save_trapsts, hwreg(HW_REG_TRAPSTS)    		 				
+	s_and_b32		s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK	//check whether this is for save  
+	s_cbranch_scc1	L_SAVE															//this is the operation for save
+
+    // *********    Handle non-CWSR traps       *******************
+    if (!EMU_RUN_HACK)
+		s_getreg_b32     s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+		s_and_b32        s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_EXCE_MASK // Check whether it is an exception
+		s_cbranch_scc1  L_EXCP_CASE   // Exception, jump back to the shader program directly.
+		s_add_u32    ttmp0, ttmp0, 4   // S_TRAP case, add 4 to ttmp0 
+		
+		L_EXCP_CASE:
+		s_and_b32    ttmp1, ttmp1, 0xFFFF
+		s_rfe_b64    [ttmp0, ttmp1]
+	end
+    // *********        End handling of non-CWSR traps   *******************
+
+/**************************************************************************/
+/*                     	save routine						              */
+/**************************************************************************/
+
+L_SAVE:	
+	
+	//check whether there is mem_viol
+	s_getreg_b32  	s_save_trapsts, hwreg(HW_REG_TRAPSTS)    		 				
+	s_and_b32		s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK			
+	s_cbranch_scc0	L_NO_PC_REWIND
+    
+	//if so, need rewind PC assuming GDS operation gets NACKed
+	s_mov_b32       s_save_tmp, 0															//clear mem_viol bit
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT, 1), s_save_tmp	//clear mem_viol bit 
+	s_and_b32 		s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
+	s_sub_u32 		s_save_pc_lo, s_save_pc_lo, 8             //pc[31:0]-8
+	s_subb_u32 		s_save_pc_hi, s_save_pc_hi, 0x0			  // -scc
+
+L_NO_PC_REWIND:
+    s_mov_b32       s_save_tmp, 0															//clear saveCtx bit
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp		//clear saveCtx bit   
+
+	//s_mov_b32		s_save_xnack_mask_lo,	xnack_mask_lo									//save XNACK_MASK  
+	//s_mov_b32		s_save_xnack_mask_hi,	xnack_mask_hi
+    s_getreg_b32	s_save_xnack_mask,  hwreg(HW_REG_SHADER_XNACK_MASK)  
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_RCNT_SHIFT, SQ_WAVE_IB_STS_RCNT_SIZE)					//save RCNT
+	s_lshl_b32		s_save_tmp, s_save_tmp, S_SAVE_PC_HI_RCNT_SHIFT
+	s_or_b32		s_save_pc_hi, s_save_pc_hi, s_save_tmp
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT, SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE)	//save FIRST_REPLAY
+	s_lshl_b32		s_save_tmp, s_save_tmp, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+	s_or_b32		s_save_pc_hi, s_save_pc_hi, s_save_tmp
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS)										//clear RCNT and FIRST_REPLAY in IB_STS
+	s_and_b32		s_save_tmp, s_save_tmp, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG
+
+	s_setreg_b32	hwreg(HW_REG_IB_STS), s_save_tmp
+    
+	/*		inform SPI the readiness and wait for SPI's go signal */
+	s_mov_b32		s_save_exec_lo,	exec_lo													//save EXEC and use EXEC for the go signal from SPI
+	s_mov_b32		s_save_exec_hi,	exec_hi
+	s_mov_b64		exec, 	0x0																//clear EXEC to get ready to receive
+	if (EMU_RUN_HACK)
+	
+	else
+		s_sendmsg	sendmsg(MSG_SAVEWAVE)													//send SPI a message and wait for SPI's write to EXEC  
+	end
+
+  L_SLEEP:		
+	s_sleep 0x2
+	
+	if (EMU_RUN_HACK)
+																							
+	else
+		s_cbranch_execz	L_SLEEP                                                         
+	end
+
+
+	/*      setup Resource Contants    */
+	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_SINGLE_WAVE))	
+		//calculate wd_addr using absolute thread id 
+		v_readlane_b32 s_save_tmp, v9, 0
+        //determine it is wave32 or wave64
+        s_getreg_b32 	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
+        s_cmp_eq_u32    s_wave_size, 0
+        s_cbranch_scc1  L_SAVE_WAVE32
+        s_lshr_b32 s_save_tmp, s_save_tmp, 6 //SAVE WAVE64
+        s_branch    L_SAVE_CON
+    L_SAVE_WAVE32:
+        s_lshr_b32 s_save_tmp, s_save_tmp, 5 //SAVE WAVE32
+    L_SAVE_CON:
+		s_mul_i32 s_save_tmp, s_save_tmp, WAVE_SPACE
+		s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
+		s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
+		s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL		
+	else
+	end
+	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_SINGLE_WAVE))
+		s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
+		s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
+		s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL		
+	else
+	end
+	
+	
+	s_mov_b32		s_save_buf_rsrc0, 	s_save_spi_init_lo														//base_addr_lo
+	s_and_b32		s_save_buf_rsrc1, 	s_save_spi_init_hi, 0x0000FFFF											//base_addr_hi
+	s_or_b32		s_save_buf_rsrc1, 	s_save_buf_rsrc1,  S_SAVE_BUF_RSRC_WORD1_STRIDE
+    s_mov_b32       s_save_buf_rsrc2,   0                                               						//NUM_RECORDS initial value = 0 (in bytes) although not neccessarily inited
+	s_mov_b32		s_save_buf_rsrc3, 	S_SAVE_BUF_RSRC_WORD3_MISC
+	s_and_b32		s_save_tmp,         s_save_spi_init_hi, S_SAVE_SPI_INIT_ATC_MASK		
+	s_lshr_b32		s_save_tmp,  		s_save_tmp, (S_SAVE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)			//get ATC bit into position
+	s_or_b32		s_save_buf_rsrc3, 	s_save_buf_rsrc3,  s_save_tmp											//or ATC
+	s_and_b32		s_save_tmp,         s_save_spi_init_hi, S_SAVE_SPI_INIT_MTYPE_MASK		
+	s_lshr_b32		s_save_tmp,  		s_save_tmp, (S_SAVE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)		//get MTYPE bits into position
+	s_or_b32		s_save_buf_rsrc3, 	s_save_buf_rsrc3,  s_save_tmp											//or MTYPE	
+	
+	s_mov_b32		s_save_m0,			m0																	//save M0
+	
+	/* 		global mem offset			*/
+	s_mov_b32		s_save_mem_offset, 	0x0																		//mem offset initial value = 0
+    s_getreg_b32 	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE) //get wave_save_size
+    s_or_b32        s_wave_size, s_save_spi_init_hi,    s_wave_size                                             //share s_wave_size with exec_hi
+
+    /*      	save VGPRs	    */
+	//////////////////////////////
+  L_SAVE_VGPR:
+  
+ 	s_mov_b32		exec_lo, 0xFFFFFFFF 											//need every thread from now on
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1  
+    s_cbranch_scc1  L_ENABLE_SAVE_VGPR_EXEC_HI   
+    s_mov_b32		exec_hi, 0x00000000
+    s_branch        L_SAVE_VGPR_NORMAL
+  L_ENABLE_SAVE_VGPR_EXEC_HI:
+	s_mov_b32		exec_hi, 0xFFFFFFFF
+  L_SAVE_VGPR_NORMAL:	
+	s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) 					//vpgr_size
+	//for wave32 and wave64, the num of vgpr function is the same?
+    s_add_u32 		s_save_alloc_size, s_save_alloc_size, 1
+	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 2 						//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)   //FIXME for GFX, zero is possible
+    //determine it is wave32 or wave64
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_SAVE_VGPR_WAVE64
+
+    //zhenxu added it for save vgpr for wave32
+	s_lshl_b32		s_save_buf_rsrc2,  s_save_alloc_size, 7							//NUM_RECORDS in bytes (32 threads*4)
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+	
+    s_mov_b32 		m0, 0x0 														//VGPR initial index value =0
+	//s_set_gpr_idx_on  m0, 0x1														//M0[7:0] = M0[7:0] and M0[15:12] = 0x1
+    //s_add_u32		s_save_alloc_size, s_save_alloc_size, 0x1000					//add 0x1000 since we compare m0 against it later, doesn't need this in gfx10
+
+  L_SAVE_VGPR_WAVE32_LOOP: 										
+	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
+	    
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	end
+
+    s_add_u32		m0, m0, 1														//next vgpr index
+	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128						//every buffer_store_dword does 128 bytes
+	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_SAVE_VGPR_WAVE32_LOOP												//VGPR save is complete?
+    s_branch    L_SAVE_LDS
+    //save vgpr for wave32 ends
+
+  L_SAVE_VGPR_WAVE64:
+	s_lshl_b32		s_save_buf_rsrc2,  s_save_alloc_size, 8							//NUM_RECORDS in bytes (64 threads*4)
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+	
+    s_mov_b32 		m0, 0x0 														//VGPR initial index value =0
+	//s_set_gpr_idx_on  m0, 0x1														//M0[7:0] = M0[7:0] and M0[15:12] = 0x1
+    //s_add_u32		s_save_alloc_size, s_save_alloc_size, 0x1000					//add 0x1000 since we compare m0 against it later, doesn't need this in gfx10
+
+  L_SAVE_VGPR_WAVE64_LOOP: 										
+	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
+	    
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	end
+
+    s_add_u32		m0, m0, 1														//next vgpr index
+	s_add_u32		s_save_mem_offset, s_save_mem_offset, 256						//every buffer_store_dword does 256 bytes
+	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_SAVE_VGPR_WAVE64_LOOP												//VGPR save is complete?
+	//s_set_gpr_idx_off
+    //
+    //Below part will be the save shared vgpr part (new for gfx10)
+    s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE) 			//shared_vgpr_size
+    s_and_b32		s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF				//shared_vgpr_size is zero?
+    s_cbranch_scc0	L_SAVE_LDS													    //no shared_vgpr used? jump to L_SAVE_LDS
+    s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 3 						//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
+    //m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
+    //save shared_vgpr will start from the index of m0
+    s_add_u32       s_save_alloc_size, s_save_alloc_size, m0
+    s_mov_b32		exec_lo, 0xFFFFFFFF
+    s_mov_b32		exec_hi, 0x00000000
+    L_SAVE_SHARED_VGPR_WAVE64_LOOP: 										
+	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
+	buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+    s_add_u32		m0, m0, 1														//next vgpr index
+	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128						//every buffer_store_dword does 256 bytes
+	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_SAVE_SHARED_VGPR_WAVE64_LOOP									//SHARED_VGPR save is complete?
+    
+	/*      	save LDS	    */
+	//////////////////////////////
+  L_SAVE_LDS:
+
+    //Only check the first wave need LDS
+	/*      the first wave in the threadgroup    */
+	s_barrier																		//FIXME  not performance-optimal "LDS is used? wait for other waves in the same TG" 
+	s_and_b32		s_save_tmp, s_wave_size, S_SAVE_SPI_INIT_FIRST_WAVE_MASK								//exec is still used here
+	s_cbranch_scc0	L_SAVE_SGPR
+	
+	s_mov_b32		exec_lo, 0xFFFFFFFF 											//need every thread from now on
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_ENABLE_SAVE_LDS_EXEC_HI   
+    s_mov_b32		exec_hi, 0x00000000
+    s_branch        L_SAVE_LDS_NORMAL
+  L_ENABLE_SAVE_LDS_EXEC_HI:
+	s_mov_b32		exec_hi, 0xFFFFFFFF
+  L_SAVE_LDS_NORMAL:	
+	s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE) 			//lds_size
+	s_and_b32		s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF				//lds_size is zero?
+	s_cbranch_scc0	L_SAVE_SGPR														//no lds used? jump to L_SAVE_VGPR
+	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 6 						//LDS size in dwords = lds_size * 64dw
+	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 2 						//LDS size in bytes
+	s_mov_b32		s_save_buf_rsrc2,  s_save_alloc_size  							//NUM_RECORDS in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+
+    //load 0~63*4(byte address) to vgpr v15
+    v_mbcnt_lo_u32_b32 v0, -1, 0
+    v_mbcnt_hi_u32_b32 v0, -1, v0
+    v_mul_u32_u24 v0, 4, v0
+
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_mov_b32 		m0, 0x0
+    s_cbranch_scc1  L_SAVE_LDS_LOOP_W64
+
+  L_SAVE_LDS_LOOP_W32:									
+	if (SAVE_LDS)
+    ds_read_b32 v1, v0
+    s_waitcnt 0														    //ensure data ready
+    buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	//buffer_store_lds_dword	s_save_buf_rsrc0, s_save_mem_offset lds:1               //save lds to memory doesn't exist in 10
+	end
+	s_add_u32		m0, m0, 128															//every buffer_store_lds does 128 bytes
+	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128							//mem offset increased by 128 bytes
+    v_add_nc_u32    v0, v0, 128
+	s_cmp_lt_u32	m0, s_save_alloc_size												//scc=(m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1  L_SAVE_LDS_LOOP_W32													//LDS save is complete?
+    s_branch        L_SAVE_SGPR
+
+  L_SAVE_LDS_LOOP_W64:									
+	if (SAVE_LDS)
+    ds_read_b32 v1, v0
+    s_waitcnt 0														    //ensure data ready
+    buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	//buffer_store_lds_dword	s_save_buf_rsrc0, s_save_mem_offset lds:1               //save lds to memory doesn't exist in 10
+	end
+	s_add_u32		m0, m0, 256															//every buffer_store_lds does 256 bytes
+	s_add_u32		s_save_mem_offset, s_save_mem_offset, 256							//mem offset increased by 256 bytes
+    v_add_nc_u32    v0, v0, 256
+	s_cmp_lt_u32	m0, s_save_alloc_size												//scc=(m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1  L_SAVE_LDS_LOOP_W64													//LDS save is complete?
+   
+	
+	/*      	save SGPRs	    */
+	//////////////////////////////
+	//s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE) 				//spgr_size
+	//s_add_u32 		s_save_alloc_size, s_save_alloc_size, 1
+	//s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 4 						//Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value) 
+	//s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 3 						//In gfx10, Number of SGPRs = (sgpr_size + 1) * 8   (non-zero value) 
+  L_SAVE_SGPR:
+    //need to look at it is wave32 or wave64
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_SAVE_SGPR_VMEM_WAVE64
+    if (SGPR_SAVE_USE_SQC)
+		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 2					//NUM_RECORDS in bytes
+    else
+        s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 7					//NUM_RECORDS in bytes (32 threads)
+    end
+    s_branch    L_SAVE_SGPR_CONT    
+  L_SAVE_SGPR_VMEM_WAVE64:
+	if (SGPR_SAVE_USE_SQC)
+		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 2					//NUM_RECORDS in bytes 
+	else
+		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 8					//NUM_RECORDS in bytes (64 threads)
+	end
+  L_SAVE_SGPR_CONT:
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+	
+	//s_mov_b32 		m0, 0x0 														//SGPR initial index value =0		
+    //s_nop           0x0                                                             //Manually inserted wait states
+	
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    
+    s_mov_b32 		m0, 0x0 														//SGPR initial index value =0		
+    s_nop           0x0                                                             //Manually inserted wait states
+
+    s_cbranch_scc1  L_SAVE_SGPR_LOOP_WAVE64
+
+  L_SAVE_SGPR_LOOP_WAVE32: 										
+	s_movrels_b32 	s0, s0 															//s0 = s[0+m0]
+    //zhenxu, adding one more argument to save sgpr function, this is only for vmem, using sqc is not change    
+	write_sgpr_to_mem_wave32(s0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)							//PV: the best performance should be using s_buffer_store_dwordx4
+	s_add_u32		m0, m0, 1														//next sgpr index
+	s_cmp_lt_u32 	m0, s_sgpr_save_num 											//scc = (m0 < s_sgpr_save_num) ? 1 : 0
+	s_cbranch_scc1 	L_SAVE_SGPR_LOOP_WAVE32												//SGPR save is complete?
+    s_branch    L_SAVE_HWREG
+
+  L_SAVE_SGPR_LOOP_WAVE64: 										
+	s_movrels_b32 	s0, s0 															//s0 = s[0+m0]
+    //zhenxu, adding one more argument to save sgpr function, this is only for vmem, using sqc is not change    
+	write_sgpr_to_mem_wave64(s0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)							//PV: the best performance should be using s_buffer_store_dwordx4
+	s_add_u32		m0, m0, 1														//next sgpr index
+	s_cmp_lt_u32 	m0, s_sgpr_save_num 											//scc = (m0 < s_sgpr_save_num) ? 1 : 0
+	s_cbranch_scc1 	L_SAVE_SGPR_LOOP_WAVE64												//SGPR save is complete?
+
+	
+	/* 		save HW registers	*/
+	//////////////////////////////
+  L_SAVE_HWREG:
+    s_mov_b32		s_save_buf_rsrc2, 0x4								//NUM_RECORDS	in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_SAVE_HWREG_WAVE64
+	
+	write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//M0
+
+	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))      
+		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
+		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
+	end
+
+	write_sgpr_to_mem_wave32(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//PC
+	write_sgpr_to_mem_wave32(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+	write_sgpr_to_mem_wave32(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//EXEC
+	write_sgpr_to_mem_wave32(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+	write_sgpr_to_mem_wave32(s_save_status, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//STATUS 
+	
+	//s_save_trapsts conflicts with s_save_alloc_size
+	s_getreg_b32    s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	write_sgpr_to_mem_wave32(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//TRAPSTS
+	
+	//write_sgpr_to_mem_wave32(s_save_xnack_mask_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_LO
+	write_sgpr_to_mem_wave32(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_HI
+	
+	//use s_save_tmp would introduce conflict here between s_save_tmp and s_save_buf_rsrc2
+	s_getreg_b32 	s_save_m0, hwreg(HW_REG_MODE)																						//MODE
+	write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+    if(SAVE_RESTORE_HWID_DDID)
+    s_getreg_b32 	s_save_m0, hwreg(HW_REG_HW_ID1)																						//HW_ID1, handler records the SE/SA/WGP/SIMD/wave of the original wave
+    write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+    end
+    s_branch   L_S_PGM_END_SAVED
+
+  L_SAVE_HWREG_WAVE64:
+    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//M0
+
+	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))      
+		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
+		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
+	end
+
+	write_sgpr_to_mem_wave64(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//PC
+	write_sgpr_to_mem_wave64(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+	write_sgpr_to_mem_wave64(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//EXEC
+	write_sgpr_to_mem_wave64(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+	write_sgpr_to_mem_wave64(s_save_status, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//STATUS 
+	
+	//s_save_trapsts conflicts with s_save_alloc_size
+	s_getreg_b32    s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	write_sgpr_to_mem_wave64(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//TRAPSTS
+	
+	//write_sgpr_to_mem_wave64(s_save_xnack_mask_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_LO
+	write_sgpr_to_mem_wave64(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_HI
+	
+	//use s_save_tmp would introduce conflict here between s_save_tmp and s_save_buf_rsrc2
+	s_getreg_b32 	s_save_m0, hwreg(HW_REG_MODE)																						//MODE
+	write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+
+
+    if(SAVE_RESTORE_HWID_DDID)
+    s_getreg_b32 	s_save_m0, hwreg(HW_REG_HW_ID1)																						//HW_ID1, handler records the SE/SA/WGP/SIMD/wave of the original wave
+    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
+
+	/* 		save DDID	*/
+	//////////////////////////////
+  L_SAVE_DDID:
+    //EXEC has been saved, no vector inst following
+    s_mov_b32	exec_lo, 0x80000000    //Set MSB to 1. Cleared when draw index is returned
+    s_sendmsg sendmsg(MSG_GET_DDID)
+
+  L_WAIT_DDID_LOOP:    
+    s_nop		7			// sleep a bit
+    s_bitcmp0_b32 exec_lo, 31	// test to see if MSB is cleared, meaning done
+    s_cbranch_scc0	L_WAIT_DDID_LOOP
+
+    s_mov_b32	s_save_m0, exec_lo
+
+
+    s_mov_b32		s_save_buf_rsrc2, 0x4								//NUM_RECORDS	in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
+	end
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_SAVE_DDID_WAVE64
+
+    write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF) 
+
+  L_SAVE_DDID_WAVE64:
+    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF) 
+
+    end
+   
+  L_S_PGM_END_SAVED:
+	/*     S_PGM_END_SAVED  */    							//FIXME  graphics ONLY
+	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_NORMAL_EXIT))	
+		s_and_b32 s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
+		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
+		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
+		s_rfe_b64 s_save_pc_lo                              //Return to the main shader program
+	else
+	end
+
+	
+    s_branch	L_END_PGM
+	
+
+				
+/**************************************************************************/
+/*                     	restore routine						              */
+/**************************************************************************/
+
+L_RESTORE:
+    /*      Setup Resource Contants    */
+    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
+		//calculate wd_addr using absolute thread id
+		v_readlane_b32 s_restore_tmp, v9, 0
+        //determine it is wave32 or wave64
+        s_getreg_b32 	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE) //change to ttmp13
+        s_cmp_eq_u32    s_restore_size, 0
+        s_cbranch_scc1  L_RESTORE_WAVE32
+        s_lshr_b32 s_restore_tmp, s_restore_tmp, 6 //SAVE WAVE64
+        s_branch    L_RESTORE_CON
+    L_RESTORE_WAVE32:
+        s_lshr_b32 s_restore_tmp, s_restore_tmp, 5 //SAVE WAVE32
+    L_RESTORE_CON:
+		s_mul_i32 s_restore_tmp, s_restore_tmp, WAVE_SPACE
+		s_add_i32 s_restore_spi_init_lo, s_restore_tmp, WG_BASE_ADDR_LO
+		s_mov_b32 s_restore_spi_init_hi, WG_BASE_ADDR_HI
+		s_and_b32 s_restore_spi_init_hi, s_restore_spi_init_hi, CTX_RESTORE_CONTROL	
+	else
+	end
+	
+    s_mov_b32		s_restore_buf_rsrc0, 	s_restore_spi_init_lo															//base_addr_lo
+	s_and_b32		s_restore_buf_rsrc1, 	s_restore_spi_init_hi, 0x0000FFFF												//base_addr_hi
+	s_or_b32		s_restore_buf_rsrc1, 	s_restore_buf_rsrc1,  S_RESTORE_BUF_RSRC_WORD1_STRIDE
+    s_mov_b32       s_restore_buf_rsrc2,   	0                                               								//NUM_RECORDS initial value = 0 (in bytes)
+	s_mov_b32		s_restore_buf_rsrc3, 	S_RESTORE_BUF_RSRC_WORD3_MISC
+	s_and_b32		s_restore_tmp,         	s_restore_spi_init_hi, S_RESTORE_SPI_INIT_ATC_MASK		
+	s_lshr_b32		s_restore_tmp,  		s_restore_tmp, (S_RESTORE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)		//get ATC bit into position
+	s_or_b32		s_restore_buf_rsrc3, 	s_restore_buf_rsrc3,  s_restore_tmp												//or ATC
+	s_and_b32		s_restore_tmp,         	s_restore_spi_init_hi, S_RESTORE_SPI_INIT_MTYPE_MASK		
+	s_lshr_b32		s_restore_tmp,  		s_restore_tmp, (S_RESTORE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)	//get MTYPE bits into position
+	s_or_b32		s_restore_buf_rsrc3, 	s_restore_buf_rsrc3,  s_restore_tmp												//or MTYPE
+    //determine it is wave32 or wave64
+    s_getreg_b32 	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
+    s_or_b32        s_restore_size, s_restore_spi_init_hi,    s_restore_size                                             //share s_wave_size with exec_hi
+	
+	/* 		global mem offset			*/
+	s_mov_b32		s_restore_mem_offset, 0x0								//mem offset initial value = 0
+
+        /*      	restore VGPRs	    */
+	//////////////////////////////
+  L_RESTORE_VGPR:
+  
+ 	s_mov_b32		exec_lo, 0xFFFFFFFF 													//need every thread from now on   //be consistent with SAVE although can be moved ahead
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_ENABLE_RESTORE_VGPR_EXEC_HI   
+    s_mov_b32		exec_hi, 0x00000000
+    s_branch        L_RESTORE_VGPR_NORMAL
+  L_ENABLE_RESTORE_VGPR_EXEC_HI:
+	s_mov_b32		exec_hi, 0xFFFFFFFF
+  L_RESTORE_VGPR_NORMAL:	
+	s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) 	//vpgr_size
+	s_add_u32 		s_restore_alloc_size, s_restore_alloc_size, 1
+	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 2 							//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
+    //determine it is wave32 or wave64
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_RESTORE_VGPR_WAVE64
+
+    s_lshl_b32		s_restore_buf_rsrc2,  s_restore_alloc_size, 7						    //NUM_RECORDS in bytes (32 threads*4)
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end	
+
+	s_mov_b32		s_restore_mem_offset_save, s_restore_mem_offset							// restore start with v1, v0 will be the last
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128
+    s_mov_b32 		m0, 1 																	//VGPR initial index value = 1
+	//s_set_gpr_idx_on  m0, 0x8																//M0[7:0] = M0[7:0] and M0[15:12] = 0x8
+    //s_add_u32		s_restore_alloc_size, s_restore_alloc_size, 0x8000						//add 0x8000 since we compare m0 against it later, might not need this in gfx10	
+
+  L_RESTORE_VGPR_WAVE32_LOOP: 										
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1	
+	end
+	s_waitcnt		vmcnt(0)																//ensure data ready
+	v_movreld_b32		v0, v0																	//v[0+m0] = v0
+    s_add_u32		m0, m0, 1																//next vgpr index
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128							//every buffer_load_dword does 128 bytes
+	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_RESTORE_VGPR_WAVE32_LOOP														//VGPR restore (except v0) is complete?
+	//s_set_gpr_idx_off
+																							/* VGPR restore on v0 */
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save	slc:1 glc:1	
+	end
+
+    s_branch    L_RESTORE_LDS
+
+  L_RESTORE_VGPR_WAVE64:
+    s_lshl_b32		s_restore_buf_rsrc2,  s_restore_alloc_size, 8						    //NUM_RECORDS in bytes (64 threads*4)
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end	
+
+	s_mov_b32		s_restore_mem_offset_save, s_restore_mem_offset							// restore start with v1, v0 will be the last
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256
+    s_mov_b32 		m0, 1 																	//VGPR initial index value = 1
+  L_RESTORE_VGPR_WAVE64_LOOP: 										
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1	
+	end
+	s_waitcnt		vmcnt(0)																//ensure data ready
+	v_movreld_b32		v0, v0																	//v[0+m0] = v0
+    s_add_u32		m0, m0, 1																//next vgpr index
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256							//every buffer_load_dword does 256 bytes
+	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_RESTORE_VGPR_WAVE64_LOOP														//VGPR restore (except v0) is complete?
+	//s_set_gpr_idx_off
+    //
+    //Below part will be the restore shared vgpr part (new for gfx10)
+    s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE) 			//shared_vgpr_size
+    s_and_b32		s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF				//shared_vgpr_size is zero?
+    s_cbranch_scc0	L_RESTORE_V0													    //no shared_vgpr used? jump to L_SAVE_LDS
+    s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 3 						//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
+    //m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
+    //restore shared_vgpr will start from the index of m0
+    s_add_u32       s_restore_alloc_size, s_restore_alloc_size, m0
+    s_mov_b32		exec_lo, 0xFFFFFFFF
+    s_mov_b32		exec_hi, 0x00000000
+    L_RESTORE_SHARED_VGPR_WAVE64_LOOP: 
+    buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1
+    s_waitcnt		vmcnt(0)																//ensure data ready
+	v_movreld_b32		v0, v0																	//v[0+m0] = v0
+    s_add_u32		m0, m0, 1																//next vgpr index
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128							//every buffer_load_dword does 256 bytes
+	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_RESTORE_SHARED_VGPR_WAVE64_LOOP														//VGPR restore (except v0) is complete?
+
+    s_mov_b32 exec_hi, 0xFFFFFFFF                                                           //restore back exec_hi before restoring V0!!
+	
+    /* VGPR restore on v0 */
+  L_RESTORE_V0:
+    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
+		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+    else
+		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save	slc:1 glc:1	
+	end
+
+
+    /*      	restore LDS	    */
+	//////////////////////////////
+  L_RESTORE_LDS:
+
+    //Only need to check the first wave    
+	/*      the first wave in the threadgroup    */
+	s_and_b32		s_restore_tmp, s_restore_size, S_RESTORE_SPI_INIT_FIRST_WAVE_MASK			
+	s_cbranch_scc0	L_RESTORE_SGPR
+	
+    s_mov_b32		exec_lo, 0xFFFFFFFF 													//need every thread from now on   //be consistent with SAVE although can be moved ahead
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_ENABLE_RESTORE_LDS_EXEC_HI   
+    s_mov_b32		exec_hi, 0x00000000
+    s_branch        L_RESTORE_LDS_NORMAL
+  L_ENABLE_RESTORE_LDS_EXEC_HI:
+	s_mov_b32		exec_hi, 0xFFFFFFFF
+  L_RESTORE_LDS_NORMAL:	
+	s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE) 				//lds_size
+	s_and_b32		s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF					//lds_size is zero?
+	s_cbranch_scc0	L_RESTORE_SGPR															//no lds used? jump to L_RESTORE_VGPR
+	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 6 							//LDS size in dwords = lds_size * 64dw
+	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 2 							//LDS size in bytes
+	s_mov_b32		s_restore_buf_rsrc2,	s_restore_alloc_size							//NUM_RECORDS in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end
+
+    s_and_b32       m0, s_wave_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_mov_b32 		m0, 0x0
+    s_cbranch_scc1  L_RESTORE_LDS_LOOP_W64
+
+  L_RESTORE_LDS_LOOP_W32:									
+	if (SAVE_LDS)
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1
+    s_waitcnt 0
+	end
+    s_add_u32		m0, m0, 128																//every buffer_load_dword does 256 bytes
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128						//mem offset increased by 256 bytes
+	s_cmp_lt_u32	m0, s_restore_alloc_size												//scc=(m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1  L_RESTORE_LDS_LOOP_W32														//LDS restore is complete?
+    s_branch        L_RESTORE_SGPR
+
+  L_RESTORE_LDS_LOOP_W64:									
+	if (SAVE_LDS)
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1
+    s_waitcnt 0
+	end
+    s_add_u32		m0, m0, 256																//every buffer_load_dword does 256 bytes
+	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256							//mem offset increased by 256 bytes
+	s_cmp_lt_u32	m0, s_restore_alloc_size												//scc=(m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1  L_RESTORE_LDS_LOOP_W64														//LDS restore is complete?
+
+	
+    /*      	restore SGPRs	    */
+	//////////////////////////////
+	//s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE) 				//spgr_size
+	//s_add_u32 		s_restore_alloc_size, s_restore_alloc_size, 1
+	//s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 4 							//Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
+	//s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 3 							//Number of SGPRs = (sgpr_size + 1) * 8   (non-zero value)
+  L_RESTORE_SGPR:
+    //need to look at it is wave32 or wave64
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_RESTORE_SGPR_VMEM_WAVE64
+	if (SGPR_SAVE_USE_SQC)
+		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 2						//NUM_RECORDS in bytes 
+	else
+        s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 7						//NUM_RECORDS in bytes (32 threads)
+    end
+    s_branch        L_RESTORE_SGPR_CONT
+  L_RESTORE_SGPR_VMEM_WAVE64:
+    if (SGPR_SAVE_USE_SQC)
+		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 2						//NUM_RECORDS in bytes 
+	else
+		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 8						//NUM_RECORDS in bytes (64 threads)
+	end
+
+  L_RESTORE_SGPR_CONT:
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end
+
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_RESTORE_SGPR_WAVE64
+
+    read_sgpr_from_mem_wave32(s_restore_tmp, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)		//save s0 to s_restore_tmp
+	s_mov_b32 		m0, 0x1
+
+  L_RESTORE_SGPR_LOOP_WAVE32:
+    read_sgpr_from_mem_wave32(s0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)															//PV: further performance improvement can be made
+	s_waitcnt		lgkmcnt(0)																//ensure data ready
+	s_movreld_b32 	s0, s0                                                                  //s[0+m0] = s0
+    s_nop 0                                                                                 // hazard SALU M0=> S_MOVREL
+	s_add_u32		m0, m0, 1																//next sgpr index
+	s_cmp_lt_u32 	m0, s_sgpr_save_num												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_RESTORE_SGPR_LOOP_WAVE32														//SGPR restore (except s0) is complete?
+	s_mov_b32		s0, s_restore_tmp															/* SGPR restore on s0 */
+    s_branch        L_RESTORE_HWREG
+  
+  L_RESTORE_SGPR_WAVE64:
+	read_sgpr_from_mem_wave64(s_restore_tmp, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)		//save s0 to s_restore_tmp
+	s_mov_b32 		m0, 0x1																				//SGPR initial index value =1	//go on with with s1
+	
+  L_RESTORE_SGPR_LOOP_WAVE64: 																					
+	read_sgpr_from_mem_wave64(s0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)															//PV: further performance improvement can be made
+	s_waitcnt		lgkmcnt(0)																//ensure data ready
+	s_movreld_b32 	s0, s0                                                                  //s[0+m0] = s0
+    s_nop 0                                                                                 // hazard SALU M0=> S_MOVREL
+	s_add_u32		m0, m0, 1																//next sgpr index
+	s_cmp_lt_u32 	m0, s_sgpr_save_num												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1 	L_RESTORE_SGPR_LOOP_WAVE64														//SGPR restore (except s0) is complete?
+	s_mov_b32		s0, s_restore_tmp															/* SGPR restore on s0 */
+
+	
+    /* 		restore HW registers	*/
+	//////////////////////////////
+  L_RESTORE_HWREG:
+    s_mov_b32		s_restore_buf_rsrc2, 0x4												//NUM_RECORDS	in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end
+
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_RESTORE_HWREG_WAVE64
+
+    read_sgpr_from_mem_wave32(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//M0
+	read_sgpr_from_mem_wave32(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//PC
+	read_sgpr_from_mem_wave32(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
+	read_sgpr_from_mem_wave32(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//EXEC
+	read_sgpr_from_mem_wave32(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
+	read_sgpr_from_mem_wave32(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//STATUS
+	read_sgpr_from_mem_wave32(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//TRAPSTS
+    //read_sgpr_from_mem_wave32(xnack_mask_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_LO
+	//read_sgpr_from_mem_wave32(xnack_mask_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_HI
+    read_sgpr_from_mem_wave32(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK
+	read_sgpr_from_mem_wave32(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//MODE
+    if(SAVE_RESTORE_HWID_DDID)
+    read_sgpr_from_mem_wave32(s_restore_hwid1, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//HW_ID1
+    end
+    s_branch        L_RESTORE_HWREG_FINISH
+
+  L_RESTORE_HWREG_WAVE64:
+	read_sgpr_from_mem_wave64(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//M0
+	read_sgpr_from_mem_wave64(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//PC
+	read_sgpr_from_mem_wave64(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
+	read_sgpr_from_mem_wave64(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//EXEC
+	read_sgpr_from_mem_wave64(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
+	read_sgpr_from_mem_wave64(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//STATUS
+	read_sgpr_from_mem_wave64(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//TRAPSTS
+    //read_sgpr_from_mem_wave64(xnack_mask_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_LO
+	//read_sgpr_from_mem_wave64(xnack_mask_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_HI
+    read_sgpr_from_mem_wave64(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK
+	read_sgpr_from_mem_wave64(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//MODE
+    if(SAVE_RESTORE_HWID_DDID)
+    read_sgpr_from_mem_wave64(s_restore_hwid1, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//HW_ID1
+    end
+  L_RESTORE_HWREG_FINISH:
+	s_waitcnt		lgkmcnt(0)																						//from now on, it is safe to restore STATUS and IB_STS
+  
+
+
+    if(SAVE_RESTORE_HWID_DDID)
+  L_RESTORE_DDID:
+    s_mov_b32      m0, s_restore_hwid1                                                      //virture ttrace support: The save-context handler records the SE/SA/WGP/SIMD/wave of the original wave
+    s_ttracedata                                                                            //and then can output it as SHADER_DATA to ttrace on restore to provide a correlation across the save-restore
+                                    
+    s_mov_b32		s_restore_buf_rsrc2, 0x4												//NUM_RECORDS	in bytes
+	if (SWIZZLE_EN)
+		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
+	else
+		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
+	end
+
+    s_and_b32       m0, s_restore_size, 1
+    s_cmp_eq_u32    m0, 1
+    s_cbranch_scc1  L_RESTORE_DDID_WAVE64
+
+    read_sgpr_from_mem_wave32(s_restore_ddid, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)	
+    s_branch        L_RESTORE_DDID_FINISH
+  L_RESTORE_DDID_WAVE64:
+    read_sgpr_from_mem_wave64(s_restore_ddid, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)	
+
+  L_RESTORE_DDID_FINISH:
+    s_waitcnt		lgkmcnt(0)
+    //s_mov_b32      m0, s_restore_ddid
+    //s_ttracedata   
+    if (RESTORE_DDID_IN_SGPR18)
+        s_mov_b32   s18, s_restore_ddid
+	end	
+    
+    end   
+
+	s_and_b32 s_restore_pc_hi, s_restore_pc_hi, 0x0000ffff    	//pc[47:32]        //Do it here in order not to affect STATUS
+
+	//for normal save & restore, the saved PC points to the next inst to execute, no adjustment needs to be made, otherwise:
+	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
+		s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 8            //pc[31:0]+8	  //two back-to-back s_trap are used (first for save and second for restore)
+		s_addc_u32	s_restore_pc_hi, s_restore_pc_hi, 0x0		 //carry bit over
+	end	
+	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_RESTORE_NORMAL))	      
+		s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 4            //pc[31:0]+4     // save is hack through s_trap but restore is normal
+		s_addc_u32	s_restore_pc_hi, s_restore_pc_hi, 0x0		 //carry bit over
+	end
+	
+	s_mov_b32 		m0, 		s_restore_m0
+	s_mov_b32 		exec_lo, 	s_restore_exec_lo
+	s_mov_b32 		exec_hi, 	s_restore_exec_hi
+	
+	s_and_b32		s_restore_m0, SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK, s_restore_trapsts
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE), s_restore_m0
+    s_setreg_b32    hwreg(HW_REG_SHADER_XNACK_MASK), s_restore_xnack_mask         //restore xnack_mask
+	s_and_b32		s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK, s_restore_trapsts
+	s_lshr_b32		s_restore_m0, s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE), s_restore_m0
+	//s_setreg_b32 	hwreg(HW_REG_TRAPSTS), 	s_restore_trapsts      //don't overwrite SAVECTX bit as it may be set through external SAVECTX during restore
+	s_setreg_b32 	hwreg(HW_REG_MODE), 	s_restore_mode
+	//reuse s_restore_m0 as a temp register
+	s_and_b32		s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_RCNT_MASK
+	s_lshr_b32		s_restore_m0, s_restore_m0, S_SAVE_PC_HI_RCNT_SHIFT
+	s_lshl_b32		s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_RCNT_SHIFT
+	s_mov_b32		s_restore_tmp, 0x0																				//IB_STS is zero
+	s_or_b32		s_restore_tmp, s_restore_tmp, s_restore_m0
+	s_and_b32		s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_FIRST_REPLAY_MASK
+	s_lshr_b32		s_restore_m0, s_restore_m0, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+	s_lshl_b32		s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT
+	s_or_b32		s_restore_tmp, s_restore_tmp, s_restore_m0
+    s_and_b32       s_restore_m0, s_restore_status, SQ_WAVE_STATUS_INST_ATC_MASK 
+    s_lshr_b32		s_restore_m0, s_restore_m0, SQ_WAVE_STATUS_INST_ATC_SHIFT
+	s_setreg_b32 	hwreg(HW_REG_IB_STS), 	s_restore_tmp
+	s_setreg_b32 	hwreg(HW_REG_STATUS), 	s_restore_status
+
+	s_barrier													//barrier to ensure the readiness of LDS before access attemps from any other wave in the same TG //FIXME not performance-optimal at this time
+	
+	
+//	s_rfe_b64 s_restore_pc_lo                              		//Return to the main shader program and resume execution
+    s_rfe_b64  s_restore_pc_lo            // s_restore_m0[0] is used to set STATUS.inst_atc 
+
+
+/**************************************************************************/
+/*                     	the END								              */
+/**************************************************************************/	
+L_END_PGM:	
+	s_endpgm
+	
+end	
+
+
+/**************************************************************************/
+/*                     	the helper functions							  */
+/**************************************************************************/
+function write_sgpr_to_mem_wave32(s, s_rsrc, s_mem_offset, use_sqc, use_mtbuf)
+	if (use_sqc)
+		s_mov_b32 exec_lo, m0					//assuming exec_lo is not needed anymore from this point on
+		s_mov_b32 m0, s_mem_offset
+		s_buffer_store_dword s, s_rsrc, m0		glc:1	
+		s_add_u32		s_mem_offset, s_mem_offset, 4
+		s_mov_b32	m0, exec_lo
+    elsif (use_mtbuf)
+        v_mov_b32	v0,	s
+        tbuffer_store_format_x v0, v0, s_rsrc, s_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+		s_add_u32		s_mem_offset, s_mem_offset, 128
+    else 
+        v_mov_b32	v0,	s
+		buffer_store_dword	v0, v0, s_rsrc, s_mem_offset	slc:1 glc:1
+        s_add_u32		s_mem_offset, s_mem_offset, 128
+	end
+end
+
+function write_sgpr_to_mem_wave64(s, s_rsrc, s_mem_offset, use_sqc, use_mtbuf)
+	if (use_sqc)
+		s_mov_b32 exec_lo, m0					//assuming exec_lo is not needed anymore from this point on
+		s_mov_b32 m0, s_mem_offset
+		s_buffer_store_dword s, s_rsrc, m0		glc:1	
+		s_add_u32		s_mem_offset, s_mem_offset, 4
+		s_mov_b32	m0, exec_lo
+    elsif (use_mtbuf)
+        v_mov_b32	v0,	s
+        tbuffer_store_format_x v0, v0, s_rsrc, s_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
+		s_add_u32		s_mem_offset, s_mem_offset, 256
+    else 
+        v_mov_b32	v0,	s
+		buffer_store_dword	v0, v0, s_rsrc, s_mem_offset	slc:1 glc:1
+        s_add_u32		s_mem_offset, s_mem_offset, 256
+	end
+end
+
+function read_sgpr_from_mem_wave32(s, s_rsrc, s_mem_offset, use_sqc)
+	s_buffer_load_dword s, s_rsrc, s_mem_offset		glc:1
+	if (use_sqc)
+		s_add_u32		s_mem_offset, s_mem_offset, 4
+	else
+        s_add_u32		s_mem_offset, s_mem_offset, 128
+	end
+end
+
+function read_sgpr_from_mem_wave64(s, s_rsrc, s_mem_offset, use_sqc)
+	s_buffer_load_dword s, s_rsrc, s_mem_offset		glc:1
+	if (use_sqc)
+		s_add_u32		s_mem_offset, s_mem_offset, 4
+	else
+        s_add_u32		s_mem_offset, s_mem_offset, 256
+	end
+end
+

drivers/gpu/drm/amd/amdkfd/kfd_crat.c

@@ -138,6 +138,8 @@ static struct kfd_gpu_cache_info carrizo_cache_info[] = {
 /* TODO - check & update Vega10 cache details */
 #define vega10_cache_info carrizo_cache_info
 #define raven_cache_info carrizo_cache_info
+/* TODO - check & update Navi10 cache details */
+#define navi10_cache_info carrizo_cache_info
 
 static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev,
 		struct crat_subtype_computeunit *cu)
@@ -666,6 +668,9 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev,
 	case CHIP_RAVEN:
 		pcache_info = raven_cache_info;
 		num_of_cache_types = ARRAY_SIZE(raven_cache_info);
+	case CHIP_NAVI10:
+		pcache_info = navi10_cache_info;
+		num_of_cache_types = ARRAY_SIZE(navi10_cache_info);
 		break;
 	default:
 		return -EINVAL;

drivers/gpu/drm/amd/amdkfd/kfd_device.c

@@ -317,6 +317,23 @@ static const struct kfd_device_info vega20_device_info = {
 	.num_sdma_queues_per_engine = 8,
 };
 
+static const struct kfd_device_info navi10_device_info = {
+	.asic_family = CHIP_NAVI10,
+	.max_pasid_bits = 16,
+	.max_no_of_hqd  = 24,
+	.doorbell_size  = 8,
+	.ih_ring_entry_size = 8 * sizeof(uint32_t),
+	.event_interrupt_class = &event_interrupt_class_v9,
+	.num_of_watch_points = 4,
+	.mqd_size_aligned = MQD_SIZE_ALIGNED,
+	.needs_iommu_device = false,
+	.supports_cwsr = true,
+	.needs_pci_atomics = false,
+	.num_sdma_engines = 2,
+	.num_xgmi_sdma_engines = 0,
+	.num_sdma_queues_per_engine = 8,
+};
+
 struct kfd_deviceid {
 	unsigned short did;
 	const struct kfd_device_info *device_info;
@@ -434,7 +451,13 @@ static const struct kfd_deviceid supported_devices[] = {
 	{ 0x66a3, &vega20_device_info },	/* Vega20 */
 	{ 0x66a4, &vega20_device_info },	/* Vega20 */
 	{ 0x66a7, &vega20_device_info },	/* Vega20 */
-	{ 0x66af, &vega20_device_info }		/* Vega20 */
+	{ 0x66af, &vega20_device_info },	/* Vega20 */
+	/* Navi10 */
+	{ 0x7310, &navi10_device_info },	/* Navi10 */
+	{ 0x7312, &navi10_device_info },	/* Navi10 */
+	{ 0x7318, &navi10_device_info },	/* Navi10 */
+	{ 0x731a, &navi10_device_info },	/* Navi10 */
+	{ 0x731f, &navi10_device_info },	/* Navi10 */
 };
 
 static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
@@ -516,10 +539,14 @@ static void kfd_cwsr_init(struct kfd_dev *kfd)
 			BUILD_BUG_ON(sizeof(cwsr_trap_gfx8_hex) > PAGE_SIZE);
 			kfd->cwsr_isa = cwsr_trap_gfx8_hex;
 			kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex);
-		} else {
+		} else if (kfd->device_info->asic_family < CHIP_NAVI10) {
 			BUILD_BUG_ON(sizeof(cwsr_trap_gfx9_hex) > PAGE_SIZE);
 			kfd->cwsr_isa = cwsr_trap_gfx9_hex;
 			kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx9_hex);
+		} else {
+			BUILD_BUG_ON(sizeof(cwsr_trap_gfx10_hex) > PAGE_SIZE);
+			kfd->cwsr_isa = cwsr_trap_gfx10_hex;
+			kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx10_hex);
 		}
 
 		kfd->cwsr_enabled = true;

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -1264,6 +1264,7 @@ static int map_queues_cpsch(struct device_queue_manager *dqm)
 		return 0;
 
 	retval = pm_send_runlist(&dqm->packets, &dqm->queues);
+	pr_debug("%s sent runlist\n", __func__);
 	if (retval) {
 		pr_err("failed to execute runlist\n");
 		return retval;
@@ -1301,7 +1302,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm,
 				KFD_FENCE_COMPLETED);
 	/* should be timed out */
 	retval = amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED,
-				QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS);
+				queue_preemption_timeout_ms);
 	if (retval)
 		return retval;
 
@@ -1785,6 +1786,9 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 	case CHIP_RAVEN:
 		device_queue_manager_init_v9(&dqm->asic_ops);
 		break;
+	case CHIP_NAVI10:
+		device_queue_manager_init_v10_navi10(&dqm->asic_ops);
+		break;
 	default:
 		WARN(1, "Unexpected ASIC family %u",
 		     dev->device_info->asic_family);
@@ -1876,7 +1880,8 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data)
 	int r = 0;
 
 	r = dqm->dev->kfd2kgd->hqd_dump(dqm->dev->kgd,
-		KFD_CIK_HIQ_PIPE, KFD_CIK_HIQ_QUEUE, &dump, &n_regs);
+					KFD_CIK_HIQ_PIPE, KFD_CIK_HIQ_QUEUE,
+					&dump, &n_regs);
 	if (!r) {
 		seq_printf(m, "  HIQ on MEC %d Pipe %d Queue %d\n",
 			   KFD_CIK_HIQ_PIPE/get_pipes_per_mec(dqm)+1,

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h

@@ -31,8 +31,6 @@
 #include "kfd_priv.h"
 #include "kfd_mqd_manager.h"
 
-#define KFD_UNMAP_LATENCY_MS			(4000)
-#define QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS (2 * KFD_UNMAP_LATENCY_MS + 1000)
 
 struct device_process_node {
 	struct qcm_process_device *qpd;
@@ -212,6 +210,8 @@ void device_queue_manager_init_vi_tonga(
 		struct device_queue_manager_asic_ops *asic_ops);
 void device_queue_manager_init_v9(
 		struct device_queue_manager_asic_ops *asic_ops);
+void device_queue_manager_init_v10_navi10(
+		struct device_queue_manager_asic_ops *asic_ops);
 void program_sh_mem_settings(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd);
 unsigned int get_queues_num(struct device_queue_manager *dqm);

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c

+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_device_queue_manager.h"
+#include "navi10_enum.h"
+#include "gc/gc_10_1_0_offset.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+
+static int update_qpd_v10(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd);
+static void init_sdma_vm_v10(struct device_queue_manager *dqm, struct queue *q,
+			    struct qcm_process_device *qpd);
+
+void device_queue_manager_init_v10_navi10(
+	struct device_queue_manager_asic_ops *asic_ops)
+{
+	asic_ops->update_qpd = update_qpd_v10;
+	asic_ops->init_sdma_vm = init_sdma_vm_v10;
+	asic_ops->mqd_manager_init = mqd_manager_init_v10;
+}
+
+static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
+{
+	uint32_t shared_base = pdd->lds_base >> 48;
+	uint32_t private_base = pdd->scratch_base >> 48;
+
+	return (shared_base << SH_MEM_BASES__SHARED_BASE__SHIFT) |
+		private_base;
+}
+
+static int update_qpd_v10(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd)
+{
+	struct kfd_process_device *pdd;
+
+	pdd = qpd_to_pdd(qpd);
+
+	/* check if sh_mem_config register already configured */
+	if (qpd->sh_mem_config == 0) {
+		qpd->sh_mem_config =
+				SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+					SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
+#if 0
+		/* TODO:
+		 *    This shouldn't be an issue with Navi10.  Verify.
+		 */
+		if (vega10_noretry)
+			qpd->sh_mem_config |=
+				1 << SH_MEM_CONFIG__RETRY_DISABLE__SHIFT;
+#endif
+
+		qpd->sh_mem_ape1_limit = 0;
+		qpd->sh_mem_ape1_base = 0;
+	}
+
+	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(pdd);
+
+	pr_debug("sh_mem_bases 0x%X\n", qpd->sh_mem_bases);
+
+	return 0;
+}
+
+static void init_sdma_vm_v10(struct device_queue_manager *dqm, struct queue *q,
+			    struct qcm_process_device *qpd)
+{
+	/* Not needed on SDMAv4 onwards any more */
+	q->properties.sdma_vm_addr = 0;
+}

drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c

@@ -405,6 +405,7 @@ int kfd_init_apertures(struct kfd_process *process)
 			case CHIP_VEGA12:
 			case CHIP_VEGA20:
 			case CHIP_RAVEN:
+			case CHIP_NAVI10:
 				kfd_init_apertures_v9(pdd, id);
 				break;
 			default:

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c

@@ -332,6 +332,9 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 	case CHIP_RAVEN:
 		kernel_queue_init_v9(&kq->ops_asic_specific);
 		break;
+	case CHIP_NAVI10:
+		kernel_queue_init_v10(&kq->ops_asic_specific);
+		break;
 	default:
 		WARN(1, "Unexpected ASIC family %u",
 		     dev->device_info->asic_family);

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h

@@ -102,5 +102,6 @@ struct kernel_queue {
 void kernel_queue_init_cik(struct kernel_queue_ops *ops);
 void kernel_queue_init_vi(struct kernel_queue_ops *ops);
 void kernel_queue_init_v9(struct kernel_queue_ops *ops);
+void kernel_queue_init_v10(struct kernel_queue_ops *ops);
 
 #endif /* KFD_KERNEL_QUEUE_H_ */

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c

+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "kfd_kernel_queue.h"
+#include "kfd_device_queue_manager.h"
+#include "kfd_pm4_headers_ai.h"
+#include "kfd_pm4_opcodes.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+
+static bool initialize_v10(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size);
+static void uninitialize_v10(struct kernel_queue *kq);
+static void submit_packet_v10(struct kernel_queue *kq);
+
+void kernel_queue_init_v10(struct kernel_queue_ops *ops)
+{
+	ops->initialize = initialize_v10;
+	ops->uninitialize = uninitialize_v10;
+	ops->submit_packet = submit_packet_v10;
+}
+
+static bool initialize_v10(struct kernel_queue *kq, struct kfd_dev *dev,
+			enum kfd_queue_type type, unsigned int queue_size)
+{
+	int retval;
+
+	retval = kfd_gtt_sa_allocate(dev, PAGE_SIZE, &kq->eop_mem);
+	if (retval != 0)
+		return false;
+
+	kq->eop_gpu_addr = kq->eop_mem->gpu_addr;
+	kq->eop_kernel_addr = kq->eop_mem->cpu_ptr;
+
+	memset(kq->eop_kernel_addr, 0, PAGE_SIZE);
+
+	return true;
+}
+
+static void uninitialize_v10(struct kernel_queue *kq)
+{
+	kfd_gtt_sa_free(kq->dev, kq->eop_mem);
+}
+
+static void submit_packet_v10(struct kernel_queue *kq)
+{
+	*kq->wptr64_kernel = kq->pending_wptr64;
+	write_kernel_doorbell64(kq->queue->properties.doorbell_ptr,
+				kq->pending_wptr64);
+}
+
+static int pm_map_process_v10(struct packet_manager *pm,
+		uint32_t *buffer, struct qcm_process_device *qpd)
+{
+	struct pm4_mes_map_process *packet;
+	uint64_t vm_page_table_base_addr = qpd->page_table_base;
+
+	packet = (struct pm4_mes_map_process *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mes_map_process));
+
+	packet->header.u32All = pm_build_pm4_header(IT_MAP_PROCESS,
+					sizeof(struct pm4_mes_map_process));
+	packet->bitfields2.diq_enable = (qpd->is_debug) ? 1 : 0;
+	packet->bitfields2.process_quantum = 1;
+	packet->bitfields2.pasid = qpd->pqm->process->pasid;
+	packet->bitfields14.gds_size = qpd->gds_size;
+	packet->bitfields14.num_gws = qpd->num_gws;
+	packet->bitfields14.num_oac = qpd->num_oac;
+	packet->bitfields14.sdma_enable = 1;
+
+	packet->bitfields14.num_queues = (qpd->is_debug) ? 0 : qpd->queue_count;
+
+	packet->sh_mem_config = qpd->sh_mem_config;
+	packet->sh_mem_bases = qpd->sh_mem_bases;
+	if (qpd->tba_addr) {
+		packet->sq_shader_tba_lo = lower_32_bits(qpd->tba_addr >> 8);
+		packet->sq_shader_tba_hi = (1 << SQ_SHADER_TBA_HI__TRAP_EN__SHIFT) |
+			upper_32_bits(qpd->tba_addr >> 8);
+		packet->sq_shader_tma_lo = lower_32_bits(qpd->tma_addr >> 8);
+		packet->sq_shader_tma_hi = upper_32_bits(qpd->tma_addr >> 8);
+	}
+
+	packet->gds_addr_lo = lower_32_bits(qpd->gds_context_area);
+	packet->gds_addr_hi = upper_32_bits(qpd->gds_context_area);
+
+	packet->vm_context_page_table_base_addr_lo32 =
+			lower_32_bits(vm_page_table_base_addr);
+	packet->vm_context_page_table_base_addr_hi32 =
+			upper_32_bits(vm_page_table_base_addr);
+
+	return 0;
+}
+
+static int pm_runlist_v10(struct packet_manager *pm, uint32_t *buffer,
+			uint64_t ib, size_t ib_size_in_dwords, bool chain)
+{
+	struct pm4_mes_runlist *packet;
+
+	int concurrent_proc_cnt = 0;
+	struct kfd_dev *kfd = pm->dqm->dev;
+
+	/* Determine the number of processes to map together to HW:
+	 * it can not exceed the number of VMIDs available to the
+	 * scheduler, and it is determined by the smaller of the number
+	 * of processes in the runlist and kfd module parameter
+	 * hws_max_conc_proc.
+	 * Note: the arbitration between the number of VMIDs and
+	 * hws_max_conc_proc has been done in
+	 * kgd2kfd_device_init().
+	 */
+	concurrent_proc_cnt = min(pm->dqm->processes_count,
+			kfd->max_proc_per_quantum);
+
+
+	packet = (struct pm4_mes_runlist *)buffer;
+
+	memset(buffer, 0, sizeof(struct pm4_mes_runlist));
+	packet->header.u32All = pm_build_pm4_header(IT_RUN_LIST,
+						sizeof(struct pm4_mes_runlist));
+
+	packet->bitfields4.ib_size = ib_size_in_dwords;
+	packet->bitfields4.chain = chain ? 1 : 0;
+	packet->bitfields4.offload_polling = 0;
+	packet->bitfields4.valid = 1;
+	packet->bitfields4.process_cnt = concurrent_proc_cnt;
+	packet->ordinal2 = lower_32_bits(ib);
+	packet->ib_base_hi = upper_32_bits(ib);
+
+	return 0;
+}
+
+static int pm_map_queues_v10(struct packet_manager *pm, uint32_t *buffer,
+		struct queue *q, bool is_static)
+{
+	struct pm4_mes_map_queues *packet;
+	bool use_static = is_static;
+
+	packet = (struct pm4_mes_map_queues *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mes_map_queues));
+
+	packet->header.u32All = pm_build_pm4_header(IT_MAP_QUEUES,
+					sizeof(struct pm4_mes_map_queues));
+	packet->bitfields2.num_queues = 1;
+	packet->bitfields2.queue_sel =
+		queue_sel__mes_map_queues__map_to_hws_determined_queue_slots_vi;
+
+	packet->bitfields2.engine_sel =
+		engine_sel__mes_map_queues__compute_vi;
+	packet->bitfields2.queue_type =
+		queue_type__mes_map_queues__normal_compute_vi;
+
+	switch (q->properties.type) {
+	case KFD_QUEUE_TYPE_COMPUTE:
+		if (use_static)
+			packet->bitfields2.queue_type =
+		queue_type__mes_map_queues__normal_latency_static_queue_vi;
+		break;
+	case KFD_QUEUE_TYPE_DIQ:
+		packet->bitfields2.queue_type =
+			queue_type__mes_map_queues__debug_interface_queue_vi;
+		break;
+	case KFD_QUEUE_TYPE_SDMA:
+	case KFD_QUEUE_TYPE_SDMA_XGMI:
+		packet->bitfields2.engine_sel = q->properties.sdma_engine_id +
+				engine_sel__mes_map_queues__sdma0_vi;
+		use_static = false; /* no static queues under SDMA */
+		break;
+	default:
+		WARN(1, "queue type %d\n", q->properties.type);
+		return -EINVAL;
+	}
+	packet->bitfields3.doorbell_offset =
+			q->properties.doorbell_off;
+
+	packet->mqd_addr_lo =
+			lower_32_bits(q->gart_mqd_addr);
+
+	packet->mqd_addr_hi =
+			upper_32_bits(q->gart_mqd_addr);
+
+	packet->wptr_addr_lo =
+			lower_32_bits((uint64_t)q->properties.write_ptr);
+
+	packet->wptr_addr_hi =
+			upper_32_bits((uint64_t)q->properties.write_ptr);
+
+	return 0;
+}
+
+static int pm_unmap_queues_v10(struct packet_manager *pm, uint32_t *buffer,
+			enum kfd_queue_type type,
+			enum kfd_unmap_queues_filter filter,
+			uint32_t filter_param, bool reset,
+			unsigned int sdma_engine)
+{
+	struct pm4_mes_unmap_queues *packet;
+
+	packet = (struct pm4_mes_unmap_queues *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mes_unmap_queues));
+
+	packet->header.u32All = pm_build_pm4_header(IT_UNMAP_QUEUES,
+					sizeof(struct pm4_mes_unmap_queues));
+	switch (type) {
+	case KFD_QUEUE_TYPE_COMPUTE:
+	case KFD_QUEUE_TYPE_DIQ:
+		packet->bitfields2.engine_sel =
+			engine_sel__mes_unmap_queues__compute;
+		break;
+	case KFD_QUEUE_TYPE_SDMA:
+	case KFD_QUEUE_TYPE_SDMA_XGMI:
+		packet->bitfields2.engine_sel =
+			engine_sel__mes_unmap_queues__sdma0 + sdma_engine;
+		break;
+	default:
+		WARN(1, "queue type %d\n", type);
+		break;
+	}
+
+	if (reset)
+		packet->bitfields2.action =
+			action__mes_unmap_queues__reset_queues;
+	else
+		packet->bitfields2.action =
+			action__mes_unmap_queues__preempt_queues;
+
+	switch (filter) {
+	case KFD_UNMAP_QUEUES_FILTER_SINGLE_QUEUE:
+		packet->bitfields2.queue_sel =
+			queue_sel__mes_unmap_queues__perform_request_on_specified_queues;
+		packet->bitfields2.num_queues = 1;
+		packet->bitfields3b.doorbell_offset0 = filter_param;
+		break;
+	case KFD_UNMAP_QUEUES_FILTER_BY_PASID:
+		packet->bitfields2.queue_sel =
+			queue_sel__mes_unmap_queues__perform_request_on_pasid_queues;
+		packet->bitfields3a.pasid = filter_param;
+		break;
+	case KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES:
+		packet->bitfields2.queue_sel =
+			queue_sel__mes_unmap_queues__unmap_all_queues;
+		break;
+	case KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES:
+		/* in this case, we do not preempt static queues */
+		packet->bitfields2.queue_sel =
+			queue_sel__mes_unmap_queues__unmap_all_non_static_queues;
+		break;
+	default:
+		WARN(1, "filter %d\n", filter);
+		break;
+	}
+
+	return 0;
+
+}
+
+static int pm_query_status_v10(struct packet_manager *pm, uint32_t *buffer,
+			uint64_t fence_address,	uint32_t fence_value)
+{
+	struct pm4_mes_query_status *packet;
+
+	packet = (struct pm4_mes_query_status *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mes_query_status));
+
+
+	packet->header.u32All = pm_build_pm4_header(IT_QUERY_STATUS,
+					sizeof(struct pm4_mes_query_status));
+
+	packet->bitfields2.context_id = 0;
+	packet->bitfields2.interrupt_sel =
+			interrupt_sel__mes_query_status__completion_status;
+	packet->bitfields2.command =
+			command__mes_query_status__fence_only_after_write_ack;
+
+	packet->addr_hi = upper_32_bits((uint64_t)fence_address);
+	packet->addr_lo = lower_32_bits((uint64_t)fence_address);
+	packet->data_hi = upper_32_bits((uint64_t)fence_value);
+	packet->data_lo = lower_32_bits((uint64_t)fence_value);
+
+	return 0;
+}
+
+
+static int pm_release_mem_v10(uint64_t gpu_addr, uint32_t *buffer)
+{
+	struct pm4_mec_release_mem *packet;
+
+	WARN_ON(!buffer);
+
+	packet = (struct pm4_mec_release_mem *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mec_release_mem));
+
+	packet->header.u32All = pm_build_pm4_header(IT_RELEASE_MEM,
+					sizeof(struct pm4_mec_release_mem));
+
+	packet->bitfields2.event_type = CACHE_FLUSH_AND_INV_TS_EVENT;
+	packet->bitfields2.event_index = event_index__mec_release_mem__end_of_pipe;
+	packet->bitfields2.tcl1_action_ena = 1;
+	packet->bitfields2.tc_action_ena = 1;
+	packet->bitfields2.cache_policy = cache_policy__mec_release_mem__lru;
+
+	packet->bitfields3.data_sel = data_sel__mec_release_mem__send_32_bit_low;
+	packet->bitfields3.int_sel =
+		int_sel__mec_release_mem__send_interrupt_after_write_confirm;
+
+	packet->bitfields4.address_lo_32b = (gpu_addr & 0xffffffff) >> 2;
+	packet->address_hi = upper_32_bits(gpu_addr);
+
+	packet->data_lo = 0;
+
+	return sizeof(struct pm4_mec_release_mem) / sizeof(unsigned int);
+}
+
+const struct packet_manager_funcs kfd_v10_pm_funcs = {
+	.map_process			= pm_map_process_v10,
+	.runlist			= pm_runlist_v10,
+	.set_resources			= pm_set_resources_vi,
+	.map_queues			= pm_map_queues_v10,
+	.unmap_queues			= pm_unmap_queues_v10,
+	.query_status			= pm_query_status_v10,
+	.release_mem			= pm_release_mem_v10,
+	.map_process_size		= sizeof(struct pm4_mes_map_process),
+	.runlist_size			= sizeof(struct pm4_mes_runlist),
+	.set_resources_size		= sizeof(struct pm4_mes_set_resources),
+	.map_queues_size		= sizeof(struct pm4_mes_map_queues),
+	.unmap_queues_size		= sizeof(struct pm4_mes_unmap_queues),
+	.query_status_size		= sizeof(struct pm4_mes_query_status),
+	.release_mem_size		= sizeof(struct pm4_mec_release_mem)
+};
+

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c

+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/printk.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include "kfd_priv.h"
+#include "kfd_mqd_manager.h"
+#include "v10_structs.h"
+#include "gc/gc_10_1_0_offset.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+#include "amdgpu_amdkfd.h"
+
+static inline struct v10_compute_mqd *get_mqd(void *mqd)
+{
+	return (struct v10_compute_mqd *)mqd;
+}
+
+static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v10_sdma_mqd *)mqd;
+}
+
+static void update_cu_mask(struct mqd_manager *mm, void *mqd,
+			   struct queue_properties *q)
+{
+	struct v10_compute_mqd *m;
+	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
+
+	if (q->cu_mask_count == 0)
+		return;
+
+	mqd_symmetrically_map_cu_mask(mm,
+		q->cu_mask, q->cu_mask_count, se_mask);
+
+	m = get_mqd(mqd);
+	m->compute_static_thread_mgmt_se0 = se_mask[0];
+	m->compute_static_thread_mgmt_se1 = se_mask[1];
+	m->compute_static_thread_mgmt_se2 = se_mask[2];
+	m->compute_static_thread_mgmt_se3 = se_mask[3];
+
+	pr_debug("update cu mask to %#x %#x %#x %#x\n",
+		m->compute_static_thread_mgmt_se0,
+		m->compute_static_thread_mgmt_se1,
+		m->compute_static_thread_mgmt_se2,
+		m->compute_static_thread_mgmt_se3);
+}
+
+static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd,
+		struct queue_properties *q)
+{
+	int retval;
+	struct kfd_mem_obj *mqd_mem_obj = NULL;
+
+	/* From V9,  for CWSR, the control stack is located on the next page
+	 * boundary after the mqd, we will use the gtt allocation function
+	 * instead of sub-allocation function.
+	 */
+	if (kfd->cwsr_enabled && (q->type == KFD_QUEUE_TYPE_COMPUTE)) {
+		mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);
+		if (!mqd_mem_obj)
+			return NULL;
+		retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->kgd,
+			ALIGN(q->ctl_stack_size, PAGE_SIZE) +
+				ALIGN(sizeof(struct v10_compute_mqd), PAGE_SIZE),
+			&(mqd_mem_obj->gtt_mem),
+			&(mqd_mem_obj->gpu_addr),
+			(void *)&(mqd_mem_obj->cpu_ptr), true);
+	} else {
+		retval = kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd),
+				&mqd_mem_obj);
+	}
+
+	if (retval) {
+		kfree(mqd_mem_obj);
+		return NULL;
+	}
+
+	return mqd_mem_obj;
+
+}
+
+static void init_mqd(struct mqd_manager *mm, void **mqd,
+			struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
+			struct queue_properties *q)
+{
+	uint64_t addr;
+	struct v10_compute_mqd *m;
+
+	m = (struct v10_compute_mqd *) mqd_mem_obj->cpu_ptr;
+	addr = mqd_mem_obj->gpu_addr;
+
+	memset(m, 0, sizeof(struct v10_compute_mqd));
+
+	m->header = 0xC0310800;
+	m->compute_pipelinestat_enable = 1;
+	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
+
+	m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
+			0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
+
+	m->cp_mqd_control = 1 << CP_MQD_CONTROL__PRIV_STATE__SHIFT;
+
+	m->cp_mqd_base_addr_lo        = lower_32_bits(addr);
+	m->cp_mqd_base_addr_hi        = upper_32_bits(addr);
+
+	m->cp_hqd_quantum = 1 << CP_HQD_QUANTUM__QUANTUM_EN__SHIFT |
+			1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
+			10 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
+
+	m->cp_hqd_pipe_priority = 1;
+	m->cp_hqd_queue_priority = 15;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL) {
+		m->cp_hqd_aql_control =
+			1 << CP_HQD_AQL_CONTROL__CONTROL0__SHIFT;
+	}
+
+	if (mm->dev->cwsr_enabled) {
+		m->cp_hqd_persistent_state |=
+			(1 << CP_HQD_PERSISTENT_STATE__QSWITCH_MODE__SHIFT);
+		m->cp_hqd_ctx_save_base_addr_lo =
+			lower_32_bits(q->ctx_save_restore_area_address);
+		m->cp_hqd_ctx_save_base_addr_hi =
+			upper_32_bits(q->ctx_save_restore_area_address);
+		m->cp_hqd_ctx_save_size = q->ctx_save_restore_area_size;
+		m->cp_hqd_cntl_stack_size = q->ctl_stack_size;
+		m->cp_hqd_cntl_stack_offset = q->ctl_stack_size;
+		m->cp_hqd_wg_state_offset = q->ctl_stack_size;
+	}
+
+	*mqd = m;
+	if (gart_addr)
+		*gart_addr = addr;
+	mm->update_mqd(mm, m, q);
+}
+
+static int load_mqd(struct mqd_manager *mm, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			struct queue_properties *p, struct mm_struct *mms)
+{
+	int r = 0;
+	/* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
+	uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
+
+	r = mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id,
+					  (uint32_t __user *)p->write_ptr,
+					  wptr_shift, 0, mms);
+	return r;
+}
+
+static void update_mqd(struct mqd_manager *mm, void *mqd,
+		      struct queue_properties *q)
+{
+	struct v10_compute_mqd *m;
+
+	m = get_mqd(mqd);
+
+	m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT;
+	m->cp_hqd_pq_control |=
+			ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1;
+	pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
+
+	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
+	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
+
+	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->cp_hqd_pq_wptr_poll_addr_lo = lower_32_bits((uint64_t)q->write_ptr);
+	m->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits((uint64_t)q->write_ptr);
+
+	m->cp_hqd_pq_doorbell_control =
+		q->doorbell_off <<
+			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
+	pr_debug("cp_hqd_pq_doorbell_control 0x%x\n",
+			m->cp_hqd_pq_doorbell_control);
+
+	m->cp_hqd_ib_control = 3 << CP_HQD_IB_CONTROL__MIN_IB_AVAIL_SIZE__SHIFT;
+
+	/*
+	 * HW does not clamp this field correctly. Maximum EOP queue size
+	 * is constrained by per-SE EOP done signal count, which is 8-bit.
+	 * Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
+	 * more than (EOP entry count - 1) so a queue size of 0x800 dwords
+	 * is safe, giving a maximum field value of 0xA.
+	 */
+	m->cp_hqd_eop_control = min(0xA,
+		ffs(q->eop_ring_buffer_size / sizeof(unsigned int)) - 1 - 1);
+	m->cp_hqd_eop_base_addr_lo =
+			lower_32_bits(q->eop_ring_buffer_address >> 8);
+	m->cp_hqd_eop_base_addr_hi =
+			upper_32_bits(q->eop_ring_buffer_address >> 8);
+
+	m->cp_hqd_iq_timer = 0;
+
+	m->cp_hqd_vmid = q->vmid;
+
+	if (q->format == KFD_QUEUE_FORMAT_AQL) {
+		/* GC 10 removed WPP_CLAMP from PQ Control */
+		m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
+				2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT |
+				1 << CP_HQD_PQ_CONTROL__QUEUE_FULL_EN__SHIFT ;
+		m->cp_hqd_pq_doorbell_control |=
+			1 << CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_BIF_DROP__SHIFT;
+	}
+	if (mm->dev->cwsr_enabled)
+		m->cp_hqd_ctx_save_control = 0;
+
+	update_cu_mask(mm, mqd, q);
+
+	q->is_active = (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0 &&
+			!q->is_evicted);
+}
+
+static int destroy_mqd(struct mqd_manager *mm, void *mqd,
+		       enum kfd_preempt_type type,
+		       unsigned int timeout, uint32_t pipe_id,
+		       uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_destroy
+		(mm->dev->kgd, mqd, type, timeout,
+		 pipe_id, queue_id);
+}
+
+static void free_mqd(struct mqd_manager *mm, void *mqd,
+			struct kfd_mem_obj *mqd_mem_obj)
+{
+	struct kfd_dev *kfd = mm->dev;
+
+	if (mqd_mem_obj->gtt_mem) {
+		amdgpu_amdkfd_free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem);
+		kfree(mqd_mem_obj);
+	} else {
+		kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
+	}
+}
+
+static bool is_occupied(struct mqd_manager *mm, void *mqd,
+			uint64_t queue_address,	uint32_t pipe_id,
+			uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_is_occupied(
+		mm->dev->kgd, queue_address,
+		pipe_id, queue_id);
+}
+
+static int get_wave_state(struct mqd_manager *mm, void *mqd,
+			  void __user *ctl_stack,
+			  u32 *ctl_stack_used_size,
+			  u32 *save_area_used_size)
+{
+	struct v10_compute_mqd *m;
+
+	/* Control stack is located one page after MQD. */
+	void *mqd_ctl_stack = (void *)((uintptr_t)mqd + PAGE_SIZE);
+
+	m = get_mqd(mqd);
+
+	*ctl_stack_used_size = m->cp_hqd_cntl_stack_size -
+		m->cp_hqd_cntl_stack_offset;
+	*save_area_used_size = m->cp_hqd_wg_state_offset -
+		m->cp_hqd_cntl_stack_size;
+
+	if (copy_to_user(ctl_stack, mqd_ctl_stack, m->cp_hqd_cntl_stack_size))
+		return -EFAULT;
+
+	return 0;
+}
+
+static void init_mqd_hiq(struct mqd_manager *mm, void **mqd,
+			struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
+			struct queue_properties *q)
+{
+	struct v10_compute_mqd *m;
+
+	init_mqd(mm, mqd, mqd_mem_obj, gart_addr, q);
+
+	m = get_mqd(*mqd);
+
+	m->cp_hqd_pq_control |= 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
+			1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
+}
+
+static void update_mqd_hiq(struct mqd_manager *mm, void *mqd,
+			struct queue_properties *q)
+{
+	struct v10_compute_mqd *m;
+
+	update_mqd(mm, mqd, q);
+
+	/* TODO: what's the point? update_mqd already does this. */
+	m = get_mqd(mqd);
+	m->cp_hqd_vmid = q->vmid;
+}
+
+static void init_mqd_sdma(struct mqd_manager *mm, void **mqd,
+		struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
+		struct queue_properties *q)
+{
+	struct v10_sdma_mqd *m;
+
+	m = (struct v10_sdma_mqd *) mqd_mem_obj->cpu_ptr;
+
+	memset(m, 0, sizeof(struct v10_sdma_mqd));
+
+	*mqd = m;
+	if (gart_addr)
+		*gart_addr = mqd_mem_obj->gpu_addr;
+
+	mm->update_mqd(mm, m, q);
+}
+
+static int load_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		uint32_t pipe_id, uint32_t queue_id,
+		struct queue_properties *p, struct mm_struct *mms)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd,
+					       (uint32_t __user *)p->write_ptr,
+					       mms);
+}
+
+#define SDMA_RLC_DUMMY_DEFAULT 0xf
+
+static void update_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		struct queue_properties *q)
+{
+	struct v10_sdma_mqd *m;
+
+	m = get_sdma_mqd(mqd);
+	m->sdmax_rlcx_rb_cntl = (ffs(q->queue_size / sizeof(unsigned int)) - 1)
+		<< SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
+		q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
+		1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
+		6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
+
+	m->sdmax_rlcx_rb_base = lower_32_bits(q->queue_address >> 8);
+	m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8);
+	m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->sdmax_rlcx_doorbell_offset =
+		q->doorbell_off << SDMA0_RLC0_DOORBELL_OFFSET__OFFSET__SHIFT;
+
+	m->sdma_engine_id = q->sdma_engine_id;
+	m->sdma_queue_id = q->sdma_queue_id;
+	m->sdmax_rlcx_dummy_reg = SDMA_RLC_DUMMY_DEFAULT;
+
+
+	q->is_active = (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0 &&
+			!q->is_evicted);
+}
+
+/*
+ *  * preempt type here is ignored because there is only one way
+ *  * to preempt sdma queue
+ */
+static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		enum kfd_preempt_type type,
+		unsigned int timeout, uint32_t pipe_id,
+		uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout);
+}
+
+static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd,
+		uint64_t queue_address, uint32_t pipe_id,
+		uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd);
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+static int debugfs_show_mqd(struct seq_file *m, void *data)
+{
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
+		     data, sizeof(struct v10_compute_mqd), false);
+	return 0;
+}
+
+static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
+{
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
+		     data, sizeof(struct v10_sdma_mqd), false);
+	return 0;
+}
+
+#endif
+
+struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
+		struct kfd_dev *dev)
+{
+	struct mqd_manager *mqd;
+
+	if (WARN_ON(type >= KFD_MQD_TYPE_MAX))
+		return NULL;
+
+	mqd = kzalloc(sizeof(*mqd), GFP_NOIO);
+	if (!mqd)
+		return NULL;
+
+	mqd->dev = dev;
+
+	switch (type) {
+	case KFD_MQD_TYPE_CP:
+		pr_debug("%s@%i\n", __func__, __LINE__);
+	case KFD_MQD_TYPE_COMPUTE:
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		mqd->allocate_mqd = allocate_mqd;
+		mqd->init_mqd = init_mqd;
+		mqd->free_mqd = free_mqd;
+		mqd->load_mqd = load_mqd;
+		mqd->update_mqd = update_mqd;
+		mqd->destroy_mqd = destroy_mqd;
+		mqd->is_occupied = is_occupied;
+		mqd->mqd_size = sizeof(struct v10_compute_mqd);
+		mqd->get_wave_state = get_wave_state;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		break;
+	case KFD_MQD_TYPE_HIQ:
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		mqd->allocate_mqd = allocate_hiq_mqd;
+		mqd->init_mqd = init_mqd_hiq;
+		mqd->free_mqd = free_mqd_hiq_sdma;
+		mqd->load_mqd = load_mqd;
+		mqd->update_mqd = update_mqd_hiq;
+		mqd->destroy_mqd = destroy_mqd;
+		mqd->is_occupied = is_occupied;
+		mqd->mqd_size = sizeof(struct v10_compute_mqd);
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		break;
+	case KFD_MQD_TYPE_DIQ:
+		mqd->allocate_mqd = allocate_hiq_mqd;
+		mqd->init_mqd = init_mqd_hiq;
+		mqd->free_mqd = free_mqd;
+		mqd->load_mqd = load_mqd;
+		mqd->update_mqd = update_mqd_hiq;
+		mqd->destroy_mqd = destroy_mqd;
+		mqd->is_occupied = is_occupied;
+		mqd->mqd_size = sizeof(struct v10_compute_mqd);
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
+		break;
+	case KFD_MQD_TYPE_SDMA:
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		mqd->allocate_mqd = allocate_sdma_mqd;
+		mqd->init_mqd = init_mqd_sdma;
+		mqd->free_mqd = free_mqd_hiq_sdma;
+		mqd->load_mqd = load_mqd_sdma;
+		mqd->update_mqd = update_mqd_sdma;
+		mqd->destroy_mqd = destroy_mqd_sdma;
+		mqd->is_occupied = is_occupied_sdma;
+		mqd->mqd_size = sizeof(struct v10_sdma_mqd);
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
+#endif
+		pr_debug("%s@%i\n", __func__, __LINE__);
+		break;
+	default:
+		kfree(mqd);
+		return NULL;
+	}
+
+	return mqd;
+}

drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c

@@ -237,6 +237,9 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm)
 	case CHIP_RAVEN:
 		pm->pmf = &kfd_v9_pm_funcs;
 		break;
+	case CHIP_NAVI10:
+		pm->pmf = &kfd_v10_pm_funcs;
+		break;
 	default:
 		WARN(1, "Unexpected ASIC family %u",
 		     dqm->dev->device_info->asic_family);

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -105,6 +105,8 @@
 
 #define KFD_KERNEL_QUEUE_SIZE 2048
 
+#define KFD_UNMAP_LATENCY_MS	(4000)
+
 /*
  * 512 = 0x200
  * The doorbell index distance between SDMA RLC (2*i) and (2*i+1) in the
@@ -167,11 +169,20 @@ extern int halt_if_hws_hang;
  */
 extern bool hws_gws_support;
 
+/*
+ * Queue preemption timeout in ms
+ */
+extern int queue_preemption_timeout_ms;
+
 enum cache_policy {
 	cache_policy_coherent,
 	cache_policy_noncoherent
 };
 
+#define KFD_IS_VI(chip) ((chip) >= CHIP_CARRIZO && (chip) <= CHIP_POLARIS11)
+#define KFD_IS_DGPU(chip) (((chip) >= CHIP_TONGA && \
+			   (chip) <= CHIP_NAVI10) || \
+			   (chip) == CHIP_HAWAII)
 #define KFD_IS_SOC15(chip) ((chip) >= CHIP_VEGA10)
 
 struct kfd_event_interrupt_class {
@@ -870,6 +881,8 @@ struct mqd_manager *mqd_manager_init_vi_tonga(enum KFD_MQD_TYPE type,
 		struct kfd_dev *dev);
 struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
 		struct kfd_dev *dev);
+struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
+		struct kfd_dev *dev);
 struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev);
 void device_queue_manager_uninit(struct device_queue_manager *dqm);
 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
@@ -959,6 +972,7 @@ struct packet_manager_funcs {
 
 extern const struct packet_manager_funcs kfd_vi_pm_funcs;
 extern const struct packet_manager_funcs kfd_v9_pm_funcs;
+extern const struct packet_manager_funcs kfd_v10_pm_funcs;
 
 int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm);
 void pm_uninit(struct packet_manager *pm);
@@ -980,6 +994,7 @@ unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size);
 int pm_set_resources_vi(struct packet_manager *pm, uint32_t *buffer,
 			struct scheduling_resources *res);
 
+
 uint64_t kfd_get_number_elems(struct kfd_dev *kfd);
 
 /* Events */

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -1203,3 +1203,4 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
 }
 
 #endif
+

drivers/gpu/drm/amd/amdkfd/kfd_topology.c

@@ -1321,6 +1321,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
 	case CHIP_RAVEN:
+	case CHIP_NAVI10:
 		dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 <<
 			HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) &
 			HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK);