1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  */
22 #include <linux/module.h>
23 #include <linux/fdtable.h>
24 #include <linux/uaccess.h>
25 #include <linux/firmware.h>
26 #include "amdgpu.h"
27 #include "amdgpu_amdkfd.h"
28 #include "amdgpu_amdkfd_arcturus.h"
29 #include "sdma0/sdma0_4_2_2_offset.h"
30 #include "sdma0/sdma0_4_2_2_sh_mask.h"
31 #include "sdma1/sdma1_4_2_2_offset.h"
32 #include "sdma1/sdma1_4_2_2_sh_mask.h"
33 #include "sdma2/sdma2_4_2_2_offset.h"
34 #include "sdma2/sdma2_4_2_2_sh_mask.h"
35 #include "sdma3/sdma3_4_2_2_offset.h"
36 #include "sdma3/sdma3_4_2_2_sh_mask.h"
37 #include "sdma4/sdma4_4_2_2_offset.h"
38 #include "sdma4/sdma4_4_2_2_sh_mask.h"
39 #include "sdma5/sdma5_4_2_2_offset.h"
40 #include "sdma5/sdma5_4_2_2_sh_mask.h"
41 #include "sdma6/sdma6_4_2_2_offset.h"
42 #include "sdma6/sdma6_4_2_2_sh_mask.h"
43 #include "sdma7/sdma7_4_2_2_offset.h"
44 #include "sdma7/sdma7_4_2_2_sh_mask.h"
45 #include "v9_structs.h"
46 #include "soc15.h"
47 #include "soc15d.h"
48 #include "amdgpu_amdkfd_gfx_v9.h"
49 #include "gfxhub_v1_0.h"
50 #include "mmhub_v9_4.h"
51 
52 #define HQD_N_REGS 56
53 #define DUMP_REG(addr) do {				\
54 		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
55 			break;				\
56 		(*dump)[i][0] = (addr) << 2;		\
57 		(*dump)[i++][1] = RREG32(addr);		\
58 	} while (0)
59 
60 static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
61 {
62 	return (struct amdgpu_device *)kgd;
63 }
64 
65 static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
66 {
67 	return (struct v9_sdma_mqd *)mqd;
68 }
69 
70 static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
71 				unsigned int engine_id,
72 				unsigned int queue_id)
73 {
74 	uint32_t sdma_engine_reg_base = 0;
75 	uint32_t sdma_rlc_reg_offset;
76 
77 	switch (engine_id) {
78 	default:
79 		dev_warn(adev->dev,
80 			 "Invalid sdma engine id (%d), using engine id 0\n",
81 			 engine_id);
82 		fallthrough;
83 	case 0:
84 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
85 				mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
86 		break;
87 	case 1:
88 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
89 				mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL;
90 		break;
91 	case 2:
92 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA2, 0,
93 				mmSDMA2_RLC0_RB_CNTL) - mmSDMA2_RLC0_RB_CNTL;
94 		break;
95 	case 3:
96 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA3, 0,
97 				mmSDMA3_RLC0_RB_CNTL) - mmSDMA3_RLC0_RB_CNTL;
98 		break;
99 	case 4:
100 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA4, 0,
101 				mmSDMA4_RLC0_RB_CNTL) - mmSDMA4_RLC0_RB_CNTL;
102 		break;
103 	case 5:
104 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA5, 0,
105 				mmSDMA5_RLC0_RB_CNTL) - mmSDMA5_RLC0_RB_CNTL;
106 		break;
107 	case 6:
108 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA6, 0,
109 				mmSDMA6_RLC0_RB_CNTL) - mmSDMA6_RLC0_RB_CNTL;
110 		break;
111 	case 7:
112 		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA7, 0,
113 				mmSDMA7_RLC0_RB_CNTL) - mmSDMA7_RLC0_RB_CNTL;
114 		break;
115 	}
116 
117 	sdma_rlc_reg_offset = sdma_engine_reg_base
118 		+ queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
119 
120 	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
121 			queue_id, sdma_rlc_reg_offset);
122 
123 	return sdma_rlc_reg_offset;
124 }
125 
126 int kgd_arcturus_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
127 			     uint32_t __user *wptr, struct mm_struct *mm)
128 {
129 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
130 	struct v9_sdma_mqd *m;
131 	uint32_t sdma_rlc_reg_offset;
132 	unsigned long end_jiffies;
133 	uint32_t data;
134 	uint64_t data64;
135 	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
136 
137 	m = get_sdma_mqd(mqd);
138 	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
139 					    m->sdma_queue_id);
140 
141 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
142 		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
143 
144 	end_jiffies = msecs_to_jiffies(2000) + jiffies;
145 	while (true) {
146 		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
147 		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
148 			break;
149 		if (time_after(jiffies, end_jiffies)) {
150 			pr_err("SDMA RLC not idle in %s\n", __func__);
151 			return -ETIME;
152 		}
153 		usleep_range(500, 1000);
154 	}
155 
156 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
157 	       m->sdmax_rlcx_doorbell_offset);
158 
159 	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
160 			     ENABLE, 1);
161 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
162 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
163 				m->sdmax_rlcx_rb_rptr);
164 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
165 				m->sdmax_rlcx_rb_rptr_hi);
166 
167 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
168 	if (read_user_wptr(mm, wptr64, data64)) {
169 		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
170 		       lower_32_bits(data64));
171 		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
172 		       upper_32_bits(data64));
173 	} else {
174 		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
175 		       m->sdmax_rlcx_rb_rptr);
176 		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
177 		       m->sdmax_rlcx_rb_rptr_hi);
178 	}
179 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
180 
181 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
182 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
183 			m->sdmax_rlcx_rb_base_hi);
184 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
185 			m->sdmax_rlcx_rb_rptr_addr_lo);
186 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
187 			m->sdmax_rlcx_rb_rptr_addr_hi);
188 
189 	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
190 			     RB_ENABLE, 1);
191 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
192 
193 	return 0;
194 }
195 
196 int kgd_arcturus_hqd_sdma_dump(struct kgd_dev *kgd,
197 			     uint32_t engine_id, uint32_t queue_id,
198 			     uint32_t (**dump)[2], uint32_t *n_regs)
199 {
200 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
201 	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
202 			engine_id, queue_id);
203 	uint32_t i = 0, reg;
204 #undef HQD_N_REGS
205 #define HQD_N_REGS (19+6+7+10)
206 
207 	*dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL);
208 	if (*dump == NULL)
209 		return -ENOMEM;
210 
211 	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
212 		DUMP_REG(sdma_rlc_reg_offset + reg);
213 	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
214 		DUMP_REG(sdma_rlc_reg_offset + reg);
215 	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
216 	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
217 		DUMP_REG(sdma_rlc_reg_offset + reg);
218 	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
219 	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
220 		DUMP_REG(sdma_rlc_reg_offset + reg);
221 
222 	WARN_ON_ONCE(i != HQD_N_REGS);
223 	*n_regs = i;
224 
225 	return 0;
226 }
227 
228 bool kgd_arcturus_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
229 {
230 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
231 	struct v9_sdma_mqd *m;
232 	uint32_t sdma_rlc_reg_offset;
233 	uint32_t sdma_rlc_rb_cntl;
234 
235 	m = get_sdma_mqd(mqd);
236 	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
237 					    m->sdma_queue_id);
238 
239 	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
240 
241 	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
242 		return true;
243 
244 	return false;
245 }
246 
247 int kgd_arcturus_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
248 				unsigned int utimeout)
249 {
250 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
251 	struct v9_sdma_mqd *m;
252 	uint32_t sdma_rlc_reg_offset;
253 	uint32_t temp;
254 	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
255 
256 	m = get_sdma_mqd(mqd);
257 	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
258 					    m->sdma_queue_id);
259 
260 	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
261 	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
262 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
263 
264 	while (true) {
265 		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
266 		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
267 			break;
268 		if (time_after(jiffies, end_jiffies)) {
269 			pr_err("SDMA RLC not idle in %s\n", __func__);
270 			return -ETIME;
271 		}
272 		usleep_range(500, 1000);
273 	}
274 
275 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
276 	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
277 		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
278 		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
279 
280 	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
281 	m->sdmax_rlcx_rb_rptr_hi =
282 		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
283 
284 	return 0;
285 }
286 
287 const struct kfd2kgd_calls arcturus_kfd2kgd = {
288 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
289 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
290 	.init_interrupts = kgd_gfx_v9_init_interrupts,
291 	.hqd_load = kgd_gfx_v9_hqd_load,
292 	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
293 	.hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
294 	.hqd_dump = kgd_gfx_v9_hqd_dump,
295 	.hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
296 	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
297 	.hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
298 	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
299 	.hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
300 	.address_watch_disable = kgd_gfx_v9_address_watch_disable,
301 	.address_watch_execute = kgd_gfx_v9_address_watch_execute,
302 	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
303 	.address_watch_get_offset = kgd_gfx_v9_address_watch_get_offset,
304 	.get_atc_vmid_pasid_mapping_info =
305 				kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
306 	.set_vm_context_page_table_base =
307 				kgd_gfx_v9_set_vm_context_page_table_base,
308 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
309 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings
310 };
311