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 
23 #undef pr_fmt
24 #define pr_fmt(fmt) "kfd2kgd: " fmt
25 
26 #include <linux/module.h>
27 #include <linux/fdtable.h>
28 #include <linux/uaccess.h>
29 #include <linux/mmu_context.h>
30 #include <linux/firmware.h>
31 #include "amdgpu.h"
32 #include "amdgpu_amdkfd.h"
33 #include "sdma0/sdma0_4_2_2_offset.h"
34 #include "sdma0/sdma0_4_2_2_sh_mask.h"
35 #include "sdma1/sdma1_4_2_2_offset.h"
36 #include "sdma1/sdma1_4_2_2_sh_mask.h"
37 #include "sdma2/sdma2_4_2_2_offset.h"
38 #include "sdma2/sdma2_4_2_2_sh_mask.h"
39 #include "sdma3/sdma3_4_2_2_offset.h"
40 #include "sdma3/sdma3_4_2_2_sh_mask.h"
41 #include "sdma4/sdma4_4_2_2_offset.h"
42 #include "sdma4/sdma4_4_2_2_sh_mask.h"
43 #include "sdma5/sdma5_4_2_2_offset.h"
44 #include "sdma5/sdma5_4_2_2_sh_mask.h"
45 #include "sdma6/sdma6_4_2_2_offset.h"
46 #include "sdma6/sdma6_4_2_2_sh_mask.h"
47 #include "sdma7/sdma7_4_2_2_offset.h"
48 #include "sdma7/sdma7_4_2_2_sh_mask.h"
49 #include "v9_structs.h"
50 #include "soc15.h"
51 #include "soc15d.h"
52 #include "amdgpu_amdkfd_gfx_v9.h"
53 
54 #define HQD_N_REGS 56
55 #define DUMP_REG(addr) do {				\
56 		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
57 			break;				\
58 		(*dump)[i][0] = (addr) << 2;		\
59 		(*dump)[i++][1] = RREG32(addr);		\
60 	} while (0)
61 
62 static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
63 {
64 	return (struct amdgpu_device *)kgd;
65 }
66 
67 static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
68 {
69 	return (struct v9_sdma_mqd *)mqd;
70 }
71 
72 static uint32_t get_sdma_base_addr(struct amdgpu_device *adev,
73 				unsigned int engine_id,
74 				unsigned int queue_id)
75 {
76 	uint32_t base[8] = {
77 		SOC15_REG_OFFSET(SDMA0, 0,
78 				 mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL,
79 		SOC15_REG_OFFSET(SDMA1, 0,
80 				 mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL,
81 		SOC15_REG_OFFSET(SDMA2, 0,
82 				 mmSDMA2_RLC0_RB_CNTL) - mmSDMA2_RLC0_RB_CNTL,
83 		SOC15_REG_OFFSET(SDMA3, 0,
84 				 mmSDMA3_RLC0_RB_CNTL) - mmSDMA3_RLC0_RB_CNTL,
85 		SOC15_REG_OFFSET(SDMA4, 0,
86 				 mmSDMA4_RLC0_RB_CNTL) - mmSDMA4_RLC0_RB_CNTL,
87 		SOC15_REG_OFFSET(SDMA5, 0,
88 				 mmSDMA5_RLC0_RB_CNTL) - mmSDMA5_RLC0_RB_CNTL,
89 		SOC15_REG_OFFSET(SDMA6, 0,
90 				 mmSDMA6_RLC0_RB_CNTL) - mmSDMA6_RLC0_RB_CNTL,
91 		SOC15_REG_OFFSET(SDMA7, 0,
92 				 mmSDMA7_RLC0_RB_CNTL) - mmSDMA7_RLC0_RB_CNTL
93 	};
94 	uint32_t retval;
95 
96 	retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL -
97 					       mmSDMA0_RLC0_RB_CNTL);
98 
99 	pr_debug("sdma base address: 0x%x\n", retval);
100 
101 	return retval;
102 }
103 
104 static u32 sdma_v4_0_get_reg_offset(struct amdgpu_device *adev,
105 		u32 instance, u32 offset)
106 {
107 	switch (instance) {
108 	case 0:
109 		return (adev->reg_offset[SDMA0_HWIP][0][0] + offset);
110 	case 1:
111 		return (adev->reg_offset[SDMA1_HWIP][0][1] + offset);
112 	case 2:
113 		return (adev->reg_offset[SDMA2_HWIP][0][1] + offset);
114 	case 3:
115 		return (adev->reg_offset[SDMA3_HWIP][0][1] + offset);
116 	case 4:
117 		return (adev->reg_offset[SDMA4_HWIP][0][1] + offset);
118 	case 5:
119 		return (adev->reg_offset[SDMA5_HWIP][0][1] + offset);
120 	case 6:
121 		return (adev->reg_offset[SDMA6_HWIP][0][1] + offset);
122 	case 7:
123 		return (adev->reg_offset[SDMA7_HWIP][0][1] + offset);
124 	default:
125 		break;
126 	}
127 	return 0;
128 }
129 
130 static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
131 			     uint32_t __user *wptr, struct mm_struct *mm)
132 {
133 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
134 	struct v9_sdma_mqd *m;
135 	uint32_t sdma_base_addr, sdmax_gfx_context_cntl;
136 	unsigned long end_jiffies;
137 	uint32_t data;
138 	uint64_t data64;
139 	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
140 
141 	m = get_sdma_mqd(mqd);
142 	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
143 					    m->sdma_queue_id);
144 	sdmax_gfx_context_cntl = sdma_v4_0_get_reg_offset(adev,
145 			m->sdma_engine_id, mmSDMA0_GFX_CONTEXT_CNTL);
146 
147 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
148 		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
149 
150 	end_jiffies = msecs_to_jiffies(2000) + jiffies;
151 	while (true) {
152 		data = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
153 		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
154 			break;
155 		if (time_after(jiffies, end_jiffies))
156 			return -ETIME;
157 		usleep_range(500, 1000);
158 	}
159 	data = RREG32(sdmax_gfx_context_cntl);
160 	data = REG_SET_FIELD(data, SDMA0_GFX_CONTEXT_CNTL,
161 			     RESUME_CTX, 0);
162 	WREG32(sdmax_gfx_context_cntl, data);
163 
164 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET,
165 	       m->sdmax_rlcx_doorbell_offset);
166 
167 	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
168 			     ENABLE, 1);
169 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data);
170 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr);
171 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI,
172 				m->sdmax_rlcx_rb_rptr_hi);
173 
174 	WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
175 	if (read_user_wptr(mm, wptr64, data64)) {
176 		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
177 		       lower_32_bits(data64));
178 		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI,
179 		       upper_32_bits(data64));
180 	} else {
181 		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
182 		       m->sdmax_rlcx_rb_rptr);
183 		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI,
184 		       m->sdmax_rlcx_rb_rptr_hi);
185 	}
186 	WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
187 
188 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
189 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
190 			m->sdmax_rlcx_rb_base_hi);
191 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
192 			m->sdmax_rlcx_rb_rptr_addr_lo);
193 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
194 			m->sdmax_rlcx_rb_rptr_addr_hi);
195 
196 	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
197 			     RB_ENABLE, 1);
198 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data);
199 
200 	return 0;
201 }
202 
203 static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
204 			     uint32_t engine_id, uint32_t queue_id,
205 			     uint32_t (**dump)[2], uint32_t *n_regs)
206 {
207 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
208 	uint32_t sdma_base_addr = get_sdma_base_addr(adev, engine_id, queue_id);
209 	uint32_t i = 0, reg;
210 #undef HQD_N_REGS
211 #define HQD_N_REGS (19+6+7+10)
212 
213 	*dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL);
214 	if (*dump == NULL)
215 		return -ENOMEM;
216 
217 	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
218 		DUMP_REG(sdma_base_addr + reg);
219 	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
220 		DUMP_REG(sdma_base_addr + reg);
221 	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
222 	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
223 		DUMP_REG(sdma_base_addr + reg);
224 	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
225 	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
226 		DUMP_REG(sdma_base_addr + reg);
227 
228 	WARN_ON_ONCE(i != HQD_N_REGS);
229 	*n_regs = i;
230 
231 	return 0;
232 }
233 
234 static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
235 {
236 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
237 	struct v9_sdma_mqd *m;
238 	uint32_t sdma_base_addr;
239 	uint32_t sdma_rlc_rb_cntl;
240 
241 	m = get_sdma_mqd(mqd);
242 	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
243 					    m->sdma_queue_id);
244 
245 	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
246 
247 	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
248 		return true;
249 
250 	return false;
251 }
252 
253 static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
254 				unsigned int utimeout)
255 {
256 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
257 	struct v9_sdma_mqd *m;
258 	uint32_t sdma_base_addr;
259 	uint32_t temp;
260 	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
261 
262 	m = get_sdma_mqd(mqd);
263 	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
264 					    m->sdma_queue_id);
265 
266 	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
267 	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
268 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
269 
270 	while (true) {
271 		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
272 		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
273 			break;
274 		if (time_after(jiffies, end_jiffies))
275 			return -ETIME;
276 		usleep_range(500, 1000);
277 	}
278 
279 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
280 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
281 		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
282 		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
283 
284 	m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR);
285 	m->sdmax_rlcx_rb_rptr_hi =
286 		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI);
287 
288 	return 0;
289 }
290 
291 static const struct kfd2kgd_calls kfd2kgd = {
292 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
293 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
294 	.init_interrupts = kgd_gfx_v9_init_interrupts,
295 	.hqd_load = kgd_gfx_v9_hqd_load,
296 	.hqd_sdma_load = kgd_hqd_sdma_load,
297 	.hqd_dump = kgd_gfx_v9_hqd_dump,
298 	.hqd_sdma_dump = kgd_hqd_sdma_dump,
299 	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
300 	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
301 	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
302 	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
303 	.address_watch_disable = kgd_gfx_v9_address_watch_disable,
304 	.address_watch_execute = kgd_gfx_v9_address_watch_execute,
305 	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
306 	.address_watch_get_offset = kgd_gfx_v9_address_watch_get_offset,
307 	.get_atc_vmid_pasid_mapping_pasid =
308 			kgd_gfx_v9_get_atc_vmid_pasid_mapping_pasid,
309 	.get_atc_vmid_pasid_mapping_valid =
310 			kgd_gfx_v9_get_atc_vmid_pasid_mapping_valid,
311 	.set_scratch_backing_va = kgd_gfx_v9_set_scratch_backing_va,
312 	.get_tile_config = kgd_gfx_v9_get_tile_config,
313 	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
314 	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
315 	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
316 	.get_hive_id = amdgpu_amdkfd_get_hive_id,
317 };
318 
319 struct kfd2kgd_calls *amdgpu_amdkfd_arcturus_get_functions(void)
320 {
321 	return (struct kfd2kgd_calls *)&kfd2kgd;
322 }
323 
324