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
get_sdma_mqd(void * mqd)60 static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
61 {
62 return (struct v9_sdma_mqd *)mqd;
63 }
64
get_sdma_rlc_reg_offset(struct amdgpu_device * adev,unsigned int engine_id,unsigned int queue_id)65 static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
66 unsigned int engine_id,
67 unsigned int queue_id)
68 {
69 uint32_t sdma_engine_reg_base = 0;
70 uint32_t sdma_rlc_reg_offset;
71
72 switch (engine_id) {
73 default:
74 dev_warn(adev->dev,
75 "Invalid sdma engine id (%d), using engine id 0\n",
76 engine_id);
77 fallthrough;
78 case 0:
79 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
80 mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
81 break;
82 case 1:
83 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
84 mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL;
85 break;
86 case 2:
87 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA2, 0,
88 mmSDMA2_RLC0_RB_CNTL) - mmSDMA2_RLC0_RB_CNTL;
89 break;
90 case 3:
91 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA3, 0,
92 mmSDMA3_RLC0_RB_CNTL) - mmSDMA3_RLC0_RB_CNTL;
93 break;
94 case 4:
95 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA4, 0,
96 mmSDMA4_RLC0_RB_CNTL) - mmSDMA4_RLC0_RB_CNTL;
97 break;
98 case 5:
99 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA5, 0,
100 mmSDMA5_RLC0_RB_CNTL) - mmSDMA5_RLC0_RB_CNTL;
101 break;
102 case 6:
103 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA6, 0,
104 mmSDMA6_RLC0_RB_CNTL) - mmSDMA6_RLC0_RB_CNTL;
105 break;
106 case 7:
107 sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA7, 0,
108 mmSDMA7_RLC0_RB_CNTL) - mmSDMA7_RLC0_RB_CNTL;
109 break;
110 }
111
112 sdma_rlc_reg_offset = sdma_engine_reg_base
113 + queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
114
115 pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
116 queue_id, sdma_rlc_reg_offset);
117
118 return sdma_rlc_reg_offset;
119 }
120
kgd_arcturus_hqd_sdma_load(struct amdgpu_device * adev,void * mqd,uint32_t __user * wptr,struct mm_struct * mm)121 int kgd_arcturus_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
122 uint32_t __user *wptr, struct mm_struct *mm)
123 {
124 struct v9_sdma_mqd *m;
125 uint32_t sdma_rlc_reg_offset;
126 unsigned long end_jiffies;
127 uint32_t data;
128 uint64_t data64;
129 uint64_t __user *wptr64 = (uint64_t __user *)wptr;
130
131 m = get_sdma_mqd(mqd);
132 sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
133 m->sdma_queue_id);
134
135 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
136 m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
137
138 end_jiffies = msecs_to_jiffies(2000) + jiffies;
139 while (true) {
140 data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
141 if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
142 break;
143 if (time_after(jiffies, end_jiffies)) {
144 pr_err("SDMA RLC not idle in %s\n", __func__);
145 return -ETIME;
146 }
147 usleep_range(500, 1000);
148 }
149
150 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
151 m->sdmax_rlcx_doorbell_offset);
152
153 data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
154 ENABLE, 1);
155 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
156 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
157 m->sdmax_rlcx_rb_rptr);
158 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
159 m->sdmax_rlcx_rb_rptr_hi);
160
161 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
162 if (read_user_wptr(mm, wptr64, data64)) {
163 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
164 lower_32_bits(data64));
165 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
166 upper_32_bits(data64));
167 } else {
168 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
169 m->sdmax_rlcx_rb_rptr);
170 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
171 m->sdmax_rlcx_rb_rptr_hi);
172 }
173 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
174
175 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
176 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
177 m->sdmax_rlcx_rb_base_hi);
178 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
179 m->sdmax_rlcx_rb_rptr_addr_lo);
180 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
181 m->sdmax_rlcx_rb_rptr_addr_hi);
182
183 data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
184 RB_ENABLE, 1);
185 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
186
187 return 0;
188 }
189
kgd_arcturus_hqd_sdma_dump(struct amdgpu_device * adev,uint32_t engine_id,uint32_t queue_id,uint32_t (** dump)[2],uint32_t * n_regs)190 int kgd_arcturus_hqd_sdma_dump(struct amdgpu_device *adev,
191 uint32_t engine_id, uint32_t queue_id,
192 uint32_t (**dump)[2], uint32_t *n_regs)
193 {
194 uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
195 engine_id, queue_id);
196 uint32_t i = 0, reg;
197 #undef HQD_N_REGS
198 #define HQD_N_REGS (19+6+7+10)
199
200 *dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL);
201 if (*dump == NULL)
202 return -ENOMEM;
203
204 for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
205 DUMP_REG(sdma_rlc_reg_offset + reg);
206 for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
207 DUMP_REG(sdma_rlc_reg_offset + reg);
208 for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
209 reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
210 DUMP_REG(sdma_rlc_reg_offset + reg);
211 for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
212 reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
213 DUMP_REG(sdma_rlc_reg_offset + reg);
214
215 WARN_ON_ONCE(i != HQD_N_REGS);
216 *n_regs = i;
217
218 return 0;
219 }
220
kgd_arcturus_hqd_sdma_is_occupied(struct amdgpu_device * adev,void * mqd)221 bool kgd_arcturus_hqd_sdma_is_occupied(struct amdgpu_device *adev,
222 void *mqd)
223 {
224 struct v9_sdma_mqd *m;
225 uint32_t sdma_rlc_reg_offset;
226 uint32_t sdma_rlc_rb_cntl;
227
228 m = get_sdma_mqd(mqd);
229 sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
230 m->sdma_queue_id);
231
232 sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
233
234 if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
235 return true;
236
237 return false;
238 }
239
kgd_arcturus_hqd_sdma_destroy(struct amdgpu_device * adev,void * mqd,unsigned int utimeout)240 int kgd_arcturus_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
241 unsigned int utimeout)
242 {
243 struct v9_sdma_mqd *m;
244 uint32_t sdma_rlc_reg_offset;
245 uint32_t temp;
246 unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
247
248 m = get_sdma_mqd(mqd);
249 sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
250 m->sdma_queue_id);
251
252 temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
253 temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
254 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
255
256 while (true) {
257 temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
258 if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
259 break;
260 if (time_after(jiffies, end_jiffies)) {
261 pr_err("SDMA RLC not idle in %s\n", __func__);
262 return -ETIME;
263 }
264 usleep_range(500, 1000);
265 }
266
267 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
268 WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
269 RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
270 SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
271
272 m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
273 m->sdmax_rlcx_rb_rptr_hi =
274 RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
275
276 return 0;
277 }
278
279 const struct kfd2kgd_calls arcturus_kfd2kgd = {
280 .program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
281 .set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
282 .init_interrupts = kgd_gfx_v9_init_interrupts,
283 .hqd_load = kgd_gfx_v9_hqd_load,
284 .hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
285 .hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
286 .hqd_dump = kgd_gfx_v9_hqd_dump,
287 .hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
288 .hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
289 .hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
290 .hqd_destroy = kgd_gfx_v9_hqd_destroy,
291 .hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
292 .wave_control_execute = kgd_gfx_v9_wave_control_execute,
293 .get_atc_vmid_pasid_mapping_info =
294 kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
295 .set_vm_context_page_table_base =
296 kgd_gfx_v9_set_vm_context_page_table_base,
297 .get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
298 .program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings
299 };
300