1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Copyright 2016-2019 HabanaLabs, Ltd.
5 * All Rights Reserved.
6 */
7
8 #include "goyaP.h"
9 #include "../include/goya/goya_coresight.h"
10 #include "../include/goya/asic_reg/goya_regs.h"
11 #include "../include/goya/asic_reg/goya_masks.h"
12
13 #include <uapi/drm/habanalabs_accel.h>
14
15 #define GOYA_PLDM_CORESIGHT_TIMEOUT_USEC (CORESIGHT_TIMEOUT_USEC * 100)
16
17 #define SPMU_SECTION_SIZE DMA_CH_0_CS_SPMU_MAX_OFFSET
18 #define SPMU_EVENT_TYPES_OFFSET 0x400
19 #define SPMU_MAX_COUNTERS 6
20
21 static u64 debug_stm_regs[GOYA_STM_LAST + 1] = {
22 [GOYA_STM_CPU] = mmCPU_STM_BASE,
23 [GOYA_STM_DMA_CH_0_CS] = mmDMA_CH_0_CS_STM_BASE,
24 [GOYA_STM_DMA_CH_1_CS] = mmDMA_CH_1_CS_STM_BASE,
25 [GOYA_STM_DMA_CH_2_CS] = mmDMA_CH_2_CS_STM_BASE,
26 [GOYA_STM_DMA_CH_3_CS] = mmDMA_CH_3_CS_STM_BASE,
27 [GOYA_STM_DMA_CH_4_CS] = mmDMA_CH_4_CS_STM_BASE,
28 [GOYA_STM_DMA_MACRO_CS] = mmDMA_MACRO_CS_STM_BASE,
29 [GOYA_STM_MME1_SBA] = mmMME1_SBA_STM_BASE,
30 [GOYA_STM_MME3_SBB] = mmMME3_SBB_STM_BASE,
31 [GOYA_STM_MME4_WACS2] = mmMME4_WACS2_STM_BASE,
32 [GOYA_STM_MME4_WACS] = mmMME4_WACS_STM_BASE,
33 [GOYA_STM_MMU_CS] = mmMMU_CS_STM_BASE,
34 [GOYA_STM_PCIE] = mmPCIE_STM_BASE,
35 [GOYA_STM_PSOC] = mmPSOC_STM_BASE,
36 [GOYA_STM_TPC0_EML] = mmTPC0_EML_STM_BASE,
37 [GOYA_STM_TPC1_EML] = mmTPC1_EML_STM_BASE,
38 [GOYA_STM_TPC2_EML] = mmTPC2_EML_STM_BASE,
39 [GOYA_STM_TPC3_EML] = mmTPC3_EML_STM_BASE,
40 [GOYA_STM_TPC4_EML] = mmTPC4_EML_STM_BASE,
41 [GOYA_STM_TPC5_EML] = mmTPC5_EML_STM_BASE,
42 [GOYA_STM_TPC6_EML] = mmTPC6_EML_STM_BASE,
43 [GOYA_STM_TPC7_EML] = mmTPC7_EML_STM_BASE
44 };
45
46 static u64 debug_etf_regs[GOYA_ETF_LAST + 1] = {
47 [GOYA_ETF_CPU_0] = mmCPU_ETF_0_BASE,
48 [GOYA_ETF_CPU_1] = mmCPU_ETF_1_BASE,
49 [GOYA_ETF_CPU_TRACE] = mmCPU_ETF_TRACE_BASE,
50 [GOYA_ETF_DMA_CH_0_CS] = mmDMA_CH_0_CS_ETF_BASE,
51 [GOYA_ETF_DMA_CH_1_CS] = mmDMA_CH_1_CS_ETF_BASE,
52 [GOYA_ETF_DMA_CH_2_CS] = mmDMA_CH_2_CS_ETF_BASE,
53 [GOYA_ETF_DMA_CH_3_CS] = mmDMA_CH_3_CS_ETF_BASE,
54 [GOYA_ETF_DMA_CH_4_CS] = mmDMA_CH_4_CS_ETF_BASE,
55 [GOYA_ETF_DMA_MACRO_CS] = mmDMA_MACRO_CS_ETF_BASE,
56 [GOYA_ETF_MME1_SBA] = mmMME1_SBA_ETF_BASE,
57 [GOYA_ETF_MME3_SBB] = mmMME3_SBB_ETF_BASE,
58 [GOYA_ETF_MME4_WACS2] = mmMME4_WACS2_ETF_BASE,
59 [GOYA_ETF_MME4_WACS] = mmMME4_WACS_ETF_BASE,
60 [GOYA_ETF_MMU_CS] = mmMMU_CS_ETF_BASE,
61 [GOYA_ETF_PCIE] = mmPCIE_ETF_BASE,
62 [GOYA_ETF_PSOC] = mmPSOC_ETF_BASE,
63 [GOYA_ETF_TPC0_EML] = mmTPC0_EML_ETF_BASE,
64 [GOYA_ETF_TPC1_EML] = mmTPC1_EML_ETF_BASE,
65 [GOYA_ETF_TPC2_EML] = mmTPC2_EML_ETF_BASE,
66 [GOYA_ETF_TPC3_EML] = mmTPC3_EML_ETF_BASE,
67 [GOYA_ETF_TPC4_EML] = mmTPC4_EML_ETF_BASE,
68 [GOYA_ETF_TPC5_EML] = mmTPC5_EML_ETF_BASE,
69 [GOYA_ETF_TPC6_EML] = mmTPC6_EML_ETF_BASE,
70 [GOYA_ETF_TPC7_EML] = mmTPC7_EML_ETF_BASE
71 };
72
73 static u64 debug_funnel_regs[GOYA_FUNNEL_LAST + 1] = {
74 [GOYA_FUNNEL_CPU] = mmCPU_FUNNEL_BASE,
75 [GOYA_FUNNEL_DMA_CH_6_1] = mmDMA_CH_FUNNEL_6_1_BASE,
76 [GOYA_FUNNEL_DMA_MACRO_3_1] = mmDMA_MACRO_FUNNEL_3_1_BASE,
77 [GOYA_FUNNEL_MME0_RTR] = mmMME0_RTR_FUNNEL_BASE,
78 [GOYA_FUNNEL_MME1_RTR] = mmMME1_RTR_FUNNEL_BASE,
79 [GOYA_FUNNEL_MME2_RTR] = mmMME2_RTR_FUNNEL_BASE,
80 [GOYA_FUNNEL_MME3_RTR] = mmMME3_RTR_FUNNEL_BASE,
81 [GOYA_FUNNEL_MME4_RTR] = mmMME4_RTR_FUNNEL_BASE,
82 [GOYA_FUNNEL_MME5_RTR] = mmMME5_RTR_FUNNEL_BASE,
83 [GOYA_FUNNEL_PCIE] = mmPCIE_FUNNEL_BASE,
84 [GOYA_FUNNEL_PSOC] = mmPSOC_FUNNEL_BASE,
85 [GOYA_FUNNEL_TPC0_EML] = mmTPC0_EML_FUNNEL_BASE,
86 [GOYA_FUNNEL_TPC1_EML] = mmTPC1_EML_FUNNEL_BASE,
87 [GOYA_FUNNEL_TPC1_RTR] = mmTPC1_RTR_FUNNEL_BASE,
88 [GOYA_FUNNEL_TPC2_EML] = mmTPC2_EML_FUNNEL_BASE,
89 [GOYA_FUNNEL_TPC2_RTR] = mmTPC2_RTR_FUNNEL_BASE,
90 [GOYA_FUNNEL_TPC3_EML] = mmTPC3_EML_FUNNEL_BASE,
91 [GOYA_FUNNEL_TPC3_RTR] = mmTPC3_RTR_FUNNEL_BASE,
92 [GOYA_FUNNEL_TPC4_EML] = mmTPC4_EML_FUNNEL_BASE,
93 [GOYA_FUNNEL_TPC4_RTR] = mmTPC4_RTR_FUNNEL_BASE,
94 [GOYA_FUNNEL_TPC5_EML] = mmTPC5_EML_FUNNEL_BASE,
95 [GOYA_FUNNEL_TPC5_RTR] = mmTPC5_RTR_FUNNEL_BASE,
96 [GOYA_FUNNEL_TPC6_EML] = mmTPC6_EML_FUNNEL_BASE,
97 [GOYA_FUNNEL_TPC6_RTR] = mmTPC6_RTR_FUNNEL_BASE,
98 [GOYA_FUNNEL_TPC7_EML] = mmTPC7_EML_FUNNEL_BASE
99 };
100
101 static u64 debug_bmon_regs[GOYA_BMON_LAST + 1] = {
102 [GOYA_BMON_CPU_RD] = mmCPU_RD_BMON_BASE,
103 [GOYA_BMON_CPU_WR] = mmCPU_WR_BMON_BASE,
104 [GOYA_BMON_DMA_CH_0_0] = mmDMA_CH_0_BMON_0_BASE,
105 [GOYA_BMON_DMA_CH_0_1] = mmDMA_CH_0_BMON_1_BASE,
106 [GOYA_BMON_DMA_CH_1_0] = mmDMA_CH_1_BMON_0_BASE,
107 [GOYA_BMON_DMA_CH_1_1] = mmDMA_CH_1_BMON_1_BASE,
108 [GOYA_BMON_DMA_CH_2_0] = mmDMA_CH_2_BMON_0_BASE,
109 [GOYA_BMON_DMA_CH_2_1] = mmDMA_CH_2_BMON_1_BASE,
110 [GOYA_BMON_DMA_CH_3_0] = mmDMA_CH_3_BMON_0_BASE,
111 [GOYA_BMON_DMA_CH_3_1] = mmDMA_CH_3_BMON_1_BASE,
112 [GOYA_BMON_DMA_CH_4_0] = mmDMA_CH_4_BMON_0_BASE,
113 [GOYA_BMON_DMA_CH_4_1] = mmDMA_CH_4_BMON_1_BASE,
114 [GOYA_BMON_DMA_MACRO_0] = mmDMA_MACRO_BMON_0_BASE,
115 [GOYA_BMON_DMA_MACRO_1] = mmDMA_MACRO_BMON_1_BASE,
116 [GOYA_BMON_DMA_MACRO_2] = mmDMA_MACRO_BMON_2_BASE,
117 [GOYA_BMON_DMA_MACRO_3] = mmDMA_MACRO_BMON_3_BASE,
118 [GOYA_BMON_DMA_MACRO_4] = mmDMA_MACRO_BMON_4_BASE,
119 [GOYA_BMON_DMA_MACRO_5] = mmDMA_MACRO_BMON_5_BASE,
120 [GOYA_BMON_DMA_MACRO_6] = mmDMA_MACRO_BMON_6_BASE,
121 [GOYA_BMON_DMA_MACRO_7] = mmDMA_MACRO_BMON_7_BASE,
122 [GOYA_BMON_MME1_SBA_0] = mmMME1_SBA_BMON0_BASE,
123 [GOYA_BMON_MME1_SBA_1] = mmMME1_SBA_BMON1_BASE,
124 [GOYA_BMON_MME3_SBB_0] = mmMME3_SBB_BMON0_BASE,
125 [GOYA_BMON_MME3_SBB_1] = mmMME3_SBB_BMON1_BASE,
126 [GOYA_BMON_MME4_WACS2_0] = mmMME4_WACS2_BMON0_BASE,
127 [GOYA_BMON_MME4_WACS2_1] = mmMME4_WACS2_BMON1_BASE,
128 [GOYA_BMON_MME4_WACS2_2] = mmMME4_WACS2_BMON2_BASE,
129 [GOYA_BMON_MME4_WACS_0] = mmMME4_WACS_BMON0_BASE,
130 [GOYA_BMON_MME4_WACS_1] = mmMME4_WACS_BMON1_BASE,
131 [GOYA_BMON_MME4_WACS_2] = mmMME4_WACS_BMON2_BASE,
132 [GOYA_BMON_MME4_WACS_3] = mmMME4_WACS_BMON3_BASE,
133 [GOYA_BMON_MME4_WACS_4] = mmMME4_WACS_BMON4_BASE,
134 [GOYA_BMON_MME4_WACS_5] = mmMME4_WACS_BMON5_BASE,
135 [GOYA_BMON_MME4_WACS_6] = mmMME4_WACS_BMON6_BASE,
136 [GOYA_BMON_MMU_0] = mmMMU_BMON_0_BASE,
137 [GOYA_BMON_MMU_1] = mmMMU_BMON_1_BASE,
138 [GOYA_BMON_PCIE_MSTR_RD] = mmPCIE_BMON_MSTR_RD_BASE,
139 [GOYA_BMON_PCIE_MSTR_WR] = mmPCIE_BMON_MSTR_WR_BASE,
140 [GOYA_BMON_PCIE_SLV_RD] = mmPCIE_BMON_SLV_RD_BASE,
141 [GOYA_BMON_PCIE_SLV_WR] = mmPCIE_BMON_SLV_WR_BASE,
142 [GOYA_BMON_TPC0_EML_0] = mmTPC0_EML_BUSMON_0_BASE,
143 [GOYA_BMON_TPC0_EML_1] = mmTPC0_EML_BUSMON_1_BASE,
144 [GOYA_BMON_TPC0_EML_2] = mmTPC0_EML_BUSMON_2_BASE,
145 [GOYA_BMON_TPC0_EML_3] = mmTPC0_EML_BUSMON_3_BASE,
146 [GOYA_BMON_TPC1_EML_0] = mmTPC1_EML_BUSMON_0_BASE,
147 [GOYA_BMON_TPC1_EML_1] = mmTPC1_EML_BUSMON_1_BASE,
148 [GOYA_BMON_TPC1_EML_2] = mmTPC1_EML_BUSMON_2_BASE,
149 [GOYA_BMON_TPC1_EML_3] = mmTPC1_EML_BUSMON_3_BASE,
150 [GOYA_BMON_TPC2_EML_0] = mmTPC2_EML_BUSMON_0_BASE,
151 [GOYA_BMON_TPC2_EML_1] = mmTPC2_EML_BUSMON_1_BASE,
152 [GOYA_BMON_TPC2_EML_2] = mmTPC2_EML_BUSMON_2_BASE,
153 [GOYA_BMON_TPC2_EML_3] = mmTPC2_EML_BUSMON_3_BASE,
154 [GOYA_BMON_TPC3_EML_0] = mmTPC3_EML_BUSMON_0_BASE,
155 [GOYA_BMON_TPC3_EML_1] = mmTPC3_EML_BUSMON_1_BASE,
156 [GOYA_BMON_TPC3_EML_2] = mmTPC3_EML_BUSMON_2_BASE,
157 [GOYA_BMON_TPC3_EML_3] = mmTPC3_EML_BUSMON_3_BASE,
158 [GOYA_BMON_TPC4_EML_0] = mmTPC4_EML_BUSMON_0_BASE,
159 [GOYA_BMON_TPC4_EML_1] = mmTPC4_EML_BUSMON_1_BASE,
160 [GOYA_BMON_TPC4_EML_2] = mmTPC4_EML_BUSMON_2_BASE,
161 [GOYA_BMON_TPC4_EML_3] = mmTPC4_EML_BUSMON_3_BASE,
162 [GOYA_BMON_TPC5_EML_0] = mmTPC5_EML_BUSMON_0_BASE,
163 [GOYA_BMON_TPC5_EML_1] = mmTPC5_EML_BUSMON_1_BASE,
164 [GOYA_BMON_TPC5_EML_2] = mmTPC5_EML_BUSMON_2_BASE,
165 [GOYA_BMON_TPC5_EML_3] = mmTPC5_EML_BUSMON_3_BASE,
166 [GOYA_BMON_TPC6_EML_0] = mmTPC6_EML_BUSMON_0_BASE,
167 [GOYA_BMON_TPC6_EML_1] = mmTPC6_EML_BUSMON_1_BASE,
168 [GOYA_BMON_TPC6_EML_2] = mmTPC6_EML_BUSMON_2_BASE,
169 [GOYA_BMON_TPC6_EML_3] = mmTPC6_EML_BUSMON_3_BASE,
170 [GOYA_BMON_TPC7_EML_0] = mmTPC7_EML_BUSMON_0_BASE,
171 [GOYA_BMON_TPC7_EML_1] = mmTPC7_EML_BUSMON_1_BASE,
172 [GOYA_BMON_TPC7_EML_2] = mmTPC7_EML_BUSMON_2_BASE,
173 [GOYA_BMON_TPC7_EML_3] = mmTPC7_EML_BUSMON_3_BASE
174 };
175
176 static u64 debug_spmu_regs[GOYA_SPMU_LAST + 1] = {
177 [GOYA_SPMU_DMA_CH_0_CS] = mmDMA_CH_0_CS_SPMU_BASE,
178 [GOYA_SPMU_DMA_CH_1_CS] = mmDMA_CH_1_CS_SPMU_BASE,
179 [GOYA_SPMU_DMA_CH_2_CS] = mmDMA_CH_2_CS_SPMU_BASE,
180 [GOYA_SPMU_DMA_CH_3_CS] = mmDMA_CH_3_CS_SPMU_BASE,
181 [GOYA_SPMU_DMA_CH_4_CS] = mmDMA_CH_4_CS_SPMU_BASE,
182 [GOYA_SPMU_DMA_MACRO_CS] = mmDMA_MACRO_CS_SPMU_BASE,
183 [GOYA_SPMU_MME1_SBA] = mmMME1_SBA_SPMU_BASE,
184 [GOYA_SPMU_MME3_SBB] = mmMME3_SBB_SPMU_BASE,
185 [GOYA_SPMU_MME4_WACS2] = mmMME4_WACS2_SPMU_BASE,
186 [GOYA_SPMU_MME4_WACS] = mmMME4_WACS_SPMU_BASE,
187 [GOYA_SPMU_MMU_CS] = mmMMU_CS_SPMU_BASE,
188 [GOYA_SPMU_PCIE] = mmPCIE_SPMU_BASE,
189 [GOYA_SPMU_TPC0_EML] = mmTPC0_EML_SPMU_BASE,
190 [GOYA_SPMU_TPC1_EML] = mmTPC1_EML_SPMU_BASE,
191 [GOYA_SPMU_TPC2_EML] = mmTPC2_EML_SPMU_BASE,
192 [GOYA_SPMU_TPC3_EML] = mmTPC3_EML_SPMU_BASE,
193 [GOYA_SPMU_TPC4_EML] = mmTPC4_EML_SPMU_BASE,
194 [GOYA_SPMU_TPC5_EML] = mmTPC5_EML_SPMU_BASE,
195 [GOYA_SPMU_TPC6_EML] = mmTPC6_EML_SPMU_BASE,
196 [GOYA_SPMU_TPC7_EML] = mmTPC7_EML_SPMU_BASE
197 };
198
goya_coresight_timeout(struct hl_device * hdev,u64 addr,int position,bool up)199 static int goya_coresight_timeout(struct hl_device *hdev, u64 addr,
200 int position, bool up)
201 {
202 int rc;
203 u32 val, timeout_usec;
204
205 if (hdev->pldm)
206 timeout_usec = GOYA_PLDM_CORESIGHT_TIMEOUT_USEC;
207 else
208 timeout_usec = CORESIGHT_TIMEOUT_USEC;
209
210 rc = hl_poll_timeout(
211 hdev,
212 addr,
213 val,
214 up ? val & BIT(position) : !(val & BIT(position)),
215 1000,
216 timeout_usec);
217
218 if (rc) {
219 dev_err(hdev->dev,
220 "Timeout while waiting for coresight, addr: 0x%llx, position: %d, up: %d\n",
221 addr, position, up);
222 return -EFAULT;
223 }
224
225 return 0;
226 }
227
goya_config_stm(struct hl_device * hdev,struct hl_debug_params * params)228 static int goya_config_stm(struct hl_device *hdev,
229 struct hl_debug_params *params)
230 {
231 struct hl_debug_params_stm *input;
232 u64 base_reg;
233 u32 frequency;
234 int rc;
235
236 if (params->reg_idx >= ARRAY_SIZE(debug_stm_regs)) {
237 dev_err(hdev->dev, "Invalid register index in STM\n");
238 return -EINVAL;
239 }
240
241 base_reg = debug_stm_regs[params->reg_idx] - CFG_BASE;
242
243 WREG32(base_reg + 0xFB0, CORESIGHT_UNLOCK);
244
245 if (params->enable) {
246 input = params->input;
247
248 if (!input)
249 return -EINVAL;
250
251 WREG32(base_reg + 0xE80, 0x80004);
252 WREG32(base_reg + 0xD64, 7);
253 WREG32(base_reg + 0xD60, 0);
254 WREG32(base_reg + 0xD00, lower_32_bits(input->he_mask));
255 WREG32(base_reg + 0xD20, lower_32_bits(input->sp_mask));
256 WREG32(base_reg + 0xD60, 1);
257 WREG32(base_reg + 0xD00, upper_32_bits(input->he_mask));
258 WREG32(base_reg + 0xD20, upper_32_bits(input->sp_mask));
259 WREG32(base_reg + 0xE70, 0x10);
260 WREG32(base_reg + 0xE60, 0);
261 WREG32(base_reg + 0xE64, 0x420000);
262 WREG32(base_reg + 0xE00, 0xFFFFFFFF);
263 WREG32(base_reg + 0xE20, 0xFFFFFFFF);
264 WREG32(base_reg + 0xEF4, input->id);
265 WREG32(base_reg + 0xDF4, 0x80);
266 frequency = hdev->asic_prop.psoc_timestamp_frequency;
267 if (frequency == 0)
268 frequency = input->frequency;
269 WREG32(base_reg + 0xE8C, frequency);
270 WREG32(base_reg + 0xE90, 0x7FF);
271 WREG32(base_reg + 0xE80, 0x27 | (input->id << 16));
272 } else {
273 WREG32(base_reg + 0xE80, 4);
274 WREG32(base_reg + 0xD64, 0);
275 WREG32(base_reg + 0xD60, 1);
276 WREG32(base_reg + 0xD00, 0);
277 WREG32(base_reg + 0xD20, 0);
278 WREG32(base_reg + 0xD60, 0);
279 WREG32(base_reg + 0xE20, 0);
280 WREG32(base_reg + 0xE00, 0);
281 WREG32(base_reg + 0xDF4, 0x80);
282 WREG32(base_reg + 0xE70, 0);
283 WREG32(base_reg + 0xE60, 0);
284 WREG32(base_reg + 0xE64, 0);
285 WREG32(base_reg + 0xE8C, 0);
286
287 rc = goya_coresight_timeout(hdev, base_reg + 0xE80, 23, false);
288 if (rc) {
289 dev_err(hdev->dev,
290 "Failed to disable STM on timeout, error %d\n",
291 rc);
292 return rc;
293 }
294
295 WREG32(base_reg + 0xE80, 4);
296 }
297
298 return 0;
299 }
300
goya_config_etf(struct hl_device * hdev,struct hl_debug_params * params)301 static int goya_config_etf(struct hl_device *hdev,
302 struct hl_debug_params *params)
303 {
304 struct hl_debug_params_etf *input;
305 u64 base_reg;
306 u32 val;
307 int rc;
308
309 if (params->reg_idx >= ARRAY_SIZE(debug_etf_regs)) {
310 dev_err(hdev->dev, "Invalid register index in ETF\n");
311 return -EINVAL;
312 }
313
314 base_reg = debug_etf_regs[params->reg_idx] - CFG_BASE;
315
316 WREG32(base_reg + 0xFB0, CORESIGHT_UNLOCK);
317
318 val = RREG32(base_reg + 0x20);
319
320 if ((!params->enable && val == 0x0) || (params->enable && val != 0x0))
321 return 0;
322
323 val = RREG32(base_reg + 0x304);
324 val |= 0x1000;
325 WREG32(base_reg + 0x304, val);
326 val |= 0x40;
327 WREG32(base_reg + 0x304, val);
328
329 rc = goya_coresight_timeout(hdev, base_reg + 0x304, 6, false);
330 if (rc) {
331 dev_err(hdev->dev,
332 "Failed to %s ETF on timeout, error %d\n",
333 params->enable ? "enable" : "disable", rc);
334 return rc;
335 }
336
337 rc = goya_coresight_timeout(hdev, base_reg + 0xC, 2, true);
338 if (rc) {
339 dev_err(hdev->dev,
340 "Failed to %s ETF on timeout, error %d\n",
341 params->enable ? "enable" : "disable", rc);
342 return rc;
343 }
344
345 WREG32(base_reg + 0x20, 0);
346
347 if (params->enable) {
348 input = params->input;
349
350 if (!input)
351 return -EINVAL;
352
353 WREG32(base_reg + 0x34, 0x3FFC);
354 WREG32(base_reg + 0x28, input->sink_mode);
355 WREG32(base_reg + 0x304, 0x4001);
356 WREG32(base_reg + 0x308, 0xA);
357 WREG32(base_reg + 0x20, 1);
358 } else {
359 WREG32(base_reg + 0x34, 0);
360 WREG32(base_reg + 0x28, 0);
361 WREG32(base_reg + 0x304, 0);
362 }
363
364 return 0;
365 }
366
goya_etr_validate_address(struct hl_device * hdev,u64 addr,u64 size)367 static int goya_etr_validate_address(struct hl_device *hdev, u64 addr,
368 u64 size)
369 {
370 struct asic_fixed_properties *prop = &hdev->asic_prop;
371 u64 range_start, range_end;
372
373 if (addr > (addr + size)) {
374 dev_err(hdev->dev,
375 "ETR buffer size %llu overflow\n", size);
376 return false;
377 }
378
379 range_start = prop->dmmu.start_addr;
380 range_end = prop->dmmu.end_addr;
381
382 return hl_mem_area_inside_range(addr, size, range_start, range_end);
383 }
384
goya_config_etr(struct hl_device * hdev,struct hl_debug_params * params)385 static int goya_config_etr(struct hl_device *hdev,
386 struct hl_debug_params *params)
387 {
388 struct hl_debug_params_etr *input;
389 u32 val;
390 int rc;
391
392 WREG32(mmPSOC_ETR_LAR, CORESIGHT_UNLOCK);
393
394 val = RREG32(mmPSOC_ETR_CTL);
395
396 if ((!params->enable && val == 0x0) || (params->enable && val != 0x0))
397 return 0;
398
399 val = RREG32(mmPSOC_ETR_FFCR);
400 val |= 0x1000;
401 WREG32(mmPSOC_ETR_FFCR, val);
402 val |= 0x40;
403 WREG32(mmPSOC_ETR_FFCR, val);
404
405 rc = goya_coresight_timeout(hdev, mmPSOC_ETR_FFCR, 6, false);
406 if (rc) {
407 dev_err(hdev->dev, "Failed to %s ETR on timeout, error %d\n",
408 params->enable ? "enable" : "disable", rc);
409 return rc;
410 }
411
412 rc = goya_coresight_timeout(hdev, mmPSOC_ETR_STS, 2, true);
413 if (rc) {
414 dev_err(hdev->dev, "Failed to %s ETR on timeout, error %d\n",
415 params->enable ? "enable" : "disable", rc);
416 return rc;
417 }
418
419 WREG32(mmPSOC_ETR_CTL, 0);
420
421 if (params->enable) {
422 input = params->input;
423
424 if (!input)
425 return -EINVAL;
426
427 if (input->buffer_size == 0) {
428 dev_err(hdev->dev,
429 "ETR buffer size should be bigger than 0\n");
430 return -EINVAL;
431 }
432
433 if (!goya_etr_validate_address(hdev,
434 input->buffer_address, input->buffer_size)) {
435 dev_err(hdev->dev, "buffer address is not valid\n");
436 return -EINVAL;
437 }
438
439 WREG32(mmPSOC_ETR_BUFWM, 0x3FFC);
440 WREG32(mmPSOC_ETR_RSZ, input->buffer_size);
441 WREG32(mmPSOC_ETR_MODE, input->sink_mode);
442 if (!hdev->asic_prop.fw_security_enabled) {
443 /* make ETR not privileged */
444 val = FIELD_PREP(PSOC_ETR_AXICTL_PROTCTRLBIT0_MASK, 0);
445 /* make ETR non-secured (inverted logic) */
446 val |= FIELD_PREP(PSOC_ETR_AXICTL_PROTCTRLBIT1_MASK, 1);
447 /* burst size 8 */
448 val |= FIELD_PREP(PSOC_ETR_AXICTL_WRBURSTLEN_MASK, 7);
449 WREG32(mmPSOC_ETR_AXICTL, val);
450 }
451 WREG32(mmPSOC_ETR_DBALO,
452 lower_32_bits(input->buffer_address));
453 WREG32(mmPSOC_ETR_DBAHI,
454 upper_32_bits(input->buffer_address));
455 WREG32(mmPSOC_ETR_FFCR, 3);
456 WREG32(mmPSOC_ETR_PSCR, 0xA);
457 WREG32(mmPSOC_ETR_CTL, 1);
458 } else {
459 WREG32(mmPSOC_ETR_BUFWM, 0);
460 WREG32(mmPSOC_ETR_RSZ, 0x400);
461 WREG32(mmPSOC_ETR_DBALO, 0);
462 WREG32(mmPSOC_ETR_DBAHI, 0);
463 WREG32(mmPSOC_ETR_PSCR, 0);
464 WREG32(mmPSOC_ETR_MODE, 0);
465 WREG32(mmPSOC_ETR_FFCR, 0);
466
467 if (params->output_size >= sizeof(u64)) {
468 u32 rwp, rwphi;
469
470 /*
471 * The trace buffer address is 40 bits wide. The end of
472 * the buffer is set in the RWP register (lower 32
473 * bits), and in the RWPHI register (upper 8 bits).
474 */
475 rwp = RREG32(mmPSOC_ETR_RWP);
476 rwphi = RREG32(mmPSOC_ETR_RWPHI) & 0xff;
477 *(u64 *) params->output = ((u64) rwphi << 32) | rwp;
478 }
479 }
480
481 return 0;
482 }
483
goya_config_funnel(struct hl_device * hdev,struct hl_debug_params * params)484 static int goya_config_funnel(struct hl_device *hdev,
485 struct hl_debug_params *params)
486 {
487 u64 base_reg;
488
489 if (params->reg_idx >= ARRAY_SIZE(debug_funnel_regs)) {
490 dev_err(hdev->dev, "Invalid register index in FUNNEL\n");
491 return -EINVAL;
492 }
493
494 base_reg = debug_funnel_regs[params->reg_idx] - CFG_BASE;
495
496 WREG32(base_reg + 0xFB0, CORESIGHT_UNLOCK);
497
498 WREG32(base_reg, params->enable ? 0x33F : 0);
499
500 return 0;
501 }
502
goya_config_bmon(struct hl_device * hdev,struct hl_debug_params * params)503 static int goya_config_bmon(struct hl_device *hdev,
504 struct hl_debug_params *params)
505 {
506 struct hl_debug_params_bmon *input;
507 u64 base_reg;
508 u32 pcie_base = 0;
509
510 if (params->reg_idx >= ARRAY_SIZE(debug_bmon_regs)) {
511 dev_err(hdev->dev, "Invalid register index in BMON\n");
512 return -EINVAL;
513 }
514
515 base_reg = debug_bmon_regs[params->reg_idx] - CFG_BASE;
516
517 WREG32(base_reg + 0x104, 1);
518
519 if (params->enable) {
520 input = params->input;
521
522 if (!input)
523 return -EINVAL;
524
525 WREG32(base_reg + 0x200, lower_32_bits(input->start_addr0));
526 WREG32(base_reg + 0x204, upper_32_bits(input->start_addr0));
527 WREG32(base_reg + 0x208, lower_32_bits(input->addr_mask0));
528 WREG32(base_reg + 0x20C, upper_32_bits(input->addr_mask0));
529 WREG32(base_reg + 0x240, lower_32_bits(input->start_addr1));
530 WREG32(base_reg + 0x244, upper_32_bits(input->start_addr1));
531 WREG32(base_reg + 0x248, lower_32_bits(input->addr_mask1));
532 WREG32(base_reg + 0x24C, upper_32_bits(input->addr_mask1));
533 WREG32(base_reg + 0x224, 0);
534 WREG32(base_reg + 0x234, 0);
535 WREG32(base_reg + 0x30C, input->bw_win);
536 WREG32(base_reg + 0x308, input->win_capture);
537
538 /* PCIE IF BMON bug WA */
539 if (params->reg_idx != GOYA_BMON_PCIE_MSTR_RD &&
540 params->reg_idx != GOYA_BMON_PCIE_MSTR_WR &&
541 params->reg_idx != GOYA_BMON_PCIE_SLV_RD &&
542 params->reg_idx != GOYA_BMON_PCIE_SLV_WR)
543 pcie_base = 0xA000000;
544
545 WREG32(base_reg + 0x700, pcie_base | 0xB00 | (input->id << 12));
546 WREG32(base_reg + 0x708, pcie_base | 0xA00 | (input->id << 12));
547 WREG32(base_reg + 0x70C, pcie_base | 0xC00 | (input->id << 12));
548
549 WREG32(base_reg + 0x100, 0x11);
550 WREG32(base_reg + 0x304, 0x1);
551 } else {
552 WREG32(base_reg + 0x200, 0);
553 WREG32(base_reg + 0x204, 0);
554 WREG32(base_reg + 0x208, 0xFFFFFFFF);
555 WREG32(base_reg + 0x20C, 0xFFFFFFFF);
556 WREG32(base_reg + 0x240, 0);
557 WREG32(base_reg + 0x244, 0);
558 WREG32(base_reg + 0x248, 0xFFFFFFFF);
559 WREG32(base_reg + 0x24C, 0xFFFFFFFF);
560 WREG32(base_reg + 0x224, 0xFFFFFFFF);
561 WREG32(base_reg + 0x234, 0x1070F);
562 WREG32(base_reg + 0x30C, 0);
563 WREG32(base_reg + 0x308, 0xFFFF);
564 WREG32(base_reg + 0x700, 0xA000B00);
565 WREG32(base_reg + 0x708, 0xA000A00);
566 WREG32(base_reg + 0x70C, 0xA000C00);
567 WREG32(base_reg + 0x100, 1);
568 WREG32(base_reg + 0x304, 0);
569 WREG32(base_reg + 0x104, 0);
570 }
571
572 return 0;
573 }
574
goya_config_spmu(struct hl_device * hdev,struct hl_debug_params * params)575 static int goya_config_spmu(struct hl_device *hdev,
576 struct hl_debug_params *params)
577 {
578 u64 base_reg;
579 u64 *output;
580 u32 output_arr_len;
581 u32 events_num;
582 u32 overflow_idx;
583 u32 cycle_cnt_idx;
584 int i;
585
586 if (params->reg_idx >= ARRAY_SIZE(debug_spmu_regs)) {
587 dev_err(hdev->dev, "Invalid register index in SPMU\n");
588 return -EINVAL;
589 }
590
591 base_reg = debug_spmu_regs[params->reg_idx] - CFG_BASE;
592
593 if (params->enable) {
594 struct hl_debug_params_spmu *input = params->input;
595
596 if (!input)
597 return -EINVAL;
598
599 if (input->event_types_num < 3) {
600 dev_err(hdev->dev,
601 "not enough event types values for SPMU enable\n");
602 return -EINVAL;
603 }
604
605 if (input->event_types_num > SPMU_MAX_COUNTERS) {
606 dev_err(hdev->dev,
607 "too many event types values for SPMU enable\n");
608 return -EINVAL;
609 }
610
611 WREG32(base_reg + 0xE04, 0x41013046);
612 WREG32(base_reg + 0xE04, 0x41013040);
613
614 for (i = 0 ; i < input->event_types_num ; i++)
615 WREG32(base_reg + SPMU_EVENT_TYPES_OFFSET + i * 4,
616 input->event_types[i]);
617
618 WREG32(base_reg + 0xE04, 0x41013041);
619 WREG32(base_reg + 0xC00, 0x8000003F);
620 } else {
621 output = params->output;
622 output_arr_len = params->output_size / 8;
623 events_num = output_arr_len - 2;
624 overflow_idx = output_arr_len - 2;
625 cycle_cnt_idx = output_arr_len - 1;
626
627 if (!output)
628 return -EINVAL;
629
630 if (output_arr_len < 3) {
631 dev_err(hdev->dev,
632 "not enough values for SPMU disable\n");
633 return -EINVAL;
634 }
635
636 if (events_num > SPMU_MAX_COUNTERS) {
637 dev_err(hdev->dev,
638 "too many events values for SPMU disable\n");
639 return -EINVAL;
640 }
641
642 WREG32(base_reg + 0xE04, 0x41013040);
643
644 for (i = 0 ; i < events_num ; i++)
645 output[i] = RREG32(base_reg + i * 8);
646
647 output[overflow_idx] = RREG32(base_reg + 0xCC0);
648
649 output[cycle_cnt_idx] = RREG32(base_reg + 0xFC);
650 output[cycle_cnt_idx] <<= 32;
651 output[cycle_cnt_idx] |= RREG32(base_reg + 0xF8);
652
653 WREG32(base_reg + 0xCC0, 0);
654 }
655
656 return 0;
657 }
658
goya_debug_coresight(struct hl_device * hdev,struct hl_ctx * ctx,void * data)659 int goya_debug_coresight(struct hl_device *hdev, struct hl_ctx *ctx, void *data)
660 {
661 struct hl_debug_params *params = data;
662 int rc = 0;
663
664 switch (params->op) {
665 case HL_DEBUG_OP_STM:
666 rc = goya_config_stm(hdev, params);
667 break;
668 case HL_DEBUG_OP_ETF:
669 rc = goya_config_etf(hdev, params);
670 break;
671 case HL_DEBUG_OP_ETR:
672 rc = goya_config_etr(hdev, params);
673 break;
674 case HL_DEBUG_OP_FUNNEL:
675 rc = goya_config_funnel(hdev, params);
676 break;
677 case HL_DEBUG_OP_BMON:
678 rc = goya_config_bmon(hdev, params);
679 break;
680 case HL_DEBUG_OP_SPMU:
681 rc = goya_config_spmu(hdev, params);
682 break;
683 case HL_DEBUG_OP_TIMESTAMP:
684 /* Do nothing as this opcode is deprecated */
685 break;
686
687 default:
688 dev_err(hdev->dev, "Unknown coresight id %d\n", params->op);
689 return -EINVAL;
690 }
691
692 /* Perform read from the device to flush all configuration */
693 RREG32(mmPCIE_DBI_DEVICE_ID_VENDOR_ID_REG);
694
695 return rc;
696 }
697
goya_halt_coresight(struct hl_device * hdev,struct hl_ctx * ctx)698 void goya_halt_coresight(struct hl_device *hdev, struct hl_ctx *ctx)
699 {
700 struct hl_debug_params params = {};
701 int i, rc;
702
703 for (i = GOYA_ETF_FIRST ; i <= GOYA_ETF_LAST ; i++) {
704 params.reg_idx = i;
705 rc = goya_config_etf(hdev, ¶ms);
706 if (rc)
707 dev_err(hdev->dev, "halt ETF failed, %d/%d\n", rc, i);
708 }
709
710 rc = goya_config_etr(hdev, ¶ms);
711 if (rc)
712 dev_err(hdev->dev, "halt ETR failed, %d\n", rc);
713 }
714