1 /*
2 * Copyright © 2014-2017 Broadcom
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 /**
25 * @file v3dx_simulator.c
26 *
27 * Implements the actual HW interaction betweeh the GL driver's VC5 simulator and the simulator.
28 *
29 * The register headers between V3D versions will have conflicting defines, so
30 * all register interactions appear in this file and are compiled per V3D version
31 * we support.
32 */
33
34 #ifdef USE_V3D_SIMULATOR
35
36 #include <assert.h>
37 #include <stdbool.h>
38 #include <stdio.h>
39
40 #include "v3d_simulator.h"
41 #include "v3d_simulator_wrapper.h"
42
43 #include "util/macros.h"
44 #include "drm-uapi/v3d_drm.h"
45
46 #define HW_REGISTER_RO(x) (x)
47 #define HW_REGISTER_RW(x) (x)
48 #if V3D_VERSION >= 41
49 #include "libs/core/v3d/registers/4.1.35.0/v3d.h"
50 #else
51 #include "libs/core/v3d/registers/3.3.0.0/v3d.h"
52 #endif
53
54 #define V3D_WRITE(reg, val) v3d_hw_write_reg(v3d, reg, val)
55 #define V3D_READ(reg) v3d_hw_read_reg(v3d, reg)
56
57 static void
v3d_invalidate_l3(struct v3d_hw * v3d)58 v3d_invalidate_l3(struct v3d_hw *v3d)
59 {
60 if (!v3d_hw_has_gca(v3d))
61 return;
62
63 #if V3D_VERSION < 40
64 uint32_t gca_ctrl = V3D_READ(V3D_GCA_CACHE_CTRL);
65
66 V3D_WRITE(V3D_GCA_CACHE_CTRL, gca_ctrl | V3D_GCA_CACHE_CTRL_FLUSH_SET);
67 V3D_WRITE(V3D_GCA_CACHE_CTRL, gca_ctrl & ~V3D_GCA_CACHE_CTRL_FLUSH_SET);
68 #endif
69 }
70
71 /* Invalidates the L2C cache. This is a read-only cache for uniforms and instructions. */
72 static void
v3d_invalidate_l2c(struct v3d_hw * v3d)73 v3d_invalidate_l2c(struct v3d_hw *v3d)
74 {
75 if (V3D_VERSION >= 33)
76 return;
77
78 V3D_WRITE(V3D_CTL_0_L2CACTL,
79 V3D_CTL_0_L2CACTL_L2CCLR_SET |
80 V3D_CTL_0_L2CACTL_L2CENA_SET);
81 }
82
83 /* Invalidates texture L2 cachelines */
84 static void
v3d_invalidate_l2t(struct v3d_hw * v3d)85 v3d_invalidate_l2t(struct v3d_hw *v3d)
86 {
87 V3D_WRITE(V3D_CTL_0_L2TFLSTA, 0);
88 V3D_WRITE(V3D_CTL_0_L2TFLEND, ~0);
89 V3D_WRITE(V3D_CTL_0_L2TCACTL,
90 V3D_CTL_0_L2TCACTL_L2TFLS_SET |
91 (0 << V3D_CTL_0_L2TCACTL_L2TFLM_LSB));
92 }
93
94 /* Flushes dirty texture cachelines from the L1 write combiner */
95 static void
v3d_flush_l1td(struct v3d_hw * v3d)96 v3d_flush_l1td(struct v3d_hw *v3d)
97 {
98 V3D_WRITE(V3D_CTL_0_L2TCACTL,
99 V3D_CTL_0_L2TCACTL_TMUWCF_SET);
100
101 assert(!(V3D_READ(V3D_CTL_0_L2TCACTL) & V3D_CTL_0_L2TCACTL_L2TFLS_SET));
102 }
103
104 /* Flushes dirty texture L2 cachelines */
105 static void
v3d_flush_l2t(struct v3d_hw * v3d)106 v3d_flush_l2t(struct v3d_hw *v3d)
107 {
108 V3D_WRITE(V3D_CTL_0_L2TFLSTA, 0);
109 V3D_WRITE(V3D_CTL_0_L2TFLEND, ~0);
110 V3D_WRITE(V3D_CTL_0_L2TCACTL,
111 V3D_CTL_0_L2TCACTL_L2TFLS_SET |
112 (2 << V3D_CTL_0_L2TCACTL_L2TFLM_LSB));
113
114 assert(!(V3D_READ(V3D_CTL_0_L2TCACTL) & V3D_CTL_0_L2TCACTL_L2TFLS_SET));
115 }
116
117 /* Invalidates the slice caches. These are read-only caches. */
118 static void
v3d_invalidate_slices(struct v3d_hw * v3d)119 v3d_invalidate_slices(struct v3d_hw *v3d)
120 {
121 V3D_WRITE(V3D_CTL_0_SLCACTL, ~0);
122 }
123
124 static void
v3d_invalidate_caches(struct v3d_hw * v3d)125 v3d_invalidate_caches(struct v3d_hw *v3d)
126 {
127 v3d_invalidate_l3(v3d);
128 v3d_invalidate_l2c(v3d);
129 v3d_invalidate_l2t(v3d);
130 v3d_invalidate_slices(v3d);
131 }
132
133 static uint32_t g_gmp_ofs;
134 static void
v3d_reload_gmp(struct v3d_hw * v3d)135 v3d_reload_gmp(struct v3d_hw *v3d)
136 {
137 /* Completely reset the GMP. */
138 V3D_WRITE(V3D_GMP_CFG,
139 V3D_GMP_CFG_PROTENABLE_SET);
140 V3D_WRITE(V3D_GMP_TABLE_ADDR, g_gmp_ofs);
141 V3D_WRITE(V3D_GMP_CLEAR_LOAD, ~0);
142 while (V3D_READ(V3D_GMP_STATUS) &
143 V3D_GMP_STATUS_CFG_BUSY_SET) {
144 ;
145 }
146 }
147
148 static UNUSED void
v3d_flush_caches(struct v3d_hw * v3d)149 v3d_flush_caches(struct v3d_hw *v3d)
150 {
151 v3d_flush_l1td(v3d);
152 v3d_flush_l2t(v3d);
153 }
154
155 int
v3dX(simulator_submit_tfu_ioctl)156 v3dX(simulator_submit_tfu_ioctl)(struct v3d_hw *v3d,
157 struct drm_v3d_submit_tfu *args)
158 {
159 int last_vtct = V3D_READ(V3D_TFU_CS) & V3D_TFU_CS_CVTCT_SET;
160
161 V3D_WRITE(V3D_TFU_IIA, args->iia);
162 V3D_WRITE(V3D_TFU_IIS, args->iis);
163 V3D_WRITE(V3D_TFU_ICA, args->ica);
164 V3D_WRITE(V3D_TFU_IUA, args->iua);
165 V3D_WRITE(V3D_TFU_IOA, args->ioa);
166 V3D_WRITE(V3D_TFU_IOS, args->ios);
167 V3D_WRITE(V3D_TFU_COEF0, args->coef[0]);
168 V3D_WRITE(V3D_TFU_COEF1, args->coef[1]);
169 V3D_WRITE(V3D_TFU_COEF2, args->coef[2]);
170 V3D_WRITE(V3D_TFU_COEF3, args->coef[3]);
171
172 V3D_WRITE(V3D_TFU_ICFG, args->icfg);
173
174 while ((V3D_READ(V3D_TFU_CS) & V3D_TFU_CS_CVTCT_SET) == last_vtct) {
175 v3d_hw_tick(v3d);
176 }
177
178 return 0;
179 }
180
181 #if V3D_VERSION >= 41
182 int
v3dX(simulator_submit_csd_ioctl)183 v3dX(simulator_submit_csd_ioctl)(struct v3d_hw *v3d,
184 struct drm_v3d_submit_csd *args,
185 uint32_t gmp_ofs)
186 {
187 g_gmp_ofs = gmp_ofs;
188 v3d_reload_gmp(v3d);
189
190 v3d_invalidate_caches(v3d);
191
192 V3D_WRITE(V3D_CSD_0_QUEUED_CFG1, args->cfg[1]);
193 V3D_WRITE(V3D_CSD_0_QUEUED_CFG2, args->cfg[2]);
194 V3D_WRITE(V3D_CSD_0_QUEUED_CFG3, args->cfg[3]);
195 V3D_WRITE(V3D_CSD_0_QUEUED_CFG4, args->cfg[4]);
196 V3D_WRITE(V3D_CSD_0_QUEUED_CFG5, args->cfg[5]);
197 V3D_WRITE(V3D_CSD_0_QUEUED_CFG6, args->cfg[6]);
198 /* CFG0 kicks off the job */
199 V3D_WRITE(V3D_CSD_0_QUEUED_CFG0, args->cfg[0]);
200
201 while (V3D_READ(V3D_CSD_0_STATUS) &
202 (V3D_CSD_0_STATUS_HAVE_CURRENT_DISPATCH_SET |
203 V3D_CSD_0_STATUS_HAVE_QUEUED_DISPATCH_SET)) {
204 v3d_hw_tick(v3d);
205 }
206
207 v3d_flush_caches(v3d);
208
209 return 0;
210 }
211 #endif
212
213 int
v3dX(simulator_get_param_ioctl)214 v3dX(simulator_get_param_ioctl)(struct v3d_hw *v3d,
215 struct drm_v3d_get_param *args)
216 {
217 static const uint32_t reg_map[] = {
218 [DRM_V3D_PARAM_V3D_UIFCFG] = V3D_HUB_CTL_UIFCFG,
219 [DRM_V3D_PARAM_V3D_HUB_IDENT1] = V3D_HUB_CTL_IDENT1,
220 [DRM_V3D_PARAM_V3D_HUB_IDENT2] = V3D_HUB_CTL_IDENT2,
221 [DRM_V3D_PARAM_V3D_HUB_IDENT3] = V3D_HUB_CTL_IDENT3,
222 [DRM_V3D_PARAM_V3D_CORE0_IDENT0] = V3D_CTL_0_IDENT0,
223 [DRM_V3D_PARAM_V3D_CORE0_IDENT1] = V3D_CTL_0_IDENT1,
224 [DRM_V3D_PARAM_V3D_CORE0_IDENT2] = V3D_CTL_0_IDENT2,
225 };
226
227 switch (args->param) {
228 case DRM_V3D_PARAM_SUPPORTS_TFU:
229 args->value = 1;
230 return 0;
231 case DRM_V3D_PARAM_SUPPORTS_CSD:
232 args->value = V3D_VERSION >= 41;
233 return 0;
234 case DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH:
235 args->value = 1;
236 return 0;
237 }
238
239 if (args->param < ARRAY_SIZE(reg_map) && reg_map[args->param]) {
240 args->value = V3D_READ(reg_map[args->param]);
241 return 0;
242 }
243
244 fprintf(stderr, "Unknown DRM_IOCTL_VC5_GET_PARAM(%lld)\n",
245 (long long)args->value);
246 abort();
247 }
248
249 static struct v3d_hw *v3d_isr_hw;
250
251 static void
v3d_isr(uint32_t hub_status)252 v3d_isr(uint32_t hub_status)
253 {
254 struct v3d_hw *v3d = v3d_isr_hw;
255
256 /* Check the per-core bits */
257 if (hub_status & (1 << 0)) {
258 uint32_t core_status = V3D_READ(V3D_CTL_0_INT_STS);
259 V3D_WRITE(V3D_CTL_0_INT_CLR, core_status);
260
261 if (core_status & V3D_CTL_0_INT_STS_INT_OUTOMEM_SET) {
262 uint32_t size = 256 * 1024;
263 uint32_t offset = v3d_simulator_get_spill(size);
264
265 v3d_reload_gmp(v3d);
266
267 V3D_WRITE(V3D_PTB_0_BPOA, offset);
268 V3D_WRITE(V3D_PTB_0_BPOS, size);
269 return;
270 }
271
272 if (core_status & V3D_CTL_0_INT_STS_INT_GMPV_SET) {
273 fprintf(stderr, "GMP violation at 0x%08x\n",
274 V3D_READ(V3D_GMP_VIO_ADDR));
275 abort();
276 } else {
277 fprintf(stderr,
278 "Unexpected ISR with core status 0x%08x\n",
279 core_status);
280 }
281 abort();
282 }
283
284 return;
285 }
286
287 void
v3dX(simulator_init_regs)288 v3dX(simulator_init_regs)(struct v3d_hw *v3d)
289 {
290 #if V3D_VERSION == 33
291 /* Set OVRTMUOUT to match kernel behavior.
292 *
293 * This means that the texture sampler uniform configuration's tmu
294 * output type field is used, instead of using the hardware default
295 * behavior based on the texture type. If you want the default
296 * behavior, you can still put "2" in the indirect texture state's
297 * output_type field.
298 */
299 V3D_WRITE(V3D_CTL_0_MISCCFG, V3D_CTL_1_MISCCFG_OVRTMUOUT_SET);
300 #endif
301
302 uint32_t core_interrupts = (V3D_CTL_0_INT_STS_INT_GMPV_SET |
303 V3D_CTL_0_INT_STS_INT_OUTOMEM_SET);
304 V3D_WRITE(V3D_CTL_0_INT_MSK_SET, ~core_interrupts);
305 V3D_WRITE(V3D_CTL_0_INT_MSK_CLR, core_interrupts);
306
307 v3d_isr_hw = v3d;
308 v3d_hw_set_isr(v3d, v3d_isr);
309 }
310
311 void
v3dX(simulator_submit_cl_ioctl)312 v3dX(simulator_submit_cl_ioctl)(struct v3d_hw *v3d,
313 struct drm_v3d_submit_cl *submit,
314 uint32_t gmp_ofs)
315 {
316 g_gmp_ofs = gmp_ofs;
317 v3d_reload_gmp(v3d);
318
319 v3d_invalidate_caches(v3d);
320
321 if (submit->qma) {
322 V3D_WRITE(V3D_CLE_0_CT0QMA, submit->qma);
323 V3D_WRITE(V3D_CLE_0_CT0QMS, submit->qms);
324 }
325 #if V3D_VERSION >= 41
326 if (submit->qts) {
327 V3D_WRITE(V3D_CLE_0_CT0QTS,
328 V3D_CLE_0_CT0QTS_CTQTSEN_SET |
329 submit->qts);
330 }
331 #endif
332 V3D_WRITE(V3D_CLE_0_CT0QBA, submit->bcl_start);
333 V3D_WRITE(V3D_CLE_0_CT0QEA, submit->bcl_end);
334
335 /* Wait for bin to complete before firing render. The kernel's
336 * scheduler implements this using the GPU scheduler blocking on the
337 * bin fence completing. (We don't use HW semaphores).
338 */
339 while (V3D_READ(V3D_CLE_0_CT0CA) !=
340 V3D_READ(V3D_CLE_0_CT0EA)) {
341 v3d_hw_tick(v3d);
342 }
343
344 v3d_invalidate_caches(v3d);
345
346 V3D_WRITE(V3D_CLE_0_CT1QBA, submit->rcl_start);
347 V3D_WRITE(V3D_CLE_0_CT1QEA, submit->rcl_end);
348
349 while (V3D_READ(V3D_CLE_0_CT1CA) !=
350 V3D_READ(V3D_CLE_0_CT1EA) ||
351 V3D_READ(V3D_CLE_1_CT1CA) !=
352 V3D_READ(V3D_CLE_1_CT1EA)) {
353 v3d_hw_tick(v3d);
354 }
355 }
356
357 #endif /* USE_V3D_SIMULATOR */
358