diff options
Diffstat (limited to 'drivers/gpu/nvgpu/gp10b/gr_gp10b.c')
-rw-r--r-- | drivers/gpu/nvgpu/gp10b/gr_gp10b.c | 2357 |
1 files changed, 2357 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/gp10b/gr_gp10b.c b/drivers/gpu/nvgpu/gp10b/gr_gp10b.c new file mode 100644 index 00000000..08988ac8 --- /dev/null +++ b/drivers/gpu/nvgpu/gp10b/gr_gp10b.c | |||
@@ -0,0 +1,2357 @@ | |||
1 | /* | ||
2 | * GP10B GPU GR | ||
3 | * | ||
4 | * Copyright (c) 2015-2017, NVIDIA CORPORATION. All rights reserved. | ||
5 | * | ||
6 | * Permission is hereby granted, free of charge, to any person obtaining a | ||
7 | * copy of this software and associated documentation files (the "Software"), | ||
8 | * to deal in the Software without restriction, including without limitation | ||
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | ||
10 | * and/or sell copies of the Software, and to permit persons to whom the | ||
11 | * Software is furnished to do so, subject to the following conditions: | ||
12 | * | ||
13 | * The above copyright notice and this permission notice shall be included in | ||
14 | * all copies or substantial portions of the Software. | ||
15 | * | ||
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | ||
21 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER | ||
22 | * DEALINGS IN THE SOFTWARE. | ||
23 | */ | ||
24 | |||
25 | #include <dt-bindings/soc/gm20b-fuse.h> | ||
26 | #include <dt-bindings/soc/gp10b-fuse.h> | ||
27 | #include <uapi/linux/nvgpu.h> | ||
28 | |||
29 | #include <nvgpu/timers.h> | ||
30 | #include <nvgpu/kmem.h> | ||
31 | #include <nvgpu/gmmu.h> | ||
32 | #include <nvgpu/dma.h> | ||
33 | #include <nvgpu/bug.h> | ||
34 | #include <nvgpu/debug.h> | ||
35 | #include <nvgpu/fuse.h> | ||
36 | #include <nvgpu/enabled.h> | ||
37 | |||
38 | #include "gk20a/gk20a.h" | ||
39 | #include "gk20a/gr_gk20a.h" | ||
40 | #include "gk20a/dbg_gpu_gk20a.h" | ||
41 | #include "gk20a/regops_gk20a.h" | ||
42 | #include "common/linux/os_linux.h" | ||
43 | |||
44 | #include "gm20b/gr_gm20b.h" | ||
45 | #include "gp10b/gr_gp10b.h" | ||
46 | |||
47 | #include <nvgpu/hw/gp10b/hw_gr_gp10b.h> | ||
48 | #include <nvgpu/hw/gp10b/hw_fifo_gp10b.h> | ||
49 | #include <nvgpu/hw/gp10b/hw_ctxsw_prog_gp10b.h> | ||
50 | #include <nvgpu/hw/gp10b/hw_mc_gp10b.h> | ||
51 | #include <nvgpu/hw/gp10b/hw_fuse_gp10b.h> | ||
52 | |||
53 | bool gr_gp10b_is_valid_class(struct gk20a *g, u32 class_num) | ||
54 | { | ||
55 | bool valid = false; | ||
56 | |||
57 | switch (class_num) { | ||
58 | case PASCAL_COMPUTE_A: | ||
59 | case PASCAL_A: | ||
60 | case PASCAL_DMA_COPY_A: | ||
61 | valid = true; | ||
62 | break; | ||
63 | |||
64 | case MAXWELL_COMPUTE_B: | ||
65 | case MAXWELL_B: | ||
66 | case FERMI_TWOD_A: | ||
67 | case KEPLER_DMA_COPY_A: | ||
68 | case MAXWELL_DMA_COPY_A: | ||
69 | valid = true; | ||
70 | break; | ||
71 | |||
72 | default: | ||
73 | break; | ||
74 | } | ||
75 | gk20a_dbg_info("class=0x%x valid=%d", class_num, valid); | ||
76 | return valid; | ||
77 | } | ||
78 | |||
79 | bool gr_gp10b_is_valid_gfx_class(struct gk20a *g, u32 class_num) | ||
80 | { | ||
81 | if (class_num == PASCAL_A || class_num == MAXWELL_B) | ||
82 | return true; | ||
83 | else | ||
84 | return false; | ||
85 | } | ||
86 | |||
87 | bool gr_gp10b_is_valid_compute_class(struct gk20a *g, u32 class_num) | ||
88 | { | ||
89 | if (class_num == PASCAL_COMPUTE_A || class_num == MAXWELL_COMPUTE_B) | ||
90 | return true; | ||
91 | else | ||
92 | return false; | ||
93 | } | ||
94 | |||
95 | |||
96 | static void gr_gp10b_sm_lrf_ecc_overcount_war(int single_err, | ||
97 | u32 sed_status, | ||
98 | u32 ded_status, | ||
99 | u32 *count_to_adjust, | ||
100 | u32 opposite_count) | ||
101 | { | ||
102 | u32 over_count = 0; | ||
103 | |||
104 | sed_status >>= gr_pri_gpc0_tpc0_sm_lrf_ecc_status_single_err_detected_qrfdp0_b(); | ||
105 | ded_status >>= gr_pri_gpc0_tpc0_sm_lrf_ecc_status_double_err_detected_qrfdp0_b(); | ||
106 | |||
107 | /* One overcount for each partition on which a SBE occurred but not a | ||
108 | DBE (or vice-versa) */ | ||
109 | if (single_err) { | ||
110 | over_count = | ||
111 | hweight32(sed_status & ~ded_status); | ||
112 | } else { | ||
113 | over_count = | ||
114 | hweight32(ded_status & ~sed_status); | ||
115 | } | ||
116 | |||
117 | /* If both a SBE and a DBE occur on the same partition, then we have an | ||
118 | overcount for the subpartition if the opposite error counts are | ||
119 | zero. */ | ||
120 | if ((sed_status & ded_status) && (opposite_count == 0)) { | ||
121 | over_count += | ||
122 | hweight32(sed_status & ded_status); | ||
123 | } | ||
124 | |||
125 | if (*count_to_adjust > over_count) | ||
126 | *count_to_adjust -= over_count; | ||
127 | else | ||
128 | *count_to_adjust = 0; | ||
129 | } | ||
130 | |||
131 | int gr_gp10b_handle_sm_exception(struct gk20a *g, | ||
132 | u32 gpc, u32 tpc, u32 sm, | ||
133 | bool *post_event, struct channel_gk20a *fault_ch, | ||
134 | u32 *hww_global_esr) | ||
135 | { | ||
136 | int ret = 0; | ||
137 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
138 | u32 tpc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_TPC_IN_GPC_STRIDE); | ||
139 | u32 offset = gpc_stride * gpc + tpc_in_gpc_stride * tpc; | ||
140 | u32 lrf_ecc_status, lrf_ecc_sed_status, lrf_ecc_ded_status; | ||
141 | u32 lrf_single_count_delta, lrf_double_count_delta; | ||
142 | u32 shm_ecc_status; | ||
143 | |||
144 | ret = gr_gk20a_handle_sm_exception(g, | ||
145 | gpc, tpc, sm, post_event, fault_ch, hww_global_esr); | ||
146 | |||
147 | /* Check for LRF ECC errors. */ | ||
148 | lrf_ecc_status = gk20a_readl(g, | ||
149 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_r() + offset); | ||
150 | lrf_ecc_sed_status = lrf_ecc_status & | ||
151 | (gr_pri_gpc0_tpc0_sm_lrf_ecc_status_single_err_detected_qrfdp0_pending_f() | | ||
152 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_single_err_detected_qrfdp1_pending_f() | | ||
153 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_single_err_detected_qrfdp2_pending_f() | | ||
154 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_single_err_detected_qrfdp3_pending_f()); | ||
155 | lrf_ecc_ded_status = lrf_ecc_status & | ||
156 | (gr_pri_gpc0_tpc0_sm_lrf_ecc_status_double_err_detected_qrfdp0_pending_f() | | ||
157 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_double_err_detected_qrfdp1_pending_f() | | ||
158 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_double_err_detected_qrfdp2_pending_f() | | ||
159 | gr_pri_gpc0_tpc0_sm_lrf_ecc_status_double_err_detected_qrfdp3_pending_f()); | ||
160 | lrf_single_count_delta = | ||
161 | gk20a_readl(g, | ||
162 | gr_pri_gpc0_tpc0_sm_lrf_ecc_single_err_count_r() + | ||
163 | offset); | ||
164 | lrf_double_count_delta = | ||
165 | gk20a_readl(g, | ||
166 | gr_pri_gpc0_tpc0_sm_lrf_ecc_double_err_count_r() + | ||
167 | offset); | ||
168 | gk20a_writel(g, | ||
169 | gr_pri_gpc0_tpc0_sm_lrf_ecc_single_err_count_r() + offset, | ||
170 | 0); | ||
171 | gk20a_writel(g, | ||
172 | gr_pri_gpc0_tpc0_sm_lrf_ecc_double_err_count_r() + offset, | ||
173 | 0); | ||
174 | if (lrf_ecc_sed_status) { | ||
175 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
176 | "Single bit error detected in SM LRF!"); | ||
177 | |||
178 | gr_gp10b_sm_lrf_ecc_overcount_war(1, | ||
179 | lrf_ecc_sed_status, | ||
180 | lrf_ecc_ded_status, | ||
181 | &lrf_single_count_delta, | ||
182 | lrf_double_count_delta); | ||
183 | g->ecc.gr.t18x.sm_lrf_single_err_count.counters[tpc] += | ||
184 | lrf_single_count_delta; | ||
185 | } | ||
186 | if (lrf_ecc_ded_status) { | ||
187 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
188 | "Double bit error detected in SM LRF!"); | ||
189 | |||
190 | gr_gp10b_sm_lrf_ecc_overcount_war(0, | ||
191 | lrf_ecc_sed_status, | ||
192 | lrf_ecc_ded_status, | ||
193 | &lrf_double_count_delta, | ||
194 | lrf_single_count_delta); | ||
195 | g->ecc.gr.t18x.sm_lrf_double_err_count.counters[tpc] += | ||
196 | lrf_double_count_delta; | ||
197 | } | ||
198 | gk20a_writel(g, gr_pri_gpc0_tpc0_sm_lrf_ecc_status_r() + offset, | ||
199 | lrf_ecc_status); | ||
200 | |||
201 | /* Check for SHM ECC errors. */ | ||
202 | shm_ecc_status = gk20a_readl(g, | ||
203 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_r() + offset); | ||
204 | if ((shm_ecc_status & | ||
205 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_single_err_corrected_shm0_pending_f()) || | ||
206 | (shm_ecc_status & | ||
207 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_single_err_corrected_shm1_pending_f()) || | ||
208 | (shm_ecc_status & | ||
209 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_single_err_detected_shm0_pending_f()) || | ||
210 | (shm_ecc_status & | ||
211 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_single_err_detected_shm1_pending_f()) ) { | ||
212 | u32 ecc_stats_reg_val; | ||
213 | |||
214 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
215 | "Single bit error detected in SM SHM!"); | ||
216 | |||
217 | ecc_stats_reg_val = | ||
218 | gk20a_readl(g, | ||
219 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_r() + offset); | ||
220 | g->ecc.gr.t18x.sm_shm_sec_count.counters[tpc] += | ||
221 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_single_corrected_v(ecc_stats_reg_val); | ||
222 | g->ecc.gr.t18x.sm_shm_sed_count.counters[tpc] += | ||
223 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_single_detected_v(ecc_stats_reg_val); | ||
224 | ecc_stats_reg_val &= ~(gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_single_corrected_m() | | ||
225 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_single_detected_m()); | ||
226 | gk20a_writel(g, | ||
227 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_r() + offset, | ||
228 | ecc_stats_reg_val); | ||
229 | } | ||
230 | if ( (shm_ecc_status & | ||
231 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_double_err_detected_shm0_pending_f()) || | ||
232 | (shm_ecc_status & | ||
233 | gr_pri_gpc0_tpc0_sm_shm_ecc_status_double_err_detected_shm1_pending_f()) ) { | ||
234 | u32 ecc_stats_reg_val; | ||
235 | |||
236 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
237 | "Double bit error detected in SM SHM!"); | ||
238 | |||
239 | ecc_stats_reg_val = | ||
240 | gk20a_readl(g, | ||
241 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_r() + offset); | ||
242 | g->ecc.gr.t18x.sm_shm_ded_count.counters[tpc] += | ||
243 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_double_detected_v(ecc_stats_reg_val); | ||
244 | ecc_stats_reg_val &= ~(gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_double_detected_m()); | ||
245 | gk20a_writel(g, | ||
246 | gr_pri_gpc0_tpc0_sm_shm_ecc_err_count_r() + offset, | ||
247 | ecc_stats_reg_val); | ||
248 | } | ||
249 | gk20a_writel(g, gr_pri_gpc0_tpc0_sm_shm_ecc_status_r() + offset, | ||
250 | shm_ecc_status); | ||
251 | |||
252 | |||
253 | return ret; | ||
254 | } | ||
255 | |||
256 | int gr_gp10b_handle_tex_exception(struct gk20a *g, u32 gpc, u32 tpc, | ||
257 | bool *post_event) | ||
258 | { | ||
259 | int ret = 0; | ||
260 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
261 | u32 tpc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_TPC_IN_GPC_STRIDE); | ||
262 | u32 offset = gpc_stride * gpc + tpc_in_gpc_stride * tpc; | ||
263 | u32 esr; | ||
264 | u32 ecc_stats_reg_val; | ||
265 | |||
266 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, ""); | ||
267 | |||
268 | esr = gk20a_readl(g, | ||
269 | gr_gpc0_tpc0_tex_m_hww_esr_r() + offset); | ||
270 | gk20a_dbg(gpu_dbg_intr | gpu_dbg_gpu_dbg, "0x%08x", esr); | ||
271 | |||
272 | if (esr & gr_gpc0_tpc0_tex_m_hww_esr_ecc_sec_pending_f()) { | ||
273 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
274 | "Single bit error detected in TEX!"); | ||
275 | |||
276 | /* Pipe 0 counters */ | ||
277 | gk20a_writel(g, | ||
278 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
279 | gr_pri_gpc0_tpc0_tex_m_routing_sel_pipe0_f()); | ||
280 | |||
281 | ecc_stats_reg_val = gk20a_readl(g, | ||
282 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset); | ||
283 | g->ecc.gr.t18x.tex_total_sec_pipe0_count.counters[tpc] += | ||
284 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_sec_v(ecc_stats_reg_val); | ||
285 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_sec_m(); | ||
286 | gk20a_writel(g, | ||
287 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset, | ||
288 | ecc_stats_reg_val); | ||
289 | |||
290 | ecc_stats_reg_val = gk20a_readl(g, | ||
291 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset); | ||
292 | g->ecc.gr.t18x.tex_unique_sec_pipe0_count.counters[tpc] += | ||
293 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_sec_v(ecc_stats_reg_val); | ||
294 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_sec_m(); | ||
295 | gk20a_writel(g, | ||
296 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset, | ||
297 | ecc_stats_reg_val); | ||
298 | |||
299 | |||
300 | /* Pipe 1 counters */ | ||
301 | gk20a_writel(g, | ||
302 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
303 | gr_pri_gpc0_tpc0_tex_m_routing_sel_pipe1_f()); | ||
304 | |||
305 | ecc_stats_reg_val = gk20a_readl(g, | ||
306 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset); | ||
307 | g->ecc.gr.t18x.tex_total_sec_pipe1_count.counters[tpc] += | ||
308 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_sec_v(ecc_stats_reg_val); | ||
309 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_sec_m(); | ||
310 | gk20a_writel(g, | ||
311 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset, | ||
312 | ecc_stats_reg_val); | ||
313 | |||
314 | ecc_stats_reg_val = gk20a_readl(g, | ||
315 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset); | ||
316 | g->ecc.gr.t18x.tex_unique_sec_pipe1_count.counters[tpc] += | ||
317 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_sec_v(ecc_stats_reg_val); | ||
318 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_sec_m(); | ||
319 | gk20a_writel(g, | ||
320 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset, | ||
321 | ecc_stats_reg_val); | ||
322 | |||
323 | |||
324 | gk20a_writel(g, | ||
325 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
326 | gr_pri_gpc0_tpc0_tex_m_routing_sel_default_f()); | ||
327 | } | ||
328 | if (esr & gr_gpc0_tpc0_tex_m_hww_esr_ecc_ded_pending_f()) { | ||
329 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_intr, | ||
330 | "Double bit error detected in TEX!"); | ||
331 | |||
332 | /* Pipe 0 counters */ | ||
333 | gk20a_writel(g, | ||
334 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
335 | gr_pri_gpc0_tpc0_tex_m_routing_sel_pipe0_f()); | ||
336 | |||
337 | ecc_stats_reg_val = gk20a_readl(g, | ||
338 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset); | ||
339 | g->ecc.gr.t18x.tex_total_ded_pipe0_count.counters[tpc] += | ||
340 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_ded_v(ecc_stats_reg_val); | ||
341 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_ded_m(); | ||
342 | gk20a_writel(g, | ||
343 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset, | ||
344 | ecc_stats_reg_val); | ||
345 | |||
346 | ecc_stats_reg_val = gk20a_readl(g, | ||
347 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset); | ||
348 | g->ecc.gr.t18x.tex_unique_ded_pipe0_count.counters[tpc] += | ||
349 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_ded_v(ecc_stats_reg_val); | ||
350 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_ded_m(); | ||
351 | gk20a_writel(g, | ||
352 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset, | ||
353 | ecc_stats_reg_val); | ||
354 | |||
355 | |||
356 | /* Pipe 1 counters */ | ||
357 | gk20a_writel(g, | ||
358 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
359 | gr_pri_gpc0_tpc0_tex_m_routing_sel_pipe1_f()); | ||
360 | |||
361 | ecc_stats_reg_val = gk20a_readl(g, | ||
362 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset); | ||
363 | g->ecc.gr.t18x.tex_total_ded_pipe1_count.counters[tpc] += | ||
364 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_ded_v(ecc_stats_reg_val); | ||
365 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_ded_m(); | ||
366 | gk20a_writel(g, | ||
367 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_total_r() + offset, | ||
368 | ecc_stats_reg_val); | ||
369 | |||
370 | ecc_stats_reg_val = gk20a_readl(g, | ||
371 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset); | ||
372 | g->ecc.gr.t18x.tex_unique_ded_pipe1_count.counters[tpc] += | ||
373 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_ded_v(ecc_stats_reg_val); | ||
374 | ecc_stats_reg_val &= ~gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_ded_m(); | ||
375 | gk20a_writel(g, | ||
376 | gr_pri_gpc0_tpc0_tex_m_ecc_cnt_unique_r() + offset, | ||
377 | ecc_stats_reg_val); | ||
378 | |||
379 | |||
380 | gk20a_writel(g, | ||
381 | gr_pri_gpc0_tpc0_tex_m_routing_r() + offset, | ||
382 | gr_pri_gpc0_tpc0_tex_m_routing_sel_default_f()); | ||
383 | } | ||
384 | |||
385 | gk20a_writel(g, | ||
386 | gr_gpc0_tpc0_tex_m_hww_esr_r() + offset, | ||
387 | esr | gr_gpc0_tpc0_tex_m_hww_esr_reset_active_f()); | ||
388 | |||
389 | return ret; | ||
390 | } | ||
391 | |||
392 | int gr_gp10b_commit_global_cb_manager(struct gk20a *g, | ||
393 | struct channel_gk20a *c, bool patch) | ||
394 | { | ||
395 | struct gr_gk20a *gr = &g->gr; | ||
396 | struct channel_ctx_gk20a *ch_ctx = &c->ch_ctx; | ||
397 | struct gr_ctx_desc *gr_ctx = ch_ctx->gr_ctx; | ||
398 | u32 attrib_offset_in_chunk = 0; | ||
399 | u32 alpha_offset_in_chunk = 0; | ||
400 | u32 pd_ab_max_output; | ||
401 | u32 gpc_index, ppc_index; | ||
402 | u32 temp, temp2; | ||
403 | u32 cbm_cfg_size_beta, cbm_cfg_size_alpha, cbm_cfg_size_steadystate; | ||
404 | u32 attrib_size_in_chunk, cb_attrib_cache_size_init; | ||
405 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
406 | u32 ppc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_PPC_IN_GPC_STRIDE); | ||
407 | u32 num_pes_per_gpc = nvgpu_get_litter_value(g, GPU_LIT_NUM_PES_PER_GPC); | ||
408 | |||
409 | gk20a_dbg_fn(""); | ||
410 | |||
411 | if (gr_ctx->graphics_preempt_mode == NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) { | ||
412 | attrib_size_in_chunk = gr->attrib_cb_default_size + | ||
413 | (gr_gpc0_ppc0_cbm_beta_cb_size_v_gfxp_v() - | ||
414 | gr_gpc0_ppc0_cbm_beta_cb_size_v_default_v()); | ||
415 | cb_attrib_cache_size_init = gr->attrib_cb_default_size + | ||
416 | (gr_gpc0_ppc0_cbm_beta_cb_size_v_gfxp_v() - | ||
417 | gr_gpc0_ppc0_cbm_beta_cb_size_v_default_v()); | ||
418 | } else { | ||
419 | attrib_size_in_chunk = gr->attrib_cb_size; | ||
420 | cb_attrib_cache_size_init = gr->attrib_cb_default_size; | ||
421 | } | ||
422 | |||
423 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_ds_tga_constraintlogic_beta_r(), | ||
424 | gr->attrib_cb_default_size, patch); | ||
425 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_ds_tga_constraintlogic_alpha_r(), | ||
426 | gr->alpha_cb_default_size, patch); | ||
427 | |||
428 | pd_ab_max_output = (gr->alpha_cb_default_size * | ||
429 | gr_gpc0_ppc0_cbm_beta_cb_size_v_granularity_v()) / | ||
430 | gr_pd_ab_dist_cfg1_max_output_granularity_v(); | ||
431 | |||
432 | if (g->gr.pd_max_batches) { | ||
433 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_pd_ab_dist_cfg1_r(), | ||
434 | gr_pd_ab_dist_cfg1_max_output_f(pd_ab_max_output) | | ||
435 | gr_pd_ab_dist_cfg1_max_batches_f(g->gr.pd_max_batches), patch); | ||
436 | } else { | ||
437 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_pd_ab_dist_cfg1_r(), | ||
438 | gr_pd_ab_dist_cfg1_max_output_f(pd_ab_max_output) | | ||
439 | gr_pd_ab_dist_cfg1_max_batches_init_f(), patch); | ||
440 | } | ||
441 | |||
442 | attrib_offset_in_chunk = alpha_offset_in_chunk + | ||
443 | gr->tpc_count * gr->alpha_cb_size; | ||
444 | |||
445 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | ||
446 | temp = gpc_stride * gpc_index; | ||
447 | temp2 = num_pes_per_gpc * gpc_index; | ||
448 | for (ppc_index = 0; ppc_index < gr->gpc_ppc_count[gpc_index]; | ||
449 | ppc_index++) { | ||
450 | cbm_cfg_size_beta = cb_attrib_cache_size_init * | ||
451 | gr->pes_tpc_count[ppc_index][gpc_index]; | ||
452 | cbm_cfg_size_alpha = gr->alpha_cb_default_size * | ||
453 | gr->pes_tpc_count[ppc_index][gpc_index]; | ||
454 | cbm_cfg_size_steadystate = gr->attrib_cb_default_size * | ||
455 | gr->pes_tpc_count[ppc_index][gpc_index]; | ||
456 | |||
457 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
458 | gr_gpc0_ppc0_cbm_beta_cb_size_r() + temp + | ||
459 | ppc_in_gpc_stride * ppc_index, | ||
460 | cbm_cfg_size_beta, patch); | ||
461 | |||
462 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
463 | gr_gpc0_ppc0_cbm_beta_cb_offset_r() + temp + | ||
464 | ppc_in_gpc_stride * ppc_index, | ||
465 | attrib_offset_in_chunk, patch); | ||
466 | |||
467 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
468 | gr_gpc0_ppc0_cbm_beta_steady_state_cb_size_r() + temp + | ||
469 | ppc_in_gpc_stride * ppc_index, | ||
470 | cbm_cfg_size_steadystate, | ||
471 | patch); | ||
472 | |||
473 | attrib_offset_in_chunk += attrib_size_in_chunk * | ||
474 | gr->pes_tpc_count[ppc_index][gpc_index]; | ||
475 | |||
476 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
477 | gr_gpc0_ppc0_cbm_alpha_cb_size_r() + temp + | ||
478 | ppc_in_gpc_stride * ppc_index, | ||
479 | cbm_cfg_size_alpha, patch); | ||
480 | |||
481 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
482 | gr_gpc0_ppc0_cbm_alpha_cb_offset_r() + temp + | ||
483 | ppc_in_gpc_stride * ppc_index, | ||
484 | alpha_offset_in_chunk, patch); | ||
485 | |||
486 | alpha_offset_in_chunk += gr->alpha_cb_size * | ||
487 | gr->pes_tpc_count[ppc_index][gpc_index]; | ||
488 | |||
489 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
490 | gr_gpcs_swdx_tc_beta_cb_size_r(ppc_index + temp2), | ||
491 | gr_gpcs_swdx_tc_beta_cb_size_v_f(cbm_cfg_size_steadystate), | ||
492 | patch); | ||
493 | } | ||
494 | } | ||
495 | |||
496 | return 0; | ||
497 | } | ||
498 | |||
499 | void gr_gp10b_commit_global_pagepool(struct gk20a *g, | ||
500 | struct channel_ctx_gk20a *ch_ctx, | ||
501 | u64 addr, u32 size, bool patch) | ||
502 | { | ||
503 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_scc_pagepool_base_r(), | ||
504 | gr_scc_pagepool_base_addr_39_8_f(addr), patch); | ||
505 | |||
506 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_scc_pagepool_r(), | ||
507 | gr_scc_pagepool_total_pages_f(size) | | ||
508 | gr_scc_pagepool_valid_true_f(), patch); | ||
509 | |||
510 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_gcc_pagepool_base_r(), | ||
511 | gr_gpcs_gcc_pagepool_base_addr_39_8_f(addr), patch); | ||
512 | |||
513 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_gcc_pagepool_r(), | ||
514 | gr_gpcs_gcc_pagepool_total_pages_f(size), patch); | ||
515 | } | ||
516 | |||
517 | int gr_gp10b_add_zbc_color(struct gk20a *g, struct gr_gk20a *gr, | ||
518 | struct zbc_entry *color_val, u32 index) | ||
519 | { | ||
520 | u32 i; | ||
521 | u32 zbc_c; | ||
522 | |||
523 | /* update l2 table */ | ||
524 | g->ops.ltc.set_zbc_color_entry(g, color_val, index); | ||
525 | |||
526 | /* update ds table */ | ||
527 | gk20a_writel(g, gr_ds_zbc_color_r_r(), | ||
528 | gr_ds_zbc_color_r_val_f(color_val->color_ds[0])); | ||
529 | gk20a_writel(g, gr_ds_zbc_color_g_r(), | ||
530 | gr_ds_zbc_color_g_val_f(color_val->color_ds[1])); | ||
531 | gk20a_writel(g, gr_ds_zbc_color_b_r(), | ||
532 | gr_ds_zbc_color_b_val_f(color_val->color_ds[2])); | ||
533 | gk20a_writel(g, gr_ds_zbc_color_a_r(), | ||
534 | gr_ds_zbc_color_a_val_f(color_val->color_ds[3])); | ||
535 | |||
536 | gk20a_writel(g, gr_ds_zbc_color_fmt_r(), | ||
537 | gr_ds_zbc_color_fmt_val_f(color_val->format)); | ||
538 | |||
539 | gk20a_writel(g, gr_ds_zbc_tbl_index_r(), | ||
540 | gr_ds_zbc_tbl_index_val_f(index + GK20A_STARTOF_ZBC_TABLE)); | ||
541 | |||
542 | /* trigger the write */ | ||
543 | gk20a_writel(g, gr_ds_zbc_tbl_ld_r(), | ||
544 | gr_ds_zbc_tbl_ld_select_c_f() | | ||
545 | gr_ds_zbc_tbl_ld_action_write_f() | | ||
546 | gr_ds_zbc_tbl_ld_trigger_active_f()); | ||
547 | |||
548 | /* update local copy */ | ||
549 | for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) { | ||
550 | gr->zbc_col_tbl[index].color_l2[i] = color_val->color_l2[i]; | ||
551 | gr->zbc_col_tbl[index].color_ds[i] = color_val->color_ds[i]; | ||
552 | } | ||
553 | gr->zbc_col_tbl[index].format = color_val->format; | ||
554 | gr->zbc_col_tbl[index].ref_cnt++; | ||
555 | |||
556 | gk20a_writel_check(g, gr_gpcs_swdx_dss_zbc_color_r_r(index), | ||
557 | color_val->color_ds[0]); | ||
558 | gk20a_writel_check(g, gr_gpcs_swdx_dss_zbc_color_g_r(index), | ||
559 | color_val->color_ds[1]); | ||
560 | gk20a_writel_check(g, gr_gpcs_swdx_dss_zbc_color_b_r(index), | ||
561 | color_val->color_ds[2]); | ||
562 | gk20a_writel_check(g, gr_gpcs_swdx_dss_zbc_color_a_r(index), | ||
563 | color_val->color_ds[3]); | ||
564 | zbc_c = gk20a_readl(g, gr_gpcs_swdx_dss_zbc_c_01_to_04_format_r() + (index & ~3)); | ||
565 | zbc_c &= ~(0x7f << ((index % 4) * 7)); | ||
566 | zbc_c |= color_val->format << ((index % 4) * 7); | ||
567 | gk20a_writel_check(g, gr_gpcs_swdx_dss_zbc_c_01_to_04_format_r() + (index & ~3), zbc_c); | ||
568 | |||
569 | return 0; | ||
570 | } | ||
571 | |||
572 | int gr_gp10b_add_zbc_depth(struct gk20a *g, struct gr_gk20a *gr, | ||
573 | struct zbc_entry *depth_val, u32 index) | ||
574 | { | ||
575 | u32 zbc_z; | ||
576 | |||
577 | /* update l2 table */ | ||
578 | g->ops.ltc.set_zbc_depth_entry(g, depth_val, index); | ||
579 | |||
580 | /* update ds table */ | ||
581 | gk20a_writel(g, gr_ds_zbc_z_r(), | ||
582 | gr_ds_zbc_z_val_f(depth_val->depth)); | ||
583 | |||
584 | gk20a_writel(g, gr_ds_zbc_z_fmt_r(), | ||
585 | gr_ds_zbc_z_fmt_val_f(depth_val->format)); | ||
586 | |||
587 | gk20a_writel(g, gr_ds_zbc_tbl_index_r(), | ||
588 | gr_ds_zbc_tbl_index_val_f(index + GK20A_STARTOF_ZBC_TABLE)); | ||
589 | |||
590 | /* trigger the write */ | ||
591 | gk20a_writel(g, gr_ds_zbc_tbl_ld_r(), | ||
592 | gr_ds_zbc_tbl_ld_select_z_f() | | ||
593 | gr_ds_zbc_tbl_ld_action_write_f() | | ||
594 | gr_ds_zbc_tbl_ld_trigger_active_f()); | ||
595 | |||
596 | /* update local copy */ | ||
597 | gr->zbc_dep_tbl[index].depth = depth_val->depth; | ||
598 | gr->zbc_dep_tbl[index].format = depth_val->format; | ||
599 | gr->zbc_dep_tbl[index].ref_cnt++; | ||
600 | |||
601 | gk20a_writel(g, gr_gpcs_swdx_dss_zbc_z_r(index), depth_val->depth); | ||
602 | zbc_z = gk20a_readl(g, gr_gpcs_swdx_dss_zbc_z_01_to_04_format_r() + (index & ~3)); | ||
603 | zbc_z &= ~(0x7f << (index % 4) * 7); | ||
604 | zbc_z |= depth_val->format << (index % 4) * 7; | ||
605 | gk20a_writel(g, gr_gpcs_swdx_dss_zbc_z_01_to_04_format_r() + (index & ~3), zbc_z); | ||
606 | |||
607 | return 0; | ||
608 | } | ||
609 | |||
610 | u32 gr_gp10b_pagepool_default_size(struct gk20a *g) | ||
611 | { | ||
612 | return gr_scc_pagepool_total_pages_hwmax_value_v(); | ||
613 | } | ||
614 | |||
615 | int gr_gp10b_calc_global_ctx_buffer_size(struct gk20a *g) | ||
616 | { | ||
617 | struct gr_gk20a *gr = &g->gr; | ||
618 | int size; | ||
619 | |||
620 | gr->attrib_cb_size = gr->attrib_cb_default_size; | ||
621 | gr->alpha_cb_size = gr->alpha_cb_default_size; | ||
622 | |||
623 | gr->attrib_cb_size = min(gr->attrib_cb_size, | ||
624 | gr_gpc0_ppc0_cbm_beta_cb_size_v_f(~0) / g->gr.tpc_count); | ||
625 | gr->alpha_cb_size = min(gr->alpha_cb_size, | ||
626 | gr_gpc0_ppc0_cbm_alpha_cb_size_v_f(~0) / g->gr.tpc_count); | ||
627 | |||
628 | size = gr->attrib_cb_size * | ||
629 | gr_gpc0_ppc0_cbm_beta_cb_size_v_granularity_v() * | ||
630 | gr->max_tpc_count; | ||
631 | |||
632 | size += gr->alpha_cb_size * | ||
633 | gr_gpc0_ppc0_cbm_alpha_cb_size_v_granularity_v() * | ||
634 | gr->max_tpc_count; | ||
635 | |||
636 | size = ALIGN(size, 128); | ||
637 | |||
638 | return size; | ||
639 | } | ||
640 | |||
641 | static void gr_gp10b_set_go_idle_timeout(struct gk20a *g, u32 data) | ||
642 | { | ||
643 | gk20a_writel(g, gr_fe_go_idle_timeout_r(), data); | ||
644 | } | ||
645 | |||
646 | static void gr_gp10b_set_coalesce_buffer_size(struct gk20a *g, u32 data) | ||
647 | { | ||
648 | u32 val; | ||
649 | |||
650 | gk20a_dbg_fn(""); | ||
651 | |||
652 | val = gk20a_readl(g, gr_gpcs_tc_debug0_r()); | ||
653 | val = set_field(val, gr_gpcs_tc_debug0_limit_coalesce_buffer_size_m(), | ||
654 | gr_gpcs_tc_debug0_limit_coalesce_buffer_size_f(data)); | ||
655 | gk20a_writel(g, gr_gpcs_tc_debug0_r(), val); | ||
656 | |||
657 | gk20a_dbg_fn("done"); | ||
658 | } | ||
659 | |||
660 | void gr_gp10b_set_bes_crop_debug3(struct gk20a *g, u32 data) | ||
661 | { | ||
662 | u32 val; | ||
663 | |||
664 | gk20a_dbg_fn(""); | ||
665 | |||
666 | val = gk20a_readl(g, gr_bes_crop_debug3_r()); | ||
667 | if ((data & 1)) { | ||
668 | val = set_field(val, | ||
669 | gr_bes_crop_debug3_blendopt_read_suppress_m(), | ||
670 | gr_bes_crop_debug3_blendopt_read_suppress_enabled_f()); | ||
671 | val = set_field(val, | ||
672 | gr_bes_crop_debug3_blendopt_fill_override_m(), | ||
673 | gr_bes_crop_debug3_blendopt_fill_override_enabled_f()); | ||
674 | } else { | ||
675 | val = set_field(val, | ||
676 | gr_bes_crop_debug3_blendopt_read_suppress_m(), | ||
677 | gr_bes_crop_debug3_blendopt_read_suppress_disabled_f()); | ||
678 | val = set_field(val, | ||
679 | gr_bes_crop_debug3_blendopt_fill_override_m(), | ||
680 | gr_bes_crop_debug3_blendopt_fill_override_disabled_f()); | ||
681 | } | ||
682 | gk20a_writel(g, gr_bes_crop_debug3_r(), val); | ||
683 | } | ||
684 | |||
685 | int gr_gp10b_handle_sw_method(struct gk20a *g, u32 addr, | ||
686 | u32 class_num, u32 offset, u32 data) | ||
687 | { | ||
688 | gk20a_dbg_fn(""); | ||
689 | |||
690 | if (class_num == PASCAL_COMPUTE_A) { | ||
691 | switch (offset << 2) { | ||
692 | case NVC0C0_SET_SHADER_EXCEPTIONS: | ||
693 | gk20a_gr_set_shader_exceptions(g, data); | ||
694 | break; | ||
695 | case NVC0C0_SET_RD_COALESCE: | ||
696 | gr_gm20b_set_rd_coalesce(g, data); | ||
697 | break; | ||
698 | default: | ||
699 | goto fail; | ||
700 | } | ||
701 | } | ||
702 | |||
703 | if (class_num == PASCAL_A) { | ||
704 | switch (offset << 2) { | ||
705 | case NVC097_SET_SHADER_EXCEPTIONS: | ||
706 | gk20a_gr_set_shader_exceptions(g, data); | ||
707 | break; | ||
708 | case NVC097_SET_CIRCULAR_BUFFER_SIZE: | ||
709 | g->ops.gr.set_circular_buffer_size(g, data); | ||
710 | break; | ||
711 | case NVC097_SET_ALPHA_CIRCULAR_BUFFER_SIZE: | ||
712 | g->ops.gr.set_alpha_circular_buffer_size(g, data); | ||
713 | break; | ||
714 | case NVC097_SET_GO_IDLE_TIMEOUT: | ||
715 | gr_gp10b_set_go_idle_timeout(g, data); | ||
716 | break; | ||
717 | case NVC097_SET_COALESCE_BUFFER_SIZE: | ||
718 | gr_gp10b_set_coalesce_buffer_size(g, data); | ||
719 | break; | ||
720 | case NVC097_SET_RD_COALESCE: | ||
721 | gr_gm20b_set_rd_coalesce(g, data); | ||
722 | break; | ||
723 | case NVC097_SET_BES_CROP_DEBUG3: | ||
724 | g->ops.gr.set_bes_crop_debug3(g, data); | ||
725 | break; | ||
726 | default: | ||
727 | goto fail; | ||
728 | } | ||
729 | } | ||
730 | return 0; | ||
731 | |||
732 | fail: | ||
733 | return -EINVAL; | ||
734 | } | ||
735 | |||
736 | void gr_gp10b_cb_size_default(struct gk20a *g) | ||
737 | { | ||
738 | struct gr_gk20a *gr = &g->gr; | ||
739 | |||
740 | if (!gr->attrib_cb_default_size) | ||
741 | gr->attrib_cb_default_size = 0x800; | ||
742 | gr->alpha_cb_default_size = | ||
743 | gr_gpc0_ppc0_cbm_alpha_cb_size_v_default_v(); | ||
744 | } | ||
745 | |||
746 | void gr_gp10b_set_alpha_circular_buffer_size(struct gk20a *g, u32 data) | ||
747 | { | ||
748 | struct gr_gk20a *gr = &g->gr; | ||
749 | u32 gpc_index, ppc_index, stride, val; | ||
750 | u32 pd_ab_max_output; | ||
751 | u32 alpha_cb_size = data * 4; | ||
752 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
753 | u32 ppc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_PPC_IN_GPC_STRIDE); | ||
754 | |||
755 | gk20a_dbg_fn(""); | ||
756 | |||
757 | if (alpha_cb_size > gr->alpha_cb_size) | ||
758 | alpha_cb_size = gr->alpha_cb_size; | ||
759 | |||
760 | gk20a_writel(g, gr_ds_tga_constraintlogic_alpha_r(), | ||
761 | (gk20a_readl(g, gr_ds_tga_constraintlogic_alpha_r()) & | ||
762 | ~gr_ds_tga_constraintlogic_alpha_cbsize_f(~0)) | | ||
763 | gr_ds_tga_constraintlogic_alpha_cbsize_f(alpha_cb_size)); | ||
764 | |||
765 | pd_ab_max_output = alpha_cb_size * | ||
766 | gr_gpc0_ppc0_cbm_alpha_cb_size_v_granularity_v() / | ||
767 | gr_pd_ab_dist_cfg1_max_output_granularity_v(); | ||
768 | |||
769 | if (g->gr.pd_max_batches) { | ||
770 | gk20a_writel(g, gr_pd_ab_dist_cfg1_r(), | ||
771 | gr_pd_ab_dist_cfg1_max_output_f(pd_ab_max_output) | | ||
772 | gr_pd_ab_dist_cfg1_max_batches_f(g->gr.pd_max_batches)); | ||
773 | } else { | ||
774 | gk20a_writel(g, gr_pd_ab_dist_cfg1_r(), | ||
775 | gr_pd_ab_dist_cfg1_max_output_f(pd_ab_max_output) | | ||
776 | gr_pd_ab_dist_cfg1_max_batches_init_f()); | ||
777 | } | ||
778 | |||
779 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | ||
780 | stride = gpc_stride * gpc_index; | ||
781 | |||
782 | for (ppc_index = 0; ppc_index < gr->gpc_ppc_count[gpc_index]; | ||
783 | ppc_index++) { | ||
784 | |||
785 | val = gk20a_readl(g, gr_gpc0_ppc0_cbm_alpha_cb_size_r() + | ||
786 | stride + | ||
787 | ppc_in_gpc_stride * ppc_index); | ||
788 | |||
789 | val = set_field(val, gr_gpc0_ppc0_cbm_alpha_cb_size_v_m(), | ||
790 | gr_gpc0_ppc0_cbm_alpha_cb_size_v_f(alpha_cb_size * | ||
791 | gr->pes_tpc_count[ppc_index][gpc_index])); | ||
792 | |||
793 | gk20a_writel(g, gr_gpc0_ppc0_cbm_alpha_cb_size_r() + | ||
794 | stride + | ||
795 | ppc_in_gpc_stride * ppc_index, val); | ||
796 | } | ||
797 | } | ||
798 | } | ||
799 | |||
800 | void gr_gp10b_set_circular_buffer_size(struct gk20a *g, u32 data) | ||
801 | { | ||
802 | struct gr_gk20a *gr = &g->gr; | ||
803 | u32 gpc_index, ppc_index, stride, val; | ||
804 | u32 cb_size_steady = data * 4, cb_size; | ||
805 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
806 | u32 ppc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_PPC_IN_GPC_STRIDE); | ||
807 | |||
808 | gk20a_dbg_fn(""); | ||
809 | |||
810 | if (cb_size_steady > gr->attrib_cb_size) | ||
811 | cb_size_steady = gr->attrib_cb_size; | ||
812 | if (gk20a_readl(g, gr_gpc0_ppc0_cbm_beta_cb_size_r()) != | ||
813 | gk20a_readl(g, | ||
814 | gr_gpc0_ppc0_cbm_beta_steady_state_cb_size_r())) { | ||
815 | cb_size = cb_size_steady + | ||
816 | (gr_gpc0_ppc0_cbm_beta_cb_size_v_gfxp_v() - | ||
817 | gr_gpc0_ppc0_cbm_beta_cb_size_v_default_v()); | ||
818 | } else { | ||
819 | cb_size = cb_size_steady; | ||
820 | } | ||
821 | |||
822 | gk20a_writel(g, gr_ds_tga_constraintlogic_beta_r(), | ||
823 | (gk20a_readl(g, gr_ds_tga_constraintlogic_beta_r()) & | ||
824 | ~gr_ds_tga_constraintlogic_beta_cbsize_f(~0)) | | ||
825 | gr_ds_tga_constraintlogic_beta_cbsize_f(cb_size_steady)); | ||
826 | |||
827 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | ||
828 | stride = gpc_stride * gpc_index; | ||
829 | |||
830 | for (ppc_index = 0; ppc_index < gr->gpc_ppc_count[gpc_index]; | ||
831 | ppc_index++) { | ||
832 | |||
833 | val = gk20a_readl(g, gr_gpc0_ppc0_cbm_beta_cb_size_r() + | ||
834 | stride + | ||
835 | ppc_in_gpc_stride * ppc_index); | ||
836 | |||
837 | val = set_field(val, | ||
838 | gr_gpc0_ppc0_cbm_beta_cb_size_v_m(), | ||
839 | gr_gpc0_ppc0_cbm_beta_cb_size_v_f(cb_size * | ||
840 | gr->pes_tpc_count[ppc_index][gpc_index])); | ||
841 | |||
842 | gk20a_writel(g, gr_gpc0_ppc0_cbm_beta_cb_size_r() + | ||
843 | stride + | ||
844 | ppc_in_gpc_stride * ppc_index, val); | ||
845 | |||
846 | gk20a_writel(g, ppc_in_gpc_stride * ppc_index + | ||
847 | gr_gpc0_ppc0_cbm_beta_steady_state_cb_size_r() + | ||
848 | stride, | ||
849 | gr_gpc0_ppc0_cbm_beta_steady_state_cb_size_v_f( | ||
850 | cb_size_steady)); | ||
851 | |||
852 | val = gk20a_readl(g, gr_gpcs_swdx_tc_beta_cb_size_r( | ||
853 | ppc_index + gpc_index)); | ||
854 | |||
855 | val = set_field(val, | ||
856 | gr_gpcs_swdx_tc_beta_cb_size_v_m(), | ||
857 | gr_gpcs_swdx_tc_beta_cb_size_v_f( | ||
858 | cb_size_steady * | ||
859 | gr->gpc_ppc_count[gpc_index])); | ||
860 | |||
861 | gk20a_writel(g, gr_gpcs_swdx_tc_beta_cb_size_r( | ||
862 | ppc_index + gpc_index), val); | ||
863 | } | ||
864 | } | ||
865 | } | ||
866 | |||
867 | int gr_gp10b_init_ctx_state(struct gk20a *g) | ||
868 | { | ||
869 | struct fecs_method_op_gk20a op = { | ||
870 | .mailbox = { .id = 0, .data = 0, | ||
871 | .clr = ~0, .ok = 0, .fail = 0}, | ||
872 | .method.data = 0, | ||
873 | .cond.ok = GR_IS_UCODE_OP_NOT_EQUAL, | ||
874 | .cond.fail = GR_IS_UCODE_OP_SKIP, | ||
875 | }; | ||
876 | int err; | ||
877 | |||
878 | gk20a_dbg_fn(""); | ||
879 | |||
880 | err = gr_gk20a_init_ctx_state(g); | ||
881 | if (err) | ||
882 | return err; | ||
883 | |||
884 | if (!g->gr.t18x.ctx_vars.preempt_image_size) { | ||
885 | op.method.addr = | ||
886 | gr_fecs_method_push_adr_discover_preemption_image_size_v(); | ||
887 | op.mailbox.ret = &g->gr.t18x.ctx_vars.preempt_image_size; | ||
888 | err = gr_gk20a_submit_fecs_method_op(g, op, false); | ||
889 | if (err) { | ||
890 | nvgpu_err(g, "query preempt image size failed"); | ||
891 | return err; | ||
892 | } | ||
893 | } | ||
894 | |||
895 | gk20a_dbg_info("preempt image size: %u", | ||
896 | g->gr.t18x.ctx_vars.preempt_image_size); | ||
897 | |||
898 | gk20a_dbg_fn("done"); | ||
899 | |||
900 | return 0; | ||
901 | } | ||
902 | |||
903 | int gr_gp10b_alloc_buffer(struct vm_gk20a *vm, size_t size, | ||
904 | struct nvgpu_mem *mem) | ||
905 | { | ||
906 | int err; | ||
907 | |||
908 | gk20a_dbg_fn(""); | ||
909 | |||
910 | err = nvgpu_dma_alloc_sys(vm->mm->g, size, mem); | ||
911 | if (err) | ||
912 | return err; | ||
913 | |||
914 | mem->gpu_va = nvgpu_gmmu_map(vm, | ||
915 | mem, | ||
916 | mem->aligned_size, | ||
917 | NVGPU_AS_MAP_BUFFER_FLAGS_CACHEABLE, | ||
918 | gk20a_mem_flag_none, | ||
919 | false, | ||
920 | mem->aperture); | ||
921 | |||
922 | if (!mem->gpu_va) { | ||
923 | err = -ENOMEM; | ||
924 | goto fail_free; | ||
925 | } | ||
926 | |||
927 | return 0; | ||
928 | |||
929 | fail_free: | ||
930 | nvgpu_dma_free(vm->mm->g, mem); | ||
931 | return err; | ||
932 | } | ||
933 | |||
934 | int gr_gp10b_set_ctxsw_preemption_mode(struct gk20a *g, | ||
935 | struct gr_ctx_desc *gr_ctx, | ||
936 | struct vm_gk20a *vm, u32 class, | ||
937 | u32 graphics_preempt_mode, | ||
938 | u32 compute_preempt_mode) | ||
939 | { | ||
940 | int err = 0; | ||
941 | |||
942 | if (g->ops.gr.is_valid_gfx_class(g, class) && | ||
943 | g->gr.t18x.ctx_vars.force_preemption_gfxp) | ||
944 | graphics_preempt_mode = NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
945 | |||
946 | if (g->ops.gr.is_valid_compute_class(g, class) && | ||
947 | g->gr.t18x.ctx_vars.force_preemption_cilp) | ||
948 | compute_preempt_mode = NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
949 | |||
950 | /* check for invalid combinations */ | ||
951 | if ((graphics_preempt_mode == 0) && (compute_preempt_mode == 0)) | ||
952 | return -EINVAL; | ||
953 | |||
954 | if ((graphics_preempt_mode == NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) && | ||
955 | (compute_preempt_mode == NVGPU_PREEMPTION_MODE_COMPUTE_CILP)) | ||
956 | return -EINVAL; | ||
957 | |||
958 | /* Do not allow lower preemption modes than current ones */ | ||
959 | if (graphics_preempt_mode && | ||
960 | (graphics_preempt_mode < gr_ctx->graphics_preempt_mode)) | ||
961 | return -EINVAL; | ||
962 | |||
963 | if (compute_preempt_mode && | ||
964 | (compute_preempt_mode < gr_ctx->compute_preempt_mode)) | ||
965 | return -EINVAL; | ||
966 | |||
967 | /* set preemption modes */ | ||
968 | switch (graphics_preempt_mode) { | ||
969 | case NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP: | ||
970 | { | ||
971 | u32 spill_size = | ||
972 | gr_gpc0_swdx_rm_spill_buffer_size_256b_default_v() * | ||
973 | gr_gpc0_swdx_rm_spill_buffer_size_256b_byte_granularity_v(); | ||
974 | u32 pagepool_size = g->ops.gr.pagepool_default_size(g) * | ||
975 | gr_scc_pagepool_total_pages_byte_granularity_v(); | ||
976 | u32 betacb_size = g->gr.attrib_cb_default_size + | ||
977 | (gr_gpc0_ppc0_cbm_beta_cb_size_v_gfxp_v() - | ||
978 | gr_gpc0_ppc0_cbm_beta_cb_size_v_default_v()); | ||
979 | u32 attrib_cb_size = (betacb_size + g->gr.alpha_cb_size) * | ||
980 | gr_gpc0_ppc0_cbm_beta_cb_size_v_granularity_v() * | ||
981 | g->gr.max_tpc_count; | ||
982 | attrib_cb_size = ALIGN(attrib_cb_size, 128); | ||
983 | |||
984 | gk20a_dbg_info("gfxp context spill_size=%d", spill_size); | ||
985 | gk20a_dbg_info("gfxp context pagepool_size=%d", pagepool_size); | ||
986 | gk20a_dbg_info("gfxp context attrib_cb_size=%d", | ||
987 | attrib_cb_size); | ||
988 | |||
989 | err = gr_gp10b_alloc_buffer(vm, | ||
990 | g->gr.t18x.ctx_vars.preempt_image_size, | ||
991 | &gr_ctx->t18x.preempt_ctxsw_buffer); | ||
992 | if (err) { | ||
993 | nvgpu_err(g, "cannot allocate preempt buffer"); | ||
994 | goto fail; | ||
995 | } | ||
996 | |||
997 | err = gr_gp10b_alloc_buffer(vm, | ||
998 | spill_size, | ||
999 | &gr_ctx->t18x.spill_ctxsw_buffer); | ||
1000 | if (err) { | ||
1001 | nvgpu_err(g, "cannot allocate spill buffer"); | ||
1002 | goto fail_free_preempt; | ||
1003 | } | ||
1004 | |||
1005 | err = gr_gp10b_alloc_buffer(vm, | ||
1006 | attrib_cb_size, | ||
1007 | &gr_ctx->t18x.betacb_ctxsw_buffer); | ||
1008 | if (err) { | ||
1009 | nvgpu_err(g, "cannot allocate beta buffer"); | ||
1010 | goto fail_free_spill; | ||
1011 | } | ||
1012 | |||
1013 | err = gr_gp10b_alloc_buffer(vm, | ||
1014 | pagepool_size, | ||
1015 | &gr_ctx->t18x.pagepool_ctxsw_buffer); | ||
1016 | if (err) { | ||
1017 | nvgpu_err(g, "cannot allocate page pool"); | ||
1018 | goto fail_free_betacb; | ||
1019 | } | ||
1020 | |||
1021 | gr_ctx->graphics_preempt_mode = graphics_preempt_mode; | ||
1022 | break; | ||
1023 | } | ||
1024 | |||
1025 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
1026 | gr_ctx->graphics_preempt_mode = graphics_preempt_mode; | ||
1027 | break; | ||
1028 | |||
1029 | default: | ||
1030 | break; | ||
1031 | } | ||
1032 | |||
1033 | if (g->ops.gr.is_valid_compute_class(g, class) || | ||
1034 | g->ops.gr.is_valid_gfx_class(g, class)) { | ||
1035 | switch (compute_preempt_mode) { | ||
1036 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
1037 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
1038 | case NVGPU_PREEMPTION_MODE_COMPUTE_CILP: | ||
1039 | gr_ctx->compute_preempt_mode = compute_preempt_mode; | ||
1040 | break; | ||
1041 | default: | ||
1042 | break; | ||
1043 | } | ||
1044 | } | ||
1045 | |||
1046 | return 0; | ||
1047 | |||
1048 | fail_free_betacb: | ||
1049 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.betacb_ctxsw_buffer); | ||
1050 | fail_free_spill: | ||
1051 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.spill_ctxsw_buffer); | ||
1052 | fail_free_preempt: | ||
1053 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.preempt_ctxsw_buffer); | ||
1054 | fail: | ||
1055 | return err; | ||
1056 | } | ||
1057 | |||
1058 | int gr_gp10b_alloc_gr_ctx(struct gk20a *g, | ||
1059 | struct gr_ctx_desc **gr_ctx, struct vm_gk20a *vm, | ||
1060 | u32 class, | ||
1061 | u32 flags) | ||
1062 | { | ||
1063 | int err; | ||
1064 | u32 graphics_preempt_mode = 0; | ||
1065 | u32 compute_preempt_mode = 0; | ||
1066 | |||
1067 | gk20a_dbg_fn(""); | ||
1068 | |||
1069 | err = gr_gk20a_alloc_gr_ctx(g, gr_ctx, vm, class, flags); | ||
1070 | if (err) | ||
1071 | return err; | ||
1072 | |||
1073 | (*gr_ctx)->t18x.ctx_id_valid = false; | ||
1074 | |||
1075 | if (flags & NVGPU_OBJ_CTX_FLAGS_SUPPORT_GFXP) | ||
1076 | graphics_preempt_mode = NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
1077 | if (flags & NVGPU_OBJ_CTX_FLAGS_SUPPORT_CILP) | ||
1078 | compute_preempt_mode = NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
1079 | |||
1080 | if (graphics_preempt_mode || compute_preempt_mode) { | ||
1081 | if (g->ops.gr.set_ctxsw_preemption_mode) { | ||
1082 | err = g->ops.gr.set_ctxsw_preemption_mode(g, *gr_ctx, vm, | ||
1083 | class, graphics_preempt_mode, compute_preempt_mode); | ||
1084 | if (err) { | ||
1085 | nvgpu_err(g, "set_ctxsw_preemption_mode failed"); | ||
1086 | goto fail_free_gk20a_ctx; | ||
1087 | } | ||
1088 | } else | ||
1089 | goto fail_free_gk20a_ctx; | ||
1090 | } | ||
1091 | |||
1092 | gk20a_dbg_fn("done"); | ||
1093 | |||
1094 | return 0; | ||
1095 | |||
1096 | fail_free_gk20a_ctx: | ||
1097 | gr_gk20a_free_gr_ctx(g, vm, *gr_ctx); | ||
1098 | *gr_ctx = NULL; | ||
1099 | |||
1100 | return err; | ||
1101 | } | ||
1102 | |||
1103 | static void dump_ctx_switch_stats(struct gk20a *g, struct vm_gk20a *vm, | ||
1104 | struct gr_ctx_desc *gr_ctx) | ||
1105 | { | ||
1106 | struct nvgpu_mem *mem = &gr_ctx->mem; | ||
1107 | |||
1108 | if (nvgpu_mem_begin(g, mem)) { | ||
1109 | WARN_ON("Cannot map context"); | ||
1110 | return; | ||
1111 | } | ||
1112 | nvgpu_err(g, "ctxsw_prog_main_image_magic_value_o : %x (expect %x)", | ||
1113 | nvgpu_mem_rd(g, mem, | ||
1114 | ctxsw_prog_main_image_magic_value_o()), | ||
1115 | ctxsw_prog_main_image_magic_value_v_value_v()); | ||
1116 | |||
1117 | nvgpu_err(g, "ctxsw_prog_main_image_context_timestamp_buffer_ptr_hi : %x", | ||
1118 | nvgpu_mem_rd(g, mem, | ||
1119 | ctxsw_prog_main_image_context_timestamp_buffer_ptr_hi_o())); | ||
1120 | |||
1121 | nvgpu_err(g, "ctxsw_prog_main_image_context_timestamp_buffer_ptr : %x", | ||
1122 | nvgpu_mem_rd(g, mem, | ||
1123 | ctxsw_prog_main_image_context_timestamp_buffer_ptr_o())); | ||
1124 | |||
1125 | nvgpu_err(g, "ctxsw_prog_main_image_context_timestamp_buffer_control : %x", | ||
1126 | nvgpu_mem_rd(g, mem, | ||
1127 | ctxsw_prog_main_image_context_timestamp_buffer_control_o())); | ||
1128 | |||
1129 | nvgpu_err(g, "NUM_SAVE_OPERATIONS : %d", | ||
1130 | nvgpu_mem_rd(g, mem, | ||
1131 | ctxsw_prog_main_image_num_save_ops_o())); | ||
1132 | nvgpu_err(g, "WFI_SAVE_OPERATIONS : %d", | ||
1133 | nvgpu_mem_rd(g, mem, | ||
1134 | ctxsw_prog_main_image_num_wfi_save_ops_o())); | ||
1135 | nvgpu_err(g, "CTA_SAVE_OPERATIONS : %d", | ||
1136 | nvgpu_mem_rd(g, mem, | ||
1137 | ctxsw_prog_main_image_num_cta_save_ops_o())); | ||
1138 | nvgpu_err(g, "GFXP_SAVE_OPERATIONS : %d", | ||
1139 | nvgpu_mem_rd(g, mem, | ||
1140 | ctxsw_prog_main_image_num_gfxp_save_ops_o())); | ||
1141 | nvgpu_err(g, "CILP_SAVE_OPERATIONS : %d", | ||
1142 | nvgpu_mem_rd(g, mem, | ||
1143 | ctxsw_prog_main_image_num_cilp_save_ops_o())); | ||
1144 | nvgpu_err(g, | ||
1145 | "image gfx preemption option (GFXP is 1) %x", | ||
1146 | nvgpu_mem_rd(g, mem, | ||
1147 | ctxsw_prog_main_image_graphics_preemption_options_o())); | ||
1148 | nvgpu_err(g, | ||
1149 | "image compute preemption option (CTA is 1) %x", | ||
1150 | nvgpu_mem_rd(g, mem, | ||
1151 | ctxsw_prog_main_image_compute_preemption_options_o())); | ||
1152 | nvgpu_mem_end(g, mem); | ||
1153 | } | ||
1154 | |||
1155 | void gr_gp10b_free_gr_ctx(struct gk20a *g, struct vm_gk20a *vm, | ||
1156 | struct gr_ctx_desc *gr_ctx) | ||
1157 | { | ||
1158 | gk20a_dbg_fn(""); | ||
1159 | |||
1160 | if (!gr_ctx) | ||
1161 | return; | ||
1162 | |||
1163 | if (g->gr.t18x.ctx_vars.dump_ctxsw_stats_on_channel_close) | ||
1164 | dump_ctx_switch_stats(g, vm, gr_ctx); | ||
1165 | |||
1166 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.pagepool_ctxsw_buffer); | ||
1167 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.betacb_ctxsw_buffer); | ||
1168 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.spill_ctxsw_buffer); | ||
1169 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.preempt_ctxsw_buffer); | ||
1170 | gr_gk20a_free_gr_ctx(g, vm, gr_ctx); | ||
1171 | gk20a_dbg_fn("done"); | ||
1172 | } | ||
1173 | |||
1174 | |||
1175 | void gr_gp10b_update_ctxsw_preemption_mode(struct gk20a *g, | ||
1176 | struct channel_ctx_gk20a *ch_ctx, | ||
1177 | struct nvgpu_mem *mem) | ||
1178 | { | ||
1179 | struct gr_ctx_desc *gr_ctx = ch_ctx->gr_ctx; | ||
1180 | struct ctx_header_desc *ctx = &ch_ctx->ctx_header; | ||
1181 | struct nvgpu_mem *ctxheader = &ctx->mem; | ||
1182 | |||
1183 | u32 gfxp_preempt_option = | ||
1184 | ctxsw_prog_main_image_graphics_preemption_options_control_gfxp_f(); | ||
1185 | u32 cilp_preempt_option = | ||
1186 | ctxsw_prog_main_image_compute_preemption_options_control_cilp_f(); | ||
1187 | u32 cta_preempt_option = | ||
1188 | ctxsw_prog_main_image_compute_preemption_options_control_cta_f(); | ||
1189 | int err; | ||
1190 | |||
1191 | gk20a_dbg_fn(""); | ||
1192 | |||
1193 | if (gr_ctx->graphics_preempt_mode == NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) { | ||
1194 | gk20a_dbg_info("GfxP: %x", gfxp_preempt_option); | ||
1195 | nvgpu_mem_wr(g, mem, | ||
1196 | ctxsw_prog_main_image_graphics_preemption_options_o(), | ||
1197 | gfxp_preempt_option); | ||
1198 | } | ||
1199 | |||
1200 | if (gr_ctx->compute_preempt_mode == NVGPU_PREEMPTION_MODE_COMPUTE_CILP) { | ||
1201 | gk20a_dbg_info("CILP: %x", cilp_preempt_option); | ||
1202 | nvgpu_mem_wr(g, mem, | ||
1203 | ctxsw_prog_main_image_compute_preemption_options_o(), | ||
1204 | cilp_preempt_option); | ||
1205 | } | ||
1206 | |||
1207 | if (gr_ctx->compute_preempt_mode == NVGPU_PREEMPTION_MODE_COMPUTE_CTA) { | ||
1208 | gk20a_dbg_info("CTA: %x", cta_preempt_option); | ||
1209 | nvgpu_mem_wr(g, mem, | ||
1210 | ctxsw_prog_main_image_compute_preemption_options_o(), | ||
1211 | cta_preempt_option); | ||
1212 | } | ||
1213 | |||
1214 | if (gr_ctx->t18x.preempt_ctxsw_buffer.gpu_va) { | ||
1215 | u32 addr; | ||
1216 | u32 size; | ||
1217 | u32 cbes_reserve; | ||
1218 | |||
1219 | if (g->ops.gr.set_preemption_buffer_va) { | ||
1220 | if (ctxheader->gpu_va) | ||
1221 | g->ops.gr.set_preemption_buffer_va(g, ctxheader, | ||
1222 | gr_ctx->t18x.preempt_ctxsw_buffer.gpu_va); | ||
1223 | else | ||
1224 | g->ops.gr.set_preemption_buffer_va(g, mem, | ||
1225 | gr_ctx->t18x.preempt_ctxsw_buffer.gpu_va); | ||
1226 | } | ||
1227 | |||
1228 | err = gr_gk20a_ctx_patch_write_begin(g, ch_ctx, true); | ||
1229 | if (err) { | ||
1230 | nvgpu_err(g, "can't map patch context"); | ||
1231 | goto out; | ||
1232 | } | ||
1233 | |||
1234 | addr = (u64_lo32(gr_ctx->t18x.betacb_ctxsw_buffer.gpu_va) >> | ||
1235 | gr_gpcs_setup_attrib_cb_base_addr_39_12_align_bits_v()) | | ||
1236 | (u64_hi32(gr_ctx->t18x.betacb_ctxsw_buffer.gpu_va) << | ||
1237 | (32 - gr_gpcs_setup_attrib_cb_base_addr_39_12_align_bits_v())); | ||
1238 | |||
1239 | gk20a_dbg_info("attrib cb addr : 0x%016x", addr); | ||
1240 | g->ops.gr.commit_global_attrib_cb(g, ch_ctx, addr, true); | ||
1241 | |||
1242 | addr = (u64_lo32(gr_ctx->t18x.pagepool_ctxsw_buffer.gpu_va) >> | ||
1243 | gr_scc_pagepool_base_addr_39_8_align_bits_v()) | | ||
1244 | (u64_hi32(gr_ctx->t18x.pagepool_ctxsw_buffer.gpu_va) << | ||
1245 | (32 - gr_scc_pagepool_base_addr_39_8_align_bits_v())); | ||
1246 | size = gr_ctx->t18x.pagepool_ctxsw_buffer.size; | ||
1247 | |||
1248 | if (size == g->ops.gr.pagepool_default_size(g)) | ||
1249 | size = gr_scc_pagepool_total_pages_hwmax_v(); | ||
1250 | |||
1251 | g->ops.gr.commit_global_pagepool(g, ch_ctx, addr, size, true); | ||
1252 | |||
1253 | addr = (u64_lo32(gr_ctx->t18x.spill_ctxsw_buffer.gpu_va) >> | ||
1254 | gr_gpc0_swdx_rm_spill_buffer_addr_39_8_align_bits_v()) | | ||
1255 | (u64_hi32(gr_ctx->t18x.spill_ctxsw_buffer.gpu_va) << | ||
1256 | (32 - gr_gpc0_swdx_rm_spill_buffer_addr_39_8_align_bits_v())); | ||
1257 | size = gr_ctx->t18x.spill_ctxsw_buffer.size / | ||
1258 | gr_gpc0_swdx_rm_spill_buffer_size_256b_byte_granularity_v(); | ||
1259 | |||
1260 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
1261 | gr_gpc0_swdx_rm_spill_buffer_addr_r(), | ||
1262 | gr_gpc0_swdx_rm_spill_buffer_addr_39_8_f(addr), | ||
1263 | true); | ||
1264 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
1265 | gr_gpc0_swdx_rm_spill_buffer_size_r(), | ||
1266 | gr_gpc0_swdx_rm_spill_buffer_size_256b_f(size), | ||
1267 | true); | ||
1268 | |||
1269 | cbes_reserve = gr_gpcs_swdx_beta_cb_ctrl_cbes_reserve_gfxp_v(); | ||
1270 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
1271 | gr_gpcs_swdx_beta_cb_ctrl_r(), | ||
1272 | gr_gpcs_swdx_beta_cb_ctrl_cbes_reserve_f( | ||
1273 | cbes_reserve), | ||
1274 | true); | ||
1275 | gr_gk20a_ctx_patch_write(g, ch_ctx, | ||
1276 | gr_gpcs_ppcs_cbm_beta_cb_ctrl_r(), | ||
1277 | gr_gpcs_ppcs_cbm_beta_cb_ctrl_cbes_reserve_f( | ||
1278 | cbes_reserve), | ||
1279 | true); | ||
1280 | |||
1281 | gr_gk20a_ctx_patch_write_end(g, ch_ctx, true); | ||
1282 | } | ||
1283 | |||
1284 | out: | ||
1285 | gk20a_dbg_fn("done"); | ||
1286 | } | ||
1287 | |||
1288 | int gr_gp10b_dump_gr_status_regs(struct gk20a *g, | ||
1289 | struct gk20a_debug_output *o) | ||
1290 | { | ||
1291 | struct gr_gk20a *gr = &g->gr; | ||
1292 | u32 gr_engine_id; | ||
1293 | |||
1294 | gr_engine_id = gk20a_fifo_get_gr_engine_id(g); | ||
1295 | |||
1296 | gk20a_debug_output(o, "NV_PGRAPH_STATUS: 0x%x\n", | ||
1297 | gk20a_readl(g, gr_status_r())); | ||
1298 | gk20a_debug_output(o, "NV_PGRAPH_STATUS1: 0x%x\n", | ||
1299 | gk20a_readl(g, gr_status_1_r())); | ||
1300 | gk20a_debug_output(o, "NV_PGRAPH_STATUS2: 0x%x\n", | ||
1301 | gk20a_readl(g, gr_status_2_r())); | ||
1302 | gk20a_debug_output(o, "NV_PGRAPH_ENGINE_STATUS: 0x%x\n", | ||
1303 | gk20a_readl(g, gr_engine_status_r())); | ||
1304 | gk20a_debug_output(o, "NV_PGRAPH_GRFIFO_STATUS : 0x%x\n", | ||
1305 | gk20a_readl(g, gr_gpfifo_status_r())); | ||
1306 | gk20a_debug_output(o, "NV_PGRAPH_GRFIFO_CONTROL : 0x%x\n", | ||
1307 | gk20a_readl(g, gr_gpfifo_ctl_r())); | ||
1308 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_HOST_INT_STATUS : 0x%x\n", | ||
1309 | gk20a_readl(g, gr_fecs_host_int_status_r())); | ||
1310 | gk20a_debug_output(o, "NV_PGRAPH_EXCEPTION : 0x%x\n", | ||
1311 | gk20a_readl(g, gr_exception_r())); | ||
1312 | gk20a_debug_output(o, "NV_PGRAPH_FECS_INTR : 0x%x\n", | ||
1313 | gk20a_readl(g, gr_fecs_intr_r())); | ||
1314 | gk20a_debug_output(o, "NV_PFIFO_ENGINE_STATUS(GR) : 0x%x\n", | ||
1315 | gk20a_readl(g, fifo_engine_status_r(gr_engine_id))); | ||
1316 | gk20a_debug_output(o, "NV_PGRAPH_ACTIVITY0: 0x%x\n", | ||
1317 | gk20a_readl(g, gr_activity_0_r())); | ||
1318 | gk20a_debug_output(o, "NV_PGRAPH_ACTIVITY1: 0x%x\n", | ||
1319 | gk20a_readl(g, gr_activity_1_r())); | ||
1320 | gk20a_debug_output(o, "NV_PGRAPH_ACTIVITY2: 0x%x\n", | ||
1321 | gk20a_readl(g, gr_activity_2_r())); | ||
1322 | gk20a_debug_output(o, "NV_PGRAPH_ACTIVITY4: 0x%x\n", | ||
1323 | gk20a_readl(g, gr_activity_4_r())); | ||
1324 | gk20a_debug_output(o, "NV_PGRAPH_PRI_SKED_ACTIVITY: 0x%x\n", | ||
1325 | gk20a_readl(g, gr_pri_sked_activity_r())); | ||
1326 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_ACTIVITY0: 0x%x\n", | ||
1327 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_activity0_r())); | ||
1328 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_ACTIVITY1: 0x%x\n", | ||
1329 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_activity1_r())); | ||
1330 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_ACTIVITY2: 0x%x\n", | ||
1331 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_activity2_r())); | ||
1332 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_ACTIVITY3: 0x%x\n", | ||
1333 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_activity3_r())); | ||
1334 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPC0_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1335 | gk20a_readl(g, gr_pri_gpc0_tpc0_tpccs_tpc_activity_0_r())); | ||
1336 | if (gr->gpc_tpc_count && gr->gpc_tpc_count[0] == 2) | ||
1337 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPC1_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1338 | gk20a_readl(g, gr_pri_gpc0_tpc1_tpccs_tpc_activity_0_r())); | ||
1339 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPCS_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1340 | gk20a_readl(g, gr_pri_gpc0_tpcs_tpccs_tpc_activity_0_r())); | ||
1341 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_GPCCS_GPC_ACTIVITY0: 0x%x\n", | ||
1342 | gk20a_readl(g, gr_pri_gpcs_gpccs_gpc_activity_0_r())); | ||
1343 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_GPCCS_GPC_ACTIVITY1: 0x%x\n", | ||
1344 | gk20a_readl(g, gr_pri_gpcs_gpccs_gpc_activity_1_r())); | ||
1345 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_GPCCS_GPC_ACTIVITY2: 0x%x\n", | ||
1346 | gk20a_readl(g, gr_pri_gpcs_gpccs_gpc_activity_2_r())); | ||
1347 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_GPCCS_GPC_ACTIVITY3: 0x%x\n", | ||
1348 | gk20a_readl(g, gr_pri_gpcs_gpccs_gpc_activity_3_r())); | ||
1349 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_TPC0_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1350 | gk20a_readl(g, gr_pri_gpcs_tpc0_tpccs_tpc_activity_0_r())); | ||
1351 | if (gr->gpc_tpc_count && gr->gpc_tpc_count[0] == 2) | ||
1352 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_TPC1_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1353 | gk20a_readl(g, gr_pri_gpcs_tpc1_tpccs_tpc_activity_0_r())); | ||
1354 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPCS_TPCS_TPCCS_TPC_ACTIVITY0: 0x%x\n", | ||
1355 | gk20a_readl(g, gr_pri_gpcs_tpcs_tpccs_tpc_activity_0_r())); | ||
1356 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_BECS_BE_ACTIVITY0: 0x%x\n", | ||
1357 | gk20a_readl(g, gr_pri_be0_becs_be_activity0_r())); | ||
1358 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE1_BECS_BE_ACTIVITY0: 0x%x\n", | ||
1359 | gk20a_readl(g, gr_pri_be1_becs_be_activity0_r())); | ||
1360 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BES_BECS_BE_ACTIVITY0: 0x%x\n", | ||
1361 | gk20a_readl(g, gr_pri_bes_becs_be_activity0_r())); | ||
1362 | gk20a_debug_output(o, "NV_PGRAPH_PRI_DS_MPIPE_STATUS: 0x%x\n", | ||
1363 | gk20a_readl(g, gr_pri_ds_mpipe_status_r())); | ||
1364 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FE_GO_IDLE_TIMEOUT : 0x%x\n", | ||
1365 | gk20a_readl(g, gr_fe_go_idle_timeout_r())); | ||
1366 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FE_GO_IDLE_INFO : 0x%x\n", | ||
1367 | gk20a_readl(g, gr_pri_fe_go_idle_info_r())); | ||
1368 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPC0_TEX_M_TEX_SUBUNITS_STATUS: 0x%x\n", | ||
1369 | gk20a_readl(g, gr_pri_gpc0_tpc0_tex_m_tex_subunits_status_r())); | ||
1370 | gk20a_debug_output(o, "NV_PGRAPH_PRI_CWD_FS: 0x%x\n", | ||
1371 | gk20a_readl(g, gr_cwd_fs_r())); | ||
1372 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FE_TPC_FS: 0x%x\n", | ||
1373 | gk20a_readl(g, gr_fe_tpc_fs_r())); | ||
1374 | gk20a_debug_output(o, "NV_PGRAPH_PRI_CWD_GPC_TPC_ID(0): 0x%x\n", | ||
1375 | gk20a_readl(g, gr_cwd_gpc_tpc_id_r(0))); | ||
1376 | gk20a_debug_output(o, "NV_PGRAPH_PRI_CWD_SM_ID(0): 0x%x\n", | ||
1377 | gk20a_readl(g, gr_cwd_sm_id_r(0))); | ||
1378 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_CTXSW_STATUS_FE_0: 0x%x\n", | ||
1379 | gk20a_readl(g, gr_fecs_ctxsw_status_fe_0_r())); | ||
1380 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_CTXSW_STATUS_1: 0x%x\n", | ||
1381 | gk20a_readl(g, gr_fecs_ctxsw_status_1_r())); | ||
1382 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_CTXSW_STATUS_GPC_0: 0x%x\n", | ||
1383 | gk20a_readl(g, gr_gpc0_gpccs_ctxsw_status_gpc_0_r())); | ||
1384 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_CTXSW_STATUS_1: 0x%x\n", | ||
1385 | gk20a_readl(g, gr_gpc0_gpccs_ctxsw_status_1_r())); | ||
1386 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_CTXSW_IDLESTATE : 0x%x\n", | ||
1387 | gk20a_readl(g, gr_fecs_ctxsw_idlestate_r())); | ||
1388 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_CTXSW_IDLESTATE : 0x%x\n", | ||
1389 | gk20a_readl(g, gr_gpc0_gpccs_ctxsw_idlestate_r())); | ||
1390 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_CURRENT_CTX : 0x%x\n", | ||
1391 | gk20a_readl(g, gr_fecs_current_ctx_r())); | ||
1392 | gk20a_debug_output(o, "NV_PGRAPH_PRI_FECS_NEW_CTX : 0x%x\n", | ||
1393 | gk20a_readl(g, gr_fecs_new_ctx_r())); | ||
1394 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_CROP_STATUS1 : 0x%x\n", | ||
1395 | gk20a_readl(g, gr_pri_be0_crop_status1_r())); | ||
1396 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BES_CROP_STATUS1 : 0x%x\n", | ||
1397 | gk20a_readl(g, gr_pri_bes_crop_status1_r())); | ||
1398 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_ZROP_STATUS : 0x%x\n", | ||
1399 | gk20a_readl(g, gr_pri_be0_zrop_status_r())); | ||
1400 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_ZROP_STATUS2 : 0x%x\n", | ||
1401 | gk20a_readl(g, gr_pri_be0_zrop_status2_r())); | ||
1402 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BES_ZROP_STATUS : 0x%x\n", | ||
1403 | gk20a_readl(g, gr_pri_bes_zrop_status_r())); | ||
1404 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BES_ZROP_STATUS2 : 0x%x\n", | ||
1405 | gk20a_readl(g, gr_pri_bes_zrop_status2_r())); | ||
1406 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_BECS_BE_EXCEPTION: 0x%x\n", | ||
1407 | gk20a_readl(g, gr_pri_be0_becs_be_exception_r())); | ||
1408 | gk20a_debug_output(o, "NV_PGRAPH_PRI_BE0_BECS_BE_EXCEPTION_EN: 0x%x\n", | ||
1409 | gk20a_readl(g, gr_pri_be0_becs_be_exception_en_r())); | ||
1410 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_EXCEPTION: 0x%x\n", | ||
1411 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_exception_r())); | ||
1412 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_GPCCS_GPC_EXCEPTION_EN: 0x%x\n", | ||
1413 | gk20a_readl(g, gr_pri_gpc0_gpccs_gpc_exception_en_r())); | ||
1414 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPC0_TPCCS_TPC_EXCEPTION: 0x%x\n", | ||
1415 | gk20a_readl(g, gr_pri_gpc0_tpc0_tpccs_tpc_exception_r())); | ||
1416 | gk20a_debug_output(o, "NV_PGRAPH_PRI_GPC0_TPC0_TPCCS_TPC_EXCEPTION_EN: 0x%x\n", | ||
1417 | gk20a_readl(g, gr_pri_gpc0_tpc0_tpccs_tpc_exception_en_r())); | ||
1418 | return 0; | ||
1419 | } | ||
1420 | |||
1421 | static bool gr_activity_empty_or_preempted(u32 val) | ||
1422 | { | ||
1423 | while(val) { | ||
1424 | u32 v = val & 7; | ||
1425 | if (v != gr_activity_4_gpc0_empty_v() && | ||
1426 | v != gr_activity_4_gpc0_preempted_v()) | ||
1427 | return false; | ||
1428 | val >>= 3; | ||
1429 | } | ||
1430 | |||
1431 | return true; | ||
1432 | } | ||
1433 | |||
1434 | int gr_gp10b_wait_empty(struct gk20a *g, unsigned long duration_ms, | ||
1435 | u32 expect_delay) | ||
1436 | { | ||
1437 | u32 delay = expect_delay; | ||
1438 | bool gr_enabled; | ||
1439 | bool ctxsw_active; | ||
1440 | bool gr_busy; | ||
1441 | u32 gr_status; | ||
1442 | u32 activity0, activity1, activity2, activity4; | ||
1443 | struct nvgpu_timeout timeout; | ||
1444 | |||
1445 | gk20a_dbg_fn(""); | ||
1446 | |||
1447 | nvgpu_timeout_init(g, &timeout, duration_ms, NVGPU_TIMER_CPU_TIMER); | ||
1448 | |||
1449 | do { | ||
1450 | /* fmodel: host gets fifo_engine_status(gr) from gr | ||
1451 | only when gr_status is read */ | ||
1452 | gr_status = gk20a_readl(g, gr_status_r()); | ||
1453 | |||
1454 | gr_enabled = gk20a_readl(g, mc_enable_r()) & | ||
1455 | mc_enable_pgraph_enabled_f(); | ||
1456 | |||
1457 | ctxsw_active = gr_status & 1<<7; | ||
1458 | |||
1459 | activity0 = gk20a_readl(g, gr_activity_0_r()); | ||
1460 | activity1 = gk20a_readl(g, gr_activity_1_r()); | ||
1461 | activity2 = gk20a_readl(g, gr_activity_2_r()); | ||
1462 | activity4 = gk20a_readl(g, gr_activity_4_r()); | ||
1463 | |||
1464 | gr_busy = !(gr_activity_empty_or_preempted(activity0) && | ||
1465 | gr_activity_empty_or_preempted(activity1) && | ||
1466 | activity2 == 0 && | ||
1467 | gr_activity_empty_or_preempted(activity4)); | ||
1468 | |||
1469 | if (!gr_enabled || (!gr_busy && !ctxsw_active)) { | ||
1470 | gk20a_dbg_fn("done"); | ||
1471 | return 0; | ||
1472 | } | ||
1473 | |||
1474 | nvgpu_usleep_range(delay, delay * 2); | ||
1475 | delay = min_t(u32, delay << 1, GR_IDLE_CHECK_MAX); | ||
1476 | } while (!nvgpu_timeout_expired(&timeout)); | ||
1477 | |||
1478 | nvgpu_err(g, | ||
1479 | "timeout, ctxsw busy : %d, gr busy : %d, %08x, %08x, %08x, %08x", | ||
1480 | ctxsw_active, gr_busy, activity0, activity1, activity2, activity4); | ||
1481 | |||
1482 | return -EAGAIN; | ||
1483 | } | ||
1484 | |||
1485 | void gr_gp10b_commit_global_attrib_cb(struct gk20a *g, | ||
1486 | struct channel_ctx_gk20a *ch_ctx, | ||
1487 | u64 addr, bool patch) | ||
1488 | { | ||
1489 | struct gr_ctx_desc *gr_ctx = ch_ctx->gr_ctx; | ||
1490 | int attrBufferSize; | ||
1491 | |||
1492 | if (gr_ctx->t18x.preempt_ctxsw_buffer.gpu_va) | ||
1493 | attrBufferSize = gr_ctx->t18x.betacb_ctxsw_buffer.size; | ||
1494 | else | ||
1495 | attrBufferSize = g->ops.gr.calc_global_ctx_buffer_size(g); | ||
1496 | |||
1497 | attrBufferSize /= gr_gpcs_tpcs_tex_rm_cb_1_size_div_128b_granularity_f(); | ||
1498 | |||
1499 | gr_gm20b_commit_global_attrib_cb(g, ch_ctx, addr, patch); | ||
1500 | |||
1501 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_tpcs_mpc_vtg_cb_global_base_addr_r(), | ||
1502 | gr_gpcs_tpcs_mpc_vtg_cb_global_base_addr_v_f(addr) | | ||
1503 | gr_gpcs_tpcs_mpc_vtg_cb_global_base_addr_valid_true_f(), patch); | ||
1504 | |||
1505 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_tpcs_tex_rm_cb_0_r(), | ||
1506 | gr_gpcs_tpcs_tex_rm_cb_0_base_addr_43_12_f(addr), patch); | ||
1507 | |||
1508 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_tpcs_tex_rm_cb_1_r(), | ||
1509 | gr_gpcs_tpcs_tex_rm_cb_1_size_div_128b_f(attrBufferSize) | | ||
1510 | gr_gpcs_tpcs_tex_rm_cb_1_valid_true_f(), patch); | ||
1511 | } | ||
1512 | |||
1513 | void gr_gp10b_commit_global_bundle_cb(struct gk20a *g, | ||
1514 | struct channel_ctx_gk20a *ch_ctx, | ||
1515 | u64 addr, u64 size, bool patch) | ||
1516 | { | ||
1517 | u32 data; | ||
1518 | |||
1519 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_scc_bundle_cb_base_r(), | ||
1520 | gr_scc_bundle_cb_base_addr_39_8_f(addr), patch); | ||
1521 | |||
1522 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_scc_bundle_cb_size_r(), | ||
1523 | gr_scc_bundle_cb_size_div_256b_f(size) | | ||
1524 | gr_scc_bundle_cb_size_valid_true_f(), patch); | ||
1525 | |||
1526 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_swdx_bundle_cb_base_r(), | ||
1527 | gr_gpcs_swdx_bundle_cb_base_addr_39_8_f(addr), patch); | ||
1528 | |||
1529 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_gpcs_swdx_bundle_cb_size_r(), | ||
1530 | gr_gpcs_swdx_bundle_cb_size_div_256b_f(size) | | ||
1531 | gr_gpcs_swdx_bundle_cb_size_valid_true_f(), patch); | ||
1532 | |||
1533 | /* data for state_limit */ | ||
1534 | data = (g->gr.bundle_cb_default_size * | ||
1535 | gr_scc_bundle_cb_size_div_256b_byte_granularity_v()) / | ||
1536 | gr_pd_ab_dist_cfg2_state_limit_scc_bundle_granularity_v(); | ||
1537 | |||
1538 | data = min_t(u32, data, g->gr.min_gpm_fifo_depth); | ||
1539 | |||
1540 | gk20a_dbg_info("bundle cb token limit : %d, state limit : %d", | ||
1541 | g->gr.bundle_cb_token_limit, data); | ||
1542 | |||
1543 | gr_gk20a_ctx_patch_write(g, ch_ctx, gr_pd_ab_dist_cfg2_r(), | ||
1544 | gr_pd_ab_dist_cfg2_token_limit_f(g->gr.bundle_cb_token_limit) | | ||
1545 | gr_pd_ab_dist_cfg2_state_limit_f(data), patch); | ||
1546 | } | ||
1547 | |||
1548 | int gr_gp10b_load_smid_config(struct gk20a *g) | ||
1549 | { | ||
1550 | u32 *tpc_sm_id; | ||
1551 | u32 i, j; | ||
1552 | u32 tpc_index, gpc_index; | ||
1553 | u32 max_gpcs = nvgpu_get_litter_value(g, GPU_LIT_NUM_GPCS); | ||
1554 | |||
1555 | tpc_sm_id = nvgpu_kcalloc(g, gr_cwd_sm_id__size_1_v(), sizeof(u32)); | ||
1556 | if (!tpc_sm_id) | ||
1557 | return -ENOMEM; | ||
1558 | |||
1559 | /* Each NV_PGRAPH_PRI_CWD_GPC_TPC_ID can store 4 TPCs.*/ | ||
1560 | for (i = 0; i <= ((g->gr.tpc_count-1) / 4); i++) { | ||
1561 | u32 reg = 0; | ||
1562 | u32 bit_stride = gr_cwd_gpc_tpc_id_gpc0_s() + | ||
1563 | gr_cwd_gpc_tpc_id_tpc0_s(); | ||
1564 | |||
1565 | for (j = 0; j < 4; j++) { | ||
1566 | u32 sm_id = (i * 4) + j; | ||
1567 | u32 bits; | ||
1568 | |||
1569 | if (sm_id >= g->gr.tpc_count) | ||
1570 | break; | ||
1571 | |||
1572 | gpc_index = g->gr.sm_to_cluster[sm_id].gpc_index; | ||
1573 | tpc_index = g->gr.sm_to_cluster[sm_id].tpc_index; | ||
1574 | |||
1575 | bits = gr_cwd_gpc_tpc_id_gpc0_f(gpc_index) | | ||
1576 | gr_cwd_gpc_tpc_id_tpc0_f(tpc_index); | ||
1577 | reg |= bits << (j * bit_stride); | ||
1578 | |||
1579 | tpc_sm_id[gpc_index + max_gpcs * ((tpc_index & 4) >> 2)] | ||
1580 | |= sm_id << (bit_stride * (tpc_index & 3)); | ||
1581 | } | ||
1582 | gk20a_writel(g, gr_cwd_gpc_tpc_id_r(i), reg); | ||
1583 | } | ||
1584 | |||
1585 | for (i = 0; i < gr_cwd_sm_id__size_1_v(); i++) | ||
1586 | gk20a_writel(g, gr_cwd_sm_id_r(i), tpc_sm_id[i]); | ||
1587 | |||
1588 | nvgpu_kfree(g, tpc_sm_id); | ||
1589 | |||
1590 | return 0; | ||
1591 | } | ||
1592 | |||
1593 | int gr_gp10b_init_fs_state(struct gk20a *g) | ||
1594 | { | ||
1595 | u32 data; | ||
1596 | |||
1597 | gk20a_dbg_fn(""); | ||
1598 | |||
1599 | data = gk20a_readl(g, gr_gpcs_tpcs_sm_texio_control_r()); | ||
1600 | data = set_field(data, gr_gpcs_tpcs_sm_texio_control_oor_addr_check_mode_m(), | ||
1601 | gr_gpcs_tpcs_sm_texio_control_oor_addr_check_mode_arm_63_48_match_f()); | ||
1602 | gk20a_writel(g, gr_gpcs_tpcs_sm_texio_control_r(), data); | ||
1603 | |||
1604 | data = gk20a_readl(g, gr_gpcs_tpcs_sm_disp_ctrl_r()); | ||
1605 | data = set_field(data, gr_gpcs_tpcs_sm_disp_ctrl_re_suppress_m(), | ||
1606 | gr_gpcs_tpcs_sm_disp_ctrl_re_suppress_disable_f()); | ||
1607 | gk20a_writel(g, gr_gpcs_tpcs_sm_disp_ctrl_r(), data); | ||
1608 | |||
1609 | if (g->gr.t18x.fecs_feature_override_ecc_val != 0) { | ||
1610 | gk20a_writel(g, | ||
1611 | gr_fecs_feature_override_ecc_r(), | ||
1612 | g->gr.t18x.fecs_feature_override_ecc_val); | ||
1613 | } | ||
1614 | |||
1615 | return gr_gm20b_init_fs_state(g); | ||
1616 | } | ||
1617 | |||
1618 | void gr_gp10b_set_gpc_tpc_mask(struct gk20a *g, u32 gpc_index) | ||
1619 | { | ||
1620 | nvgpu_tegra_fuse_write_bypass(g, 0x1); | ||
1621 | nvgpu_tegra_fuse_write_access_sw(g, 0x0); | ||
1622 | |||
1623 | if (g->gr.gpc_tpc_mask[gpc_index] == 0x1) | ||
1624 | nvgpu_tegra_fuse_write_opt_gpu_tpc0_disable(g, 0x2); | ||
1625 | else if (g->gr.gpc_tpc_mask[gpc_index] == 0x2) | ||
1626 | nvgpu_tegra_fuse_write_opt_gpu_tpc0_disable(g, 0x1); | ||
1627 | else | ||
1628 | nvgpu_tegra_fuse_write_opt_gpu_tpc0_disable(g, 0x0); | ||
1629 | } | ||
1630 | |||
1631 | void gr_gp10b_get_access_map(struct gk20a *g, | ||
1632 | u32 **whitelist, int *num_entries) | ||
1633 | { | ||
1634 | static u32 wl_addr_gp10b[] = { | ||
1635 | /* this list must be sorted (low to high) */ | ||
1636 | 0x404468, /* gr_pri_mme_max_instructions */ | ||
1637 | 0x418300, /* gr_pri_gpcs_rasterarb_line_class */ | ||
1638 | 0x418800, /* gr_pri_gpcs_setup_debug */ | ||
1639 | 0x418e00, /* gr_pri_gpcs_swdx_config */ | ||
1640 | 0x418e40, /* gr_pri_gpcs_swdx_tc_bundle_ctrl */ | ||
1641 | 0x418e44, /* gr_pri_gpcs_swdx_tc_bundle_ctrl */ | ||
1642 | 0x418e48, /* gr_pri_gpcs_swdx_tc_bundle_ctrl */ | ||
1643 | 0x418e4c, /* gr_pri_gpcs_swdx_tc_bundle_ctrl */ | ||
1644 | 0x418e50, /* gr_pri_gpcs_swdx_tc_bundle_ctrl */ | ||
1645 | 0x418e58, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1646 | 0x418e5c, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1647 | 0x418e60, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1648 | 0x418e64, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1649 | 0x418e68, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1650 | 0x418e6c, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1651 | 0x418e70, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1652 | 0x418e74, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1653 | 0x418e78, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1654 | 0x418e7c, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1655 | 0x418e80, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1656 | 0x418e84, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1657 | 0x418e88, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1658 | 0x418e8c, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1659 | 0x418e90, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1660 | 0x418e94, /* gr_pri_gpcs_swdx_tc_bundle_addr */ | ||
1661 | 0x419864, /* gr_pri_gpcs_tpcs_pe_l2_evict_policy */ | ||
1662 | 0x419a04, /* gr_pri_gpcs_tpcs_tex_lod_dbg */ | ||
1663 | 0x419a08, /* gr_pri_gpcs_tpcs_tex_samp_dbg */ | ||
1664 | 0x419e10, /* gr_pri_gpcs_tpcs_sm_dbgr_control0 */ | ||
1665 | 0x419f78, /* gr_pri_gpcs_tpcs_sm_disp_ctrl */ | ||
1666 | }; | ||
1667 | |||
1668 | *whitelist = wl_addr_gp10b; | ||
1669 | *num_entries = ARRAY_SIZE(wl_addr_gp10b); | ||
1670 | } | ||
1671 | |||
1672 | static int gr_gp10b_disable_channel_or_tsg(struct gk20a *g, struct channel_gk20a *fault_ch) | ||
1673 | { | ||
1674 | int ret = 0; | ||
1675 | |||
1676 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, ""); | ||
1677 | |||
1678 | ret = gk20a_disable_channel_tsg(g, fault_ch); | ||
1679 | if (ret) { | ||
1680 | nvgpu_err(g, | ||
1681 | "CILP: failed to disable channel/TSG!"); | ||
1682 | return ret; | ||
1683 | } | ||
1684 | |||
1685 | ret = g->ops.fifo.update_runlist(g, fault_ch->runlist_id, ~0, true, false); | ||
1686 | if (ret) { | ||
1687 | nvgpu_err(g, | ||
1688 | "CILP: failed to restart runlist 0!"); | ||
1689 | return ret; | ||
1690 | } | ||
1691 | |||
1692 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, "CILP: restarted runlist"); | ||
1693 | |||
1694 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1695 | "CILP: tsgid: 0x%x", fault_ch->tsgid); | ||
1696 | |||
1697 | if (gk20a_is_channel_marked_as_tsg(fault_ch)) { | ||
1698 | gk20a_fifo_issue_preempt(g, fault_ch->tsgid, true); | ||
1699 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1700 | "CILP: preempted tsg"); | ||
1701 | } else { | ||
1702 | gk20a_fifo_issue_preempt(g, fault_ch->chid, false); | ||
1703 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1704 | "CILP: preempted channel"); | ||
1705 | } | ||
1706 | |||
1707 | return ret; | ||
1708 | } | ||
1709 | |||
1710 | int gr_gp10b_set_cilp_preempt_pending(struct gk20a *g, | ||
1711 | struct channel_gk20a *fault_ch) | ||
1712 | { | ||
1713 | int ret; | ||
1714 | struct gr_ctx_desc *gr_ctx = fault_ch->ch_ctx.gr_ctx; | ||
1715 | |||
1716 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, ""); | ||
1717 | |||
1718 | if (!gr_ctx) | ||
1719 | return -EINVAL; | ||
1720 | |||
1721 | if (gr_ctx->t18x.cilp_preempt_pending) { | ||
1722 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1723 | "CILP is already pending for chid %d", | ||
1724 | fault_ch->chid); | ||
1725 | return 0; | ||
1726 | } | ||
1727 | |||
1728 | /* get ctx_id from the ucode image */ | ||
1729 | if (!gr_ctx->t18x.ctx_id_valid) { | ||
1730 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1731 | "CILP: looking up ctx id"); | ||
1732 | ret = gr_gk20a_get_ctx_id(g, fault_ch, &gr_ctx->t18x.ctx_id); | ||
1733 | if (ret) { | ||
1734 | nvgpu_err(g, "CILP: error looking up ctx id!"); | ||
1735 | return ret; | ||
1736 | } | ||
1737 | gr_ctx->t18x.ctx_id_valid = true; | ||
1738 | } | ||
1739 | |||
1740 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1741 | "CILP: ctx id is 0x%x", gr_ctx->t18x.ctx_id); | ||
1742 | |||
1743 | /* send ucode method to set ctxsw interrupt */ | ||
1744 | ret = gr_gk20a_submit_fecs_sideband_method_op(g, | ||
1745 | (struct fecs_method_op_gk20a) { | ||
1746 | .method.data = gr_ctx->t18x.ctx_id, | ||
1747 | .method.addr = | ||
1748 | gr_fecs_method_push_adr_configure_interrupt_completion_option_v(), | ||
1749 | .mailbox = { | ||
1750 | .id = 1 /* sideband */, .data = 0, | ||
1751 | .clr = ~0, .ret = NULL, | ||
1752 | .ok = gr_fecs_ctxsw_mailbox_value_pass_v(), | ||
1753 | .fail = 0}, | ||
1754 | .cond.ok = GR_IS_UCODE_OP_EQUAL, | ||
1755 | .cond.fail = GR_IS_UCODE_OP_SKIP}); | ||
1756 | |||
1757 | if (ret) { | ||
1758 | nvgpu_err(g, "CILP: failed to enable ctxsw interrupt!"); | ||
1759 | return ret; | ||
1760 | } | ||
1761 | |||
1762 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1763 | "CILP: enabled ctxsw completion interrupt"); | ||
1764 | |||
1765 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1766 | "CILP: disabling channel %d", | ||
1767 | fault_ch->chid); | ||
1768 | |||
1769 | ret = gr_gp10b_disable_channel_or_tsg(g, fault_ch); | ||
1770 | if (ret) { | ||
1771 | nvgpu_err(g, "CILP: failed to disable channel!!"); | ||
1772 | return ret; | ||
1773 | } | ||
1774 | |||
1775 | /* set cilp_preempt_pending = true and record the channel */ | ||
1776 | gr_ctx->t18x.cilp_preempt_pending = true; | ||
1777 | g->gr.t18x.cilp_preempt_pending_chid = fault_ch->chid; | ||
1778 | |||
1779 | if (gk20a_is_channel_marked_as_tsg(fault_ch)) { | ||
1780 | struct tsg_gk20a *tsg = &g->fifo.tsg[fault_ch->tsgid]; | ||
1781 | |||
1782 | gk20a_tsg_event_id_post_event(tsg, | ||
1783 | NVGPU_EVENT_ID_CILP_PREEMPTION_STARTED); | ||
1784 | } else { | ||
1785 | gk20a_channel_event_id_post_event(fault_ch, | ||
1786 | NVGPU_EVENT_ID_CILP_PREEMPTION_STARTED); | ||
1787 | } | ||
1788 | |||
1789 | return 0; | ||
1790 | } | ||
1791 | |||
1792 | static int gr_gp10b_clear_cilp_preempt_pending(struct gk20a *g, | ||
1793 | struct channel_gk20a *fault_ch) | ||
1794 | { | ||
1795 | struct gr_ctx_desc *gr_ctx = fault_ch->ch_ctx.gr_ctx; | ||
1796 | |||
1797 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, ""); | ||
1798 | |||
1799 | if (!gr_ctx) | ||
1800 | return -EINVAL; | ||
1801 | |||
1802 | /* The ucode is self-clearing, so all we need to do here is | ||
1803 | to clear cilp_preempt_pending. */ | ||
1804 | if (!gr_ctx->t18x.cilp_preempt_pending) { | ||
1805 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1806 | "CILP is already cleared for chid %d\n", | ||
1807 | fault_ch->chid); | ||
1808 | return 0; | ||
1809 | } | ||
1810 | |||
1811 | gr_ctx->t18x.cilp_preempt_pending = false; | ||
1812 | g->gr.t18x.cilp_preempt_pending_chid = -1; | ||
1813 | |||
1814 | return 0; | ||
1815 | } | ||
1816 | |||
1817 | /* @brief pre-process work on the SM exceptions to determine if we clear them or not. | ||
1818 | * | ||
1819 | * On Pascal, if we are in CILP preemtion mode, preempt the channel and handle errors with special processing | ||
1820 | */ | ||
1821 | int gr_gp10b_pre_process_sm_exception(struct gk20a *g, | ||
1822 | u32 gpc, u32 tpc, u32 sm, u32 global_esr, u32 warp_esr, | ||
1823 | bool sm_debugger_attached, struct channel_gk20a *fault_ch, | ||
1824 | bool *early_exit, bool *ignore_debugger) | ||
1825 | { | ||
1826 | int ret; | ||
1827 | bool cilp_enabled = false; | ||
1828 | u32 global_mask = 0, dbgr_control0, global_esr_copy; | ||
1829 | u32 gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_GPC_STRIDE); | ||
1830 | u32 tpc_in_gpc_stride = nvgpu_get_litter_value(g, GPU_LIT_TPC_IN_GPC_STRIDE); | ||
1831 | u32 offset = gpc_stride * gpc + tpc_in_gpc_stride * tpc; | ||
1832 | |||
1833 | *early_exit = false; | ||
1834 | *ignore_debugger = false; | ||
1835 | |||
1836 | if (fault_ch) | ||
1837 | cilp_enabled = (fault_ch->ch_ctx.gr_ctx->compute_preempt_mode == | ||
1838 | NVGPU_PREEMPTION_MODE_COMPUTE_CILP); | ||
1839 | |||
1840 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "SM Exception received on gpc %d tpc %d = %u\n", | ||
1841 | gpc, tpc, global_esr); | ||
1842 | |||
1843 | if (cilp_enabled && sm_debugger_attached) { | ||
1844 | if (global_esr & gr_gpc0_tpc0_sm_hww_global_esr_bpt_int_pending_f()) | ||
1845 | gk20a_writel(g, gr_gpc0_tpc0_sm_hww_global_esr_r() + offset, | ||
1846 | gr_gpc0_tpc0_sm_hww_global_esr_bpt_int_pending_f()); | ||
1847 | |||
1848 | if (global_esr & gr_gpc0_tpc0_sm_hww_global_esr_single_step_complete_pending_f()) | ||
1849 | gk20a_writel(g, gr_gpc0_tpc0_sm_hww_global_esr_r() + offset, | ||
1850 | gr_gpc0_tpc0_sm_hww_global_esr_single_step_complete_pending_f()); | ||
1851 | |||
1852 | global_mask = gr_gpc0_tpc0_sm_hww_global_esr_sm_to_sm_fault_pending_f() | | ||
1853 | gr_gpcs_tpcs_sm_hww_global_esr_l1_error_pending_f() | | ||
1854 | gr_gpcs_tpcs_sm_hww_global_esr_multiple_warp_errors_pending_f() | | ||
1855 | gr_gpcs_tpcs_sm_hww_global_esr_physical_stack_overflow_error_pending_f() | | ||
1856 | gr_gpcs_tpcs_sm_hww_global_esr_timeout_error_pending_f() | | ||
1857 | gr_gpcs_tpcs_sm_hww_global_esr_bpt_pause_pending_f(); | ||
1858 | |||
1859 | if (warp_esr != 0 || (global_esr & global_mask) != 0) { | ||
1860 | *ignore_debugger = true; | ||
1861 | |||
1862 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1863 | "CILP: starting wait for LOCKED_DOWN on gpc %d tpc %d\n", | ||
1864 | gpc, tpc); | ||
1865 | |||
1866 | if (gk20a_dbg_gpu_broadcast_stop_trigger(fault_ch)) { | ||
1867 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1868 | "CILP: Broadcasting STOP_TRIGGER from gpc %d tpc %d\n", | ||
1869 | gpc, tpc); | ||
1870 | g->ops.gr.suspend_all_sms(g, global_mask, false); | ||
1871 | |||
1872 | gk20a_dbg_gpu_clear_broadcast_stop_trigger(fault_ch); | ||
1873 | } else { | ||
1874 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1875 | "CILP: STOP_TRIGGER from gpc %d tpc %d\n", | ||
1876 | gpc, tpc); | ||
1877 | g->ops.gr.suspend_single_sm(g, gpc, tpc, sm, global_mask, true); | ||
1878 | } | ||
1879 | |||
1880 | /* reset the HWW errors after locking down */ | ||
1881 | global_esr_copy = g->ops.gr.get_sm_hww_global_esr(g, | ||
1882 | gpc, tpc, sm); | ||
1883 | g->ops.gr.clear_sm_hww(g, | ||
1884 | gpc, tpc, sm, global_esr_copy); | ||
1885 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1886 | "CILP: HWWs cleared for gpc %d tpc %d\n", | ||
1887 | gpc, tpc); | ||
1888 | |||
1889 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "CILP: Setting CILP preempt pending\n"); | ||
1890 | ret = gr_gp10b_set_cilp_preempt_pending(g, fault_ch); | ||
1891 | if (ret) { | ||
1892 | nvgpu_err(g, "CILP: error while setting CILP preempt pending!"); | ||
1893 | return ret; | ||
1894 | } | ||
1895 | |||
1896 | dbgr_control0 = gk20a_readl(g, gr_gpc0_tpc0_sm_dbgr_control0_r() + offset); | ||
1897 | if (dbgr_control0 & gr_gpcs_tpcs_sm_dbgr_control0_single_step_mode_enable_f()) { | ||
1898 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1899 | "CILP: clearing SINGLE_STEP_MODE before resume for gpc %d tpc %d\n", | ||
1900 | gpc, tpc); | ||
1901 | dbgr_control0 = set_field(dbgr_control0, | ||
1902 | gr_gpcs_tpcs_sm_dbgr_control0_single_step_mode_m(), | ||
1903 | gr_gpcs_tpcs_sm_dbgr_control0_single_step_mode_disable_f()); | ||
1904 | gk20a_writel(g, gr_gpc0_tpc0_sm_dbgr_control0_r() + offset, dbgr_control0); | ||
1905 | } | ||
1906 | |||
1907 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, | ||
1908 | "CILP: resume for gpc %d tpc %d\n", | ||
1909 | gpc, tpc); | ||
1910 | g->ops.gr.resume_single_sm(g, gpc, tpc, sm); | ||
1911 | |||
1912 | *ignore_debugger = true; | ||
1913 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "CILP: All done on gpc %d, tpc %d\n", gpc, tpc); | ||
1914 | } | ||
1915 | |||
1916 | *early_exit = true; | ||
1917 | } | ||
1918 | return 0; | ||
1919 | } | ||
1920 | |||
1921 | static int gr_gp10b_get_cilp_preempt_pending_chid(struct gk20a *g, int *__chid) | ||
1922 | { | ||
1923 | struct gr_ctx_desc *gr_ctx; | ||
1924 | struct channel_gk20a *ch; | ||
1925 | int chid; | ||
1926 | int ret = -EINVAL; | ||
1927 | |||
1928 | chid = g->gr.t18x.cilp_preempt_pending_chid; | ||
1929 | |||
1930 | ch = gk20a_channel_get(gk20a_fifo_channel_from_chid(g, chid)); | ||
1931 | if (!ch) | ||
1932 | return ret; | ||
1933 | |||
1934 | gr_ctx = ch->ch_ctx.gr_ctx; | ||
1935 | |||
1936 | if (gr_ctx->t18x.cilp_preempt_pending) { | ||
1937 | *__chid = chid; | ||
1938 | ret = 0; | ||
1939 | } | ||
1940 | |||
1941 | gk20a_channel_put(ch); | ||
1942 | |||
1943 | return ret; | ||
1944 | } | ||
1945 | |||
1946 | int gr_gp10b_handle_fecs_error(struct gk20a *g, | ||
1947 | struct channel_gk20a *__ch, | ||
1948 | struct gr_gk20a_isr_data *isr_data) | ||
1949 | { | ||
1950 | u32 gr_fecs_intr = gk20a_readl(g, gr_fecs_host_int_status_r()); | ||
1951 | struct channel_gk20a *ch; | ||
1952 | int chid = -1; | ||
1953 | int ret = 0; | ||
1954 | |||
1955 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, ""); | ||
1956 | |||
1957 | /* | ||
1958 | * INTR1 (bit 1 of the HOST_INT_STATUS_CTXSW_INTR) | ||
1959 | * indicates that a CILP ctxsw save has finished | ||
1960 | */ | ||
1961 | if (gr_fecs_intr & gr_fecs_host_int_status_ctxsw_intr_f(2)) { | ||
1962 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
1963 | "CILP: ctxsw save completed!\n"); | ||
1964 | |||
1965 | /* now clear the interrupt */ | ||
1966 | gk20a_writel(g, gr_fecs_host_int_clear_r(), | ||
1967 | gr_fecs_host_int_clear_ctxsw_intr1_clear_f()); | ||
1968 | |||
1969 | ret = gr_gp10b_get_cilp_preempt_pending_chid(g, &chid); | ||
1970 | if (ret) | ||
1971 | goto clean_up; | ||
1972 | |||
1973 | ch = gk20a_channel_get( | ||
1974 | gk20a_fifo_channel_from_chid(g, chid)); | ||
1975 | if (!ch) | ||
1976 | goto clean_up; | ||
1977 | |||
1978 | |||
1979 | /* set preempt_pending to false */ | ||
1980 | ret = gr_gp10b_clear_cilp_preempt_pending(g, ch); | ||
1981 | if (ret) { | ||
1982 | nvgpu_err(g, "CILP: error while unsetting CILP preempt pending!"); | ||
1983 | gk20a_channel_put(ch); | ||
1984 | goto clean_up; | ||
1985 | } | ||
1986 | |||
1987 | /* Post events to UMD */ | ||
1988 | gk20a_dbg_gpu_post_events(ch); | ||
1989 | |||
1990 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
1991 | struct tsg_gk20a *tsg = &g->fifo.tsg[ch->tsgid]; | ||
1992 | |||
1993 | gk20a_tsg_event_id_post_event(tsg, | ||
1994 | NVGPU_EVENT_ID_CILP_PREEMPTION_COMPLETE); | ||
1995 | } else { | ||
1996 | gk20a_channel_event_id_post_event(ch, | ||
1997 | NVGPU_EVENT_ID_CILP_PREEMPTION_COMPLETE); | ||
1998 | } | ||
1999 | |||
2000 | gk20a_channel_put(ch); | ||
2001 | } | ||
2002 | |||
2003 | clean_up: | ||
2004 | /* handle any remaining interrupts */ | ||
2005 | return gk20a_gr_handle_fecs_error(g, __ch, isr_data); | ||
2006 | } | ||
2007 | |||
2008 | u32 gp10b_gr_get_sm_hww_warp_esr(struct gk20a *g, | ||
2009 | u32 gpc, u32 tpc, u32 sm) | ||
2010 | { | ||
2011 | u32 offset = gk20a_gr_gpc_offset(g, gpc) + gk20a_gr_tpc_offset(g, tpc); | ||
2012 | u32 hww_warp_esr = gk20a_readl(g, | ||
2013 | gr_gpc0_tpc0_sm_hww_warp_esr_r() + offset); | ||
2014 | |||
2015 | if (!(hww_warp_esr & gr_gpc0_tpc0_sm_hww_warp_esr_addr_valid_m())) | ||
2016 | hww_warp_esr = set_field(hww_warp_esr, | ||
2017 | gr_gpc0_tpc0_sm_hww_warp_esr_addr_error_type_m(), | ||
2018 | gr_gpc0_tpc0_sm_hww_warp_esr_addr_error_type_none_f()); | ||
2019 | |||
2020 | return hww_warp_esr; | ||
2021 | } | ||
2022 | |||
2023 | u32 get_ecc_override_val(struct gk20a *g) | ||
2024 | { | ||
2025 | u32 val; | ||
2026 | |||
2027 | val = gk20a_readl(g, fuse_opt_ecc_en_r()); | ||
2028 | if (val) | ||
2029 | return gk20a_readl(g, gr_fecs_feature_override_ecc_r()); | ||
2030 | |||
2031 | return 0; | ||
2032 | } | ||
2033 | |||
2034 | static bool gr_gp10b_suspend_context(struct channel_gk20a *ch, | ||
2035 | bool *cilp_preempt_pending) | ||
2036 | { | ||
2037 | struct gk20a *g = ch->g; | ||
2038 | struct channel_ctx_gk20a *ch_ctx = &ch->ch_ctx; | ||
2039 | struct gr_ctx_desc *gr_ctx = ch_ctx->gr_ctx; | ||
2040 | bool ctx_resident = false; | ||
2041 | int err = 0; | ||
2042 | |||
2043 | *cilp_preempt_pending = false; | ||
2044 | |||
2045 | if (gk20a_is_channel_ctx_resident(ch)) { | ||
2046 | g->ops.gr.suspend_all_sms(g, 0, false); | ||
2047 | |||
2048 | if (gr_ctx->compute_preempt_mode == NVGPU_PREEMPTION_MODE_COMPUTE_CILP) { | ||
2049 | err = gr_gp10b_set_cilp_preempt_pending(g, ch); | ||
2050 | if (err) | ||
2051 | nvgpu_err(g, "unable to set CILP preempt pending"); | ||
2052 | else | ||
2053 | *cilp_preempt_pending = true; | ||
2054 | |||
2055 | g->ops.gr.resume_all_sms(g); | ||
2056 | } | ||
2057 | |||
2058 | ctx_resident = true; | ||
2059 | } else { | ||
2060 | gk20a_disable_channel_tsg(g, ch); | ||
2061 | } | ||
2062 | |||
2063 | return ctx_resident; | ||
2064 | } | ||
2065 | |||
2066 | int gr_gp10b_suspend_contexts(struct gk20a *g, | ||
2067 | struct dbg_session_gk20a *dbg_s, | ||
2068 | int *ctx_resident_ch_fd) | ||
2069 | { | ||
2070 | u32 delay = GR_IDLE_CHECK_DEFAULT; | ||
2071 | bool cilp_preempt_pending = false; | ||
2072 | struct channel_gk20a *cilp_preempt_pending_ch = NULL; | ||
2073 | struct channel_gk20a *ch; | ||
2074 | struct dbg_session_channel_data *ch_data; | ||
2075 | int err = 0; | ||
2076 | int local_ctx_resident_ch_fd = -1; | ||
2077 | bool ctx_resident; | ||
2078 | |||
2079 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); | ||
2080 | |||
2081 | err = gr_gk20a_disable_ctxsw(g); | ||
2082 | if (err) { | ||
2083 | nvgpu_err(g, "unable to stop gr ctxsw"); | ||
2084 | nvgpu_mutex_release(&g->dbg_sessions_lock); | ||
2085 | goto clean_up; | ||
2086 | } | ||
2087 | |||
2088 | nvgpu_mutex_acquire(&dbg_s->ch_list_lock); | ||
2089 | |||
2090 | list_for_each_entry(ch_data, &dbg_s->ch_list, ch_entry) { | ||
2091 | ch = g->fifo.channel + ch_data->chid; | ||
2092 | |||
2093 | ctx_resident = gr_gp10b_suspend_context(ch, | ||
2094 | &cilp_preempt_pending); | ||
2095 | if (ctx_resident) | ||
2096 | local_ctx_resident_ch_fd = ch_data->channel_fd; | ||
2097 | if (cilp_preempt_pending) | ||
2098 | cilp_preempt_pending_ch = ch; | ||
2099 | } | ||
2100 | |||
2101 | nvgpu_mutex_release(&dbg_s->ch_list_lock); | ||
2102 | |||
2103 | err = gr_gk20a_enable_ctxsw(g); | ||
2104 | if (err) { | ||
2105 | nvgpu_mutex_release(&g->dbg_sessions_lock); | ||
2106 | goto clean_up; | ||
2107 | } | ||
2108 | |||
2109 | nvgpu_mutex_release(&g->dbg_sessions_lock); | ||
2110 | |||
2111 | if (cilp_preempt_pending_ch) { | ||
2112 | struct channel_ctx_gk20a *ch_ctx = | ||
2113 | &cilp_preempt_pending_ch->ch_ctx; | ||
2114 | struct gr_ctx_desc *gr_ctx = ch_ctx->gr_ctx; | ||
2115 | struct nvgpu_timeout timeout; | ||
2116 | |||
2117 | gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, | ||
2118 | "CILP preempt pending, waiting %lu msecs for preemption", | ||
2119 | gk20a_get_gr_idle_timeout(g)); | ||
2120 | |||
2121 | nvgpu_timeout_init(g, &timeout, gk20a_get_gr_idle_timeout(g), | ||
2122 | NVGPU_TIMER_CPU_TIMER); | ||
2123 | do { | ||
2124 | if (!gr_ctx->t18x.cilp_preempt_pending) | ||
2125 | break; | ||
2126 | |||
2127 | nvgpu_usleep_range(delay, delay * 2); | ||
2128 | delay = min_t(u32, delay << 1, GR_IDLE_CHECK_MAX); | ||
2129 | } while (!nvgpu_timeout_expired(&timeout)); | ||
2130 | |||
2131 | /* If cilp is still pending at this point, timeout */ | ||
2132 | if (gr_ctx->t18x.cilp_preempt_pending) | ||
2133 | err = -ETIMEDOUT; | ||
2134 | } | ||
2135 | |||
2136 | *ctx_resident_ch_fd = local_ctx_resident_ch_fd; | ||
2137 | |||
2138 | clean_up: | ||
2139 | return err; | ||
2140 | } | ||
2141 | |||
2142 | int gr_gp10b_set_boosted_ctx(struct channel_gk20a *ch, | ||
2143 | bool boost) | ||
2144 | { | ||
2145 | struct gr_ctx_desc *gr_ctx = ch->ch_ctx.gr_ctx; | ||
2146 | struct gk20a *g = ch->g; | ||
2147 | struct nvgpu_mem *mem = &gr_ctx->mem; | ||
2148 | int err = 0; | ||
2149 | |||
2150 | gr_ctx->boosted_ctx = boost; | ||
2151 | |||
2152 | if (nvgpu_mem_begin(g, mem)) | ||
2153 | return -ENOMEM; | ||
2154 | |||
2155 | err = gk20a_disable_channel_tsg(g, ch); | ||
2156 | if (err) | ||
2157 | goto unmap_ctx; | ||
2158 | |||
2159 | err = gk20a_fifo_preempt(g, ch); | ||
2160 | if (err) | ||
2161 | goto enable_ch; | ||
2162 | |||
2163 | if (g->ops.gr.update_boosted_ctx) | ||
2164 | g->ops.gr.update_boosted_ctx(g, mem, gr_ctx); | ||
2165 | else | ||
2166 | err = -ENOSYS; | ||
2167 | |||
2168 | enable_ch: | ||
2169 | gk20a_enable_channel_tsg(g, ch); | ||
2170 | unmap_ctx: | ||
2171 | nvgpu_mem_end(g, mem); | ||
2172 | |||
2173 | return err; | ||
2174 | } | ||
2175 | |||
2176 | void gr_gp10b_update_boosted_ctx(struct gk20a *g, struct nvgpu_mem *mem, | ||
2177 | struct gr_ctx_desc *gr_ctx) { | ||
2178 | u32 v; | ||
2179 | |||
2180 | v = ctxsw_prog_main_image_pmu_options_boost_clock_frequencies_f( | ||
2181 | gr_ctx->boosted_ctx); | ||
2182 | nvgpu_mem_wr(g, mem, ctxsw_prog_main_image_pmu_options_o(), v); | ||
2183 | } | ||
2184 | |||
2185 | int gr_gp10b_set_preemption_mode(struct channel_gk20a *ch, | ||
2186 | u32 graphics_preempt_mode, | ||
2187 | u32 compute_preempt_mode) | ||
2188 | { | ||
2189 | struct gr_ctx_desc *gr_ctx = ch->ch_ctx.gr_ctx; | ||
2190 | struct channel_ctx_gk20a *ch_ctx = &ch->ch_ctx; | ||
2191 | struct gk20a *g = ch->g; | ||
2192 | struct tsg_gk20a *tsg; | ||
2193 | struct vm_gk20a *vm; | ||
2194 | struct nvgpu_mem *mem = &gr_ctx->mem; | ||
2195 | struct ctx_header_desc *ctx = &ch->ch_ctx.ctx_header; | ||
2196 | struct nvgpu_mem *ctxheader = &ctx->mem; | ||
2197 | u32 class; | ||
2198 | int err = 0; | ||
2199 | |||
2200 | class = ch->obj_class; | ||
2201 | if (!class) | ||
2202 | return -EINVAL; | ||
2203 | |||
2204 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
2205 | tsg = &g->fifo.tsg[ch->tsgid]; | ||
2206 | vm = tsg->vm; | ||
2207 | } else { | ||
2208 | vm = ch->vm; | ||
2209 | } | ||
2210 | |||
2211 | /* skip setting anything if both modes are already set */ | ||
2212 | if (graphics_preempt_mode && | ||
2213 | (graphics_preempt_mode == gr_ctx->graphics_preempt_mode)) | ||
2214 | graphics_preempt_mode = 0; | ||
2215 | |||
2216 | if (compute_preempt_mode && | ||
2217 | (compute_preempt_mode == gr_ctx->compute_preempt_mode)) | ||
2218 | compute_preempt_mode = 0; | ||
2219 | |||
2220 | if (graphics_preempt_mode == 0 && compute_preempt_mode == 0) | ||
2221 | return 0; | ||
2222 | |||
2223 | if (g->ops.gr.set_ctxsw_preemption_mode) { | ||
2224 | |||
2225 | gk20a_dbg(gpu_dbg_sched, "chid=%d tsgid=%d pid=%d " | ||
2226 | "graphics_preempt=%d compute_preempt=%d", | ||
2227 | ch->chid, | ||
2228 | ch->tsgid, | ||
2229 | ch->tgid, | ||
2230 | graphics_preempt_mode, | ||
2231 | compute_preempt_mode); | ||
2232 | err = g->ops.gr.set_ctxsw_preemption_mode(g, gr_ctx, vm, class, | ||
2233 | graphics_preempt_mode, compute_preempt_mode); | ||
2234 | if (err) { | ||
2235 | nvgpu_err(g, "set_ctxsw_preemption_mode failed"); | ||
2236 | return err; | ||
2237 | } | ||
2238 | } | ||
2239 | |||
2240 | if (nvgpu_mem_begin(g, mem)) | ||
2241 | return -ENOMEM; | ||
2242 | |||
2243 | if (nvgpu_mem_begin(g, ctxheader)) | ||
2244 | goto unamp_ctx_header; | ||
2245 | |||
2246 | err = gk20a_disable_channel_tsg(g, ch); | ||
2247 | if (err) | ||
2248 | goto unmap_ctx; | ||
2249 | |||
2250 | err = gk20a_fifo_preempt(g, ch); | ||
2251 | if (err) | ||
2252 | goto enable_ch; | ||
2253 | |||
2254 | if (g->ops.gr.update_ctxsw_preemption_mode) { | ||
2255 | g->ops.gr.update_ctxsw_preemption_mode(ch->g, | ||
2256 | ch_ctx, mem); | ||
2257 | |||
2258 | err = gr_gk20a_ctx_patch_write_begin(g, ch_ctx, true); | ||
2259 | if (err) { | ||
2260 | nvgpu_err(g, "can't map patch context"); | ||
2261 | goto enable_ch; | ||
2262 | } | ||
2263 | g->ops.gr.commit_global_cb_manager(g, ch, true); | ||
2264 | gr_gk20a_ctx_patch_write_end(g, ch_ctx, true); | ||
2265 | } | ||
2266 | |||
2267 | enable_ch: | ||
2268 | gk20a_enable_channel_tsg(g, ch); | ||
2269 | unmap_ctx: | ||
2270 | nvgpu_mem_end(g, ctxheader); | ||
2271 | unamp_ctx_header: | ||
2272 | nvgpu_mem_end(g, mem); | ||
2273 | |||
2274 | return err; | ||
2275 | } | ||
2276 | |||
2277 | int gr_gp10b_get_preemption_mode_flags(struct gk20a *g, | ||
2278 | struct nvgpu_preemption_modes_rec *preemption_modes_rec) | ||
2279 | { | ||
2280 | preemption_modes_rec->graphics_preemption_mode_flags = ( | ||
2281 | NVGPU_PREEMPTION_MODE_GRAPHICS_WFI | | ||
2282 | NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP); | ||
2283 | preemption_modes_rec->compute_preemption_mode_flags = ( | ||
2284 | NVGPU_PREEMPTION_MODE_COMPUTE_WFI | | ||
2285 | NVGPU_PREEMPTION_MODE_COMPUTE_CTA | | ||
2286 | NVGPU_PREEMPTION_MODE_COMPUTE_CILP); | ||
2287 | |||
2288 | preemption_modes_rec->default_graphics_preempt_mode = | ||
2289 | NVGPU_PREEMPTION_MODE_GRAPHICS_WFI; | ||
2290 | preemption_modes_rec->default_compute_preempt_mode = | ||
2291 | NVGPU_PREEMPTION_MODE_COMPUTE_WFI; | ||
2292 | |||
2293 | return 0; | ||
2294 | } | ||
2295 | |||
2296 | int gr_gp10b_init_preemption_state(struct gk20a *g) | ||
2297 | { | ||
2298 | u32 debug_2; | ||
2299 | struct gr_gk20a *gr = &g->gr; | ||
2300 | u32 sysclk_cycles = gr->gfxp_wfi_timeout_count; | ||
2301 | gk20a_writel(g, gr_fe_gfxp_wfi_timeout_r(), | ||
2302 | gr_fe_gfxp_wfi_timeout_count_f(sysclk_cycles)); | ||
2303 | |||
2304 | debug_2 = gk20a_readl(g, gr_debug_2_r()); | ||
2305 | debug_2 = set_field(debug_2, | ||
2306 | gr_debug_2_gfxp_wfi_always_injects_wfi_m(), | ||
2307 | gr_debug_2_gfxp_wfi_always_injects_wfi_enabled_f()); | ||
2308 | gk20a_writel(g, gr_debug_2_r(), debug_2); | ||
2309 | |||
2310 | return 0; | ||
2311 | } | ||
2312 | |||
2313 | void gr_gp10b_set_preemption_buffer_va(struct gk20a *g, | ||
2314 | struct nvgpu_mem *mem, u64 gpu_va) | ||
2315 | { | ||
2316 | u32 va = u64_lo32(gpu_va >> 8); | ||
2317 | |||
2318 | nvgpu_mem_wr(g, mem, | ||
2319 | ctxsw_prog_main_image_full_preemption_ptr_o(), va); | ||
2320 | |||
2321 | } | ||
2322 | |||
2323 | void gr_gp10b_init_czf_bypass(struct gk20a *g) | ||
2324 | { | ||
2325 | g->gr.czf_bypass = gr_gpc0_prop_debug1_czf_bypass_init_v(); | ||
2326 | } | ||
2327 | |||
2328 | int gr_gp10b_set_czf_bypass(struct gk20a *g, struct channel_gk20a *ch) | ||
2329 | { | ||
2330 | struct nvgpu_dbg_gpu_reg_op ops; | ||
2331 | |||
2332 | ops.op = REGOP(WRITE_32); | ||
2333 | ops.type = REGOP(TYPE_GR_CTX); | ||
2334 | ops.status = REGOP(STATUS_SUCCESS); | ||
2335 | ops.value_hi = 0; | ||
2336 | ops.and_n_mask_lo = gr_gpc0_prop_debug1_czf_bypass_m(); | ||
2337 | ops.and_n_mask_hi = 0; | ||
2338 | ops.offset = gr_gpc0_prop_debug1_r(); | ||
2339 | ops.value_lo = gr_gpc0_prop_debug1_czf_bypass_f( | ||
2340 | g->gr.czf_bypass); | ||
2341 | |||
2342 | return __gr_gk20a_exec_ctx_ops(ch, &ops, 1, 1, 0, false); | ||
2343 | } | ||
2344 | |||
2345 | void gr_gp10b_init_ctxsw_hdr_data(struct gk20a *g, struct nvgpu_mem *mem) | ||
2346 | { | ||
2347 | gk20a_gr_init_ctxsw_hdr_data(g, mem); | ||
2348 | |||
2349 | nvgpu_mem_wr(g, mem, | ||
2350 | ctxsw_prog_main_image_num_wfi_save_ops_o(), 0); | ||
2351 | nvgpu_mem_wr(g, mem, | ||
2352 | ctxsw_prog_main_image_num_cta_save_ops_o(), 0); | ||
2353 | nvgpu_mem_wr(g, mem, | ||
2354 | ctxsw_prog_main_image_num_gfxp_save_ops_o(), 0); | ||
2355 | nvgpu_mem_wr(g, mem, | ||
2356 | ctxsw_prog_main_image_num_cilp_save_ops_o(), 0); | ||
2357 | } | ||