diff options
Diffstat (limited to 'drivers/gpu/nvgpu/gk20a/pmu_gk20a.c')
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/pmu_gk20a.c | 830 |
1 files changed, 830 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c new file mode 100644 index 00000000..d8478b2d --- /dev/null +++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c | |||
@@ -0,0 +1,830 @@ | |||
1 | /* | ||
2 | * GK20A PMU (aka. gPMU outside gk20a context) | ||
3 | * | ||
4 | * Copyright (c) 2011-2017, 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 <uapi/linux/nvgpu.h> | ||
26 | |||
27 | #include <nvgpu/nvgpu_common.h> | ||
28 | #include <nvgpu/timers.h> | ||
29 | #include <nvgpu/kmem.h> | ||
30 | #include <nvgpu/dma.h> | ||
31 | #include <nvgpu/log.h> | ||
32 | #include <nvgpu/bug.h> | ||
33 | #include <nvgpu/firmware.h> | ||
34 | #include <nvgpu/falcon.h> | ||
35 | #include <nvgpu/mm.h> | ||
36 | |||
37 | #include "gk20a.h" | ||
38 | #include "gr_gk20a.h" | ||
39 | |||
40 | #include <nvgpu/hw/gk20a/hw_mc_gk20a.h> | ||
41 | #include <nvgpu/hw/gk20a/hw_pwr_gk20a.h> | ||
42 | #include <nvgpu/hw/gk20a/hw_top_gk20a.h> | ||
43 | |||
44 | #ifdef CONFIG_TEGRA_19x_GPU | ||
45 | #include "nvgpu_gpuid_t19x.h" | ||
46 | #endif | ||
47 | |||
48 | #define gk20a_dbg_pmu(fmt, arg...) \ | ||
49 | gk20a_dbg(gpu_dbg_pmu, fmt, ##arg) | ||
50 | |||
51 | bool nvgpu_find_hex_in_string(char *strings, struct gk20a *g, u32 *hex_pos) | ||
52 | { | ||
53 | u32 i = 0, j = strlen(strings); | ||
54 | |||
55 | for (; i < j; i++) { | ||
56 | if (strings[i] == '%') | ||
57 | if (strings[i + 1] == 'x' || strings[i + 1] == 'X') { | ||
58 | *hex_pos = i; | ||
59 | return true; | ||
60 | } | ||
61 | } | ||
62 | *hex_pos = -1; | ||
63 | return false; | ||
64 | } | ||
65 | |||
66 | static void print_pmu_trace(struct nvgpu_pmu *pmu) | ||
67 | { | ||
68 | struct gk20a *g = pmu->g; | ||
69 | u32 i = 0, j = 0, k, l, m, count; | ||
70 | char part_str[40], buf[0x40]; | ||
71 | void *tracebuffer; | ||
72 | char *trace; | ||
73 | u32 *trace1; | ||
74 | |||
75 | /* allocate system memory to copy pmu trace buffer */ | ||
76 | tracebuffer = nvgpu_kzalloc(g, GK20A_PMU_TRACE_BUFSIZE); | ||
77 | if (tracebuffer == NULL) | ||
78 | return; | ||
79 | |||
80 | /* read pmu traces into system memory buffer */ | ||
81 | nvgpu_mem_rd_n(g, &pmu->trace_buf, 0, tracebuffer, | ||
82 | GK20A_PMU_TRACE_BUFSIZE); | ||
83 | |||
84 | trace = (char *)tracebuffer; | ||
85 | trace1 = (u32 *)tracebuffer; | ||
86 | |||
87 | nvgpu_err(g, "dump PMU trace buffer"); | ||
88 | for (i = 0; i < GK20A_PMU_TRACE_BUFSIZE; i += 0x40) { | ||
89 | for (j = 0; j < 0x40; j++) | ||
90 | if (trace1[(i / 4) + j]) | ||
91 | break; | ||
92 | if (j == 0x40) | ||
93 | break; | ||
94 | count = scnprintf(buf, 0x40, "Index %x: ", trace1[(i / 4)]); | ||
95 | l = 0; | ||
96 | m = 0; | ||
97 | while (nvgpu_find_hex_in_string((trace+i+20+m), g, &k)) { | ||
98 | if (k >= 40) | ||
99 | break; | ||
100 | strncpy(part_str, (trace+i+20+m), k); | ||
101 | part_str[k] = 0; | ||
102 | count += scnprintf((buf + count), 0x40, "%s0x%x", | ||
103 | part_str, trace1[(i / 4) + 1 + l]); | ||
104 | l++; | ||
105 | m += k + 2; | ||
106 | } | ||
107 | |||
108 | scnprintf((buf + count), 0x40, "%s", (trace+i+20+m)); | ||
109 | nvgpu_err(g, "%s", buf); | ||
110 | } | ||
111 | |||
112 | nvgpu_kfree(g, tracebuffer); | ||
113 | } | ||
114 | |||
115 | void pmu_enable_irq(struct nvgpu_pmu *pmu, bool enable) | ||
116 | { | ||
117 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
118 | u32 intr_mask; | ||
119 | u32 intr_dest; | ||
120 | |||
121 | gk20a_dbg_fn(""); | ||
122 | |||
123 | g->ops.mc.intr_unit_config(g, MC_INTR_UNIT_DISABLE, true, | ||
124 | mc_intr_mask_0_pmu_enabled_f()); | ||
125 | g->ops.mc.intr_unit_config(g, MC_INTR_UNIT_DISABLE, false, | ||
126 | mc_intr_mask_1_pmu_enabled_f()); | ||
127 | |||
128 | nvgpu_flcn_set_irq(pmu->flcn, false, 0x0, 0x0); | ||
129 | |||
130 | if (enable) { | ||
131 | /* dest 0=falcon, 1=host; level 0=irq0, 1=irq1 */ | ||
132 | intr_dest = pwr_falcon_irqdest_host_gptmr_f(0) | | ||
133 | pwr_falcon_irqdest_host_wdtmr_f(1) | | ||
134 | pwr_falcon_irqdest_host_mthd_f(0) | | ||
135 | pwr_falcon_irqdest_host_ctxsw_f(0) | | ||
136 | pwr_falcon_irqdest_host_halt_f(1) | | ||
137 | pwr_falcon_irqdest_host_exterr_f(0) | | ||
138 | pwr_falcon_irqdest_host_swgen0_f(1) | | ||
139 | pwr_falcon_irqdest_host_swgen1_f(0) | | ||
140 | pwr_falcon_irqdest_host_ext_f(0xff) | | ||
141 | pwr_falcon_irqdest_target_gptmr_f(1) | | ||
142 | pwr_falcon_irqdest_target_wdtmr_f(0) | | ||
143 | pwr_falcon_irqdest_target_mthd_f(0) | | ||
144 | pwr_falcon_irqdest_target_ctxsw_f(0) | | ||
145 | pwr_falcon_irqdest_target_halt_f(0) | | ||
146 | pwr_falcon_irqdest_target_exterr_f(0) | | ||
147 | pwr_falcon_irqdest_target_swgen0_f(0) | | ||
148 | pwr_falcon_irqdest_target_swgen1_f(0) | | ||
149 | pwr_falcon_irqdest_target_ext_f(0xff); | ||
150 | |||
151 | /* 0=disable, 1=enable */ | ||
152 | intr_mask = pwr_falcon_irqmset_gptmr_f(1) | | ||
153 | pwr_falcon_irqmset_wdtmr_f(1) | | ||
154 | pwr_falcon_irqmset_mthd_f(0) | | ||
155 | pwr_falcon_irqmset_ctxsw_f(0) | | ||
156 | pwr_falcon_irqmset_halt_f(1) | | ||
157 | pwr_falcon_irqmset_exterr_f(1) | | ||
158 | pwr_falcon_irqmset_swgen0_f(1) | | ||
159 | pwr_falcon_irqmset_swgen1_f(1); | ||
160 | |||
161 | nvgpu_flcn_set_irq(pmu->flcn, true, intr_mask, intr_dest); | ||
162 | |||
163 | g->ops.mc.intr_unit_config(g, MC_INTR_UNIT_ENABLE, true, | ||
164 | mc_intr_mask_0_pmu_enabled_f()); | ||
165 | } | ||
166 | |||
167 | gk20a_dbg_fn("done"); | ||
168 | } | ||
169 | |||
170 | |||
171 | |||
172 | int pmu_bootstrap(struct nvgpu_pmu *pmu) | ||
173 | { | ||
174 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
175 | struct mm_gk20a *mm = &g->mm; | ||
176 | struct pmu_ucode_desc *desc = pmu->desc; | ||
177 | u64 addr_code, addr_data, addr_load; | ||
178 | u32 i, blocks, addr_args; | ||
179 | |||
180 | gk20a_dbg_fn(""); | ||
181 | |||
182 | gk20a_writel(g, pwr_falcon_itfen_r(), | ||
183 | gk20a_readl(g, pwr_falcon_itfen_r()) | | ||
184 | pwr_falcon_itfen_ctxen_enable_f()); | ||
185 | gk20a_writel(g, pwr_pmu_new_instblk_r(), | ||
186 | pwr_pmu_new_instblk_ptr_f( | ||
187 | nvgpu_inst_block_addr(g, &mm->pmu.inst_block) >> 12) | | ||
188 | pwr_pmu_new_instblk_valid_f(1) | | ||
189 | pwr_pmu_new_instblk_target_sys_coh_f()); | ||
190 | |||
191 | /* TBD: load all other surfaces */ | ||
192 | g->ops.pmu_ver.set_pmu_cmdline_args_trace_size( | ||
193 | pmu, GK20A_PMU_TRACE_BUFSIZE); | ||
194 | g->ops.pmu_ver.set_pmu_cmdline_args_trace_dma_base(pmu); | ||
195 | g->ops.pmu_ver.set_pmu_cmdline_args_trace_dma_idx( | ||
196 | pmu, GK20A_PMU_DMAIDX_VIRT); | ||
197 | |||
198 | g->ops.pmu_ver.set_pmu_cmdline_args_cpu_freq(pmu, | ||
199 | g->ops.clk.get_rate(g, CTRL_CLK_DOMAIN_PWRCLK)); | ||
200 | |||
201 | addr_args = (pwr_falcon_hwcfg_dmem_size_v( | ||
202 | gk20a_readl(g, pwr_falcon_hwcfg_r())) | ||
203 | << GK20A_PMU_DMEM_BLKSIZE2) - | ||
204 | g->ops.pmu_ver.get_pmu_cmdline_args_size(pmu); | ||
205 | |||
206 | nvgpu_flcn_copy_to_dmem(pmu->flcn, addr_args, | ||
207 | (u8 *)(g->ops.pmu_ver.get_pmu_cmdline_args_ptr(pmu)), | ||
208 | g->ops.pmu_ver.get_pmu_cmdline_args_size(pmu), 0); | ||
209 | |||
210 | gk20a_writel(g, pwr_falcon_dmemc_r(0), | ||
211 | pwr_falcon_dmemc_offs_f(0) | | ||
212 | pwr_falcon_dmemc_blk_f(0) | | ||
213 | pwr_falcon_dmemc_aincw_f(1)); | ||
214 | |||
215 | addr_code = u64_lo32((pmu->ucode.gpu_va + | ||
216 | desc->app_start_offset + | ||
217 | desc->app_resident_code_offset) >> 8) ; | ||
218 | addr_data = u64_lo32((pmu->ucode.gpu_va + | ||
219 | desc->app_start_offset + | ||
220 | desc->app_resident_data_offset) >> 8); | ||
221 | addr_load = u64_lo32((pmu->ucode.gpu_va + | ||
222 | desc->bootloader_start_offset) >> 8); | ||
223 | |||
224 | gk20a_writel(g, pwr_falcon_dmemd_r(0), GK20A_PMU_DMAIDX_UCODE); | ||
225 | gk20a_writel(g, pwr_falcon_dmemd_r(0), addr_code); | ||
226 | gk20a_writel(g, pwr_falcon_dmemd_r(0), desc->app_size); | ||
227 | gk20a_writel(g, pwr_falcon_dmemd_r(0), desc->app_resident_code_size); | ||
228 | gk20a_writel(g, pwr_falcon_dmemd_r(0), desc->app_imem_entry); | ||
229 | gk20a_writel(g, pwr_falcon_dmemd_r(0), addr_data); | ||
230 | gk20a_writel(g, pwr_falcon_dmemd_r(0), desc->app_resident_data_size); | ||
231 | gk20a_writel(g, pwr_falcon_dmemd_r(0), addr_code); | ||
232 | gk20a_writel(g, pwr_falcon_dmemd_r(0), 0x1); | ||
233 | gk20a_writel(g, pwr_falcon_dmemd_r(0), addr_args); | ||
234 | |||
235 | g->ops.pmu.write_dmatrfbase(g, | ||
236 | addr_load - (desc->bootloader_imem_offset >> 8)); | ||
237 | |||
238 | blocks = ((desc->bootloader_size + 0xFF) & ~0xFF) >> 8; | ||
239 | |||
240 | for (i = 0; i < blocks; i++) { | ||
241 | gk20a_writel(g, pwr_falcon_dmatrfmoffs_r(), | ||
242 | desc->bootloader_imem_offset + (i << 8)); | ||
243 | gk20a_writel(g, pwr_falcon_dmatrffboffs_r(), | ||
244 | desc->bootloader_imem_offset + (i << 8)); | ||
245 | gk20a_writel(g, pwr_falcon_dmatrfcmd_r(), | ||
246 | pwr_falcon_dmatrfcmd_imem_f(1) | | ||
247 | pwr_falcon_dmatrfcmd_write_f(0) | | ||
248 | pwr_falcon_dmatrfcmd_size_f(6) | | ||
249 | pwr_falcon_dmatrfcmd_ctxdma_f(GK20A_PMU_DMAIDX_UCODE)); | ||
250 | } | ||
251 | |||
252 | nvgpu_flcn_bootstrap(g->pmu.flcn, desc->bootloader_entry_point); | ||
253 | |||
254 | gk20a_writel(g, pwr_falcon_os_r(), desc->app_version); | ||
255 | |||
256 | return 0; | ||
257 | } | ||
258 | |||
259 | void gk20a_pmu_pg_idle_counter_config(struct gk20a *g, u32 pg_engine_id) | ||
260 | { | ||
261 | gk20a_writel(g, pwr_pmu_pg_idlefilth_r(pg_engine_id), | ||
262 | PMU_PG_IDLE_THRESHOLD); | ||
263 | gk20a_writel(g, pwr_pmu_pg_ppuidlefilth_r(pg_engine_id), | ||
264 | PMU_PG_POST_POWERUP_IDLE_THRESHOLD); | ||
265 | } | ||
266 | |||
267 | int gk20a_pmu_mutex_acquire(struct nvgpu_pmu *pmu, u32 id, u32 *token) | ||
268 | { | ||
269 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
270 | struct pmu_mutex *mutex; | ||
271 | u32 data, owner, max_retry; | ||
272 | |||
273 | if (!pmu->initialized) | ||
274 | return -EINVAL; | ||
275 | |||
276 | BUG_ON(!token); | ||
277 | BUG_ON(!PMU_MUTEX_ID_IS_VALID(id)); | ||
278 | BUG_ON(id > pmu->mutex_cnt); | ||
279 | |||
280 | mutex = &pmu->mutex[id]; | ||
281 | |||
282 | owner = pwr_pmu_mutex_value_v( | ||
283 | gk20a_readl(g, pwr_pmu_mutex_r(mutex->index))); | ||
284 | |||
285 | if (*token != PMU_INVALID_MUTEX_OWNER_ID && *token == owner) { | ||
286 | BUG_ON(mutex->ref_cnt == 0); | ||
287 | gk20a_dbg_pmu("already acquired by owner : 0x%08x", *token); | ||
288 | mutex->ref_cnt++; | ||
289 | return 0; | ||
290 | } | ||
291 | |||
292 | max_retry = 40; | ||
293 | do { | ||
294 | data = pwr_pmu_mutex_id_value_v( | ||
295 | gk20a_readl(g, pwr_pmu_mutex_id_r())); | ||
296 | if (data == pwr_pmu_mutex_id_value_init_v() || | ||
297 | data == pwr_pmu_mutex_id_value_not_avail_v()) { | ||
298 | nvgpu_warn(g, | ||
299 | "fail to generate mutex token: val 0x%08x", | ||
300 | owner); | ||
301 | nvgpu_usleep_range(20, 40); | ||
302 | continue; | ||
303 | } | ||
304 | |||
305 | owner = data; | ||
306 | gk20a_writel(g, pwr_pmu_mutex_r(mutex->index), | ||
307 | pwr_pmu_mutex_value_f(owner)); | ||
308 | |||
309 | data = pwr_pmu_mutex_value_v( | ||
310 | gk20a_readl(g, pwr_pmu_mutex_r(mutex->index))); | ||
311 | |||
312 | if (owner == data) { | ||
313 | mutex->ref_cnt = 1; | ||
314 | gk20a_dbg_pmu("mutex acquired: id=%d, token=0x%x", | ||
315 | mutex->index, *token); | ||
316 | *token = owner; | ||
317 | return 0; | ||
318 | } else { | ||
319 | gk20a_dbg_info("fail to acquire mutex idx=0x%08x", | ||
320 | mutex->index); | ||
321 | |||
322 | data = gk20a_readl(g, pwr_pmu_mutex_id_release_r()); | ||
323 | data = set_field(data, | ||
324 | pwr_pmu_mutex_id_release_value_m(), | ||
325 | pwr_pmu_mutex_id_release_value_f(owner)); | ||
326 | gk20a_writel(g, pwr_pmu_mutex_id_release_r(), data); | ||
327 | |||
328 | nvgpu_usleep_range(20, 40); | ||
329 | continue; | ||
330 | } | ||
331 | } while (max_retry-- > 0); | ||
332 | |||
333 | return -EBUSY; | ||
334 | } | ||
335 | |||
336 | int gk20a_pmu_mutex_release(struct nvgpu_pmu *pmu, u32 id, u32 *token) | ||
337 | { | ||
338 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
339 | struct pmu_mutex *mutex; | ||
340 | u32 owner, data; | ||
341 | |||
342 | if (!pmu->initialized) | ||
343 | return -EINVAL; | ||
344 | |||
345 | BUG_ON(!token); | ||
346 | BUG_ON(!PMU_MUTEX_ID_IS_VALID(id)); | ||
347 | BUG_ON(id > pmu->mutex_cnt); | ||
348 | |||
349 | mutex = &pmu->mutex[id]; | ||
350 | |||
351 | owner = pwr_pmu_mutex_value_v( | ||
352 | gk20a_readl(g, pwr_pmu_mutex_r(mutex->index))); | ||
353 | |||
354 | if (*token != owner) { | ||
355 | nvgpu_err(g, "requester 0x%08x NOT match owner 0x%08x", | ||
356 | *token, owner); | ||
357 | return -EINVAL; | ||
358 | } | ||
359 | |||
360 | if (--mutex->ref_cnt > 0) | ||
361 | return -EBUSY; | ||
362 | |||
363 | gk20a_writel(g, pwr_pmu_mutex_r(mutex->index), | ||
364 | pwr_pmu_mutex_value_initial_lock_f()); | ||
365 | |||
366 | data = gk20a_readl(g, pwr_pmu_mutex_id_release_r()); | ||
367 | data = set_field(data, pwr_pmu_mutex_id_release_value_m(), | ||
368 | pwr_pmu_mutex_id_release_value_f(owner)); | ||
369 | gk20a_writel(g, pwr_pmu_mutex_id_release_r(), data); | ||
370 | |||
371 | gk20a_dbg_pmu("mutex released: id=%d, token=0x%x", | ||
372 | mutex->index, *token); | ||
373 | |||
374 | return 0; | ||
375 | } | ||
376 | |||
377 | int gk20a_pmu_queue_head(struct nvgpu_pmu *pmu, struct pmu_queue *queue, | ||
378 | u32 *head, bool set) | ||
379 | { | ||
380 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
381 | u32 queue_head_size = 0; | ||
382 | |||
383 | if (g->ops.pmu.pmu_get_queue_head_size) | ||
384 | queue_head_size = g->ops.pmu.pmu_get_queue_head_size(); | ||
385 | |||
386 | BUG_ON(!head || !queue_head_size); | ||
387 | |||
388 | if (PMU_IS_COMMAND_QUEUE(queue->id)) { | ||
389 | |||
390 | if (queue->index >= queue_head_size) | ||
391 | return -EINVAL; | ||
392 | |||
393 | if (!set) | ||
394 | *head = pwr_pmu_queue_head_address_v( | ||
395 | gk20a_readl(g, | ||
396 | g->ops.pmu.pmu_get_queue_head(queue->index))); | ||
397 | else | ||
398 | gk20a_writel(g, | ||
399 | g->ops.pmu.pmu_get_queue_head(queue->index), | ||
400 | pwr_pmu_queue_head_address_f(*head)); | ||
401 | } else { | ||
402 | if (!set) | ||
403 | *head = pwr_pmu_msgq_head_val_v( | ||
404 | gk20a_readl(g, pwr_pmu_msgq_head_r())); | ||
405 | else | ||
406 | gk20a_writel(g, | ||
407 | pwr_pmu_msgq_head_r(), | ||
408 | pwr_pmu_msgq_head_val_f(*head)); | ||
409 | } | ||
410 | |||
411 | return 0; | ||
412 | } | ||
413 | |||
414 | int gk20a_pmu_queue_tail(struct nvgpu_pmu *pmu, struct pmu_queue *queue, | ||
415 | u32 *tail, bool set) | ||
416 | { | ||
417 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
418 | u32 queue_tail_size = 0; | ||
419 | |||
420 | if (g->ops.pmu.pmu_get_queue_tail_size) | ||
421 | queue_tail_size = g->ops.pmu.pmu_get_queue_tail_size(); | ||
422 | |||
423 | BUG_ON(!tail || !queue_tail_size); | ||
424 | |||
425 | if (PMU_IS_COMMAND_QUEUE(queue->id)) { | ||
426 | |||
427 | if (queue->index >= queue_tail_size) | ||
428 | return -EINVAL; | ||
429 | |||
430 | if (!set) | ||
431 | *tail = pwr_pmu_queue_tail_address_v( | ||
432 | gk20a_readl(g, | ||
433 | g->ops.pmu.pmu_get_queue_tail(queue->index))); | ||
434 | else | ||
435 | gk20a_writel(g, | ||
436 | g->ops.pmu.pmu_get_queue_tail(queue->index), | ||
437 | pwr_pmu_queue_tail_address_f(*tail)); | ||
438 | |||
439 | } else { | ||
440 | if (!set) | ||
441 | *tail = pwr_pmu_msgq_tail_val_v( | ||
442 | gk20a_readl(g, pwr_pmu_msgq_tail_r())); | ||
443 | else | ||
444 | gk20a_writel(g, | ||
445 | pwr_pmu_msgq_tail_r(), | ||
446 | pwr_pmu_msgq_tail_val_f(*tail)); | ||
447 | } | ||
448 | |||
449 | return 0; | ||
450 | } | ||
451 | |||
452 | void gk20a_pmu_msgq_tail(struct nvgpu_pmu *pmu, u32 *tail, bool set) | ||
453 | { | ||
454 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
455 | u32 queue_tail_size = 0; | ||
456 | |||
457 | if (g->ops.pmu.pmu_get_queue_tail_size) | ||
458 | queue_tail_size = g->ops.pmu.pmu_get_queue_tail_size(); | ||
459 | |||
460 | BUG_ON(!tail || !queue_tail_size); | ||
461 | |||
462 | if (!set) | ||
463 | *tail = pwr_pmu_msgq_tail_val_v( | ||
464 | gk20a_readl(g, pwr_pmu_msgq_tail_r())); | ||
465 | else | ||
466 | gk20a_writel(g, | ||
467 | pwr_pmu_msgq_tail_r(), | ||
468 | pwr_pmu_msgq_tail_val_f(*tail)); | ||
469 | } | ||
470 | |||
471 | int gk20a_init_pmu_setup_hw1(struct gk20a *g) | ||
472 | { | ||
473 | struct nvgpu_pmu *pmu = &g->pmu; | ||
474 | int err = 0; | ||
475 | |||
476 | gk20a_dbg_fn(""); | ||
477 | |||
478 | nvgpu_mutex_acquire(&pmu->isr_mutex); | ||
479 | nvgpu_flcn_reset(pmu->flcn); | ||
480 | pmu->isr_enabled = true; | ||
481 | nvgpu_mutex_release(&pmu->isr_mutex); | ||
482 | |||
483 | /* setup apertures - virtual */ | ||
484 | gk20a_writel(g, pwr_fbif_transcfg_r(GK20A_PMU_DMAIDX_UCODE), | ||
485 | pwr_fbif_transcfg_mem_type_virtual_f()); | ||
486 | gk20a_writel(g, pwr_fbif_transcfg_r(GK20A_PMU_DMAIDX_VIRT), | ||
487 | pwr_fbif_transcfg_mem_type_virtual_f()); | ||
488 | /* setup apertures - physical */ | ||
489 | gk20a_writel(g, pwr_fbif_transcfg_r(GK20A_PMU_DMAIDX_PHYS_VID), | ||
490 | pwr_fbif_transcfg_mem_type_physical_f() | | ||
491 | pwr_fbif_transcfg_target_local_fb_f()); | ||
492 | gk20a_writel(g, pwr_fbif_transcfg_r(GK20A_PMU_DMAIDX_PHYS_SYS_COH), | ||
493 | pwr_fbif_transcfg_mem_type_physical_f() | | ||
494 | pwr_fbif_transcfg_target_coherent_sysmem_f()); | ||
495 | gk20a_writel(g, pwr_fbif_transcfg_r(GK20A_PMU_DMAIDX_PHYS_SYS_NCOH), | ||
496 | pwr_fbif_transcfg_mem_type_physical_f() | | ||
497 | pwr_fbif_transcfg_target_noncoherent_sysmem_f()); | ||
498 | |||
499 | err = g->ops.pmu.pmu_nsbootstrap(pmu); | ||
500 | |||
501 | return err; | ||
502 | |||
503 | } | ||
504 | |||
505 | void gk20a_write_dmatrfbase(struct gk20a *g, u32 addr) | ||
506 | { | ||
507 | gk20a_writel(g, pwr_falcon_dmatrfbase_r(), addr); | ||
508 | } | ||
509 | |||
510 | bool gk20a_pmu_is_engine_in_reset(struct gk20a *g) | ||
511 | { | ||
512 | u32 pmc_enable; | ||
513 | bool status = false; | ||
514 | |||
515 | pmc_enable = gk20a_readl(g, mc_enable_r()); | ||
516 | if (mc_enable_pwr_v(pmc_enable) == | ||
517 | mc_enable_pwr_disabled_v()) | ||
518 | status = true; | ||
519 | |||
520 | return status; | ||
521 | } | ||
522 | |||
523 | int gk20a_pmu_engine_reset(struct gk20a *g, bool do_reset) | ||
524 | { | ||
525 | if (do_reset) | ||
526 | g->ops.mc.enable(g, mc_enable_pwr_enabled_f()); | ||
527 | else | ||
528 | g->ops.mc.disable(g, mc_enable_pwr_enabled_f()); | ||
529 | |||
530 | return 0; | ||
531 | } | ||
532 | |||
533 | bool gk20a_is_pmu_supported(struct gk20a *g) | ||
534 | { | ||
535 | return true; | ||
536 | } | ||
537 | |||
538 | u32 gk20a_pmu_pg_engines_list(struct gk20a *g) | ||
539 | { | ||
540 | return BIT(PMU_PG_ELPG_ENGINE_ID_GRAPHICS); | ||
541 | } | ||
542 | |||
543 | u32 gk20a_pmu_pg_feature_list(struct gk20a *g, u32 pg_engine_id) | ||
544 | { | ||
545 | if (pg_engine_id == PMU_PG_ELPG_ENGINE_ID_GRAPHICS) | ||
546 | return PMU_PG_FEATURE_GR_POWER_GATING_ENABLED; | ||
547 | |||
548 | return 0; | ||
549 | } | ||
550 | |||
551 | static void pmu_handle_zbc_msg(struct gk20a *g, struct pmu_msg *msg, | ||
552 | void *param, u32 handle, u32 status) | ||
553 | { | ||
554 | struct nvgpu_pmu *pmu = param; | ||
555 | gk20a_dbg_pmu("reply ZBC_TABLE_UPDATE"); | ||
556 | pmu->zbc_save_done = 1; | ||
557 | } | ||
558 | |||
559 | void gk20a_pmu_save_zbc(struct gk20a *g, u32 entries) | ||
560 | { | ||
561 | struct nvgpu_pmu *pmu = &g->pmu; | ||
562 | struct pmu_cmd cmd; | ||
563 | u32 seq; | ||
564 | |||
565 | if (!pmu->pmu_ready || !entries || !pmu->zbc_ready) | ||
566 | return; | ||
567 | |||
568 | memset(&cmd, 0, sizeof(struct pmu_cmd)); | ||
569 | cmd.hdr.unit_id = PMU_UNIT_PG; | ||
570 | cmd.hdr.size = PMU_CMD_HDR_SIZE + sizeof(struct pmu_zbc_cmd); | ||
571 | cmd.cmd.zbc.cmd_type = g->pmu_ver_cmd_id_zbc_table_update; | ||
572 | cmd.cmd.zbc.entry_mask = ZBC_MASK(entries); | ||
573 | |||
574 | pmu->zbc_save_done = 0; | ||
575 | |||
576 | gk20a_dbg_pmu("cmd post ZBC_TABLE_UPDATE"); | ||
577 | nvgpu_pmu_cmd_post(g, &cmd, NULL, NULL, PMU_COMMAND_QUEUE_HPQ, | ||
578 | pmu_handle_zbc_msg, pmu, &seq, ~0); | ||
579 | pmu_wait_message_cond(pmu, gk20a_get_gr_idle_timeout(g), | ||
580 | &pmu->zbc_save_done, 1); | ||
581 | if (!pmu->zbc_save_done) | ||
582 | nvgpu_err(g, "ZBC save timeout"); | ||
583 | } | ||
584 | |||
585 | int nvgpu_pmu_handle_therm_event(struct nvgpu_pmu *pmu, | ||
586 | struct nv_pmu_therm_msg *msg) | ||
587 | { | ||
588 | gk20a_dbg_fn(""); | ||
589 | |||
590 | switch (msg->msg_type) { | ||
591 | case NV_PMU_THERM_MSG_ID_EVENT_HW_SLOWDOWN_NOTIFICATION: | ||
592 | #ifdef CONFIG_ARCH_TEGRA_18x_SOC | ||
593 | if (msg->hw_slct_msg.mask == BIT(NV_PMU_THERM_EVENT_THERMAL_1)) | ||
594 | nvgpu_clk_arb_schedule_alarm(gk20a_from_pmu(pmu), | ||
595 | (0x1UL << NVGPU_GPU_EVENT_ALARM_THERMAL_ABOVE_THRESHOLD)); | ||
596 | else | ||
597 | #endif | ||
598 | gk20a_dbg_pmu("Unwanted/Unregistered thermal event received %d", | ||
599 | msg->hw_slct_msg.mask); | ||
600 | break; | ||
601 | default: | ||
602 | gk20a_dbg_pmu("unkown therm event received %d", msg->msg_type); | ||
603 | break; | ||
604 | } | ||
605 | |||
606 | return 0; | ||
607 | } | ||
608 | |||
609 | void gk20a_pmu_dump_elpg_stats(struct nvgpu_pmu *pmu) | ||
610 | { | ||
611 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
612 | |||
613 | gk20a_dbg_pmu("pwr_pmu_idle_mask_supp_r(3): 0x%08x", | ||
614 | gk20a_readl(g, pwr_pmu_idle_mask_supp_r(3))); | ||
615 | gk20a_dbg_pmu("pwr_pmu_idle_mask_1_supp_r(3): 0x%08x", | ||
616 | gk20a_readl(g, pwr_pmu_idle_mask_1_supp_r(3))); | ||
617 | gk20a_dbg_pmu("pwr_pmu_idle_ctrl_supp_r(3): 0x%08x", | ||
618 | gk20a_readl(g, pwr_pmu_idle_ctrl_supp_r(3))); | ||
619 | gk20a_dbg_pmu("pwr_pmu_pg_idle_cnt_r(0): 0x%08x", | ||
620 | gk20a_readl(g, pwr_pmu_pg_idle_cnt_r(0))); | ||
621 | gk20a_dbg_pmu("pwr_pmu_pg_intren_r(0): 0x%08x", | ||
622 | gk20a_readl(g, pwr_pmu_pg_intren_r(0))); | ||
623 | |||
624 | gk20a_dbg_pmu("pwr_pmu_idle_count_r(3): 0x%08x", | ||
625 | gk20a_readl(g, pwr_pmu_idle_count_r(3))); | ||
626 | gk20a_dbg_pmu("pwr_pmu_idle_count_r(4): 0x%08x", | ||
627 | gk20a_readl(g, pwr_pmu_idle_count_r(4))); | ||
628 | gk20a_dbg_pmu("pwr_pmu_idle_count_r(7): 0x%08x", | ||
629 | gk20a_readl(g, pwr_pmu_idle_count_r(7))); | ||
630 | } | ||
631 | |||
632 | void gk20a_pmu_dump_falcon_stats(struct nvgpu_pmu *pmu) | ||
633 | { | ||
634 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
635 | unsigned int i; | ||
636 | |||
637 | for (i = 0; i < pwr_pmu_mailbox__size_1_v(); i++) | ||
638 | nvgpu_err(g, "pwr_pmu_mailbox_r(%d) : 0x%x", | ||
639 | i, gk20a_readl(g, pwr_pmu_mailbox_r(i))); | ||
640 | |||
641 | for (i = 0; i < pwr_pmu_debug__size_1_v(); i++) | ||
642 | nvgpu_err(g, "pwr_pmu_debug_r(%d) : 0x%x", | ||
643 | i, gk20a_readl(g, pwr_pmu_debug_r(i))); | ||
644 | |||
645 | i = gk20a_readl(g, pwr_pmu_bar0_error_status_r()); | ||
646 | nvgpu_err(g, "pwr_pmu_bar0_error_status_r : 0x%x", i); | ||
647 | if (i != 0) { | ||
648 | nvgpu_err(g, "pwr_pmu_bar0_addr_r : 0x%x", | ||
649 | gk20a_readl(g, pwr_pmu_bar0_addr_r())); | ||
650 | nvgpu_err(g, "pwr_pmu_bar0_data_r : 0x%x", | ||
651 | gk20a_readl(g, pwr_pmu_bar0_data_r())); | ||
652 | nvgpu_err(g, "pwr_pmu_bar0_timeout_r : 0x%x", | ||
653 | gk20a_readl(g, pwr_pmu_bar0_timeout_r())); | ||
654 | nvgpu_err(g, "pwr_pmu_bar0_ctl_r : 0x%x", | ||
655 | gk20a_readl(g, pwr_pmu_bar0_ctl_r())); | ||
656 | } | ||
657 | |||
658 | i = gk20a_readl(g, pwr_pmu_bar0_fecs_error_r()); | ||
659 | nvgpu_err(g, "pwr_pmu_bar0_fecs_error_r : 0x%x", i); | ||
660 | |||
661 | i = gk20a_readl(g, pwr_falcon_exterrstat_r()); | ||
662 | nvgpu_err(g, "pwr_falcon_exterrstat_r : 0x%x", i); | ||
663 | if (pwr_falcon_exterrstat_valid_v(i) == | ||
664 | pwr_falcon_exterrstat_valid_true_v()) { | ||
665 | nvgpu_err(g, "pwr_falcon_exterraddr_r : 0x%x", | ||
666 | gk20a_readl(g, pwr_falcon_exterraddr_r())); | ||
667 | nvgpu_err(g, "pmc_enable : 0x%x", | ||
668 | gk20a_readl(g, mc_enable_r())); | ||
669 | } | ||
670 | |||
671 | /* Print PMU F/W debug prints */ | ||
672 | print_pmu_trace(pmu); | ||
673 | } | ||
674 | |||
675 | bool gk20a_pmu_is_interrupted(struct nvgpu_pmu *pmu) | ||
676 | { | ||
677 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
678 | u32 servicedpmuint; | ||
679 | |||
680 | servicedpmuint = pwr_falcon_irqstat_halt_true_f() | | ||
681 | pwr_falcon_irqstat_exterr_true_f() | | ||
682 | pwr_falcon_irqstat_swgen0_true_f(); | ||
683 | |||
684 | if (gk20a_readl(g, pwr_falcon_irqstat_r()) & servicedpmuint) | ||
685 | return true; | ||
686 | |||
687 | return false; | ||
688 | } | ||
689 | |||
690 | void gk20a_pmu_isr(struct gk20a *g) | ||
691 | { | ||
692 | struct nvgpu_pmu *pmu = &g->pmu; | ||
693 | struct pmu_queue *queue; | ||
694 | u32 intr, mask; | ||
695 | bool recheck = false; | ||
696 | |||
697 | gk20a_dbg_fn(""); | ||
698 | |||
699 | nvgpu_mutex_acquire(&pmu->isr_mutex); | ||
700 | if (!pmu->isr_enabled) { | ||
701 | nvgpu_mutex_release(&pmu->isr_mutex); | ||
702 | return; | ||
703 | } | ||
704 | |||
705 | mask = gk20a_readl(g, pwr_falcon_irqmask_r()) & | ||
706 | gk20a_readl(g, pwr_falcon_irqdest_r()); | ||
707 | |||
708 | intr = gk20a_readl(g, pwr_falcon_irqstat_r()); | ||
709 | |||
710 | gk20a_dbg_pmu("received falcon interrupt: 0x%08x", intr); | ||
711 | |||
712 | intr = gk20a_readl(g, pwr_falcon_irqstat_r()) & mask; | ||
713 | if (!intr || pmu->pmu_state == PMU_STATE_OFF) { | ||
714 | gk20a_writel(g, pwr_falcon_irqsclr_r(), intr); | ||
715 | nvgpu_mutex_release(&pmu->isr_mutex); | ||
716 | return; | ||
717 | } | ||
718 | |||
719 | if (intr & pwr_falcon_irqstat_halt_true_f()) { | ||
720 | nvgpu_err(g, "pmu halt intr not implemented"); | ||
721 | nvgpu_pmu_dump_falcon_stats(pmu); | ||
722 | if (gk20a_readl(g, pwr_pmu_mailbox_r | ||
723 | (PMU_MODE_MISMATCH_STATUS_MAILBOX_R)) == | ||
724 | PMU_MODE_MISMATCH_STATUS_VAL) | ||
725 | if (g->ops.pmu.dump_secure_fuses) | ||
726 | g->ops.pmu.dump_secure_fuses(g); | ||
727 | } | ||
728 | if (intr & pwr_falcon_irqstat_exterr_true_f()) { | ||
729 | nvgpu_err(g, | ||
730 | "pmu exterr intr not implemented. Clearing interrupt."); | ||
731 | nvgpu_pmu_dump_falcon_stats(pmu); | ||
732 | |||
733 | gk20a_writel(g, pwr_falcon_exterrstat_r(), | ||
734 | gk20a_readl(g, pwr_falcon_exterrstat_r()) & | ||
735 | ~pwr_falcon_exterrstat_valid_m()); | ||
736 | } | ||
737 | if (intr & pwr_falcon_irqstat_swgen0_true_f()) { | ||
738 | nvgpu_pmu_process_message(pmu); | ||
739 | recheck = true; | ||
740 | } | ||
741 | |||
742 | gk20a_writel(g, pwr_falcon_irqsclr_r(), intr); | ||
743 | |||
744 | if (recheck) { | ||
745 | queue = &pmu->queue[PMU_MESSAGE_QUEUE]; | ||
746 | if (!nvgpu_pmu_queue_is_empty(pmu, queue)) | ||
747 | gk20a_writel(g, pwr_falcon_irqsset_r(), | ||
748 | pwr_falcon_irqsset_swgen0_set_f()); | ||
749 | } | ||
750 | |||
751 | nvgpu_mutex_release(&pmu->isr_mutex); | ||
752 | } | ||
753 | |||
754 | void gk20a_pmu_init_perfmon_counter(struct gk20a *g) | ||
755 | { | ||
756 | u32 data; | ||
757 | |||
758 | /* use counter #3 for GR && CE2 busy cycles */ | ||
759 | gk20a_writel(g, pwr_pmu_idle_mask_r(3), | ||
760 | pwr_pmu_idle_mask_gr_enabled_f() | | ||
761 | pwr_pmu_idle_mask_ce_2_enabled_f()); | ||
762 | |||
763 | /* disable idle filtering for counters 3 and 6 */ | ||
764 | data = gk20a_readl(g, pwr_pmu_idle_ctrl_r(3)); | ||
765 | data = set_field(data, pwr_pmu_idle_ctrl_value_m() | | ||
766 | pwr_pmu_idle_ctrl_filter_m(), | ||
767 | pwr_pmu_idle_ctrl_value_busy_f() | | ||
768 | pwr_pmu_idle_ctrl_filter_disabled_f()); | ||
769 | gk20a_writel(g, pwr_pmu_idle_ctrl_r(3), data); | ||
770 | |||
771 | /* use counter #6 for total cycles */ | ||
772 | data = gk20a_readl(g, pwr_pmu_idle_ctrl_r(6)); | ||
773 | data = set_field(data, pwr_pmu_idle_ctrl_value_m() | | ||
774 | pwr_pmu_idle_ctrl_filter_m(), | ||
775 | pwr_pmu_idle_ctrl_value_always_f() | | ||
776 | pwr_pmu_idle_ctrl_filter_disabled_f()); | ||
777 | gk20a_writel(g, pwr_pmu_idle_ctrl_r(6), data); | ||
778 | |||
779 | /* | ||
780 | * We don't want to disturb counters #3 and #6, which are used by | ||
781 | * perfmon, so we add wiring also to counters #1 and #2 for | ||
782 | * exposing raw counter readings. | ||
783 | */ | ||
784 | gk20a_writel(g, pwr_pmu_idle_mask_r(1), | ||
785 | pwr_pmu_idle_mask_gr_enabled_f() | | ||
786 | pwr_pmu_idle_mask_ce_2_enabled_f()); | ||
787 | |||
788 | data = gk20a_readl(g, pwr_pmu_idle_ctrl_r(1)); | ||
789 | data = set_field(data, pwr_pmu_idle_ctrl_value_m() | | ||
790 | pwr_pmu_idle_ctrl_filter_m(), | ||
791 | pwr_pmu_idle_ctrl_value_busy_f() | | ||
792 | pwr_pmu_idle_ctrl_filter_disabled_f()); | ||
793 | gk20a_writel(g, pwr_pmu_idle_ctrl_r(1), data); | ||
794 | |||
795 | data = gk20a_readl(g, pwr_pmu_idle_ctrl_r(2)); | ||
796 | data = set_field(data, pwr_pmu_idle_ctrl_value_m() | | ||
797 | pwr_pmu_idle_ctrl_filter_m(), | ||
798 | pwr_pmu_idle_ctrl_value_always_f() | | ||
799 | pwr_pmu_idle_ctrl_filter_disabled_f()); | ||
800 | gk20a_writel(g, pwr_pmu_idle_ctrl_r(2), data); | ||
801 | } | ||
802 | |||
803 | u32 gk20a_pmu_read_idle_counter(struct gk20a *g, u32 counter_id) | ||
804 | { | ||
805 | return pwr_pmu_idle_count_value_v( | ||
806 | gk20a_readl(g, pwr_pmu_idle_count_r(counter_id))); | ||
807 | } | ||
808 | |||
809 | void gk20a_pmu_reset_idle_counter(struct gk20a *g, u32 counter_id) | ||
810 | { | ||
811 | gk20a_writel(g, pwr_pmu_idle_count_r(counter_id), | ||
812 | pwr_pmu_idle_count_reset_f(1)); | ||
813 | } | ||
814 | |||
815 | void gk20a_pmu_elpg_statistics(struct gk20a *g, u32 pg_engine_id, | ||
816 | struct pmu_pg_stats_data *pg_stat_data) | ||
817 | { | ||
818 | struct nvgpu_pmu *pmu = &g->pmu; | ||
819 | struct pmu_pg_stats stats; | ||
820 | |||
821 | nvgpu_flcn_copy_from_dmem(pmu->flcn, | ||
822 | pmu->stat_dmem_offset[pg_engine_id], | ||
823 | (u8 *)&stats, sizeof(struct pmu_pg_stats), 0); | ||
824 | |||
825 | pg_stat_data->ingating_time = stats.pg_ingating_time_us; | ||
826 | pg_stat_data->ungating_time = stats.pg_ungating_time_us; | ||
827 | pg_stat_data->gating_cnt = stats.pg_gating_cnt; | ||
828 | pg_stat_data->avg_entry_latency_us = stats.pg_avg_entry_time_us; | ||
829 | pg_stat_data->avg_exit_latency_us = stats.pg_avg_exit_time_us; | ||
830 | } | ||