summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/nvgpu/gk20a/gk20a.c
blob: 2d7266ce665a612207c45f0f280f1402cc6b3a80 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
/*
 * GK20A Graphics
 *
 * Copyright (c) 2011-2017, NVIDIA CORPORATION.  All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

#include <linux/reboot.h>

#include <nvgpu/nvgpu_common.h>
#include <nvgpu/kmem.h>
#include <nvgpu/allocator.h>
#include <nvgpu/timers.h>
#include <nvgpu/soc.h>
#include <nvgpu/enabled.h>
#include <nvgpu/pmu.h>
#include <nvgpu/gmmu.h>
#include <nvgpu/ltc.h>
#include <nvgpu/vidmem.h>
#include <nvgpu/mm.h>
#include <nvgpu/ctxsw_trace.h>

#include <trace/events/gk20a.h>

#include "gk20a.h"
#include "channel_sync_gk20a.h"

#include "dbg_gpu_gk20a.h"
#include "mc_gk20a.h"
#include "hal.h"
#include "vgpu/vgpu.h"
#include "bus_gk20a.h"
#ifdef CONFIG_ARCH_TEGRA_18x_SOC
#include "pstate/pstate.h"
#endif

#ifdef CONFIG_TEGRA_19x_GPU
#include "nvgpu_gpuid_t19x.h"
#endif

void __nvgpu_check_gpu_state(struct gk20a *g)
{
	u32 boot_0 = g->ops.mc.boot_0(g, NULL, NULL, NULL);

	if (boot_0 == 0xffffffff) {
		pr_err("nvgpu: GPU has disappeared from bus!!\n");
		pr_err("nvgpu: Rebooting system!!\n");
		kernel_restart(NULL);
	}
}

void __gk20a_warn_on_no_regs(void)
{
	WARN_ONCE(1, "Attempted access to GPU regs after unmapping!");
}

static int gk20a_detect_chip(struct gk20a *g)
{
	struct nvgpu_gpu_params *p = &g->params;
	u32 val;

	if (p->gpu_arch)
		return 0;

	val = gk20a_mc_boot_0(g, &p->gpu_arch, &p->gpu_impl, &p->gpu_rev);

	gk20a_dbg_info("arch: %x, impl: %x, rev: %x\n",
			g->params.gpu_arch,
			g->params.gpu_impl,
			g->params.gpu_rev);

	return gpu_init_hal(g);
}

int gk20a_prepare_poweroff(struct gk20a *g)
{
	int ret = 0;

	gk20a_dbg_fn("");

	ret = gk20a_channel_suspend(g);
	if (ret)
		return ret;

	/* disable elpg before gr or fifo suspend */
	if (g->ops.pmu.is_pmu_supported(g))
		ret |= nvgpu_pmu_destroy(g);

	ret |= gk20a_gr_suspend(g);
	ret |= nvgpu_mm_suspend(g);
	ret |= gk20a_fifo_suspend(g);

	gk20a_ce_suspend(g);

	/* Disable GPCPLL */
	if (g->ops.clk.suspend_clk_support)
		ret |= g->ops.clk.suspend_clk_support(g);

#ifdef CONFIG_ARCH_TEGRA_18x_SOC
	if (nvgpu_is_enabled(g, NVGPU_PMU_PSTATE))
		gk20a_deinit_pstate_support(g);
#endif
	g->power_on = false;

	return ret;
}

int gk20a_finalize_poweron(struct gk20a *g)
{
	int err;
#if defined(CONFIG_TEGRA_GK20A_NVHOST) && defined(CONFIG_TEGRA_19x_GPU)
	u32 nr_pages;
#endif

	gk20a_dbg_fn("");

	if (g->power_on)
		return 0;

	g->power_on = true;

	err = gk20a_detect_chip(g);
	if (err)
		goto done;

	/*
	 * Before probing the GPU make sure the GPU's state is cleared. This is
	 * relevant for rebind operations.
	 */
	if (g->ops.xve.reset_gpu && !g->gpu_reset_done) {
		g->ops.xve.reset_gpu(g);
		g->gpu_reset_done = true;
	}

	/*
	 * Do this early so any early VMs that get made are capable of mapping
	 * buffers.
	 */
	err = nvgpu_pd_cache_init(g);
	if (err)
		return err;

	/* init interface layer support for PMU falcon */
	nvgpu_flcn_sw_init(g, FALCON_ID_PMU);
	nvgpu_flcn_sw_init(g, FALCON_ID_SEC2);
	nvgpu_flcn_sw_init(g, FALCON_ID_NVDEC);

	if (g->ops.bios.init)
		err = g->ops.bios.init(g);
	if (err)
		goto done;

	g->ops.bus.init_hw(g);

	if (g->ops.clk.disable_slowboot)
		g->ops.clk.disable_slowboot(g);

	gk20a_enable_priv_ring(g);

	/* TBD: move this after graphics init in which blcg/slcg is enabled.
	   This function removes SlowdownOnBoot which applies 32x divider
	   on gpcpll bypass path. The purpose of slowdown is to save power
	   during boot but it also significantly slows down gk20a init on
	   simulation and emulation. We should remove SOB after graphics power
	   saving features (blcg/slcg) are enabled. For now, do it here. */
	if (g->ops.clk.init_clk_support) {
		err = g->ops.clk.init_clk_support(g);
		if (err) {
			nvgpu_err(g, "failed to init gk20a clk");
			goto done;
		}
	}

	if (g->ops.fb.mem_unlock) {
		err = g->ops.fb.mem_unlock(g);
		if (err) {
			nvgpu_err(g, "failed to unlock memory");
			goto done;
		}
	}

	err = g->ops.fifo.reset_enable_hw(g);

	if (err) {
		nvgpu_err(g, "failed to reset gk20a fifo");
		goto done;
	}

	err = nvgpu_init_ltc_support(g);
	if (err) {
		nvgpu_err(g, "failed to init ltc");
		goto done;
	}

	err = nvgpu_init_mm_support(g);
	if (err) {
		nvgpu_err(g, "failed to init gk20a mm");
		goto done;
	}

	err = gk20a_init_fifo_support(g);
	if (err) {
		nvgpu_err(g, "failed to init gk20a fifo");
		goto done;
	}

	if (g->ops.therm.elcg_init_idle_filters)
		g->ops.therm.elcg_init_idle_filters(g);

	g->ops.mc.intr_enable(g);

	err = gk20a_enable_gr_hw(g);
	if (err) {
		nvgpu_err(g, "failed to enable gr");
		goto done;
	}

	if (g->ops.pmu.is_pmu_supported(g)) {
		if (g->ops.pmu.prepare_ucode)
			err = g->ops.pmu.prepare_ucode(g);
		if (err) {
			nvgpu_err(g, "failed to init pmu ucode");
			goto done;
		}
	}

#ifdef CONFIG_ARCH_TEGRA_18x_SOC
	if (nvgpu_is_enabled(g, NVGPU_PMU_PSTATE)) {
		err = gk20a_init_pstate_support(g);
		if (err) {
			nvgpu_err(g, "failed to init pstates");
			goto done;
		}
	}
#endif

	if (g->ops.pmu.is_pmu_supported(g)) {
		err = nvgpu_init_pmu_support(g);
		if (err) {
			nvgpu_err(g, "failed to init gk20a pmu");
			goto done;
		}
	}

	err = gk20a_init_gr_support(g);
	if (err) {
		nvgpu_err(g, "failed to init gk20a gr");
		goto done;
	}

#ifdef CONFIG_ARCH_TEGRA_18x_SOC
	if (nvgpu_is_enabled(g, NVGPU_PMU_PSTATE)) {
		err = gk20a_init_pstate_pmu_support(g);
		if (err) {
			nvgpu_err(g, "failed to init pstates");
			goto done;
		}
	}

	err = nvgpu_clk_arb_init_arbiter(g);
	if (err) {
		nvgpu_err(g, "failed to init clk arb");
		goto done;
	}
#endif

	err = gk20a_init_therm_support(g);
	if (err) {
		nvgpu_err(g, "failed to init gk20a therm");
		goto done;
	}

	err = g->ops.chip_init_gpu_characteristics(g);
	if (err) {
		nvgpu_err(g, "failed to init gk20a gpu characteristics");
		goto done;
	}

#ifdef CONFIG_GK20A_CTXSW_TRACE
	err = gk20a_ctxsw_trace_init(g);
	if (err)
		nvgpu_warn(g, "could not initialize ctxsw tracing");
#endif

	/* Restore the debug setting */
	g->ops.fb.set_debug_mode(g, g->mmu_debug_ctrl);

	gk20a_channel_resume(g);

	gk20a_init_ce_support(g);

	nvgpu_init_mm_ce_context(g);

	if (g->ops.xve.available_speeds) {
		u32 speed;

		if (!nvgpu_is_enabled(g, NVGPU_SUPPORT_ASPM) && g->ops.xve.disable_aspm)
			g->ops.xve.disable_aspm(g);

		g->ops.xve.available_speeds(g, &speed);

		/* Set to max speed */
		speed = 1 << (fls(speed) - 1);
		err = g->ops.xve.set_speed(g, speed);
		if (err) {
			nvgpu_err(g, "Failed to set PCIe bus speed!");
			goto done;
		}
	}

	nvgpu_vidmem_thread_unpause(&g->mm);

#if defined(CONFIG_TEGRA_GK20A_NVHOST) && defined(CONFIG_TEGRA_19x_GPU)
	if (gk20a_platform_has_syncpoints(g) && g->syncpt_unit_size) {
		if (!nvgpu_mem_is_valid(&g->syncpt_mem)) {
			nr_pages = DIV_ROUND_UP(g->syncpt_unit_size, PAGE_SIZE);
			__nvgpu_mem_create_from_phys(g, &g->syncpt_mem,
					g->syncpt_unit_base, nr_pages);
		}
	}
#endif

done:
	if (err)
		g->power_on = false;

	return err;
}

/*
 * Check if the device can go busy. Basically if the driver is currently
 * in the process of dying then do not let new places make the driver busy.
 */
int gk20a_can_busy(struct gk20a *g)
{
	if (nvgpu_is_enabled(g, NVGPU_DRIVER_IS_DYING))
		return 0;
	return 1;
}

int gk20a_wait_for_idle(struct gk20a *g)
{
	int wait_length = 150; /* 3 second overall max wait. */
	int target_usage_count = 0;

	if (!g)
		return -ENODEV;

	if (g->user_railgate_disabled)
		target_usage_count = 1;

	while ((nvgpu_atomic_read(&g->usage_count) != target_usage_count)
			&& (wait_length-- >= 0))
		nvgpu_msleep(20);

	if (wait_length < 0) {
		pr_warn("%s: Timed out waiting for idle (%d)!\n",
			__func__, nvgpu_atomic_read(&g->usage_count));
		return -ETIMEDOUT;
	}

	return 0;
}

int gk20a_init_gpu_characteristics(struct gk20a *g)
{
	struct nvgpu_gpu_characteristics *gpu = &g->gpu_characteristics;

	gpu->L2_cache_size = g->ops.ltc.determine_L2_size_bytes(g);
	gpu->on_board_video_memory_size = 0; /* integrated GPU */

	gpu->num_gpc = g->gr.gpc_count;
	gpu->max_gpc_count = g->gr.max_gpc_count;

	gpu->num_tpc_per_gpc = g->gr.max_tpc_per_gpc_count;

	gpu->bus_type = NVGPU_GPU_BUS_TYPE_AXI; /* always AXI for now */

	gpu->compression_page_size = g->ops.fb.compression_page_size(g);

	__nvgpu_set_enabled(g, NVGPU_SUPPORT_PARTIAL_MAPPINGS, true);
	__nvgpu_set_enabled(g, NVGPU_SUPPORT_MAP_DIRECT_KIND_CTRL, true);

	if (IS_ENABLED(CONFIG_SYNC))
		__nvgpu_set_enabled(g, NVGPU_SUPPORT_SYNC_FENCE_FDS, true);

	if (g->ops.mm.support_sparse && g->ops.mm.support_sparse(g))
		__nvgpu_set_enabled(g, NVGPU_SUPPORT_SPARSE_ALLOCS, true);

	if (gk20a_platform_has_syncpoints(g))
		__nvgpu_set_enabled(g, NVGPU_HAS_SYNCPOINTS, true);

	/*
	 * Fast submits are supported as long as the user doesn't request
	 * anything that depends on job tracking. (Here, fast means strictly no
	 * metadata, just the gpfifo contents are copied and gp_put updated).
	 */
	__nvgpu_set_enabled(g,
			NVGPU_SUPPORT_DETERMINISTIC_SUBMIT_NO_JOBTRACKING,
			true);

	/*
	 * Sync framework requires deferred job cleanup, wrapping syncs in FDs,
	 * and other heavy stuff, which prevents deterministic submits. This is
	 * supported otherwise, provided that the user doesn't request anything
	 * that depends on deferred cleanup.
	 */
	if (!gk20a_channel_sync_needs_sync_framework(g))
		__nvgpu_set_enabled(g,
				NVGPU_SUPPORT_DETERMINISTIC_SUBMIT_FULL,
				true);

	__nvgpu_set_enabled(g, NVGPU_SUPPORT_DETERMINISTIC_OPTS, true);

	__nvgpu_set_enabled(g, NVGPU_SUPPORT_USERSPACE_MANAGED_AS, true);
	__nvgpu_set_enabled(g, NVGPU_SUPPORT_TSG, true);

	if (g->ops.clk_arb.get_arbiter_clk_domains)
		__nvgpu_set_enabled(g, NVGPU_SUPPORT_CLOCK_CONTROLS, true);

	gpu->gpc_mask = (1 << g->gr.gpc_count)-1;

	g->ops.gr.detect_sm_arch(g);

	if (g->ops.gr.init_cyclestats)
		g->ops.gr.init_cyclestats(g);

	gpu->gpu_ioctl_nr_last = NVGPU_GPU_IOCTL_LAST;
	gpu->tsg_ioctl_nr_last = NVGPU_TSG_IOCTL_LAST;
	gpu->dbg_gpu_ioctl_nr_last = NVGPU_DBG_GPU_IOCTL_LAST;
	gpu->ioctl_channel_nr_last = NVGPU_IOCTL_CHANNEL_LAST;
	gpu->as_ioctl_nr_last = NVGPU_AS_IOCTL_LAST;
	gpu->event_ioctl_nr_last = NVGPU_EVENT_IOCTL_LAST;
	gpu->gpu_va_bit_count = 40;

	strlcpy(gpu->chipname, g->name, sizeof(gpu->chipname));
	gpu->max_fbps_count = g->ops.gr.get_max_fbps_count(g);
	gpu->fbp_en_mask = g->ops.gr.get_fbp_en_mask(g);
	gpu->max_ltc_per_fbp =  g->ops.gr.get_max_ltc_per_fbp(g);
	gpu->max_lts_per_ltc = g->ops.gr.get_max_lts_per_ltc(g);
	g->ops.gr.get_rop_l2_en_mask(g);
	gpu->gr_compbit_store_base_hw = g->gr.compbit_store.base_hw;
	gpu->gr_gobs_per_comptagline_per_slice =
		g->gr.gobs_per_comptagline_per_slice;
	gpu->num_ltc = g->ltc_count;
	gpu->lts_per_ltc = g->gr.slices_per_ltc;
	gpu->cbc_cache_line_size = g->gr.cacheline_size;
	gpu->cbc_comptags_per_line = g->gr.comptags_per_cacheline;

	gpu->map_buffer_batch_limit = 256;

	if (g->ops.clk.get_maxrate)
		gpu->max_freq = g->ops.clk.get_maxrate(&g->clk);

	g->ops.gr.get_preemption_mode_flags(g, &g->gr.preemption_mode_rec);
	gpu->graphics_preemption_mode_flags =
		g->gr.preemption_mode_rec.graphics_preemption_mode_flags;
	gpu->compute_preemption_mode_flags =
		g->gr.preemption_mode_rec.compute_preemption_mode_flags;
	gpu->default_graphics_preempt_mode =
		g->gr.preemption_mode_rec.default_graphics_preempt_mode;
	gpu->default_compute_preempt_mode =
		g->gr.preemption_mode_rec.default_compute_preempt_mode;

	gpu->local_video_memory_size = g->mm.vidmem.size;

	gpu->pci_vendor_id = g->pci_vendor_id;
	gpu->pci_device_id = g->pci_device_id;
	gpu->pci_subsystem_vendor_id = g->pci_subsystem_vendor_id;
	gpu->pci_subsystem_device_id = g->pci_subsystem_device_id;
	gpu->pci_class = g->pci_class;
	gpu->pci_revision = g->pci_revision;

	return 0;
}

/*
 * Free the gk20a struct.
 */
static void gk20a_free_cb(struct nvgpu_ref *refcount)
{
	struct gk20a *g = container_of(refcount,
		struct gk20a, refcount);

	gk20a_dbg(gpu_dbg_shutdown, "Freeing GK20A struct!");

	gk20a_ce_destroy(g);

	if (g->remove_support)
		g->remove_support(g);

	if (g->free)
		g->free(g);
}

/**
 * gk20a_get() - Increment ref count on driver
 *
 * @g The driver to increment
 * This will fail if the driver is in the process of being released. In that
 * case it will return NULL. Otherwise a pointer to the driver passed in will
 * be returned.
 */
struct gk20a * __must_check gk20a_get(struct gk20a *g)
{
	int success;

	/*
	 * Handle the possibility we are still freeing the gk20a struct while
	 * gk20a_get() is called. Unlikely but plausible race condition. Ideally
	 * the code will never be in such a situation that this race is
	 * possible.
	 */
	success = nvgpu_ref_get_unless_zero(&g->refcount);

	gk20a_dbg(gpu_dbg_shutdown, "GET: refs currently %d %s",
		nvgpu_atomic_read(&g->refcount.refcount),
			success ? "" : "(FAILED)");

	return success ? g : NULL;
}

/**
 * gk20a_put() - Decrement ref count on driver
 *
 * @g - The driver to decrement
 *
 * Decrement the driver ref-count. If neccesary also free the underlying driver
 * memory
 */
void gk20a_put(struct gk20a *g)
{
	/*
	 * Note - this is racy, two instances of this could run before the
	 * actual kref_put(0 runs, you could see something like:
	 *
	 *  ... PUT: refs currently 2
	 *  ... PUT: refs currently 2
	 *  ... Freeing GK20A struct!
	 */
	gk20a_dbg(gpu_dbg_shutdown, "PUT: refs currently %d",
		nvgpu_atomic_read(&g->refcount.refcount));

	nvgpu_ref_put(&g->refcount, gk20a_free_cb);
}