1 /*
2  * Copyright (c) 2014-2016, NVIDIA CORPORATION. All rights reserved.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
20  * DEALINGS IN THE SOFTWARE.
21  *
22  * Shamelessly ripped off from ChromeOS's gk20a/clk_pllg.c
23  *
24  */
25 #define gk20a_clk(p) container_of((p), struct gk20a_clk, base)
26 #include "priv.h"
27 
28 #include <core/tegra.h>
29 #include <subdev/timer.h>
30 
31 #define KHZ (1000)
32 #define MHZ (KHZ * 1000)
33 
34 #define MASK(w)	((1 << w) - 1)
35 
36 #define SYS_GPCPLL_CFG_BASE			0x00137000
37 #define GPC_BCASE_GPCPLL_CFG_BASE		0x00132800
38 
39 #define GPCPLL_CFG		(SYS_GPCPLL_CFG_BASE + 0)
40 #define GPCPLL_CFG_ENABLE	BIT(0)
41 #define GPCPLL_CFG_IDDQ		BIT(1)
42 #define GPCPLL_CFG_LOCK_DET_OFF	BIT(4)
43 #define GPCPLL_CFG_LOCK		BIT(17)
44 
45 #define GPCPLL_COEFF		(SYS_GPCPLL_CFG_BASE + 4)
46 #define GPCPLL_COEFF_M_SHIFT	0
47 #define GPCPLL_COEFF_M_WIDTH	8
48 #define GPCPLL_COEFF_N_SHIFT	8
49 #define GPCPLL_COEFF_N_WIDTH	8
50 #define GPCPLL_COEFF_P_SHIFT	16
51 #define GPCPLL_COEFF_P_WIDTH	6
52 
53 #define GPCPLL_CFG2			(SYS_GPCPLL_CFG_BASE + 0xc)
54 #define GPCPLL_CFG2_SETUP2_SHIFT	16
55 #define GPCPLL_CFG2_PLL_STEPA_SHIFT	24
56 
57 #define GPCPLL_CFG3			(SYS_GPCPLL_CFG_BASE + 0x18)
58 #define GPCPLL_CFG3_PLL_STEPB_SHIFT	16
59 
60 #define GPCPLL_NDIV_SLOWDOWN			(SYS_GPCPLL_CFG_BASE + 0x1c)
61 #define GPCPLL_NDIV_SLOWDOWN_NDIV_LO_SHIFT	0
62 #define GPCPLL_NDIV_SLOWDOWN_NDIV_MID_SHIFT	8
63 #define GPCPLL_NDIV_SLOWDOWN_STEP_SIZE_LO2MID_SHIFT	16
64 #define GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT	22
65 #define GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT	31
66 
67 #define SEL_VCO				(SYS_GPCPLL_CFG_BASE + 0x100)
68 #define SEL_VCO_GPC2CLK_OUT_SHIFT	0
69 
70 #define GPC2CLK_OUT			(SYS_GPCPLL_CFG_BASE + 0x250)
71 #define GPC2CLK_OUT_SDIV14_INDIV4_WIDTH	1
72 #define GPC2CLK_OUT_SDIV14_INDIV4_SHIFT	31
73 #define GPC2CLK_OUT_SDIV14_INDIV4_MODE	1
74 #define GPC2CLK_OUT_VCODIV_WIDTH	6
75 #define GPC2CLK_OUT_VCODIV_SHIFT	8
76 #define GPC2CLK_OUT_VCODIV1		0
77 #define GPC2CLK_OUT_VCODIV_MASK		(MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
78 					GPC2CLK_OUT_VCODIV_SHIFT)
79 #define	GPC2CLK_OUT_BYPDIV_WIDTH	6
80 #define GPC2CLK_OUT_BYPDIV_SHIFT	0
81 #define GPC2CLK_OUT_BYPDIV31		0x3c
82 #define GPC2CLK_OUT_INIT_MASK	((MASK(GPC2CLK_OUT_SDIV14_INDIV4_WIDTH) << \
83 		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT)\
84 		| (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << GPC2CLK_OUT_VCODIV_SHIFT)\
85 		| (MASK(GPC2CLK_OUT_BYPDIV_WIDTH) << GPC2CLK_OUT_BYPDIV_SHIFT))
86 #define GPC2CLK_OUT_INIT_VAL	((GPC2CLK_OUT_SDIV14_INDIV4_MODE << \
87 		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT) \
88 		| (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT) \
89 		| (GPC2CLK_OUT_BYPDIV31 << GPC2CLK_OUT_BYPDIV_SHIFT))
90 
91 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG	(GPC_BCASE_GPCPLL_CFG_BASE + 0xa0)
92 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT	24
93 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK \
94 	    (0x1 << GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT)
95 
96 static const u8 pl_to_div[] = {
97 /* PL:   0, 1, 2, 3, 4, 5, 6,  7,  8,  9, 10, 11, 12, 13, 14 */
98 /* p: */ 1, 2, 3, 4, 5, 6, 8, 10, 12, 16, 12, 16, 20, 24, 32,
99 };
100 
101 /* All frequencies in Khz */
102 struct gk20a_clk_pllg_params {
103 	u32 min_vco, max_vco;
104 	u32 min_u, max_u;
105 	u32 min_m, max_m;
106 	u32 min_n, max_n;
107 	u32 min_pl, max_pl;
108 };
109 
110 static const struct gk20a_clk_pllg_params gk20a_pllg_params = {
111 	.min_vco = 1000000, .max_vco = 2064000,
112 	.min_u = 12000, .max_u = 38000,
113 	.min_m = 1, .max_m = 255,
114 	.min_n = 8, .max_n = 255,
115 	.min_pl = 1, .max_pl = 32,
116 };
117 
118 struct gk20a_clk {
119 	struct nvkm_clk base;
120 	const struct gk20a_clk_pllg_params *params;
121 	u32 m, n, pl;
122 	u32 parent_rate;
123 };
124 
125 static void
126 gk20a_pllg_read_mnp(struct gk20a_clk *clk)
127 {
128 	struct nvkm_device *device = clk->base.subdev.device;
129 	u32 val;
130 
131 	val = nvkm_rd32(device, GPCPLL_COEFF);
132 	clk->m = (val >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
133 	clk->n = (val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH);
134 	clk->pl = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
135 }
136 
137 static u32
138 gk20a_pllg_calc_rate(struct gk20a_clk *clk)
139 {
140 	u32 rate;
141 	u32 divider;
142 
143 	rate = clk->parent_rate * clk->n;
144 	divider = clk->m * pl_to_div[clk->pl];
145 
146 	return rate / divider / 2;
147 }
148 
149 static int
150 gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
151 {
152 	struct nvkm_subdev *subdev = &clk->base.subdev;
153 	u32 target_clk_f, ref_clk_f, target_freq;
154 	u32 min_vco_f, max_vco_f;
155 	u32 low_pl, high_pl, best_pl;
156 	u32 target_vco_f;
157 	u32 best_m, best_n;
158 	u32 best_delta = ~0;
159 	u32 pl;
160 
161 	target_clk_f = rate * 2 / KHZ;
162 	ref_clk_f = clk->parent_rate / KHZ;
163 
164 	max_vco_f = clk->params->max_vco;
165 	min_vco_f = clk->params->min_vco;
166 	best_m = clk->params->max_m;
167 	best_n = clk->params->min_n;
168 	best_pl = clk->params->min_pl;
169 
170 	target_vco_f = target_clk_f + target_clk_f / 50;
171 	if (max_vco_f < target_vco_f)
172 		max_vco_f = target_vco_f;
173 
174 	/* min_pl <= high_pl <= max_pl */
175 	high_pl = (max_vco_f + target_vco_f - 1) / target_vco_f;
176 	high_pl = min(high_pl, clk->params->max_pl);
177 	high_pl = max(high_pl, clk->params->min_pl);
178 
179 	/* min_pl <= low_pl <= max_pl */
180 	low_pl = min_vco_f / target_vco_f;
181 	low_pl = min(low_pl, clk->params->max_pl);
182 	low_pl = max(low_pl, clk->params->min_pl);
183 
184 	/* Find Indices of high_pl and low_pl */
185 	for (pl = 0; pl < ARRAY_SIZE(pl_to_div) - 1; pl++) {
186 		if (pl_to_div[pl] >= low_pl) {
187 			low_pl = pl;
188 			break;
189 		}
190 	}
191 	for (pl = 0; pl < ARRAY_SIZE(pl_to_div) - 1; pl++) {
192 		if (pl_to_div[pl] >= high_pl) {
193 			high_pl = pl;
194 			break;
195 		}
196 	}
197 
198 	nvkm_debug(subdev, "low_PL %d(div%d), high_PL %d(div%d)", low_pl,
199 		   pl_to_div[low_pl], high_pl, pl_to_div[high_pl]);
200 
201 	/* Select lowest possible VCO */
202 	for (pl = low_pl; pl <= high_pl; pl++) {
203 		u32 m, n, n2;
204 
205 		target_vco_f = target_clk_f * pl_to_div[pl];
206 		for (m = clk->params->min_m; m <= clk->params->max_m; m++) {
207 			u32 u_f, vco_f;
208 
209 			u_f = ref_clk_f / m;
210 
211 			if (u_f < clk->params->min_u)
212 				break;
213 			if (u_f > clk->params->max_u)
214 				continue;
215 
216 			n = (target_vco_f * m) / ref_clk_f;
217 			n2 = ((target_vco_f * m) + (ref_clk_f - 1)) / ref_clk_f;
218 
219 			if (n > clk->params->max_n)
220 				break;
221 
222 			for (; n <= n2; n++) {
223 				if (n < clk->params->min_n)
224 					continue;
225 				if (n > clk->params->max_n)
226 					break;
227 
228 				vco_f = ref_clk_f * n / m;
229 
230 				if (vco_f >= min_vco_f && vco_f <= max_vco_f) {
231 					u32 delta, lwv;
232 
233 					lwv = (vco_f + (pl_to_div[pl] / 2))
234 						/ pl_to_div[pl];
235 					delta = abs(lwv - target_clk_f);
236 
237 					if (delta < best_delta) {
238 						best_delta = delta;
239 						best_m = m;
240 						best_n = n;
241 						best_pl = pl;
242 
243 						if (best_delta == 0)
244 							goto found_match;
245 					}
246 				}
247 			}
248 		}
249 	}
250 
251 found_match:
252 	WARN_ON(best_delta == ~0);
253 
254 	if (best_delta != 0)
255 		nvkm_debug(subdev,
256 			   "no best match for target @ %dMHz on gpc_pll",
257 			   target_clk_f / KHZ);
258 
259 	clk->m = best_m;
260 	clk->n = best_n;
261 	clk->pl = best_pl;
262 
263 	target_freq = gk20a_pllg_calc_rate(clk);
264 
265 	nvkm_debug(subdev,
266 		   "actual target freq %d MHz, M %d, N %d, PL %d(div%d)\n",
267 		   target_freq / MHZ, clk->m, clk->n, clk->pl,
268 		   pl_to_div[clk->pl]);
269 	return 0;
270 }
271 
272 static int
273 gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
274 {
275 	struct nvkm_subdev *subdev = &clk->base.subdev;
276 	struct nvkm_device *device = subdev->device;
277 	u32 val;
278 	int ramp_timeout;
279 
280 	/* get old coefficients */
281 	val = nvkm_rd32(device, GPCPLL_COEFF);
282 	/* do nothing if NDIV is the same */
283 	if (n == ((val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH)))
284 		return 0;
285 
286 	/* setup */
287 	nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
288 		0x2b << GPCPLL_CFG2_PLL_STEPA_SHIFT);
289 	nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
290 		0xb << GPCPLL_CFG3_PLL_STEPB_SHIFT);
291 
292 	/* pll slowdown mode */
293 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
294 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
295 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
296 
297 	/* new ndiv ready for ramp */
298 	val = nvkm_rd32(device, GPCPLL_COEFF);
299 	val &= ~(MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT);
300 	val |= (n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
301 	udelay(1);
302 	nvkm_wr32(device, GPCPLL_COEFF, val);
303 
304 	/* dynamic ramp to new ndiv */
305 	val = nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
306 	val |= 0x1 << GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT;
307 	udelay(1);
308 	nvkm_wr32(device, GPCPLL_NDIV_SLOWDOWN, val);
309 
310 	for (ramp_timeout = 500; ramp_timeout > 0; ramp_timeout--) {
311 		udelay(1);
312 		val = nvkm_rd32(device, GPC_BCAST_NDIV_SLOWDOWN_DEBUG);
313 		if (val & GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK)
314 			break;
315 	}
316 
317 	/* exit slowdown mode */
318 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
319 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT) |
320 		BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
321 	nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
322 
323 	if (ramp_timeout <= 0) {
324 		nvkm_error(subdev, "gpcpll dynamic ramp timeout\n");
325 		return -ETIMEDOUT;
326 	}
327 
328 	return 0;
329 }
330 
331 static void
332 gk20a_pllg_enable(struct gk20a_clk *clk)
333 {
334 	struct nvkm_device *device = clk->base.subdev.device;
335 
336 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
337 	nvkm_rd32(device, GPCPLL_CFG);
338 }
339 
340 static void
341 gk20a_pllg_disable(struct gk20a_clk *clk)
342 {
343 	struct nvkm_device *device = clk->base.subdev.device;
344 
345 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
346 	nvkm_rd32(device, GPCPLL_CFG);
347 }
348 
349 static int
350 _gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool allow_slide)
351 {
352 	struct nvkm_subdev *subdev = &clk->base.subdev;
353 	struct nvkm_device *device = subdev->device;
354 	u32 val, cfg;
355 	u32 m_old, pl_old, n_lo;
356 
357 	/* get old coefficients */
358 	val = nvkm_rd32(device, GPCPLL_COEFF);
359 	m_old = (val >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
360 	pl_old = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
361 
362 	/* do NDIV slide if there is no change in M and PL */
363 	cfg = nvkm_rd32(device, GPCPLL_CFG);
364 	if (allow_slide && clk->m == m_old && clk->pl == pl_old &&
365 	    (cfg & GPCPLL_CFG_ENABLE)) {
366 		return gk20a_pllg_slide(clk, clk->n);
367 	}
368 
369 	/* slide down to NDIV_LO */
370 	n_lo = DIV_ROUND_UP(m_old * clk->params->min_vco,
371 			    clk->parent_rate / KHZ);
372 	if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
373 		int ret = gk20a_pllg_slide(clk, n_lo);
374 
375 		if (ret)
376 			return ret;
377 	}
378 
379 	/* split FO-to-bypass jump in halfs by setting out divider 1:2 */
380 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
381 		0x2 << GPC2CLK_OUT_VCODIV_SHIFT);
382 
383 	/* put PLL in bypass before programming it */
384 	val = nvkm_rd32(device, SEL_VCO);
385 	val &= ~(BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
386 	udelay(2);
387 	nvkm_wr32(device, SEL_VCO, val);
388 
389 	/* get out from IDDQ */
390 	val = nvkm_rd32(device, GPCPLL_CFG);
391 	if (val & GPCPLL_CFG_IDDQ) {
392 		val &= ~GPCPLL_CFG_IDDQ;
393 		nvkm_wr32(device, GPCPLL_CFG, val);
394 		nvkm_rd32(device, GPCPLL_CFG);
395 		udelay(2);
396 	}
397 
398 	gk20a_pllg_disable(clk);
399 
400 	nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
401 		   clk->m, clk->n, clk->pl);
402 
403 	n_lo = DIV_ROUND_UP(clk->m * clk->params->min_vco,
404 			    clk->parent_rate / KHZ);
405 	val = clk->m << GPCPLL_COEFF_M_SHIFT;
406 	val |= (allow_slide ? n_lo : clk->n) << GPCPLL_COEFF_N_SHIFT;
407 	val |= clk->pl << GPCPLL_COEFF_P_SHIFT;
408 	nvkm_wr32(device, GPCPLL_COEFF, val);
409 
410 	gk20a_pllg_enable(clk);
411 
412 	val = nvkm_rd32(device, GPCPLL_CFG);
413 	if (val & GPCPLL_CFG_LOCK_DET_OFF) {
414 		val &= ~GPCPLL_CFG_LOCK_DET_OFF;
415 		nvkm_wr32(device, GPCPLL_CFG, val);
416 	}
417 
418 	if (nvkm_usec(device, 300,
419 		if (nvkm_rd32(device, GPCPLL_CFG) & GPCPLL_CFG_LOCK)
420 			break;
421 	) < 0)
422 		return -ETIMEDOUT;
423 
424 	/* switch to VCO mode */
425 	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
426 		  BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
427 
428 	/* restore out divider 1:1 */
429 	val = nvkm_rd32(device, GPC2CLK_OUT);
430 	val &= ~GPC2CLK_OUT_VCODIV_MASK;
431 	udelay(2);
432 	nvkm_wr32(device, GPC2CLK_OUT, val);
433 
434 	/* slide up to new NDIV */
435 	return allow_slide ? gk20a_pllg_slide(clk, clk->n) : 0;
436 }
437 
438 static int
439 gk20a_pllg_program_mnp(struct gk20a_clk *clk)
440 {
441 	int err;
442 
443 	err = _gk20a_pllg_program_mnp(clk, true);
444 	if (err)
445 		err = _gk20a_pllg_program_mnp(clk, false);
446 
447 	return err;
448 }
449 
450 #define GK20A_CLK_GPC_MDIV 1000
451 
452 static struct nvkm_pstate
453 gk20a_pstates[] = {
454 	{
455 		.base = {
456 			.domain[nv_clk_src_gpc] = 72000,
457 			.voltage = 0,
458 		},
459 	},
460 	{
461 		.base = {
462 			.domain[nv_clk_src_gpc] = 108000,
463 			.voltage = 1,
464 		},
465 	},
466 	{
467 		.base = {
468 			.domain[nv_clk_src_gpc] = 180000,
469 			.voltage = 2,
470 		},
471 	},
472 	{
473 		.base = {
474 			.domain[nv_clk_src_gpc] = 252000,
475 			.voltage = 3,
476 		},
477 	},
478 	{
479 		.base = {
480 			.domain[nv_clk_src_gpc] = 324000,
481 			.voltage = 4,
482 		},
483 	},
484 	{
485 		.base = {
486 			.domain[nv_clk_src_gpc] = 396000,
487 			.voltage = 5,
488 		},
489 	},
490 	{
491 		.base = {
492 			.domain[nv_clk_src_gpc] = 468000,
493 			.voltage = 6,
494 		},
495 	},
496 	{
497 		.base = {
498 			.domain[nv_clk_src_gpc] = 540000,
499 			.voltage = 7,
500 		},
501 	},
502 	{
503 		.base = {
504 			.domain[nv_clk_src_gpc] = 612000,
505 			.voltage = 8,
506 		},
507 	},
508 	{
509 		.base = {
510 			.domain[nv_clk_src_gpc] = 648000,
511 			.voltage = 9,
512 		},
513 	},
514 	{
515 		.base = {
516 			.domain[nv_clk_src_gpc] = 684000,
517 			.voltage = 10,
518 		},
519 	},
520 	{
521 		.base = {
522 			.domain[nv_clk_src_gpc] = 708000,
523 			.voltage = 11,
524 		},
525 	},
526 	{
527 		.base = {
528 			.domain[nv_clk_src_gpc] = 756000,
529 			.voltage = 12,
530 		},
531 	},
532 	{
533 		.base = {
534 			.domain[nv_clk_src_gpc] = 804000,
535 			.voltage = 13,
536 		},
537 	},
538 	{
539 		.base = {
540 			.domain[nv_clk_src_gpc] = 852000,
541 			.voltage = 14,
542 		},
543 	},
544 };
545 
546 static int
547 gk20a_clk_read(struct nvkm_clk *base, enum nv_clk_src src)
548 {
549 	struct gk20a_clk *clk = gk20a_clk(base);
550 	struct nvkm_subdev *subdev = &clk->base.subdev;
551 	struct nvkm_device *device = subdev->device;
552 
553 	switch (src) {
554 	case nv_clk_src_crystal:
555 		return device->crystal;
556 	case nv_clk_src_gpc:
557 		gk20a_pllg_read_mnp(clk);
558 		return gk20a_pllg_calc_rate(clk) / GK20A_CLK_GPC_MDIV;
559 	default:
560 		nvkm_error(subdev, "invalid clock source %d\n", src);
561 		return -EINVAL;
562 	}
563 }
564 
565 static int
566 gk20a_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
567 {
568 	struct gk20a_clk *clk = gk20a_clk(base);
569 
570 	return gk20a_pllg_calc_mnp(clk, cstate->domain[nv_clk_src_gpc] *
571 					 GK20A_CLK_GPC_MDIV);
572 }
573 
574 static int
575 gk20a_clk_prog(struct nvkm_clk *base)
576 {
577 	struct gk20a_clk *clk = gk20a_clk(base);
578 
579 	return gk20a_pllg_program_mnp(clk);
580 }
581 
582 static void
583 gk20a_clk_tidy(struct nvkm_clk *base)
584 {
585 }
586 
587 static void
588 gk20a_clk_fini(struct nvkm_clk *base)
589 {
590 	struct nvkm_device *device = base->subdev.device;
591 	struct gk20a_clk *clk = gk20a_clk(base);
592 	u32 val;
593 
594 	/* slide to VCO min */
595 	val = nvkm_rd32(device, GPCPLL_CFG);
596 	if (val & GPCPLL_CFG_ENABLE) {
597 		u32 coef, m, n_lo;
598 
599 		coef = nvkm_rd32(device, GPCPLL_COEFF);
600 		m = (coef >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
601 		n_lo = DIV_ROUND_UP(m * clk->params->min_vco,
602 				    clk->parent_rate / KHZ);
603 		gk20a_pllg_slide(clk, n_lo);
604 	}
605 
606 	/* put PLL in bypass before disabling it */
607 	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
608 
609 	gk20a_pllg_disable(clk);
610 }
611 
612 static int
613 gk20a_clk_init(struct nvkm_clk *base)
614 {
615 	struct gk20a_clk *clk = gk20a_clk(base);
616 	struct nvkm_subdev *subdev = &clk->base.subdev;
617 	struct nvkm_device *device = subdev->device;
618 	int ret;
619 
620 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK, GPC2CLK_OUT_INIT_VAL);
621 
622 	ret = gk20a_clk_prog(&clk->base);
623 	if (ret) {
624 		nvkm_error(subdev, "cannot initialize clock\n");
625 		return ret;
626 	}
627 
628 	return 0;
629 }
630 
631 static const struct nvkm_clk_func
632 gk20a_clk = {
633 	.init = gk20a_clk_init,
634 	.fini = gk20a_clk_fini,
635 	.read = gk20a_clk_read,
636 	.calc = gk20a_clk_calc,
637 	.prog = gk20a_clk_prog,
638 	.tidy = gk20a_clk_tidy,
639 	.pstates = gk20a_pstates,
640 	.nr_pstates = ARRAY_SIZE(gk20a_pstates),
641 	.domains = {
642 		{ nv_clk_src_crystal, 0xff },
643 		{ nv_clk_src_gpc, 0xff, 0, "core", GK20A_CLK_GPC_MDIV },
644 		{ nv_clk_src_max }
645 	}
646 };
647 
648 int
649 gk20a_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
650 {
651 	struct nvkm_device_tegra *tdev = device->func->tegra(device);
652 	struct gk20a_clk *clk;
653 	int ret, i;
654 
655 	if (!(clk = kzalloc(sizeof(*clk), GFP_KERNEL)))
656 		return -ENOMEM;
657 	*pclk = &clk->base;
658 
659 	/* Finish initializing the pstates */
660 	for (i = 0; i < ARRAY_SIZE(gk20a_pstates); i++) {
661 		INIT_LIST_HEAD(&gk20a_pstates[i].list);
662 		gk20a_pstates[i].pstate = i + 1;
663 	}
664 
665 	clk->params = &gk20a_pllg_params;
666 	clk->parent_rate = clk_get_rate(tdev->clk);
667 
668 	ret = nvkm_clk_ctor(&gk20a_clk, device, index, true, &clk->base);
669 	nvkm_info(&clk->base.subdev, "parent clock rate: %d Khz\n",
670 		  clk->parent_rate / KHZ);
671 	return ret;
672 }
673