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 #include "priv.h"
26 #include "gk20a.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 GPCPLL_CFG		(SYS_GPCPLL_CFG_BASE + 0)
37 #define GPCPLL_CFG_ENABLE	BIT(0)
38 #define GPCPLL_CFG_IDDQ		BIT(1)
39 #define GPCPLL_CFG_LOCK_DET_OFF	BIT(4)
40 #define GPCPLL_CFG_LOCK		BIT(17)
41 
42 #define GPCPLL_COEFF		(SYS_GPCPLL_CFG_BASE + 4)
43 #define GPCPLL_COEFF_M_SHIFT	0
44 #define GPCPLL_COEFF_M_WIDTH	8
45 #define GPCPLL_COEFF_N_SHIFT	8
46 #define GPCPLL_COEFF_N_WIDTH	8
47 #define GPCPLL_COEFF_P_SHIFT	16
48 #define GPCPLL_COEFF_P_WIDTH	6
49 
50 #define GPCPLL_CFG2			(SYS_GPCPLL_CFG_BASE + 0xc)
51 #define GPCPLL_CFG2_SETUP2_SHIFT	16
52 #define GPCPLL_CFG2_PLL_STEPA_SHIFT	24
53 
54 #define GPCPLL_CFG3			(SYS_GPCPLL_CFG_BASE + 0x18)
55 #define GPCPLL_CFG3_PLL_STEPB_SHIFT	16
56 
57 #define GPC_BCASE_GPCPLL_CFG_BASE		0x00132800
58 #define GPCPLL_NDIV_SLOWDOWN			(SYS_GPCPLL_CFG_BASE + 0x1c)
59 #define GPCPLL_NDIV_SLOWDOWN_NDIV_LO_SHIFT	0
60 #define GPCPLL_NDIV_SLOWDOWN_NDIV_MID_SHIFT	8
61 #define GPCPLL_NDIV_SLOWDOWN_STEP_SIZE_LO2MID_SHIFT	16
62 #define GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT	22
63 #define GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT	31
64 
65 #define SEL_VCO				(SYS_GPCPLL_CFG_BASE + 0x100)
66 #define SEL_VCO_GPC2CLK_OUT_SHIFT	0
67 
68 #define GPC2CLK_OUT			(SYS_GPCPLL_CFG_BASE + 0x250)
69 #define GPC2CLK_OUT_SDIV14_INDIV4_WIDTH	1
70 #define GPC2CLK_OUT_SDIV14_INDIV4_SHIFT	31
71 #define GPC2CLK_OUT_SDIV14_INDIV4_MODE	1
72 #define GPC2CLK_OUT_VCODIV_WIDTH	6
73 #define GPC2CLK_OUT_VCODIV_SHIFT	8
74 #define GPC2CLK_OUT_VCODIV1		0
75 #define GPC2CLK_OUT_VCODIV_MASK		(MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
76 					GPC2CLK_OUT_VCODIV_SHIFT)
77 #define GPC2CLK_OUT_BYPDIV_WIDTH	6
78 #define GPC2CLK_OUT_BYPDIV_SHIFT	0
79 #define GPC2CLK_OUT_BYPDIV31		0x3c
80 #define GPC2CLK_OUT_INIT_MASK	((MASK(GPC2CLK_OUT_SDIV14_INDIV4_WIDTH) << \
81 		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT)\
82 		| (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << GPC2CLK_OUT_VCODIV_SHIFT)\
83 		| (MASK(GPC2CLK_OUT_BYPDIV_WIDTH) << GPC2CLK_OUT_BYPDIV_SHIFT))
84 #define GPC2CLK_OUT_INIT_VAL	((GPC2CLK_OUT_SDIV14_INDIV4_MODE << \
85 		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT) \
86 		| (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT) \
87 		| (GPC2CLK_OUT_BYPDIV31 << GPC2CLK_OUT_BYPDIV_SHIFT))
88 
89 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG	(GPC_BCASE_GPCPLL_CFG_BASE + 0xa0)
90 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT	24
91 #define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK \
92 	    (0x1 << GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT)
93 
94 static const u8 _pl_to_div[] = {
95 /* PL:   0, 1, 2, 3, 4, 5, 6,  7,  8,  9, 10, 11, 12, 13, 14 */
96 /* p: */ 1, 2, 3, 4, 5, 6, 8, 10, 12, 16, 12, 16, 20, 24, 32,
97 };
98 
99 static u32 pl_to_div(u32 pl)
100 {
101 	if (pl >= ARRAY_SIZE(_pl_to_div))
102 		return 1;
103 
104 	return _pl_to_div[pl];
105 }
106 
107 static u32 div_to_pl(u32 div)
108 {
109 	u32 pl;
110 
111 	for (pl = 0; pl < ARRAY_SIZE(_pl_to_div) - 1; pl++) {
112 		if (_pl_to_div[pl] >= div)
113 			return pl;
114 	}
115 
116 	return ARRAY_SIZE(_pl_to_div) - 1;
117 }
118 
119 static const struct gk20a_clk_pllg_params gk20a_pllg_params = {
120 	.min_vco = 1000000, .max_vco = 2064000,
121 	.min_u = 12000, .max_u = 38000,
122 	.min_m = 1, .max_m = 255,
123 	.min_n = 8, .max_n = 255,
124 	.min_pl = 1, .max_pl = 32,
125 };
126 
127 static void
128 gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll)
129 {
130 	struct nvkm_device *device = clk->base.subdev.device;
131 	u32 val;
132 
133 	val = nvkm_rd32(device, GPCPLL_COEFF);
134 	pll->m = (val >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
135 	pll->n = (val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH);
136 	pll->pl = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
137 }
138 
139 static void
140 gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
141 {
142 	struct nvkm_device *device = clk->base.subdev.device;
143 	u32 val;
144 
145 	val = (pll->m & MASK(GPCPLL_COEFF_M_WIDTH)) << GPCPLL_COEFF_M_SHIFT;
146 	val |= (pll->n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
147 	val |= (pll->pl & MASK(GPCPLL_COEFF_P_WIDTH)) << GPCPLL_COEFF_P_SHIFT;
148 	nvkm_wr32(device, GPCPLL_COEFF, val);
149 }
150 
151 static u32
152 gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct gk20a_pll *pll)
153 {
154 	u32 rate;
155 	u32 divider;
156 
157 	rate = clk->parent_rate * pll->n;
158 	divider = pll->m * clk->pl_to_div(pll->pl);
159 
160 	return rate / divider / 2;
161 }
162 
163 static int
164 gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate,
165 		    struct gk20a_pll *pll)
166 {
167 	struct nvkm_subdev *subdev = &clk->base.subdev;
168 	u32 target_clk_f, ref_clk_f, target_freq;
169 	u32 min_vco_f, max_vco_f;
170 	u32 low_pl, high_pl, best_pl;
171 	u32 target_vco_f;
172 	u32 best_m, best_n;
173 	u32 best_delta = ~0;
174 	u32 pl;
175 
176 	target_clk_f = rate * 2 / KHZ;
177 	ref_clk_f = clk->parent_rate / KHZ;
178 
179 	target_vco_f = target_clk_f + target_clk_f / 50;
180 	max_vco_f = max(clk->params->max_vco, target_vco_f);
181 	min_vco_f = clk->params->min_vco;
182 	best_m = clk->params->max_m;
183 	best_n = clk->params->min_n;
184 	best_pl = clk->params->min_pl;
185 
186 	/* min_pl <= high_pl <= max_pl */
187 	high_pl = (max_vco_f + target_vco_f - 1) / target_vco_f;
188 	high_pl = min(high_pl, clk->params->max_pl);
189 	high_pl = max(high_pl, clk->params->min_pl);
190 	high_pl = clk->div_to_pl(high_pl);
191 
192 	/* min_pl <= low_pl <= max_pl */
193 	low_pl = min_vco_f / target_vco_f;
194 	low_pl = min(low_pl, clk->params->max_pl);
195 	low_pl = max(low_pl, clk->params->min_pl);
196 	low_pl = clk->div_to_pl(low_pl);
197 
198 	nvkm_debug(subdev, "low_PL %d(div%d), high_PL %d(div%d)", low_pl,
199 		   clk->pl_to_div(low_pl), high_pl, clk->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 * clk->pl_to_div(pl);
206 
207 		for (m = clk->params->min_m; m <= clk->params->max_m; m++) {
208 			u32 u_f = ref_clk_f / m;
209 
210 			if (u_f < clk->params->min_u)
211 				break;
212 			if (u_f > clk->params->max_u)
213 				continue;
214 
215 			n = (target_vco_f * m) / ref_clk_f;
216 			n2 = ((target_vco_f * m) + (ref_clk_f - 1)) / ref_clk_f;
217 
218 			if (n > clk->params->max_n)
219 				break;
220 
221 			for (; n <= n2; n++) {
222 				u32 vco_f;
223 
224 				if (n < clk->params->min_n)
225 					continue;
226 				if (n > clk->params->max_n)
227 					break;
228 
229 				vco_f = ref_clk_f * n / m;
230 
231 				if (vco_f >= min_vco_f && vco_f <= max_vco_f) {
232 					u32 delta, lwv;
233 
234 					lwv = (vco_f + (clk->pl_to_div(pl) / 2))
235 						/ clk->pl_to_div(pl);
236 					delta = abs(lwv - target_clk_f);
237 
238 					if (delta < best_delta) {
239 						best_delta = delta;
240 						best_m = m;
241 						best_n = n;
242 						best_pl = pl;
243 
244 						if (best_delta == 0)
245 							goto found_match;
246 					}
247 				}
248 			}
249 		}
250 	}
251 
252 found_match:
253 	WARN_ON(best_delta == ~0);
254 
255 	if (best_delta != 0)
256 		nvkm_debug(subdev,
257 			   "no best match for target @ %dMHz on gpc_pll",
258 			   target_clk_f / KHZ);
259 
260 	pll->m = best_m;
261 	pll->n = best_n;
262 	pll->pl = best_pl;
263 
264 	target_freq = gk20a_pllg_calc_rate(clk, pll);
265 
266 	nvkm_debug(subdev,
267 		   "actual target freq %d KHz, M %d, N %d, PL %d(div%d)\n",
268 		   target_freq / KHZ, pll->m, pll->n, pll->pl,
269 		   clk->pl_to_div(pll->pl));
270 	return 0;
271 }
272 
273 static int
274 gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
275 {
276 	struct nvkm_subdev *subdev = &clk->base.subdev;
277 	struct nvkm_device *device = subdev->device;
278 	struct gk20a_pll pll;
279 	int ret = 0;
280 
281 	/* get old coefficients */
282 	gk20a_pllg_read_mnp(clk, &pll);
283 	/* do nothing if NDIV is the same */
284 	if (n == pll.n)
285 		return 0;
286 
287 	/* pll slowdown mode */
288 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
289 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
290 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
291 
292 	/* new ndiv ready for ramp */
293 	pll.n = n;
294 	udelay(1);
295 	gk20a_pllg_write_mnp(clk, &pll);
296 
297 	/* dynamic ramp to new ndiv */
298 	udelay(1);
299 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
300 		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT),
301 		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT));
302 
303 	/* wait for ramping to complete */
304 	if (nvkm_wait_usec(device, 500, GPC_BCAST_NDIV_SLOWDOWN_DEBUG,
305 		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK,
306 		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK) < 0)
307 		ret = -ETIMEDOUT;
308 
309 	/* exit slowdown mode */
310 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
311 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT) |
312 		BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
313 	nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
314 
315 	return ret;
316 }
317 
318 static void
319 gk20a_pllg_enable(struct gk20a_clk *clk)
320 {
321 	struct nvkm_device *device = clk->base.subdev.device;
322 
323 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
324 	nvkm_rd32(device, GPCPLL_CFG);
325 }
326 
327 static void
328 gk20a_pllg_disable(struct gk20a_clk *clk)
329 {
330 	struct nvkm_device *device = clk->base.subdev.device;
331 
332 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
333 	nvkm_rd32(device, GPCPLL_CFG);
334 }
335 
336 static int
337 _gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll,
338 			bool allow_slide)
339 {
340 	struct nvkm_subdev *subdev = &clk->base.subdev;
341 	struct nvkm_device *device = subdev->device;
342 	u32 val, cfg;
343 	struct gk20a_pll old_pll;
344 	u32 n_lo;
345 
346 	/* get old coefficients */
347 	gk20a_pllg_read_mnp(clk, &old_pll);
348 
349 	/* do NDIV slide if there is no change in M and PL */
350 	cfg = nvkm_rd32(device, GPCPLL_CFG);
351 	if (allow_slide && pll->m == old_pll.m &&
352 	    pll->pl == old_pll.pl && (cfg & GPCPLL_CFG_ENABLE)) {
353 		return gk20a_pllg_slide(clk, pll->n);
354 	}
355 
356 	/* slide down to NDIV_LO */
357 	if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
358 		int ret;
359 
360 		n_lo = DIV_ROUND_UP(old_pll.m * clk->params->min_vco,
361 				    clk->parent_rate / KHZ);
362 		ret = gk20a_pllg_slide(clk, n_lo);
363 
364 		if (ret)
365 			return ret;
366 	}
367 
368 	/* split FO-to-bypass jump in halfs by setting out divider 1:2 */
369 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
370 		0x2 << GPC2CLK_OUT_VCODIV_SHIFT);
371 
372 	/* put PLL in bypass before programming it */
373 	val = nvkm_rd32(device, SEL_VCO);
374 	val &= ~(BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
375 	udelay(2);
376 	nvkm_wr32(device, SEL_VCO, val);
377 
378 	/* get out from IDDQ */
379 	val = nvkm_rd32(device, GPCPLL_CFG);
380 	if (val & GPCPLL_CFG_IDDQ) {
381 		val &= ~GPCPLL_CFG_IDDQ;
382 		nvkm_wr32(device, GPCPLL_CFG, val);
383 		nvkm_rd32(device, GPCPLL_CFG);
384 		udelay(2);
385 	}
386 
387 	gk20a_pllg_disable(clk);
388 
389 	nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
390 		   pll->m, pll->n, pll->pl);
391 
392 	old_pll = *pll;
393 	if (allow_slide)
394 		old_pll.n = DIV_ROUND_UP(pll->m * clk->params->min_vco,
395 					 clk->parent_rate / KHZ);
396 	gk20a_pllg_write_mnp(clk, &old_pll);
397 
398 	gk20a_pllg_enable(clk);
399 
400 	val = nvkm_rd32(device, GPCPLL_CFG);
401 	if (val & GPCPLL_CFG_LOCK_DET_OFF) {
402 		val &= ~GPCPLL_CFG_LOCK_DET_OFF;
403 		nvkm_wr32(device, GPCPLL_CFG, val);
404 	}
405 
406 	if (nvkm_usec(device, 300,
407 		if (nvkm_rd32(device, GPCPLL_CFG) & GPCPLL_CFG_LOCK)
408 			break;
409 	) < 0)
410 		return -ETIMEDOUT;
411 
412 	/* switch to VCO mode */
413 	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
414 		  BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
415 
416 	/* restore out divider 1:1 */
417 	val = nvkm_rd32(device, GPC2CLK_OUT);
418 	if ((val & GPC2CLK_OUT_VCODIV_MASK) !=
419 	    (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT)) {
420 		val &= ~GPC2CLK_OUT_VCODIV_MASK;
421 		val |= GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT;
422 		udelay(2);
423 		nvkm_wr32(device, GPC2CLK_OUT, val);
424 		/* Intentional 2nd write to assure linear divider operation */
425 		nvkm_wr32(device, GPC2CLK_OUT, val);
426 		nvkm_rd32(device, GPC2CLK_OUT);
427 	}
428 
429 	/* slide up to new NDIV */
430 	return allow_slide ? gk20a_pllg_slide(clk, pll->n) : 0;
431 }
432 
433 static int
434 gk20a_pllg_program_mnp(struct gk20a_clk *clk)
435 {
436 	int err;
437 
438 	err = _gk20a_pllg_program_mnp(clk, &clk->pll, true);
439 	if (err)
440 		err = _gk20a_pllg_program_mnp(clk, &clk->pll, false);
441 
442 	return err;
443 }
444 
445 static struct nvkm_pstate
446 gk20a_pstates[] = {
447 	{
448 		.base = {
449 			.domain[nv_clk_src_gpc] = 72000,
450 			.voltage = 0,
451 		},
452 	},
453 	{
454 		.base = {
455 			.domain[nv_clk_src_gpc] = 108000,
456 			.voltage = 1,
457 		},
458 	},
459 	{
460 		.base = {
461 			.domain[nv_clk_src_gpc] = 180000,
462 			.voltage = 2,
463 		},
464 	},
465 	{
466 		.base = {
467 			.domain[nv_clk_src_gpc] = 252000,
468 			.voltage = 3,
469 		},
470 	},
471 	{
472 		.base = {
473 			.domain[nv_clk_src_gpc] = 324000,
474 			.voltage = 4,
475 		},
476 	},
477 	{
478 		.base = {
479 			.domain[nv_clk_src_gpc] = 396000,
480 			.voltage = 5,
481 		},
482 	},
483 	{
484 		.base = {
485 			.domain[nv_clk_src_gpc] = 468000,
486 			.voltage = 6,
487 		},
488 	},
489 	{
490 		.base = {
491 			.domain[nv_clk_src_gpc] = 540000,
492 			.voltage = 7,
493 		},
494 	},
495 	{
496 		.base = {
497 			.domain[nv_clk_src_gpc] = 612000,
498 			.voltage = 8,
499 		},
500 	},
501 	{
502 		.base = {
503 			.domain[nv_clk_src_gpc] = 648000,
504 			.voltage = 9,
505 		},
506 	},
507 	{
508 		.base = {
509 			.domain[nv_clk_src_gpc] = 684000,
510 			.voltage = 10,
511 		},
512 	},
513 	{
514 		.base = {
515 			.domain[nv_clk_src_gpc] = 708000,
516 			.voltage = 11,
517 		},
518 	},
519 	{
520 		.base = {
521 			.domain[nv_clk_src_gpc] = 756000,
522 			.voltage = 12,
523 		},
524 	},
525 	{
526 		.base = {
527 			.domain[nv_clk_src_gpc] = 804000,
528 			.voltage = 13,
529 		},
530 	},
531 	{
532 		.base = {
533 			.domain[nv_clk_src_gpc] = 852000,
534 			.voltage = 14,
535 		},
536 	},
537 };
538 
539 int
540 gk20a_clk_read(struct nvkm_clk *base, enum nv_clk_src src)
541 {
542 	struct gk20a_clk *clk = gk20a_clk(base);
543 	struct nvkm_subdev *subdev = &clk->base.subdev;
544 	struct nvkm_device *device = subdev->device;
545 	struct gk20a_pll pll;
546 
547 	switch (src) {
548 	case nv_clk_src_crystal:
549 		return device->crystal;
550 	case nv_clk_src_gpc:
551 		gk20a_pllg_read_mnp(clk, &pll);
552 		return gk20a_pllg_calc_rate(clk, &pll) / GK20A_CLK_GPC_MDIV;
553 	default:
554 		nvkm_error(subdev, "invalid clock source %d\n", src);
555 		return -EINVAL;
556 	}
557 }
558 
559 int
560 gk20a_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
561 {
562 	struct gk20a_clk *clk = gk20a_clk(base);
563 
564 	return gk20a_pllg_calc_mnp(clk, cstate->domain[nv_clk_src_gpc] *
565 					 GK20A_CLK_GPC_MDIV, &clk->pll);
566 }
567 
568 int
569 gk20a_clk_prog(struct nvkm_clk *base)
570 {
571 	struct gk20a_clk *clk = gk20a_clk(base);
572 
573 	return gk20a_pllg_program_mnp(clk);
574 }
575 
576 void
577 gk20a_clk_tidy(struct nvkm_clk *base)
578 {
579 }
580 
581 int
582 gk20a_clk_setup_slide(struct gk20a_clk *clk)
583 {
584 	struct nvkm_subdev *subdev = &clk->base.subdev;
585 	struct nvkm_device *device = subdev->device;
586 	u32 step_a, step_b;
587 
588 	switch (clk->parent_rate) {
589 	case 12000000:
590 	case 12800000:
591 	case 13000000:
592 		step_a = 0x2b;
593 		step_b = 0x0b;
594 		break;
595 	case 19200000:
596 		step_a = 0x12;
597 		step_b = 0x08;
598 		break;
599 	case 38400000:
600 		step_a = 0x04;
601 		step_b = 0x05;
602 		break;
603 	default:
604 		nvkm_error(subdev, "invalid parent clock rate %u KHz",
605 			   clk->parent_rate / KHZ);
606 		return -EINVAL;
607 	}
608 
609 	nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
610 		step_a << GPCPLL_CFG2_PLL_STEPA_SHIFT);
611 	nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
612 		step_b << GPCPLL_CFG3_PLL_STEPB_SHIFT);
613 
614 	return 0;
615 }
616 
617 void
618 gk20a_clk_fini(struct nvkm_clk *base)
619 {
620 	struct nvkm_device *device = base->subdev.device;
621 	struct gk20a_clk *clk = gk20a_clk(base);
622 	u32 val;
623 
624 	/* slide to VCO min */
625 	val = nvkm_rd32(device, GPCPLL_CFG);
626 	if (val & GPCPLL_CFG_ENABLE) {
627 		struct gk20a_pll pll;
628 		u32 n_lo;
629 
630 		gk20a_pllg_read_mnp(clk, &pll);
631 		n_lo = DIV_ROUND_UP(pll.m * clk->params->min_vco,
632 				    clk->parent_rate / KHZ);
633 		gk20a_pllg_slide(clk, n_lo);
634 	}
635 
636 	/* put PLL in bypass before disabling it */
637 	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
638 
639 	gk20a_pllg_disable(clk);
640 }
641 
642 static int
643 gk20a_clk_init(struct nvkm_clk *base)
644 {
645 	struct gk20a_clk *clk = gk20a_clk(base);
646 	struct nvkm_subdev *subdev = &clk->base.subdev;
647 	struct nvkm_device *device = subdev->device;
648 	int ret;
649 
650 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
651 		  GPC2CLK_OUT_INIT_VAL);
652 
653 	ret = gk20a_clk_setup_slide(clk);
654 	if (ret)
655 		return ret;
656 
657 	/* Start with lowest frequency */
658 	base->func->calc(base, &base->func->pstates[0].base);
659 	ret = base->func->prog(&clk->base);
660 	if (ret) {
661 		nvkm_error(subdev, "cannot initialize clock\n");
662 		return ret;
663 	}
664 
665 	return 0;
666 }
667 
668 static const struct nvkm_clk_func
669 gk20a_clk = {
670 	.init = gk20a_clk_init,
671 	.fini = gk20a_clk_fini,
672 	.read = gk20a_clk_read,
673 	.calc = gk20a_clk_calc,
674 	.prog = gk20a_clk_prog,
675 	.tidy = gk20a_clk_tidy,
676 	.pstates = gk20a_pstates,
677 	.nr_pstates = ARRAY_SIZE(gk20a_pstates),
678 	.domains = {
679 		{ nv_clk_src_crystal, 0xff },
680 		{ nv_clk_src_gpc, 0xff, 0, "core", GK20A_CLK_GPC_MDIV },
681 		{ nv_clk_src_max }
682 	}
683 };
684 
685 int
686 _gk20a_clk_ctor(struct nvkm_device *device, int index,
687 		const struct nvkm_clk_func *func,
688 		const struct gk20a_clk_pllg_params *params,
689 		struct gk20a_clk *clk)
690 {
691 	struct nvkm_device_tegra *tdev = device->func->tegra(device);
692 	int ret;
693 	int i;
694 
695 	/* Finish initializing the pstates */
696 	for (i = 0; i < func->nr_pstates; i++) {
697 		INIT_LIST_HEAD(&func->pstates[i].list);
698 		func->pstates[i].pstate = i + 1;
699 	}
700 
701 	clk->params = params;
702 	clk->parent_rate = clk_get_rate(tdev->clk);
703 
704 	ret = nvkm_clk_ctor(func, device, index, true, &clk->base);
705 	if (ret)
706 		return ret;
707 
708 	nvkm_debug(&clk->base.subdev, "parent clock rate: %d Khz\n",
709 		   clk->parent_rate / KHZ);
710 
711 	return 0;
712 }
713 
714 int
715 gk20a_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
716 {
717 	struct gk20a_clk *clk;
718 	int ret;
719 
720 	clk = kzalloc(sizeof(*clk), GFP_KERNEL);
721 	if (!clk)
722 		return -ENOMEM;
723 	*pclk = &clk->base;
724 
725 	ret = _gk20a_clk_ctor(device, index, &gk20a_clk, &gk20a_pllg_params,
726 			      clk);
727 
728 	clk->pl_to_div = pl_to_div;
729 	clk->div_to_pl = div_to_pl;
730 
731 	return ret;
732 }
733