2 * Copyright 2012 Red Hat Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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 COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
26 #include <engine/fifo.h>
27 #include <subdev/bios.h>
28 #include <subdev/bios/pll.h>
29 #include <subdev/timer.h>
35 struct nva3_clock_priv {
36 struct nouveau_clock base;
37 struct nva3_clock_info eng[nv_clk_src_max];
40 static u32 read_clk(struct nva3_clock_priv *, int, bool);
41 static u32 read_pll(struct nva3_clock_priv *, int, u32);
44 read_vco(struct nva3_clock_priv *priv, int clk)
46 u32 sctl = nv_rd32(priv, 0x4120 + (clk * 4));
48 switch (sctl & 0x00000030) {
50 return nv_device(priv)->crystal;
52 return read_pll(priv, 0x41, 0x00e820);
54 return read_pll(priv, 0x42, 0x00e8a0);
61 read_clk(struct nva3_clock_priv *priv, int clk, bool ignore_en)
65 /* refclk for the 0xe8xx plls is a fixed frequency */
67 if (nv_device(priv)->chipset == 0xaf) {
68 /* no joke.. seriously.. sigh.. */
69 return nv_rd32(priv, 0x00471c) * 1000;
72 return nv_device(priv)->crystal;
75 sctl = nv_rd32(priv, 0x4120 + (clk * 4));
76 if (!ignore_en && !(sctl & 0x00000100))
80 if (sctl & 0x00000400)
84 switch (sctl & 0x00003000) {
86 if (!(sctl & 0x00000200))
87 return nv_device(priv)->crystal;
90 if (sctl & 0x00000040)
95 if (!(sctl & 0x00000001))
98 sclk = read_vco(priv, clk);
99 sdiv = ((sctl & 0x003f0000) >> 16) + 2;
100 return (sclk * 2) / sdiv;
107 read_pll(struct nva3_clock_priv *priv, int clk, u32 pll)
109 u32 ctrl = nv_rd32(priv, pll + 0);
110 u32 sclk = 0, P = 1, N = 1, M = 1;
112 if (!(ctrl & 0x00000008)) {
113 if (ctrl & 0x00000001) {
114 u32 coef = nv_rd32(priv, pll + 4);
115 M = (coef & 0x000000ff) >> 0;
116 N = (coef & 0x0000ff00) >> 8;
117 P = (coef & 0x003f0000) >> 16;
119 /* no post-divider on these..
120 * XXX: it looks more like two post-"dividers" that
121 * cross each other out in the default RPLL config */
122 if ((pll & 0x00ff00) == 0x00e800)
125 sclk = read_clk(priv, 0x00 + clk, false);
128 sclk = read_clk(priv, 0x10 + clk, false);
132 return sclk * N / (M * P);
137 nva3_clock_read(struct nouveau_clock *clk, enum nv_clk_src src)
139 struct nva3_clock_priv *priv = (void *)clk;
143 case nv_clk_src_crystal:
144 return nv_device(priv)->crystal;
145 case nv_clk_src_core:
146 case nv_clk_src_core_intm:
147 return read_pll(priv, 0x00, 0x4200);
148 case nv_clk_src_shader:
149 return read_pll(priv, 0x01, 0x4220);
151 return read_pll(priv, 0x02, 0x4000);
152 case nv_clk_src_disp:
153 return read_clk(priv, 0x20, false);
154 case nv_clk_src_vdec:
155 return read_clk(priv, 0x21, false);
156 case nv_clk_src_daemon:
157 return read_clk(priv, 0x25, false);
158 case nv_clk_src_host:
159 hsrc = (nv_rd32(priv, 0xc040) & 0x30000000) >> 28;
162 return read_clk(priv, 0x1d, false);
167 nv_error(clk, "unknown HOST clock source %d\n", hsrc);
171 nv_error(clk, "invalid clock source %d\n", src);
179 nva3_clk_info(struct nouveau_clock *clock, int clk, u32 khz,
180 struct nva3_clock_info *info)
182 struct nva3_clock_priv *priv = (void *)clock;
183 u32 oclk, sclk, sdiv, diff;
189 info->clk = 0x00000100;
192 info->clk = 0x00002100;
195 info->clk = 0x00002140;
198 sclk = read_vco(priv, clk);
199 sdiv = min((sclk * 2) / khz, (u32)65);
200 oclk = (sclk * 2) / sdiv;
201 diff = ((khz + 3000) - oclk);
203 /* When imprecise, play it safe and aim for a clock lower than
204 * desired rather than higher */
207 oclk = (sclk * 2) / sdiv;
210 /* divider can go as low as 2, limited here because NVIDIA
211 * and the VBIOS on my NVA8 seem to prefer using the PLL
212 * for 810MHz - is there a good reason?
213 * XXX: PLLs with refclk 810MHz? */
215 info->clk = (((sdiv - 2) << 16) | 0x00003100);
226 nva3_pll_info(struct nouveau_clock *clock, int clk, u32 pll, u32 khz,
227 struct nva3_clock_info *info)
229 struct nouveau_bios *bios = nouveau_bios(clock);
230 struct nva3_clock_priv *priv = (void *)clock;
231 struct nvbios_pll limits;
237 /* If we can get a within [-2, 3) MHz of a divider, we'll disable the
238 * PLL and use the divider instead. */
239 ret = nva3_clk_info(clock, clk, khz, info);
241 if (!pll || (diff >= -2000 && diff < 3000)) {
246 ret = nvbios_pll_parse(bios, pll, &limits);
250 ret = nva3_clk_info(clock, clk - 0x10, limits.refclk, info);
251 if (ret != limits.refclk)
254 ret = nva3_pll_calc(nv_subdev(priv), &limits, khz, &N, NULL, &M, &P);
256 info->pll = (P << 16) | (N << 8) | M;
260 info->fb_delay = max(((khz + 7566) / 15133), (u32) 18);
262 return ret ? ret : -ERANGE;
266 calc_clk(struct nva3_clock_priv *priv, struct nouveau_cstate *cstate,
267 int clk, u32 pll, int idx)
269 int ret = nva3_pll_info(&priv->base, clk, pll, cstate->domain[idx],
277 calc_host(struct nva3_clock_priv *priv, struct nouveau_cstate *cstate)
280 u32 kHz = cstate->domain[nv_clk_src_host];
281 struct nva3_clock_info *info = &priv->eng[nv_clk_src_host];
285 info->host_out = NVA3_HOST_277;
289 info->host_out = NVA3_HOST_CLK;
291 ret = nva3_clk_info(&priv->base, 0x1d, kHz, info);
298 nva3_clock_pre(struct nouveau_clock *clk, unsigned long *flags)
300 struct nouveau_fifo *pfifo = nouveau_fifo(clk);
302 /* halt and idle execution engines */
303 nv_mask(clk, 0x020060, 0x00070000, 0x00000000);
304 nv_mask(clk, 0x002504, 0x00000001, 0x00000001);
305 /* Wait until the interrupt handler is finished */
306 if (!nv_wait(clk, 0x000100, 0xffffffff, 0x00000000))
310 pfifo->pause(pfifo, flags);
312 if (!nv_wait(clk, 0x002504, 0x00000010, 0x00000010))
314 if (!nv_wait(clk, 0x00251c, 0x0000003f, 0x0000003f))
321 nva3_clock_post(struct nouveau_clock *clk, unsigned long *flags)
323 struct nouveau_fifo *pfifo = nouveau_fifo(clk);
326 pfifo->start(pfifo, flags);
328 nv_mask(clk, 0x002504, 0x00000001, 0x00000000);
329 nv_mask(clk, 0x020060, 0x00070000, 0x00040000);
333 disable_clk_src(struct nva3_clock_priv *priv, u32 src)
335 nv_mask(priv, src, 0x00000100, 0x00000000);
336 nv_mask(priv, src, 0x00000001, 0x00000000);
340 prog_pll(struct nva3_clock_priv *priv, int clk, u32 pll, int idx)
342 struct nva3_clock_info *info = &priv->eng[idx];
343 const u32 src0 = 0x004120 + (clk * 4);
344 const u32 src1 = 0x004160 + (clk * 4);
345 const u32 ctrl = pll + 0;
346 const u32 coef = pll + 4;
350 /* Always start from a non-PLL clock */
351 bypass = nv_rd32(priv, ctrl) & 0x00000008;
353 nv_mask(priv, src1, 0x00000101, 0x00000101);
354 nv_mask(priv, ctrl, 0x00000008, 0x00000008);
358 nv_mask(priv, src0, 0x003f3141, 0x00000101 | info->clk);
359 nv_wr32(priv, coef, info->pll);
360 nv_mask(priv, ctrl, 0x00000015, 0x00000015);
361 nv_mask(priv, ctrl, 0x00000010, 0x00000000);
362 if (!nv_wait(priv, ctrl, 0x00020000, 0x00020000)) {
363 nv_mask(priv, ctrl, 0x00000010, 0x00000010);
364 nv_mask(priv, src0, 0x00000101, 0x00000000);
367 nv_mask(priv, ctrl, 0x00000010, 0x00000010);
368 nv_mask(priv, ctrl, 0x00000008, 0x00000000);
369 disable_clk_src(priv, src1);
371 nv_mask(priv, src1, 0x003f3141, 0x00000101 | info->clk);
372 nv_mask(priv, ctrl, 0x00000018, 0x00000018);
374 nv_mask(priv, ctrl, 0x00000001, 0x00000000);
375 disable_clk_src(priv, src0);
380 prog_clk(struct nva3_clock_priv *priv, int clk, int idx)
382 struct nva3_clock_info *info = &priv->eng[idx];
383 nv_mask(priv, 0x004120 + (clk * 4), 0x003f3141, 0x00000101 | info->clk);
387 prog_host(struct nva3_clock_priv *priv)
389 struct nva3_clock_info *info = &priv->eng[nv_clk_src_host];
390 u32 hsrc = (nv_rd32(priv, 0xc040));
392 switch (info->host_out) {
394 if ((hsrc & 0x30000000) == 0) {
395 nv_wr32(priv, 0xc040, hsrc | 0x20000000);
396 disable_clk_src(priv, 0x4194);
400 prog_clk(priv, 0x1d, nv_clk_src_host);
401 if ((hsrc & 0x30000000) >= 0x20000000) {
402 nv_wr32(priv, 0xc040, hsrc & ~0x30000000);
409 /* This seems to be a clock gating factor on idle, always set to 64 */
410 nv_wr32(priv, 0xc044, 0x3e);
414 prog_core(struct nva3_clock_priv *priv, int idx)
416 struct nva3_clock_info *info = &priv->eng[idx];
417 u32 fb_delay = nv_rd32(priv, 0x10002c);
419 if (fb_delay < info->fb_delay)
420 nv_wr32(priv, 0x10002c, info->fb_delay);
422 prog_pll(priv, 0x00, 0x004200, idx);
424 if (fb_delay > info->fb_delay)
425 nv_wr32(priv, 0x10002c, info->fb_delay);
429 nva3_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate)
431 struct nva3_clock_priv *priv = (void *)clk;
432 struct nva3_clock_info *core = &priv->eng[nv_clk_src_core];
435 if ((ret = calc_clk(priv, cstate, 0x10, 0x4200, nv_clk_src_core)) ||
436 (ret = calc_clk(priv, cstate, 0x11, 0x4220, nv_clk_src_shader)) ||
437 (ret = calc_clk(priv, cstate, 0x20, 0x0000, nv_clk_src_disp)) ||
438 (ret = calc_clk(priv, cstate, 0x21, 0x0000, nv_clk_src_vdec)) ||
439 (ret = calc_host(priv, cstate)))
442 /* XXX: Should be reading the highest bit in the VBIOS clock to decide
443 * whether to use a PLL or not... but using a PLL defeats the purpose */
445 ret = nva3_clk_info(clk, 0x10,
446 cstate->domain[nv_clk_src_core_intm],
447 &priv->eng[nv_clk_src_core_intm]);
456 nva3_clock_prog(struct nouveau_clock *clk)
458 struct nva3_clock_priv *priv = (void *)clk;
459 struct nva3_clock_info *core = &priv->eng[nv_clk_src_core];
462 unsigned long *f = &flags;
464 ret = nva3_clock_pre(clk, f);
469 prog_core(priv, nv_clk_src_core_intm);
471 prog_core(priv, nv_clk_src_core);
472 prog_pll(priv, 0x01, 0x004220, nv_clk_src_shader);
473 prog_clk(priv, 0x20, nv_clk_src_disp);
474 prog_clk(priv, 0x21, nv_clk_src_vdec);
481 nva3_clock_post(clk, f);
487 nva3_clock_tidy(struct nouveau_clock *clk)
491 static struct nouveau_clocks
493 { nv_clk_src_crystal , 0xff },
494 { nv_clk_src_core , 0x00, 0, "core", 1000 },
495 { nv_clk_src_shader , 0x01, 0, "shader", 1000 },
496 { nv_clk_src_mem , 0x02, 0, "memory", 1000 },
497 { nv_clk_src_vdec , 0x03 },
498 { nv_clk_src_disp , 0x04 },
499 { nv_clk_src_host , 0x05 },
500 { nv_clk_src_core_intm, 0x06 },
505 nva3_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
506 struct nouveau_oclass *oclass, void *data, u32 size,
507 struct nouveau_object **pobject)
509 struct nva3_clock_priv *priv;
512 ret = nouveau_clock_create(parent, engine, oclass, nva3_domain, NULL, 0,
514 *pobject = nv_object(priv);
518 priv->base.read = nva3_clock_read;
519 priv->base.calc = nva3_clock_calc;
520 priv->base.prog = nva3_clock_prog;
521 priv->base.tidy = nva3_clock_tidy;
525 struct nouveau_oclass
526 nva3_clock_oclass = {
527 .handle = NV_SUBDEV(CLOCK, 0xa3),
528 .ofuncs = &(struct nouveau_ofuncs) {
529 .ctor = nva3_clock_ctor,
530 .dtor = _nouveau_clock_dtor,
531 .init = _nouveau_clock_init,
532 .fini = _nouveau_clock_fini,