xref: /linux/drivers/cpufreq/powernow-k8.c (revision fab183d632628381b466a41479489541ac0e29a0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *   (c) 2003-2012 Advanced Micro Devices, Inc.
4  *
5  *  Maintainer:
6  *  Andreas Herrmann <herrmann.der.user@googlemail.com>
7  *
8  *  Based on the powernow-k7.c module written by Dave Jones.
9  *  (C) 2003 Dave Jones on behalf of SuSE Labs
10  *  (C) 2004 Dominik Brodowski <linux@brodo.de>
11  *  (C) 2004 Pavel Machek <pavel@ucw.cz>
12  *  Based upon datasheets & sample CPUs kindly provided by AMD.
13  *
14  *  Valuable input gratefully received from Dave Jones, Pavel Machek,
15  *  Dominik Brodowski, Jacob Shin, and others.
16  *  Originally developed by Paul Devriendt.
17  *
18  *  Processor information obtained from Chapter 9 (Power and Thermal
19  *  Management) of the "BIOS and Kernel Developer's Guide (BKDG) for
20  *  the AMD Athlon 64 and AMD Opteron Processors" and section "2.x
21  *  Power Management" in BKDGs for newer AMD CPU families.
22  *
23  *  Tables for specific CPUs can be inferred from AMD's processor
24  *  power and thermal data sheets, (e.g. 30417.pdf, 30430.pdf, 43375.pdf)
25  */
26 
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 
29 #include <linux/kernel.h>
30 #include <linux/smp.h>
31 #include <linux/module.h>
32 #include <linux/init.h>
33 #include <linux/cpufreq.h>
34 #include <linux/slab.h>
35 #include <linux/string.h>
36 #include <linux/cpumask.h>
37 #include <linux/io.h>
38 #include <linux/delay.h>
39 
40 #include <asm/msr.h>
41 #include <asm/cpu_device_id.h>
42 #include <asm/cpuid/api.h>
43 
44 #include <linux/acpi.h>
45 #include <linux/mutex.h>
46 #include <acpi/processor.h>
47 
48 #define VERSION "version 2.20.00"
49 #include "powernow-k8.h"
50 
51 /* serialize freq changes  */
52 static DEFINE_MUTEX(fidvid_mutex);
53 
54 static DEFINE_PER_CPU(struct powernow_k8_data *, powernow_data);
55 
56 static struct cpufreq_driver cpufreq_amd64_driver;
57 
58 /* Return a frequency in MHz, given an input fid */
find_freq_from_fid(u32 fid)59 static u32 find_freq_from_fid(u32 fid)
60 {
61 	return 800 + (fid * 100);
62 }
63 
64 /* Return a frequency in KHz, given an input fid */
find_khz_freq_from_fid(u32 fid)65 static u32 find_khz_freq_from_fid(u32 fid)
66 {
67 	return 1000 * find_freq_from_fid(fid);
68 }
69 
70 /* Return the vco fid for an input fid
71  *
72  * Each "low" fid has corresponding "high" fid, and you can get to "low" fids
73  * only from corresponding high fids. This returns "high" fid corresponding to
74  * "low" one.
75  */
convert_fid_to_vco_fid(u32 fid)76 static u32 convert_fid_to_vco_fid(u32 fid)
77 {
78 	if (fid < HI_FID_TABLE_BOTTOM)
79 		return 8 + (2 * fid);
80 	else
81 		return fid;
82 }
83 
84 /*
85  * Return 1 if the pending bit is set. Unless we just instructed the processor
86  * to transition to a new state, seeing this bit set is really bad news.
87  */
pending_bit_stuck(void)88 static int pending_bit_stuck(void)
89 {
90 	u64 msr;
91 
92 	rdmsrq(MSR_FIDVID_STATUS, msr);
93 	return msr & MSR_S_LO_CHANGE_PENDING ? 1 : 0;
94 }
95 
96 /*
97  * Update the global current fid / vid values from the status msr.
98  * Returns 1 on error.
99  */
query_current_values_with_pending_wait(struct powernow_k8_data * data)100 static int query_current_values_with_pending_wait(struct powernow_k8_data *data)
101 {
102 	struct msr msr;
103 	u32 i = 0;
104 
105 	do {
106 		if (i++ > 10000) {
107 			pr_debug("detected change pending stuck\n");
108 			return 1;
109 		}
110 		rdmsrq(MSR_FIDVID_STATUS, msr.q);
111 	} while (msr.l & MSR_S_LO_CHANGE_PENDING);
112 
113 	data->currvid = msr.h & MSR_S_HI_CURRENT_VID;
114 	data->currfid = msr.l & MSR_S_LO_CURRENT_FID;
115 
116 	return 0;
117 }
118 
119 /* the isochronous relief time */
count_off_irt(struct powernow_k8_data * data)120 static void count_off_irt(struct powernow_k8_data *data)
121 {
122 	udelay((1 << data->irt) * 10);
123 }
124 
125 /* the voltage stabilization time */
count_off_vst(struct powernow_k8_data * data)126 static void count_off_vst(struct powernow_k8_data *data)
127 {
128 	udelay(data->vstable * VST_UNITS_20US);
129 }
130 
131 /* need to init the control msr to a safe value (for each cpu) */
fidvid_msr_init(void)132 static void fidvid_msr_init(void)
133 {
134 	struct msr msr;
135 	u8 fid, vid;
136 
137 	rdmsrq(MSR_FIDVID_STATUS, msr.q);
138 	vid = msr.h & MSR_S_HI_CURRENT_VID;
139 	fid = msr.l & MSR_S_LO_CURRENT_FID;
140 	msr.l = fid | (vid << MSR_C_LO_VID_SHIFT);
141 	msr.h = MSR_C_HI_STP_GNT_BENIGN;
142 	pr_debug("cpu%d, init lo 0x%x, hi 0x%x\n", smp_processor_id(), msr.l, msr.h);
143 	wrmsrq(MSR_FIDVID_CTL, msr.q);
144 }
145 
146 /* write the new fid value along with the other control fields to the msr */
write_new_fid(struct powernow_k8_data * data,u32 fid)147 static int write_new_fid(struct powernow_k8_data *data, u32 fid)
148 {
149 	struct msr msr;
150 	u32 savevid = data->currvid;
151 	u32 i = 0;
152 
153 	if ((fid & INVALID_FID_MASK) || (data->currvid & INVALID_VID_MASK)) {
154 		pr_err("internal error - overflow on fid write\n");
155 		return 1;
156 	}
157 
158 	msr.l = fid;
159 	msr.l |= (data->currvid << MSR_C_LO_VID_SHIFT);
160 	msr.l |= MSR_C_LO_INIT_FID_VID;
161 	msr.h = data->plllock * PLL_LOCK_CONVERSION;
162 
163 	pr_debug("writing fid 0x%x, lo 0x%x, hi 0x%x\n", fid, msr.l, msr.h);
164 
165 	do {
166 		wrmsrq(MSR_FIDVID_CTL, msr.q);
167 		if (i++ > 100) {
168 			pr_err("Hardware error - pending bit very stuck - no further pstate changes possible\n");
169 			return 1;
170 		}
171 	} while (query_current_values_with_pending_wait(data));
172 
173 	count_off_irt(data);
174 
175 	if (savevid != data->currvid) {
176 		pr_err("vid change on fid trans, old 0x%x, new 0x%x\n",
177 		       savevid, data->currvid);
178 		return 1;
179 	}
180 
181 	if (fid != data->currfid) {
182 		pr_err("fid trans failed, fid 0x%x, curr 0x%x\n", fid,
183 			data->currfid);
184 		return 1;
185 	}
186 
187 	return 0;
188 }
189 
190 /* Write a new vid to the hardware */
write_new_vid(struct powernow_k8_data * data,u32 vid)191 static int write_new_vid(struct powernow_k8_data *data, u32 vid)
192 {
193 	struct msr msr;
194 	u32 savefid = data->currfid;
195 	int i = 0;
196 
197 	if ((data->currfid & INVALID_FID_MASK) || (vid & INVALID_VID_MASK)) {
198 		pr_err("internal error - overflow on vid write\n");
199 		return 1;
200 	}
201 
202 	msr.l = data->currfid;
203 	msr.l |= (vid << MSR_C_LO_VID_SHIFT);
204 	msr.l |= MSR_C_LO_INIT_FID_VID;
205 	msr.h = STOP_GRANT_5NS;
206 
207 	pr_debug("writing vid 0x%x, lo 0x%x, hi 0x%x\n", vid, msr.l, msr.h);
208 
209 	do {
210 		wrmsrq(MSR_FIDVID_CTL, msr.q);
211 		if (i++ > 100) {
212 			pr_err("internal error - pending bit very stuck - no further pstate changes possible\n");
213 			return 1;
214 		}
215 	} while (query_current_values_with_pending_wait(data));
216 
217 	if (savefid != data->currfid) {
218 		pr_err("fid changed on vid trans, old 0x%x new 0x%x\n",
219 			savefid, data->currfid);
220 		return 1;
221 	}
222 
223 	if (vid != data->currvid) {
224 		pr_err("vid trans failed, vid 0x%x, curr 0x%x\n",
225 				vid, data->currvid);
226 		return 1;
227 	}
228 
229 	return 0;
230 }
231 
232 /*
233  * Reduce the vid by the max of step or reqvid.
234  * Decreasing vid codes represent increasing voltages:
235  * vid of 0 is 1.550V, vid of 0x1e is 0.800V, vid of VID_OFF is off.
236  */
decrease_vid_code_by_step(struct powernow_k8_data * data,u32 reqvid,u32 step)237 static int decrease_vid_code_by_step(struct powernow_k8_data *data,
238 		u32 reqvid, u32 step)
239 {
240 	if ((data->currvid - reqvid) > step)
241 		reqvid = data->currvid - step;
242 
243 	if (write_new_vid(data, reqvid))
244 		return 1;
245 
246 	count_off_vst(data);
247 
248 	return 0;
249 }
250 
251 /* Change Opteron/Athlon64 fid and vid, by the 3 phases. */
transition_fid_vid(struct powernow_k8_data * data,u32 reqfid,u32 reqvid)252 static int transition_fid_vid(struct powernow_k8_data *data,
253 		u32 reqfid, u32 reqvid)
254 {
255 	if (core_voltage_pre_transition(data, reqvid, reqfid))
256 		return 1;
257 
258 	if (core_frequency_transition(data, reqfid))
259 		return 1;
260 
261 	if (core_voltage_post_transition(data, reqvid))
262 		return 1;
263 
264 	if (query_current_values_with_pending_wait(data))
265 		return 1;
266 
267 	if ((reqfid != data->currfid) || (reqvid != data->currvid)) {
268 		pr_err("failed (cpu%d): req 0x%x 0x%x, curr 0x%x 0x%x\n",
269 				smp_processor_id(),
270 				reqfid, reqvid, data->currfid, data->currvid);
271 		return 1;
272 	}
273 
274 	pr_debug("transitioned (cpu%d): new fid 0x%x, vid 0x%x\n",
275 		smp_processor_id(), data->currfid, data->currvid);
276 
277 	return 0;
278 }
279 
280 /* Phase 1 - core voltage transition ... setup voltage */
core_voltage_pre_transition(struct powernow_k8_data * data,u32 reqvid,u32 reqfid)281 static int core_voltage_pre_transition(struct powernow_k8_data *data,
282 		u32 reqvid, u32 reqfid)
283 {
284 	struct msr msr;
285 	u32 rvosteps = data->rvo;
286 	u32 savefid = data->currfid;
287 	u32 maxvid, rvomult = 1;
288 
289 	pr_debug("ph1 (cpu%d): start, currfid 0x%x, currvid 0x%x, reqvid 0x%x, rvo 0x%x\n",
290 		smp_processor_id(),
291 		data->currfid, data->currvid, reqvid, data->rvo);
292 
293 	if ((savefid < LO_FID_TABLE_TOP) && (reqfid < LO_FID_TABLE_TOP))
294 		rvomult = 2;
295 	rvosteps *= rvomult;
296 	rdmsrq(MSR_FIDVID_STATUS, msr.q);
297 	maxvid = 0x1f & (msr.h >> 16);
298 	pr_debug("ph1 maxvid=0x%x\n", maxvid);
299 	if (reqvid < maxvid) /* lower numbers are higher voltages */
300 		reqvid = maxvid;
301 
302 	while (data->currvid > reqvid) {
303 		pr_debug("ph1: curr 0x%x, req vid 0x%x\n",
304 			data->currvid, reqvid);
305 		if (decrease_vid_code_by_step(data, reqvid, data->vidmvs))
306 			return 1;
307 	}
308 
309 	while ((rvosteps > 0) &&
310 			((rvomult * data->rvo + data->currvid) > reqvid)) {
311 		if (data->currvid == maxvid) {
312 			rvosteps = 0;
313 		} else {
314 			pr_debug("ph1: changing vid for rvo, req 0x%x\n",
315 				data->currvid - 1);
316 			if (decrease_vid_code_by_step(data, data->currvid-1, 1))
317 				return 1;
318 			rvosteps--;
319 		}
320 	}
321 
322 	if (query_current_values_with_pending_wait(data))
323 		return 1;
324 
325 	if (savefid != data->currfid) {
326 		pr_err("ph1 err, currfid changed 0x%x\n", data->currfid);
327 		return 1;
328 	}
329 
330 	pr_debug("ph1 complete, currfid 0x%x, currvid 0x%x\n",
331 		data->currfid, data->currvid);
332 
333 	return 0;
334 }
335 
336 /* Phase 2 - core frequency transition */
core_frequency_transition(struct powernow_k8_data * data,u32 reqfid)337 static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid)
338 {
339 	u32 vcoreqfid, vcocurrfid, vcofiddiff;
340 	u32 fid_interval, savevid = data->currvid;
341 
342 	if (data->currfid == reqfid) {
343 		pr_err("ph2 null fid transition 0x%x\n", data->currfid);
344 		return 0;
345 	}
346 
347 	pr_debug("ph2 (cpu%d): starting, currfid 0x%x, currvid 0x%x, reqfid 0x%x\n",
348 		smp_processor_id(),
349 		data->currfid, data->currvid, reqfid);
350 
351 	vcoreqfid = convert_fid_to_vco_fid(reqfid);
352 	vcocurrfid = convert_fid_to_vco_fid(data->currfid);
353 	vcofiddiff = vcocurrfid > vcoreqfid ? vcocurrfid - vcoreqfid
354 	    : vcoreqfid - vcocurrfid;
355 
356 	if ((reqfid <= LO_FID_TABLE_TOP) && (data->currfid <= LO_FID_TABLE_TOP))
357 		vcofiddiff = 0;
358 
359 	while (vcofiddiff > 2) {
360 		(data->currfid & 1) ? (fid_interval = 1) : (fid_interval = 2);
361 
362 		if (reqfid > data->currfid) {
363 			if (data->currfid > LO_FID_TABLE_TOP) {
364 				if (write_new_fid(data,
365 						data->currfid + fid_interval))
366 					return 1;
367 			} else {
368 				if (write_new_fid
369 				    (data,
370 				     2 + convert_fid_to_vco_fid(data->currfid)))
371 					return 1;
372 			}
373 		} else {
374 			if (write_new_fid(data, data->currfid - fid_interval))
375 				return 1;
376 		}
377 
378 		vcocurrfid = convert_fid_to_vco_fid(data->currfid);
379 		vcofiddiff = vcocurrfid > vcoreqfid ? vcocurrfid - vcoreqfid
380 		    : vcoreqfid - vcocurrfid;
381 	}
382 
383 	if (write_new_fid(data, reqfid))
384 		return 1;
385 
386 	if (query_current_values_with_pending_wait(data))
387 		return 1;
388 
389 	if (data->currfid != reqfid) {
390 		pr_err("ph2: mismatch, failed fid transition, curr 0x%x, req 0x%x\n",
391 			data->currfid, reqfid);
392 		return 1;
393 	}
394 
395 	if (savevid != data->currvid) {
396 		pr_err("ph2: vid changed, save 0x%x, curr 0x%x\n",
397 			savevid, data->currvid);
398 		return 1;
399 	}
400 
401 	pr_debug("ph2 complete, currfid 0x%x, currvid 0x%x\n",
402 		data->currfid, data->currvid);
403 
404 	return 0;
405 }
406 
407 /* Phase 3 - core voltage transition flow ... jump to the final vid. */
core_voltage_post_transition(struct powernow_k8_data * data,u32 reqvid)408 static int core_voltage_post_transition(struct powernow_k8_data *data,
409 		u32 reqvid)
410 {
411 	u32 savefid = data->currfid;
412 	u32 savereqvid = reqvid;
413 
414 	pr_debug("ph3 (cpu%d): starting, currfid 0x%x, currvid 0x%x\n",
415 		smp_processor_id(),
416 		data->currfid, data->currvid);
417 
418 	if (reqvid != data->currvid) {
419 		if (write_new_vid(data, reqvid))
420 			return 1;
421 
422 		if (savefid != data->currfid) {
423 			pr_err("ph3: bad fid change, save 0x%x, curr 0x%x\n",
424 				savefid, data->currfid);
425 			return 1;
426 		}
427 
428 		if (data->currvid != reqvid) {
429 			pr_err("ph3: failed vid transition\n, req 0x%x, curr 0x%x",
430 				reqvid, data->currvid);
431 			return 1;
432 		}
433 	}
434 
435 	if (query_current_values_with_pending_wait(data))
436 		return 1;
437 
438 	if (savereqvid != data->currvid) {
439 		pr_debug("ph3 failed, currvid 0x%x\n", data->currvid);
440 		return 1;
441 	}
442 
443 	if (savefid != data->currfid) {
444 		pr_debug("ph3 failed, currfid changed 0x%x\n",
445 			data->currfid);
446 		return 1;
447 	}
448 
449 	pr_debug("ph3 complete, currfid 0x%x, currvid 0x%x\n",
450 		data->currfid, data->currvid);
451 
452 	return 0;
453 }
454 
455 static const struct x86_cpu_id powernow_k8_ids[] = {
456 	/* IO based frequency switching */
457 	X86_MATCH_VENDOR_FAM(AMD, 0xf, NULL),
458 	{}
459 };
460 MODULE_DEVICE_TABLE(x86cpu, powernow_k8_ids);
461 
check_supported_cpu(void * _rc)462 static void check_supported_cpu(void *_rc)
463 {
464 	u32 eax, ebx, ecx, edx;
465 	int *rc = _rc;
466 
467 	*rc = -ENODEV;
468 
469 	eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
470 
471 	if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) {
472 		if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) ||
473 		    ((eax & CPUID_XMOD) > CPUID_XMOD_REV_MASK)) {
474 			pr_info("Processor cpuid %x not supported\n", eax);
475 			return;
476 		}
477 
478 		eax = cpuid_eax(CPUID_GET_MAX_CAPABILITIES);
479 		if (eax < CPUID_FREQ_VOLT_CAPABILITIES) {
480 			pr_info("No frequency change capabilities detected\n");
481 			return;
482 		}
483 
484 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
485 		if ((edx & P_STATE_TRANSITION_CAPABLE)
486 			!= P_STATE_TRANSITION_CAPABLE) {
487 			pr_info_once("Power state transitions not supported\n");
488 			return;
489 		}
490 		*rc = 0;
491 	}
492 }
493 
check_pst_table(struct powernow_k8_data * data,struct pst_s * pst,u8 maxvid)494 static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst,
495 		u8 maxvid)
496 {
497 	unsigned int j;
498 	u8 lastfid = 0xff;
499 
500 	for (j = 0; j < data->numps; j++) {
501 		if (pst[j].vid > LEAST_VID) {
502 			pr_err(FW_BUG "vid %d invalid : 0x%x\n", j,
503 				pst[j].vid);
504 			return -EINVAL;
505 		}
506 		if (pst[j].vid < data->rvo) {
507 			/* vid + rvo >= 0 */
508 			pr_err(FW_BUG "0 vid exceeded with pstate %d\n", j);
509 			return -ENODEV;
510 		}
511 		if (pst[j].vid < maxvid + data->rvo) {
512 			/* vid + rvo >= maxvid */
513 			pr_err(FW_BUG "maxvid exceeded with pstate %d\n", j);
514 			return -ENODEV;
515 		}
516 		if (pst[j].fid > MAX_FID) {
517 			pr_err(FW_BUG "maxfid exceeded with pstate %d\n", j);
518 			return -ENODEV;
519 		}
520 		if (j && (pst[j].fid < HI_FID_TABLE_BOTTOM)) {
521 			/* Only first fid is allowed to be in "low" range */
522 			pr_err(FW_BUG "two low fids - %d : 0x%x\n", j,
523 				pst[j].fid);
524 			return -EINVAL;
525 		}
526 		if (pst[j].fid < lastfid)
527 			lastfid = pst[j].fid;
528 	}
529 	if (lastfid & 1) {
530 		pr_err(FW_BUG "lastfid invalid\n");
531 		return -EINVAL;
532 	}
533 	if (lastfid > LO_FID_TABLE_TOP)
534 		pr_info(FW_BUG "first fid not from lo freq table\n");
535 
536 	return 0;
537 }
538 
invalidate_entry(struct cpufreq_frequency_table * powernow_table,unsigned int entry)539 static void invalidate_entry(struct cpufreq_frequency_table *powernow_table,
540 		unsigned int entry)
541 {
542 	powernow_table[entry].frequency = CPUFREQ_ENTRY_INVALID;
543 }
544 
print_basics(struct powernow_k8_data * data)545 static void print_basics(struct powernow_k8_data *data)
546 {
547 	int j;
548 	for (j = 0; j < data->numps; j++) {
549 		if (data->powernow_table[j].frequency !=
550 				CPUFREQ_ENTRY_INVALID) {
551 			pr_info("fid 0x%x (%d MHz), vid 0x%x\n",
552 				data->powernow_table[j].driver_data & 0xff,
553 				data->powernow_table[j].frequency/1000,
554 				data->powernow_table[j].driver_data >> 8);
555 		}
556 	}
557 	if (data->batps)
558 		pr_info("Only %d pstates on battery\n", data->batps);
559 }
560 
fill_powernow_table(struct powernow_k8_data * data,struct pst_s * pst,u8 maxvid)561 static int fill_powernow_table(struct powernow_k8_data *data,
562 		struct pst_s *pst, u8 maxvid)
563 {
564 	struct cpufreq_frequency_table *powernow_table;
565 	unsigned int j;
566 
567 	if (data->batps) {
568 		/* use ACPI support to get full speed on mains power */
569 		pr_warn("Only %d pstates usable (use ACPI driver for full range\n",
570 			data->batps);
571 		data->numps = data->batps;
572 	}
573 
574 	for (j = 1; j < data->numps; j++) {
575 		if (pst[j-1].fid >= pst[j].fid) {
576 			pr_err("PST out of sequence\n");
577 			return -EINVAL;
578 		}
579 	}
580 
581 	if (data->numps < 2) {
582 		pr_err("no p states to transition\n");
583 		return -ENODEV;
584 	}
585 
586 	if (check_pst_table(data, pst, maxvid))
587 		return -EINVAL;
588 
589 	powernow_table = kzalloc((sizeof(*powernow_table)
590 		* (data->numps + 1)), GFP_KERNEL);
591 	if (!powernow_table)
592 		return -ENOMEM;
593 
594 	for (j = 0; j < data->numps; j++) {
595 		int freq;
596 		powernow_table[j].driver_data = pst[j].fid; /* lower 8 bits */
597 		powernow_table[j].driver_data |= (pst[j].vid << 8); /* upper 8 bits */
598 		freq = find_khz_freq_from_fid(pst[j].fid);
599 		powernow_table[j].frequency = freq;
600 	}
601 	powernow_table[data->numps].frequency = CPUFREQ_TABLE_END;
602 	powernow_table[data->numps].driver_data = 0;
603 
604 	if (query_current_values_with_pending_wait(data)) {
605 		kfree(powernow_table);
606 		return -EIO;
607 	}
608 
609 	pr_debug("cfid 0x%x, cvid 0x%x\n", data->currfid, data->currvid);
610 	data->powernow_table = powernow_table;
611 	if (cpumask_first(topology_core_cpumask(data->cpu)) == data->cpu)
612 		print_basics(data);
613 
614 	for (j = 0; j < data->numps; j++)
615 		if ((pst[j].fid == data->currfid) &&
616 		    (pst[j].vid == data->currvid))
617 			return 0;
618 
619 	pr_debug("currfid/vid do not match PST, ignoring\n");
620 	return 0;
621 }
622 
623 /* Find and validate the PSB/PST table in BIOS. */
find_psb_table(struct powernow_k8_data * data)624 static int find_psb_table(struct powernow_k8_data *data)
625 {
626 	struct psb_s *psb;
627 	unsigned int i;
628 	u32 mvs;
629 	u8 maxvid;
630 	u32 cpst = 0;
631 	u32 thiscpuid;
632 
633 	for (i = 0xc0000; i < 0xffff0; i += 0x10) {
634 		/* Scan BIOS looking for the signature. */
635 		/* It can not be at ffff0 - it is too big. */
636 
637 		psb = phys_to_virt(i);
638 		if (memcmp(psb, PSB_ID_STRING, PSB_ID_STRING_LEN) != 0)
639 			continue;
640 
641 		pr_debug("found PSB header at 0x%p\n", psb);
642 
643 		pr_debug("table vers: 0x%x\n", psb->tableversion);
644 		if (psb->tableversion != PSB_VERSION_1_4) {
645 			pr_err(FW_BUG "PSB table is not v1.4\n");
646 			return -ENODEV;
647 		}
648 
649 		pr_debug("flags: 0x%x\n", psb->flags1);
650 		if (psb->flags1) {
651 			pr_err(FW_BUG "unknown flags\n");
652 			return -ENODEV;
653 		}
654 
655 		data->vstable = psb->vstable;
656 		pr_debug("voltage stabilization time: %d(*20us)\n",
657 				data->vstable);
658 
659 		pr_debug("flags2: 0x%x\n", psb->flags2);
660 		data->rvo = psb->flags2 & 3;
661 		data->irt = ((psb->flags2) >> 2) & 3;
662 		mvs = ((psb->flags2) >> 4) & 3;
663 		data->vidmvs = 1 << mvs;
664 		data->batps = ((psb->flags2) >> 6) & 3;
665 
666 		pr_debug("ramp voltage offset: %d\n", data->rvo);
667 		pr_debug("isochronous relief time: %d\n", data->irt);
668 		pr_debug("maximum voltage step: %d - 0x%x\n", mvs, data->vidmvs);
669 
670 		pr_debug("numpst: 0x%x\n", psb->num_tables);
671 		cpst = psb->num_tables;
672 		if ((psb->cpuid == 0x00000fc0) ||
673 		    (psb->cpuid == 0x00000fe0)) {
674 			thiscpuid = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
675 			if ((thiscpuid == 0x00000fc0) ||
676 			    (thiscpuid == 0x00000fe0))
677 				cpst = 1;
678 		}
679 		if (cpst != 1) {
680 			pr_err(FW_BUG "numpst must be 1\n");
681 			return -ENODEV;
682 		}
683 
684 		data->plllock = psb->plllocktime;
685 		pr_debug("plllocktime: 0x%x (units 1us)\n", psb->plllocktime);
686 		pr_debug("maxfid: 0x%x\n", psb->maxfid);
687 		pr_debug("maxvid: 0x%x\n", psb->maxvid);
688 		maxvid = psb->maxvid;
689 
690 		data->numps = psb->numps;
691 		pr_debug("numpstates: 0x%x\n", data->numps);
692 		return fill_powernow_table(data,
693 				(struct pst_s *)(psb+1), maxvid);
694 	}
695 	/*
696 	 * If you see this message, complain to BIOS manufacturer. If
697 	 * he tells you "we do not support Linux" or some similar
698 	 * nonsense, remember that Windows 2000 uses the same legacy
699 	 * mechanism that the old Linux PSB driver uses. Tell them it
700 	 * is broken with Windows 2000.
701 	 *
702 	 * The reference to the AMD documentation is chapter 9 in the
703 	 * BIOS and Kernel Developer's Guide, which is available on
704 	 * www.amd.com
705 	 */
706 	pr_err(FW_BUG "No PSB or ACPI _PSS objects\n");
707 	pr_err("Make sure that your BIOS is up to date and Cool'N'Quiet support is enabled in BIOS setup\n");
708 	return -ENODEV;
709 }
710 
powernow_k8_acpi_pst_values(struct powernow_k8_data * data,unsigned int index)711 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data,
712 		unsigned int index)
713 {
714 	u64 control;
715 
716 	if (!data->acpi_data.state_count)
717 		return;
718 
719 	control = data->acpi_data.states[index].control;
720 	data->irt = (control >> IRT_SHIFT) & IRT_MASK;
721 	data->rvo = (control >> RVO_SHIFT) & RVO_MASK;
722 	data->exttype = (control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
723 	data->plllock = (control >> PLL_L_SHIFT) & PLL_L_MASK;
724 	data->vidmvs = 1 << ((control >> MVS_SHIFT) & MVS_MASK);
725 	data->vstable = (control >> VST_SHIFT) & VST_MASK;
726 }
727 
powernow_k8_cpu_init_acpi(struct powernow_k8_data * data)728 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
729 {
730 	struct cpufreq_frequency_table *powernow_table;
731 	int ret_val = -ENODEV;
732 	u64 control, status;
733 
734 	if (acpi_processor_register_performance(&data->acpi_data, data->cpu)) {
735 		pr_debug("register performance failed: bad ACPI data\n");
736 		return -EIO;
737 	}
738 
739 	/* verify the data contained in the ACPI structures */
740 	if (data->acpi_data.state_count <= 1) {
741 		pr_debug("No ACPI P-States\n");
742 		goto err_out;
743 	}
744 
745 	control = data->acpi_data.control_register.space_id;
746 	status = data->acpi_data.status_register.space_id;
747 
748 	if ((control != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
749 	    (status != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
750 		pr_debug("Invalid control/status registers (%llx - %llx)\n",
751 			control, status);
752 		goto err_out;
753 	}
754 
755 	/* fill in data->powernow_table */
756 	powernow_table = kzalloc((sizeof(*powernow_table)
757 		* (data->acpi_data.state_count + 1)), GFP_KERNEL);
758 	if (!powernow_table)
759 		goto err_out;
760 
761 	/* fill in data */
762 	data->numps = data->acpi_data.state_count;
763 	powernow_k8_acpi_pst_values(data, 0);
764 
765 	ret_val = fill_powernow_table_fidvid(data, powernow_table);
766 	if (ret_val)
767 		goto err_out_mem;
768 
769 	powernow_table[data->acpi_data.state_count].frequency =
770 		CPUFREQ_TABLE_END;
771 	data->powernow_table = powernow_table;
772 
773 	if (cpumask_first(topology_core_cpumask(data->cpu)) == data->cpu)
774 		print_basics(data);
775 
776 	/* notify BIOS that we exist */
777 	acpi_processor_notify_smm(THIS_MODULE);
778 
779 	if (!zalloc_cpumask_var(&data->acpi_data.shared_cpu_map, GFP_KERNEL)) {
780 		pr_err("unable to alloc powernow_k8_data cpumask\n");
781 		ret_val = -ENOMEM;
782 		goto err_out_mem;
783 	}
784 
785 	return 0;
786 
787 err_out_mem:
788 	kfree(powernow_table);
789 
790 err_out:
791 	acpi_processor_unregister_performance(data->cpu);
792 
793 	/* data->acpi_data.state_count informs us at ->exit()
794 	 * whether ACPI was used */
795 	data->acpi_data.state_count = 0;
796 
797 	return ret_val;
798 }
799 
fill_powernow_table_fidvid(struct powernow_k8_data * data,struct cpufreq_frequency_table * powernow_table)800 static int fill_powernow_table_fidvid(struct powernow_k8_data *data,
801 		struct cpufreq_frequency_table *powernow_table)
802 {
803 	int i;
804 
805 	for (i = 0; i < data->acpi_data.state_count; i++) {
806 		u32 fid;
807 		u32 vid;
808 		u32 freq, index;
809 		u64 status, control;
810 
811 		if (data->exttype) {
812 			status =  data->acpi_data.states[i].status;
813 			fid = status & EXT_FID_MASK;
814 			vid = (status >> VID_SHIFT) & EXT_VID_MASK;
815 		} else {
816 			control =  data->acpi_data.states[i].control;
817 			fid = control & FID_MASK;
818 			vid = (control >> VID_SHIFT) & VID_MASK;
819 		}
820 
821 		pr_debug("   %d : fid 0x%x, vid 0x%x\n", i, fid, vid);
822 
823 		index = fid | (vid<<8);
824 		powernow_table[i].driver_data = index;
825 
826 		freq = find_khz_freq_from_fid(fid);
827 		powernow_table[i].frequency = freq;
828 
829 		/* verify frequency is OK */
830 		if ((freq > (MAX_FREQ * 1000)) || (freq < (MIN_FREQ * 1000))) {
831 			pr_debug("invalid freq %u kHz, ignoring\n", freq);
832 			invalidate_entry(powernow_table, i);
833 			continue;
834 		}
835 
836 		/* verify voltage is OK -
837 		 * BIOSs are using "off" to indicate invalid */
838 		if (vid == VID_OFF) {
839 			pr_debug("invalid vid %u, ignoring\n", vid);
840 			invalidate_entry(powernow_table, i);
841 			continue;
842 		}
843 
844 		if (freq != (data->acpi_data.states[i].core_frequency * 1000)) {
845 			pr_info("invalid freq entries %u kHz vs. %u kHz\n",
846 				freq, (unsigned int)
847 				(data->acpi_data.states[i].core_frequency
848 				 * 1000));
849 			invalidate_entry(powernow_table, i);
850 			continue;
851 		}
852 	}
853 	return 0;
854 }
855 
powernow_k8_cpu_exit_acpi(struct powernow_k8_data * data)856 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data)
857 {
858 	if (data->acpi_data.state_count)
859 		acpi_processor_unregister_performance(data->cpu);
860 	free_cpumask_var(data->acpi_data.shared_cpu_map);
861 }
862 
get_transition_latency(struct powernow_k8_data * data)863 static int get_transition_latency(struct powernow_k8_data *data)
864 {
865 	int max_latency = 0;
866 	int i;
867 	for (i = 0; i < data->acpi_data.state_count; i++) {
868 		int cur_latency = data->acpi_data.states[i].transition_latency
869 			+ data->acpi_data.states[i].bus_master_latency;
870 		if (cur_latency > max_latency)
871 			max_latency = cur_latency;
872 	}
873 	if (max_latency == 0) {
874 		pr_err(FW_WARN "Invalid zero transition latency\n");
875 		max_latency = 1;
876 	}
877 	/* value in usecs, needs to be in nanoseconds */
878 	return 1000 * max_latency;
879 }
880 
881 /* Take a frequency, and issue the fid/vid transition command */
transition_frequency_fidvid(struct powernow_k8_data * data,unsigned int index,struct cpufreq_policy * policy)882 static int transition_frequency_fidvid(struct powernow_k8_data *data,
883 		unsigned int index,
884 		struct cpufreq_policy *policy)
885 {
886 	u32 fid = 0;
887 	u32 vid = 0;
888 	int res;
889 	struct cpufreq_freqs freqs;
890 
891 	pr_debug("cpu %d transition to index %u\n", smp_processor_id(), index);
892 
893 	/* fid/vid correctness check for k8 */
894 	/* fid are the lower 8 bits of the index we stored into
895 	 * the cpufreq frequency table in find_psb_table, vid
896 	 * are the upper 8 bits.
897 	 */
898 	fid = data->powernow_table[index].driver_data & 0xFF;
899 	vid = (data->powernow_table[index].driver_data & 0xFF00) >> 8;
900 
901 	pr_debug("table matched fid 0x%x, giving vid 0x%x\n", fid, vid);
902 
903 	if (query_current_values_with_pending_wait(data))
904 		return 1;
905 
906 	if ((data->currvid == vid) && (data->currfid == fid)) {
907 		pr_debug("target matches current values (fid 0x%x, vid 0x%x)\n",
908 			fid, vid);
909 		return 0;
910 	}
911 
912 	pr_debug("cpu %d, changing to fid 0x%x, vid 0x%x\n",
913 		smp_processor_id(), fid, vid);
914 	freqs.old = find_khz_freq_from_fid(data->currfid);
915 	freqs.new = find_khz_freq_from_fid(fid);
916 
917 	cpufreq_freq_transition_begin(policy, &freqs);
918 	res = transition_fid_vid(data, fid, vid);
919 	cpufreq_freq_transition_end(policy, &freqs, res);
920 
921 	return res;
922 }
923 
924 struct powernowk8_target_arg {
925 	struct cpufreq_policy		*pol;
926 	unsigned			newstate;
927 };
928 
powernowk8_target_fn(void * arg)929 static long powernowk8_target_fn(void *arg)
930 {
931 	struct powernowk8_target_arg *pta = arg;
932 	struct cpufreq_policy *pol = pta->pol;
933 	unsigned newstate = pta->newstate;
934 	struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
935 	u32 checkfid;
936 	u32 checkvid;
937 	int ret;
938 
939 	if (!data)
940 		return -EINVAL;
941 
942 	checkfid = data->currfid;
943 	checkvid = data->currvid;
944 
945 	if (pending_bit_stuck()) {
946 		pr_err("failing targ, change pending bit set\n");
947 		return -EIO;
948 	}
949 
950 	pr_debug("targ: cpu %d, %d kHz, min %d, max %d\n",
951 		pol->cpu, data->powernow_table[newstate].frequency, pol->min,
952 		pol->max);
953 
954 	if (query_current_values_with_pending_wait(data))
955 		return -EIO;
956 
957 	pr_debug("targ: curr fid 0x%x, vid 0x%x\n",
958 		data->currfid, data->currvid);
959 
960 	if ((checkvid != data->currvid) ||
961 	    (checkfid != data->currfid)) {
962 		pr_info("error - out of sync, fix 0x%x 0x%x, vid 0x%x 0x%x\n",
963 		       checkfid, data->currfid,
964 		       checkvid, data->currvid);
965 	}
966 
967 	mutex_lock(&fidvid_mutex);
968 
969 	powernow_k8_acpi_pst_values(data, newstate);
970 
971 	ret = transition_frequency_fidvid(data, newstate, pol);
972 
973 	if (ret) {
974 		pr_err("transition frequency failed\n");
975 		mutex_unlock(&fidvid_mutex);
976 		return 1;
977 	}
978 	mutex_unlock(&fidvid_mutex);
979 
980 	pol->cur = find_khz_freq_from_fid(data->currfid);
981 
982 	return 0;
983 }
984 
985 /* Driver entry point to switch to the target frequency */
powernowk8_target(struct cpufreq_policy * pol,unsigned index)986 static int powernowk8_target(struct cpufreq_policy *pol, unsigned index)
987 {
988 	struct powernowk8_target_arg pta = { .pol = pol, .newstate = index };
989 
990 	return work_on_cpu(pol->cpu, powernowk8_target_fn, &pta);
991 }
992 
993 struct init_on_cpu {
994 	struct powernow_k8_data *data;
995 	int rc;
996 };
997 
powernowk8_cpu_init_on_cpu(void * _init_on_cpu)998 static void powernowk8_cpu_init_on_cpu(void *_init_on_cpu)
999 {
1000 	struct init_on_cpu *init_on_cpu = _init_on_cpu;
1001 
1002 	if (pending_bit_stuck()) {
1003 		pr_err("failing init, change pending bit set\n");
1004 		init_on_cpu->rc = -ENODEV;
1005 		return;
1006 	}
1007 
1008 	if (query_current_values_with_pending_wait(init_on_cpu->data)) {
1009 		init_on_cpu->rc = -ENODEV;
1010 		return;
1011 	}
1012 
1013 	fidvid_msr_init();
1014 
1015 	init_on_cpu->rc = 0;
1016 }
1017 
1018 #define MISSING_PSS_MSG \
1019 	FW_BUG "No compatible ACPI _PSS objects found.\n" \
1020 	FW_BUG "First, make sure Cool'N'Quiet is enabled in the BIOS.\n" \
1021 	FW_BUG "If that doesn't help, try upgrading your BIOS.\n"
1022 
1023 /* per CPU init entry point to the driver */
powernowk8_cpu_init(struct cpufreq_policy * pol)1024 static int powernowk8_cpu_init(struct cpufreq_policy *pol)
1025 {
1026 	struct powernow_k8_data *data;
1027 	struct init_on_cpu init_on_cpu;
1028 	int rc, cpu;
1029 
1030 	smp_call_function_single(pol->cpu, check_supported_cpu, &rc, 1);
1031 	if (rc)
1032 		return -ENODEV;
1033 
1034 	data = kzalloc_obj(*data);
1035 	if (!data)
1036 		return -ENOMEM;
1037 
1038 	data->cpu = pol->cpu;
1039 
1040 	if (powernow_k8_cpu_init_acpi(data)) {
1041 		/*
1042 		 * Use the PSB BIOS structure. This is only available on
1043 		 * an UP version, and is deprecated by AMD.
1044 		 */
1045 		if (num_online_cpus() != 1) {
1046 			pr_err_once(MISSING_PSS_MSG);
1047 			goto err_out;
1048 		}
1049 		if (pol->cpu != 0) {
1050 			pr_err(FW_BUG "No ACPI _PSS objects for CPU other than CPU0. Complain to your BIOS vendor.\n");
1051 			goto err_out;
1052 		}
1053 		rc = find_psb_table(data);
1054 		if (rc)
1055 			goto err_out;
1056 
1057 		/* Take a crude guess here.
1058 		 * That guess was in microseconds, so multiply with 1000 */
1059 		pol->cpuinfo.transition_latency = (
1060 			 ((data->rvo + 8) * data->vstable * VST_UNITS_20US) +
1061 			 ((1 << data->irt) * 30)) * 1000;
1062 	} else /* ACPI _PSS objects available */
1063 		pol->cpuinfo.transition_latency = get_transition_latency(data);
1064 
1065 	/* only run on specific CPU from here on */
1066 	init_on_cpu.data = data;
1067 	smp_call_function_single(data->cpu, powernowk8_cpu_init_on_cpu,
1068 				 &init_on_cpu, 1);
1069 	rc = init_on_cpu.rc;
1070 	if (rc != 0)
1071 		goto err_out_exit_acpi;
1072 
1073 	cpumask_copy(pol->cpus, topology_core_cpumask(pol->cpu));
1074 	data->available_cores = pol->cpus;
1075 	pol->freq_table = data->powernow_table;
1076 
1077 	pr_debug("cpu_init done, current fid 0x%x, vid 0x%x\n",
1078 		data->currfid, data->currvid);
1079 
1080 	/* Point all the CPUs in this policy to the same data */
1081 	for_each_cpu(cpu, pol->cpus)
1082 		per_cpu(powernow_data, cpu) = data;
1083 
1084 	return 0;
1085 
1086 err_out_exit_acpi:
1087 	powernow_k8_cpu_exit_acpi(data);
1088 	kfree(data->powernow_table);
1089 
1090 err_out:
1091 	kfree(data);
1092 	return -ENODEV;
1093 }
1094 
powernowk8_cpu_exit(struct cpufreq_policy * pol)1095 static void powernowk8_cpu_exit(struct cpufreq_policy *pol)
1096 {
1097 	struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1098 	int cpu;
1099 
1100 	if (!data)
1101 		return;
1102 
1103 	powernow_k8_cpu_exit_acpi(data);
1104 
1105 	kfree(data->powernow_table);
1106 	kfree(data);
1107 	/* pol->cpus will be empty here, use related_cpus instead. */
1108 	for_each_cpu(cpu, pol->related_cpus)
1109 		per_cpu(powernow_data, cpu) = NULL;
1110 }
1111 
query_values_on_cpu(void * _err)1112 static void query_values_on_cpu(void *_err)
1113 {
1114 	int *err = _err;
1115 	struct powernow_k8_data *data = __this_cpu_read(powernow_data);
1116 
1117 	*err = query_current_values_with_pending_wait(data);
1118 }
1119 
powernowk8_get(unsigned int cpu)1120 static unsigned int powernowk8_get(unsigned int cpu)
1121 {
1122 	struct powernow_k8_data *data = per_cpu(powernow_data, cpu);
1123 	unsigned int khz = 0;
1124 	int err;
1125 
1126 	if (!data)
1127 		return 0;
1128 
1129 	smp_call_function_single(cpu, query_values_on_cpu, &err, true);
1130 	if (err)
1131 		goto out;
1132 
1133 	khz = find_khz_freq_from_fid(data->currfid);
1134 
1135 
1136 out:
1137 	return khz;
1138 }
1139 
1140 static struct cpufreq_driver cpufreq_amd64_driver = {
1141 	.flags		= CPUFREQ_ASYNC_NOTIFICATION,
1142 	.verify		= cpufreq_generic_frequency_table_verify,
1143 	.target_index	= powernowk8_target,
1144 	.bios_limit	= acpi_processor_get_bios_limit,
1145 	.init		= powernowk8_cpu_init,
1146 	.exit		= powernowk8_cpu_exit,
1147 	.get		= powernowk8_get,
1148 	.name		= "powernow-k8",
1149 };
1150 
__request_acpi_cpufreq(void)1151 static void __request_acpi_cpufreq(void)
1152 {
1153 	const char drv[] = "acpi-cpufreq";
1154 	const char *cur_drv;
1155 
1156 	cur_drv = cpufreq_get_current_driver();
1157 	if (!cur_drv)
1158 		goto request;
1159 
1160 	if (strncmp(cur_drv, drv, min_t(size_t, strlen(cur_drv), strlen(drv))))
1161 		pr_warn("WTF driver: %s\n", cur_drv);
1162 
1163 	return;
1164 
1165  request:
1166 	pr_warn("This CPU is not supported anymore, using acpi-cpufreq instead.\n");
1167 	request_module(drv);
1168 }
1169 
1170 /* driver entry point for init */
powernowk8_init(void)1171 static int powernowk8_init(void)
1172 {
1173 	unsigned int i, supported_cpus = 0;
1174 	int ret;
1175 
1176 	if (!x86_match_cpu(powernow_k8_ids))
1177 		return -ENODEV;
1178 
1179 	if (boot_cpu_has(X86_FEATURE_HW_PSTATE)) {
1180 		__request_acpi_cpufreq();
1181 		return -ENODEV;
1182 	}
1183 
1184 	cpus_read_lock();
1185 	for_each_online_cpu(i) {
1186 		smp_call_function_single(i, check_supported_cpu, &ret, 1);
1187 		if (!ret)
1188 			supported_cpus++;
1189 	}
1190 
1191 	if (supported_cpus != num_online_cpus()) {
1192 		cpus_read_unlock();
1193 		return -ENODEV;
1194 	}
1195 	cpus_read_unlock();
1196 
1197 	ret = cpufreq_register_driver(&cpufreq_amd64_driver);
1198 	if (ret)
1199 		return ret;
1200 
1201 	pr_info("Found %d %s (%d cpu cores) (" VERSION ")\n",
1202 		num_online_nodes(), boot_cpu_data.x86_model_id, supported_cpus);
1203 
1204 	return ret;
1205 }
1206 
1207 /* driver entry point for term */
powernowk8_exit(void)1208 static void __exit powernowk8_exit(void)
1209 {
1210 	pr_debug("exit\n");
1211 
1212 	cpufreq_unregister_driver(&cpufreq_amd64_driver);
1213 }
1214 
1215 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com>");
1216 MODULE_AUTHOR("Mark Langsdorf <mark.langsdorf@amd.com>");
1217 MODULE_DESCRIPTION("AMD Athlon 64 and Opteron processor frequency driver.");
1218 MODULE_LICENSE("GPL");
1219 
1220 late_initcall(powernowk8_init);
1221 module_exit(powernowk8_exit);
1222