Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / arch / x86 / kernel / cpu / cpufreq / powernow-k8.c
1 /*
2  *   (c) 2003-2006 Advanced Micro Devices, Inc.
3  *  Your use of this code is subject to the terms and conditions of the
4  *  GNU general public license version 2. See "COPYING" or
5  *  http://www.gnu.org/licenses/gpl.html
6  *
7  *  Support : mark.langsdorf@amd.com
8  *
9  *  Based on the powernow-k7.c module written by Dave Jones.
10  *  (C) 2003 Dave Jones on behalf of SuSE Labs
11  *  (C) 2004 Dominik Brodowski <linux@brodo.de>
12  *  (C) 2004 Pavel Machek <pavel@suse.cz>
13  *  Licensed under the terms of the GNU GPL License version 2.
14  *  Based upon datasheets & sample CPUs kindly provided by AMD.
15  *
16  *  Valuable input gratefully received from Dave Jones, Pavel Machek,
17  *  Dominik Brodowski, Jacob Shin, and others.
18  *  Originally developed by Paul Devriendt.
19  *  Processor information obtained from Chapter 9 (Power and Thermal Management)
20  *  of the "BIOS and Kernel Developer's Guide for the AMD Athlon 64 and AMD
21  *  Opteron Processors" available for download from www.amd.com
22  *
23  *  Tables for specific CPUs can be inferred from
24  *     http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30430.pdf
25  */
26
27 #include <linux/kernel.h>
28 #include <linux/smp.h>
29 #include <linux/module.h>
30 #include <linux/init.h>
31 #include <linux/cpufreq.h>
32 #include <linux/slab.h>
33 #include <linux/string.h>
34 #include <linux/cpumask.h>
35 #include <linux/sched.h>        /* for current / set_cpus_allowed() */
36 #include <linux/io.h>
37 #include <linux/delay.h>
38
39 #include <asm/msr.h>
40
41 #include <linux/acpi.h>
42 #include <linux/mutex.h>
43 #include <acpi/processor.h>
44
45 #define PFX "powernow-k8: "
46 #define VERSION "version 2.20.00"
47 #include "powernow-k8.h"
48
49 /* serialize freq changes  */
50 static DEFINE_MUTEX(fidvid_mutex);
51
52 static DEFINE_PER_CPU(struct powernow_k8_data *, powernow_data);
53
54 static int cpu_family = CPU_OPTERON;
55
56 #ifndef CONFIG_SMP
57 static inline const struct cpumask *cpu_core_mask(int cpu)
58 {
59         return cpumask_of(0);
60 }
61 #endif
62
63 /* Return a frequency in MHz, given an input fid */
64 static u32 find_freq_from_fid(u32 fid)
65 {
66         return 800 + (fid * 100);
67 }
68
69 /* Return a frequency in KHz, given an input fid */
70 static u32 find_khz_freq_from_fid(u32 fid)
71 {
72         return 1000 * find_freq_from_fid(fid);
73 }
74
75 static u32 find_khz_freq_from_pstate(struct cpufreq_frequency_table *data,
76                 u32 pstate)
77 {
78         return data[pstate].frequency;
79 }
80
81 /* Return the vco fid for an input fid
82  *
83  * Each "low" fid has corresponding "high" fid, and you can get to "low" fids
84  * only from corresponding high fids. This returns "high" fid corresponding to
85  * "low" one.
86  */
87 static u32 convert_fid_to_vco_fid(u32 fid)
88 {
89         if (fid < HI_FID_TABLE_BOTTOM)
90                 return 8 + (2 * fid);
91         else
92                 return fid;
93 }
94
95 /*
96  * Return 1 if the pending bit is set. Unless we just instructed the processor
97  * to transition to a new state, seeing this bit set is really bad news.
98  */
99 static int pending_bit_stuck(void)
100 {
101         u32 lo, hi;
102
103         if (cpu_family == CPU_HW_PSTATE)
104                 return 0;
105
106         rdmsr(MSR_FIDVID_STATUS, lo, hi);
107         return lo & MSR_S_LO_CHANGE_PENDING ? 1 : 0;
108 }
109
110 /*
111  * Update the global current fid / vid values from the status msr.
112  * Returns 1 on error.
113  */
114 static int query_current_values_with_pending_wait(struct powernow_k8_data *data)
115 {
116         u32 lo, hi;
117         u32 i = 0;
118
119         if (cpu_family == CPU_HW_PSTATE) {
120                 if (data->currpstate == HW_PSTATE_INVALID) {
121                         /* read (initial) hw pstate if not yet set */
122                         rdmsr(MSR_PSTATE_STATUS, lo, hi);
123                         i = lo & HW_PSTATE_MASK;
124
125                         /*
126                          * a workaround for family 11h erratum 311 might cause
127                          * an "out-of-range Pstate if the core is in Pstate-0
128                          */
129                         if (i >= data->numps)
130                                 data->currpstate = HW_PSTATE_0;
131                         else
132                                 data->currpstate = i;
133                 }
134                 return 0;
135         }
136         do {
137                 if (i++ > 10000) {
138                         dprintk("detected change pending stuck\n");
139                         return 1;
140                 }
141                 rdmsr(MSR_FIDVID_STATUS, lo, hi);
142         } while (lo & MSR_S_LO_CHANGE_PENDING);
143
144         data->currvid = hi & MSR_S_HI_CURRENT_VID;
145         data->currfid = lo & MSR_S_LO_CURRENT_FID;
146
147         return 0;
148 }
149
150 /* the isochronous relief time */
151 static void count_off_irt(struct powernow_k8_data *data)
152 {
153         udelay((1 << data->irt) * 10);
154         return;
155 }
156
157 /* the voltage stabilization time */
158 static void count_off_vst(struct powernow_k8_data *data)
159 {
160         udelay(data->vstable * VST_UNITS_20US);
161         return;
162 }
163
164 /* need to init the control msr to a safe value (for each cpu) */
165 static void fidvid_msr_init(void)
166 {
167         u32 lo, hi;
168         u8 fid, vid;
169
170         rdmsr(MSR_FIDVID_STATUS, lo, hi);
171         vid = hi & MSR_S_HI_CURRENT_VID;
172         fid = lo & MSR_S_LO_CURRENT_FID;
173         lo = fid | (vid << MSR_C_LO_VID_SHIFT);
174         hi = MSR_C_HI_STP_GNT_BENIGN;
175         dprintk("cpu%d, init lo 0x%x, hi 0x%x\n", smp_processor_id(), lo, hi);
176         wrmsr(MSR_FIDVID_CTL, lo, hi);
177 }
178
179 /* write the new fid value along with the other control fields to the msr */
180 static int write_new_fid(struct powernow_k8_data *data, u32 fid)
181 {
182         u32 lo;
183         u32 savevid = data->currvid;
184         u32 i = 0;
185
186         if ((fid & INVALID_FID_MASK) || (data->currvid & INVALID_VID_MASK)) {
187                 printk(KERN_ERR PFX "internal error - overflow on fid write\n");
188                 return 1;
189         }
190
191         lo = fid;
192         lo |= (data->currvid << MSR_C_LO_VID_SHIFT);
193         lo |= MSR_C_LO_INIT_FID_VID;
194
195         dprintk("writing fid 0x%x, lo 0x%x, hi 0x%x\n",
196                 fid, lo, data->plllock * PLL_LOCK_CONVERSION);
197
198         do {
199                 wrmsr(MSR_FIDVID_CTL, lo, data->plllock * PLL_LOCK_CONVERSION);
200                 if (i++ > 100) {
201                         printk(KERN_ERR PFX
202                                 "Hardware error - pending bit very stuck - "
203                                 "no further pstate changes possible\n");
204                         return 1;
205                 }
206         } while (query_current_values_with_pending_wait(data));
207
208         count_off_irt(data);
209
210         if (savevid != data->currvid) {
211                 printk(KERN_ERR PFX
212                         "vid change on fid trans, old 0x%x, new 0x%x\n",
213                         savevid, data->currvid);
214                 return 1;
215         }
216
217         if (fid != data->currfid) {
218                 printk(KERN_ERR PFX
219                         "fid trans failed, fid 0x%x, curr 0x%x\n", fid,
220                         data->currfid);
221                 return 1;
222         }
223
224         return 0;
225 }
226
227 /* Write a new vid to the hardware */
228 static int write_new_vid(struct powernow_k8_data *data, u32 vid)
229 {
230         u32 lo;
231         u32 savefid = data->currfid;
232         int i = 0;
233
234         if ((data->currfid & INVALID_FID_MASK) || (vid & INVALID_VID_MASK)) {
235                 printk(KERN_ERR PFX "internal error - overflow on vid write\n");
236                 return 1;
237         }
238
239         lo = data->currfid;
240         lo |= (vid << MSR_C_LO_VID_SHIFT);
241         lo |= MSR_C_LO_INIT_FID_VID;
242
243         dprintk("writing vid 0x%x, lo 0x%x, hi 0x%x\n",
244                 vid, lo, STOP_GRANT_5NS);
245
246         do {
247                 wrmsr(MSR_FIDVID_CTL, lo, STOP_GRANT_5NS);
248                 if (i++ > 100) {
249                         printk(KERN_ERR PFX "internal error - pending bit "
250                                         "very stuck - no further pstate "
251                                         "changes possible\n");
252                         return 1;
253                 }
254         } while (query_current_values_with_pending_wait(data));
255
256         if (savefid != data->currfid) {
257                 printk(KERN_ERR PFX "fid changed on vid trans, old "
258                         "0x%x new 0x%x\n",
259                        savefid, data->currfid);
260                 return 1;
261         }
262
263         if (vid != data->currvid) {
264                 printk(KERN_ERR PFX "vid trans failed, vid 0x%x, "
265                                 "curr 0x%x\n",
266                                 vid, data->currvid);
267                 return 1;
268         }
269
270         return 0;
271 }
272
273 /*
274  * Reduce the vid by the max of step or reqvid.
275  * Decreasing vid codes represent increasing voltages:
276  * vid of 0 is 1.550V, vid of 0x1e is 0.800V, vid of VID_OFF is off.
277  */
278 static int decrease_vid_code_by_step(struct powernow_k8_data *data,
279                 u32 reqvid, u32 step)
280 {
281         if ((data->currvid - reqvid) > step)
282                 reqvid = data->currvid - step;
283
284         if (write_new_vid(data, reqvid))
285                 return 1;
286
287         count_off_vst(data);
288
289         return 0;
290 }
291
292 /* Change hardware pstate by single MSR write */
293 static int transition_pstate(struct powernow_k8_data *data, u32 pstate)
294 {
295         wrmsr(MSR_PSTATE_CTRL, pstate, 0);
296         data->currpstate = pstate;
297         return 0;
298 }
299
300 /* Change Opteron/Athlon64 fid and vid, by the 3 phases. */
301 static int transition_fid_vid(struct powernow_k8_data *data,
302                 u32 reqfid, u32 reqvid)
303 {
304         if (core_voltage_pre_transition(data, reqvid))
305                 return 1;
306
307         if (core_frequency_transition(data, reqfid))
308                 return 1;
309
310         if (core_voltage_post_transition(data, reqvid))
311                 return 1;
312
313         if (query_current_values_with_pending_wait(data))
314                 return 1;
315
316         if ((reqfid != data->currfid) || (reqvid != data->currvid)) {
317                 printk(KERN_ERR PFX "failed (cpu%d): req 0x%x 0x%x, "
318                                 "curr 0x%x 0x%x\n",
319                                 smp_processor_id(),
320                                 reqfid, reqvid, data->currfid, data->currvid);
321                 return 1;
322         }
323
324         dprintk("transitioned (cpu%d): new fid 0x%x, vid 0x%x\n",
325                 smp_processor_id(), data->currfid, data->currvid);
326
327         return 0;
328 }
329
330 /* Phase 1 - core voltage transition ... setup voltage */
331 static int core_voltage_pre_transition(struct powernow_k8_data *data,
332                 u32 reqvid)
333 {
334         u32 rvosteps = data->rvo;
335         u32 savefid = data->currfid;
336         u32 maxvid, lo;
337
338         dprintk("ph1 (cpu%d): start, currfid 0x%x, currvid 0x%x, "
339                 "reqvid 0x%x, rvo 0x%x\n",
340                 smp_processor_id(),
341                 data->currfid, data->currvid, reqvid, data->rvo);
342
343         rdmsr(MSR_FIDVID_STATUS, lo, maxvid);
344         maxvid = 0x1f & (maxvid >> 16);
345         dprintk("ph1 maxvid=0x%x\n", maxvid);
346         if (reqvid < maxvid) /* lower numbers are higher voltages */
347                 reqvid = maxvid;
348
349         while (data->currvid > reqvid) {
350                 dprintk("ph1: curr 0x%x, req vid 0x%x\n",
351                         data->currvid, reqvid);
352                 if (decrease_vid_code_by_step(data, reqvid, data->vidmvs))
353                         return 1;
354         }
355
356         while ((rvosteps > 0) && ((data->rvo + data->currvid) > reqvid)) {
357                 if (data->currvid == maxvid) {
358                         rvosteps = 0;
359                 } else {
360                         dprintk("ph1: changing vid for rvo, req 0x%x\n",
361                                 data->currvid - 1);
362                         if (decrease_vid_code_by_step(data, data->currvid-1, 1))
363                                 return 1;
364                         rvosteps--;
365                 }
366         }
367
368         if (query_current_values_with_pending_wait(data))
369                 return 1;
370
371         if (savefid != data->currfid) {
372                 printk(KERN_ERR PFX "ph1 err, currfid changed 0x%x\n",
373                                 data->currfid);
374                 return 1;
375         }
376
377         dprintk("ph1 complete, currfid 0x%x, currvid 0x%x\n",
378                 data->currfid, data->currvid);
379
380         return 0;
381 }
382
383 /* Phase 2 - core frequency transition */
384 static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid)
385 {
386         u32 vcoreqfid, vcocurrfid, vcofiddiff;
387         u32 fid_interval, savevid = data->currvid;
388
389         if ((reqfid < HI_FID_TABLE_BOTTOM) &&
390             (data->currfid < HI_FID_TABLE_BOTTOM)) {
391                 printk(KERN_ERR PFX "ph2: illegal lo-lo transition "
392                                 "0x%x 0x%x\n", reqfid, data->currfid);
393                 return 1;
394         }
395
396         if (data->currfid == reqfid) {
397                 printk(KERN_ERR PFX "ph2 null fid transition 0x%x\n",
398                                 data->currfid);
399                 return 0;
400         }
401
402         dprintk("ph2 (cpu%d): starting, currfid 0x%x, currvid 0x%x, "
403                 "reqfid 0x%x\n",
404                 smp_processor_id(),
405                 data->currfid, data->currvid, reqfid);
406
407         vcoreqfid = convert_fid_to_vco_fid(reqfid);
408         vcocurrfid = convert_fid_to_vco_fid(data->currfid);
409         vcofiddiff = vcocurrfid > vcoreqfid ? vcocurrfid - vcoreqfid
410             : vcoreqfid - vcocurrfid;
411
412         while (vcofiddiff > 2) {
413                 (data->currfid & 1) ? (fid_interval = 1) : (fid_interval = 2);
414
415                 if (reqfid > data->currfid) {
416                         if (data->currfid > LO_FID_TABLE_TOP) {
417                                 if (write_new_fid(data,
418                                                 data->currfid + fid_interval))
419                                         return 1;
420                         } else {
421                                 if (write_new_fid
422                                     (data,
423                                      2 + convert_fid_to_vco_fid(data->currfid)))
424                                         return 1;
425                         }
426                 } else {
427                         if (write_new_fid(data, data->currfid - fid_interval))
428                                 return 1;
429                 }
430
431                 vcocurrfid = convert_fid_to_vco_fid(data->currfid);
432                 vcofiddiff = vcocurrfid > vcoreqfid ? vcocurrfid - vcoreqfid
433                     : vcoreqfid - vcocurrfid;
434         }
435
436         if (write_new_fid(data, reqfid))
437                 return 1;
438
439         if (query_current_values_with_pending_wait(data))
440                 return 1;
441
442         if (data->currfid != reqfid) {
443                 printk(KERN_ERR PFX
444                         "ph2: mismatch, failed fid transition, "
445                         "curr 0x%x, req 0x%x\n",
446                         data->currfid, reqfid);
447                 return 1;
448         }
449
450         if (savevid != data->currvid) {
451                 printk(KERN_ERR PFX "ph2: vid changed, save 0x%x, curr 0x%x\n",
452                         savevid, data->currvid);
453                 return 1;
454         }
455
456         dprintk("ph2 complete, currfid 0x%x, currvid 0x%x\n",
457                 data->currfid, data->currvid);
458
459         return 0;
460 }
461
462 /* Phase 3 - core voltage transition flow ... jump to the final vid. */
463 static int core_voltage_post_transition(struct powernow_k8_data *data,
464                 u32 reqvid)
465 {
466         u32 savefid = data->currfid;
467         u32 savereqvid = reqvid;
468
469         dprintk("ph3 (cpu%d): starting, currfid 0x%x, currvid 0x%x\n",
470                 smp_processor_id(),
471                 data->currfid, data->currvid);
472
473         if (reqvid != data->currvid) {
474                 if (write_new_vid(data, reqvid))
475                         return 1;
476
477                 if (savefid != data->currfid) {
478                         printk(KERN_ERR PFX
479                                "ph3: bad fid change, save 0x%x, curr 0x%x\n",
480                                savefid, data->currfid);
481                         return 1;
482                 }
483
484                 if (data->currvid != reqvid) {
485                         printk(KERN_ERR PFX
486                                "ph3: failed vid transition\n, "
487                                "req 0x%x, curr 0x%x",
488                                reqvid, data->currvid);
489                         return 1;
490                 }
491         }
492
493         if (query_current_values_with_pending_wait(data))
494                 return 1;
495
496         if (savereqvid != data->currvid) {
497                 dprintk("ph3 failed, currvid 0x%x\n", data->currvid);
498                 return 1;
499         }
500
501         if (savefid != data->currfid) {
502                 dprintk("ph3 failed, currfid changed 0x%x\n",
503                         data->currfid);
504                 return 1;
505         }
506
507         dprintk("ph3 complete, currfid 0x%x, currvid 0x%x\n",
508                 data->currfid, data->currvid);
509
510         return 0;
511 }
512
513 static int check_supported_cpu(unsigned int cpu)
514 {
515         cpumask_t oldmask;
516         u32 eax, ebx, ecx, edx;
517         unsigned int rc = 0;
518
519         oldmask = current->cpus_allowed;
520         set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
521
522         if (smp_processor_id() != cpu) {
523                 printk(KERN_ERR PFX "limiting to cpu %u failed\n", cpu);
524                 goto out;
525         }
526
527         if (current_cpu_data.x86_vendor != X86_VENDOR_AMD)
528                 goto out;
529
530         eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
531         if (((eax & CPUID_XFAM) != CPUID_XFAM_K8) &&
532             ((eax & CPUID_XFAM) < CPUID_XFAM_10H))
533                 goto out;
534
535         if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) {
536                 if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) ||
537                     ((eax & CPUID_XMOD) > CPUID_XMOD_REV_MASK)) {
538                         printk(KERN_INFO PFX
539                                 "Processor cpuid %x not supported\n", eax);
540                         goto out;
541                 }
542
543                 eax = cpuid_eax(CPUID_GET_MAX_CAPABILITIES);
544                 if (eax < CPUID_FREQ_VOLT_CAPABILITIES) {
545                         printk(KERN_INFO PFX
546                                "No frequency change capabilities detected\n");
547                         goto out;
548                 }
549
550                 cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
551                 if ((edx & P_STATE_TRANSITION_CAPABLE)
552                         != P_STATE_TRANSITION_CAPABLE) {
553                         printk(KERN_INFO PFX
554                                 "Power state transitions not supported\n");
555                         goto out;
556                 }
557         } else { /* must be a HW Pstate capable processor */
558                 cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
559                 if ((edx & USE_HW_PSTATE) == USE_HW_PSTATE)
560                         cpu_family = CPU_HW_PSTATE;
561                 else
562                         goto out;
563         }
564
565         rc = 1;
566
567 out:
568         set_cpus_allowed_ptr(current, &oldmask);
569         return rc;
570 }
571
572 static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst,
573                 u8 maxvid)
574 {
575         unsigned int j;
576         u8 lastfid = 0xff;
577
578         for (j = 0; j < data->numps; j++) {
579                 if (pst[j].vid > LEAST_VID) {
580                         printk(KERN_ERR FW_BUG PFX "vid %d invalid : 0x%x\n",
581                                j, pst[j].vid);
582                         return -EINVAL;
583                 }
584                 if (pst[j].vid < data->rvo) {
585                         /* vid + rvo >= 0 */
586                         printk(KERN_ERR FW_BUG PFX "0 vid exceeded with pstate"
587                                " %d\n", j);
588                         return -ENODEV;
589                 }
590                 if (pst[j].vid < maxvid + data->rvo) {
591                         /* vid + rvo >= maxvid */
592                         printk(KERN_ERR FW_BUG PFX "maxvid exceeded with pstate"
593                                " %d\n", j);
594                         return -ENODEV;
595                 }
596                 if (pst[j].fid > MAX_FID) {
597                         printk(KERN_ERR FW_BUG PFX "maxfid exceeded with pstate"
598                                " %d\n", j);
599                         return -ENODEV;
600                 }
601                 if (j && (pst[j].fid < HI_FID_TABLE_BOTTOM)) {
602                         /* Only first fid is allowed to be in "low" range */
603                         printk(KERN_ERR FW_BUG PFX "two low fids - %d : "
604                                "0x%x\n", j, pst[j].fid);
605                         return -EINVAL;
606                 }
607                 if (pst[j].fid < lastfid)
608                         lastfid = pst[j].fid;
609         }
610         if (lastfid & 1) {
611                 printk(KERN_ERR FW_BUG PFX "lastfid invalid\n");
612                 return -EINVAL;
613         }
614         if (lastfid > LO_FID_TABLE_TOP)
615                 printk(KERN_INFO FW_BUG PFX
616                         "first fid not from lo freq table\n");
617
618         return 0;
619 }
620
621 static void invalidate_entry(struct powernow_k8_data *data, unsigned int entry)
622 {
623         data->powernow_table[entry].frequency = CPUFREQ_ENTRY_INVALID;
624 }
625
626 static void print_basics(struct powernow_k8_data *data)
627 {
628         int j;
629         for (j = 0; j < data->numps; j++) {
630                 if (data->powernow_table[j].frequency !=
631                                 CPUFREQ_ENTRY_INVALID) {
632                         if (cpu_family == CPU_HW_PSTATE) {
633                                 printk(KERN_INFO PFX
634                                         "   %d : pstate %d (%d MHz)\n", j,
635                                         data->powernow_table[j].index,
636                                         data->powernow_table[j].frequency/1000);
637                         } else {
638                                 printk(KERN_INFO PFX
639                                         "   %d : fid 0x%x (%d MHz), vid 0x%x\n",
640                                         j,
641                                         data->powernow_table[j].index & 0xff,
642                                         data->powernow_table[j].frequency/1000,
643                                         data->powernow_table[j].index >> 8);
644                         }
645                 }
646         }
647         if (data->batps)
648                 printk(KERN_INFO PFX "Only %d pstates on battery\n",
649                                 data->batps);
650 }
651
652 static u32 freq_from_fid_did(u32 fid, u32 did)
653 {
654         u32 mhz = 0;
655
656         if (boot_cpu_data.x86 == 0x10)
657                 mhz = (100 * (fid + 0x10)) >> did;
658         else if (boot_cpu_data.x86 == 0x11)
659                 mhz = (100 * (fid + 8)) >> did;
660         else
661                 BUG();
662
663         return mhz * 1000;
664 }
665
666 static int fill_powernow_table(struct powernow_k8_data *data,
667                 struct pst_s *pst, u8 maxvid)
668 {
669         struct cpufreq_frequency_table *powernow_table;
670         unsigned int j;
671
672         if (data->batps) {
673                 /* use ACPI support to get full speed on mains power */
674                 printk(KERN_WARNING PFX
675                         "Only %d pstates usable (use ACPI driver for full "
676                         "range\n", data->batps);
677                 data->numps = data->batps;
678         }
679
680         for (j = 1; j < data->numps; j++) {
681                 if (pst[j-1].fid >= pst[j].fid) {
682                         printk(KERN_ERR PFX "PST out of sequence\n");
683                         return -EINVAL;
684                 }
685         }
686
687         if (data->numps < 2) {
688                 printk(KERN_ERR PFX "no p states to transition\n");
689                 return -ENODEV;
690         }
691
692         if (check_pst_table(data, pst, maxvid))
693                 return -EINVAL;
694
695         powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
696                 * (data->numps + 1)), GFP_KERNEL);
697         if (!powernow_table) {
698                 printk(KERN_ERR PFX "powernow_table memory alloc failure\n");
699                 return -ENOMEM;
700         }
701
702         for (j = 0; j < data->numps; j++) {
703                 int freq;
704                 powernow_table[j].index = pst[j].fid; /* lower 8 bits */
705                 powernow_table[j].index |= (pst[j].vid << 8); /* upper 8 bits */
706                 freq = find_khz_freq_from_fid(pst[j].fid);
707                 powernow_table[j].frequency = freq;
708         }
709         powernow_table[data->numps].frequency = CPUFREQ_TABLE_END;
710         powernow_table[data->numps].index = 0;
711
712         if (query_current_values_with_pending_wait(data)) {
713                 kfree(powernow_table);
714                 return -EIO;
715         }
716
717         dprintk("cfid 0x%x, cvid 0x%x\n", data->currfid, data->currvid);
718         data->powernow_table = powernow_table;
719         if (cpumask_first(cpu_core_mask(data->cpu)) == data->cpu)
720                 print_basics(data);
721
722         for (j = 0; j < data->numps; j++)
723                 if ((pst[j].fid == data->currfid) &&
724                     (pst[j].vid == data->currvid))
725                         return 0;
726
727         dprintk("currfid/vid do not match PST, ignoring\n");
728         return 0;
729 }
730
731 /* Find and validate the PSB/PST table in BIOS. */
732 static int find_psb_table(struct powernow_k8_data *data)
733 {
734         struct psb_s *psb;
735         unsigned int i;
736         u32 mvs;
737         u8 maxvid;
738         u32 cpst = 0;
739         u32 thiscpuid;
740
741         for (i = 0xc0000; i < 0xffff0; i += 0x10) {
742                 /* Scan BIOS looking for the signature. */
743                 /* It can not be at ffff0 - it is too big. */
744
745                 psb = phys_to_virt(i);
746                 if (memcmp(psb, PSB_ID_STRING, PSB_ID_STRING_LEN) != 0)
747                         continue;
748
749                 dprintk("found PSB header at 0x%p\n", psb);
750
751                 dprintk("table vers: 0x%x\n", psb->tableversion);
752                 if (psb->tableversion != PSB_VERSION_1_4) {
753                         printk(KERN_ERR FW_BUG PFX "PSB table is not v1.4\n");
754                         return -ENODEV;
755                 }
756
757                 dprintk("flags: 0x%x\n", psb->flags1);
758                 if (psb->flags1) {
759                         printk(KERN_ERR FW_BUG PFX "unknown flags\n");
760                         return -ENODEV;
761                 }
762
763                 data->vstable = psb->vstable;
764                 dprintk("voltage stabilization time: %d(*20us)\n",
765                                 data->vstable);
766
767                 dprintk("flags2: 0x%x\n", psb->flags2);
768                 data->rvo = psb->flags2 & 3;
769                 data->irt = ((psb->flags2) >> 2) & 3;
770                 mvs = ((psb->flags2) >> 4) & 3;
771                 data->vidmvs = 1 << mvs;
772                 data->batps = ((psb->flags2) >> 6) & 3;
773
774                 dprintk("ramp voltage offset: %d\n", data->rvo);
775                 dprintk("isochronous relief time: %d\n", data->irt);
776                 dprintk("maximum voltage step: %d - 0x%x\n", mvs, data->vidmvs);
777
778                 dprintk("numpst: 0x%x\n", psb->num_tables);
779                 cpst = psb->num_tables;
780                 if ((psb->cpuid == 0x00000fc0) ||
781                     (psb->cpuid == 0x00000fe0)) {
782                         thiscpuid = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
783                         if ((thiscpuid == 0x00000fc0) ||
784                             (thiscpuid == 0x00000fe0))
785                                 cpst = 1;
786                 }
787                 if (cpst != 1) {
788                         printk(KERN_ERR FW_BUG PFX "numpst must be 1\n");
789                         return -ENODEV;
790                 }
791
792                 data->plllock = psb->plllocktime;
793                 dprintk("plllocktime: 0x%x (units 1us)\n", psb->plllocktime);
794                 dprintk("maxfid: 0x%x\n", psb->maxfid);
795                 dprintk("maxvid: 0x%x\n", psb->maxvid);
796                 maxvid = psb->maxvid;
797
798                 data->numps = psb->numps;
799                 dprintk("numpstates: 0x%x\n", data->numps);
800                 return fill_powernow_table(data,
801                                 (struct pst_s *)(psb+1), maxvid);
802         }
803         /*
804          * If you see this message, complain to BIOS manufacturer. If
805          * he tells you "we do not support Linux" or some similar
806          * nonsense, remember that Windows 2000 uses the same legacy
807          * mechanism that the old Linux PSB driver uses. Tell them it
808          * is broken with Windows 2000.
809          *
810          * The reference to the AMD documentation is chapter 9 in the
811          * BIOS and Kernel Developer's Guide, which is available on
812          * www.amd.com
813          */
814         printk(KERN_ERR FW_BUG PFX "No PSB or ACPI _PSS objects\n");
815         return -ENODEV;
816 }
817
818 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data,
819                 unsigned int index)
820 {
821         acpi_integer control;
822
823         if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE))
824                 return;
825
826         control = data->acpi_data.states[index].control; data->irt = (control
827                         >> IRT_SHIFT) & IRT_MASK; data->rvo = (control >>
828                                 RVO_SHIFT) & RVO_MASK; data->exttype = (control
829                                         >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
830         data->plllock = (control >> PLL_L_SHIFT) & PLL_L_MASK; data->vidmvs = 1
831                 << ((control >> MVS_SHIFT) & MVS_MASK); data->vstable =
832                 (control >> VST_SHIFT) & VST_MASK; }
833
834 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
835 {
836         struct cpufreq_frequency_table *powernow_table;
837         int ret_val = -ENODEV;
838         acpi_integer space_id;
839
840         if (acpi_processor_register_performance(&data->acpi_data, data->cpu)) {
841                 dprintk("register performance failed: bad ACPI data\n");
842                 return -EIO;
843         }
844
845         /* verify the data contained in the ACPI structures */
846         if (data->acpi_data.state_count <= 1) {
847                 dprintk("No ACPI P-States\n");
848                 goto err_out;
849         }
850
851         space_id = data->acpi_data.control_register.space_id;
852         if ((space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
853                 (space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
854                 dprintk("Invalid control/status registers (%x - %x)\n",
855                         data->acpi_data.control_register.space_id,
856                         space_id);
857                 goto err_out;
858         }
859
860         /* fill in data->powernow_table */
861         powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
862                 * (data->acpi_data.state_count + 1)), GFP_KERNEL);
863         if (!powernow_table) {
864                 dprintk("powernow_table memory alloc failure\n");
865                 goto err_out;
866         }
867
868         if (cpu_family == CPU_HW_PSTATE)
869                 ret_val = fill_powernow_table_pstate(data, powernow_table);
870         else
871                 ret_val = fill_powernow_table_fidvid(data, powernow_table);
872         if (ret_val)
873                 goto err_out_mem;
874
875         powernow_table[data->acpi_data.state_count].frequency =
876                 CPUFREQ_TABLE_END;
877         powernow_table[data->acpi_data.state_count].index = 0;
878         data->powernow_table = powernow_table;
879
880         /* fill in data */
881         data->numps = data->acpi_data.state_count;
882         if (cpumask_first(cpu_core_mask(data->cpu)) == data->cpu)
883                 print_basics(data);
884         powernow_k8_acpi_pst_values(data, 0);
885
886         /* notify BIOS that we exist */
887         acpi_processor_notify_smm(THIS_MODULE);
888
889         if (!alloc_cpumask_var(&data->acpi_data.shared_cpu_map, GFP_KERNEL)) {
890                 printk(KERN_ERR PFX
891                                 "unable to alloc powernow_k8_data cpumask\n");
892                 ret_val = -ENOMEM;
893                 goto err_out_mem;
894         }
895
896         return 0;
897
898 err_out_mem:
899         kfree(powernow_table);
900
901 err_out:
902         acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
903
904         /* data->acpi_data.state_count informs us at ->exit()
905          * whether ACPI was used */
906         data->acpi_data.state_count = 0;
907
908         return ret_val;
909 }
910
911 static int fill_powernow_table_pstate(struct powernow_k8_data *data,
912                 struct cpufreq_frequency_table *powernow_table)
913 {
914         int i;
915         u32 hi = 0, lo = 0;
916         rdmsr(MSR_PSTATE_CUR_LIMIT, hi, lo);
917         data->max_hw_pstate = (hi & HW_PSTATE_MAX_MASK) >> HW_PSTATE_MAX_SHIFT;
918
919         for (i = 0; i < data->acpi_data.state_count; i++) {
920                 u32 index;
921
922                 index = data->acpi_data.states[i].control & HW_PSTATE_MASK;
923                 if (index > data->max_hw_pstate) {
924                         printk(KERN_ERR PFX "invalid pstate %d - "
925                                         "bad value %d.\n", i, index);
926                         printk(KERN_ERR PFX "Please report to BIOS "
927                                         "manufacturer\n");
928                         invalidate_entry(data, i);
929                         continue;
930                 }
931                 rdmsr(MSR_PSTATE_DEF_BASE + index, lo, hi);
932                 if (!(hi & HW_PSTATE_VALID_MASK)) {
933                         dprintk("invalid pstate %d, ignoring\n", index);
934                         invalidate_entry(data, i);
935                         continue;
936                 }
937
938                 powernow_table[i].index = index;
939
940                 /* Frequency may be rounded for these */
941                 if (boot_cpu_data.x86 == 0x10 || boot_cpu_data.x86 == 0x11) {
942                         powernow_table[i].frequency =
943                                 freq_from_fid_did(lo & 0x3f, (lo >> 6) & 7);
944                 } else
945                         powernow_table[i].frequency =
946                                 data->acpi_data.states[i].core_frequency * 1000;
947         }
948         return 0;
949 }
950
951 static int fill_powernow_table_fidvid(struct powernow_k8_data *data,
952                 struct cpufreq_frequency_table *powernow_table)
953 {
954         int i;
955         int cntlofreq = 0;
956
957         for (i = 0; i < data->acpi_data.state_count; i++) {
958                 u32 fid;
959                 u32 vid;
960                 u32 freq, index;
961                 acpi_integer status, control;
962
963                 if (data->exttype) {
964                         status =  data->acpi_data.states[i].status;
965                         fid = status & EXT_FID_MASK;
966                         vid = (status >> VID_SHIFT) & EXT_VID_MASK;
967                 } else {
968                         control =  data->acpi_data.states[i].control;
969                         fid = control & FID_MASK;
970                         vid = (control >> VID_SHIFT) & VID_MASK;
971                 }
972
973                 dprintk("   %d : fid 0x%x, vid 0x%x\n", i, fid, vid);
974
975                 index = fid | (vid<<8);
976                 powernow_table[i].index = index;
977
978                 freq = find_khz_freq_from_fid(fid);
979                 powernow_table[i].frequency = freq;
980
981                 /* verify frequency is OK */
982                 if ((freq > (MAX_FREQ * 1000)) || (freq < (MIN_FREQ * 1000))) {
983                         dprintk("invalid freq %u kHz, ignoring\n", freq);
984                         invalidate_entry(data, i);
985                         continue;
986                 }
987
988                 /* verify voltage is OK -
989                  * BIOSs are using "off" to indicate invalid */
990                 if (vid == VID_OFF) {
991                         dprintk("invalid vid %u, ignoring\n", vid);
992                         invalidate_entry(data, i);
993                         continue;
994                 }
995
996                 /* verify only 1 entry from the lo frequency table */
997                 if (fid < HI_FID_TABLE_BOTTOM) {
998                         if (cntlofreq) {
999                                 /* if both entries are the same,
1000                                  * ignore this one ... */
1001                                 if ((freq != powernow_table[cntlofreq].frequency) ||
1002                                     (index != powernow_table[cntlofreq].index)) {
1003                                         printk(KERN_ERR PFX
1004                                                 "Too many lo freq table "
1005                                                 "entries\n");
1006                                         return 1;
1007                                 }
1008
1009                                 dprintk("double low frequency table entry, "
1010                                                 "ignoring it.\n");
1011                                 invalidate_entry(data, i);
1012                                 continue;
1013                         } else
1014                                 cntlofreq = i;
1015                 }
1016
1017                 if (freq != (data->acpi_data.states[i].core_frequency * 1000)) {
1018                         printk(KERN_INFO PFX "invalid freq entries "
1019                                 "%u kHz vs. %u kHz\n", freq,
1020                                 (unsigned int)
1021                                 (data->acpi_data.states[i].core_frequency
1022                                  * 1000));
1023                         invalidate_entry(data, i);
1024                         continue;
1025                 }
1026         }
1027         return 0;
1028 }
1029
1030 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data)
1031 {
1032         if (data->acpi_data.state_count)
1033                 acpi_processor_unregister_performance(&data->acpi_data,
1034                                 data->cpu);
1035         free_cpumask_var(data->acpi_data.shared_cpu_map);
1036 }
1037
1038 static int get_transition_latency(struct powernow_k8_data *data)
1039 {
1040         int max_latency = 0;
1041         int i;
1042         for (i = 0; i < data->acpi_data.state_count; i++) {
1043                 int cur_latency = data->acpi_data.states[i].transition_latency
1044                         + data->acpi_data.states[i].bus_master_latency;
1045                 if (cur_latency > max_latency)
1046                         max_latency = cur_latency;
1047         }
1048         /* value in usecs, needs to be in nanoseconds */
1049         return 1000 * max_latency;
1050 }
1051
1052 /* Take a frequency, and issue the fid/vid transition command */
1053 static int transition_frequency_fidvid(struct powernow_k8_data *data,
1054                 unsigned int index)
1055 {
1056         u32 fid = 0;
1057         u32 vid = 0;
1058         int res, i;
1059         struct cpufreq_freqs freqs;
1060
1061         dprintk("cpu %d transition to index %u\n", smp_processor_id(), index);
1062
1063         /* fid/vid correctness check for k8 */
1064         /* fid are the lower 8 bits of the index we stored into
1065          * the cpufreq frequency table in find_psb_table, vid
1066          * are the upper 8 bits.
1067          */
1068         fid = data->powernow_table[index].index & 0xFF;
1069         vid = (data->powernow_table[index].index & 0xFF00) >> 8;
1070
1071         dprintk("table matched fid 0x%x, giving vid 0x%x\n", fid, vid);
1072
1073         if (query_current_values_with_pending_wait(data))
1074                 return 1;
1075
1076         if ((data->currvid == vid) && (data->currfid == fid)) {
1077                 dprintk("target matches current values (fid 0x%x, vid 0x%x)\n",
1078                         fid, vid);
1079                 return 0;
1080         }
1081
1082         if ((fid < HI_FID_TABLE_BOTTOM) &&
1083             (data->currfid < HI_FID_TABLE_BOTTOM)) {
1084                 printk(KERN_ERR PFX
1085                        "ignoring illegal change in lo freq table-%x to 0x%x\n",
1086                        data->currfid, fid);
1087                 return 1;
1088         }
1089
1090         dprintk("cpu %d, changing to fid 0x%x, vid 0x%x\n",
1091                 smp_processor_id(), fid, vid);
1092         freqs.old = find_khz_freq_from_fid(data->currfid);
1093         freqs.new = find_khz_freq_from_fid(fid);
1094
1095         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1096                 freqs.cpu = i;
1097                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1098         }
1099
1100         res = transition_fid_vid(data, fid, vid);
1101         freqs.new = find_khz_freq_from_fid(data->currfid);
1102
1103         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1104                 freqs.cpu = i;
1105                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1106         }
1107         return res;
1108 }
1109
1110 /* Take a frequency, and issue the hardware pstate transition command */
1111 static int transition_frequency_pstate(struct powernow_k8_data *data,
1112                 unsigned int index)
1113 {
1114         u32 pstate = 0;
1115         int res, i;
1116         struct cpufreq_freqs freqs;
1117
1118         dprintk("cpu %d transition to index %u\n", smp_processor_id(), index);
1119
1120         /* get MSR index for hardware pstate transition */
1121         pstate = index & HW_PSTATE_MASK;
1122         if (pstate > data->max_hw_pstate)
1123                 return 0;
1124         freqs.old = find_khz_freq_from_pstate(data->powernow_table,
1125                         data->currpstate);
1126         freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
1127
1128         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1129                 freqs.cpu = i;
1130                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1131         }
1132
1133         res = transition_pstate(data, pstate);
1134         freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
1135
1136         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1137                 freqs.cpu = i;
1138                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1139         }
1140         return res;
1141 }
1142
1143 /* Driver entry point to switch to the target frequency */
1144 static int powernowk8_target(struct cpufreq_policy *pol,
1145                 unsigned targfreq, unsigned relation)
1146 {
1147         cpumask_t oldmask;
1148         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1149         u32 checkfid;
1150         u32 checkvid;
1151         unsigned int newstate;
1152         int ret = -EIO;
1153
1154         if (!data)
1155                 return -EINVAL;
1156
1157         checkfid = data->currfid;
1158         checkvid = data->currvid;
1159
1160         /* only run on specific CPU from here on */
1161         oldmask = current->cpus_allowed;
1162         set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
1163
1164         if (smp_processor_id() != pol->cpu) {
1165                 printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
1166                 goto err_out;
1167         }
1168
1169         if (pending_bit_stuck()) {
1170                 printk(KERN_ERR PFX "failing targ, change pending bit set\n");
1171                 goto err_out;
1172         }
1173
1174         dprintk("targ: cpu %d, %d kHz, min %d, max %d, relation %d\n",
1175                 pol->cpu, targfreq, pol->min, pol->max, relation);
1176
1177         if (query_current_values_with_pending_wait(data))
1178                 goto err_out;
1179
1180         if (cpu_family != CPU_HW_PSTATE) {
1181                 dprintk("targ: curr fid 0x%x, vid 0x%x\n",
1182                 data->currfid, data->currvid);
1183
1184                 if ((checkvid != data->currvid) ||
1185                     (checkfid != data->currfid)) {
1186                         printk(KERN_INFO PFX
1187                                 "error - out of sync, fix 0x%x 0x%x, "
1188                                 "vid 0x%x 0x%x\n",
1189                                 checkfid, data->currfid,
1190                                 checkvid, data->currvid);
1191                 }
1192         }
1193
1194         if (cpufreq_frequency_table_target(pol, data->powernow_table,
1195                                 targfreq, relation, &newstate))
1196                 goto err_out;
1197
1198         mutex_lock(&fidvid_mutex);
1199
1200         powernow_k8_acpi_pst_values(data, newstate);
1201
1202         if (cpu_family == CPU_HW_PSTATE)
1203                 ret = transition_frequency_pstate(data, newstate);
1204         else
1205                 ret = transition_frequency_fidvid(data, newstate);
1206         if (ret) {
1207                 printk(KERN_ERR PFX "transition frequency failed\n");
1208                 ret = 1;
1209                 mutex_unlock(&fidvid_mutex);
1210                 goto err_out;
1211         }
1212         mutex_unlock(&fidvid_mutex);
1213
1214         if (cpu_family == CPU_HW_PSTATE)
1215                 pol->cur = find_khz_freq_from_pstate(data->powernow_table,
1216                                 newstate);
1217         else
1218                 pol->cur = find_khz_freq_from_fid(data->currfid);
1219         ret = 0;
1220
1221 err_out:
1222         set_cpus_allowed_ptr(current, &oldmask);
1223         return ret;
1224 }
1225
1226 /* Driver entry point to verify the policy and range of frequencies */
1227 static int powernowk8_verify(struct cpufreq_policy *pol)
1228 {
1229         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1230
1231         if (!data)
1232                 return -EINVAL;
1233
1234         return cpufreq_frequency_table_verify(pol, data->powernow_table);
1235 }
1236
1237 static const char ACPI_PSS_BIOS_BUG_MSG[] =
1238         KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n"
1239         KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n";
1240
1241 /* per CPU init entry point to the driver */
1242 static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
1243 {
1244         struct powernow_k8_data *data;
1245         cpumask_t oldmask;
1246         int rc;
1247
1248         if (!cpu_online(pol->cpu))
1249                 return -ENODEV;
1250
1251         if (!check_supported_cpu(pol->cpu))
1252                 return -ENODEV;
1253
1254         data = kzalloc(sizeof(struct powernow_k8_data), GFP_KERNEL);
1255         if (!data) {
1256                 printk(KERN_ERR PFX "unable to alloc powernow_k8_data");
1257                 return -ENOMEM;
1258         }
1259
1260         data->cpu = pol->cpu;
1261         data->currpstate = HW_PSTATE_INVALID;
1262
1263         if (powernow_k8_cpu_init_acpi(data)) {
1264                 /*
1265                  * Use the PSB BIOS structure. This is only availabe on
1266                  * an UP version, and is deprecated by AMD.
1267                  */
1268                 if (num_online_cpus() != 1) {
1269                         printk_once(ACPI_PSS_BIOS_BUG_MSG);
1270                         goto err_out;
1271                 }
1272                 if (pol->cpu != 0) {
1273                         printk(KERN_ERR FW_BUG PFX "No ACPI _PSS objects for "
1274                                "CPU other than CPU0. Complain to your BIOS "
1275                                "vendor.\n");
1276                         goto err_out;
1277                 }
1278                 rc = find_psb_table(data);
1279                 if (rc)
1280                         goto err_out;
1281
1282                 /* Take a crude guess here.
1283                  * That guess was in microseconds, so multiply with 1000 */
1284                 pol->cpuinfo.transition_latency = (
1285                          ((data->rvo + 8) * data->vstable * VST_UNITS_20US) +
1286                          ((1 << data->irt) * 30)) * 1000;
1287         } else /* ACPI _PSS objects available */
1288                 pol->cpuinfo.transition_latency = get_transition_latency(data);
1289
1290         /* only run on specific CPU from here on */
1291         oldmask = current->cpus_allowed;
1292         set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
1293
1294         if (smp_processor_id() != pol->cpu) {
1295                 printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
1296                 goto err_out_unmask;
1297         }
1298
1299         if (pending_bit_stuck()) {
1300                 printk(KERN_ERR PFX "failing init, change pending bit set\n");
1301                 goto err_out_unmask;
1302         }
1303
1304         if (query_current_values_with_pending_wait(data))
1305                 goto err_out_unmask;
1306
1307         if (cpu_family == CPU_OPTERON)
1308                 fidvid_msr_init();
1309
1310         /* run on any CPU again */
1311         set_cpus_allowed_ptr(current, &oldmask);
1312
1313         if (cpu_family == CPU_HW_PSTATE)
1314                 cpumask_copy(pol->cpus, cpumask_of(pol->cpu));
1315         else
1316                 cpumask_copy(pol->cpus, cpu_core_mask(pol->cpu));
1317         data->available_cores = pol->cpus;
1318
1319         if (cpu_family == CPU_HW_PSTATE)
1320                 pol->cur = find_khz_freq_from_pstate(data->powernow_table,
1321                                 data->currpstate);
1322         else
1323                 pol->cur = find_khz_freq_from_fid(data->currfid);
1324         dprintk("policy current frequency %d kHz\n", pol->cur);
1325
1326         /* min/max the cpu is capable of */
1327         if (cpufreq_frequency_table_cpuinfo(pol, data->powernow_table)) {
1328                 printk(KERN_ERR FW_BUG PFX "invalid powernow_table\n");
1329                 powernow_k8_cpu_exit_acpi(data);
1330                 kfree(data->powernow_table);
1331                 kfree(data);
1332                 return -EINVAL;
1333         }
1334
1335         cpufreq_frequency_table_get_attr(data->powernow_table, pol->cpu);
1336
1337         if (cpu_family == CPU_HW_PSTATE)
1338                 dprintk("cpu_init done, current pstate 0x%x\n",
1339                                 data->currpstate);
1340         else
1341                 dprintk("cpu_init done, current fid 0x%x, vid 0x%x\n",
1342                         data->currfid, data->currvid);
1343
1344         per_cpu(powernow_data, pol->cpu) = data;
1345
1346         return 0;
1347
1348 err_out_unmask:
1349         set_cpus_allowed_ptr(current, &oldmask);
1350         powernow_k8_cpu_exit_acpi(data);
1351
1352 err_out:
1353         kfree(data);
1354         return -ENODEV;
1355 }
1356
1357 static int __devexit powernowk8_cpu_exit(struct cpufreq_policy *pol)
1358 {
1359         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1360
1361         if (!data)
1362                 return -EINVAL;
1363
1364         powernow_k8_cpu_exit_acpi(data);
1365
1366         cpufreq_frequency_table_put_attr(pol->cpu);
1367
1368         kfree(data->powernow_table);
1369         kfree(data);
1370
1371         return 0;
1372 }
1373
1374 static unsigned int powernowk8_get(unsigned int cpu)
1375 {
1376         struct powernow_k8_data *data;
1377         cpumask_t oldmask = current->cpus_allowed;
1378         unsigned int khz = 0;
1379         unsigned int first;
1380
1381         first = cpumask_first(cpu_core_mask(cpu));
1382         data = per_cpu(powernow_data, first);
1383
1384         if (!data)
1385                 return -EINVAL;
1386
1387         set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
1388         if (smp_processor_id() != cpu) {
1389                 printk(KERN_ERR PFX
1390                         "limiting to CPU %d failed in powernowk8_get\n", cpu);
1391                 set_cpus_allowed_ptr(current, &oldmask);
1392                 return 0;
1393         }
1394
1395         if (query_current_values_with_pending_wait(data))
1396                 goto out;
1397
1398         if (cpu_family == CPU_HW_PSTATE)
1399                 khz = find_khz_freq_from_pstate(data->powernow_table,
1400                                                 data->currpstate);
1401         else
1402                 khz = find_khz_freq_from_fid(data->currfid);
1403
1404
1405 out:
1406         set_cpus_allowed_ptr(current, &oldmask);
1407         return khz;
1408 }
1409
1410 static struct freq_attr *powernow_k8_attr[] = {
1411         &cpufreq_freq_attr_scaling_available_freqs,
1412         NULL,
1413 };
1414
1415 static struct cpufreq_driver cpufreq_amd64_driver = {
1416         .verify = powernowk8_verify,
1417         .target = powernowk8_target,
1418         .init = powernowk8_cpu_init,
1419         .exit = __devexit_p(powernowk8_cpu_exit),
1420         .get = powernowk8_get,
1421         .name = "powernow-k8",
1422         .owner = THIS_MODULE,
1423         .attr = powernow_k8_attr,
1424 };
1425
1426 /* driver entry point for init */
1427 static int __cpuinit powernowk8_init(void)
1428 {
1429         unsigned int i, supported_cpus = 0;
1430
1431         for_each_online_cpu(i) {
1432                 if (check_supported_cpu(i))
1433                         supported_cpus++;
1434         }
1435
1436         if (supported_cpus == num_online_cpus()) {
1437                 printk(KERN_INFO PFX "Found %d %s "
1438                         "processors (%d cpu cores) (" VERSION ")\n",
1439                         num_online_nodes(),
1440                         boot_cpu_data.x86_model_id, supported_cpus);
1441                 return cpufreq_register_driver(&cpufreq_amd64_driver);
1442         }
1443
1444         return -ENODEV;
1445 }
1446
1447 /* driver entry point for term */
1448 static void __exit powernowk8_exit(void)
1449 {
1450         dprintk("exit\n");
1451
1452         cpufreq_unregister_driver(&cpufreq_amd64_driver);
1453 }
1454
1455 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com> and "
1456                 "Mark Langsdorf <mark.langsdorf@amd.com>");
1457 MODULE_DESCRIPTION("AMD Athlon 64 and Opteron processor frequency driver.");
1458 MODULE_LICENSE("GPL");
1459
1460 late_initcall(powernowk8_init);
1461 module_exit(powernowk8_exit);