[CPUFREQ] fix timer teardown in ondemand governor
[sfrench/cifs-2.6.git] / 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 int fill_powernow_table(struct powernow_k8_data *data,
653                 struct pst_s *pst, u8 maxvid)
654 {
655         struct cpufreq_frequency_table *powernow_table;
656         unsigned int j;
657
658         if (data->batps) {
659                 /* use ACPI support to get full speed on mains power */
660                 printk(KERN_WARNING PFX
661                         "Only %d pstates usable (use ACPI driver for full "
662                         "range\n", data->batps);
663                 data->numps = data->batps;
664         }
665
666         for (j = 1; j < data->numps; j++) {
667                 if (pst[j-1].fid >= pst[j].fid) {
668                         printk(KERN_ERR PFX "PST out of sequence\n");
669                         return -EINVAL;
670                 }
671         }
672
673         if (data->numps < 2) {
674                 printk(KERN_ERR PFX "no p states to transition\n");
675                 return -ENODEV;
676         }
677
678         if (check_pst_table(data, pst, maxvid))
679                 return -EINVAL;
680
681         powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
682                 * (data->numps + 1)), GFP_KERNEL);
683         if (!powernow_table) {
684                 printk(KERN_ERR PFX "powernow_table memory alloc failure\n");
685                 return -ENOMEM;
686         }
687
688         for (j = 0; j < data->numps; j++) {
689                 int freq;
690                 powernow_table[j].index = pst[j].fid; /* lower 8 bits */
691                 powernow_table[j].index |= (pst[j].vid << 8); /* upper 8 bits */
692                 freq = find_khz_freq_from_fid(pst[j].fid);
693                 powernow_table[j].frequency = freq;
694         }
695         powernow_table[data->numps].frequency = CPUFREQ_TABLE_END;
696         powernow_table[data->numps].index = 0;
697
698         if (query_current_values_with_pending_wait(data)) {
699                 kfree(powernow_table);
700                 return -EIO;
701         }
702
703         dprintk("cfid 0x%x, cvid 0x%x\n", data->currfid, data->currvid);
704         data->powernow_table = powernow_table;
705         if (cpumask_first(cpu_core_mask(data->cpu)) == data->cpu)
706                 print_basics(data);
707
708         for (j = 0; j < data->numps; j++)
709                 if ((pst[j].fid == data->currfid) &&
710                     (pst[j].vid == data->currvid))
711                         return 0;
712
713         dprintk("currfid/vid do not match PST, ignoring\n");
714         return 0;
715 }
716
717 /* Find and validate the PSB/PST table in BIOS. */
718 static int find_psb_table(struct powernow_k8_data *data)
719 {
720         struct psb_s *psb;
721         unsigned int i;
722         u32 mvs;
723         u8 maxvid;
724         u32 cpst = 0;
725         u32 thiscpuid;
726
727         for (i = 0xc0000; i < 0xffff0; i += 0x10) {
728                 /* Scan BIOS looking for the signature. */
729                 /* It can not be at ffff0 - it is too big. */
730
731                 psb = phys_to_virt(i);
732                 if (memcmp(psb, PSB_ID_STRING, PSB_ID_STRING_LEN) != 0)
733                         continue;
734
735                 dprintk("found PSB header at 0x%p\n", psb);
736
737                 dprintk("table vers: 0x%x\n", psb->tableversion);
738                 if (psb->tableversion != PSB_VERSION_1_4) {
739                         printk(KERN_ERR FW_BUG PFX "PSB table is not v1.4\n");
740                         return -ENODEV;
741                 }
742
743                 dprintk("flags: 0x%x\n", psb->flags1);
744                 if (psb->flags1) {
745                         printk(KERN_ERR FW_BUG PFX "unknown flags\n");
746                         return -ENODEV;
747                 }
748
749                 data->vstable = psb->vstable;
750                 dprintk("voltage stabilization time: %d(*20us)\n",
751                                 data->vstable);
752
753                 dprintk("flags2: 0x%x\n", psb->flags2);
754                 data->rvo = psb->flags2 & 3;
755                 data->irt = ((psb->flags2) >> 2) & 3;
756                 mvs = ((psb->flags2) >> 4) & 3;
757                 data->vidmvs = 1 << mvs;
758                 data->batps = ((psb->flags2) >> 6) & 3;
759
760                 dprintk("ramp voltage offset: %d\n", data->rvo);
761                 dprintk("isochronous relief time: %d\n", data->irt);
762                 dprintk("maximum voltage step: %d - 0x%x\n", mvs, data->vidmvs);
763
764                 dprintk("numpst: 0x%x\n", psb->num_tables);
765                 cpst = psb->num_tables;
766                 if ((psb->cpuid == 0x00000fc0) ||
767                     (psb->cpuid == 0x00000fe0)) {
768                         thiscpuid = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
769                         if ((thiscpuid == 0x00000fc0) ||
770                             (thiscpuid == 0x00000fe0))
771                                 cpst = 1;
772                 }
773                 if (cpst != 1) {
774                         printk(KERN_ERR FW_BUG PFX "numpst must be 1\n");
775                         return -ENODEV;
776                 }
777
778                 data->plllock = psb->plllocktime;
779                 dprintk("plllocktime: 0x%x (units 1us)\n", psb->plllocktime);
780                 dprintk("maxfid: 0x%x\n", psb->maxfid);
781                 dprintk("maxvid: 0x%x\n", psb->maxvid);
782                 maxvid = psb->maxvid;
783
784                 data->numps = psb->numps;
785                 dprintk("numpstates: 0x%x\n", data->numps);
786                 return fill_powernow_table(data,
787                                 (struct pst_s *)(psb+1), maxvid);
788         }
789         /*
790          * If you see this message, complain to BIOS manufacturer. If
791          * he tells you "we do not support Linux" or some similar
792          * nonsense, remember that Windows 2000 uses the same legacy
793          * mechanism that the old Linux PSB driver uses. Tell them it
794          * is broken with Windows 2000.
795          *
796          * The reference to the AMD documentation is chapter 9 in the
797          * BIOS and Kernel Developer's Guide, which is available on
798          * www.amd.com
799          */
800         printk(KERN_ERR FW_BUG PFX "No PSB or ACPI _PSS objects\n");
801         return -ENODEV;
802 }
803
804 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data,
805                 unsigned int index)
806 {
807         acpi_integer control;
808
809         if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE))
810                 return;
811
812         control = data->acpi_data.states[index].control; data->irt = (control
813                         >> IRT_SHIFT) & IRT_MASK; data->rvo = (control >>
814                                 RVO_SHIFT) & RVO_MASK; data->exttype = (control
815                                         >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
816         data->plllock = (control >> PLL_L_SHIFT) & PLL_L_MASK; data->vidmvs = 1
817                 << ((control >> MVS_SHIFT) & MVS_MASK); data->vstable =
818                 (control >> VST_SHIFT) & VST_MASK; }
819
820 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
821 {
822         struct cpufreq_frequency_table *powernow_table;
823         int ret_val = -ENODEV;
824         acpi_integer space_id;
825
826         if (acpi_processor_register_performance(&data->acpi_data, data->cpu)) {
827                 dprintk("register performance failed: bad ACPI data\n");
828                 return -EIO;
829         }
830
831         /* verify the data contained in the ACPI structures */
832         if (data->acpi_data.state_count <= 1) {
833                 dprintk("No ACPI P-States\n");
834                 goto err_out;
835         }
836
837         space_id = data->acpi_data.control_register.space_id;
838         if ((space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
839                 (space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
840                 dprintk("Invalid control/status registers (%x - %x)\n",
841                         data->acpi_data.control_register.space_id,
842                         space_id);
843                 goto err_out;
844         }
845
846         /* fill in data->powernow_table */
847         powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
848                 * (data->acpi_data.state_count + 1)), GFP_KERNEL);
849         if (!powernow_table) {
850                 dprintk("powernow_table memory alloc failure\n");
851                 goto err_out;
852         }
853
854         if (cpu_family == CPU_HW_PSTATE)
855                 ret_val = fill_powernow_table_pstate(data, powernow_table);
856         else
857                 ret_val = fill_powernow_table_fidvid(data, powernow_table);
858         if (ret_val)
859                 goto err_out_mem;
860
861         powernow_table[data->acpi_data.state_count].frequency =
862                 CPUFREQ_TABLE_END;
863         powernow_table[data->acpi_data.state_count].index = 0;
864         data->powernow_table = powernow_table;
865
866         /* fill in data */
867         data->numps = data->acpi_data.state_count;
868         if (cpumask_first(cpu_core_mask(data->cpu)) == data->cpu)
869                 print_basics(data);
870         powernow_k8_acpi_pst_values(data, 0);
871
872         /* notify BIOS that we exist */
873         acpi_processor_notify_smm(THIS_MODULE);
874
875         if (!alloc_cpumask_var(&data->acpi_data.shared_cpu_map, GFP_KERNEL)) {
876                 printk(KERN_ERR PFX
877                                 "unable to alloc powernow_k8_data cpumask\n");
878                 ret_val = -ENOMEM;
879                 goto err_out_mem;
880         }
881
882         return 0;
883
884 err_out_mem:
885         kfree(powernow_table);
886
887 err_out:
888         acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
889
890         /* data->acpi_data.state_count informs us at ->exit()
891          * whether ACPI was used */
892         data->acpi_data.state_count = 0;
893
894         return ret_val;
895 }
896
897 static int fill_powernow_table_pstate(struct powernow_k8_data *data,
898                 struct cpufreq_frequency_table *powernow_table)
899 {
900         int i;
901         u32 hi = 0, lo = 0;
902         rdmsr(MSR_PSTATE_CUR_LIMIT, hi, lo);
903         data->max_hw_pstate = (hi & HW_PSTATE_MAX_MASK) >> HW_PSTATE_MAX_SHIFT;
904
905         for (i = 0; i < data->acpi_data.state_count; i++) {
906                 u32 index;
907
908                 index = data->acpi_data.states[i].control & HW_PSTATE_MASK;
909                 if (index > data->max_hw_pstate) {
910                         printk(KERN_ERR PFX "invalid pstate %d - "
911                                         "bad value %d.\n", i, index);
912                         printk(KERN_ERR PFX "Please report to BIOS "
913                                         "manufacturer\n");
914                         invalidate_entry(data, i);
915                         continue;
916                 }
917                 rdmsr(MSR_PSTATE_DEF_BASE + index, lo, hi);
918                 if (!(hi & HW_PSTATE_VALID_MASK)) {
919                         dprintk("invalid pstate %d, ignoring\n", index);
920                         invalidate_entry(data, i);
921                         continue;
922                 }
923
924                 powernow_table[i].index = index;
925
926                 powernow_table[i].frequency =
927                         data->acpi_data.states[i].core_frequency * 1000;
928         }
929         return 0;
930 }
931
932 static int fill_powernow_table_fidvid(struct powernow_k8_data *data,
933                 struct cpufreq_frequency_table *powernow_table)
934 {
935         int i;
936         int cntlofreq = 0;
937
938         for (i = 0; i < data->acpi_data.state_count; i++) {
939                 u32 fid;
940                 u32 vid;
941                 u32 freq, index;
942                 acpi_integer status, control;
943
944                 if (data->exttype) {
945                         status =  data->acpi_data.states[i].status;
946                         fid = status & EXT_FID_MASK;
947                         vid = (status >> VID_SHIFT) & EXT_VID_MASK;
948                 } else {
949                         control =  data->acpi_data.states[i].control;
950                         fid = control & FID_MASK;
951                         vid = (control >> VID_SHIFT) & VID_MASK;
952                 }
953
954                 dprintk("   %d : fid 0x%x, vid 0x%x\n", i, fid, vid);
955
956                 index = fid | (vid<<8);
957                 powernow_table[i].index = index;
958
959                 freq = find_khz_freq_from_fid(fid);
960                 powernow_table[i].frequency = freq;
961
962                 /* verify frequency is OK */
963                 if ((freq > (MAX_FREQ * 1000)) || (freq < (MIN_FREQ * 1000))) {
964                         dprintk("invalid freq %u kHz, ignoring\n", freq);
965                         invalidate_entry(data, i);
966                         continue;
967                 }
968
969                 /* verify voltage is OK -
970                  * BIOSs are using "off" to indicate invalid */
971                 if (vid == VID_OFF) {
972                         dprintk("invalid vid %u, ignoring\n", vid);
973                         invalidate_entry(data, i);
974                         continue;
975                 }
976
977                 /* verify only 1 entry from the lo frequency table */
978                 if (fid < HI_FID_TABLE_BOTTOM) {
979                         if (cntlofreq) {
980                                 /* if both entries are the same,
981                                  * ignore this one ... */
982                                 if ((freq != powernow_table[cntlofreq].frequency) ||
983                                     (index != powernow_table[cntlofreq].index)) {
984                                         printk(KERN_ERR PFX
985                                                 "Too many lo freq table "
986                                                 "entries\n");
987                                         return 1;
988                                 }
989
990                                 dprintk("double low frequency table entry, "
991                                                 "ignoring it.\n");
992                                 invalidate_entry(data, i);
993                                 continue;
994                         } else
995                                 cntlofreq = i;
996                 }
997
998                 if (freq != (data->acpi_data.states[i].core_frequency * 1000)) {
999                         printk(KERN_INFO PFX "invalid freq entries "
1000                                 "%u kHz vs. %u kHz\n", freq,
1001                                 (unsigned int)
1002                                 (data->acpi_data.states[i].core_frequency
1003                                  * 1000));
1004                         invalidate_entry(data, i);
1005                         continue;
1006                 }
1007         }
1008         return 0;
1009 }
1010
1011 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data)
1012 {
1013         if (data->acpi_data.state_count)
1014                 acpi_processor_unregister_performance(&data->acpi_data,
1015                                 data->cpu);
1016         free_cpumask_var(data->acpi_data.shared_cpu_map);
1017 }
1018
1019 static int get_transition_latency(struct powernow_k8_data *data)
1020 {
1021         int max_latency = 0;
1022         int i;
1023         for (i = 0; i < data->acpi_data.state_count; i++) {
1024                 int cur_latency = data->acpi_data.states[i].transition_latency
1025                         + data->acpi_data.states[i].bus_master_latency;
1026                 if (cur_latency > max_latency)
1027                         max_latency = cur_latency;
1028         }
1029         /* value in usecs, needs to be in nanoseconds */
1030         return 1000 * max_latency;
1031 }
1032
1033 /* Take a frequency, and issue the fid/vid transition command */
1034 static int transition_frequency_fidvid(struct powernow_k8_data *data,
1035                 unsigned int index)
1036 {
1037         u32 fid = 0;
1038         u32 vid = 0;
1039         int res, i;
1040         struct cpufreq_freqs freqs;
1041
1042         dprintk("cpu %d transition to index %u\n", smp_processor_id(), index);
1043
1044         /* fid/vid correctness check for k8 */
1045         /* fid are the lower 8 bits of the index we stored into
1046          * the cpufreq frequency table in find_psb_table, vid
1047          * are the upper 8 bits.
1048          */
1049         fid = data->powernow_table[index].index & 0xFF;
1050         vid = (data->powernow_table[index].index & 0xFF00) >> 8;
1051
1052         dprintk("table matched fid 0x%x, giving vid 0x%x\n", fid, vid);
1053
1054         if (query_current_values_with_pending_wait(data))
1055                 return 1;
1056
1057         if ((data->currvid == vid) && (data->currfid == fid)) {
1058                 dprintk("target matches current values (fid 0x%x, vid 0x%x)\n",
1059                         fid, vid);
1060                 return 0;
1061         }
1062
1063         if ((fid < HI_FID_TABLE_BOTTOM) &&
1064             (data->currfid < HI_FID_TABLE_BOTTOM)) {
1065                 printk(KERN_ERR PFX
1066                        "ignoring illegal change in lo freq table-%x to 0x%x\n",
1067                        data->currfid, fid);
1068                 return 1;
1069         }
1070
1071         dprintk("cpu %d, changing to fid 0x%x, vid 0x%x\n",
1072                 smp_processor_id(), fid, vid);
1073         freqs.old = find_khz_freq_from_fid(data->currfid);
1074         freqs.new = find_khz_freq_from_fid(fid);
1075
1076         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1077                 freqs.cpu = i;
1078                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1079         }
1080
1081         res = transition_fid_vid(data, fid, vid);
1082         freqs.new = find_khz_freq_from_fid(data->currfid);
1083
1084         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1085                 freqs.cpu = i;
1086                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1087         }
1088         return res;
1089 }
1090
1091 /* Take a frequency, and issue the hardware pstate transition command */
1092 static int transition_frequency_pstate(struct powernow_k8_data *data,
1093                 unsigned int index)
1094 {
1095         u32 pstate = 0;
1096         int res, i;
1097         struct cpufreq_freqs freqs;
1098
1099         dprintk("cpu %d transition to index %u\n", smp_processor_id(), index);
1100
1101         /* get MSR index for hardware pstate transition */
1102         pstate = index & HW_PSTATE_MASK;
1103         if (pstate > data->max_hw_pstate)
1104                 return 0;
1105         freqs.old = find_khz_freq_from_pstate(data->powernow_table,
1106                         data->currpstate);
1107         freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
1108
1109         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1110                 freqs.cpu = i;
1111                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1112         }
1113
1114         res = transition_pstate(data, pstate);
1115         freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
1116
1117         for_each_cpu_mask_nr(i, *(data->available_cores)) {
1118                 freqs.cpu = i;
1119                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1120         }
1121         return res;
1122 }
1123
1124 /* Driver entry point to switch to the target frequency */
1125 static int powernowk8_target(struct cpufreq_policy *pol,
1126                 unsigned targfreq, unsigned relation)
1127 {
1128         cpumask_t oldmask;
1129         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1130         u32 checkfid;
1131         u32 checkvid;
1132         unsigned int newstate;
1133         int ret = -EIO;
1134
1135         if (!data)
1136                 return -EINVAL;
1137
1138         checkfid = data->currfid;
1139         checkvid = data->currvid;
1140
1141         /* only run on specific CPU from here on */
1142         oldmask = current->cpus_allowed;
1143         set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
1144
1145         if (smp_processor_id() != pol->cpu) {
1146                 printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
1147                 goto err_out;
1148         }
1149
1150         if (pending_bit_stuck()) {
1151                 printk(KERN_ERR PFX "failing targ, change pending bit set\n");
1152                 goto err_out;
1153         }
1154
1155         dprintk("targ: cpu %d, %d kHz, min %d, max %d, relation %d\n",
1156                 pol->cpu, targfreq, pol->min, pol->max, relation);
1157
1158         if (query_current_values_with_pending_wait(data))
1159                 goto err_out;
1160
1161         if (cpu_family != CPU_HW_PSTATE) {
1162                 dprintk("targ: curr fid 0x%x, vid 0x%x\n",
1163                 data->currfid, data->currvid);
1164
1165                 if ((checkvid != data->currvid) ||
1166                     (checkfid != data->currfid)) {
1167                         printk(KERN_INFO PFX
1168                                 "error - out of sync, fix 0x%x 0x%x, "
1169                                 "vid 0x%x 0x%x\n",
1170                                 checkfid, data->currfid,
1171                                 checkvid, data->currvid);
1172                 }
1173         }
1174
1175         if (cpufreq_frequency_table_target(pol, data->powernow_table,
1176                                 targfreq, relation, &newstate))
1177                 goto err_out;
1178
1179         mutex_lock(&fidvid_mutex);
1180
1181         powernow_k8_acpi_pst_values(data, newstate);
1182
1183         if (cpu_family == CPU_HW_PSTATE)
1184                 ret = transition_frequency_pstate(data, newstate);
1185         else
1186                 ret = transition_frequency_fidvid(data, newstate);
1187         if (ret) {
1188                 printk(KERN_ERR PFX "transition frequency failed\n");
1189                 ret = 1;
1190                 mutex_unlock(&fidvid_mutex);
1191                 goto err_out;
1192         }
1193         mutex_unlock(&fidvid_mutex);
1194
1195         if (cpu_family == CPU_HW_PSTATE)
1196                 pol->cur = find_khz_freq_from_pstate(data->powernow_table,
1197                                 newstate);
1198         else
1199                 pol->cur = find_khz_freq_from_fid(data->currfid);
1200         ret = 0;
1201
1202 err_out:
1203         set_cpus_allowed_ptr(current, &oldmask);
1204         return ret;
1205 }
1206
1207 /* Driver entry point to verify the policy and range of frequencies */
1208 static int powernowk8_verify(struct cpufreq_policy *pol)
1209 {
1210         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1211
1212         if (!data)
1213                 return -EINVAL;
1214
1215         return cpufreq_frequency_table_verify(pol, data->powernow_table);
1216 }
1217
1218 /* per CPU init entry point to the driver */
1219 static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
1220 {
1221         struct powernow_k8_data *data;
1222         cpumask_t oldmask;
1223         int rc;
1224         static int print_once;
1225
1226         if (!cpu_online(pol->cpu))
1227                 return -ENODEV;
1228
1229         if (!check_supported_cpu(pol->cpu))
1230                 return -ENODEV;
1231
1232         data = kzalloc(sizeof(struct powernow_k8_data), GFP_KERNEL);
1233         if (!data) {
1234                 printk(KERN_ERR PFX "unable to alloc powernow_k8_data");
1235                 return -ENOMEM;
1236         }
1237
1238         data->cpu = pol->cpu;
1239         data->currpstate = HW_PSTATE_INVALID;
1240
1241         if (powernow_k8_cpu_init_acpi(data)) {
1242                 /*
1243                  * Use the PSB BIOS structure. This is only availabe on
1244                  * an UP version, and is deprecated by AMD.
1245                  */
1246                 if (num_online_cpus() != 1) {
1247                         /*
1248                          * Replace this one with print_once as soon as such a
1249                          * thing gets introduced
1250                          */
1251                         if (!print_once) {
1252                                 WARN_ONCE(1, KERN_ERR FW_BUG PFX "Your BIOS "
1253                                         "does not provide ACPI _PSS objects "
1254                                         "in a way that Linux understands. "
1255                                         "Please report this to the Linux ACPI"
1256                                         " maintainers and complain to your "
1257                                         "BIOS vendor.\n");
1258                                 print_once++;
1259                         }
1260                         goto err_out;
1261                 }
1262                 if (pol->cpu != 0) {
1263                         printk(KERN_ERR FW_BUG PFX "No ACPI _PSS objects for "
1264                                "CPU other than CPU0. Complain to your BIOS "
1265                                "vendor.\n");
1266                         goto err_out;
1267                 }
1268                 rc = find_psb_table(data);
1269                 if (rc)
1270                         goto err_out;
1271
1272                 /* Take a crude guess here.
1273                  * That guess was in microseconds, so multiply with 1000 */
1274                 pol->cpuinfo.transition_latency = (
1275                          ((data->rvo + 8) * data->vstable * VST_UNITS_20US) +
1276                          ((1 << data->irt) * 30)) * 1000;
1277         } else /* ACPI _PSS objects available */
1278                 pol->cpuinfo.transition_latency = get_transition_latency(data);
1279
1280         /* only run on specific CPU from here on */
1281         oldmask = current->cpus_allowed;
1282         set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
1283
1284         if (smp_processor_id() != pol->cpu) {
1285                 printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
1286                 goto err_out_unmask;
1287         }
1288
1289         if (pending_bit_stuck()) {
1290                 printk(KERN_ERR PFX "failing init, change pending bit set\n");
1291                 goto err_out_unmask;
1292         }
1293
1294         if (query_current_values_with_pending_wait(data))
1295                 goto err_out_unmask;
1296
1297         if (cpu_family == CPU_OPTERON)
1298                 fidvid_msr_init();
1299
1300         /* run on any CPU again */
1301         set_cpus_allowed_ptr(current, &oldmask);
1302
1303         if (cpu_family == CPU_HW_PSTATE)
1304                 cpumask_copy(pol->cpus, cpumask_of(pol->cpu));
1305         else
1306                 cpumask_copy(pol->cpus, cpu_core_mask(pol->cpu));
1307         data->available_cores = pol->cpus;
1308
1309         if (cpu_family == CPU_HW_PSTATE)
1310                 pol->cur = find_khz_freq_from_pstate(data->powernow_table,
1311                                 data->currpstate);
1312         else
1313                 pol->cur = find_khz_freq_from_fid(data->currfid);
1314         dprintk("policy current frequency %d kHz\n", pol->cur);
1315
1316         /* min/max the cpu is capable of */
1317         if (cpufreq_frequency_table_cpuinfo(pol, data->powernow_table)) {
1318                 printk(KERN_ERR FW_BUG PFX "invalid powernow_table\n");
1319                 powernow_k8_cpu_exit_acpi(data);
1320                 kfree(data->powernow_table);
1321                 kfree(data);
1322                 return -EINVAL;
1323         }
1324
1325         cpufreq_frequency_table_get_attr(data->powernow_table, pol->cpu);
1326
1327         if (cpu_family == CPU_HW_PSTATE)
1328                 dprintk("cpu_init done, current pstate 0x%x\n",
1329                                 data->currpstate);
1330         else
1331                 dprintk("cpu_init done, current fid 0x%x, vid 0x%x\n",
1332                         data->currfid, data->currvid);
1333
1334         per_cpu(powernow_data, pol->cpu) = data;
1335
1336         return 0;
1337
1338 err_out_unmask:
1339         set_cpus_allowed_ptr(current, &oldmask);
1340         powernow_k8_cpu_exit_acpi(data);
1341
1342 err_out:
1343         kfree(data);
1344         return -ENODEV;
1345 }
1346
1347 static int __devexit powernowk8_cpu_exit(struct cpufreq_policy *pol)
1348 {
1349         struct powernow_k8_data *data = per_cpu(powernow_data, pol->cpu);
1350
1351         if (!data)
1352                 return -EINVAL;
1353
1354         powernow_k8_cpu_exit_acpi(data);
1355
1356         cpufreq_frequency_table_put_attr(pol->cpu);
1357
1358         kfree(data->powernow_table);
1359         kfree(data);
1360
1361         return 0;
1362 }
1363
1364 static unsigned int powernowk8_get(unsigned int cpu)
1365 {
1366         struct powernow_k8_data *data;
1367         cpumask_t oldmask = current->cpus_allowed;
1368         unsigned int khz = 0;
1369         unsigned int first;
1370
1371         first = cpumask_first(cpu_core_mask(cpu));
1372         data = per_cpu(powernow_data, first);
1373
1374         if (!data)
1375                 return -EINVAL;
1376
1377         set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
1378         if (smp_processor_id() != cpu) {
1379                 printk(KERN_ERR PFX
1380                         "limiting to CPU %d failed in powernowk8_get\n", cpu);
1381                 set_cpus_allowed_ptr(current, &oldmask);
1382                 return 0;
1383         }
1384
1385         if (query_current_values_with_pending_wait(data))
1386                 goto out;
1387
1388         if (cpu_family == CPU_HW_PSTATE)
1389                 khz = find_khz_freq_from_pstate(data->powernow_table,
1390                                                 data->currpstate);
1391         else
1392                 khz = find_khz_freq_from_fid(data->currfid);
1393
1394
1395 out:
1396         set_cpus_allowed_ptr(current, &oldmask);
1397         return khz;
1398 }
1399
1400 static struct freq_attr *powernow_k8_attr[] = {
1401         &cpufreq_freq_attr_scaling_available_freqs,
1402         NULL,
1403 };
1404
1405 static struct cpufreq_driver cpufreq_amd64_driver = {
1406         .verify = powernowk8_verify,
1407         .target = powernowk8_target,
1408         .init = powernowk8_cpu_init,
1409         .exit = __devexit_p(powernowk8_cpu_exit),
1410         .get = powernowk8_get,
1411         .name = "powernow-k8",
1412         .owner = THIS_MODULE,
1413         .attr = powernow_k8_attr,
1414 };
1415
1416 /* driver entry point for init */
1417 static int __cpuinit powernowk8_init(void)
1418 {
1419         unsigned int i, supported_cpus = 0;
1420
1421         for_each_online_cpu(i) {
1422                 if (check_supported_cpu(i))
1423                         supported_cpus++;
1424         }
1425
1426         if (supported_cpus == num_online_cpus()) {
1427                 printk(KERN_INFO PFX "Found %d %s "
1428                         "processors (%d cpu cores) (" VERSION ")\n",
1429                         num_online_nodes(),
1430                         boot_cpu_data.x86_model_id, supported_cpus);
1431                 return cpufreq_register_driver(&cpufreq_amd64_driver);
1432         }
1433
1434         return -ENODEV;
1435 }
1436
1437 /* driver entry point for term */
1438 static void __exit powernowk8_exit(void)
1439 {
1440         dprintk("exit\n");
1441
1442         cpufreq_unregister_driver(&cpufreq_amd64_driver);
1443 }
1444
1445 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com> and "
1446                 "Mark Langsdorf <mark.langsdorf@amd.com>");
1447 MODULE_DESCRIPTION("AMD Athlon 64 and Opteron processor frequency driver.");
1448 MODULE_LICENSE("GPL");
1449
1450 late_initcall(powernowk8_init);
1451 module_exit(powernowk8_exit);