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
7 * Support : mark.langsdorf@amd.com
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.
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
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
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() */
39 #include <asm/delay.h>
41 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
42 #include <linux/acpi.h>
43 #include <linux/mutex.h>
44 #include <acpi/processor.h>
47 #define PFX "powernow-k8: "
48 #define VERSION "version 2.20.00"
49 #include "powernow-k8.h"
51 /* serialize freq changes */
52 static DEFINE_MUTEX(fidvid_mutex
);
54 static DEFINE_PER_CPU(struct powernow_k8_data
*, powernow_data
);
56 static int cpu_family
= CPU_OPTERON
;
59 DEFINE_PER_CPU(cpumask_t
, cpu_core_map
);
62 /* Return a frequency in MHz, given an input fid */
63 static u32
find_freq_from_fid(u32 fid
)
65 return 800 + (fid
* 100);
68 /* Return a frequency in KHz, given an input fid */
69 static u32
find_khz_freq_from_fid(u32 fid
)
71 return 1000 * find_freq_from_fid(fid
);
74 static u32
find_khz_freq_from_pstate(struct cpufreq_frequency_table
*data
, u32 pstate
)
76 return data
[pstate
].frequency
;
79 /* Return the vco fid for an input fid
81 * Each "low" fid has corresponding "high" fid, and you can get to "low" fids
82 * only from corresponding high fids. This returns "high" fid corresponding to
85 static u32
convert_fid_to_vco_fid(u32 fid
)
87 if (fid
< HI_FID_TABLE_BOTTOM
)
94 * Return 1 if the pending bit is set. Unless we just instructed the processor
95 * to transition to a new state, seeing this bit set is really bad news.
97 static int pending_bit_stuck(void)
101 if (cpu_family
== CPU_HW_PSTATE
)
104 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
105 return lo
& MSR_S_LO_CHANGE_PENDING
? 1 : 0;
109 * Update the global current fid / vid values from the status msr.
110 * Returns 1 on error.
112 static int query_current_values_with_pending_wait(struct powernow_k8_data
*data
)
117 if (cpu_family
== CPU_HW_PSTATE
) {
118 if (data
->currpstate
== HW_PSTATE_INVALID
) {
119 /* read (initial) hw pstate if not yet set */
120 rdmsr(MSR_PSTATE_STATUS
, lo
, hi
);
121 i
= lo
& HW_PSTATE_MASK
;
124 * a workaround for family 11h erratum 311 might cause
125 * an "out-of-range Pstate if the core is in Pstate-0
127 if (i
>= data
->numps
)
128 data
->currpstate
= HW_PSTATE_0
;
130 data
->currpstate
= i
;
136 dprintk("detected change pending stuck\n");
139 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
140 } while (lo
& MSR_S_LO_CHANGE_PENDING
);
142 data
->currvid
= hi
& MSR_S_HI_CURRENT_VID
;
143 data
->currfid
= lo
& MSR_S_LO_CURRENT_FID
;
148 /* the isochronous relief time */
149 static void count_off_irt(struct powernow_k8_data
*data
)
151 udelay((1 << data
->irt
) * 10);
155 /* the voltage stabilization time */
156 static void count_off_vst(struct powernow_k8_data
*data
)
158 udelay(data
->vstable
* VST_UNITS_20US
);
162 /* need to init the control msr to a safe value (for each cpu) */
163 static void fidvid_msr_init(void)
168 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
169 vid
= hi
& MSR_S_HI_CURRENT_VID
;
170 fid
= lo
& MSR_S_LO_CURRENT_FID
;
171 lo
= fid
| (vid
<< MSR_C_LO_VID_SHIFT
);
172 hi
= MSR_C_HI_STP_GNT_BENIGN
;
173 dprintk("cpu%d, init lo 0x%x, hi 0x%x\n", smp_processor_id(), lo
, hi
);
174 wrmsr(MSR_FIDVID_CTL
, lo
, hi
);
177 /* write the new fid value along with the other control fields to the msr */
178 static int write_new_fid(struct powernow_k8_data
*data
, u32 fid
)
181 u32 savevid
= data
->currvid
;
184 if ((fid
& INVALID_FID_MASK
) || (data
->currvid
& INVALID_VID_MASK
)) {
185 printk(KERN_ERR PFX
"internal error - overflow on fid write\n");
189 lo
= fid
| (data
->currvid
<< MSR_C_LO_VID_SHIFT
) | MSR_C_LO_INIT_FID_VID
;
191 dprintk("writing fid 0x%x, lo 0x%x, hi 0x%x\n",
192 fid
, lo
, data
->plllock
* PLL_LOCK_CONVERSION
);
195 wrmsr(MSR_FIDVID_CTL
, lo
, data
->plllock
* PLL_LOCK_CONVERSION
);
197 printk(KERN_ERR PFX
"Hardware error - pending bit very stuck - no further pstate changes possible\n");
200 } while (query_current_values_with_pending_wait(data
));
204 if (savevid
!= data
->currvid
) {
205 printk(KERN_ERR PFX
"vid change on fid trans, old 0x%x, new 0x%x\n",
206 savevid
, data
->currvid
);
210 if (fid
!= data
->currfid
) {
211 printk(KERN_ERR PFX
"fid trans failed, fid 0x%x, curr 0x%x\n", fid
,
219 /* Write a new vid to the hardware */
220 static int write_new_vid(struct powernow_k8_data
*data
, u32 vid
)
223 u32 savefid
= data
->currfid
;
226 if ((data
->currfid
& INVALID_FID_MASK
) || (vid
& INVALID_VID_MASK
)) {
227 printk(KERN_ERR PFX
"internal error - overflow on vid write\n");
231 lo
= data
->currfid
| (vid
<< MSR_C_LO_VID_SHIFT
) | MSR_C_LO_INIT_FID_VID
;
233 dprintk("writing vid 0x%x, lo 0x%x, hi 0x%x\n",
234 vid
, lo
, STOP_GRANT_5NS
);
237 wrmsr(MSR_FIDVID_CTL
, lo
, STOP_GRANT_5NS
);
239 printk(KERN_ERR PFX
"internal error - pending bit very stuck - no further pstate changes possible\n");
242 } while (query_current_values_with_pending_wait(data
));
244 if (savefid
!= data
->currfid
) {
245 printk(KERN_ERR PFX
"fid changed on vid trans, old 0x%x new 0x%x\n",
246 savefid
, data
->currfid
);
250 if (vid
!= data
->currvid
) {
251 printk(KERN_ERR PFX
"vid trans failed, vid 0x%x, curr 0x%x\n", vid
,
260 * Reduce the vid by the max of step or reqvid.
261 * Decreasing vid codes represent increasing voltages:
262 * vid of 0 is 1.550V, vid of 0x1e is 0.800V, vid of VID_OFF is off.
264 static int decrease_vid_code_by_step(struct powernow_k8_data
*data
, u32 reqvid
, u32 step
)
266 if ((data
->currvid
- reqvid
) > step
)
267 reqvid
= data
->currvid
- step
;
269 if (write_new_vid(data
, reqvid
))
277 /* Change hardware pstate by single MSR write */
278 static int transition_pstate(struct powernow_k8_data
*data
, u32 pstate
)
280 wrmsr(MSR_PSTATE_CTRL
, pstate
, 0);
281 data
->currpstate
= pstate
;
285 /* Change Opteron/Athlon64 fid and vid, by the 3 phases. */
286 static int transition_fid_vid(struct powernow_k8_data
*data
, u32 reqfid
, u32 reqvid
)
288 if (core_voltage_pre_transition(data
, reqvid
))
291 if (core_frequency_transition(data
, reqfid
))
294 if (core_voltage_post_transition(data
, reqvid
))
297 if (query_current_values_with_pending_wait(data
))
300 if ((reqfid
!= data
->currfid
) || (reqvid
!= data
->currvid
)) {
301 printk(KERN_ERR PFX
"failed (cpu%d): req 0x%x 0x%x, curr 0x%x 0x%x\n",
303 reqfid
, reqvid
, data
->currfid
, data
->currvid
);
307 dprintk("transitioned (cpu%d): new fid 0x%x, vid 0x%x\n",
308 smp_processor_id(), data
->currfid
, data
->currvid
);
313 /* Phase 1 - core voltage transition ... setup voltage */
314 static int core_voltage_pre_transition(struct powernow_k8_data
*data
, u32 reqvid
)
316 u32 rvosteps
= data
->rvo
;
317 u32 savefid
= data
->currfid
;
320 dprintk("ph1 (cpu%d): start, currfid 0x%x, currvid 0x%x, reqvid 0x%x, rvo 0x%x\n",
322 data
->currfid
, data
->currvid
, reqvid
, data
->rvo
);
324 rdmsr(MSR_FIDVID_STATUS
, lo
, maxvid
);
325 maxvid
= 0x1f & (maxvid
>> 16);
326 dprintk("ph1 maxvid=0x%x\n", maxvid
);
327 if (reqvid
< maxvid
) /* lower numbers are higher voltages */
330 while (data
->currvid
> reqvid
) {
331 dprintk("ph1: curr 0x%x, req vid 0x%x\n",
332 data
->currvid
, reqvid
);
333 if (decrease_vid_code_by_step(data
, reqvid
, data
->vidmvs
))
337 while ((rvosteps
> 0) && ((data
->rvo
+ data
->currvid
) > reqvid
)) {
338 if (data
->currvid
== maxvid
) {
341 dprintk("ph1: changing vid for rvo, req 0x%x\n",
343 if (decrease_vid_code_by_step(data
, data
->currvid
- 1, 1))
349 if (query_current_values_with_pending_wait(data
))
352 if (savefid
!= data
->currfid
) {
353 printk(KERN_ERR PFX
"ph1 err, currfid changed 0x%x\n", data
->currfid
);
357 dprintk("ph1 complete, currfid 0x%x, currvid 0x%x\n",
358 data
->currfid
, data
->currvid
);
363 /* Phase 2 - core frequency transition */
364 static int core_frequency_transition(struct powernow_k8_data
*data
, u32 reqfid
)
366 u32 vcoreqfid
, vcocurrfid
, vcofiddiff
, fid_interval
, savevid
= data
->currvid
;
368 if ((reqfid
< HI_FID_TABLE_BOTTOM
) && (data
->currfid
< HI_FID_TABLE_BOTTOM
)) {
369 printk(KERN_ERR PFX
"ph2: illegal lo-lo transition 0x%x 0x%x\n",
370 reqfid
, data
->currfid
);
374 if (data
->currfid
== reqfid
) {
375 printk(KERN_ERR PFX
"ph2 null fid transition 0x%x\n", data
->currfid
);
379 dprintk("ph2 (cpu%d): starting, currfid 0x%x, currvid 0x%x, reqfid 0x%x\n",
381 data
->currfid
, data
->currvid
, reqfid
);
383 vcoreqfid
= convert_fid_to_vco_fid(reqfid
);
384 vcocurrfid
= convert_fid_to_vco_fid(data
->currfid
);
385 vcofiddiff
= vcocurrfid
> vcoreqfid
? vcocurrfid
- vcoreqfid
386 : vcoreqfid
- vcocurrfid
;
388 while (vcofiddiff
> 2) {
389 (data
->currfid
& 1) ? (fid_interval
= 1) : (fid_interval
= 2);
391 if (reqfid
> data
->currfid
) {
392 if (data
->currfid
> LO_FID_TABLE_TOP
) {
393 if (write_new_fid(data
, data
->currfid
+ fid_interval
)) {
398 (data
, 2 + convert_fid_to_vco_fid(data
->currfid
))) {
403 if (write_new_fid(data
, data
->currfid
- fid_interval
))
407 vcocurrfid
= convert_fid_to_vco_fid(data
->currfid
);
408 vcofiddiff
= vcocurrfid
> vcoreqfid
? vcocurrfid
- vcoreqfid
409 : vcoreqfid
- vcocurrfid
;
412 if (write_new_fid(data
, reqfid
))
415 if (query_current_values_with_pending_wait(data
))
418 if (data
->currfid
!= reqfid
) {
420 "ph2: mismatch, failed fid transition, curr 0x%x, req 0x%x\n",
421 data
->currfid
, reqfid
);
425 if (savevid
!= data
->currvid
) {
426 printk(KERN_ERR PFX
"ph2: vid changed, save 0x%x, curr 0x%x\n",
427 savevid
, data
->currvid
);
431 dprintk("ph2 complete, currfid 0x%x, currvid 0x%x\n",
432 data
->currfid
, data
->currvid
);
437 /* Phase 3 - core voltage transition flow ... jump to the final vid. */
438 static int core_voltage_post_transition(struct powernow_k8_data
*data
, u32 reqvid
)
440 u32 savefid
= data
->currfid
;
441 u32 savereqvid
= reqvid
;
443 dprintk("ph3 (cpu%d): starting, currfid 0x%x, currvid 0x%x\n",
445 data
->currfid
, data
->currvid
);
447 if (reqvid
!= data
->currvid
) {
448 if (write_new_vid(data
, reqvid
))
451 if (savefid
!= data
->currfid
) {
453 "ph3: bad fid change, save 0x%x, curr 0x%x\n",
454 savefid
, data
->currfid
);
458 if (data
->currvid
!= reqvid
) {
460 "ph3: failed vid transition\n, req 0x%x, curr 0x%x",
461 reqvid
, data
->currvid
);
466 if (query_current_values_with_pending_wait(data
))
469 if (savereqvid
!= data
->currvid
) {
470 dprintk("ph3 failed, currvid 0x%x\n", data
->currvid
);
474 if (savefid
!= data
->currfid
) {
475 dprintk("ph3 failed, currfid changed 0x%x\n",
480 dprintk("ph3 complete, currfid 0x%x, currvid 0x%x\n",
481 data
->currfid
, data
->currvid
);
486 static int check_supported_cpu(unsigned int cpu
)
489 u32 eax
, ebx
, ecx
, edx
;
492 oldmask
= current
->cpus_allowed
;
493 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(cpu
));
495 if (smp_processor_id() != cpu
) {
496 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", cpu
);
500 if (current_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
)
503 eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
504 if (((eax
& CPUID_XFAM
) != CPUID_XFAM_K8
) &&
505 ((eax
& CPUID_XFAM
) < CPUID_XFAM_10H
))
508 if ((eax
& CPUID_XFAM
) == CPUID_XFAM_K8
) {
509 if (((eax
& CPUID_USE_XFAM_XMOD
) != CPUID_USE_XFAM_XMOD
) ||
510 ((eax
& CPUID_XMOD
) > CPUID_XMOD_REV_MASK
)) {
511 printk(KERN_INFO PFX
"Processor cpuid %x not supported\n", eax
);
515 eax
= cpuid_eax(CPUID_GET_MAX_CAPABILITIES
);
516 if (eax
< CPUID_FREQ_VOLT_CAPABILITIES
) {
518 "No frequency change capabilities detected\n");
522 cpuid(CPUID_FREQ_VOLT_CAPABILITIES
, &eax
, &ebx
, &ecx
, &edx
);
523 if ((edx
& P_STATE_TRANSITION_CAPABLE
) != P_STATE_TRANSITION_CAPABLE
) {
524 printk(KERN_INFO PFX
"Power state transitions not supported\n");
527 } else { /* must be a HW Pstate capable processor */
528 cpuid(CPUID_FREQ_VOLT_CAPABILITIES
, &eax
, &ebx
, &ecx
, &edx
);
529 if ((edx
& USE_HW_PSTATE
) == USE_HW_PSTATE
)
530 cpu_family
= CPU_HW_PSTATE
;
538 set_cpus_allowed_ptr(current
, &oldmask
);
542 static int check_pst_table(struct powernow_k8_data
*data
, struct pst_s
*pst
, u8 maxvid
)
547 for (j
= 0; j
< data
->numps
; j
++) {
548 if (pst
[j
].vid
> LEAST_VID
) {
549 printk(KERN_ERR FW_BUG PFX
"vid %d invalid : 0x%x\n",
553 if (pst
[j
].vid
< data
->rvo
) { /* vid + rvo >= 0 */
554 printk(KERN_ERR FW_BUG PFX
"0 vid exceeded with pstate"
558 if (pst
[j
].vid
< maxvid
+ data
->rvo
) { /* vid + rvo >= maxvid */
559 printk(KERN_ERR FW_BUG PFX
"maxvid exceeded with pstate"
563 if (pst
[j
].fid
> MAX_FID
) {
564 printk(KERN_ERR FW_BUG PFX
"maxfid exceeded with pstate"
568 if (j
&& (pst
[j
].fid
< HI_FID_TABLE_BOTTOM
)) {
569 /* Only first fid is allowed to be in "low" range */
570 printk(KERN_ERR FW_BUG PFX
"two low fids - %d : "
571 "0x%x\n", j
, pst
[j
].fid
);
574 if (pst
[j
].fid
< lastfid
)
575 lastfid
= pst
[j
].fid
;
578 printk(KERN_ERR FW_BUG PFX
"lastfid invalid\n");
581 if (lastfid
> LO_FID_TABLE_TOP
)
582 printk(KERN_INFO FW_BUG PFX
"first fid not from lo freq table\n");
587 static void print_basics(struct powernow_k8_data
*data
)
590 for (j
= 0; j
< data
->numps
; j
++) {
591 if (data
->powernow_table
[j
].frequency
!= CPUFREQ_ENTRY_INVALID
) {
592 if (cpu_family
== CPU_HW_PSTATE
) {
593 printk(KERN_INFO PFX
" %d : pstate %d (%d MHz)\n",
595 data
->powernow_table
[j
].index
,
596 data
->powernow_table
[j
].frequency
/1000);
598 printk(KERN_INFO PFX
" %d : fid 0x%x (%d MHz), vid 0x%x\n",
600 data
->powernow_table
[j
].index
& 0xff,
601 data
->powernow_table
[j
].frequency
/1000,
602 data
->powernow_table
[j
].index
>> 8);
607 printk(KERN_INFO PFX
"Only %d pstates on battery\n", data
->batps
);
610 static int fill_powernow_table(struct powernow_k8_data
*data
, struct pst_s
*pst
, u8 maxvid
)
612 struct cpufreq_frequency_table
*powernow_table
;
615 if (data
->batps
) { /* use ACPI support to get full speed on mains power */
616 printk(KERN_WARNING PFX
"Only %d pstates usable (use ACPI driver for full range\n", data
->batps
);
617 data
->numps
= data
->batps
;
620 for ( j
=1; j
<data
->numps
; j
++ ) {
621 if (pst
[j
-1].fid
>= pst
[j
].fid
) {
622 printk(KERN_ERR PFX
"PST out of sequence\n");
627 if (data
->numps
< 2) {
628 printk(KERN_ERR PFX
"no p states to transition\n");
632 if (check_pst_table(data
, pst
, maxvid
))
635 powernow_table
= kmalloc((sizeof(struct cpufreq_frequency_table
)
636 * (data
->numps
+ 1)), GFP_KERNEL
);
637 if (!powernow_table
) {
638 printk(KERN_ERR PFX
"powernow_table memory alloc failure\n");
642 for (j
= 0; j
< data
->numps
; j
++) {
643 powernow_table
[j
].index
= pst
[j
].fid
; /* lower 8 bits */
644 powernow_table
[j
].index
|= (pst
[j
].vid
<< 8); /* upper 8 bits */
645 powernow_table
[j
].frequency
= find_khz_freq_from_fid(pst
[j
].fid
);
647 powernow_table
[data
->numps
].frequency
= CPUFREQ_TABLE_END
;
648 powernow_table
[data
->numps
].index
= 0;
650 if (query_current_values_with_pending_wait(data
)) {
651 kfree(powernow_table
);
655 dprintk("cfid 0x%x, cvid 0x%x\n", data
->currfid
, data
->currvid
);
656 data
->powernow_table
= powernow_table
;
657 if (first_cpu(per_cpu(cpu_core_map
, data
->cpu
)) == data
->cpu
)
660 for (j
= 0; j
< data
->numps
; j
++)
661 if ((pst
[j
].fid
==data
->currfid
) && (pst
[j
].vid
==data
->currvid
))
664 dprintk("currfid/vid do not match PST, ignoring\n");
668 /* Find and validate the PSB/PST table in BIOS. */
669 static int find_psb_table(struct powernow_k8_data
*data
)
678 for (i
= 0xc0000; i
< 0xffff0; i
+= 0x10) {
679 /* Scan BIOS looking for the signature. */
680 /* It can not be at ffff0 - it is too big. */
682 psb
= phys_to_virt(i
);
683 if (memcmp(psb
, PSB_ID_STRING
, PSB_ID_STRING_LEN
) != 0)
686 dprintk("found PSB header at 0x%p\n", psb
);
688 dprintk("table vers: 0x%x\n", psb
->tableversion
);
689 if (psb
->tableversion
!= PSB_VERSION_1_4
) {
690 printk(KERN_ERR FW_BUG PFX
"PSB table is not v1.4\n");
694 dprintk("flags: 0x%x\n", psb
->flags1
);
696 printk(KERN_ERR FW_BUG PFX
"unknown flags\n");
700 data
->vstable
= psb
->vstable
;
701 dprintk("voltage stabilization time: %d(*20us)\n", data
->vstable
);
703 dprintk("flags2: 0x%x\n", psb
->flags2
);
704 data
->rvo
= psb
->flags2
& 3;
705 data
->irt
= ((psb
->flags2
) >> 2) & 3;
706 mvs
= ((psb
->flags2
) >> 4) & 3;
707 data
->vidmvs
= 1 << mvs
;
708 data
->batps
= ((psb
->flags2
) >> 6) & 3;
710 dprintk("ramp voltage offset: %d\n", data
->rvo
);
711 dprintk("isochronous relief time: %d\n", data
->irt
);
712 dprintk("maximum voltage step: %d - 0x%x\n", mvs
, data
->vidmvs
);
714 dprintk("numpst: 0x%x\n", psb
->num_tables
);
715 cpst
= psb
->num_tables
;
716 if ((psb
->cpuid
== 0x00000fc0) || (psb
->cpuid
== 0x00000fe0) ){
717 thiscpuid
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
718 if ((thiscpuid
== 0x00000fc0) || (thiscpuid
== 0x00000fe0) ) {
723 printk(KERN_ERR FW_BUG PFX
"numpst must be 1\n");
727 data
->plllock
= psb
->plllocktime
;
728 dprintk("plllocktime: 0x%x (units 1us)\n", psb
->plllocktime
);
729 dprintk("maxfid: 0x%x\n", psb
->maxfid
);
730 dprintk("maxvid: 0x%x\n", psb
->maxvid
);
731 maxvid
= psb
->maxvid
;
733 data
->numps
= psb
->numps
;
734 dprintk("numpstates: 0x%x\n", data
->numps
);
735 return fill_powernow_table(data
, (struct pst_s
*)(psb
+1), maxvid
);
738 * If you see this message, complain to BIOS manufacturer. If
739 * he tells you "we do not support Linux" or some similar
740 * nonsense, remember that Windows 2000 uses the same legacy
741 * mechanism that the old Linux PSB driver uses. Tell them it
742 * is broken with Windows 2000.
744 * The reference to the AMD documentation is chapter 9 in the
745 * BIOS and Kernel Developer's Guide, which is available on
748 printk(KERN_ERR PFX
"BIOS error - no PSB or ACPI _PSS objects\n");
752 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
753 static void powernow_k8_acpi_pst_values(struct powernow_k8_data
*data
, unsigned int index
)
755 if (!data
->acpi_data
.state_count
|| (cpu_family
== CPU_HW_PSTATE
))
758 data
->irt
= (data
->acpi_data
.states
[index
].control
>> IRT_SHIFT
) & IRT_MASK
;
759 data
->rvo
= (data
->acpi_data
.states
[index
].control
>> RVO_SHIFT
) & RVO_MASK
;
760 data
->exttype
= (data
->acpi_data
.states
[index
].control
>> EXT_TYPE_SHIFT
) & EXT_TYPE_MASK
;
761 data
->plllock
= (data
->acpi_data
.states
[index
].control
>> PLL_L_SHIFT
) & PLL_L_MASK
;
762 data
->vidmvs
= 1 << ((data
->acpi_data
.states
[index
].control
>> MVS_SHIFT
) & MVS_MASK
);
763 data
->vstable
= (data
->acpi_data
.states
[index
].control
>> VST_SHIFT
) & VST_MASK
;
766 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data
*data
)
768 struct cpufreq_frequency_table
*powernow_table
;
769 int ret_val
= -ENODEV
;
771 if (acpi_processor_register_performance(&data
->acpi_data
, data
->cpu
)) {
772 dprintk("register performance failed: bad ACPI data\n");
776 /* verify the data contained in the ACPI structures */
777 if (data
->acpi_data
.state_count
<= 1) {
778 dprintk("No ACPI P-States\n");
782 if ((data
->acpi_data
.control_register
.space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
) ||
783 (data
->acpi_data
.status_register
.space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
)) {
784 dprintk("Invalid control/status registers (%x - %x)\n",
785 data
->acpi_data
.control_register
.space_id
,
786 data
->acpi_data
.status_register
.space_id
);
790 /* fill in data->powernow_table */
791 powernow_table
= kmalloc((sizeof(struct cpufreq_frequency_table
)
792 * (data
->acpi_data
.state_count
+ 1)), GFP_KERNEL
);
793 if (!powernow_table
) {
794 dprintk("powernow_table memory alloc failure\n");
798 if (cpu_family
== CPU_HW_PSTATE
)
799 ret_val
= fill_powernow_table_pstate(data
, powernow_table
);
801 ret_val
= fill_powernow_table_fidvid(data
, powernow_table
);
805 powernow_table
[data
->acpi_data
.state_count
].frequency
= CPUFREQ_TABLE_END
;
806 powernow_table
[data
->acpi_data
.state_count
].index
= 0;
807 data
->powernow_table
= powernow_table
;
810 data
->numps
= data
->acpi_data
.state_count
;
811 if (first_cpu(per_cpu(cpu_core_map
, data
->cpu
)) == data
->cpu
)
813 powernow_k8_acpi_pst_values(data
, 0);
815 /* notify BIOS that we exist */
816 acpi_processor_notify_smm(THIS_MODULE
);
818 if (!alloc_cpumask_var(&data
->acpi_data
.shared_cpu_map
, GFP_KERNEL
)) {
820 "unable to alloc powernow_k8_data cpumask\n");
828 kfree(powernow_table
);
831 acpi_processor_unregister_performance(&data
->acpi_data
, data
->cpu
);
833 /* data->acpi_data.state_count informs us at ->exit() whether ACPI was used */
834 data
->acpi_data
.state_count
= 0;
839 static int fill_powernow_table_pstate(struct powernow_k8_data
*data
, struct cpufreq_frequency_table
*powernow_table
)
843 rdmsr(MSR_PSTATE_CUR_LIMIT
, hi
, lo
);
844 data
->max_hw_pstate
= (hi
& HW_PSTATE_MAX_MASK
) >> HW_PSTATE_MAX_SHIFT
;
846 for (i
= 0; i
< data
->acpi_data
.state_count
; i
++) {
849 index
= data
->acpi_data
.states
[i
].control
& HW_PSTATE_MASK
;
850 if (index
> data
->max_hw_pstate
) {
851 printk(KERN_ERR PFX
"invalid pstate %d - bad value %d.\n", i
, index
);
852 printk(KERN_ERR PFX
"Please report to BIOS manufacturer\n");
853 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
856 rdmsr(MSR_PSTATE_DEF_BASE
+ index
, lo
, hi
);
857 if (!(hi
& HW_PSTATE_VALID_MASK
)) {
858 dprintk("invalid pstate %d, ignoring\n", index
);
859 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
863 powernow_table
[i
].index
= index
;
865 powernow_table
[i
].frequency
= data
->acpi_data
.states
[i
].core_frequency
* 1000;
870 static int fill_powernow_table_fidvid(struct powernow_k8_data
*data
, struct cpufreq_frequency_table
*powernow_table
)
874 for (i
= 0; i
< data
->acpi_data
.state_count
; i
++) {
879 fid
= data
->acpi_data
.states
[i
].status
& EXT_FID_MASK
;
880 vid
= (data
->acpi_data
.states
[i
].status
>> VID_SHIFT
) & EXT_VID_MASK
;
882 fid
= data
->acpi_data
.states
[i
].control
& FID_MASK
;
883 vid
= (data
->acpi_data
.states
[i
].control
>> VID_SHIFT
) & VID_MASK
;
886 dprintk(" %d : fid 0x%x, vid 0x%x\n", i
, fid
, vid
);
888 powernow_table
[i
].index
= fid
; /* lower 8 bits */
889 powernow_table
[i
].index
|= (vid
<< 8); /* upper 8 bits */
890 powernow_table
[i
].frequency
= find_khz_freq_from_fid(fid
);
892 /* verify frequency is OK */
893 if ((powernow_table
[i
].frequency
> (MAX_FREQ
* 1000)) ||
894 (powernow_table
[i
].frequency
< (MIN_FREQ
* 1000))) {
895 dprintk("invalid freq %u kHz, ignoring\n", powernow_table
[i
].frequency
);
896 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
900 /* verify voltage is OK - BIOSs are using "off" to indicate invalid */
901 if (vid
== VID_OFF
) {
902 dprintk("invalid vid %u, ignoring\n", vid
);
903 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
907 /* verify only 1 entry from the lo frequency table */
908 if (fid
< HI_FID_TABLE_BOTTOM
) {
910 /* if both entries are the same, ignore this one ... */
911 if ((powernow_table
[i
].frequency
!= powernow_table
[cntlofreq
].frequency
) ||
912 (powernow_table
[i
].index
!= powernow_table
[cntlofreq
].index
)) {
913 printk(KERN_ERR PFX
"Too many lo freq table entries\n");
917 dprintk("double low frequency table entry, ignoring it.\n");
918 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
924 if (powernow_table
[i
].frequency
!= (data
->acpi_data
.states
[i
].core_frequency
* 1000)) {
925 printk(KERN_INFO PFX
"invalid freq entries %u kHz vs. %u kHz\n",
926 powernow_table
[i
].frequency
,
927 (unsigned int) (data
->acpi_data
.states
[i
].core_frequency
* 1000));
928 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
935 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data
*data
)
937 if (data
->acpi_data
.state_count
)
938 acpi_processor_unregister_performance(&data
->acpi_data
, data
->cpu
);
939 free_cpumask_var(data
->acpi_data
.shared_cpu_map
);
943 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data
*data
) { return -ENODEV
; }
944 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data
*data
) { return; }
945 static void powernow_k8_acpi_pst_values(struct powernow_k8_data
*data
, unsigned int index
) { return; }
946 #endif /* CONFIG_X86_POWERNOW_K8_ACPI */
948 /* Take a frequency, and issue the fid/vid transition command */
949 static int transition_frequency_fidvid(struct powernow_k8_data
*data
, unsigned int index
)
954 struct cpufreq_freqs freqs
;
956 dprintk("cpu %d transition to index %u\n", smp_processor_id(), index
);
958 /* fid/vid correctness check for k8 */
959 /* fid are the lower 8 bits of the index we stored into
960 * the cpufreq frequency table in find_psb_table, vid
961 * are the upper 8 bits.
963 fid
= data
->powernow_table
[index
].index
& 0xFF;
964 vid
= (data
->powernow_table
[index
].index
& 0xFF00) >> 8;
966 dprintk("table matched fid 0x%x, giving vid 0x%x\n", fid
, vid
);
968 if (query_current_values_with_pending_wait(data
))
971 if ((data
->currvid
== vid
) && (data
->currfid
== fid
)) {
972 dprintk("target matches current values (fid 0x%x, vid 0x%x)\n",
977 if ((fid
< HI_FID_TABLE_BOTTOM
) && (data
->currfid
< HI_FID_TABLE_BOTTOM
)) {
979 "ignoring illegal change in lo freq table-%x to 0x%x\n",
984 dprintk("cpu %d, changing to fid 0x%x, vid 0x%x\n",
985 smp_processor_id(), fid
, vid
);
986 freqs
.old
= find_khz_freq_from_fid(data
->currfid
);
987 freqs
.new = find_khz_freq_from_fid(fid
);
989 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
991 cpufreq_notify_transition(&freqs
, CPUFREQ_PRECHANGE
);
994 res
= transition_fid_vid(data
, fid
, vid
);
995 freqs
.new = find_khz_freq_from_fid(data
->currfid
);
997 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
999 cpufreq_notify_transition(&freqs
, CPUFREQ_POSTCHANGE
);
1004 /* Take a frequency, and issue the hardware pstate transition command */
1005 static int transition_frequency_pstate(struct powernow_k8_data
*data
, unsigned int index
)
1009 struct cpufreq_freqs freqs
;
1011 dprintk("cpu %d transition to index %u\n", smp_processor_id(), index
);
1013 /* get MSR index for hardware pstate transition */
1014 pstate
= index
& HW_PSTATE_MASK
;
1015 if (pstate
> data
->max_hw_pstate
)
1017 freqs
.old
= find_khz_freq_from_pstate(data
->powernow_table
, data
->currpstate
);
1018 freqs
.new = find_khz_freq_from_pstate(data
->powernow_table
, pstate
);
1020 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1022 cpufreq_notify_transition(&freqs
, CPUFREQ_PRECHANGE
);
1025 res
= transition_pstate(data
, pstate
);
1026 freqs
.new = find_khz_freq_from_pstate(data
->powernow_table
, pstate
);
1028 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1030 cpufreq_notify_transition(&freqs
, CPUFREQ_POSTCHANGE
);
1035 /* Driver entry point to switch to the target frequency */
1036 static int powernowk8_target(struct cpufreq_policy
*pol
, unsigned targfreq
, unsigned relation
)
1039 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1042 unsigned int newstate
;
1048 checkfid
= data
->currfid
;
1049 checkvid
= data
->currvid
;
1051 /* only run on specific CPU from here on */
1052 oldmask
= current
->cpus_allowed
;
1053 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(pol
->cpu
));
1055 if (smp_processor_id() != pol
->cpu
) {
1056 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", pol
->cpu
);
1060 if (pending_bit_stuck()) {
1061 printk(KERN_ERR PFX
"failing targ, change pending bit set\n");
1065 dprintk("targ: cpu %d, %d kHz, min %d, max %d, relation %d\n",
1066 pol
->cpu
, targfreq
, pol
->min
, pol
->max
, relation
);
1068 if (query_current_values_with_pending_wait(data
))
1071 if (cpu_family
!= CPU_HW_PSTATE
) {
1072 dprintk("targ: curr fid 0x%x, vid 0x%x\n",
1073 data
->currfid
, data
->currvid
);
1075 if ((checkvid
!= data
->currvid
) || (checkfid
!= data
->currfid
)) {
1076 printk(KERN_INFO PFX
1077 "error - out of sync, fix 0x%x 0x%x, vid 0x%x 0x%x\n",
1078 checkfid
, data
->currfid
, checkvid
, data
->currvid
);
1082 if (cpufreq_frequency_table_target(pol
, data
->powernow_table
, targfreq
, relation
, &newstate
))
1085 mutex_lock(&fidvid_mutex
);
1087 powernow_k8_acpi_pst_values(data
, newstate
);
1089 if (cpu_family
== CPU_HW_PSTATE
)
1090 ret
= transition_frequency_pstate(data
, newstate
);
1092 ret
= transition_frequency_fidvid(data
, newstate
);
1094 printk(KERN_ERR PFX
"transition frequency failed\n");
1096 mutex_unlock(&fidvid_mutex
);
1099 mutex_unlock(&fidvid_mutex
);
1101 if (cpu_family
== CPU_HW_PSTATE
)
1102 pol
->cur
= find_khz_freq_from_pstate(data
->powernow_table
, newstate
);
1104 pol
->cur
= find_khz_freq_from_fid(data
->currfid
);
1108 set_cpus_allowed_ptr(current
, &oldmask
);
1112 /* Driver entry point to verify the policy and range of frequencies */
1113 static int powernowk8_verify(struct cpufreq_policy
*pol
)
1115 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1120 return cpufreq_frequency_table_verify(pol
, data
->powernow_table
);
1123 /* per CPU init entry point to the driver */
1124 static int __cpuinit
powernowk8_cpu_init(struct cpufreq_policy
*pol
)
1126 struct powernow_k8_data
*data
;
1130 if (!cpu_online(pol
->cpu
))
1133 if (!check_supported_cpu(pol
->cpu
))
1136 data
= kzalloc(sizeof(struct powernow_k8_data
), GFP_KERNEL
);
1138 printk(KERN_ERR PFX
"unable to alloc powernow_k8_data");
1142 data
->cpu
= pol
->cpu
;
1143 data
->currpstate
= HW_PSTATE_INVALID
;
1145 rc
= powernow_k8_cpu_init_acpi(data
);
1148 * Use the PSB BIOS structure. This is only availabe on
1149 * an UP version, and is deprecated by AMD.
1151 if (num_online_cpus() != 1) {
1152 #ifndef CONFIG_ACPI_PROCESSOR
1153 printk(KERN_ERR PFX
"ACPI Processor support is required "
1154 "for SMP systems but is absent. Please load the "
1155 "ACPI Processor module before starting this "
1158 printk(KERN_ERR FW_BUG PFX
"Your BIOS does not provide"
1159 " ACPI _PSS objects in a way that Linux "
1160 "understands. Please report this to the Linux "
1161 "ACPI maintainers and complain to your BIOS "
1166 if (pol
->cpu
!= 0) {
1167 printk(KERN_ERR FW_BUG PFX
"No ACPI _PSS objects for "
1168 "CPU other than CPU0. Complain to your BIOS "
1172 rc
= find_psb_table(data
);
1178 /* only run on specific CPU from here on */
1179 oldmask
= current
->cpus_allowed
;
1180 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(pol
->cpu
));
1182 if (smp_processor_id() != pol
->cpu
) {
1183 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", pol
->cpu
);
1187 if (pending_bit_stuck()) {
1188 printk(KERN_ERR PFX
"failing init, change pending bit set\n");
1192 if (query_current_values_with_pending_wait(data
))
1195 if (cpu_family
== CPU_OPTERON
)
1198 /* run on any CPU again */
1199 set_cpus_allowed_ptr(current
, &oldmask
);
1201 if (cpu_family
== CPU_HW_PSTATE
)
1202 cpumask_copy(pol
->cpus
, cpumask_of(pol
->cpu
));
1204 cpumask_copy(pol
->cpus
, &per_cpu(cpu_core_map
, pol
->cpu
));
1205 data
->available_cores
= pol
->cpus
;
1207 /* Take a crude guess here.
1208 * That guess was in microseconds, so multiply with 1000 */
1209 pol
->cpuinfo
.transition_latency
= (((data
->rvo
+ 8) * data
->vstable
* VST_UNITS_20US
)
1210 + (3 * (1 << data
->irt
) * 10)) * 1000;
1212 if (cpu_family
== CPU_HW_PSTATE
)
1213 pol
->cur
= find_khz_freq_from_pstate(data
->powernow_table
, data
->currpstate
);
1215 pol
->cur
= find_khz_freq_from_fid(data
->currfid
);
1216 dprintk("policy current frequency %d kHz\n", pol
->cur
);
1218 /* min/max the cpu is capable of */
1219 if (cpufreq_frequency_table_cpuinfo(pol
, data
->powernow_table
)) {
1220 printk(KERN_ERR FW_BUG PFX
"invalid powernow_table\n");
1221 powernow_k8_cpu_exit_acpi(data
);
1222 kfree(data
->powernow_table
);
1227 cpufreq_frequency_table_get_attr(data
->powernow_table
, pol
->cpu
);
1229 if (cpu_family
== CPU_HW_PSTATE
)
1230 dprintk("cpu_init done, current pstate 0x%x\n", data
->currpstate
);
1232 dprintk("cpu_init done, current fid 0x%x, vid 0x%x\n",
1233 data
->currfid
, data
->currvid
);
1235 per_cpu(powernow_data
, pol
->cpu
) = data
;
1240 set_cpus_allowed_ptr(current
, &oldmask
);
1241 powernow_k8_cpu_exit_acpi(data
);
1247 static int __devexit
powernowk8_cpu_exit (struct cpufreq_policy
*pol
)
1249 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1254 powernow_k8_cpu_exit_acpi(data
);
1256 cpufreq_frequency_table_put_attr(pol
->cpu
);
1258 kfree(data
->powernow_table
);
1264 static unsigned int powernowk8_get (unsigned int cpu
)
1266 struct powernow_k8_data
*data
;
1267 cpumask_t oldmask
= current
->cpus_allowed
;
1268 unsigned int khz
= 0;
1271 first
= first_cpu(per_cpu(cpu_core_map
, cpu
));
1272 data
= per_cpu(powernow_data
, first
);
1277 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(cpu
));
1278 if (smp_processor_id() != cpu
) {
1280 "limiting to CPU %d failed in powernowk8_get\n", cpu
);
1281 set_cpus_allowed_ptr(current
, &oldmask
);
1285 if (query_current_values_with_pending_wait(data
))
1288 if (cpu_family
== CPU_HW_PSTATE
)
1289 khz
= find_khz_freq_from_pstate(data
->powernow_table
,
1292 khz
= find_khz_freq_from_fid(data
->currfid
);
1296 set_cpus_allowed_ptr(current
, &oldmask
);
1300 static struct freq_attr
* powernow_k8_attr
[] = {
1301 &cpufreq_freq_attr_scaling_available_freqs
,
1305 static struct cpufreq_driver cpufreq_amd64_driver
= {
1306 .verify
= powernowk8_verify
,
1307 .target
= powernowk8_target
,
1308 .init
= powernowk8_cpu_init
,
1309 .exit
= __devexit_p(powernowk8_cpu_exit
),
1310 .get
= powernowk8_get
,
1311 .name
= "powernow-k8",
1312 .owner
= THIS_MODULE
,
1313 .attr
= powernow_k8_attr
,
1316 /* driver entry point for init */
1317 static int __cpuinit
powernowk8_init(void)
1319 unsigned int i
, supported_cpus
= 0;
1321 for_each_online_cpu(i
) {
1322 if (check_supported_cpu(i
))
1326 if (supported_cpus
== num_online_cpus()) {
1327 printk(KERN_INFO PFX
"Found %d %s "
1328 "processors (%d cpu cores) (" VERSION
")\n",
1330 boot_cpu_data
.x86_model_id
, supported_cpus
);
1331 return cpufreq_register_driver(&cpufreq_amd64_driver
);
1337 /* driver entry point for term */
1338 static void __exit
powernowk8_exit(void)
1342 cpufreq_unregister_driver(&cpufreq_amd64_driver
);
1345 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com> and Mark Langsdorf <mark.langsdorf@amd.com>");
1346 MODULE_DESCRIPTION("AMD Athlon 64 and Opteron processor frequency driver.");
1347 MODULE_LICENSE("GPL");
1349 late_initcall(powernowk8_init
);
1350 module_exit(powernowk8_exit
);