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 static inline const struct cpumask
*cpu_core_mask(int cpu
)
65 /* Return a frequency in MHz, given an input fid */
66 static u32
find_freq_from_fid(u32 fid
)
68 return 800 + (fid
* 100);
71 /* Return a frequency in KHz, given an input fid */
72 static u32
find_khz_freq_from_fid(u32 fid
)
74 return 1000 * find_freq_from_fid(fid
);
77 static u32
find_khz_freq_from_pstate(struct cpufreq_frequency_table
*data
, u32 pstate
)
79 return data
[pstate
].frequency
;
82 /* Return the vco fid for an input fid
84 * Each "low" fid has corresponding "high" fid, and you can get to "low" fids
85 * only from corresponding high fids. This returns "high" fid corresponding to
88 static u32
convert_fid_to_vco_fid(u32 fid
)
90 if (fid
< HI_FID_TABLE_BOTTOM
)
97 * Return 1 if the pending bit is set. Unless we just instructed the processor
98 * to transition to a new state, seeing this bit set is really bad news.
100 static int pending_bit_stuck(void)
104 if (cpu_family
== CPU_HW_PSTATE
)
107 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
108 return lo
& MSR_S_LO_CHANGE_PENDING
? 1 : 0;
112 * Update the global current fid / vid values from the status msr.
113 * Returns 1 on error.
115 static int query_current_values_with_pending_wait(struct powernow_k8_data
*data
)
120 if (cpu_family
== CPU_HW_PSTATE
) {
121 if (data
->currpstate
== HW_PSTATE_INVALID
) {
122 /* read (initial) hw pstate if not yet set */
123 rdmsr(MSR_PSTATE_STATUS
, lo
, hi
);
124 i
= lo
& HW_PSTATE_MASK
;
127 * a workaround for family 11h erratum 311 might cause
128 * an "out-of-range Pstate if the core is in Pstate-0
130 if (i
>= data
->numps
)
131 data
->currpstate
= HW_PSTATE_0
;
133 data
->currpstate
= i
;
139 dprintk("detected change pending stuck\n");
142 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
143 } while (lo
& MSR_S_LO_CHANGE_PENDING
);
145 data
->currvid
= hi
& MSR_S_HI_CURRENT_VID
;
146 data
->currfid
= lo
& MSR_S_LO_CURRENT_FID
;
151 /* the isochronous relief time */
152 static void count_off_irt(struct powernow_k8_data
*data
)
154 udelay((1 << data
->irt
) * 10);
158 /* the voltage stabilization time */
159 static void count_off_vst(struct powernow_k8_data
*data
)
161 udelay(data
->vstable
* VST_UNITS_20US
);
165 /* need to init the control msr to a safe value (for each cpu) */
166 static void fidvid_msr_init(void)
171 rdmsr(MSR_FIDVID_STATUS
, lo
, hi
);
172 vid
= hi
& MSR_S_HI_CURRENT_VID
;
173 fid
= lo
& MSR_S_LO_CURRENT_FID
;
174 lo
= fid
| (vid
<< MSR_C_LO_VID_SHIFT
);
175 hi
= MSR_C_HI_STP_GNT_BENIGN
;
176 dprintk("cpu%d, init lo 0x%x, hi 0x%x\n", smp_processor_id(), lo
, hi
);
177 wrmsr(MSR_FIDVID_CTL
, lo
, hi
);
180 /* write the new fid value along with the other control fields to the msr */
181 static int write_new_fid(struct powernow_k8_data
*data
, u32 fid
)
184 u32 savevid
= data
->currvid
;
187 if ((fid
& INVALID_FID_MASK
) || (data
->currvid
& INVALID_VID_MASK
)) {
188 printk(KERN_ERR PFX
"internal error - overflow on fid write\n");
192 lo
= fid
| (data
->currvid
<< MSR_C_LO_VID_SHIFT
) | MSR_C_LO_INIT_FID_VID
;
194 dprintk("writing fid 0x%x, lo 0x%x, hi 0x%x\n",
195 fid
, lo
, data
->plllock
* PLL_LOCK_CONVERSION
);
198 wrmsr(MSR_FIDVID_CTL
, lo
, data
->plllock
* PLL_LOCK_CONVERSION
);
200 printk(KERN_ERR PFX
"Hardware error - pending bit very stuck - no further pstate changes possible\n");
203 } while (query_current_values_with_pending_wait(data
));
207 if (savevid
!= data
->currvid
) {
208 printk(KERN_ERR PFX
"vid change on fid trans, old 0x%x, new 0x%x\n",
209 savevid
, data
->currvid
);
213 if (fid
!= data
->currfid
) {
214 printk(KERN_ERR PFX
"fid trans failed, fid 0x%x, curr 0x%x\n", fid
,
222 /* Write a new vid to the hardware */
223 static int write_new_vid(struct powernow_k8_data
*data
, u32 vid
)
226 u32 savefid
= data
->currfid
;
229 if ((data
->currfid
& INVALID_FID_MASK
) || (vid
& INVALID_VID_MASK
)) {
230 printk(KERN_ERR PFX
"internal error - overflow on vid write\n");
234 lo
= data
->currfid
| (vid
<< MSR_C_LO_VID_SHIFT
) | MSR_C_LO_INIT_FID_VID
;
236 dprintk("writing vid 0x%x, lo 0x%x, hi 0x%x\n",
237 vid
, lo
, STOP_GRANT_5NS
);
240 wrmsr(MSR_FIDVID_CTL
, lo
, STOP_GRANT_5NS
);
242 printk(KERN_ERR PFX
"internal error - pending bit very stuck - no further pstate changes possible\n");
245 } while (query_current_values_with_pending_wait(data
));
247 if (savefid
!= data
->currfid
) {
248 printk(KERN_ERR PFX
"fid changed on vid trans, old 0x%x new 0x%x\n",
249 savefid
, data
->currfid
);
253 if (vid
!= data
->currvid
) {
254 printk(KERN_ERR PFX
"vid trans failed, vid 0x%x, curr 0x%x\n", vid
,
263 * Reduce the vid by the max of step or reqvid.
264 * Decreasing vid codes represent increasing voltages:
265 * vid of 0 is 1.550V, vid of 0x1e is 0.800V, vid of VID_OFF is off.
267 static int decrease_vid_code_by_step(struct powernow_k8_data
*data
, u32 reqvid
, u32 step
)
269 if ((data
->currvid
- reqvid
) > step
)
270 reqvid
= data
->currvid
- step
;
272 if (write_new_vid(data
, reqvid
))
280 /* Change hardware pstate by single MSR write */
281 static int transition_pstate(struct powernow_k8_data
*data
, u32 pstate
)
283 wrmsr(MSR_PSTATE_CTRL
, pstate
, 0);
284 data
->currpstate
= pstate
;
288 /* Change Opteron/Athlon64 fid and vid, by the 3 phases. */
289 static int transition_fid_vid(struct powernow_k8_data
*data
, u32 reqfid
, u32 reqvid
)
291 if (core_voltage_pre_transition(data
, reqvid
))
294 if (core_frequency_transition(data
, reqfid
))
297 if (core_voltage_post_transition(data
, reqvid
))
300 if (query_current_values_with_pending_wait(data
))
303 if ((reqfid
!= data
->currfid
) || (reqvid
!= data
->currvid
)) {
304 printk(KERN_ERR PFX
"failed (cpu%d): req 0x%x 0x%x, curr 0x%x 0x%x\n",
306 reqfid
, reqvid
, data
->currfid
, data
->currvid
);
310 dprintk("transitioned (cpu%d): new fid 0x%x, vid 0x%x\n",
311 smp_processor_id(), data
->currfid
, data
->currvid
);
316 /* Phase 1 - core voltage transition ... setup voltage */
317 static int core_voltage_pre_transition(struct powernow_k8_data
*data
, u32 reqvid
)
319 u32 rvosteps
= data
->rvo
;
320 u32 savefid
= data
->currfid
;
323 dprintk("ph1 (cpu%d): start, currfid 0x%x, currvid 0x%x, reqvid 0x%x, rvo 0x%x\n",
325 data
->currfid
, data
->currvid
, reqvid
, data
->rvo
);
327 rdmsr(MSR_FIDVID_STATUS
, lo
, maxvid
);
328 maxvid
= 0x1f & (maxvid
>> 16);
329 dprintk("ph1 maxvid=0x%x\n", maxvid
);
330 if (reqvid
< maxvid
) /* lower numbers are higher voltages */
333 while (data
->currvid
> reqvid
) {
334 dprintk("ph1: curr 0x%x, req vid 0x%x\n",
335 data
->currvid
, reqvid
);
336 if (decrease_vid_code_by_step(data
, reqvid
, data
->vidmvs
))
340 while ((rvosteps
> 0) && ((data
->rvo
+ data
->currvid
) > reqvid
)) {
341 if (data
->currvid
== maxvid
) {
344 dprintk("ph1: changing vid for rvo, req 0x%x\n",
346 if (decrease_vid_code_by_step(data
, data
->currvid
- 1, 1))
352 if (query_current_values_with_pending_wait(data
))
355 if (savefid
!= data
->currfid
) {
356 printk(KERN_ERR PFX
"ph1 err, currfid changed 0x%x\n", data
->currfid
);
360 dprintk("ph1 complete, currfid 0x%x, currvid 0x%x\n",
361 data
->currfid
, data
->currvid
);
366 /* Phase 2 - core frequency transition */
367 static int core_frequency_transition(struct powernow_k8_data
*data
, u32 reqfid
)
369 u32 vcoreqfid
, vcocurrfid
, vcofiddiff
, fid_interval
, savevid
= data
->currvid
;
371 if ((reqfid
< HI_FID_TABLE_BOTTOM
) && (data
->currfid
< HI_FID_TABLE_BOTTOM
)) {
372 printk(KERN_ERR PFX
"ph2: illegal lo-lo transition 0x%x 0x%x\n",
373 reqfid
, data
->currfid
);
377 if (data
->currfid
== reqfid
) {
378 printk(KERN_ERR PFX
"ph2 null fid transition 0x%x\n", data
->currfid
);
382 dprintk("ph2 (cpu%d): starting, currfid 0x%x, currvid 0x%x, reqfid 0x%x\n",
384 data
->currfid
, data
->currvid
, reqfid
);
386 vcoreqfid
= convert_fid_to_vco_fid(reqfid
);
387 vcocurrfid
= convert_fid_to_vco_fid(data
->currfid
);
388 vcofiddiff
= vcocurrfid
> vcoreqfid
? vcocurrfid
- vcoreqfid
389 : vcoreqfid
- vcocurrfid
;
391 while (vcofiddiff
> 2) {
392 (data
->currfid
& 1) ? (fid_interval
= 1) : (fid_interval
= 2);
394 if (reqfid
> data
->currfid
) {
395 if (data
->currfid
> LO_FID_TABLE_TOP
) {
396 if (write_new_fid(data
, data
->currfid
+ fid_interval
)) {
401 (data
, 2 + convert_fid_to_vco_fid(data
->currfid
))) {
406 if (write_new_fid(data
, data
->currfid
- fid_interval
))
410 vcocurrfid
= convert_fid_to_vco_fid(data
->currfid
);
411 vcofiddiff
= vcocurrfid
> vcoreqfid
? vcocurrfid
- vcoreqfid
412 : vcoreqfid
- vcocurrfid
;
415 if (write_new_fid(data
, reqfid
))
418 if (query_current_values_with_pending_wait(data
))
421 if (data
->currfid
!= reqfid
) {
423 "ph2: mismatch, failed fid transition, curr 0x%x, req 0x%x\n",
424 data
->currfid
, reqfid
);
428 if (savevid
!= data
->currvid
) {
429 printk(KERN_ERR PFX
"ph2: vid changed, save 0x%x, curr 0x%x\n",
430 savevid
, data
->currvid
);
434 dprintk("ph2 complete, currfid 0x%x, currvid 0x%x\n",
435 data
->currfid
, data
->currvid
);
440 /* Phase 3 - core voltage transition flow ... jump to the final vid. */
441 static int core_voltage_post_transition(struct powernow_k8_data
*data
, u32 reqvid
)
443 u32 savefid
= data
->currfid
;
444 u32 savereqvid
= reqvid
;
446 dprintk("ph3 (cpu%d): starting, currfid 0x%x, currvid 0x%x\n",
448 data
->currfid
, data
->currvid
);
450 if (reqvid
!= data
->currvid
) {
451 if (write_new_vid(data
, reqvid
))
454 if (savefid
!= data
->currfid
) {
456 "ph3: bad fid change, save 0x%x, curr 0x%x\n",
457 savefid
, data
->currfid
);
461 if (data
->currvid
!= reqvid
) {
463 "ph3: failed vid transition\n, req 0x%x, curr 0x%x",
464 reqvid
, data
->currvid
);
469 if (query_current_values_with_pending_wait(data
))
472 if (savereqvid
!= data
->currvid
) {
473 dprintk("ph3 failed, currvid 0x%x\n", data
->currvid
);
477 if (savefid
!= data
->currfid
) {
478 dprintk("ph3 failed, currfid changed 0x%x\n",
483 dprintk("ph3 complete, currfid 0x%x, currvid 0x%x\n",
484 data
->currfid
, data
->currvid
);
489 static int check_supported_cpu(unsigned int cpu
)
492 u32 eax
, ebx
, ecx
, edx
;
495 oldmask
= current
->cpus_allowed
;
496 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(cpu
));
498 if (smp_processor_id() != cpu
) {
499 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", cpu
);
503 if (current_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
)
506 eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
507 if (((eax
& CPUID_XFAM
) != CPUID_XFAM_K8
) &&
508 ((eax
& CPUID_XFAM
) < CPUID_XFAM_10H
))
511 if ((eax
& CPUID_XFAM
) == CPUID_XFAM_K8
) {
512 if (((eax
& CPUID_USE_XFAM_XMOD
) != CPUID_USE_XFAM_XMOD
) ||
513 ((eax
& CPUID_XMOD
) > CPUID_XMOD_REV_MASK
)) {
514 printk(KERN_INFO PFX
"Processor cpuid %x not supported\n", eax
);
518 eax
= cpuid_eax(CPUID_GET_MAX_CAPABILITIES
);
519 if (eax
< CPUID_FREQ_VOLT_CAPABILITIES
) {
521 "No frequency change capabilities detected\n");
525 cpuid(CPUID_FREQ_VOLT_CAPABILITIES
, &eax
, &ebx
, &ecx
, &edx
);
526 if ((edx
& P_STATE_TRANSITION_CAPABLE
) != P_STATE_TRANSITION_CAPABLE
) {
527 printk(KERN_INFO PFX
"Power state transitions not supported\n");
530 } else { /* must be a HW Pstate capable processor */
531 cpuid(CPUID_FREQ_VOLT_CAPABILITIES
, &eax
, &ebx
, &ecx
, &edx
);
532 if ((edx
& USE_HW_PSTATE
) == USE_HW_PSTATE
)
533 cpu_family
= CPU_HW_PSTATE
;
541 set_cpus_allowed_ptr(current
, &oldmask
);
545 static int check_pst_table(struct powernow_k8_data
*data
, struct pst_s
*pst
, u8 maxvid
)
550 for (j
= 0; j
< data
->numps
; j
++) {
551 if (pst
[j
].vid
> LEAST_VID
) {
552 printk(KERN_ERR FW_BUG PFX
"vid %d invalid : 0x%x\n",
556 if (pst
[j
].vid
< data
->rvo
) { /* vid + rvo >= 0 */
557 printk(KERN_ERR FW_BUG PFX
"0 vid exceeded with pstate"
561 if (pst
[j
].vid
< maxvid
+ data
->rvo
) { /* vid + rvo >= maxvid */
562 printk(KERN_ERR FW_BUG PFX
"maxvid exceeded with pstate"
566 if (pst
[j
].fid
> MAX_FID
) {
567 printk(KERN_ERR FW_BUG PFX
"maxfid exceeded with pstate"
571 if (j
&& (pst
[j
].fid
< HI_FID_TABLE_BOTTOM
)) {
572 /* Only first fid is allowed to be in "low" range */
573 printk(KERN_ERR FW_BUG PFX
"two low fids - %d : "
574 "0x%x\n", j
, pst
[j
].fid
);
577 if (pst
[j
].fid
< lastfid
)
578 lastfid
= pst
[j
].fid
;
581 printk(KERN_ERR FW_BUG PFX
"lastfid invalid\n");
584 if (lastfid
> LO_FID_TABLE_TOP
)
585 printk(KERN_INFO FW_BUG PFX
"first fid not from lo freq table\n");
590 static void print_basics(struct powernow_k8_data
*data
)
593 for (j
= 0; j
< data
->numps
; j
++) {
594 if (data
->powernow_table
[j
].frequency
!= CPUFREQ_ENTRY_INVALID
) {
595 if (cpu_family
== CPU_HW_PSTATE
) {
596 printk(KERN_INFO PFX
" %d : pstate %d (%d MHz)\n",
598 data
->powernow_table
[j
].index
,
599 data
->powernow_table
[j
].frequency
/1000);
601 printk(KERN_INFO PFX
" %d : fid 0x%x (%d MHz), vid 0x%x\n",
603 data
->powernow_table
[j
].index
& 0xff,
604 data
->powernow_table
[j
].frequency
/1000,
605 data
->powernow_table
[j
].index
>> 8);
610 printk(KERN_INFO PFX
"Only %d pstates on battery\n", data
->batps
);
613 static int fill_powernow_table(struct powernow_k8_data
*data
, struct pst_s
*pst
, u8 maxvid
)
615 struct cpufreq_frequency_table
*powernow_table
;
618 if (data
->batps
) { /* use ACPI support to get full speed on mains power */
619 printk(KERN_WARNING PFX
"Only %d pstates usable (use ACPI driver for full range\n", data
->batps
);
620 data
->numps
= data
->batps
;
623 for ( j
=1; j
<data
->numps
; j
++ ) {
624 if (pst
[j
-1].fid
>= pst
[j
].fid
) {
625 printk(KERN_ERR PFX
"PST out of sequence\n");
630 if (data
->numps
< 2) {
631 printk(KERN_ERR PFX
"no p states to transition\n");
635 if (check_pst_table(data
, pst
, maxvid
))
638 powernow_table
= kmalloc((sizeof(struct cpufreq_frequency_table
)
639 * (data
->numps
+ 1)), GFP_KERNEL
);
640 if (!powernow_table
) {
641 printk(KERN_ERR PFX
"powernow_table memory alloc failure\n");
645 for (j
= 0; j
< data
->numps
; j
++) {
646 powernow_table
[j
].index
= pst
[j
].fid
; /* lower 8 bits */
647 powernow_table
[j
].index
|= (pst
[j
].vid
<< 8); /* upper 8 bits */
648 powernow_table
[j
].frequency
= find_khz_freq_from_fid(pst
[j
].fid
);
650 powernow_table
[data
->numps
].frequency
= CPUFREQ_TABLE_END
;
651 powernow_table
[data
->numps
].index
= 0;
653 if (query_current_values_with_pending_wait(data
)) {
654 kfree(powernow_table
);
658 dprintk("cfid 0x%x, cvid 0x%x\n", data
->currfid
, data
->currvid
);
659 data
->powernow_table
= powernow_table
;
660 if (cpumask_first(cpu_core_mask(data
->cpu
)) == data
->cpu
)
663 for (j
= 0; j
< data
->numps
; j
++)
664 if ((pst
[j
].fid
==data
->currfid
) && (pst
[j
].vid
==data
->currvid
))
667 dprintk("currfid/vid do not match PST, ignoring\n");
671 /* Find and validate the PSB/PST table in BIOS. */
672 static int find_psb_table(struct powernow_k8_data
*data
)
681 for (i
= 0xc0000; i
< 0xffff0; i
+= 0x10) {
682 /* Scan BIOS looking for the signature. */
683 /* It can not be at ffff0 - it is too big. */
685 psb
= phys_to_virt(i
);
686 if (memcmp(psb
, PSB_ID_STRING
, PSB_ID_STRING_LEN
) != 0)
689 dprintk("found PSB header at 0x%p\n", psb
);
691 dprintk("table vers: 0x%x\n", psb
->tableversion
);
692 if (psb
->tableversion
!= PSB_VERSION_1_4
) {
693 printk(KERN_ERR FW_BUG PFX
"PSB table is not v1.4\n");
697 dprintk("flags: 0x%x\n", psb
->flags1
);
699 printk(KERN_ERR FW_BUG PFX
"unknown flags\n");
703 data
->vstable
= psb
->vstable
;
704 dprintk("voltage stabilization time: %d(*20us)\n", data
->vstable
);
706 dprintk("flags2: 0x%x\n", psb
->flags2
);
707 data
->rvo
= psb
->flags2
& 3;
708 data
->irt
= ((psb
->flags2
) >> 2) & 3;
709 mvs
= ((psb
->flags2
) >> 4) & 3;
710 data
->vidmvs
= 1 << mvs
;
711 data
->batps
= ((psb
->flags2
) >> 6) & 3;
713 dprintk("ramp voltage offset: %d\n", data
->rvo
);
714 dprintk("isochronous relief time: %d\n", data
->irt
);
715 dprintk("maximum voltage step: %d - 0x%x\n", mvs
, data
->vidmvs
);
717 dprintk("numpst: 0x%x\n", psb
->num_tables
);
718 cpst
= psb
->num_tables
;
719 if ((psb
->cpuid
== 0x00000fc0) || (psb
->cpuid
== 0x00000fe0) ){
720 thiscpuid
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
721 if ((thiscpuid
== 0x00000fc0) || (thiscpuid
== 0x00000fe0) ) {
726 printk(KERN_ERR FW_BUG PFX
"numpst must be 1\n");
730 data
->plllock
= psb
->plllocktime
;
731 dprintk("plllocktime: 0x%x (units 1us)\n", psb
->plllocktime
);
732 dprintk("maxfid: 0x%x\n", psb
->maxfid
);
733 dprintk("maxvid: 0x%x\n", psb
->maxvid
);
734 maxvid
= psb
->maxvid
;
736 data
->numps
= psb
->numps
;
737 dprintk("numpstates: 0x%x\n", data
->numps
);
738 return fill_powernow_table(data
, (struct pst_s
*)(psb
+1), maxvid
);
741 * If you see this message, complain to BIOS manufacturer. If
742 * he tells you "we do not support Linux" or some similar
743 * nonsense, remember that Windows 2000 uses the same legacy
744 * mechanism that the old Linux PSB driver uses. Tell them it
745 * is broken with Windows 2000.
747 * The reference to the AMD documentation is chapter 9 in the
748 * BIOS and Kernel Developer's Guide, which is available on
751 printk(KERN_ERR PFX
"BIOS error - no PSB or ACPI _PSS objects\n");
755 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
756 static void powernow_k8_acpi_pst_values(struct powernow_k8_data
*data
, unsigned int index
)
758 if (!data
->acpi_data
.state_count
|| (cpu_family
== CPU_HW_PSTATE
))
761 data
->irt
= (data
->acpi_data
.states
[index
].control
>> IRT_SHIFT
) & IRT_MASK
;
762 data
->rvo
= (data
->acpi_data
.states
[index
].control
>> RVO_SHIFT
) & RVO_MASK
;
763 data
->exttype
= (data
->acpi_data
.states
[index
].control
>> EXT_TYPE_SHIFT
) & EXT_TYPE_MASK
;
764 data
->plllock
= (data
->acpi_data
.states
[index
].control
>> PLL_L_SHIFT
) & PLL_L_MASK
;
765 data
->vidmvs
= 1 << ((data
->acpi_data
.states
[index
].control
>> MVS_SHIFT
) & MVS_MASK
);
766 data
->vstable
= (data
->acpi_data
.states
[index
].control
>> VST_SHIFT
) & VST_MASK
;
769 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data
*data
)
771 struct cpufreq_frequency_table
*powernow_table
;
772 int ret_val
= -ENODEV
;
774 if (acpi_processor_register_performance(&data
->acpi_data
, data
->cpu
)) {
775 dprintk("register performance failed: bad ACPI data\n");
779 /* verify the data contained in the ACPI structures */
780 if (data
->acpi_data
.state_count
<= 1) {
781 dprintk("No ACPI P-States\n");
785 if ((data
->acpi_data
.control_register
.space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
) ||
786 (data
->acpi_data
.status_register
.space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
)) {
787 dprintk("Invalid control/status registers (%x - %x)\n",
788 data
->acpi_data
.control_register
.space_id
,
789 data
->acpi_data
.status_register
.space_id
);
793 /* fill in data->powernow_table */
794 powernow_table
= kmalloc((sizeof(struct cpufreq_frequency_table
)
795 * (data
->acpi_data
.state_count
+ 1)), GFP_KERNEL
);
796 if (!powernow_table
) {
797 dprintk("powernow_table memory alloc failure\n");
801 if (cpu_family
== CPU_HW_PSTATE
)
802 ret_val
= fill_powernow_table_pstate(data
, powernow_table
);
804 ret_val
= fill_powernow_table_fidvid(data
, powernow_table
);
808 powernow_table
[data
->acpi_data
.state_count
].frequency
= CPUFREQ_TABLE_END
;
809 powernow_table
[data
->acpi_data
.state_count
].index
= 0;
810 data
->powernow_table
= powernow_table
;
813 data
->numps
= data
->acpi_data
.state_count
;
814 if (cpumask_first(cpu_core_mask(data
->cpu
)) == data
->cpu
)
816 powernow_k8_acpi_pst_values(data
, 0);
818 /* notify BIOS that we exist */
819 acpi_processor_notify_smm(THIS_MODULE
);
821 if (!alloc_cpumask_var(&data
->acpi_data
.shared_cpu_map
, GFP_KERNEL
)) {
823 "unable to alloc powernow_k8_data cpumask\n");
831 kfree(powernow_table
);
834 acpi_processor_unregister_performance(&data
->acpi_data
, data
->cpu
);
836 /* data->acpi_data.state_count informs us at ->exit() whether ACPI was used */
837 data
->acpi_data
.state_count
= 0;
842 static int fill_powernow_table_pstate(struct powernow_k8_data
*data
, struct cpufreq_frequency_table
*powernow_table
)
846 rdmsr(MSR_PSTATE_CUR_LIMIT
, hi
, lo
);
847 data
->max_hw_pstate
= (hi
& HW_PSTATE_MAX_MASK
) >> HW_PSTATE_MAX_SHIFT
;
849 for (i
= 0; i
< data
->acpi_data
.state_count
; i
++) {
852 index
= data
->acpi_data
.states
[i
].control
& HW_PSTATE_MASK
;
853 if (index
> data
->max_hw_pstate
) {
854 printk(KERN_ERR PFX
"invalid pstate %d - bad value %d.\n", i
, index
);
855 printk(KERN_ERR PFX
"Please report to BIOS manufacturer\n");
856 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
859 rdmsr(MSR_PSTATE_DEF_BASE
+ index
, lo
, hi
);
860 if (!(hi
& HW_PSTATE_VALID_MASK
)) {
861 dprintk("invalid pstate %d, ignoring\n", index
);
862 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
866 powernow_table
[i
].index
= index
;
868 powernow_table
[i
].frequency
= data
->acpi_data
.states
[i
].core_frequency
* 1000;
873 static int fill_powernow_table_fidvid(struct powernow_k8_data
*data
, struct cpufreq_frequency_table
*powernow_table
)
877 for (i
= 0; i
< data
->acpi_data
.state_count
; i
++) {
882 fid
= data
->acpi_data
.states
[i
].status
& EXT_FID_MASK
;
883 vid
= (data
->acpi_data
.states
[i
].status
>> VID_SHIFT
) & EXT_VID_MASK
;
885 fid
= data
->acpi_data
.states
[i
].control
& FID_MASK
;
886 vid
= (data
->acpi_data
.states
[i
].control
>> VID_SHIFT
) & VID_MASK
;
889 dprintk(" %d : fid 0x%x, vid 0x%x\n", i
, fid
, vid
);
891 powernow_table
[i
].index
= fid
; /* lower 8 bits */
892 powernow_table
[i
].index
|= (vid
<< 8); /* upper 8 bits */
893 powernow_table
[i
].frequency
= find_khz_freq_from_fid(fid
);
895 /* verify frequency is OK */
896 if ((powernow_table
[i
].frequency
> (MAX_FREQ
* 1000)) ||
897 (powernow_table
[i
].frequency
< (MIN_FREQ
* 1000))) {
898 dprintk("invalid freq %u kHz, ignoring\n", powernow_table
[i
].frequency
);
899 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
903 /* verify voltage is OK - BIOSs are using "off" to indicate invalid */
904 if (vid
== VID_OFF
) {
905 dprintk("invalid vid %u, ignoring\n", vid
);
906 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
910 /* verify only 1 entry from the lo frequency table */
911 if (fid
< HI_FID_TABLE_BOTTOM
) {
913 /* if both entries are the same, ignore this one ... */
914 if ((powernow_table
[i
].frequency
!= powernow_table
[cntlofreq
].frequency
) ||
915 (powernow_table
[i
].index
!= powernow_table
[cntlofreq
].index
)) {
916 printk(KERN_ERR PFX
"Too many lo freq table entries\n");
920 dprintk("double low frequency table entry, ignoring it.\n");
921 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
927 if (powernow_table
[i
].frequency
!= (data
->acpi_data
.states
[i
].core_frequency
* 1000)) {
928 printk(KERN_INFO PFX
"invalid freq entries %u kHz vs. %u kHz\n",
929 powernow_table
[i
].frequency
,
930 (unsigned int) (data
->acpi_data
.states
[i
].core_frequency
* 1000));
931 powernow_table
[i
].frequency
= CPUFREQ_ENTRY_INVALID
;
938 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data
*data
)
940 if (data
->acpi_data
.state_count
)
941 acpi_processor_unregister_performance(&data
->acpi_data
, data
->cpu
);
942 free_cpumask_var(data
->acpi_data
.shared_cpu_map
);
945 static int get_transition_latency(struct powernow_k8_data
*data
)
949 for (i
= 0; i
< data
->acpi_data
.state_count
; i
++) {
950 int cur_latency
= data
->acpi_data
.states
[i
].transition_latency
951 + data
->acpi_data
.states
[i
].bus_master_latency
;
952 if (cur_latency
> max_latency
)
953 max_latency
= cur_latency
;
955 /* value in usecs, needs to be in nanoseconds */
956 return 1000 * max_latency
;
960 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data
*data
) { return -ENODEV
; }
961 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data
*data
) { return; }
962 static void powernow_k8_acpi_pst_values(struct powernow_k8_data
*data
, unsigned int index
) { return; }
963 static int get_transition_latency(struct powernow_k8_data
*data
) { return 0; }
964 #endif /* CONFIG_X86_POWERNOW_K8_ACPI */
966 /* Take a frequency, and issue the fid/vid transition command */
967 static int transition_frequency_fidvid(struct powernow_k8_data
*data
, unsigned int index
)
972 struct cpufreq_freqs freqs
;
974 dprintk("cpu %d transition to index %u\n", smp_processor_id(), index
);
976 /* fid/vid correctness check for k8 */
977 /* fid are the lower 8 bits of the index we stored into
978 * the cpufreq frequency table in find_psb_table, vid
979 * are the upper 8 bits.
981 fid
= data
->powernow_table
[index
].index
& 0xFF;
982 vid
= (data
->powernow_table
[index
].index
& 0xFF00) >> 8;
984 dprintk("table matched fid 0x%x, giving vid 0x%x\n", fid
, vid
);
986 if (query_current_values_with_pending_wait(data
))
989 if ((data
->currvid
== vid
) && (data
->currfid
== fid
)) {
990 dprintk("target matches current values (fid 0x%x, vid 0x%x)\n",
995 if ((fid
< HI_FID_TABLE_BOTTOM
) && (data
->currfid
< HI_FID_TABLE_BOTTOM
)) {
997 "ignoring illegal change in lo freq table-%x to 0x%x\n",
1002 dprintk("cpu %d, changing to fid 0x%x, vid 0x%x\n",
1003 smp_processor_id(), fid
, vid
);
1004 freqs
.old
= find_khz_freq_from_fid(data
->currfid
);
1005 freqs
.new = find_khz_freq_from_fid(fid
);
1007 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1009 cpufreq_notify_transition(&freqs
, CPUFREQ_PRECHANGE
);
1012 res
= transition_fid_vid(data
, fid
, vid
);
1013 freqs
.new = find_khz_freq_from_fid(data
->currfid
);
1015 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1017 cpufreq_notify_transition(&freqs
, CPUFREQ_POSTCHANGE
);
1022 /* Take a frequency, and issue the hardware pstate transition command */
1023 static int transition_frequency_pstate(struct powernow_k8_data
*data
, unsigned int index
)
1027 struct cpufreq_freqs freqs
;
1029 dprintk("cpu %d transition to index %u\n", smp_processor_id(), index
);
1031 /* get MSR index for hardware pstate transition */
1032 pstate
= index
& HW_PSTATE_MASK
;
1033 if (pstate
> data
->max_hw_pstate
)
1035 freqs
.old
= find_khz_freq_from_pstate(data
->powernow_table
, data
->currpstate
);
1036 freqs
.new = find_khz_freq_from_pstate(data
->powernow_table
, pstate
);
1038 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1040 cpufreq_notify_transition(&freqs
, CPUFREQ_PRECHANGE
);
1043 res
= transition_pstate(data
, pstate
);
1044 freqs
.new = find_khz_freq_from_pstate(data
->powernow_table
, pstate
);
1046 for_each_cpu_mask_nr(i
, *(data
->available_cores
)) {
1048 cpufreq_notify_transition(&freqs
, CPUFREQ_POSTCHANGE
);
1053 /* Driver entry point to switch to the target frequency */
1054 static int powernowk8_target(struct cpufreq_policy
*pol
, unsigned targfreq
, unsigned relation
)
1057 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1060 unsigned int newstate
;
1066 checkfid
= data
->currfid
;
1067 checkvid
= data
->currvid
;
1069 /* only run on specific CPU from here on */
1070 oldmask
= current
->cpus_allowed
;
1071 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(pol
->cpu
));
1073 if (smp_processor_id() != pol
->cpu
) {
1074 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", pol
->cpu
);
1078 if (pending_bit_stuck()) {
1079 printk(KERN_ERR PFX
"failing targ, change pending bit set\n");
1083 dprintk("targ: cpu %d, %d kHz, min %d, max %d, relation %d\n",
1084 pol
->cpu
, targfreq
, pol
->min
, pol
->max
, relation
);
1086 if (query_current_values_with_pending_wait(data
))
1089 if (cpu_family
!= CPU_HW_PSTATE
) {
1090 dprintk("targ: curr fid 0x%x, vid 0x%x\n",
1091 data
->currfid
, data
->currvid
);
1093 if ((checkvid
!= data
->currvid
) || (checkfid
!= data
->currfid
)) {
1094 printk(KERN_INFO PFX
1095 "error - out of sync, fix 0x%x 0x%x, vid 0x%x 0x%x\n",
1096 checkfid
, data
->currfid
, checkvid
, data
->currvid
);
1100 if (cpufreq_frequency_table_target(pol
, data
->powernow_table
, targfreq
, relation
, &newstate
))
1103 mutex_lock(&fidvid_mutex
);
1105 powernow_k8_acpi_pst_values(data
, newstate
);
1107 if (cpu_family
== CPU_HW_PSTATE
)
1108 ret
= transition_frequency_pstate(data
, newstate
);
1110 ret
= transition_frequency_fidvid(data
, newstate
);
1112 printk(KERN_ERR PFX
"transition frequency failed\n");
1114 mutex_unlock(&fidvid_mutex
);
1117 mutex_unlock(&fidvid_mutex
);
1119 if (cpu_family
== CPU_HW_PSTATE
)
1120 pol
->cur
= find_khz_freq_from_pstate(data
->powernow_table
, newstate
);
1122 pol
->cur
= find_khz_freq_from_fid(data
->currfid
);
1126 set_cpus_allowed_ptr(current
, &oldmask
);
1130 /* Driver entry point to verify the policy and range of frequencies */
1131 static int powernowk8_verify(struct cpufreq_policy
*pol
)
1133 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1138 return cpufreq_frequency_table_verify(pol
, data
->powernow_table
);
1141 /* per CPU init entry point to the driver */
1142 static int __cpuinit
powernowk8_cpu_init(struct cpufreq_policy
*pol
)
1144 struct powernow_k8_data
*data
;
1148 if (!cpu_online(pol
->cpu
))
1151 if (!check_supported_cpu(pol
->cpu
))
1154 data
= kzalloc(sizeof(struct powernow_k8_data
), GFP_KERNEL
);
1156 printk(KERN_ERR PFX
"unable to alloc powernow_k8_data");
1160 data
->cpu
= pol
->cpu
;
1161 data
->currpstate
= HW_PSTATE_INVALID
;
1163 if (powernow_k8_cpu_init_acpi(data
)) {
1165 * Use the PSB BIOS structure. This is only availabe on
1166 * an UP version, and is deprecated by AMD.
1168 if (num_online_cpus() != 1) {
1169 #ifndef CONFIG_ACPI_PROCESSOR
1170 printk(KERN_ERR PFX
"ACPI Processor support is required "
1171 "for SMP systems but is absent. Please load the "
1172 "ACPI Processor module before starting this "
1175 printk(KERN_ERR FW_BUG PFX
"Your BIOS does not provide"
1176 " ACPI _PSS objects in a way that Linux "
1177 "understands. Please report this to the Linux "
1178 "ACPI maintainers and complain to your BIOS "
1184 if (pol
->cpu
!= 0) {
1185 printk(KERN_ERR FW_BUG PFX
"No ACPI _PSS objects for "
1186 "CPU other than CPU0. Complain to your BIOS "
1191 rc
= find_psb_table(data
);
1196 /* Take a crude guess here.
1197 * That guess was in microseconds, so multiply with 1000 */
1198 pol
->cpuinfo
.transition_latency
= (
1199 ((data
->rvo
+ 8) * data
->vstable
* VST_UNITS_20US
) +
1200 ((1 << data
->irt
) * 30)) * 1000;
1201 } else /* ACPI _PSS objects available */
1202 pol
->cpuinfo
.transition_latency
= get_transition_latency(data
);
1204 /* only run on specific CPU from here on */
1205 oldmask
= current
->cpus_allowed
;
1206 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(pol
->cpu
));
1208 if (smp_processor_id() != pol
->cpu
) {
1209 printk(KERN_ERR PFX
"limiting to cpu %u failed\n", pol
->cpu
);
1213 if (pending_bit_stuck()) {
1214 printk(KERN_ERR PFX
"failing init, change pending bit set\n");
1218 if (query_current_values_with_pending_wait(data
))
1221 if (cpu_family
== CPU_OPTERON
)
1224 /* run on any CPU again */
1225 set_cpus_allowed_ptr(current
, &oldmask
);
1227 if (cpu_family
== CPU_HW_PSTATE
)
1228 cpumask_copy(pol
->cpus
, cpumask_of(pol
->cpu
));
1230 cpumask_copy(pol
->cpus
, cpu_core_mask(pol
->cpu
));
1231 data
->available_cores
= pol
->cpus
;
1233 if (cpu_family
== CPU_HW_PSTATE
)
1234 pol
->cur
= find_khz_freq_from_pstate(data
->powernow_table
, data
->currpstate
);
1236 pol
->cur
= find_khz_freq_from_fid(data
->currfid
);
1237 dprintk("policy current frequency %d kHz\n", pol
->cur
);
1239 /* min/max the cpu is capable of */
1240 if (cpufreq_frequency_table_cpuinfo(pol
, data
->powernow_table
)) {
1241 printk(KERN_ERR FW_BUG PFX
"invalid powernow_table\n");
1242 powernow_k8_cpu_exit_acpi(data
);
1243 kfree(data
->powernow_table
);
1248 cpufreq_frequency_table_get_attr(data
->powernow_table
, pol
->cpu
);
1250 if (cpu_family
== CPU_HW_PSTATE
)
1251 dprintk("cpu_init done, current pstate 0x%x\n", data
->currpstate
);
1253 dprintk("cpu_init done, current fid 0x%x, vid 0x%x\n",
1254 data
->currfid
, data
->currvid
);
1256 per_cpu(powernow_data
, pol
->cpu
) = data
;
1261 set_cpus_allowed_ptr(current
, &oldmask
);
1262 powernow_k8_cpu_exit_acpi(data
);
1268 static int __devexit
powernowk8_cpu_exit (struct cpufreq_policy
*pol
)
1270 struct powernow_k8_data
*data
= per_cpu(powernow_data
, pol
->cpu
);
1275 powernow_k8_cpu_exit_acpi(data
);
1277 cpufreq_frequency_table_put_attr(pol
->cpu
);
1279 kfree(data
->powernow_table
);
1285 static unsigned int powernowk8_get (unsigned int cpu
)
1287 struct powernow_k8_data
*data
;
1288 cpumask_t oldmask
= current
->cpus_allowed
;
1289 unsigned int khz
= 0;
1292 first
= cpumask_first(cpu_core_mask(cpu
));
1293 data
= per_cpu(powernow_data
, first
);
1298 set_cpus_allowed_ptr(current
, &cpumask_of_cpu(cpu
));
1299 if (smp_processor_id() != cpu
) {
1301 "limiting to CPU %d failed in powernowk8_get\n", cpu
);
1302 set_cpus_allowed_ptr(current
, &oldmask
);
1306 if (query_current_values_with_pending_wait(data
))
1309 if (cpu_family
== CPU_HW_PSTATE
)
1310 khz
= find_khz_freq_from_pstate(data
->powernow_table
,
1313 khz
= find_khz_freq_from_fid(data
->currfid
);
1317 set_cpus_allowed_ptr(current
, &oldmask
);
1321 static struct freq_attr
* powernow_k8_attr
[] = {
1322 &cpufreq_freq_attr_scaling_available_freqs
,
1326 static struct cpufreq_driver cpufreq_amd64_driver
= {
1327 .verify
= powernowk8_verify
,
1328 .target
= powernowk8_target
,
1329 .init
= powernowk8_cpu_init
,
1330 .exit
= __devexit_p(powernowk8_cpu_exit
),
1331 .get
= powernowk8_get
,
1332 .name
= "powernow-k8",
1333 .owner
= THIS_MODULE
,
1334 .attr
= powernow_k8_attr
,
1337 /* driver entry point for init */
1338 static int __cpuinit
powernowk8_init(void)
1340 unsigned int i
, supported_cpus
= 0;
1342 for_each_online_cpu(i
) {
1343 if (check_supported_cpu(i
))
1347 if (supported_cpus
== num_online_cpus()) {
1348 printk(KERN_INFO PFX
"Found %d %s "
1349 "processors (%d cpu cores) (" VERSION
")\n",
1351 boot_cpu_data
.x86_model_id
, supported_cpus
);
1352 return cpufreq_register_driver(&cpufreq_amd64_driver
);
1358 /* driver entry point for term */
1359 static void __exit
powernowk8_exit(void)
1363 cpufreq_unregister_driver(&cpufreq_amd64_driver
);
1366 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com> and Mark Langsdorf <mark.langsdorf@amd.com>");
1367 MODULE_DESCRIPTION("AMD Athlon 64 and Opteron processor frequency driver.");
1368 MODULE_LICENSE("GPL");
1370 late_initcall(powernowk8_init
);
1371 module_exit(powernowk8_exit
);