xref: /openbmc/linux/drivers/platform/x86/intel_ips.c (revision 293d5b43)
1 /*
2  * Copyright (c) 2009-2010 Intel Corporation
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, write to the Free Software Foundation, Inc.,
15  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
16  *
17  * The full GNU General Public License is included in this distribution in
18  * the file called "COPYING".
19  *
20  * Authors:
21  *	Jesse Barnes <jbarnes@virtuousgeek.org>
22  */
23 
24 /*
25  * Some Intel Ibex Peak based platforms support so-called "intelligent
26  * power sharing", which allows the CPU and GPU to cooperate to maximize
27  * performance within a given TDP (thermal design point).  This driver
28  * performs the coordination between the CPU and GPU, monitors thermal and
29  * power statistics in the platform, and initializes power monitoring
30  * hardware.  It also provides a few tunables to control behavior.  Its
31  * primary purpose is to safely allow CPU and GPU turbo modes to be enabled
32  * by tracking power and thermal budget; secondarily it can boost turbo
33  * performance by allocating more power or thermal budget to the CPU or GPU
34  * based on available headroom and activity.
35  *
36  * The basic algorithm is driven by a 5s moving average of temperature.  If
37  * thermal headroom is available, the CPU and/or GPU power clamps may be
38  * adjusted upwards.  If we hit the thermal ceiling or a thermal trigger,
39  * we scale back the clamp.  Aside from trigger events (when we're critically
40  * close or over our TDP) we don't adjust the clamps more than once every
41  * five seconds.
42  *
43  * The thermal device (device 31, function 6) has a set of registers that
44  * are updated by the ME firmware.  The ME should also take the clamp values
45  * written to those registers and write them to the CPU, but we currently
46  * bypass that functionality and write the CPU MSR directly.
47  *
48  * UNSUPPORTED:
49  *   - dual MCP configs
50  *
51  * TODO:
52  *   - handle CPU hotplug
53  *   - provide turbo enable/disable api
54  *
55  * Related documents:
56  *   - CDI 403777, 403778 - Auburndale EDS vol 1 & 2
57  *   - CDI 401376 - Ibex Peak EDS
58  *   - ref 26037, 26641 - IPS BIOS spec
59  *   - ref 26489 - Nehalem BIOS writer's guide
60  *   - ref 26921 - Ibex Peak BIOS Specification
61  */
62 
63 #include <linux/debugfs.h>
64 #include <linux/delay.h>
65 #include <linux/interrupt.h>
66 #include <linux/kernel.h>
67 #include <linux/kthread.h>
68 #include <linux/module.h>
69 #include <linux/pci.h>
70 #include <linux/sched.h>
71 #include <linux/seq_file.h>
72 #include <linux/string.h>
73 #include <linux/tick.h>
74 #include <linux/timer.h>
75 #include <linux/dmi.h>
76 #include <drm/i915_drm.h>
77 #include <asm/msr.h>
78 #include <asm/processor.h>
79 #include "intel_ips.h"
80 
81 #include <linux/io-64-nonatomic-lo-hi.h>
82 
83 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
84 
85 /*
86  * Package level MSRs for monitor/control
87  */
88 #define PLATFORM_INFO	0xce
89 #define   PLATFORM_TDP		(1<<29)
90 #define   PLATFORM_RATIO	(1<<28)
91 
92 #define IA32_MISC_ENABLE	0x1a0
93 #define   IA32_MISC_TURBO_EN	(1ULL<<38)
94 
95 #define TURBO_POWER_CURRENT_LIMIT	0x1ac
96 #define   TURBO_TDC_OVR_EN	(1UL<<31)
97 #define   TURBO_TDC_MASK	(0x000000007fff0000UL)
98 #define   TURBO_TDC_SHIFT	(16)
99 #define   TURBO_TDP_OVR_EN	(1UL<<15)
100 #define   TURBO_TDP_MASK	(0x0000000000003fffUL)
101 
102 /*
103  * Core/thread MSRs for monitoring
104  */
105 #define IA32_PERF_CTL		0x199
106 #define   IA32_PERF_TURBO_DIS	(1ULL<<32)
107 
108 /*
109  * Thermal PCI device regs
110  */
111 #define THM_CFG_TBAR	0x10
112 #define THM_CFG_TBAR_HI	0x14
113 
114 #define THM_TSIU	0x00
115 #define THM_TSE		0x01
116 #define   TSE_EN	0xb8
117 #define THM_TSS		0x02
118 #define THM_TSTR	0x03
119 #define THM_TSTTP	0x04
120 #define THM_TSCO	0x08
121 #define THM_TSES	0x0c
122 #define THM_TSGPEN	0x0d
123 #define   TSGPEN_HOT_LOHI	(1<<1)
124 #define   TSGPEN_CRIT_LOHI	(1<<2)
125 #define THM_TSPC	0x0e
126 #define THM_PPEC	0x10
127 #define THM_CTA		0x12
128 #define THM_PTA		0x14
129 #define   PTA_SLOPE_MASK	(0xff00)
130 #define   PTA_SLOPE_SHIFT	8
131 #define   PTA_OFFSET_MASK	(0x00ff)
132 #define THM_MGTA	0x16
133 #define   MGTA_SLOPE_MASK	(0xff00)
134 #define   MGTA_SLOPE_SHIFT	8
135 #define   MGTA_OFFSET_MASK	(0x00ff)
136 #define THM_TRC		0x1a
137 #define   TRC_CORE2_EN	(1<<15)
138 #define   TRC_THM_EN	(1<<12)
139 #define   TRC_C6_WAR	(1<<8)
140 #define   TRC_CORE1_EN	(1<<7)
141 #define   TRC_CORE_PWR	(1<<6)
142 #define   TRC_PCH_EN	(1<<5)
143 #define   TRC_MCH_EN	(1<<4)
144 #define   TRC_DIMM4	(1<<3)
145 #define   TRC_DIMM3	(1<<2)
146 #define   TRC_DIMM2	(1<<1)
147 #define   TRC_DIMM1	(1<<0)
148 #define THM_TES		0x20
149 #define THM_TEN		0x21
150 #define   TEN_UPDATE_EN	1
151 #define THM_PSC		0x24
152 #define   PSC_NTG	(1<<0) /* No GFX turbo support */
153 #define   PSC_NTPC	(1<<1) /* No CPU turbo support */
154 #define   PSC_PP_DEF	(0<<2) /* Perf policy up to driver */
155 #define   PSP_PP_PC	(1<<2) /* BIOS prefers CPU perf */
156 #define   PSP_PP_BAL	(2<<2) /* BIOS wants balanced perf */
157 #define   PSP_PP_GFX	(3<<2) /* BIOS prefers GFX perf */
158 #define   PSP_PBRT	(1<<4) /* BIOS run time support */
159 #define THM_CTV1	0x30
160 #define   CTV_TEMP_ERROR (1<<15)
161 #define   CTV_TEMP_MASK	0x3f
162 #define   CTV_
163 #define THM_CTV2	0x32
164 #define THM_CEC		0x34 /* undocumented power accumulator in joules */
165 #define THM_AE		0x3f
166 #define THM_HTS		0x50 /* 32 bits */
167 #define   HTS_PCPL_MASK	(0x7fe00000)
168 #define   HTS_PCPL_SHIFT 21
169 #define   HTS_GPL_MASK  (0x001ff000)
170 #define   HTS_GPL_SHIFT 12
171 #define   HTS_PP_MASK	(0x00000c00)
172 #define   HTS_PP_SHIFT  10
173 #define   HTS_PP_DEF	0
174 #define   HTS_PP_PROC	1
175 #define   HTS_PP_BAL	2
176 #define   HTS_PP_GFX	3
177 #define   HTS_PCTD_DIS	(1<<9)
178 #define   HTS_GTD_DIS	(1<<8)
179 #define   HTS_PTL_MASK  (0x000000fe)
180 #define   HTS_PTL_SHIFT 1
181 #define   HTS_NVV	(1<<0)
182 #define THM_HTSHI	0x54 /* 16 bits */
183 #define   HTS2_PPL_MASK		(0x03ff)
184 #define   HTS2_PRST_MASK	(0x3c00)
185 #define   HTS2_PRST_SHIFT	10
186 #define   HTS2_PRST_UNLOADED	0
187 #define   HTS2_PRST_RUNNING	1
188 #define   HTS2_PRST_TDISOP	2 /* turbo disabled due to power */
189 #define   HTS2_PRST_TDISHT	3 /* turbo disabled due to high temp */
190 #define   HTS2_PRST_TDISUSR	4 /* user disabled turbo */
191 #define   HTS2_PRST_TDISPLAT	5 /* platform disabled turbo */
192 #define   HTS2_PRST_TDISPM	6 /* power management disabled turbo */
193 #define   HTS2_PRST_TDISERR	7 /* some kind of error disabled turbo */
194 #define THM_PTL		0x56
195 #define THM_MGTV	0x58
196 #define   TV_MASK	0x000000000000ff00
197 #define   TV_SHIFT	8
198 #define THM_PTV		0x60
199 #define   PTV_MASK	0x00ff
200 #define THM_MMGPC	0x64
201 #define THM_MPPC	0x66
202 #define THM_MPCPC	0x68
203 #define THM_TSPIEN	0x82
204 #define   TSPIEN_AUX_LOHI	(1<<0)
205 #define   TSPIEN_HOT_LOHI	(1<<1)
206 #define   TSPIEN_CRIT_LOHI	(1<<2)
207 #define   TSPIEN_AUX2_LOHI	(1<<3)
208 #define THM_TSLOCK	0x83
209 #define THM_ATR		0x84
210 #define THM_TOF		0x87
211 #define THM_STS		0x98
212 #define   STS_PCPL_MASK		(0x7fe00000)
213 #define   STS_PCPL_SHIFT	21
214 #define   STS_GPL_MASK		(0x001ff000)
215 #define   STS_GPL_SHIFT		12
216 #define   STS_PP_MASK		(0x00000c00)
217 #define   STS_PP_SHIFT		10
218 #define   STS_PP_DEF		0
219 #define   STS_PP_PROC		1
220 #define   STS_PP_BAL		2
221 #define   STS_PP_GFX		3
222 #define   STS_PCTD_DIS		(1<<9)
223 #define   STS_GTD_DIS		(1<<8)
224 #define   STS_PTL_MASK		(0x000000fe)
225 #define   STS_PTL_SHIFT		1
226 #define   STS_NVV		(1<<0)
227 #define THM_SEC		0x9c
228 #define   SEC_ACK	(1<<0)
229 #define THM_TC3		0xa4
230 #define THM_TC1		0xa8
231 #define   STS_PPL_MASK		(0x0003ff00)
232 #define   STS_PPL_SHIFT		16
233 #define THM_TC2		0xac
234 #define THM_DTV		0xb0
235 #define THM_ITV		0xd8
236 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
237 #define   ITV_ME_SEQNO_SHIFT (16)
238 #define   ITV_MCH_TEMP_MASK 0x0000ff00
239 #define   ITV_MCH_TEMP_SHIFT (8)
240 #define   ITV_PCH_TEMP_MASK 0x000000ff
241 
242 #define thm_readb(off) readb(ips->regmap + (off))
243 #define thm_readw(off) readw(ips->regmap + (off))
244 #define thm_readl(off) readl(ips->regmap + (off))
245 #define thm_readq(off) readq(ips->regmap + (off))
246 
247 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
248 #define thm_writew(off, val) writew((val), ips->regmap + (off))
249 #define thm_writel(off, val) writel((val), ips->regmap + (off))
250 
251 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
252 static bool late_i915_load = false;
253 
254 /* For initial average collection */
255 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
256 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
257 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
258 
259 /* Per-SKU limits */
260 struct ips_mcp_limits {
261 	int cpu_family;
262 	int cpu_model; /* includes extended model... */
263 	int mcp_power_limit; /* mW units */
264 	int core_power_limit;
265 	int mch_power_limit;
266 	int core_temp_limit; /* degrees C */
267 	int mch_temp_limit;
268 };
269 
270 /* Max temps are -10 degrees C to avoid PROCHOT# */
271 
272 static struct ips_mcp_limits ips_sv_limits = {
273 	.mcp_power_limit = 35000,
274 	.core_power_limit = 29000,
275 	.mch_power_limit = 20000,
276 	.core_temp_limit = 95,
277 	.mch_temp_limit = 90
278 };
279 
280 static struct ips_mcp_limits ips_lv_limits = {
281 	.mcp_power_limit = 25000,
282 	.core_power_limit = 21000,
283 	.mch_power_limit = 13000,
284 	.core_temp_limit = 95,
285 	.mch_temp_limit = 90
286 };
287 
288 static struct ips_mcp_limits ips_ulv_limits = {
289 	.mcp_power_limit = 18000,
290 	.core_power_limit = 14000,
291 	.mch_power_limit = 11000,
292 	.core_temp_limit = 95,
293 	.mch_temp_limit = 90
294 };
295 
296 struct ips_driver {
297 	struct pci_dev *dev;
298 	void *regmap;
299 	struct task_struct *monitor;
300 	struct task_struct *adjust;
301 	struct dentry *debug_root;
302 
303 	/* Average CPU core temps (all averages in .01 degrees C for precision) */
304 	u16 ctv1_avg_temp;
305 	u16 ctv2_avg_temp;
306 	/* GMCH average */
307 	u16 mch_avg_temp;
308 	/* Average for the CPU (both cores?) */
309 	u16 mcp_avg_temp;
310 	/* Average power consumption (in mW) */
311 	u32 cpu_avg_power;
312 	u32 mch_avg_power;
313 
314 	/* Offset values */
315 	u16 cta_val;
316 	u16 pta_val;
317 	u16 mgta_val;
318 
319 	/* Maximums & prefs, protected by turbo status lock */
320 	spinlock_t turbo_status_lock;
321 	u16 mcp_temp_limit;
322 	u16 mcp_power_limit;
323 	u16 core_power_limit;
324 	u16 mch_power_limit;
325 	bool cpu_turbo_enabled;
326 	bool __cpu_turbo_on;
327 	bool gpu_turbo_enabled;
328 	bool __gpu_turbo_on;
329 	bool gpu_preferred;
330 	bool poll_turbo_status;
331 	bool second_cpu;
332 	bool turbo_toggle_allowed;
333 	struct ips_mcp_limits *limits;
334 
335 	/* Optional MCH interfaces for if i915 is in use */
336 	unsigned long (*read_mch_val)(void);
337 	bool (*gpu_raise)(void);
338 	bool (*gpu_lower)(void);
339 	bool (*gpu_busy)(void);
340 	bool (*gpu_turbo_disable)(void);
341 
342 	/* For restoration at unload */
343 	u64 orig_turbo_limit;
344 	u64 orig_turbo_ratios;
345 };
346 
347 static bool
348 ips_gpu_turbo_enabled(struct ips_driver *ips);
349 
350 /**
351  * ips_cpu_busy - is CPU busy?
352  * @ips: IPS driver struct
353  *
354  * Check CPU for load to see whether we should increase its thermal budget.
355  *
356  * RETURNS:
357  * True if the CPU could use more power, false otherwise.
358  */
359 static bool ips_cpu_busy(struct ips_driver *ips)
360 {
361 	if ((avenrun[0] >> FSHIFT) > 1)
362 		return true;
363 
364 	return false;
365 }
366 
367 /**
368  * ips_cpu_raise - raise CPU power clamp
369  * @ips: IPS driver struct
370  *
371  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
372  * this platform.
373  *
374  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
375  * long as we haven't hit the TDP limit for the SKU).
376  */
377 static void ips_cpu_raise(struct ips_driver *ips)
378 {
379 	u64 turbo_override;
380 	u16 cur_tdp_limit, new_tdp_limit;
381 
382 	if (!ips->cpu_turbo_enabled)
383 		return;
384 
385 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
386 
387 	cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
388 	new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
389 
390 	/* Clamp to SKU TDP limit */
391 	if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
392 		new_tdp_limit = cur_tdp_limit;
393 
394 	thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
395 
396 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
397 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
398 
399 	turbo_override &= ~TURBO_TDP_MASK;
400 	turbo_override |= new_tdp_limit;
401 
402 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
403 }
404 
405 /**
406  * ips_cpu_lower - lower CPU power clamp
407  * @ips: IPS driver struct
408  *
409  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
410  *
411  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
412  * as low as the platform limits will allow (though we could go lower there
413  * wouldn't be much point).
414  */
415 static void ips_cpu_lower(struct ips_driver *ips)
416 {
417 	u64 turbo_override;
418 	u16 cur_limit, new_limit;
419 
420 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
421 
422 	cur_limit = turbo_override & TURBO_TDP_MASK;
423 	new_limit = cur_limit - 8; /* 1W decrease */
424 
425 	/* Clamp to SKU TDP limit */
426 	if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
427 		new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
428 
429 	thm_writew(THM_MPCPC, (new_limit * 10) / 8);
430 
431 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
432 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
433 
434 	turbo_override &= ~TURBO_TDP_MASK;
435 	turbo_override |= new_limit;
436 
437 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
438 }
439 
440 /**
441  * do_enable_cpu_turbo - internal turbo enable function
442  * @data: unused
443  *
444  * Internal function for actually updating MSRs.  When we enable/disable
445  * turbo, we need to do it on each CPU; this function is the one called
446  * by on_each_cpu() when needed.
447  */
448 static void do_enable_cpu_turbo(void *data)
449 {
450 	u64 perf_ctl;
451 
452 	rdmsrl(IA32_PERF_CTL, perf_ctl);
453 	if (perf_ctl & IA32_PERF_TURBO_DIS) {
454 		perf_ctl &= ~IA32_PERF_TURBO_DIS;
455 		wrmsrl(IA32_PERF_CTL, perf_ctl);
456 	}
457 }
458 
459 /**
460  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
461  * @ips: IPS driver struct
462  *
463  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
464  * all logical threads.
465  */
466 static void ips_enable_cpu_turbo(struct ips_driver *ips)
467 {
468 	/* Already on, no need to mess with MSRs */
469 	if (ips->__cpu_turbo_on)
470 		return;
471 
472 	if (ips->turbo_toggle_allowed)
473 		on_each_cpu(do_enable_cpu_turbo, ips, 1);
474 
475 	ips->__cpu_turbo_on = true;
476 }
477 
478 /**
479  * do_disable_cpu_turbo - internal turbo disable function
480  * @data: unused
481  *
482  * Internal function for actually updating MSRs.  When we enable/disable
483  * turbo, we need to do it on each CPU; this function is the one called
484  * by on_each_cpu() when needed.
485  */
486 static void do_disable_cpu_turbo(void *data)
487 {
488 	u64 perf_ctl;
489 
490 	rdmsrl(IA32_PERF_CTL, perf_ctl);
491 	if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
492 		perf_ctl |= IA32_PERF_TURBO_DIS;
493 		wrmsrl(IA32_PERF_CTL, perf_ctl);
494 	}
495 }
496 
497 /**
498  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
499  * @ips: IPS driver struct
500  *
501  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
502  * all logical threads.
503  */
504 static void ips_disable_cpu_turbo(struct ips_driver *ips)
505 {
506 	/* Already off, leave it */
507 	if (!ips->__cpu_turbo_on)
508 		return;
509 
510 	if (ips->turbo_toggle_allowed)
511 		on_each_cpu(do_disable_cpu_turbo, ips, 1);
512 
513 	ips->__cpu_turbo_on = false;
514 }
515 
516 /**
517  * ips_gpu_busy - is GPU busy?
518  * @ips: IPS driver struct
519  *
520  * Check GPU for load to see whether we should increase its thermal budget.
521  * We need to call into the i915 driver in this case.
522  *
523  * RETURNS:
524  * True if the GPU could use more power, false otherwise.
525  */
526 static bool ips_gpu_busy(struct ips_driver *ips)
527 {
528 	if (!ips_gpu_turbo_enabled(ips))
529 		return false;
530 
531 	return ips->gpu_busy();
532 }
533 
534 /**
535  * ips_gpu_raise - raise GPU power clamp
536  * @ips: IPS driver struct
537  *
538  * Raise the GPU frequency/power if possible.  We need to call into the
539  * i915 driver in this case.
540  */
541 static void ips_gpu_raise(struct ips_driver *ips)
542 {
543 	if (!ips_gpu_turbo_enabled(ips))
544 		return;
545 
546 	if (!ips->gpu_raise())
547 		ips->gpu_turbo_enabled = false;
548 
549 	return;
550 }
551 
552 /**
553  * ips_gpu_lower - lower GPU power clamp
554  * @ips: IPS driver struct
555  *
556  * Lower GPU frequency/power if possible.  Need to call i915.
557  */
558 static void ips_gpu_lower(struct ips_driver *ips)
559 {
560 	if (!ips_gpu_turbo_enabled(ips))
561 		return;
562 
563 	if (!ips->gpu_lower())
564 		ips->gpu_turbo_enabled = false;
565 
566 	return;
567 }
568 
569 /**
570  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
571  * @ips: IPS driver struct
572  *
573  * Call into the graphics driver indicating that it can safely use
574  * turbo mode.
575  */
576 static void ips_enable_gpu_turbo(struct ips_driver *ips)
577 {
578 	if (ips->__gpu_turbo_on)
579 		return;
580 	ips->__gpu_turbo_on = true;
581 }
582 
583 /**
584  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
585  * @ips: IPS driver struct
586  *
587  * Request that the graphics driver disable turbo mode.
588  */
589 static void ips_disable_gpu_turbo(struct ips_driver *ips)
590 {
591 	/* Avoid calling i915 if turbo is already disabled */
592 	if (!ips->__gpu_turbo_on)
593 		return;
594 
595 	if (!ips->gpu_turbo_disable())
596 		dev_err(&ips->dev->dev, "failed to disable graphics turbo\n");
597 	else
598 		ips->__gpu_turbo_on = false;
599 }
600 
601 /**
602  * mcp_exceeded - check whether we're outside our thermal & power limits
603  * @ips: IPS driver struct
604  *
605  * Check whether the MCP is over its thermal or power budget.
606  */
607 static bool mcp_exceeded(struct ips_driver *ips)
608 {
609 	unsigned long flags;
610 	bool ret = false;
611 	u32 temp_limit;
612 	u32 avg_power;
613 
614 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
615 
616 	temp_limit = ips->mcp_temp_limit * 100;
617 	if (ips->mcp_avg_temp > temp_limit)
618 		ret = true;
619 
620 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
621 	if (avg_power > ips->mcp_power_limit)
622 		ret = true;
623 
624 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
625 
626 	return ret;
627 }
628 
629 /**
630  * cpu_exceeded - check whether a CPU core is outside its limits
631  * @ips: IPS driver struct
632  * @cpu: CPU number to check
633  *
634  * Check a given CPU's average temp or power is over its limit.
635  */
636 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
637 {
638 	unsigned long flags;
639 	int avg;
640 	bool ret = false;
641 
642 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
643 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
644 	if (avg > (ips->limits->core_temp_limit * 100))
645 		ret = true;
646 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
647 		ret = true;
648 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
649 
650 	if (ret)
651 		dev_info(&ips->dev->dev,
652 			 "CPU power or thermal limit exceeded\n");
653 
654 	return ret;
655 }
656 
657 /**
658  * mch_exceeded - check whether the GPU is over budget
659  * @ips: IPS driver struct
660  *
661  * Check the MCH temp & power against their maximums.
662  */
663 static bool mch_exceeded(struct ips_driver *ips)
664 {
665 	unsigned long flags;
666 	bool ret = false;
667 
668 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
669 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
670 		ret = true;
671 	if (ips->mch_avg_power > ips->mch_power_limit)
672 		ret = true;
673 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
674 
675 	return ret;
676 }
677 
678 /**
679  * verify_limits - verify BIOS provided limits
680  * @ips: IPS structure
681  *
682  * BIOS can optionally provide non-default limits for power and temp.  Check
683  * them here and use the defaults if the BIOS values are not provided or
684  * are otherwise unusable.
685  */
686 static void verify_limits(struct ips_driver *ips)
687 {
688 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
689 	    ips->mcp_power_limit > 35000)
690 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
691 
692 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
693 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
694 	    ips->mcp_temp_limit > 150)
695 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
696 					  ips->limits->mch_temp_limit);
697 }
698 
699 /**
700  * update_turbo_limits - get various limits & settings from regs
701  * @ips: IPS driver struct
702  *
703  * Update the IPS power & temp limits, along with turbo enable flags,
704  * based on latest register contents.
705  *
706  * Used at init time and for runtime BIOS support, which requires polling
707  * the regs for updates (as a result of AC->DC transition for example).
708  *
709  * LOCKING:
710  * Caller must hold turbo_status_lock (outside of init)
711  */
712 static void update_turbo_limits(struct ips_driver *ips)
713 {
714 	u32 hts = thm_readl(THM_HTS);
715 
716 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
717 	/*
718 	 * Disable turbo for now, until we can figure out why the power figures
719 	 * are wrong
720 	 */
721 	ips->cpu_turbo_enabled = false;
722 
723 	if (ips->gpu_busy)
724 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
725 
726 	ips->core_power_limit = thm_readw(THM_MPCPC);
727 	ips->mch_power_limit = thm_readw(THM_MMGPC);
728 	ips->mcp_temp_limit = thm_readw(THM_PTL);
729 	ips->mcp_power_limit = thm_readw(THM_MPPC);
730 
731 	verify_limits(ips);
732 	/* Ignore BIOS CPU vs GPU pref */
733 }
734 
735 /**
736  * ips_adjust - adjust power clamp based on thermal state
737  * @data: ips driver structure
738  *
739  * Wake up every 5s or so and check whether we should adjust the power clamp.
740  * Check CPU and GPU load to determine which needs adjustment.  There are
741  * several things to consider here:
742  *   - do we need to adjust up or down?
743  *   - is CPU busy?
744  *   - is GPU busy?
745  *   - is CPU in turbo?
746  *   - is GPU in turbo?
747  *   - is CPU or GPU preferred? (CPU is default)
748  *
749  * So, given the above, we do the following:
750  *   - up (TDP available)
751  *     - CPU not busy, GPU not busy - nothing
752  *     - CPU busy, GPU not busy - adjust CPU up
753  *     - CPU not busy, GPU busy - adjust GPU up
754  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
755  *       non-preferred unit if necessary
756  *   - down (at TDP limit)
757  *     - adjust both CPU and GPU down if possible
758  *
759 		cpu+ gpu+	cpu+gpu-	cpu-gpu+	cpu-gpu-
760 cpu < gpu <	cpu+gpu+	cpu+		gpu+		nothing
761 cpu < gpu >=	cpu+gpu-(mcp<)	cpu+gpu-(mcp<)	gpu-		gpu-
762 cpu >= gpu <	cpu-gpu+(mcp<)	cpu-		cpu-gpu+(mcp<)	cpu-
763 cpu >= gpu >=	cpu-gpu-	cpu-gpu-	cpu-gpu-	cpu-gpu-
764  *
765  */
766 static int ips_adjust(void *data)
767 {
768 	struct ips_driver *ips = data;
769 	unsigned long flags;
770 
771 	dev_dbg(&ips->dev->dev, "starting ips-adjust thread\n");
772 
773 	/*
774 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
775 	 * often isn't recommended due to ME interaction.
776 	 */
777 	do {
778 		bool cpu_busy = ips_cpu_busy(ips);
779 		bool gpu_busy = ips_gpu_busy(ips);
780 
781 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
782 		if (ips->poll_turbo_status)
783 			update_turbo_limits(ips);
784 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
785 
786 		/* Update turbo status if necessary */
787 		if (ips->cpu_turbo_enabled)
788 			ips_enable_cpu_turbo(ips);
789 		else
790 			ips_disable_cpu_turbo(ips);
791 
792 		if (ips->gpu_turbo_enabled)
793 			ips_enable_gpu_turbo(ips);
794 		else
795 			ips_disable_gpu_turbo(ips);
796 
797 		/* We're outside our comfort zone, crank them down */
798 		if (mcp_exceeded(ips)) {
799 			ips_cpu_lower(ips);
800 			ips_gpu_lower(ips);
801 			goto sleep;
802 		}
803 
804 		if (!cpu_exceeded(ips, 0) && cpu_busy)
805 			ips_cpu_raise(ips);
806 		else
807 			ips_cpu_lower(ips);
808 
809 		if (!mch_exceeded(ips) && gpu_busy)
810 			ips_gpu_raise(ips);
811 		else
812 			ips_gpu_lower(ips);
813 
814 sleep:
815 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
816 	} while (!kthread_should_stop());
817 
818 	dev_dbg(&ips->dev->dev, "ips-adjust thread stopped\n");
819 
820 	return 0;
821 }
822 
823 /*
824  * Helpers for reading out temp/power values and calculating their
825  * averages for the decision making and monitoring functions.
826  */
827 
828 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
829 {
830 	u64 total = 0;
831 	int i;
832 	u16 avg;
833 
834 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
835 		total += (u64)(array[i] * 100);
836 
837 	do_div(total, IPS_SAMPLE_COUNT);
838 
839 	avg = (u16)total;
840 
841 	return avg;
842 }
843 
844 static u16 read_mgtv(struct ips_driver *ips)
845 {
846 	u16 ret;
847 	u64 slope, offset;
848 	u64 val;
849 
850 	val = thm_readq(THM_MGTV);
851 	val = (val & TV_MASK) >> TV_SHIFT;
852 
853 	slope = offset = thm_readw(THM_MGTA);
854 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
855 	offset = offset & MGTA_OFFSET_MASK;
856 
857 	ret = ((val * slope + 0x40) >> 7) + offset;
858 
859 	return 0; /* MCH temp reporting buggy */
860 }
861 
862 static u16 read_ptv(struct ips_driver *ips)
863 {
864 	u16 val, slope, offset;
865 
866 	slope = (ips->pta_val & PTA_SLOPE_MASK) >> PTA_SLOPE_SHIFT;
867 	offset = ips->pta_val & PTA_OFFSET_MASK;
868 
869 	val = thm_readw(THM_PTV) & PTV_MASK;
870 
871 	return val;
872 }
873 
874 static u16 read_ctv(struct ips_driver *ips, int cpu)
875 {
876 	int reg = cpu ? THM_CTV2 : THM_CTV1;
877 	u16 val;
878 
879 	val = thm_readw(reg);
880 	if (!(val & CTV_TEMP_ERROR))
881 		val = (val) >> 6; /* discard fractional component */
882 	else
883 		val = 0;
884 
885 	return val;
886 }
887 
888 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
889 {
890 	u32 val;
891 	u32 ret;
892 
893 	/*
894 	 * CEC is in joules/65535.  Take difference over time to
895 	 * get watts.
896 	 */
897 	val = thm_readl(THM_CEC);
898 
899 	/* period is in ms and we want mW */
900 	ret = (((val - *last) * 1000) / period);
901 	ret = (ret * 1000) / 65535;
902 	*last = val;
903 
904 	return 0;
905 }
906 
907 static const u16 temp_decay_factor = 2;
908 static u16 update_average_temp(u16 avg, u16 val)
909 {
910 	u16 ret;
911 
912 	/* Multiply by 100 for extra precision */
913 	ret = (val * 100 / temp_decay_factor) +
914 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
915 	return ret;
916 }
917 
918 static const u16 power_decay_factor = 2;
919 static u16 update_average_power(u32 avg, u32 val)
920 {
921 	u32 ret;
922 
923 	ret = (val / power_decay_factor) +
924 		(((power_decay_factor - 1) * avg) / power_decay_factor);
925 
926 	return ret;
927 }
928 
929 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
930 {
931 	u64 total = 0;
932 	u32 avg;
933 	int i;
934 
935 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
936 		total += array[i];
937 
938 	do_div(total, IPS_SAMPLE_COUNT);
939 	avg = (u32)total;
940 
941 	return avg;
942 }
943 
944 static void monitor_timeout(unsigned long arg)
945 {
946 	wake_up_process((struct task_struct *)arg);
947 }
948 
949 /**
950  * ips_monitor - temp/power monitoring thread
951  * @data: ips driver structure
952  *
953  * This is the main function for the IPS driver.  It monitors power and
954  * tempurature in the MCP and adjusts CPU and GPU power clams accordingly.
955  *
956  * We keep a 5s moving average of power consumption and tempurature.  Using
957  * that data, along with CPU vs GPU preference, we adjust the power clamps
958  * up or down.
959  */
960 static int ips_monitor(void *data)
961 {
962 	struct ips_driver *ips = data;
963 	struct timer_list timer;
964 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
965 	int i;
966 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
967 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
968 	u8 cur_seqno, last_seqno;
969 
970 	mcp_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
971 	ctv1_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
972 	ctv2_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
973 	mch_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
974 	cpu_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
975 	mchp_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
976 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
977 			!cpu_samples || !mchp_samples) {
978 		dev_err(&ips->dev->dev,
979 			"failed to allocate sample array, ips disabled\n");
980 		kfree(mcp_samples);
981 		kfree(ctv1_samples);
982 		kfree(ctv2_samples);
983 		kfree(mch_samples);
984 		kfree(cpu_samples);
985 		kfree(mchp_samples);
986 		return -ENOMEM;
987 	}
988 
989 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
990 		ITV_ME_SEQNO_SHIFT;
991 	seqno_timestamp = get_jiffies_64();
992 
993 	old_cpu_power = thm_readl(THM_CEC);
994 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
995 
996 	/* Collect an initial average */
997 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
998 		u32 mchp, cpu_power;
999 		u16 val;
1000 
1001 		mcp_samples[i] = read_ptv(ips);
1002 
1003 		val = read_ctv(ips, 0);
1004 		ctv1_samples[i] = val;
1005 
1006 		val = read_ctv(ips, 1);
1007 		ctv2_samples[i] = val;
1008 
1009 		val = read_mgtv(ips);
1010 		mch_samples[i] = val;
1011 
1012 		cpu_power = get_cpu_power(ips, &old_cpu_power,
1013 					  IPS_SAMPLE_PERIOD);
1014 		cpu_samples[i] = cpu_power;
1015 
1016 		if (ips->read_mch_val) {
1017 			mchp = ips->read_mch_val();
1018 			mchp_samples[i] = mchp;
1019 		}
1020 
1021 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1022 		if (kthread_should_stop())
1023 			break;
1024 	}
1025 
1026 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1027 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1028 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1029 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1030 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1031 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1032 	kfree(mcp_samples);
1033 	kfree(ctv1_samples);
1034 	kfree(ctv2_samples);
1035 	kfree(mch_samples);
1036 	kfree(cpu_samples);
1037 	kfree(mchp_samples);
1038 
1039 	/* Start the adjustment thread now that we have data */
1040 	wake_up_process(ips->adjust);
1041 
1042 	/*
1043 	 * Ok, now we have an initial avg.  From here on out, we track the
1044 	 * running avg using a decaying average calculation.  This allows
1045 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1046 	 */
1047 	old_cpu_power = thm_readl(THM_CEC);
1048 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1049 	last_sample_period = IPS_SAMPLE_PERIOD;
1050 
1051 	setup_deferrable_timer_on_stack(&timer, monitor_timeout,
1052 					(unsigned long)current);
1053 	do {
1054 		u32 cpu_val, mch_val;
1055 		u16 val;
1056 
1057 		/* MCP itself */
1058 		val = read_ptv(ips);
1059 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1060 
1061 		/* Processor 0 */
1062 		val = read_ctv(ips, 0);
1063 		ips->ctv1_avg_temp =
1064 			update_average_temp(ips->ctv1_avg_temp, val);
1065 		/* Power */
1066 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1067 					last_sample_period);
1068 		ips->cpu_avg_power =
1069 			update_average_power(ips->cpu_avg_power, cpu_val);
1070 
1071 		if (ips->second_cpu) {
1072 			/* Processor 1 */
1073 			val = read_ctv(ips, 1);
1074 			ips->ctv2_avg_temp =
1075 				update_average_temp(ips->ctv2_avg_temp, val);
1076 		}
1077 
1078 		/* MCH */
1079 		val = read_mgtv(ips);
1080 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1081 		/* Power */
1082 		if (ips->read_mch_val) {
1083 			mch_val = ips->read_mch_val();
1084 			ips->mch_avg_power =
1085 				update_average_power(ips->mch_avg_power,
1086 						     mch_val);
1087 		}
1088 
1089 		/*
1090 		 * Make sure ME is updating thermal regs.
1091 		 * Note:
1092 		 * If it's been more than a second since the last update,
1093 		 * the ME is probably hung.
1094 		 */
1095 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1096 			ITV_ME_SEQNO_SHIFT;
1097 		if (cur_seqno == last_seqno &&
1098 		    time_after(jiffies, seqno_timestamp + HZ)) {
1099 			dev_warn(&ips->dev->dev, "ME failed to update for more than 1s, likely hung\n");
1100 		} else {
1101 			seqno_timestamp = get_jiffies_64();
1102 			last_seqno = cur_seqno;
1103 		}
1104 
1105 		last_msecs = jiffies_to_msecs(jiffies);
1106 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1107 
1108 		__set_current_state(TASK_INTERRUPTIBLE);
1109 		mod_timer(&timer, expire);
1110 		schedule();
1111 
1112 		/* Calculate actual sample period for power averaging */
1113 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1114 		if (!last_sample_period)
1115 			last_sample_period = 1;
1116 	} while (!kthread_should_stop());
1117 
1118 	del_timer_sync(&timer);
1119 	destroy_timer_on_stack(&timer);
1120 
1121 	dev_dbg(&ips->dev->dev, "ips-monitor thread stopped\n");
1122 
1123 	return 0;
1124 }
1125 
1126 #if 0
1127 #define THM_DUMPW(reg) \
1128 	{ \
1129 	u16 val = thm_readw(reg); \
1130 	dev_dbg(&ips->dev->dev, #reg ": 0x%04x\n", val); \
1131 	}
1132 #define THM_DUMPL(reg) \
1133 	{ \
1134 	u32 val = thm_readl(reg); \
1135 	dev_dbg(&ips->dev->dev, #reg ": 0x%08x\n", val); \
1136 	}
1137 #define THM_DUMPQ(reg) \
1138 	{ \
1139 	u64 val = thm_readq(reg); \
1140 	dev_dbg(&ips->dev->dev, #reg ": 0x%016x\n", val); \
1141 	}
1142 
1143 static void dump_thermal_info(struct ips_driver *ips)
1144 {
1145 	u16 ptl;
1146 
1147 	ptl = thm_readw(THM_PTL);
1148 	dev_dbg(&ips->dev->dev, "Processor temp limit: %d\n", ptl);
1149 
1150 	THM_DUMPW(THM_CTA);
1151 	THM_DUMPW(THM_TRC);
1152 	THM_DUMPW(THM_CTV1);
1153 	THM_DUMPL(THM_STS);
1154 	THM_DUMPW(THM_PTV);
1155 	THM_DUMPQ(THM_MGTV);
1156 }
1157 #endif
1158 
1159 /**
1160  * ips_irq_handler - handle temperature triggers and other IPS events
1161  * @irq: irq number
1162  * @arg: unused
1163  *
1164  * Handle temperature limit trigger events, generally by lowering the clamps.
1165  * If we're at a critical limit, we clamp back to the lowest possible value
1166  * to prevent emergency shutdown.
1167  */
1168 static irqreturn_t ips_irq_handler(int irq, void *arg)
1169 {
1170 	struct ips_driver *ips = arg;
1171 	u8 tses = thm_readb(THM_TSES);
1172 	u8 tes = thm_readb(THM_TES);
1173 
1174 	if (!tses && !tes)
1175 		return IRQ_NONE;
1176 
1177 	dev_info(&ips->dev->dev, "TSES: 0x%02x\n", tses);
1178 	dev_info(&ips->dev->dev, "TES: 0x%02x\n", tes);
1179 
1180 	/* STS update from EC? */
1181 	if (tes & 1) {
1182 		u32 sts, tc1;
1183 
1184 		sts = thm_readl(THM_STS);
1185 		tc1 = thm_readl(THM_TC1);
1186 
1187 		if (sts & STS_NVV) {
1188 			spin_lock(&ips->turbo_status_lock);
1189 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1190 				STS_PCPL_SHIFT;
1191 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1192 				STS_GPL_SHIFT;
1193 			/* ignore EC CPU vs GPU pref */
1194 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1195 			/*
1196 			 * Disable turbo for now, until we can figure
1197 			 * out why the power figures are wrong
1198 			 */
1199 			ips->cpu_turbo_enabled = false;
1200 			if (ips->gpu_busy)
1201 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1202 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1203 				STS_PTL_SHIFT;
1204 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1205 				STS_PPL_SHIFT;
1206 			verify_limits(ips);
1207 			spin_unlock(&ips->turbo_status_lock);
1208 
1209 			thm_writeb(THM_SEC, SEC_ACK);
1210 		}
1211 		thm_writeb(THM_TES, tes);
1212 	}
1213 
1214 	/* Thermal trip */
1215 	if (tses) {
1216 		dev_warn(&ips->dev->dev,
1217 			 "thermal trip occurred, tses: 0x%04x\n", tses);
1218 		thm_writeb(THM_TSES, tses);
1219 	}
1220 
1221 	return IRQ_HANDLED;
1222 }
1223 
1224 #ifndef CONFIG_DEBUG_FS
1225 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1226 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1227 #else
1228 
1229 /* Expose current state and limits in debugfs if possible */
1230 
1231 struct ips_debugfs_node {
1232 	struct ips_driver *ips;
1233 	char *name;
1234 	int (*show)(struct seq_file *m, void *data);
1235 };
1236 
1237 static int show_cpu_temp(struct seq_file *m, void *data)
1238 {
1239 	struct ips_driver *ips = m->private;
1240 
1241 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1242 		   ips->ctv1_avg_temp % 100);
1243 
1244 	return 0;
1245 }
1246 
1247 static int show_cpu_power(struct seq_file *m, void *data)
1248 {
1249 	struct ips_driver *ips = m->private;
1250 
1251 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1252 
1253 	return 0;
1254 }
1255 
1256 static int show_cpu_clamp(struct seq_file *m, void *data)
1257 {
1258 	u64 turbo_override;
1259 	int tdp, tdc;
1260 
1261 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1262 
1263 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1264 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1265 
1266 	/* Convert to .1W/A units */
1267 	tdp = tdp * 10 / 8;
1268 	tdc = tdc * 10 / 8;
1269 
1270 	/* Watts Amperes */
1271 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1272 		   tdc / 10, tdc % 10);
1273 
1274 	return 0;
1275 }
1276 
1277 static int show_mch_temp(struct seq_file *m, void *data)
1278 {
1279 	struct ips_driver *ips = m->private;
1280 
1281 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1282 		   ips->mch_avg_temp % 100);
1283 
1284 	return 0;
1285 }
1286 
1287 static int show_mch_power(struct seq_file *m, void *data)
1288 {
1289 	struct ips_driver *ips = m->private;
1290 
1291 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1292 
1293 	return 0;
1294 }
1295 
1296 static struct ips_debugfs_node ips_debug_files[] = {
1297 	{ NULL, "cpu_temp", show_cpu_temp },
1298 	{ NULL, "cpu_power", show_cpu_power },
1299 	{ NULL, "cpu_clamp", show_cpu_clamp },
1300 	{ NULL, "mch_temp", show_mch_temp },
1301 	{ NULL, "mch_power", show_mch_power },
1302 };
1303 
1304 static int ips_debugfs_open(struct inode *inode, struct file *file)
1305 {
1306 	struct ips_debugfs_node *node = inode->i_private;
1307 
1308 	return single_open(file, node->show, node->ips);
1309 }
1310 
1311 static const struct file_operations ips_debugfs_ops = {
1312 	.owner = THIS_MODULE,
1313 	.open = ips_debugfs_open,
1314 	.read = seq_read,
1315 	.llseek = seq_lseek,
1316 	.release = single_release,
1317 };
1318 
1319 static void ips_debugfs_cleanup(struct ips_driver *ips)
1320 {
1321 	if (ips->debug_root)
1322 		debugfs_remove_recursive(ips->debug_root);
1323 	return;
1324 }
1325 
1326 static void ips_debugfs_init(struct ips_driver *ips)
1327 {
1328 	int i;
1329 
1330 	ips->debug_root = debugfs_create_dir("ips", NULL);
1331 	if (!ips->debug_root) {
1332 		dev_err(&ips->dev->dev,
1333 			"failed to create debugfs entries: %ld\n",
1334 			PTR_ERR(ips->debug_root));
1335 		return;
1336 	}
1337 
1338 	for (i = 0; i < ARRAY_SIZE(ips_debug_files); i++) {
1339 		struct dentry *ent;
1340 		struct ips_debugfs_node *node = &ips_debug_files[i];
1341 
1342 		node->ips = ips;
1343 		ent = debugfs_create_file(node->name, S_IFREG | S_IRUGO,
1344 					  ips->debug_root, node,
1345 					  &ips_debugfs_ops);
1346 		if (!ent) {
1347 			dev_err(&ips->dev->dev,
1348 				"failed to create debug file: %ld\n",
1349 				PTR_ERR(ent));
1350 			goto err_cleanup;
1351 		}
1352 	}
1353 
1354 	return;
1355 
1356 err_cleanup:
1357 	ips_debugfs_cleanup(ips);
1358 	return;
1359 }
1360 #endif /* CONFIG_DEBUG_FS */
1361 
1362 /**
1363  * ips_detect_cpu - detect whether CPU supports IPS
1364  *
1365  * Walk our list and see if we're on a supported CPU.  If we find one,
1366  * return the limits for it.
1367  */
1368 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1369 {
1370 	u64 turbo_power, misc_en;
1371 	struct ips_mcp_limits *limits = NULL;
1372 	u16 tdp;
1373 
1374 	if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1375 		dev_info(&ips->dev->dev, "Non-IPS CPU detected.\n");
1376 		goto out;
1377 	}
1378 
1379 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1380 	/*
1381 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1382 	 * turbo manually or we'll get an illegal MSR access, even though
1383 	 * turbo will still be available.
1384 	 */
1385 	if (misc_en & IA32_MISC_TURBO_EN)
1386 		ips->turbo_toggle_allowed = true;
1387 	else
1388 		ips->turbo_toggle_allowed = false;
1389 
1390 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1391 		limits = &ips_sv_limits;
1392 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1393 		limits = &ips_lv_limits;
1394 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1395 		limits = &ips_ulv_limits;
1396 	else {
1397 		dev_info(&ips->dev->dev, "No CPUID match found.\n");
1398 		goto out;
1399 	}
1400 
1401 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1402 	tdp = turbo_power & TURBO_TDP_MASK;
1403 
1404 	/* Sanity check TDP against CPU */
1405 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1406 		dev_info(&ips->dev->dev, "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1407 			 tdp / 8, limits->core_power_limit / 1000);
1408 		limits->core_power_limit = (tdp / 8) * 1000;
1409 	}
1410 
1411 out:
1412 	return limits;
1413 }
1414 
1415 /**
1416  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1417  * @ips: IPS driver
1418  *
1419  * The i915 driver exports several interfaces to allow the IPS driver to
1420  * monitor and control graphics turbo mode.  If we can find them, we can
1421  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1422  * thermal and power limits in the MCP.
1423  */
1424 static bool ips_get_i915_syms(struct ips_driver *ips)
1425 {
1426 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1427 	if (!ips->read_mch_val)
1428 		goto out_err;
1429 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1430 	if (!ips->gpu_raise)
1431 		goto out_put_mch;
1432 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1433 	if (!ips->gpu_lower)
1434 		goto out_put_raise;
1435 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1436 	if (!ips->gpu_busy)
1437 		goto out_put_lower;
1438 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1439 	if (!ips->gpu_turbo_disable)
1440 		goto out_put_busy;
1441 
1442 	return true;
1443 
1444 out_put_busy:
1445 	symbol_put(i915_gpu_busy);
1446 out_put_lower:
1447 	symbol_put(i915_gpu_lower);
1448 out_put_raise:
1449 	symbol_put(i915_gpu_raise);
1450 out_put_mch:
1451 	symbol_put(i915_read_mch_val);
1452 out_err:
1453 	return false;
1454 }
1455 
1456 static bool
1457 ips_gpu_turbo_enabled(struct ips_driver *ips)
1458 {
1459 	if (!ips->gpu_busy && late_i915_load) {
1460 		if (ips_get_i915_syms(ips)) {
1461 			dev_info(&ips->dev->dev,
1462 				 "i915 driver attached, reenabling gpu turbo\n");
1463 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1464 		}
1465 	}
1466 
1467 	return ips->gpu_turbo_enabled;
1468 }
1469 
1470 void
1471 ips_link_to_i915_driver(void)
1472 {
1473 	/* We can't cleanly get at the various ips_driver structs from
1474 	 * this caller (the i915 driver), so just set a flag saying
1475 	 * that it's time to try getting the symbols again.
1476 	 */
1477 	late_i915_load = true;
1478 }
1479 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1480 
1481 static const struct pci_device_id ips_id_table[] = {
1482 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL,
1483 		     PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1484 	{ 0, }
1485 };
1486 
1487 MODULE_DEVICE_TABLE(pci, ips_id_table);
1488 
1489 static int ips_blacklist_callback(const struct dmi_system_id *id)
1490 {
1491 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1492 	return 1;
1493 }
1494 
1495 static const struct dmi_system_id ips_blacklist[] = {
1496 	{
1497 		.callback = ips_blacklist_callback,
1498 		.ident = "HP ProBook",
1499 		.matches = {
1500 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1501 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1502 		},
1503 	},
1504 	{ }	/* terminating entry */
1505 };
1506 
1507 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1508 {
1509 	u64 platform_info;
1510 	struct ips_driver *ips;
1511 	u32 hts;
1512 	int ret = 0;
1513 	u16 htshi, trc, trc_required_mask;
1514 	u8 tse;
1515 
1516 	if (dmi_check_system(ips_blacklist))
1517 		return -ENODEV;
1518 
1519 	ips = kzalloc(sizeof(struct ips_driver), GFP_KERNEL);
1520 	if (!ips)
1521 		return -ENOMEM;
1522 
1523 	pci_set_drvdata(dev, ips);
1524 	ips->dev = dev;
1525 
1526 	ips->limits = ips_detect_cpu(ips);
1527 	if (!ips->limits) {
1528 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1529 		ret = -ENXIO;
1530 		goto error_free;
1531 	}
1532 
1533 	spin_lock_init(&ips->turbo_status_lock);
1534 
1535 	ret = pci_enable_device(dev);
1536 	if (ret) {
1537 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1538 		goto error_free;
1539 	}
1540 
1541 	if (!pci_resource_start(dev, 0)) {
1542 		dev_err(&dev->dev, "TBAR not assigned, aborting\n");
1543 		ret = -ENXIO;
1544 		goto error_free;
1545 	}
1546 
1547 	ret = pci_request_regions(dev, "ips thermal sensor");
1548 	if (ret) {
1549 		dev_err(&dev->dev, "thermal resource busy, aborting\n");
1550 		goto error_free;
1551 	}
1552 
1553 
1554 	ips->regmap = ioremap(pci_resource_start(dev, 0),
1555 			      pci_resource_len(dev, 0));
1556 	if (!ips->regmap) {
1557 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1558 		ret = -EBUSY;
1559 		goto error_release;
1560 	}
1561 
1562 	tse = thm_readb(THM_TSE);
1563 	if (tse != TSE_EN) {
1564 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1565 		ret = -ENXIO;
1566 		goto error_unmap;
1567 	}
1568 
1569 	trc = thm_readw(THM_TRC);
1570 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1571 	if ((trc & trc_required_mask) != trc_required_mask) {
1572 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1573 		ret = -ENXIO;
1574 		goto error_unmap;
1575 	}
1576 
1577 	if (trc & TRC_CORE2_EN)
1578 		ips->second_cpu = true;
1579 
1580 	update_turbo_limits(ips);
1581 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1582 		ips->mcp_power_limit / 10);
1583 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1584 		ips->core_power_limit / 10);
1585 	/* BIOS may update limits at runtime */
1586 	if (thm_readl(THM_PSC) & PSP_PBRT)
1587 		ips->poll_turbo_status = true;
1588 
1589 	if (!ips_get_i915_syms(ips)) {
1590 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1591 		ips->gpu_turbo_enabled = false;
1592 	} else {
1593 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1594 		ips->gpu_turbo_enabled = true;
1595 	}
1596 
1597 	/*
1598 	 * Check PLATFORM_INFO MSR to make sure this chip is
1599 	 * turbo capable.
1600 	 */
1601 	rdmsrl(PLATFORM_INFO, platform_info);
1602 	if (!(platform_info & PLATFORM_TDP)) {
1603 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1604 		ret = -ENODEV;
1605 		goto error_unmap;
1606 	}
1607 
1608 	/*
1609 	 * IRQ handler for ME interaction
1610 	 * Note: don't use MSI here as the PCH has bugs.
1611 	 */
1612 	pci_disable_msi(dev);
1613 	ret = request_irq(dev->irq, ips_irq_handler, IRQF_SHARED, "ips",
1614 			  ips);
1615 	if (ret) {
1616 		dev_err(&dev->dev, "request irq failed, aborting\n");
1617 		goto error_unmap;
1618 	}
1619 
1620 	/* Enable aux, hot & critical interrupts */
1621 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1622 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1623 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1624 
1625 	/* Collect adjustment values */
1626 	ips->cta_val = thm_readw(THM_CTA);
1627 	ips->pta_val = thm_readw(THM_PTA);
1628 	ips->mgta_val = thm_readw(THM_MGTA);
1629 
1630 	/* Save turbo limits & ratios */
1631 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1632 
1633 	ips_disable_cpu_turbo(ips);
1634 	ips->cpu_turbo_enabled = false;
1635 
1636 	/* Create thermal adjust thread */
1637 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1638 	if (IS_ERR(ips->adjust)) {
1639 		dev_err(&dev->dev,
1640 			"failed to create thermal adjust thread, aborting\n");
1641 		ret = -ENOMEM;
1642 		goto error_free_irq;
1643 
1644 	}
1645 
1646 	/*
1647 	 * Set up the work queue and monitor thread. The monitor thread
1648 	 * will wake up ips_adjust thread.
1649 	 */
1650 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1651 	if (IS_ERR(ips->monitor)) {
1652 		dev_err(&dev->dev,
1653 			"failed to create thermal monitor thread, aborting\n");
1654 		ret = -ENOMEM;
1655 		goto error_thread_cleanup;
1656 	}
1657 
1658 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1659 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1660 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1661 
1662 	thm_writew(THM_HTSHI, htshi);
1663 	thm_writel(THM_HTS, hts);
1664 
1665 	ips_debugfs_init(ips);
1666 
1667 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1668 		 ips->mcp_temp_limit);
1669 	return ret;
1670 
1671 error_thread_cleanup:
1672 	kthread_stop(ips->adjust);
1673 error_free_irq:
1674 	free_irq(ips->dev->irq, ips);
1675 error_unmap:
1676 	iounmap(ips->regmap);
1677 error_release:
1678 	pci_release_regions(dev);
1679 error_free:
1680 	kfree(ips);
1681 	return ret;
1682 }
1683 
1684 static void ips_remove(struct pci_dev *dev)
1685 {
1686 	struct ips_driver *ips = pci_get_drvdata(dev);
1687 	u64 turbo_override;
1688 
1689 	if (!ips)
1690 		return;
1691 
1692 	ips_debugfs_cleanup(ips);
1693 
1694 	/* Release i915 driver */
1695 	if (ips->read_mch_val)
1696 		symbol_put(i915_read_mch_val);
1697 	if (ips->gpu_raise)
1698 		symbol_put(i915_gpu_raise);
1699 	if (ips->gpu_lower)
1700 		symbol_put(i915_gpu_lower);
1701 	if (ips->gpu_busy)
1702 		symbol_put(i915_gpu_busy);
1703 	if (ips->gpu_turbo_disable)
1704 		symbol_put(i915_gpu_turbo_disable);
1705 
1706 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1707 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1708 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1709 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1710 
1711 	free_irq(ips->dev->irq, ips);
1712 	if (ips->adjust)
1713 		kthread_stop(ips->adjust);
1714 	if (ips->monitor)
1715 		kthread_stop(ips->monitor);
1716 	iounmap(ips->regmap);
1717 	pci_release_regions(dev);
1718 	kfree(ips);
1719 	dev_dbg(&dev->dev, "IPS driver removed\n");
1720 }
1721 
1722 static void ips_shutdown(struct pci_dev *dev)
1723 {
1724 }
1725 
1726 static struct pci_driver ips_pci_driver = {
1727 	.name = "intel ips",
1728 	.id_table = ips_id_table,
1729 	.probe = ips_probe,
1730 	.remove = ips_remove,
1731 	.shutdown = ips_shutdown,
1732 };
1733 
1734 module_pci_driver(ips_pci_driver);
1735 
1736 MODULE_LICENSE("GPL");
1737 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1738 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1739