xref: /linux/drivers/platform/x86/intel_ips.c (revision 9d64fc08f6fe59a7d71e84f650dd2c0f080254dd)
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  * The full GNU General Public License is included in this distribution in
14  * the file called "COPYING".
15  *
16  * Authors:
17  *	Jesse Barnes <jbarnes@virtuousgeek.org>
18  */
19 
20 /*
21  * Some Intel Ibex Peak based platforms support so-called "intelligent
22  * power sharing", which allows the CPU and GPU to cooperate to maximize
23  * performance within a given TDP (thermal design point).  This driver
24  * performs the coordination between the CPU and GPU, monitors thermal and
25  * power statistics in the platform, and initializes power monitoring
26  * hardware.  It also provides a few tunables to control behavior.  Its
27  * primary purpose is to safely allow CPU and GPU turbo modes to be enabled
28  * by tracking power and thermal budget; secondarily it can boost turbo
29  * performance by allocating more power or thermal budget to the CPU or GPU
30  * based on available headroom and activity.
31  *
32  * The basic algorithm is driven by a 5s moving average of temperature.  If
33  * thermal headroom is available, the CPU and/or GPU power clamps may be
34  * adjusted upwards.  If we hit the thermal ceiling or a thermal trigger,
35  * we scale back the clamp.  Aside from trigger events (when we're critically
36  * close or over our TDP) we don't adjust the clamps more than once every
37  * five seconds.
38  *
39  * The thermal device (device 31, function 6) has a set of registers that
40  * are updated by the ME firmware.  The ME should also take the clamp values
41  * written to those registers and write them to the CPU, but we currently
42  * bypass that functionality and write the CPU MSR directly.
43  *
44  * UNSUPPORTED:
45  *   - dual MCP configs
46  *
47  * TODO:
48  *   - handle CPU hotplug
49  *   - provide turbo enable/disable api
50  *
51  * Related documents:
52  *   - CDI 403777, 403778 - Auburndale EDS vol 1 & 2
53  *   - CDI 401376 - Ibex Peak EDS
54  *   - ref 26037, 26641 - IPS BIOS spec
55  *   - ref 26489 - Nehalem BIOS writer's guide
56  *   - ref 26921 - Ibex Peak BIOS Specification
57  */
58 
59 #include <linux/debugfs.h>
60 #include <linux/delay.h>
61 #include <linux/interrupt.h>
62 #include <linux/kernel.h>
63 #include <linux/kthread.h>
64 #include <linux/module.h>
65 #include <linux/pci.h>
66 #include <linux/sched.h>
67 #include <linux/sched/loadavg.h>
68 #include <linux/seq_file.h>
69 #include <linux/string.h>
70 #include <linux/tick.h>
71 #include <linux/timer.h>
72 #include <linux/dmi.h>
73 #include <drm/i915_drm.h>
74 #include <asm/msr.h>
75 #include <asm/processor.h>
76 #include "intel_ips.h"
77 
78 #include <linux/io-64-nonatomic-lo-hi.h>
79 
80 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
81 
82 /*
83  * Package level MSRs for monitor/control
84  */
85 #define PLATFORM_INFO	0xce
86 #define   PLATFORM_TDP		(1<<29)
87 #define   PLATFORM_RATIO	(1<<28)
88 
89 #define IA32_MISC_ENABLE	0x1a0
90 #define   IA32_MISC_TURBO_EN	(1ULL<<38)
91 
92 #define TURBO_POWER_CURRENT_LIMIT	0x1ac
93 #define   TURBO_TDC_OVR_EN	(1UL<<31)
94 #define   TURBO_TDC_MASK	(0x000000007fff0000UL)
95 #define   TURBO_TDC_SHIFT	(16)
96 #define   TURBO_TDP_OVR_EN	(1UL<<15)
97 #define   TURBO_TDP_MASK	(0x0000000000003fffUL)
98 
99 /*
100  * Core/thread MSRs for monitoring
101  */
102 #define IA32_PERF_CTL		0x199
103 #define   IA32_PERF_TURBO_DIS	(1ULL<<32)
104 
105 /*
106  * Thermal PCI device regs
107  */
108 #define THM_CFG_TBAR	0x10
109 #define THM_CFG_TBAR_HI	0x14
110 
111 #define THM_TSIU	0x00
112 #define THM_TSE		0x01
113 #define   TSE_EN	0xb8
114 #define THM_TSS		0x02
115 #define THM_TSTR	0x03
116 #define THM_TSTTP	0x04
117 #define THM_TSCO	0x08
118 #define THM_TSES	0x0c
119 #define THM_TSGPEN	0x0d
120 #define   TSGPEN_HOT_LOHI	(1<<1)
121 #define   TSGPEN_CRIT_LOHI	(1<<2)
122 #define THM_TSPC	0x0e
123 #define THM_PPEC	0x10
124 #define THM_CTA		0x12
125 #define THM_PTA		0x14
126 #define   PTA_SLOPE_MASK	(0xff00)
127 #define   PTA_SLOPE_SHIFT	8
128 #define   PTA_OFFSET_MASK	(0x00ff)
129 #define THM_MGTA	0x16
130 #define   MGTA_SLOPE_MASK	(0xff00)
131 #define   MGTA_SLOPE_SHIFT	8
132 #define   MGTA_OFFSET_MASK	(0x00ff)
133 #define THM_TRC		0x1a
134 #define   TRC_CORE2_EN	(1<<15)
135 #define   TRC_THM_EN	(1<<12)
136 #define   TRC_C6_WAR	(1<<8)
137 #define   TRC_CORE1_EN	(1<<7)
138 #define   TRC_CORE_PWR	(1<<6)
139 #define   TRC_PCH_EN	(1<<5)
140 #define   TRC_MCH_EN	(1<<4)
141 #define   TRC_DIMM4	(1<<3)
142 #define   TRC_DIMM3	(1<<2)
143 #define   TRC_DIMM2	(1<<1)
144 #define   TRC_DIMM1	(1<<0)
145 #define THM_TES		0x20
146 #define THM_TEN		0x21
147 #define   TEN_UPDATE_EN	1
148 #define THM_PSC		0x24
149 #define   PSC_NTG	(1<<0) /* No GFX turbo support */
150 #define   PSC_NTPC	(1<<1) /* No CPU turbo support */
151 #define   PSC_PP_DEF	(0<<2) /* Perf policy up to driver */
152 #define   PSP_PP_PC	(1<<2) /* BIOS prefers CPU perf */
153 #define   PSP_PP_BAL	(2<<2) /* BIOS wants balanced perf */
154 #define   PSP_PP_GFX	(3<<2) /* BIOS prefers GFX perf */
155 #define   PSP_PBRT	(1<<4) /* BIOS run time support */
156 #define THM_CTV1	0x30
157 #define   CTV_TEMP_ERROR (1<<15)
158 #define   CTV_TEMP_MASK	0x3f
159 #define   CTV_
160 #define THM_CTV2	0x32
161 #define THM_CEC		0x34 /* undocumented power accumulator in joules */
162 #define THM_AE		0x3f
163 #define THM_HTS		0x50 /* 32 bits */
164 #define   HTS_PCPL_MASK	(0x7fe00000)
165 #define   HTS_PCPL_SHIFT 21
166 #define   HTS_GPL_MASK  (0x001ff000)
167 #define   HTS_GPL_SHIFT 12
168 #define   HTS_PP_MASK	(0x00000c00)
169 #define   HTS_PP_SHIFT  10
170 #define   HTS_PP_DEF	0
171 #define   HTS_PP_PROC	1
172 #define   HTS_PP_BAL	2
173 #define   HTS_PP_GFX	3
174 #define   HTS_PCTD_DIS	(1<<9)
175 #define   HTS_GTD_DIS	(1<<8)
176 #define   HTS_PTL_MASK  (0x000000fe)
177 #define   HTS_PTL_SHIFT 1
178 #define   HTS_NVV	(1<<0)
179 #define THM_HTSHI	0x54 /* 16 bits */
180 #define   HTS2_PPL_MASK		(0x03ff)
181 #define   HTS2_PRST_MASK	(0x3c00)
182 #define   HTS2_PRST_SHIFT	10
183 #define   HTS2_PRST_UNLOADED	0
184 #define   HTS2_PRST_RUNNING	1
185 #define   HTS2_PRST_TDISOP	2 /* turbo disabled due to power */
186 #define   HTS2_PRST_TDISHT	3 /* turbo disabled due to high temp */
187 #define   HTS2_PRST_TDISUSR	4 /* user disabled turbo */
188 #define   HTS2_PRST_TDISPLAT	5 /* platform disabled turbo */
189 #define   HTS2_PRST_TDISPM	6 /* power management disabled turbo */
190 #define   HTS2_PRST_TDISERR	7 /* some kind of error disabled turbo */
191 #define THM_PTL		0x56
192 #define THM_MGTV	0x58
193 #define   TV_MASK	0x000000000000ff00
194 #define   TV_SHIFT	8
195 #define THM_PTV		0x60
196 #define   PTV_MASK	0x00ff
197 #define THM_MMGPC	0x64
198 #define THM_MPPC	0x66
199 #define THM_MPCPC	0x68
200 #define THM_TSPIEN	0x82
201 #define   TSPIEN_AUX_LOHI	(1<<0)
202 #define   TSPIEN_HOT_LOHI	(1<<1)
203 #define   TSPIEN_CRIT_LOHI	(1<<2)
204 #define   TSPIEN_AUX2_LOHI	(1<<3)
205 #define THM_TSLOCK	0x83
206 #define THM_ATR		0x84
207 #define THM_TOF		0x87
208 #define THM_STS		0x98
209 #define   STS_PCPL_MASK		(0x7fe00000)
210 #define   STS_PCPL_SHIFT	21
211 #define   STS_GPL_MASK		(0x001ff000)
212 #define   STS_GPL_SHIFT		12
213 #define   STS_PP_MASK		(0x00000c00)
214 #define   STS_PP_SHIFT		10
215 #define   STS_PP_DEF		0
216 #define   STS_PP_PROC		1
217 #define   STS_PP_BAL		2
218 #define   STS_PP_GFX		3
219 #define   STS_PCTD_DIS		(1<<9)
220 #define   STS_GTD_DIS		(1<<8)
221 #define   STS_PTL_MASK		(0x000000fe)
222 #define   STS_PTL_SHIFT		1
223 #define   STS_NVV		(1<<0)
224 #define THM_SEC		0x9c
225 #define   SEC_ACK	(1<<0)
226 #define THM_TC3		0xa4
227 #define THM_TC1		0xa8
228 #define   STS_PPL_MASK		(0x0003ff00)
229 #define   STS_PPL_SHIFT		16
230 #define THM_TC2		0xac
231 #define THM_DTV		0xb0
232 #define THM_ITV		0xd8
233 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
234 #define   ITV_ME_SEQNO_SHIFT (16)
235 #define   ITV_MCH_TEMP_MASK 0x0000ff00
236 #define   ITV_MCH_TEMP_SHIFT (8)
237 #define   ITV_PCH_TEMP_MASK 0x000000ff
238 
239 #define thm_readb(off) readb(ips->regmap + (off))
240 #define thm_readw(off) readw(ips->regmap + (off))
241 #define thm_readl(off) readl(ips->regmap + (off))
242 #define thm_readq(off) readq(ips->regmap + (off))
243 
244 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
245 #define thm_writew(off, val) writew((val), ips->regmap + (off))
246 #define thm_writel(off, val) writel((val), ips->regmap + (off))
247 
248 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
249 static bool late_i915_load = false;
250 
251 /* For initial average collection */
252 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
253 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
254 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
255 
256 /* Per-SKU limits */
257 struct ips_mcp_limits {
258 	int mcp_power_limit; /* mW units */
259 	int core_power_limit;
260 	int mch_power_limit;
261 	int core_temp_limit; /* degrees C */
262 	int mch_temp_limit;
263 };
264 
265 /* Max temps are -10 degrees C to avoid PROCHOT# */
266 
267 static struct ips_mcp_limits ips_sv_limits = {
268 	.mcp_power_limit = 35000,
269 	.core_power_limit = 29000,
270 	.mch_power_limit = 20000,
271 	.core_temp_limit = 95,
272 	.mch_temp_limit = 90
273 };
274 
275 static struct ips_mcp_limits ips_lv_limits = {
276 	.mcp_power_limit = 25000,
277 	.core_power_limit = 21000,
278 	.mch_power_limit = 13000,
279 	.core_temp_limit = 95,
280 	.mch_temp_limit = 90
281 };
282 
283 static struct ips_mcp_limits ips_ulv_limits = {
284 	.mcp_power_limit = 18000,
285 	.core_power_limit = 14000,
286 	.mch_power_limit = 11000,
287 	.core_temp_limit = 95,
288 	.mch_temp_limit = 90
289 };
290 
291 struct ips_driver {
292 	struct device *dev;
293 	void __iomem *regmap;
294 	int irq;
295 
296 	struct task_struct *monitor;
297 	struct task_struct *adjust;
298 	struct dentry *debug_root;
299 
300 	/* Average CPU core temps (all averages in .01 degrees C for precision) */
301 	u16 ctv1_avg_temp;
302 	u16 ctv2_avg_temp;
303 	/* GMCH average */
304 	u16 mch_avg_temp;
305 	/* Average for the CPU (both cores?) */
306 	u16 mcp_avg_temp;
307 	/* Average power consumption (in mW) */
308 	u32 cpu_avg_power;
309 	u32 mch_avg_power;
310 
311 	/* Offset values */
312 	u16 cta_val;
313 	u16 pta_val;
314 	u16 mgta_val;
315 
316 	/* Maximums & prefs, protected by turbo status lock */
317 	spinlock_t turbo_status_lock;
318 	u16 mcp_temp_limit;
319 	u16 mcp_power_limit;
320 	u16 core_power_limit;
321 	u16 mch_power_limit;
322 	bool cpu_turbo_enabled;
323 	bool __cpu_turbo_on;
324 	bool gpu_turbo_enabled;
325 	bool __gpu_turbo_on;
326 	bool gpu_preferred;
327 	bool poll_turbo_status;
328 	bool second_cpu;
329 	bool turbo_toggle_allowed;
330 	struct ips_mcp_limits *limits;
331 
332 	/* Optional MCH interfaces for if i915 is in use */
333 	unsigned long (*read_mch_val)(void);
334 	bool (*gpu_raise)(void);
335 	bool (*gpu_lower)(void);
336 	bool (*gpu_busy)(void);
337 	bool (*gpu_turbo_disable)(void);
338 
339 	/* For restoration at unload */
340 	u64 orig_turbo_limit;
341 	u64 orig_turbo_ratios;
342 };
343 
344 static bool
345 ips_gpu_turbo_enabled(struct ips_driver *ips);
346 
347 /**
348  * ips_cpu_busy - is CPU busy?
349  * @ips: IPS driver struct
350  *
351  * Check CPU for load to see whether we should increase its thermal budget.
352  *
353  * RETURNS:
354  * True if the CPU could use more power, false otherwise.
355  */
356 static bool ips_cpu_busy(struct ips_driver *ips)
357 {
358 	if ((avenrun[0] >> FSHIFT) > 1)
359 		return true;
360 
361 	return false;
362 }
363 
364 /**
365  * ips_cpu_raise - raise CPU power clamp
366  * @ips: IPS driver struct
367  *
368  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
369  * this platform.
370  *
371  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
372  * long as we haven't hit the TDP limit for the SKU).
373  */
374 static void ips_cpu_raise(struct ips_driver *ips)
375 {
376 	u64 turbo_override;
377 	u16 cur_tdp_limit, new_tdp_limit;
378 
379 	if (!ips->cpu_turbo_enabled)
380 		return;
381 
382 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
383 
384 	cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
385 	new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
386 
387 	/* Clamp to SKU TDP limit */
388 	if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
389 		new_tdp_limit = cur_tdp_limit;
390 
391 	thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
392 
393 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
394 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
395 
396 	turbo_override &= ~TURBO_TDP_MASK;
397 	turbo_override |= new_tdp_limit;
398 
399 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
400 }
401 
402 /**
403  * ips_cpu_lower - lower CPU power clamp
404  * @ips: IPS driver struct
405  *
406  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
407  *
408  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
409  * as low as the platform limits will allow (though we could go lower there
410  * wouldn't be much point).
411  */
412 static void ips_cpu_lower(struct ips_driver *ips)
413 {
414 	u64 turbo_override;
415 	u16 cur_limit, new_limit;
416 
417 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
418 
419 	cur_limit = turbo_override & TURBO_TDP_MASK;
420 	new_limit = cur_limit - 8; /* 1W decrease */
421 
422 	/* Clamp to SKU TDP limit */
423 	if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
424 		new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
425 
426 	thm_writew(THM_MPCPC, (new_limit * 10) / 8);
427 
428 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
429 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
430 
431 	turbo_override &= ~TURBO_TDP_MASK;
432 	turbo_override |= new_limit;
433 
434 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
435 }
436 
437 /**
438  * do_enable_cpu_turbo - internal turbo enable function
439  * @data: unused
440  *
441  * Internal function for actually updating MSRs.  When we enable/disable
442  * turbo, we need to do it on each CPU; this function is the one called
443  * by on_each_cpu() when needed.
444  */
445 static void do_enable_cpu_turbo(void *data)
446 {
447 	u64 perf_ctl;
448 
449 	rdmsrl(IA32_PERF_CTL, perf_ctl);
450 	if (perf_ctl & IA32_PERF_TURBO_DIS) {
451 		perf_ctl &= ~IA32_PERF_TURBO_DIS;
452 		wrmsrl(IA32_PERF_CTL, perf_ctl);
453 	}
454 }
455 
456 /**
457  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
458  * @ips: IPS driver struct
459  *
460  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
461  * all logical threads.
462  */
463 static void ips_enable_cpu_turbo(struct ips_driver *ips)
464 {
465 	/* Already on, no need to mess with MSRs */
466 	if (ips->__cpu_turbo_on)
467 		return;
468 
469 	if (ips->turbo_toggle_allowed)
470 		on_each_cpu(do_enable_cpu_turbo, ips, 1);
471 
472 	ips->__cpu_turbo_on = true;
473 }
474 
475 /**
476  * do_disable_cpu_turbo - internal turbo disable function
477  * @data: unused
478  *
479  * Internal function for actually updating MSRs.  When we enable/disable
480  * turbo, we need to do it on each CPU; this function is the one called
481  * by on_each_cpu() when needed.
482  */
483 static void do_disable_cpu_turbo(void *data)
484 {
485 	u64 perf_ctl;
486 
487 	rdmsrl(IA32_PERF_CTL, perf_ctl);
488 	if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
489 		perf_ctl |= IA32_PERF_TURBO_DIS;
490 		wrmsrl(IA32_PERF_CTL, perf_ctl);
491 	}
492 }
493 
494 /**
495  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
496  * @ips: IPS driver struct
497  *
498  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
499  * all logical threads.
500  */
501 static void ips_disable_cpu_turbo(struct ips_driver *ips)
502 {
503 	/* Already off, leave it */
504 	if (!ips->__cpu_turbo_on)
505 		return;
506 
507 	if (ips->turbo_toggle_allowed)
508 		on_each_cpu(do_disable_cpu_turbo, ips, 1);
509 
510 	ips->__cpu_turbo_on = false;
511 }
512 
513 /**
514  * ips_gpu_busy - is GPU busy?
515  * @ips: IPS driver struct
516  *
517  * Check GPU for load to see whether we should increase its thermal budget.
518  * We need to call into the i915 driver in this case.
519  *
520  * RETURNS:
521  * True if the GPU could use more power, false otherwise.
522  */
523 static bool ips_gpu_busy(struct ips_driver *ips)
524 {
525 	if (!ips_gpu_turbo_enabled(ips))
526 		return false;
527 
528 	return ips->gpu_busy();
529 }
530 
531 /**
532  * ips_gpu_raise - raise GPU power clamp
533  * @ips: IPS driver struct
534  *
535  * Raise the GPU frequency/power if possible.  We need to call into the
536  * i915 driver in this case.
537  */
538 static void ips_gpu_raise(struct ips_driver *ips)
539 {
540 	if (!ips_gpu_turbo_enabled(ips))
541 		return;
542 
543 	if (!ips->gpu_raise())
544 		ips->gpu_turbo_enabled = false;
545 
546 	return;
547 }
548 
549 /**
550  * ips_gpu_lower - lower GPU power clamp
551  * @ips: IPS driver struct
552  *
553  * Lower GPU frequency/power if possible.  Need to call i915.
554  */
555 static void ips_gpu_lower(struct ips_driver *ips)
556 {
557 	if (!ips_gpu_turbo_enabled(ips))
558 		return;
559 
560 	if (!ips->gpu_lower())
561 		ips->gpu_turbo_enabled = false;
562 
563 	return;
564 }
565 
566 /**
567  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
568  * @ips: IPS driver struct
569  *
570  * Call into the graphics driver indicating that it can safely use
571  * turbo mode.
572  */
573 static void ips_enable_gpu_turbo(struct ips_driver *ips)
574 {
575 	if (ips->__gpu_turbo_on)
576 		return;
577 	ips->__gpu_turbo_on = true;
578 }
579 
580 /**
581  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
582  * @ips: IPS driver struct
583  *
584  * Request that the graphics driver disable turbo mode.
585  */
586 static void ips_disable_gpu_turbo(struct ips_driver *ips)
587 {
588 	/* Avoid calling i915 if turbo is already disabled */
589 	if (!ips->__gpu_turbo_on)
590 		return;
591 
592 	if (!ips->gpu_turbo_disable())
593 		dev_err(ips->dev, "failed to disable graphics turbo\n");
594 	else
595 		ips->__gpu_turbo_on = false;
596 }
597 
598 /**
599  * mcp_exceeded - check whether we're outside our thermal & power limits
600  * @ips: IPS driver struct
601  *
602  * Check whether the MCP is over its thermal or power budget.
603  */
604 static bool mcp_exceeded(struct ips_driver *ips)
605 {
606 	unsigned long flags;
607 	bool ret = false;
608 	u32 temp_limit;
609 	u32 avg_power;
610 
611 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
612 
613 	temp_limit = ips->mcp_temp_limit * 100;
614 	if (ips->mcp_avg_temp > temp_limit)
615 		ret = true;
616 
617 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
618 	if (avg_power > ips->mcp_power_limit)
619 		ret = true;
620 
621 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
622 
623 	return ret;
624 }
625 
626 /**
627  * cpu_exceeded - check whether a CPU core is outside its limits
628  * @ips: IPS driver struct
629  * @cpu: CPU number to check
630  *
631  * Check a given CPU's average temp or power is over its limit.
632  */
633 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
634 {
635 	unsigned long flags;
636 	int avg;
637 	bool ret = false;
638 
639 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
640 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
641 	if (avg > (ips->limits->core_temp_limit * 100))
642 		ret = true;
643 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
644 		ret = true;
645 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
646 
647 	if (ret)
648 		dev_info(ips->dev, "CPU power or thermal limit exceeded\n");
649 
650 	return ret;
651 }
652 
653 /**
654  * mch_exceeded - check whether the GPU is over budget
655  * @ips: IPS driver struct
656  *
657  * Check the MCH temp & power against their maximums.
658  */
659 static bool mch_exceeded(struct ips_driver *ips)
660 {
661 	unsigned long flags;
662 	bool ret = false;
663 
664 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
665 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
666 		ret = true;
667 	if (ips->mch_avg_power > ips->mch_power_limit)
668 		ret = true;
669 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
670 
671 	return ret;
672 }
673 
674 /**
675  * verify_limits - verify BIOS provided limits
676  * @ips: IPS structure
677  *
678  * BIOS can optionally provide non-default limits for power and temp.  Check
679  * them here and use the defaults if the BIOS values are not provided or
680  * are otherwise unusable.
681  */
682 static void verify_limits(struct ips_driver *ips)
683 {
684 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
685 	    ips->mcp_power_limit > 35000)
686 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
687 
688 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
689 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
690 	    ips->mcp_temp_limit > 150)
691 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
692 					  ips->limits->mch_temp_limit);
693 }
694 
695 /**
696  * update_turbo_limits - get various limits & settings from regs
697  * @ips: IPS driver struct
698  *
699  * Update the IPS power & temp limits, along with turbo enable flags,
700  * based on latest register contents.
701  *
702  * Used at init time and for runtime BIOS support, which requires polling
703  * the regs for updates (as a result of AC->DC transition for example).
704  *
705  * LOCKING:
706  * Caller must hold turbo_status_lock (outside of init)
707  */
708 static void update_turbo_limits(struct ips_driver *ips)
709 {
710 	u32 hts = thm_readl(THM_HTS);
711 
712 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
713 	/*
714 	 * Disable turbo for now, until we can figure out why the power figures
715 	 * are wrong
716 	 */
717 	ips->cpu_turbo_enabled = false;
718 
719 	if (ips->gpu_busy)
720 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
721 
722 	ips->core_power_limit = thm_readw(THM_MPCPC);
723 	ips->mch_power_limit = thm_readw(THM_MMGPC);
724 	ips->mcp_temp_limit = thm_readw(THM_PTL);
725 	ips->mcp_power_limit = thm_readw(THM_MPPC);
726 
727 	verify_limits(ips);
728 	/* Ignore BIOS CPU vs GPU pref */
729 }
730 
731 /**
732  * ips_adjust - adjust power clamp based on thermal state
733  * @data: ips driver structure
734  *
735  * Wake up every 5s or so and check whether we should adjust the power clamp.
736  * Check CPU and GPU load to determine which needs adjustment.  There are
737  * several things to consider here:
738  *   - do we need to adjust up or down?
739  *   - is CPU busy?
740  *   - is GPU busy?
741  *   - is CPU in turbo?
742  *   - is GPU in turbo?
743  *   - is CPU or GPU preferred? (CPU is default)
744  *
745  * So, given the above, we do the following:
746  *   - up (TDP available)
747  *     - CPU not busy, GPU not busy - nothing
748  *     - CPU busy, GPU not busy - adjust CPU up
749  *     - CPU not busy, GPU busy - adjust GPU up
750  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
751  *       non-preferred unit if necessary
752  *   - down (at TDP limit)
753  *     - adjust both CPU and GPU down if possible
754  *
755 		cpu+ gpu+	cpu+gpu-	cpu-gpu+	cpu-gpu-
756 cpu < gpu <	cpu+gpu+	cpu+		gpu+		nothing
757 cpu < gpu >=	cpu+gpu-(mcp<)	cpu+gpu-(mcp<)	gpu-		gpu-
758 cpu >= gpu <	cpu-gpu+(mcp<)	cpu-		cpu-gpu+(mcp<)	cpu-
759 cpu >= gpu >=	cpu-gpu-	cpu-gpu-	cpu-gpu-	cpu-gpu-
760  *
761  */
762 static int ips_adjust(void *data)
763 {
764 	struct ips_driver *ips = data;
765 	unsigned long flags;
766 
767 	dev_dbg(ips->dev, "starting ips-adjust thread\n");
768 
769 	/*
770 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
771 	 * often isn't recommended due to ME interaction.
772 	 */
773 	do {
774 		bool cpu_busy = ips_cpu_busy(ips);
775 		bool gpu_busy = ips_gpu_busy(ips);
776 
777 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
778 		if (ips->poll_turbo_status)
779 			update_turbo_limits(ips);
780 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
781 
782 		/* Update turbo status if necessary */
783 		if (ips->cpu_turbo_enabled)
784 			ips_enable_cpu_turbo(ips);
785 		else
786 			ips_disable_cpu_turbo(ips);
787 
788 		if (ips->gpu_turbo_enabled)
789 			ips_enable_gpu_turbo(ips);
790 		else
791 			ips_disable_gpu_turbo(ips);
792 
793 		/* We're outside our comfort zone, crank them down */
794 		if (mcp_exceeded(ips)) {
795 			ips_cpu_lower(ips);
796 			ips_gpu_lower(ips);
797 			goto sleep;
798 		}
799 
800 		if (!cpu_exceeded(ips, 0) && cpu_busy)
801 			ips_cpu_raise(ips);
802 		else
803 			ips_cpu_lower(ips);
804 
805 		if (!mch_exceeded(ips) && gpu_busy)
806 			ips_gpu_raise(ips);
807 		else
808 			ips_gpu_lower(ips);
809 
810 sleep:
811 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
812 	} while (!kthread_should_stop());
813 
814 	dev_dbg(ips->dev, "ips-adjust thread stopped\n");
815 
816 	return 0;
817 }
818 
819 /*
820  * Helpers for reading out temp/power values and calculating their
821  * averages for the decision making and monitoring functions.
822  */
823 
824 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
825 {
826 	u64 total = 0;
827 	int i;
828 	u16 avg;
829 
830 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
831 		total += (u64)(array[i] * 100);
832 
833 	do_div(total, IPS_SAMPLE_COUNT);
834 
835 	avg = (u16)total;
836 
837 	return avg;
838 }
839 
840 static u16 read_mgtv(struct ips_driver *ips)
841 {
842 	u16 ret;
843 	u64 slope, offset;
844 	u64 val;
845 
846 	val = thm_readq(THM_MGTV);
847 	val = (val & TV_MASK) >> TV_SHIFT;
848 
849 	slope = offset = thm_readw(THM_MGTA);
850 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
851 	offset = offset & MGTA_OFFSET_MASK;
852 
853 	ret = ((val * slope + 0x40) >> 7) + offset;
854 
855 	return 0; /* MCH temp reporting buggy */
856 }
857 
858 static u16 read_ptv(struct ips_driver *ips)
859 {
860 	u16 val, slope, offset;
861 
862 	slope = (ips->pta_val & PTA_SLOPE_MASK) >> PTA_SLOPE_SHIFT;
863 	offset = ips->pta_val & PTA_OFFSET_MASK;
864 
865 	val = thm_readw(THM_PTV) & PTV_MASK;
866 
867 	return val;
868 }
869 
870 static u16 read_ctv(struct ips_driver *ips, int cpu)
871 {
872 	int reg = cpu ? THM_CTV2 : THM_CTV1;
873 	u16 val;
874 
875 	val = thm_readw(reg);
876 	if (!(val & CTV_TEMP_ERROR))
877 		val = (val) >> 6; /* discard fractional component */
878 	else
879 		val = 0;
880 
881 	return val;
882 }
883 
884 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
885 {
886 	u32 val;
887 	u32 ret;
888 
889 	/*
890 	 * CEC is in joules/65535.  Take difference over time to
891 	 * get watts.
892 	 */
893 	val = thm_readl(THM_CEC);
894 
895 	/* period is in ms and we want mW */
896 	ret = (((val - *last) * 1000) / period);
897 	ret = (ret * 1000) / 65535;
898 	*last = val;
899 
900 	return 0;
901 }
902 
903 static const u16 temp_decay_factor = 2;
904 static u16 update_average_temp(u16 avg, u16 val)
905 {
906 	u16 ret;
907 
908 	/* Multiply by 100 for extra precision */
909 	ret = (val * 100 / temp_decay_factor) +
910 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
911 	return ret;
912 }
913 
914 static const u16 power_decay_factor = 2;
915 static u16 update_average_power(u32 avg, u32 val)
916 {
917 	u32 ret;
918 
919 	ret = (val / power_decay_factor) +
920 		(((power_decay_factor - 1) * avg) / power_decay_factor);
921 
922 	return ret;
923 }
924 
925 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
926 {
927 	u64 total = 0;
928 	u32 avg;
929 	int i;
930 
931 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
932 		total += array[i];
933 
934 	do_div(total, IPS_SAMPLE_COUNT);
935 	avg = (u32)total;
936 
937 	return avg;
938 }
939 
940 static void monitor_timeout(unsigned long arg)
941 {
942 	wake_up_process((struct task_struct *)arg);
943 }
944 
945 /**
946  * ips_monitor - temp/power monitoring thread
947  * @data: ips driver structure
948  *
949  * This is the main function for the IPS driver.  It monitors power and
950  * tempurature in the MCP and adjusts CPU and GPU power clams accordingly.
951  *
952  * We keep a 5s moving average of power consumption and tempurature.  Using
953  * that data, along with CPU vs GPU preference, we adjust the power clamps
954  * up or down.
955  */
956 static int ips_monitor(void *data)
957 {
958 	struct ips_driver *ips = data;
959 	struct timer_list timer;
960 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
961 	int i;
962 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
963 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
964 	u8 cur_seqno, last_seqno;
965 
966 	mcp_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
967 	ctv1_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
968 	ctv2_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
969 	mch_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
970 	cpu_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
971 	mchp_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
972 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
973 			!cpu_samples || !mchp_samples) {
974 		dev_err(ips->dev,
975 			"failed to allocate sample array, ips disabled\n");
976 		kfree(mcp_samples);
977 		kfree(ctv1_samples);
978 		kfree(ctv2_samples);
979 		kfree(mch_samples);
980 		kfree(cpu_samples);
981 		kfree(mchp_samples);
982 		return -ENOMEM;
983 	}
984 
985 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
986 		ITV_ME_SEQNO_SHIFT;
987 	seqno_timestamp = get_jiffies_64();
988 
989 	old_cpu_power = thm_readl(THM_CEC);
990 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
991 
992 	/* Collect an initial average */
993 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
994 		u32 mchp, cpu_power;
995 		u16 val;
996 
997 		mcp_samples[i] = read_ptv(ips);
998 
999 		val = read_ctv(ips, 0);
1000 		ctv1_samples[i] = val;
1001 
1002 		val = read_ctv(ips, 1);
1003 		ctv2_samples[i] = val;
1004 
1005 		val = read_mgtv(ips);
1006 		mch_samples[i] = val;
1007 
1008 		cpu_power = get_cpu_power(ips, &old_cpu_power,
1009 					  IPS_SAMPLE_PERIOD);
1010 		cpu_samples[i] = cpu_power;
1011 
1012 		if (ips->read_mch_val) {
1013 			mchp = ips->read_mch_val();
1014 			mchp_samples[i] = mchp;
1015 		}
1016 
1017 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1018 		if (kthread_should_stop())
1019 			break;
1020 	}
1021 
1022 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1023 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1024 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1025 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1026 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1027 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1028 	kfree(mcp_samples);
1029 	kfree(ctv1_samples);
1030 	kfree(ctv2_samples);
1031 	kfree(mch_samples);
1032 	kfree(cpu_samples);
1033 	kfree(mchp_samples);
1034 
1035 	/* Start the adjustment thread now that we have data */
1036 	wake_up_process(ips->adjust);
1037 
1038 	/*
1039 	 * Ok, now we have an initial avg.  From here on out, we track the
1040 	 * running avg using a decaying average calculation.  This allows
1041 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1042 	 */
1043 	old_cpu_power = thm_readl(THM_CEC);
1044 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1045 	last_sample_period = IPS_SAMPLE_PERIOD;
1046 
1047 	setup_deferrable_timer_on_stack(&timer, monitor_timeout,
1048 					(unsigned long)current);
1049 	do {
1050 		u32 cpu_val, mch_val;
1051 		u16 val;
1052 
1053 		/* MCP itself */
1054 		val = read_ptv(ips);
1055 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1056 
1057 		/* Processor 0 */
1058 		val = read_ctv(ips, 0);
1059 		ips->ctv1_avg_temp =
1060 			update_average_temp(ips->ctv1_avg_temp, val);
1061 		/* Power */
1062 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1063 					last_sample_period);
1064 		ips->cpu_avg_power =
1065 			update_average_power(ips->cpu_avg_power, cpu_val);
1066 
1067 		if (ips->second_cpu) {
1068 			/* Processor 1 */
1069 			val = read_ctv(ips, 1);
1070 			ips->ctv2_avg_temp =
1071 				update_average_temp(ips->ctv2_avg_temp, val);
1072 		}
1073 
1074 		/* MCH */
1075 		val = read_mgtv(ips);
1076 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1077 		/* Power */
1078 		if (ips->read_mch_val) {
1079 			mch_val = ips->read_mch_val();
1080 			ips->mch_avg_power =
1081 				update_average_power(ips->mch_avg_power,
1082 						     mch_val);
1083 		}
1084 
1085 		/*
1086 		 * Make sure ME is updating thermal regs.
1087 		 * Note:
1088 		 * If it's been more than a second since the last update,
1089 		 * the ME is probably hung.
1090 		 */
1091 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1092 			ITV_ME_SEQNO_SHIFT;
1093 		if (cur_seqno == last_seqno &&
1094 		    time_after(jiffies, seqno_timestamp + HZ)) {
1095 			dev_warn(ips->dev,
1096 				 "ME failed to update for more than 1s, likely hung\n");
1097 		} else {
1098 			seqno_timestamp = get_jiffies_64();
1099 			last_seqno = cur_seqno;
1100 		}
1101 
1102 		last_msecs = jiffies_to_msecs(jiffies);
1103 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1104 
1105 		__set_current_state(TASK_INTERRUPTIBLE);
1106 		mod_timer(&timer, expire);
1107 		schedule();
1108 
1109 		/* Calculate actual sample period for power averaging */
1110 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1111 		if (!last_sample_period)
1112 			last_sample_period = 1;
1113 	} while (!kthread_should_stop());
1114 
1115 	del_timer_sync(&timer);
1116 	destroy_timer_on_stack(&timer);
1117 
1118 	dev_dbg(ips->dev, "ips-monitor thread stopped\n");
1119 
1120 	return 0;
1121 }
1122 
1123 #if 0
1124 #define THM_DUMPW(reg) \
1125 	{ \
1126 	u16 val = thm_readw(reg); \
1127 	dev_dbg(ips->dev, #reg ": 0x%04x\n", val); \
1128 	}
1129 #define THM_DUMPL(reg) \
1130 	{ \
1131 	u32 val = thm_readl(reg); \
1132 	dev_dbg(ips->dev, #reg ": 0x%08x\n", val); \
1133 	}
1134 #define THM_DUMPQ(reg) \
1135 	{ \
1136 	u64 val = thm_readq(reg); \
1137 	dev_dbg(ips->dev, #reg ": 0x%016x\n", val); \
1138 	}
1139 
1140 static void dump_thermal_info(struct ips_driver *ips)
1141 {
1142 	u16 ptl;
1143 
1144 	ptl = thm_readw(THM_PTL);
1145 	dev_dbg(ips->dev, "Processor temp limit: %d\n", ptl);
1146 
1147 	THM_DUMPW(THM_CTA);
1148 	THM_DUMPW(THM_TRC);
1149 	THM_DUMPW(THM_CTV1);
1150 	THM_DUMPL(THM_STS);
1151 	THM_DUMPW(THM_PTV);
1152 	THM_DUMPQ(THM_MGTV);
1153 }
1154 #endif
1155 
1156 /**
1157  * ips_irq_handler - handle temperature triggers and other IPS events
1158  * @irq: irq number
1159  * @arg: unused
1160  *
1161  * Handle temperature limit trigger events, generally by lowering the clamps.
1162  * If we're at a critical limit, we clamp back to the lowest possible value
1163  * to prevent emergency shutdown.
1164  */
1165 static irqreturn_t ips_irq_handler(int irq, void *arg)
1166 {
1167 	struct ips_driver *ips = arg;
1168 	u8 tses = thm_readb(THM_TSES);
1169 	u8 tes = thm_readb(THM_TES);
1170 
1171 	if (!tses && !tes)
1172 		return IRQ_NONE;
1173 
1174 	dev_info(ips->dev, "TSES: 0x%02x\n", tses);
1175 	dev_info(ips->dev, "TES: 0x%02x\n", tes);
1176 
1177 	/* STS update from EC? */
1178 	if (tes & 1) {
1179 		u32 sts, tc1;
1180 
1181 		sts = thm_readl(THM_STS);
1182 		tc1 = thm_readl(THM_TC1);
1183 
1184 		if (sts & STS_NVV) {
1185 			spin_lock(&ips->turbo_status_lock);
1186 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1187 				STS_PCPL_SHIFT;
1188 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1189 				STS_GPL_SHIFT;
1190 			/* ignore EC CPU vs GPU pref */
1191 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1192 			/*
1193 			 * Disable turbo for now, until we can figure
1194 			 * out why the power figures are wrong
1195 			 */
1196 			ips->cpu_turbo_enabled = false;
1197 			if (ips->gpu_busy)
1198 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1199 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1200 				STS_PTL_SHIFT;
1201 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1202 				STS_PPL_SHIFT;
1203 			verify_limits(ips);
1204 			spin_unlock(&ips->turbo_status_lock);
1205 
1206 			thm_writeb(THM_SEC, SEC_ACK);
1207 		}
1208 		thm_writeb(THM_TES, tes);
1209 	}
1210 
1211 	/* Thermal trip */
1212 	if (tses) {
1213 		dev_warn(ips->dev, "thermal trip occurred, tses: 0x%04x\n",
1214 			 tses);
1215 		thm_writeb(THM_TSES, tses);
1216 	}
1217 
1218 	return IRQ_HANDLED;
1219 }
1220 
1221 #ifndef CONFIG_DEBUG_FS
1222 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1223 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1224 #else
1225 
1226 /* Expose current state and limits in debugfs if possible */
1227 
1228 struct ips_debugfs_node {
1229 	struct ips_driver *ips;
1230 	char *name;
1231 	int (*show)(struct seq_file *m, void *data);
1232 };
1233 
1234 static int show_cpu_temp(struct seq_file *m, void *data)
1235 {
1236 	struct ips_driver *ips = m->private;
1237 
1238 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1239 		   ips->ctv1_avg_temp % 100);
1240 
1241 	return 0;
1242 }
1243 
1244 static int show_cpu_power(struct seq_file *m, void *data)
1245 {
1246 	struct ips_driver *ips = m->private;
1247 
1248 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1249 
1250 	return 0;
1251 }
1252 
1253 static int show_cpu_clamp(struct seq_file *m, void *data)
1254 {
1255 	u64 turbo_override;
1256 	int tdp, tdc;
1257 
1258 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1259 
1260 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1261 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1262 
1263 	/* Convert to .1W/A units */
1264 	tdp = tdp * 10 / 8;
1265 	tdc = tdc * 10 / 8;
1266 
1267 	/* Watts Amperes */
1268 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1269 		   tdc / 10, tdc % 10);
1270 
1271 	return 0;
1272 }
1273 
1274 static int show_mch_temp(struct seq_file *m, void *data)
1275 {
1276 	struct ips_driver *ips = m->private;
1277 
1278 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1279 		   ips->mch_avg_temp % 100);
1280 
1281 	return 0;
1282 }
1283 
1284 static int show_mch_power(struct seq_file *m, void *data)
1285 {
1286 	struct ips_driver *ips = m->private;
1287 
1288 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1289 
1290 	return 0;
1291 }
1292 
1293 static struct ips_debugfs_node ips_debug_files[] = {
1294 	{ NULL, "cpu_temp", show_cpu_temp },
1295 	{ NULL, "cpu_power", show_cpu_power },
1296 	{ NULL, "cpu_clamp", show_cpu_clamp },
1297 	{ NULL, "mch_temp", show_mch_temp },
1298 	{ NULL, "mch_power", show_mch_power },
1299 };
1300 
1301 static int ips_debugfs_open(struct inode *inode, struct file *file)
1302 {
1303 	struct ips_debugfs_node *node = inode->i_private;
1304 
1305 	return single_open(file, node->show, node->ips);
1306 }
1307 
1308 static const struct file_operations ips_debugfs_ops = {
1309 	.owner = THIS_MODULE,
1310 	.open = ips_debugfs_open,
1311 	.read = seq_read,
1312 	.llseek = seq_lseek,
1313 	.release = single_release,
1314 };
1315 
1316 static void ips_debugfs_cleanup(struct ips_driver *ips)
1317 {
1318 	if (ips->debug_root)
1319 		debugfs_remove_recursive(ips->debug_root);
1320 	return;
1321 }
1322 
1323 static void ips_debugfs_init(struct ips_driver *ips)
1324 {
1325 	int i;
1326 
1327 	ips->debug_root = debugfs_create_dir("ips", NULL);
1328 	if (!ips->debug_root) {
1329 		dev_err(ips->dev, "failed to create debugfs entries: %ld\n",
1330 			PTR_ERR(ips->debug_root));
1331 		return;
1332 	}
1333 
1334 	for (i = 0; i < ARRAY_SIZE(ips_debug_files); i++) {
1335 		struct dentry *ent;
1336 		struct ips_debugfs_node *node = &ips_debug_files[i];
1337 
1338 		node->ips = ips;
1339 		ent = debugfs_create_file(node->name, S_IFREG | S_IRUGO,
1340 					  ips->debug_root, node,
1341 					  &ips_debugfs_ops);
1342 		if (!ent) {
1343 			dev_err(ips->dev, "failed to create debug file: %ld\n",
1344 				PTR_ERR(ent));
1345 			goto err_cleanup;
1346 		}
1347 	}
1348 
1349 	return;
1350 
1351 err_cleanup:
1352 	ips_debugfs_cleanup(ips);
1353 	return;
1354 }
1355 #endif /* CONFIG_DEBUG_FS */
1356 
1357 /**
1358  * ips_detect_cpu - detect whether CPU supports IPS
1359  *
1360  * Walk our list and see if we're on a supported CPU.  If we find one,
1361  * return the limits for it.
1362  */
1363 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1364 {
1365 	u64 turbo_power, misc_en;
1366 	struct ips_mcp_limits *limits = NULL;
1367 	u16 tdp;
1368 
1369 	if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1370 		dev_info(ips->dev, "Non-IPS CPU detected.\n");
1371 		return NULL;
1372 	}
1373 
1374 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1375 	/*
1376 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1377 	 * turbo manually or we'll get an illegal MSR access, even though
1378 	 * turbo will still be available.
1379 	 */
1380 	if (misc_en & IA32_MISC_TURBO_EN)
1381 		ips->turbo_toggle_allowed = true;
1382 	else
1383 		ips->turbo_toggle_allowed = false;
1384 
1385 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1386 		limits = &ips_sv_limits;
1387 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1388 		limits = &ips_lv_limits;
1389 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1390 		limits = &ips_ulv_limits;
1391 	else {
1392 		dev_info(ips->dev, "No CPUID match found.\n");
1393 		return NULL;
1394 	}
1395 
1396 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1397 	tdp = turbo_power & TURBO_TDP_MASK;
1398 
1399 	/* Sanity check TDP against CPU */
1400 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1401 		dev_info(ips->dev,
1402 			 "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1403 			 tdp / 8, limits->core_power_limit / 1000);
1404 		limits->core_power_limit = (tdp / 8) * 1000;
1405 	}
1406 
1407 	return limits;
1408 }
1409 
1410 /**
1411  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1412  * @ips: IPS driver
1413  *
1414  * The i915 driver exports several interfaces to allow the IPS driver to
1415  * monitor and control graphics turbo mode.  If we can find them, we can
1416  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1417  * thermal and power limits in the MCP.
1418  */
1419 static bool ips_get_i915_syms(struct ips_driver *ips)
1420 {
1421 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1422 	if (!ips->read_mch_val)
1423 		goto out_err;
1424 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1425 	if (!ips->gpu_raise)
1426 		goto out_put_mch;
1427 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1428 	if (!ips->gpu_lower)
1429 		goto out_put_raise;
1430 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1431 	if (!ips->gpu_busy)
1432 		goto out_put_lower;
1433 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1434 	if (!ips->gpu_turbo_disable)
1435 		goto out_put_busy;
1436 
1437 	return true;
1438 
1439 out_put_busy:
1440 	symbol_put(i915_gpu_busy);
1441 out_put_lower:
1442 	symbol_put(i915_gpu_lower);
1443 out_put_raise:
1444 	symbol_put(i915_gpu_raise);
1445 out_put_mch:
1446 	symbol_put(i915_read_mch_val);
1447 out_err:
1448 	return false;
1449 }
1450 
1451 static bool
1452 ips_gpu_turbo_enabled(struct ips_driver *ips)
1453 {
1454 	if (!ips->gpu_busy && late_i915_load) {
1455 		if (ips_get_i915_syms(ips)) {
1456 			dev_info(ips->dev,
1457 				 "i915 driver attached, reenabling gpu turbo\n");
1458 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1459 		}
1460 	}
1461 
1462 	return ips->gpu_turbo_enabled;
1463 }
1464 
1465 void
1466 ips_link_to_i915_driver(void)
1467 {
1468 	/* We can't cleanly get at the various ips_driver structs from
1469 	 * this caller (the i915 driver), so just set a flag saying
1470 	 * that it's time to try getting the symbols again.
1471 	 */
1472 	late_i915_load = true;
1473 }
1474 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1475 
1476 static const struct pci_device_id ips_id_table[] = {
1477 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1478 	{ 0, }
1479 };
1480 
1481 MODULE_DEVICE_TABLE(pci, ips_id_table);
1482 
1483 static int ips_blacklist_callback(const struct dmi_system_id *id)
1484 {
1485 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1486 	return 1;
1487 }
1488 
1489 static const struct dmi_system_id ips_blacklist[] = {
1490 	{
1491 		.callback = ips_blacklist_callback,
1492 		.ident = "HP ProBook",
1493 		.matches = {
1494 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1495 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1496 		},
1497 	},
1498 	{ }	/* terminating entry */
1499 };
1500 
1501 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1502 {
1503 	u64 platform_info;
1504 	struct ips_driver *ips;
1505 	u32 hts;
1506 	int ret = 0;
1507 	u16 htshi, trc, trc_required_mask;
1508 	u8 tse;
1509 
1510 	if (dmi_check_system(ips_blacklist))
1511 		return -ENODEV;
1512 
1513 	ips = devm_kzalloc(&dev->dev, sizeof(*ips), GFP_KERNEL);
1514 	if (!ips)
1515 		return -ENOMEM;
1516 
1517 	spin_lock_init(&ips->turbo_status_lock);
1518 	ips->dev = &dev->dev;
1519 
1520 	ips->limits = ips_detect_cpu(ips);
1521 	if (!ips->limits) {
1522 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1523 		return -ENXIO;
1524 	}
1525 
1526 	ret = pcim_enable_device(dev);
1527 	if (ret) {
1528 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1529 		return ret;
1530 	}
1531 
1532 	ret = pcim_iomap_regions(dev, 1 << 0, pci_name(dev));
1533 	if (ret) {
1534 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1535 		return ret;
1536 	}
1537 	ips->regmap = pcim_iomap_table(dev)[0];
1538 
1539 	pci_set_drvdata(dev, ips);
1540 
1541 	tse = thm_readb(THM_TSE);
1542 	if (tse != TSE_EN) {
1543 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1544 		return -ENXIO;
1545 	}
1546 
1547 	trc = thm_readw(THM_TRC);
1548 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1549 	if ((trc & trc_required_mask) != trc_required_mask) {
1550 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1551 		return -ENXIO;
1552 	}
1553 
1554 	if (trc & TRC_CORE2_EN)
1555 		ips->second_cpu = true;
1556 
1557 	update_turbo_limits(ips);
1558 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1559 		ips->mcp_power_limit / 10);
1560 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1561 		ips->core_power_limit / 10);
1562 	/* BIOS may update limits at runtime */
1563 	if (thm_readl(THM_PSC) & PSP_PBRT)
1564 		ips->poll_turbo_status = true;
1565 
1566 	if (!ips_get_i915_syms(ips)) {
1567 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1568 		ips->gpu_turbo_enabled = false;
1569 	} else {
1570 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1571 		ips->gpu_turbo_enabled = true;
1572 	}
1573 
1574 	/*
1575 	 * Check PLATFORM_INFO MSR to make sure this chip is
1576 	 * turbo capable.
1577 	 */
1578 	rdmsrl(PLATFORM_INFO, platform_info);
1579 	if (!(platform_info & PLATFORM_TDP)) {
1580 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1581 		return -ENODEV;
1582 	}
1583 
1584 	/*
1585 	 * IRQ handler for ME interaction
1586 	 * Note: don't use MSI here as the PCH has bugs.
1587 	 */
1588 	ret = pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_LEGACY);
1589 	if (ret < 0)
1590 		return ret;
1591 
1592 	ips->irq = pci_irq_vector(dev, 0);
1593 
1594 	ret = request_irq(ips->irq, ips_irq_handler, IRQF_SHARED, "ips", ips);
1595 	if (ret) {
1596 		dev_err(&dev->dev, "request irq failed, aborting\n");
1597 		return ret;
1598 	}
1599 
1600 	/* Enable aux, hot & critical interrupts */
1601 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1602 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1603 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1604 
1605 	/* Collect adjustment values */
1606 	ips->cta_val = thm_readw(THM_CTA);
1607 	ips->pta_val = thm_readw(THM_PTA);
1608 	ips->mgta_val = thm_readw(THM_MGTA);
1609 
1610 	/* Save turbo limits & ratios */
1611 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1612 
1613 	ips_disable_cpu_turbo(ips);
1614 	ips->cpu_turbo_enabled = false;
1615 
1616 	/* Create thermal adjust thread */
1617 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1618 	if (IS_ERR(ips->adjust)) {
1619 		dev_err(&dev->dev,
1620 			"failed to create thermal adjust thread, aborting\n");
1621 		ret = -ENOMEM;
1622 		goto error_free_irq;
1623 
1624 	}
1625 
1626 	/*
1627 	 * Set up the work queue and monitor thread. The monitor thread
1628 	 * will wake up ips_adjust thread.
1629 	 */
1630 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1631 	if (IS_ERR(ips->monitor)) {
1632 		dev_err(&dev->dev,
1633 			"failed to create thermal monitor thread, aborting\n");
1634 		ret = -ENOMEM;
1635 		goto error_thread_cleanup;
1636 	}
1637 
1638 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1639 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1640 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1641 
1642 	thm_writew(THM_HTSHI, htshi);
1643 	thm_writel(THM_HTS, hts);
1644 
1645 	ips_debugfs_init(ips);
1646 
1647 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1648 		 ips->mcp_temp_limit);
1649 	return ret;
1650 
1651 error_thread_cleanup:
1652 	kthread_stop(ips->adjust);
1653 error_free_irq:
1654 	free_irq(ips->irq, ips);
1655 	pci_free_irq_vectors(dev);
1656 	return ret;
1657 }
1658 
1659 static void ips_remove(struct pci_dev *dev)
1660 {
1661 	struct ips_driver *ips = pci_get_drvdata(dev);
1662 	u64 turbo_override;
1663 
1664 	if (!ips)
1665 		return;
1666 
1667 	ips_debugfs_cleanup(ips);
1668 
1669 	/* Release i915 driver */
1670 	if (ips->read_mch_val)
1671 		symbol_put(i915_read_mch_val);
1672 	if (ips->gpu_raise)
1673 		symbol_put(i915_gpu_raise);
1674 	if (ips->gpu_lower)
1675 		symbol_put(i915_gpu_lower);
1676 	if (ips->gpu_busy)
1677 		symbol_put(i915_gpu_busy);
1678 	if (ips->gpu_turbo_disable)
1679 		symbol_put(i915_gpu_turbo_disable);
1680 
1681 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1682 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1683 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1684 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1685 
1686 	free_irq(ips->irq, ips);
1687 	pci_free_irq_vectors(dev);
1688 	if (ips->adjust)
1689 		kthread_stop(ips->adjust);
1690 	if (ips->monitor)
1691 		kthread_stop(ips->monitor);
1692 	dev_dbg(&dev->dev, "IPS driver removed\n");
1693 }
1694 
1695 static struct pci_driver ips_pci_driver = {
1696 	.name = "intel ips",
1697 	.id_table = ips_id_table,
1698 	.probe = ips_probe,
1699 	.remove = ips_remove,
1700 };
1701 
1702 module_pci_driver(ips_pci_driver);
1703 
1704 MODULE_LICENSE("GPL");
1705 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1706 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1707