xref: /linux/drivers/platform/x86/intel_ips.c (revision 2a6b6c9a226279b4f6668450ddb21ae655558087)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2009-2010 Intel Corporation
4  *
5  * Authors:
6  *	Jesse Barnes <jbarnes@virtuousgeek.org>
7  */
8 
9 /*
10  * Some Intel Ibex Peak based platforms support so-called "intelligent
11  * power sharing", which allows the CPU and GPU to cooperate to maximize
12  * performance within a given TDP (thermal design point).  This driver
13  * performs the coordination between the CPU and GPU, monitors thermal and
14  * power statistics in the platform, and initializes power monitoring
15  * hardware.  It also provides a few tunables to control behavior.  Its
16  * primary purpose is to safely allow CPU and GPU turbo modes to be enabled
17  * by tracking power and thermal budget; secondarily it can boost turbo
18  * performance by allocating more power or thermal budget to the CPU or GPU
19  * based on available headroom and activity.
20  *
21  * The basic algorithm is driven by a 5s moving average of temperature.  If
22  * thermal headroom is available, the CPU and/or GPU power clamps may be
23  * adjusted upwards.  If we hit the thermal ceiling or a thermal trigger,
24  * we scale back the clamp.  Aside from trigger events (when we're critically
25  * close or over our TDP) we don't adjust the clamps more than once every
26  * five seconds.
27  *
28  * The thermal device (device 31, function 6) has a set of registers that
29  * are updated by the ME firmware.  The ME should also take the clamp values
30  * written to those registers and write them to the CPU, but we currently
31  * bypass that functionality and write the CPU MSR directly.
32  *
33  * UNSUPPORTED:
34  *   - dual MCP configs
35  *
36  * TODO:
37  *   - handle CPU hotplug
38  *   - provide turbo enable/disable api
39  *
40  * Related documents:
41  *   - CDI 403777, 403778 - Auburndale EDS vol 1 & 2
42  *   - CDI 401376 - Ibex Peak EDS
43  *   - ref 26037, 26641 - IPS BIOS spec
44  *   - ref 26489 - Nehalem BIOS writer's guide
45  *   - ref 26921 - Ibex Peak BIOS Specification
46  */
47 
48 #include <linux/debugfs.h>
49 #include <linux/delay.h>
50 #include <linux/interrupt.h>
51 #include <linux/kernel.h>
52 #include <linux/kthread.h>
53 #include <linux/module.h>
54 #include <linux/pci.h>
55 #include <linux/sched.h>
56 #include <linux/sched/loadavg.h>
57 #include <linux/seq_file.h>
58 #include <linux/string.h>
59 #include <linux/tick.h>
60 #include <linux/timer.h>
61 #include <linux/dmi.h>
62 #include <drm/intel/i915_drm.h>
63 #include <asm/msr.h>
64 #include <asm/processor.h>
65 #include <asm/cpu_device_id.h>
66 #include "intel_ips.h"
67 
68 #include <linux/io-64-nonatomic-lo-hi.h>
69 
70 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
71 
72 /*
73  * Package level MSRs for monitor/control
74  */
75 #define PLATFORM_INFO	0xce
76 #define   PLATFORM_TDP		(1<<29)
77 #define   PLATFORM_RATIO	(1<<28)
78 
79 #define IA32_MISC_ENABLE	0x1a0
80 #define   IA32_MISC_TURBO_EN	(1ULL<<38)
81 
82 #define TURBO_POWER_CURRENT_LIMIT	0x1ac
83 #define   TURBO_TDC_OVR_EN	(1UL<<31)
84 #define   TURBO_TDC_MASK	(0x000000007fff0000UL)
85 #define   TURBO_TDC_SHIFT	(16)
86 #define   TURBO_TDP_OVR_EN	(1UL<<15)
87 #define   TURBO_TDP_MASK	(0x0000000000003fffUL)
88 
89 /*
90  * Core/thread MSRs for monitoring
91  */
92 #define IA32_PERF_CTL		0x199
93 #define   IA32_PERF_TURBO_DIS	(1ULL<<32)
94 
95 /*
96  * Thermal PCI device regs
97  */
98 #define THM_CFG_TBAR	0x10
99 #define THM_CFG_TBAR_HI	0x14
100 
101 #define THM_TSIU	0x00
102 #define THM_TSE		0x01
103 #define   TSE_EN	0xb8
104 #define THM_TSS		0x02
105 #define THM_TSTR	0x03
106 #define THM_TSTTP	0x04
107 #define THM_TSCO	0x08
108 #define THM_TSES	0x0c
109 #define THM_TSGPEN	0x0d
110 #define   TSGPEN_HOT_LOHI	(1<<1)
111 #define   TSGPEN_CRIT_LOHI	(1<<2)
112 #define THM_TSPC	0x0e
113 #define THM_PPEC	0x10
114 #define THM_CTA		0x12
115 #define THM_PTA		0x14
116 #define   PTA_SLOPE_MASK	(0xff00)
117 #define   PTA_SLOPE_SHIFT	8
118 #define   PTA_OFFSET_MASK	(0x00ff)
119 #define THM_MGTA	0x16
120 #define   MGTA_SLOPE_MASK	(0xff00)
121 #define   MGTA_SLOPE_SHIFT	8
122 #define   MGTA_OFFSET_MASK	(0x00ff)
123 #define THM_TRC		0x1a
124 #define   TRC_CORE2_EN	(1<<15)
125 #define   TRC_THM_EN	(1<<12)
126 #define   TRC_C6_WAR	(1<<8)
127 #define   TRC_CORE1_EN	(1<<7)
128 #define   TRC_CORE_PWR	(1<<6)
129 #define   TRC_PCH_EN	(1<<5)
130 #define   TRC_MCH_EN	(1<<4)
131 #define   TRC_DIMM4	(1<<3)
132 #define   TRC_DIMM3	(1<<2)
133 #define   TRC_DIMM2	(1<<1)
134 #define   TRC_DIMM1	(1<<0)
135 #define THM_TES		0x20
136 #define THM_TEN		0x21
137 #define   TEN_UPDATE_EN	1
138 #define THM_PSC		0x24
139 #define   PSC_NTG	(1<<0) /* No GFX turbo support */
140 #define   PSC_NTPC	(1<<1) /* No CPU turbo support */
141 #define   PSC_PP_DEF	(0<<2) /* Perf policy up to driver */
142 #define   PSP_PP_PC	(1<<2) /* BIOS prefers CPU perf */
143 #define   PSP_PP_BAL	(2<<2) /* BIOS wants balanced perf */
144 #define   PSP_PP_GFX	(3<<2) /* BIOS prefers GFX perf */
145 #define   PSP_PBRT	(1<<4) /* BIOS run time support */
146 #define THM_CTV1	0x30
147 #define   CTV_TEMP_ERROR (1<<15)
148 #define   CTV_TEMP_MASK	0x3f
149 #define   CTV_
150 #define THM_CTV2	0x32
151 #define THM_CEC		0x34 /* undocumented power accumulator in joules */
152 #define THM_AE		0x3f
153 #define THM_HTS		0x50 /* 32 bits */
154 #define   HTS_PCPL_MASK	(0x7fe00000)
155 #define   HTS_PCPL_SHIFT 21
156 #define   HTS_GPL_MASK  (0x001ff000)
157 #define   HTS_GPL_SHIFT 12
158 #define   HTS_PP_MASK	(0x00000c00)
159 #define   HTS_PP_SHIFT  10
160 #define   HTS_PP_DEF	0
161 #define   HTS_PP_PROC	1
162 #define   HTS_PP_BAL	2
163 #define   HTS_PP_GFX	3
164 #define   HTS_PCTD_DIS	(1<<9)
165 #define   HTS_GTD_DIS	(1<<8)
166 #define   HTS_PTL_MASK  (0x000000fe)
167 #define   HTS_PTL_SHIFT 1
168 #define   HTS_NVV	(1<<0)
169 #define THM_HTSHI	0x54 /* 16 bits */
170 #define   HTS2_PPL_MASK		(0x03ff)
171 #define   HTS2_PRST_MASK	(0x3c00)
172 #define   HTS2_PRST_SHIFT	10
173 #define   HTS2_PRST_UNLOADED	0
174 #define   HTS2_PRST_RUNNING	1
175 #define   HTS2_PRST_TDISOP	2 /* turbo disabled due to power */
176 #define   HTS2_PRST_TDISHT	3 /* turbo disabled due to high temp */
177 #define   HTS2_PRST_TDISUSR	4 /* user disabled turbo */
178 #define   HTS2_PRST_TDISPLAT	5 /* platform disabled turbo */
179 #define   HTS2_PRST_TDISPM	6 /* power management disabled turbo */
180 #define   HTS2_PRST_TDISERR	7 /* some kind of error disabled turbo */
181 #define THM_PTL		0x56
182 #define THM_MGTV	0x58
183 #define   TV_MASK	0x000000000000ff00
184 #define   TV_SHIFT	8
185 #define THM_PTV		0x60
186 #define   PTV_MASK	0x00ff
187 #define THM_MMGPC	0x64
188 #define THM_MPPC	0x66
189 #define THM_MPCPC	0x68
190 #define THM_TSPIEN	0x82
191 #define   TSPIEN_AUX_LOHI	(1<<0)
192 #define   TSPIEN_HOT_LOHI	(1<<1)
193 #define   TSPIEN_CRIT_LOHI	(1<<2)
194 #define   TSPIEN_AUX2_LOHI	(1<<3)
195 #define THM_TSLOCK	0x83
196 #define THM_ATR		0x84
197 #define THM_TOF		0x87
198 #define THM_STS		0x98
199 #define   STS_PCPL_MASK		(0x7fe00000)
200 #define   STS_PCPL_SHIFT	21
201 #define   STS_GPL_MASK		(0x001ff000)
202 #define   STS_GPL_SHIFT		12
203 #define   STS_PP_MASK		(0x00000c00)
204 #define   STS_PP_SHIFT		10
205 #define   STS_PP_DEF		0
206 #define   STS_PP_PROC		1
207 #define   STS_PP_BAL		2
208 #define   STS_PP_GFX		3
209 #define   STS_PCTD_DIS		(1<<9)
210 #define   STS_GTD_DIS		(1<<8)
211 #define   STS_PTL_MASK		(0x000000fe)
212 #define   STS_PTL_SHIFT		1
213 #define   STS_NVV		(1<<0)
214 #define THM_SEC		0x9c
215 #define   SEC_ACK	(1<<0)
216 #define THM_TC3		0xa4
217 #define THM_TC1		0xa8
218 #define   STS_PPL_MASK		(0x0003ff00)
219 #define   STS_PPL_SHIFT		16
220 #define THM_TC2		0xac
221 #define THM_DTV		0xb0
222 #define THM_ITV		0xd8
223 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
224 #define   ITV_ME_SEQNO_SHIFT (16)
225 #define   ITV_MCH_TEMP_MASK 0x0000ff00
226 #define   ITV_MCH_TEMP_SHIFT (8)
227 #define   ITV_PCH_TEMP_MASK 0x000000ff
228 
229 #define thm_readb(off) readb(ips->regmap + (off))
230 #define thm_readw(off) readw(ips->regmap + (off))
231 #define thm_readl(off) readl(ips->regmap + (off))
232 #define thm_readq(off) readq(ips->regmap + (off))
233 
234 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
235 #define thm_writew(off, val) writew((val), ips->regmap + (off))
236 #define thm_writel(off, val) writel((val), ips->regmap + (off))
237 
238 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
239 static bool late_i915_load = false;
240 
241 /* For initial average collection */
242 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
243 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
244 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
245 
246 /* Per-SKU limits */
247 struct ips_mcp_limits {
248 	int mcp_power_limit; /* mW units */
249 	int core_power_limit;
250 	int mch_power_limit;
251 	int core_temp_limit; /* degrees C */
252 	int mch_temp_limit;
253 };
254 
255 /* Max temps are -10 degrees C to avoid PROCHOT# */
256 
257 static struct ips_mcp_limits ips_sv_limits = {
258 	.mcp_power_limit = 35000,
259 	.core_power_limit = 29000,
260 	.mch_power_limit = 20000,
261 	.core_temp_limit = 95,
262 	.mch_temp_limit = 90
263 };
264 
265 static struct ips_mcp_limits ips_lv_limits = {
266 	.mcp_power_limit = 25000,
267 	.core_power_limit = 21000,
268 	.mch_power_limit = 13000,
269 	.core_temp_limit = 95,
270 	.mch_temp_limit = 90
271 };
272 
273 static struct ips_mcp_limits ips_ulv_limits = {
274 	.mcp_power_limit = 18000,
275 	.core_power_limit = 14000,
276 	.mch_power_limit = 11000,
277 	.core_temp_limit = 95,
278 	.mch_temp_limit = 90
279 };
280 
281 struct ips_driver {
282 	struct device *dev;
283 	void __iomem *regmap;
284 	int irq;
285 
286 	struct task_struct *monitor;
287 	struct task_struct *adjust;
288 	struct dentry *debug_root;
289 	struct timer_list timer;
290 
291 	/* Average CPU core temps (all averages in .01 degrees C for precision) */
292 	u16 ctv1_avg_temp;
293 	u16 ctv2_avg_temp;
294 	/* GMCH average */
295 	u16 mch_avg_temp;
296 	/* Average for the CPU (both cores?) */
297 	u16 mcp_avg_temp;
298 	/* Average power consumption (in mW) */
299 	u32 cpu_avg_power;
300 	u32 mch_avg_power;
301 
302 	/* Offset values */
303 	u16 cta_val;
304 	u16 pta_val;
305 	u16 mgta_val;
306 
307 	/* Maximums & prefs, protected by turbo status lock */
308 	spinlock_t turbo_status_lock;
309 	u16 mcp_temp_limit;
310 	u16 mcp_power_limit;
311 	u16 core_power_limit;
312 	u16 mch_power_limit;
313 	bool cpu_turbo_enabled;
314 	bool __cpu_turbo_on;
315 	bool gpu_turbo_enabled;
316 	bool __gpu_turbo_on;
317 	bool gpu_preferred;
318 	bool poll_turbo_status;
319 	bool second_cpu;
320 	bool turbo_toggle_allowed;
321 	struct ips_mcp_limits *limits;
322 
323 	/* Optional MCH interfaces for if i915 is in use */
324 	unsigned long (*read_mch_val)(void);
325 	bool (*gpu_raise)(void);
326 	bool (*gpu_lower)(void);
327 	bool (*gpu_busy)(void);
328 	bool (*gpu_turbo_disable)(void);
329 
330 	/* For restoration at unload */
331 	u64 orig_turbo_limit;
332 	u64 orig_turbo_ratios;
333 };
334 
335 static bool
336 ips_gpu_turbo_enabled(struct ips_driver *ips);
337 
338 /**
339  * ips_cpu_busy - is CPU busy?
340  * @ips: IPS driver struct
341  *
342  * Check CPU for load to see whether we should increase its thermal budget.
343  *
344  * RETURNS:
345  * True if the CPU could use more power, false otherwise.
346  */
347 static bool ips_cpu_busy(struct ips_driver *ips)
348 {
349 	if ((avenrun[0] >> FSHIFT) > 1)
350 		return true;
351 
352 	return false;
353 }
354 
355 /**
356  * ips_cpu_raise - raise CPU power clamp
357  * @ips: IPS driver struct
358  *
359  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
360  * this platform.
361  *
362  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
363  * long as we haven't hit the TDP limit for the SKU).
364  */
365 static void ips_cpu_raise(struct ips_driver *ips)
366 {
367 	u64 turbo_override;
368 	u16 cur_tdp_limit, new_tdp_limit;
369 
370 	if (!ips->cpu_turbo_enabled)
371 		return;
372 
373 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
374 
375 	cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
376 	new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
377 
378 	/* Clamp to SKU TDP limit */
379 	if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
380 		new_tdp_limit = cur_tdp_limit;
381 
382 	thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
383 
384 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
385 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
386 
387 	turbo_override &= ~TURBO_TDP_MASK;
388 	turbo_override |= new_tdp_limit;
389 
390 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
391 }
392 
393 /**
394  * ips_cpu_lower - lower CPU power clamp
395  * @ips: IPS driver struct
396  *
397  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
398  *
399  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
400  * as low as the platform limits will allow (though we could go lower there
401  * wouldn't be much point).
402  */
403 static void ips_cpu_lower(struct ips_driver *ips)
404 {
405 	u64 turbo_override;
406 	u16 cur_limit, new_limit;
407 
408 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
409 
410 	cur_limit = turbo_override & TURBO_TDP_MASK;
411 	new_limit = cur_limit - 8; /* 1W decrease */
412 
413 	/* Clamp to SKU TDP limit */
414 	if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
415 		new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
416 
417 	thm_writew(THM_MPCPC, (new_limit * 10) / 8);
418 
419 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
420 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
421 
422 	turbo_override &= ~TURBO_TDP_MASK;
423 	turbo_override |= new_limit;
424 
425 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
426 }
427 
428 /**
429  * do_enable_cpu_turbo - internal turbo enable function
430  * @data: unused
431  *
432  * Internal function for actually updating MSRs.  When we enable/disable
433  * turbo, we need to do it on each CPU; this function is the one called
434  * by on_each_cpu() when needed.
435  */
436 static void do_enable_cpu_turbo(void *data)
437 {
438 	u64 perf_ctl;
439 
440 	rdmsrl(IA32_PERF_CTL, perf_ctl);
441 	if (perf_ctl & IA32_PERF_TURBO_DIS) {
442 		perf_ctl &= ~IA32_PERF_TURBO_DIS;
443 		wrmsrl(IA32_PERF_CTL, perf_ctl);
444 	}
445 }
446 
447 /**
448  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
449  * @ips: IPS driver struct
450  *
451  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
452  * all logical threads.
453  */
454 static void ips_enable_cpu_turbo(struct ips_driver *ips)
455 {
456 	/* Already on, no need to mess with MSRs */
457 	if (ips->__cpu_turbo_on)
458 		return;
459 
460 	if (ips->turbo_toggle_allowed)
461 		on_each_cpu(do_enable_cpu_turbo, ips, 1);
462 
463 	ips->__cpu_turbo_on = true;
464 }
465 
466 /**
467  * do_disable_cpu_turbo - internal turbo disable function
468  * @data: unused
469  *
470  * Internal function for actually updating MSRs.  When we enable/disable
471  * turbo, we need to do it on each CPU; this function is the one called
472  * by on_each_cpu() when needed.
473  */
474 static void do_disable_cpu_turbo(void *data)
475 {
476 	u64 perf_ctl;
477 
478 	rdmsrl(IA32_PERF_CTL, perf_ctl);
479 	if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
480 		perf_ctl |= IA32_PERF_TURBO_DIS;
481 		wrmsrl(IA32_PERF_CTL, perf_ctl);
482 	}
483 }
484 
485 /**
486  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
487  * @ips: IPS driver struct
488  *
489  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
490  * all logical threads.
491  */
492 static void ips_disable_cpu_turbo(struct ips_driver *ips)
493 {
494 	/* Already off, leave it */
495 	if (!ips->__cpu_turbo_on)
496 		return;
497 
498 	if (ips->turbo_toggle_allowed)
499 		on_each_cpu(do_disable_cpu_turbo, ips, 1);
500 
501 	ips->__cpu_turbo_on = false;
502 }
503 
504 /**
505  * ips_gpu_busy - is GPU busy?
506  * @ips: IPS driver struct
507  *
508  * Check GPU for load to see whether we should increase its thermal budget.
509  * We need to call into the i915 driver in this case.
510  *
511  * RETURNS:
512  * True if the GPU could use more power, false otherwise.
513  */
514 static bool ips_gpu_busy(struct ips_driver *ips)
515 {
516 	if (!ips_gpu_turbo_enabled(ips))
517 		return false;
518 
519 	return ips->gpu_busy();
520 }
521 
522 /**
523  * ips_gpu_raise - raise GPU power clamp
524  * @ips: IPS driver struct
525  *
526  * Raise the GPU frequency/power if possible.  We need to call into the
527  * i915 driver in this case.
528  */
529 static void ips_gpu_raise(struct ips_driver *ips)
530 {
531 	if (!ips_gpu_turbo_enabled(ips))
532 		return;
533 
534 	if (!ips->gpu_raise())
535 		ips->gpu_turbo_enabled = false;
536 
537 	return;
538 }
539 
540 /**
541  * ips_gpu_lower - lower GPU power clamp
542  * @ips: IPS driver struct
543  *
544  * Lower GPU frequency/power if possible.  Need to call i915.
545  */
546 static void ips_gpu_lower(struct ips_driver *ips)
547 {
548 	if (!ips_gpu_turbo_enabled(ips))
549 		return;
550 
551 	if (!ips->gpu_lower())
552 		ips->gpu_turbo_enabled = false;
553 
554 	return;
555 }
556 
557 /**
558  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
559  * @ips: IPS driver struct
560  *
561  * Call into the graphics driver indicating that it can safely use
562  * turbo mode.
563  */
564 static void ips_enable_gpu_turbo(struct ips_driver *ips)
565 {
566 	if (ips->__gpu_turbo_on)
567 		return;
568 	ips->__gpu_turbo_on = true;
569 }
570 
571 /**
572  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
573  * @ips: IPS driver struct
574  *
575  * Request that the graphics driver disable turbo mode.
576  */
577 static void ips_disable_gpu_turbo(struct ips_driver *ips)
578 {
579 	/* Avoid calling i915 if turbo is already disabled */
580 	if (!ips->__gpu_turbo_on)
581 		return;
582 
583 	if (!ips->gpu_turbo_disable())
584 		dev_err(ips->dev, "failed to disable graphics turbo\n");
585 	else
586 		ips->__gpu_turbo_on = false;
587 }
588 
589 /**
590  * mcp_exceeded - check whether we're outside our thermal & power limits
591  * @ips: IPS driver struct
592  *
593  * Check whether the MCP is over its thermal or power budget.
594  *
595  * Returns: %true if the temp or power has exceeded its maximum, else %false
596  */
597 static bool mcp_exceeded(struct ips_driver *ips)
598 {
599 	unsigned long flags;
600 	bool ret = false;
601 	u32 temp_limit;
602 	u32 avg_power;
603 
604 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
605 
606 	temp_limit = ips->mcp_temp_limit * 100;
607 	if (ips->mcp_avg_temp > temp_limit)
608 		ret = true;
609 
610 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
611 	if (avg_power > ips->mcp_power_limit)
612 		ret = true;
613 
614 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
615 
616 	return ret;
617 }
618 
619 /**
620  * cpu_exceeded - check whether a CPU core is outside its limits
621  * @ips: IPS driver struct
622  * @cpu: CPU number to check
623  *
624  * Check a given CPU's average temp or power is over its limit.
625  *
626  * Returns: %true if the temp or power has exceeded its maximum, else %false
627  */
628 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
629 {
630 	unsigned long flags;
631 	int avg;
632 	bool ret = false;
633 
634 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
635 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
636 	if (avg > (ips->limits->core_temp_limit * 100))
637 		ret = true;
638 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
639 		ret = true;
640 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
641 
642 	if (ret)
643 		dev_info(ips->dev, "CPU power or thermal limit exceeded\n");
644 
645 	return ret;
646 }
647 
648 /**
649  * mch_exceeded - check whether the GPU is over budget
650  * @ips: IPS driver struct
651  *
652  * Check the MCH temp & power against their maximums.
653  *
654  * Returns: %true if the temp or power has exceeded its maximum, else %false
655  */
656 static bool mch_exceeded(struct ips_driver *ips)
657 {
658 	unsigned long flags;
659 	bool ret = false;
660 
661 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
662 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
663 		ret = true;
664 	if (ips->mch_avg_power > ips->mch_power_limit)
665 		ret = true;
666 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
667 
668 	return ret;
669 }
670 
671 /**
672  * verify_limits - verify BIOS provided limits
673  * @ips: IPS structure
674  *
675  * BIOS can optionally provide non-default limits for power and temp.  Check
676  * them here and use the defaults if the BIOS values are not provided or
677  * are otherwise unusable.
678  */
679 static void verify_limits(struct ips_driver *ips)
680 {
681 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
682 	    ips->mcp_power_limit > 35000)
683 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
684 
685 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
686 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
687 	    ips->mcp_temp_limit > 150)
688 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
689 					  ips->limits->mch_temp_limit);
690 }
691 
692 /**
693  * update_turbo_limits - get various limits & settings from regs
694  * @ips: IPS driver struct
695  *
696  * Update the IPS power & temp limits, along with turbo enable flags,
697  * based on latest register contents.
698  *
699  * Used at init time and for runtime BIOS support, which requires polling
700  * the regs for updates (as a result of AC->DC transition for example).
701  *
702  * LOCKING:
703  * Caller must hold turbo_status_lock (outside of init)
704  */
705 static void update_turbo_limits(struct ips_driver *ips)
706 {
707 	u32 hts = thm_readl(THM_HTS);
708 
709 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
710 	/*
711 	 * Disable turbo for now, until we can figure out why the power figures
712 	 * are wrong
713 	 */
714 	ips->cpu_turbo_enabled = false;
715 
716 	if (ips->gpu_busy)
717 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
718 
719 	ips->core_power_limit = thm_readw(THM_MPCPC);
720 	ips->mch_power_limit = thm_readw(THM_MMGPC);
721 	ips->mcp_temp_limit = thm_readw(THM_PTL);
722 	ips->mcp_power_limit = thm_readw(THM_MPPC);
723 
724 	verify_limits(ips);
725 	/* Ignore BIOS CPU vs GPU pref */
726 }
727 
728 /**
729  * ips_adjust - adjust power clamp based on thermal state
730  * @data: ips driver structure
731  *
732  * Wake up every 5s or so and check whether we should adjust the power clamp.
733  * Check CPU and GPU load to determine which needs adjustment.  There are
734  * several things to consider here:
735  *   - do we need to adjust up or down?
736  *   - is CPU busy?
737  *   - is GPU busy?
738  *   - is CPU in turbo?
739  *   - is GPU in turbo?
740  *   - is CPU or GPU preferred? (CPU is default)
741  *
742  * So, given the above, we do the following:
743  *   - up (TDP available)
744  *     - CPU not busy, GPU not busy - nothing
745  *     - CPU busy, GPU not busy - adjust CPU up
746  *     - CPU not busy, GPU busy - adjust GPU up
747  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
748  *       non-preferred unit if necessary
749  *   - down (at TDP limit)
750  *     - adjust both CPU and GPU down if possible
751  *
752  *              |cpu+ gpu+      cpu+gpu-        cpu-gpu+        cpu-gpu-
753  * cpu < gpu <  |cpu+gpu+       cpu+            gpu+            nothing
754  * cpu < gpu >= |cpu+gpu-(mcp<) cpu+gpu-(mcp<)  gpu-            gpu-
755  * cpu >= gpu < |cpu-gpu+(mcp<) cpu-            cpu-gpu+(mcp<)  cpu-
756  * cpu >= gpu >=|cpu-gpu-       cpu-gpu-        cpu-gpu-        cpu-gpu-
757  *
758  * Returns: %0
759  */
760 static int ips_adjust(void *data)
761 {
762 	struct ips_driver *ips = data;
763 	unsigned long flags;
764 
765 	dev_dbg(ips->dev, "starting ips-adjust thread\n");
766 
767 	/*
768 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
769 	 * often isn't recommended due to ME interaction.
770 	 */
771 	do {
772 		bool cpu_busy = ips_cpu_busy(ips);
773 		bool gpu_busy = ips_gpu_busy(ips);
774 
775 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
776 		if (ips->poll_turbo_status)
777 			update_turbo_limits(ips);
778 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
779 
780 		/* Update turbo status if necessary */
781 		if (ips->cpu_turbo_enabled)
782 			ips_enable_cpu_turbo(ips);
783 		else
784 			ips_disable_cpu_turbo(ips);
785 
786 		if (ips->gpu_turbo_enabled)
787 			ips_enable_gpu_turbo(ips);
788 		else
789 			ips_disable_gpu_turbo(ips);
790 
791 		/* We're outside our comfort zone, crank them down */
792 		if (mcp_exceeded(ips)) {
793 			ips_cpu_lower(ips);
794 			ips_gpu_lower(ips);
795 			goto sleep;
796 		}
797 
798 		if (!cpu_exceeded(ips, 0) && cpu_busy)
799 			ips_cpu_raise(ips);
800 		else
801 			ips_cpu_lower(ips);
802 
803 		if (!mch_exceeded(ips) && gpu_busy)
804 			ips_gpu_raise(ips);
805 		else
806 			ips_gpu_lower(ips);
807 
808 sleep:
809 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
810 	} while (!kthread_should_stop());
811 
812 	dev_dbg(ips->dev, "ips-adjust thread stopped\n");
813 
814 	return 0;
815 }
816 
817 /*
818  * Helpers for reading out temp/power values and calculating their
819  * averages for the decision making and monitoring functions.
820  */
821 
822 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
823 {
824 	u64 total = 0;
825 	int i;
826 	u16 avg;
827 
828 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
829 		total += (u64)(array[i] * 100);
830 
831 	do_div(total, IPS_SAMPLE_COUNT);
832 
833 	avg = (u16)total;
834 
835 	return avg;
836 }
837 
838 static u16 read_mgtv(struct ips_driver *ips)
839 {
840 	u16 __maybe_unused ret;
841 	u64 slope, offset;
842 	u64 val;
843 
844 	val = thm_readq(THM_MGTV);
845 	val = (val & TV_MASK) >> TV_SHIFT;
846 
847 	slope = offset = thm_readw(THM_MGTA);
848 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
849 	offset = offset & MGTA_OFFSET_MASK;
850 
851 	ret = ((val * slope + 0x40) >> 7) + offset;
852 
853 	return 0; /* MCH temp reporting buggy */
854 }
855 
856 static u16 read_ptv(struct ips_driver *ips)
857 {
858 	u16 val;
859 
860 	val = thm_readw(THM_PTV) & PTV_MASK;
861 
862 	return val;
863 }
864 
865 static u16 read_ctv(struct ips_driver *ips, int cpu)
866 {
867 	int reg = cpu ? THM_CTV2 : THM_CTV1;
868 	u16 val;
869 
870 	val = thm_readw(reg);
871 	if (!(val & CTV_TEMP_ERROR))
872 		val = (val) >> 6; /* discard fractional component */
873 	else
874 		val = 0;
875 
876 	return val;
877 }
878 
879 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
880 {
881 	u32 val;
882 	u32 ret;
883 
884 	/*
885 	 * CEC is in joules/65535.  Take difference over time to
886 	 * get watts.
887 	 */
888 	val = thm_readl(THM_CEC);
889 
890 	/* period is in ms and we want mW */
891 	ret = (((val - *last) * 1000) / period);
892 	ret = (ret * 1000) / 65535;
893 	*last = val;
894 
895 	return 0;
896 }
897 
898 static const u16 temp_decay_factor = 2;
899 static u16 update_average_temp(u16 avg, u16 val)
900 {
901 	u16 ret;
902 
903 	/* Multiply by 100 for extra precision */
904 	ret = (val * 100 / temp_decay_factor) +
905 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
906 	return ret;
907 }
908 
909 static const u16 power_decay_factor = 2;
910 static u16 update_average_power(u32 avg, u32 val)
911 {
912 	u32 ret;
913 
914 	ret = (val / power_decay_factor) +
915 		(((power_decay_factor - 1) * avg) / power_decay_factor);
916 
917 	return ret;
918 }
919 
920 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
921 {
922 	u64 total = 0;
923 	u32 avg;
924 	int i;
925 
926 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
927 		total += array[i];
928 
929 	do_div(total, IPS_SAMPLE_COUNT);
930 	avg = (u32)total;
931 
932 	return avg;
933 }
934 
935 static void monitor_timeout(struct timer_list *t)
936 {
937 	struct ips_driver *ips = from_timer(ips, t, timer);
938 	wake_up_process(ips->monitor);
939 }
940 
941 /**
942  * ips_monitor - temp/power monitoring thread
943  * @data: ips driver structure
944  *
945  * This is the main function for the IPS driver.  It monitors power and
946  * temperature in the MCP and adjusts CPU and GPU power clamps accordingly.
947  *
948  * We keep a 5s moving average of power consumption and temperature.  Using
949  * that data, along with CPU vs GPU preference, we adjust the power clamps
950  * up or down.
951  *
952  * Returns: %0 on success or -errno on error
953  */
954 static int ips_monitor(void *data)
955 {
956 	struct ips_driver *ips = data;
957 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
958 	int i;
959 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
960 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
961 	u8 cur_seqno, last_seqno;
962 
963 	mcp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
964 	ctv1_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
965 	ctv2_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
966 	mch_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
967 	cpu_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
968 	mchp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
969 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
970 			!cpu_samples || !mchp_samples) {
971 		dev_err(ips->dev,
972 			"failed to allocate sample array, ips disabled\n");
973 		kfree(mcp_samples);
974 		kfree(ctv1_samples);
975 		kfree(ctv2_samples);
976 		kfree(mch_samples);
977 		kfree(cpu_samples);
978 		kfree(mchp_samples);
979 		return -ENOMEM;
980 	}
981 
982 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
983 		ITV_ME_SEQNO_SHIFT;
984 	seqno_timestamp = get_jiffies_64();
985 
986 	old_cpu_power = thm_readl(THM_CEC);
987 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
988 
989 	/* Collect an initial average */
990 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
991 		u32 mchp, cpu_power;
992 		u16 val;
993 
994 		mcp_samples[i] = read_ptv(ips);
995 
996 		val = read_ctv(ips, 0);
997 		ctv1_samples[i] = val;
998 
999 		val = read_ctv(ips, 1);
1000 		ctv2_samples[i] = val;
1001 
1002 		val = read_mgtv(ips);
1003 		mch_samples[i] = val;
1004 
1005 		cpu_power = get_cpu_power(ips, &old_cpu_power,
1006 					  IPS_SAMPLE_PERIOD);
1007 		cpu_samples[i] = cpu_power;
1008 
1009 		if (ips->read_mch_val) {
1010 			mchp = ips->read_mch_val();
1011 			mchp_samples[i] = mchp;
1012 		}
1013 
1014 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1015 		if (kthread_should_stop())
1016 			break;
1017 	}
1018 
1019 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1020 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1021 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1022 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1023 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1024 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1025 	kfree(mcp_samples);
1026 	kfree(ctv1_samples);
1027 	kfree(ctv2_samples);
1028 	kfree(mch_samples);
1029 	kfree(cpu_samples);
1030 	kfree(mchp_samples);
1031 
1032 	/* Start the adjustment thread now that we have data */
1033 	wake_up_process(ips->adjust);
1034 
1035 	/*
1036 	 * Ok, now we have an initial avg.  From here on out, we track the
1037 	 * running avg using a decaying average calculation.  This allows
1038 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1039 	 */
1040 	old_cpu_power = thm_readl(THM_CEC);
1041 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1042 	last_sample_period = IPS_SAMPLE_PERIOD;
1043 
1044 	timer_setup(&ips->timer, monitor_timeout, TIMER_DEFERRABLE);
1045 	do {
1046 		u32 cpu_val, mch_val;
1047 		u16 val;
1048 
1049 		/* MCP itself */
1050 		val = read_ptv(ips);
1051 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1052 
1053 		/* Processor 0 */
1054 		val = read_ctv(ips, 0);
1055 		ips->ctv1_avg_temp =
1056 			update_average_temp(ips->ctv1_avg_temp, val);
1057 		/* Power */
1058 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1059 					last_sample_period);
1060 		ips->cpu_avg_power =
1061 			update_average_power(ips->cpu_avg_power, cpu_val);
1062 
1063 		if (ips->second_cpu) {
1064 			/* Processor 1 */
1065 			val = read_ctv(ips, 1);
1066 			ips->ctv2_avg_temp =
1067 				update_average_temp(ips->ctv2_avg_temp, val);
1068 		}
1069 
1070 		/* MCH */
1071 		val = read_mgtv(ips);
1072 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1073 		/* Power */
1074 		if (ips->read_mch_val) {
1075 			mch_val = ips->read_mch_val();
1076 			ips->mch_avg_power =
1077 				update_average_power(ips->mch_avg_power,
1078 						     mch_val);
1079 		}
1080 
1081 		/*
1082 		 * Make sure ME is updating thermal regs.
1083 		 * Note:
1084 		 * If it's been more than a second since the last update,
1085 		 * the ME is probably hung.
1086 		 */
1087 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1088 			ITV_ME_SEQNO_SHIFT;
1089 		if (cur_seqno == last_seqno &&
1090 		    time_after(jiffies, seqno_timestamp + HZ)) {
1091 			dev_warn(ips->dev,
1092 				 "ME failed to update for more than 1s, likely hung\n");
1093 		} else {
1094 			seqno_timestamp = get_jiffies_64();
1095 			last_seqno = cur_seqno;
1096 		}
1097 
1098 		last_msecs = jiffies_to_msecs(jiffies);
1099 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1100 
1101 		__set_current_state(TASK_INTERRUPTIBLE);
1102 		mod_timer(&ips->timer, expire);
1103 		schedule();
1104 
1105 		/* Calculate actual sample period for power averaging */
1106 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1107 		if (!last_sample_period)
1108 			last_sample_period = 1;
1109 	} while (!kthread_should_stop());
1110 
1111 	del_timer_sync(&ips->timer);
1112 
1113 	dev_dbg(ips->dev, "ips-monitor thread stopped\n");
1114 
1115 	return 0;
1116 }
1117 
1118 /**
1119  * ips_irq_handler - handle temperature triggers and other IPS events
1120  * @irq: irq number
1121  * @arg: unused
1122  *
1123  * Handle temperature limit trigger events, generally by lowering the clamps.
1124  * If we're at a critical limit, we clamp back to the lowest possible value
1125  * to prevent emergency shutdown.
1126  *
1127  * Returns: IRQ_NONE or IRQ_HANDLED
1128  */
1129 static irqreturn_t ips_irq_handler(int irq, void *arg)
1130 {
1131 	struct ips_driver *ips = arg;
1132 	u8 tses = thm_readb(THM_TSES);
1133 	u8 tes = thm_readb(THM_TES);
1134 
1135 	if (!tses && !tes)
1136 		return IRQ_NONE;
1137 
1138 	dev_info(ips->dev, "TSES: 0x%02x\n", tses);
1139 	dev_info(ips->dev, "TES: 0x%02x\n", tes);
1140 
1141 	/* STS update from EC? */
1142 	if (tes & 1) {
1143 		u32 sts, tc1;
1144 
1145 		sts = thm_readl(THM_STS);
1146 		tc1 = thm_readl(THM_TC1);
1147 
1148 		if (sts & STS_NVV) {
1149 			spin_lock(&ips->turbo_status_lock);
1150 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1151 				STS_PCPL_SHIFT;
1152 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1153 				STS_GPL_SHIFT;
1154 			/* ignore EC CPU vs GPU pref */
1155 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1156 			/*
1157 			 * Disable turbo for now, until we can figure
1158 			 * out why the power figures are wrong
1159 			 */
1160 			ips->cpu_turbo_enabled = false;
1161 			if (ips->gpu_busy)
1162 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1163 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1164 				STS_PTL_SHIFT;
1165 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1166 				STS_PPL_SHIFT;
1167 			verify_limits(ips);
1168 			spin_unlock(&ips->turbo_status_lock);
1169 
1170 			thm_writeb(THM_SEC, SEC_ACK);
1171 		}
1172 		thm_writeb(THM_TES, tes);
1173 	}
1174 
1175 	/* Thermal trip */
1176 	if (tses) {
1177 		dev_warn(ips->dev, "thermal trip occurred, tses: 0x%04x\n",
1178 			 tses);
1179 		thm_writeb(THM_TSES, tses);
1180 	}
1181 
1182 	return IRQ_HANDLED;
1183 }
1184 
1185 #ifndef CONFIG_DEBUG_FS
1186 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1187 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1188 #else
1189 
1190 /* Expose current state and limits in debugfs if possible */
1191 
1192 static int cpu_temp_show(struct seq_file *m, void *data)
1193 {
1194 	struct ips_driver *ips = m->private;
1195 
1196 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1197 		   ips->ctv1_avg_temp % 100);
1198 
1199 	return 0;
1200 }
1201 DEFINE_SHOW_ATTRIBUTE(cpu_temp);
1202 
1203 static int cpu_power_show(struct seq_file *m, void *data)
1204 {
1205 	struct ips_driver *ips = m->private;
1206 
1207 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1208 
1209 	return 0;
1210 }
1211 DEFINE_SHOW_ATTRIBUTE(cpu_power);
1212 
1213 static int cpu_clamp_show(struct seq_file *m, void *data)
1214 {
1215 	u64 turbo_override;
1216 	int tdp, tdc;
1217 
1218 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1219 
1220 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1221 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1222 
1223 	/* Convert to .1W/A units */
1224 	tdp = tdp * 10 / 8;
1225 	tdc = tdc * 10 / 8;
1226 
1227 	/* Watts Amperes */
1228 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1229 		   tdc / 10, tdc % 10);
1230 
1231 	return 0;
1232 }
1233 DEFINE_SHOW_ATTRIBUTE(cpu_clamp);
1234 
1235 static int mch_temp_show(struct seq_file *m, void *data)
1236 {
1237 	struct ips_driver *ips = m->private;
1238 
1239 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1240 		   ips->mch_avg_temp % 100);
1241 
1242 	return 0;
1243 }
1244 DEFINE_SHOW_ATTRIBUTE(mch_temp);
1245 
1246 static int mch_power_show(struct seq_file *m, void *data)
1247 {
1248 	struct ips_driver *ips = m->private;
1249 
1250 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1251 
1252 	return 0;
1253 }
1254 DEFINE_SHOW_ATTRIBUTE(mch_power);
1255 
1256 static void ips_debugfs_cleanup(struct ips_driver *ips)
1257 {
1258 	debugfs_remove_recursive(ips->debug_root);
1259 }
1260 
1261 static void ips_debugfs_init(struct ips_driver *ips)
1262 {
1263 	ips->debug_root = debugfs_create_dir("ips", NULL);
1264 
1265 	debugfs_create_file("cpu_temp", 0444, ips->debug_root, ips, &cpu_temp_fops);
1266 	debugfs_create_file("cpu_power", 0444, ips->debug_root, ips, &cpu_power_fops);
1267 	debugfs_create_file("cpu_clamp", 0444, ips->debug_root, ips, &cpu_clamp_fops);
1268 	debugfs_create_file("mch_temp", 0444, ips->debug_root, ips, &mch_temp_fops);
1269 	debugfs_create_file("mch_power", 0444, ips->debug_root, ips, &mch_power_fops);
1270 }
1271 #endif /* CONFIG_DEBUG_FS */
1272 
1273 /**
1274  * ips_detect_cpu - detect whether CPU supports IPS
1275  * @ips: IPS driver struct
1276  *
1277  * Walk our list and see if we're on a supported CPU.  If we find one,
1278  * return the limits for it.
1279  *
1280  * Returns: the &ips_mcp_limits struct that matches the boot CPU or %NULL
1281  */
1282 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1283 {
1284 	u64 turbo_power, misc_en;
1285 	struct ips_mcp_limits *limits = NULL;
1286 	u16 tdp;
1287 
1288 	if (!(boot_cpu_data.x86_vfm == INTEL_WESTMERE)) {
1289 		dev_info(ips->dev, "Non-IPS CPU detected.\n");
1290 		return NULL;
1291 	}
1292 
1293 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1294 	/*
1295 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1296 	 * turbo manually or we'll get an illegal MSR access, even though
1297 	 * turbo will still be available.
1298 	 */
1299 	if (misc_en & IA32_MISC_TURBO_EN)
1300 		ips->turbo_toggle_allowed = true;
1301 	else
1302 		ips->turbo_toggle_allowed = false;
1303 
1304 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1305 		limits = &ips_sv_limits;
1306 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1307 		limits = &ips_lv_limits;
1308 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1309 		limits = &ips_ulv_limits;
1310 	else {
1311 		dev_info(ips->dev, "No CPUID match found.\n");
1312 		return NULL;
1313 	}
1314 
1315 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1316 	tdp = turbo_power & TURBO_TDP_MASK;
1317 
1318 	/* Sanity check TDP against CPU */
1319 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1320 		dev_info(ips->dev,
1321 			 "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1322 			 tdp / 8, limits->core_power_limit / 1000);
1323 		limits->core_power_limit = (tdp / 8) * 1000;
1324 	}
1325 
1326 	return limits;
1327 }
1328 
1329 /**
1330  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1331  * @ips: IPS driver
1332  *
1333  * The i915 driver exports several interfaces to allow the IPS driver to
1334  * monitor and control graphics turbo mode.  If we can find them, we can
1335  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1336  * thermal and power limits in the MCP.
1337  *
1338  * Returns: %true if the required symbols are found, else %false
1339  */
1340 static bool ips_get_i915_syms(struct ips_driver *ips)
1341 {
1342 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1343 	if (!ips->read_mch_val)
1344 		goto out_err;
1345 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1346 	if (!ips->gpu_raise)
1347 		goto out_put_mch;
1348 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1349 	if (!ips->gpu_lower)
1350 		goto out_put_raise;
1351 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1352 	if (!ips->gpu_busy)
1353 		goto out_put_lower;
1354 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1355 	if (!ips->gpu_turbo_disable)
1356 		goto out_put_busy;
1357 
1358 	return true;
1359 
1360 out_put_busy:
1361 	symbol_put(i915_gpu_busy);
1362 out_put_lower:
1363 	symbol_put(i915_gpu_lower);
1364 out_put_raise:
1365 	symbol_put(i915_gpu_raise);
1366 out_put_mch:
1367 	symbol_put(i915_read_mch_val);
1368 out_err:
1369 	return false;
1370 }
1371 
1372 static bool
1373 ips_gpu_turbo_enabled(struct ips_driver *ips)
1374 {
1375 	if (!ips->gpu_busy && late_i915_load) {
1376 		if (ips_get_i915_syms(ips)) {
1377 			dev_info(ips->dev,
1378 				 "i915 driver attached, reenabling gpu turbo\n");
1379 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1380 		}
1381 	}
1382 
1383 	return ips->gpu_turbo_enabled;
1384 }
1385 
1386 void
1387 ips_link_to_i915_driver(void)
1388 {
1389 	/* We can't cleanly get at the various ips_driver structs from
1390 	 * this caller (the i915 driver), so just set a flag saying
1391 	 * that it's time to try getting the symbols again.
1392 	 */
1393 	late_i915_load = true;
1394 }
1395 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1396 
1397 static const struct pci_device_id ips_id_table[] = {
1398 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1399 	{ 0, }
1400 };
1401 
1402 MODULE_DEVICE_TABLE(pci, ips_id_table);
1403 
1404 static int ips_blacklist_callback(const struct dmi_system_id *id)
1405 {
1406 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1407 	return 1;
1408 }
1409 
1410 static const struct dmi_system_id ips_blacklist[] = {
1411 	{
1412 		.callback = ips_blacklist_callback,
1413 		.ident = "HP ProBook",
1414 		.matches = {
1415 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1416 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1417 		},
1418 	},
1419 	{ }	/* terminating entry */
1420 };
1421 
1422 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1423 {
1424 	u64 platform_info;
1425 	struct ips_driver *ips;
1426 	u32 hts;
1427 	int ret = 0;
1428 	u16 htshi, trc, trc_required_mask;
1429 	u8 tse;
1430 
1431 	if (dmi_check_system(ips_blacklist))
1432 		return -ENODEV;
1433 
1434 	ips = devm_kzalloc(&dev->dev, sizeof(*ips), GFP_KERNEL);
1435 	if (!ips)
1436 		return -ENOMEM;
1437 
1438 	spin_lock_init(&ips->turbo_status_lock);
1439 	ips->dev = &dev->dev;
1440 
1441 	ips->limits = ips_detect_cpu(ips);
1442 	if (!ips->limits) {
1443 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1444 		return -ENXIO;
1445 	}
1446 
1447 	ret = pcim_enable_device(dev);
1448 	if (ret) {
1449 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1450 		return ret;
1451 	}
1452 
1453 	ret = pcim_iomap_regions(dev, 1 << 0, pci_name(dev));
1454 	if (ret) {
1455 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1456 		return ret;
1457 	}
1458 	ips->regmap = pcim_iomap_table(dev)[0];
1459 
1460 	pci_set_drvdata(dev, ips);
1461 
1462 	tse = thm_readb(THM_TSE);
1463 	if (tse != TSE_EN) {
1464 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1465 		return -ENXIO;
1466 	}
1467 
1468 	trc = thm_readw(THM_TRC);
1469 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1470 	if ((trc & trc_required_mask) != trc_required_mask) {
1471 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1472 		return -ENXIO;
1473 	}
1474 
1475 	if (trc & TRC_CORE2_EN)
1476 		ips->second_cpu = true;
1477 
1478 	update_turbo_limits(ips);
1479 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1480 		ips->mcp_power_limit / 10);
1481 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1482 		ips->core_power_limit / 10);
1483 	/* BIOS may update limits at runtime */
1484 	if (thm_readl(THM_PSC) & PSP_PBRT)
1485 		ips->poll_turbo_status = true;
1486 
1487 	if (!ips_get_i915_syms(ips)) {
1488 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1489 		ips->gpu_turbo_enabled = false;
1490 	} else {
1491 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1492 		ips->gpu_turbo_enabled = true;
1493 	}
1494 
1495 	/*
1496 	 * Check PLATFORM_INFO MSR to make sure this chip is
1497 	 * turbo capable.
1498 	 */
1499 	rdmsrl(PLATFORM_INFO, platform_info);
1500 	if (!(platform_info & PLATFORM_TDP)) {
1501 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1502 		return -ENODEV;
1503 	}
1504 
1505 	/*
1506 	 * IRQ handler for ME interaction
1507 	 * Note: don't use MSI here as the PCH has bugs.
1508 	 */
1509 	ret = pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_INTX);
1510 	if (ret < 0)
1511 		return ret;
1512 
1513 	ips->irq = pci_irq_vector(dev, 0);
1514 
1515 	ret = request_irq(ips->irq, ips_irq_handler, IRQF_SHARED, "ips", ips);
1516 	if (ret) {
1517 		dev_err(&dev->dev, "request irq failed, aborting\n");
1518 		return ret;
1519 	}
1520 
1521 	/* Enable aux, hot & critical interrupts */
1522 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1523 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1524 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1525 
1526 	/* Collect adjustment values */
1527 	ips->cta_val = thm_readw(THM_CTA);
1528 	ips->pta_val = thm_readw(THM_PTA);
1529 	ips->mgta_val = thm_readw(THM_MGTA);
1530 
1531 	/* Save turbo limits & ratios */
1532 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1533 
1534 	ips_disable_cpu_turbo(ips);
1535 	ips->cpu_turbo_enabled = false;
1536 
1537 	/* Create thermal adjust thread */
1538 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1539 	if (IS_ERR(ips->adjust)) {
1540 		dev_err(&dev->dev,
1541 			"failed to create thermal adjust thread, aborting\n");
1542 		ret = -ENOMEM;
1543 		goto error_free_irq;
1544 
1545 	}
1546 
1547 	/*
1548 	 * Set up the work queue and monitor thread. The monitor thread
1549 	 * will wake up ips_adjust thread.
1550 	 */
1551 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1552 	if (IS_ERR(ips->monitor)) {
1553 		dev_err(&dev->dev,
1554 			"failed to create thermal monitor thread, aborting\n");
1555 		ret = -ENOMEM;
1556 		goto error_thread_cleanup;
1557 	}
1558 
1559 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1560 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1561 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1562 
1563 	thm_writew(THM_HTSHI, htshi);
1564 	thm_writel(THM_HTS, hts);
1565 
1566 	ips_debugfs_init(ips);
1567 
1568 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1569 		 ips->mcp_temp_limit);
1570 	return ret;
1571 
1572 error_thread_cleanup:
1573 	kthread_stop(ips->adjust);
1574 error_free_irq:
1575 	free_irq(ips->irq, ips);
1576 	pci_free_irq_vectors(dev);
1577 	return ret;
1578 }
1579 
1580 static void ips_remove(struct pci_dev *dev)
1581 {
1582 	struct ips_driver *ips = pci_get_drvdata(dev);
1583 	u64 turbo_override;
1584 
1585 	ips_debugfs_cleanup(ips);
1586 
1587 	/* Release i915 driver */
1588 	if (ips->read_mch_val)
1589 		symbol_put(i915_read_mch_val);
1590 	if (ips->gpu_raise)
1591 		symbol_put(i915_gpu_raise);
1592 	if (ips->gpu_lower)
1593 		symbol_put(i915_gpu_lower);
1594 	if (ips->gpu_busy)
1595 		symbol_put(i915_gpu_busy);
1596 	if (ips->gpu_turbo_disable)
1597 		symbol_put(i915_gpu_turbo_disable);
1598 
1599 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1600 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1601 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1602 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1603 
1604 	free_irq(ips->irq, ips);
1605 	pci_free_irq_vectors(dev);
1606 	if (ips->adjust)
1607 		kthread_stop(ips->adjust);
1608 	if (ips->monitor)
1609 		kthread_stop(ips->monitor);
1610 	dev_dbg(&dev->dev, "IPS driver removed\n");
1611 }
1612 
1613 static struct pci_driver ips_pci_driver = {
1614 	.name = "intel ips",
1615 	.id_table = ips_id_table,
1616 	.probe = ips_probe,
1617 	.remove = ips_remove,
1618 };
1619 
1620 module_pci_driver(ips_pci_driver);
1621 
1622 MODULE_LICENSE("GPL v2");
1623 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1624 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1625