xref: /linux/drivers/platform/x86/intel_ips.c (revision 76f623d2d4283cc36a9c8a5b585df74638f1efa5)
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/i915_drm.h>
63 #include <asm/msr.h>
64 #include <asm/processor.h>
65 #include "intel_ips.h"
66 
67 #include <linux/io-64-nonatomic-lo-hi.h>
68 
69 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
70 
71 /*
72  * Package level MSRs for monitor/control
73  */
74 #define PLATFORM_INFO	0xce
75 #define   PLATFORM_TDP		(1<<29)
76 #define   PLATFORM_RATIO	(1<<28)
77 
78 #define IA32_MISC_ENABLE	0x1a0
79 #define   IA32_MISC_TURBO_EN	(1ULL<<38)
80 
81 #define TURBO_POWER_CURRENT_LIMIT	0x1ac
82 #define   TURBO_TDC_OVR_EN	(1UL<<31)
83 #define   TURBO_TDC_MASK	(0x000000007fff0000UL)
84 #define   TURBO_TDC_SHIFT	(16)
85 #define   TURBO_TDP_OVR_EN	(1UL<<15)
86 #define   TURBO_TDP_MASK	(0x0000000000003fffUL)
87 
88 /*
89  * Core/thread MSRs for monitoring
90  */
91 #define IA32_PERF_CTL		0x199
92 #define   IA32_PERF_TURBO_DIS	(1ULL<<32)
93 
94 /*
95  * Thermal PCI device regs
96  */
97 #define THM_CFG_TBAR	0x10
98 #define THM_CFG_TBAR_HI	0x14
99 
100 #define THM_TSIU	0x00
101 #define THM_TSE		0x01
102 #define   TSE_EN	0xb8
103 #define THM_TSS		0x02
104 #define THM_TSTR	0x03
105 #define THM_TSTTP	0x04
106 #define THM_TSCO	0x08
107 #define THM_TSES	0x0c
108 #define THM_TSGPEN	0x0d
109 #define   TSGPEN_HOT_LOHI	(1<<1)
110 #define   TSGPEN_CRIT_LOHI	(1<<2)
111 #define THM_TSPC	0x0e
112 #define THM_PPEC	0x10
113 #define THM_CTA		0x12
114 #define THM_PTA		0x14
115 #define   PTA_SLOPE_MASK	(0xff00)
116 #define   PTA_SLOPE_SHIFT	8
117 #define   PTA_OFFSET_MASK	(0x00ff)
118 #define THM_MGTA	0x16
119 #define   MGTA_SLOPE_MASK	(0xff00)
120 #define   MGTA_SLOPE_SHIFT	8
121 #define   MGTA_OFFSET_MASK	(0x00ff)
122 #define THM_TRC		0x1a
123 #define   TRC_CORE2_EN	(1<<15)
124 #define   TRC_THM_EN	(1<<12)
125 #define   TRC_C6_WAR	(1<<8)
126 #define   TRC_CORE1_EN	(1<<7)
127 #define   TRC_CORE_PWR	(1<<6)
128 #define   TRC_PCH_EN	(1<<5)
129 #define   TRC_MCH_EN	(1<<4)
130 #define   TRC_DIMM4	(1<<3)
131 #define   TRC_DIMM3	(1<<2)
132 #define   TRC_DIMM2	(1<<1)
133 #define   TRC_DIMM1	(1<<0)
134 #define THM_TES		0x20
135 #define THM_TEN		0x21
136 #define   TEN_UPDATE_EN	1
137 #define THM_PSC		0x24
138 #define   PSC_NTG	(1<<0) /* No GFX turbo support */
139 #define   PSC_NTPC	(1<<1) /* No CPU turbo support */
140 #define   PSC_PP_DEF	(0<<2) /* Perf policy up to driver */
141 #define   PSP_PP_PC	(1<<2) /* BIOS prefers CPU perf */
142 #define   PSP_PP_BAL	(2<<2) /* BIOS wants balanced perf */
143 #define   PSP_PP_GFX	(3<<2) /* BIOS prefers GFX perf */
144 #define   PSP_PBRT	(1<<4) /* BIOS run time support */
145 #define THM_CTV1	0x30
146 #define   CTV_TEMP_ERROR (1<<15)
147 #define   CTV_TEMP_MASK	0x3f
148 #define   CTV_
149 #define THM_CTV2	0x32
150 #define THM_CEC		0x34 /* undocumented power accumulator in joules */
151 #define THM_AE		0x3f
152 #define THM_HTS		0x50 /* 32 bits */
153 #define   HTS_PCPL_MASK	(0x7fe00000)
154 #define   HTS_PCPL_SHIFT 21
155 #define   HTS_GPL_MASK  (0x001ff000)
156 #define   HTS_GPL_SHIFT 12
157 #define   HTS_PP_MASK	(0x00000c00)
158 #define   HTS_PP_SHIFT  10
159 #define   HTS_PP_DEF	0
160 #define   HTS_PP_PROC	1
161 #define   HTS_PP_BAL	2
162 #define   HTS_PP_GFX	3
163 #define   HTS_PCTD_DIS	(1<<9)
164 #define   HTS_GTD_DIS	(1<<8)
165 #define   HTS_PTL_MASK  (0x000000fe)
166 #define   HTS_PTL_SHIFT 1
167 #define   HTS_NVV	(1<<0)
168 #define THM_HTSHI	0x54 /* 16 bits */
169 #define   HTS2_PPL_MASK		(0x03ff)
170 #define   HTS2_PRST_MASK	(0x3c00)
171 #define   HTS2_PRST_SHIFT	10
172 #define   HTS2_PRST_UNLOADED	0
173 #define   HTS2_PRST_RUNNING	1
174 #define   HTS2_PRST_TDISOP	2 /* turbo disabled due to power */
175 #define   HTS2_PRST_TDISHT	3 /* turbo disabled due to high temp */
176 #define   HTS2_PRST_TDISUSR	4 /* user disabled turbo */
177 #define   HTS2_PRST_TDISPLAT	5 /* platform disabled turbo */
178 #define   HTS2_PRST_TDISPM	6 /* power management disabled turbo */
179 #define   HTS2_PRST_TDISERR	7 /* some kind of error disabled turbo */
180 #define THM_PTL		0x56
181 #define THM_MGTV	0x58
182 #define   TV_MASK	0x000000000000ff00
183 #define   TV_SHIFT	8
184 #define THM_PTV		0x60
185 #define   PTV_MASK	0x00ff
186 #define THM_MMGPC	0x64
187 #define THM_MPPC	0x66
188 #define THM_MPCPC	0x68
189 #define THM_TSPIEN	0x82
190 #define   TSPIEN_AUX_LOHI	(1<<0)
191 #define   TSPIEN_HOT_LOHI	(1<<1)
192 #define   TSPIEN_CRIT_LOHI	(1<<2)
193 #define   TSPIEN_AUX2_LOHI	(1<<3)
194 #define THM_TSLOCK	0x83
195 #define THM_ATR		0x84
196 #define THM_TOF		0x87
197 #define THM_STS		0x98
198 #define   STS_PCPL_MASK		(0x7fe00000)
199 #define   STS_PCPL_SHIFT	21
200 #define   STS_GPL_MASK		(0x001ff000)
201 #define   STS_GPL_SHIFT		12
202 #define   STS_PP_MASK		(0x00000c00)
203 #define   STS_PP_SHIFT		10
204 #define   STS_PP_DEF		0
205 #define   STS_PP_PROC		1
206 #define   STS_PP_BAL		2
207 #define   STS_PP_GFX		3
208 #define   STS_PCTD_DIS		(1<<9)
209 #define   STS_GTD_DIS		(1<<8)
210 #define   STS_PTL_MASK		(0x000000fe)
211 #define   STS_PTL_SHIFT		1
212 #define   STS_NVV		(1<<0)
213 #define THM_SEC		0x9c
214 #define   SEC_ACK	(1<<0)
215 #define THM_TC3		0xa4
216 #define THM_TC1		0xa8
217 #define   STS_PPL_MASK		(0x0003ff00)
218 #define   STS_PPL_SHIFT		16
219 #define THM_TC2		0xac
220 #define THM_DTV		0xb0
221 #define THM_ITV		0xd8
222 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
223 #define   ITV_ME_SEQNO_SHIFT (16)
224 #define   ITV_MCH_TEMP_MASK 0x0000ff00
225 #define   ITV_MCH_TEMP_SHIFT (8)
226 #define   ITV_PCH_TEMP_MASK 0x000000ff
227 
228 #define thm_readb(off) readb(ips->regmap + (off))
229 #define thm_readw(off) readw(ips->regmap + (off))
230 #define thm_readl(off) readl(ips->regmap + (off))
231 #define thm_readq(off) readq(ips->regmap + (off))
232 
233 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
234 #define thm_writew(off, val) writew((val), ips->regmap + (off))
235 #define thm_writel(off, val) writel((val), ips->regmap + (off))
236 
237 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
238 static bool late_i915_load = false;
239 
240 /* For initial average collection */
241 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
242 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
243 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
244 
245 /* Per-SKU limits */
246 struct ips_mcp_limits {
247 	int mcp_power_limit; /* mW units */
248 	int core_power_limit;
249 	int mch_power_limit;
250 	int core_temp_limit; /* degrees C */
251 	int mch_temp_limit;
252 };
253 
254 /* Max temps are -10 degrees C to avoid PROCHOT# */
255 
256 static struct ips_mcp_limits ips_sv_limits = {
257 	.mcp_power_limit = 35000,
258 	.core_power_limit = 29000,
259 	.mch_power_limit = 20000,
260 	.core_temp_limit = 95,
261 	.mch_temp_limit = 90
262 };
263 
264 static struct ips_mcp_limits ips_lv_limits = {
265 	.mcp_power_limit = 25000,
266 	.core_power_limit = 21000,
267 	.mch_power_limit = 13000,
268 	.core_temp_limit = 95,
269 	.mch_temp_limit = 90
270 };
271 
272 static struct ips_mcp_limits ips_ulv_limits = {
273 	.mcp_power_limit = 18000,
274 	.core_power_limit = 14000,
275 	.mch_power_limit = 11000,
276 	.core_temp_limit = 95,
277 	.mch_temp_limit = 90
278 };
279 
280 struct ips_driver {
281 	struct device *dev;
282 	void __iomem *regmap;
283 	int irq;
284 
285 	struct task_struct *monitor;
286 	struct task_struct *adjust;
287 	struct dentry *debug_root;
288 	struct timer_list timer;
289 
290 	/* Average CPU core temps (all averages in .01 degrees C for precision) */
291 	u16 ctv1_avg_temp;
292 	u16 ctv2_avg_temp;
293 	/* GMCH average */
294 	u16 mch_avg_temp;
295 	/* Average for the CPU (both cores?) */
296 	u16 mcp_avg_temp;
297 	/* Average power consumption (in mW) */
298 	u32 cpu_avg_power;
299 	u32 mch_avg_power;
300 
301 	/* Offset values */
302 	u16 cta_val;
303 	u16 pta_val;
304 	u16 mgta_val;
305 
306 	/* Maximums & prefs, protected by turbo status lock */
307 	spinlock_t turbo_status_lock;
308 	u16 mcp_temp_limit;
309 	u16 mcp_power_limit;
310 	u16 core_power_limit;
311 	u16 mch_power_limit;
312 	bool cpu_turbo_enabled;
313 	bool __cpu_turbo_on;
314 	bool gpu_turbo_enabled;
315 	bool __gpu_turbo_on;
316 	bool gpu_preferred;
317 	bool poll_turbo_status;
318 	bool second_cpu;
319 	bool turbo_toggle_allowed;
320 	struct ips_mcp_limits *limits;
321 
322 	/* Optional MCH interfaces for if i915 is in use */
323 	unsigned long (*read_mch_val)(void);
324 	bool (*gpu_raise)(void);
325 	bool (*gpu_lower)(void);
326 	bool (*gpu_busy)(void);
327 	bool (*gpu_turbo_disable)(void);
328 
329 	/* For restoration at unload */
330 	u64 orig_turbo_limit;
331 	u64 orig_turbo_ratios;
332 };
333 
334 static bool
335 ips_gpu_turbo_enabled(struct ips_driver *ips);
336 
337 /**
338  * ips_cpu_busy - is CPU busy?
339  * @ips: IPS driver struct
340  *
341  * Check CPU for load to see whether we should increase its thermal budget.
342  *
343  * RETURNS:
344  * True if the CPU could use more power, false otherwise.
345  */
346 static bool ips_cpu_busy(struct ips_driver *ips)
347 {
348 	if ((avenrun[0] >> FSHIFT) > 1)
349 		return true;
350 
351 	return false;
352 }
353 
354 /**
355  * ips_cpu_raise - raise CPU power clamp
356  * @ips: IPS driver struct
357  *
358  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
359  * this platform.
360  *
361  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
362  * long as we haven't hit the TDP limit for the SKU).
363  */
364 static void ips_cpu_raise(struct ips_driver *ips)
365 {
366 	u64 turbo_override;
367 	u16 cur_tdp_limit, new_tdp_limit;
368 
369 	if (!ips->cpu_turbo_enabled)
370 		return;
371 
372 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
373 
374 	cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
375 	new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
376 
377 	/* Clamp to SKU TDP limit */
378 	if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
379 		new_tdp_limit = cur_tdp_limit;
380 
381 	thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
382 
383 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
384 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
385 
386 	turbo_override &= ~TURBO_TDP_MASK;
387 	turbo_override |= new_tdp_limit;
388 
389 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
390 }
391 
392 /**
393  * ips_cpu_lower - lower CPU power clamp
394  * @ips: IPS driver struct
395  *
396  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
397  *
398  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
399  * as low as the platform limits will allow (though we could go lower there
400  * wouldn't be much point).
401  */
402 static void ips_cpu_lower(struct ips_driver *ips)
403 {
404 	u64 turbo_override;
405 	u16 cur_limit, new_limit;
406 
407 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
408 
409 	cur_limit = turbo_override & TURBO_TDP_MASK;
410 	new_limit = cur_limit - 8; /* 1W decrease */
411 
412 	/* Clamp to SKU TDP limit */
413 	if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
414 		new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
415 
416 	thm_writew(THM_MPCPC, (new_limit * 10) / 8);
417 
418 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
419 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
420 
421 	turbo_override &= ~TURBO_TDP_MASK;
422 	turbo_override |= new_limit;
423 
424 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
425 }
426 
427 /**
428  * do_enable_cpu_turbo - internal turbo enable function
429  * @data: unused
430  *
431  * Internal function for actually updating MSRs.  When we enable/disable
432  * turbo, we need to do it on each CPU; this function is the one called
433  * by on_each_cpu() when needed.
434  */
435 static void do_enable_cpu_turbo(void *data)
436 {
437 	u64 perf_ctl;
438 
439 	rdmsrl(IA32_PERF_CTL, perf_ctl);
440 	if (perf_ctl & IA32_PERF_TURBO_DIS) {
441 		perf_ctl &= ~IA32_PERF_TURBO_DIS;
442 		wrmsrl(IA32_PERF_CTL, perf_ctl);
443 	}
444 }
445 
446 /**
447  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
448  * @ips: IPS driver struct
449  *
450  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
451  * all logical threads.
452  */
453 static void ips_enable_cpu_turbo(struct ips_driver *ips)
454 {
455 	/* Already on, no need to mess with MSRs */
456 	if (ips->__cpu_turbo_on)
457 		return;
458 
459 	if (ips->turbo_toggle_allowed)
460 		on_each_cpu(do_enable_cpu_turbo, ips, 1);
461 
462 	ips->__cpu_turbo_on = true;
463 }
464 
465 /**
466  * do_disable_cpu_turbo - internal turbo disable function
467  * @data: unused
468  *
469  * Internal function for actually updating MSRs.  When we enable/disable
470  * turbo, we need to do it on each CPU; this function is the one called
471  * by on_each_cpu() when needed.
472  */
473 static void do_disable_cpu_turbo(void *data)
474 {
475 	u64 perf_ctl;
476 
477 	rdmsrl(IA32_PERF_CTL, perf_ctl);
478 	if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
479 		perf_ctl |= IA32_PERF_TURBO_DIS;
480 		wrmsrl(IA32_PERF_CTL, perf_ctl);
481 	}
482 }
483 
484 /**
485  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
486  * @ips: IPS driver struct
487  *
488  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
489  * all logical threads.
490  */
491 static void ips_disable_cpu_turbo(struct ips_driver *ips)
492 {
493 	/* Already off, leave it */
494 	if (!ips->__cpu_turbo_on)
495 		return;
496 
497 	if (ips->turbo_toggle_allowed)
498 		on_each_cpu(do_disable_cpu_turbo, ips, 1);
499 
500 	ips->__cpu_turbo_on = false;
501 }
502 
503 /**
504  * ips_gpu_busy - is GPU busy?
505  * @ips: IPS driver struct
506  *
507  * Check GPU for load to see whether we should increase its thermal budget.
508  * We need to call into the i915 driver in this case.
509  *
510  * RETURNS:
511  * True if the GPU could use more power, false otherwise.
512  */
513 static bool ips_gpu_busy(struct ips_driver *ips)
514 {
515 	if (!ips_gpu_turbo_enabled(ips))
516 		return false;
517 
518 	return ips->gpu_busy();
519 }
520 
521 /**
522  * ips_gpu_raise - raise GPU power clamp
523  * @ips: IPS driver struct
524  *
525  * Raise the GPU frequency/power if possible.  We need to call into the
526  * i915 driver in this case.
527  */
528 static void ips_gpu_raise(struct ips_driver *ips)
529 {
530 	if (!ips_gpu_turbo_enabled(ips))
531 		return;
532 
533 	if (!ips->gpu_raise())
534 		ips->gpu_turbo_enabled = false;
535 
536 	return;
537 }
538 
539 /**
540  * ips_gpu_lower - lower GPU power clamp
541  * @ips: IPS driver struct
542  *
543  * Lower GPU frequency/power if possible.  Need to call i915.
544  */
545 static void ips_gpu_lower(struct ips_driver *ips)
546 {
547 	if (!ips_gpu_turbo_enabled(ips))
548 		return;
549 
550 	if (!ips->gpu_lower())
551 		ips->gpu_turbo_enabled = false;
552 
553 	return;
554 }
555 
556 /**
557  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
558  * @ips: IPS driver struct
559  *
560  * Call into the graphics driver indicating that it can safely use
561  * turbo mode.
562  */
563 static void ips_enable_gpu_turbo(struct ips_driver *ips)
564 {
565 	if (ips->__gpu_turbo_on)
566 		return;
567 	ips->__gpu_turbo_on = true;
568 }
569 
570 /**
571  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
572  * @ips: IPS driver struct
573  *
574  * Request that the graphics driver disable turbo mode.
575  */
576 static void ips_disable_gpu_turbo(struct ips_driver *ips)
577 {
578 	/* Avoid calling i915 if turbo is already disabled */
579 	if (!ips->__gpu_turbo_on)
580 		return;
581 
582 	if (!ips->gpu_turbo_disable())
583 		dev_err(ips->dev, "failed to disable graphics turbo\n");
584 	else
585 		ips->__gpu_turbo_on = false;
586 }
587 
588 /**
589  * mcp_exceeded - check whether we're outside our thermal & power limits
590  * @ips: IPS driver struct
591  *
592  * Check whether the MCP is over its thermal or power budget.
593  *
594  * Returns: %true if the temp or power has exceeded its maximum, else %false
595  */
596 static bool mcp_exceeded(struct ips_driver *ips)
597 {
598 	unsigned long flags;
599 	bool ret = false;
600 	u32 temp_limit;
601 	u32 avg_power;
602 
603 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
604 
605 	temp_limit = ips->mcp_temp_limit * 100;
606 	if (ips->mcp_avg_temp > temp_limit)
607 		ret = true;
608 
609 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
610 	if (avg_power > ips->mcp_power_limit)
611 		ret = true;
612 
613 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
614 
615 	return ret;
616 }
617 
618 /**
619  * cpu_exceeded - check whether a CPU core is outside its limits
620  * @ips: IPS driver struct
621  * @cpu: CPU number to check
622  *
623  * Check a given CPU's average temp or power is over its limit.
624  *
625  * Returns: %true if the temp or power has exceeded its maximum, else %false
626  */
627 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
628 {
629 	unsigned long flags;
630 	int avg;
631 	bool ret = false;
632 
633 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
634 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
635 	if (avg > (ips->limits->core_temp_limit * 100))
636 		ret = true;
637 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
638 		ret = true;
639 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
640 
641 	if (ret)
642 		dev_info(ips->dev, "CPU power or thermal limit exceeded\n");
643 
644 	return ret;
645 }
646 
647 /**
648  * mch_exceeded - check whether the GPU is over budget
649  * @ips: IPS driver struct
650  *
651  * Check the MCH temp & power against their maximums.
652  *
653  * Returns: %true if the temp or power has exceeded its maximum, else %false
654  */
655 static bool mch_exceeded(struct ips_driver *ips)
656 {
657 	unsigned long flags;
658 	bool ret = false;
659 
660 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
661 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
662 		ret = true;
663 	if (ips->mch_avg_power > ips->mch_power_limit)
664 		ret = true;
665 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
666 
667 	return ret;
668 }
669 
670 /**
671  * verify_limits - verify BIOS provided limits
672  * @ips: IPS structure
673  *
674  * BIOS can optionally provide non-default limits for power and temp.  Check
675  * them here and use the defaults if the BIOS values are not provided or
676  * are otherwise unusable.
677  */
678 static void verify_limits(struct ips_driver *ips)
679 {
680 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
681 	    ips->mcp_power_limit > 35000)
682 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
683 
684 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
685 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
686 	    ips->mcp_temp_limit > 150)
687 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
688 					  ips->limits->mch_temp_limit);
689 }
690 
691 /**
692  * update_turbo_limits - get various limits & settings from regs
693  * @ips: IPS driver struct
694  *
695  * Update the IPS power & temp limits, along with turbo enable flags,
696  * based on latest register contents.
697  *
698  * Used at init time and for runtime BIOS support, which requires polling
699  * the regs for updates (as a result of AC->DC transition for example).
700  *
701  * LOCKING:
702  * Caller must hold turbo_status_lock (outside of init)
703  */
704 static void update_turbo_limits(struct ips_driver *ips)
705 {
706 	u32 hts = thm_readl(THM_HTS);
707 
708 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
709 	/*
710 	 * Disable turbo for now, until we can figure out why the power figures
711 	 * are wrong
712 	 */
713 	ips->cpu_turbo_enabled = false;
714 
715 	if (ips->gpu_busy)
716 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
717 
718 	ips->core_power_limit = thm_readw(THM_MPCPC);
719 	ips->mch_power_limit = thm_readw(THM_MMGPC);
720 	ips->mcp_temp_limit = thm_readw(THM_PTL);
721 	ips->mcp_power_limit = thm_readw(THM_MPPC);
722 
723 	verify_limits(ips);
724 	/* Ignore BIOS CPU vs GPU pref */
725 }
726 
727 /**
728  * ips_adjust - adjust power clamp based on thermal state
729  * @data: ips driver structure
730  *
731  * Wake up every 5s or so and check whether we should adjust the power clamp.
732  * Check CPU and GPU load to determine which needs adjustment.  There are
733  * several things to consider here:
734  *   - do we need to adjust up or down?
735  *   - is CPU busy?
736  *   - is GPU busy?
737  *   - is CPU in turbo?
738  *   - is GPU in turbo?
739  *   - is CPU or GPU preferred? (CPU is default)
740  *
741  * So, given the above, we do the following:
742  *   - up (TDP available)
743  *     - CPU not busy, GPU not busy - nothing
744  *     - CPU busy, GPU not busy - adjust CPU up
745  *     - CPU not busy, GPU busy - adjust GPU up
746  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
747  *       non-preferred unit if necessary
748  *   - down (at TDP limit)
749  *     - adjust both CPU and GPU down if possible
750  *
751  *              |cpu+ gpu+      cpu+gpu-        cpu-gpu+        cpu-gpu-
752  * cpu < gpu <  |cpu+gpu+       cpu+            gpu+            nothing
753  * cpu < gpu >= |cpu+gpu-(mcp<) cpu+gpu-(mcp<)  gpu-            gpu-
754  * cpu >= gpu < |cpu-gpu+(mcp<) cpu-            cpu-gpu+(mcp<)  cpu-
755  * cpu >= gpu >=|cpu-gpu-       cpu-gpu-        cpu-gpu-        cpu-gpu-
756  *
757  * Returns: %0
758  */
759 static int ips_adjust(void *data)
760 {
761 	struct ips_driver *ips = data;
762 	unsigned long flags;
763 
764 	dev_dbg(ips->dev, "starting ips-adjust thread\n");
765 
766 	/*
767 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
768 	 * often isn't recommended due to ME interaction.
769 	 */
770 	do {
771 		bool cpu_busy = ips_cpu_busy(ips);
772 		bool gpu_busy = ips_gpu_busy(ips);
773 
774 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
775 		if (ips->poll_turbo_status)
776 			update_turbo_limits(ips);
777 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
778 
779 		/* Update turbo status if necessary */
780 		if (ips->cpu_turbo_enabled)
781 			ips_enable_cpu_turbo(ips);
782 		else
783 			ips_disable_cpu_turbo(ips);
784 
785 		if (ips->gpu_turbo_enabled)
786 			ips_enable_gpu_turbo(ips);
787 		else
788 			ips_disable_gpu_turbo(ips);
789 
790 		/* We're outside our comfort zone, crank them down */
791 		if (mcp_exceeded(ips)) {
792 			ips_cpu_lower(ips);
793 			ips_gpu_lower(ips);
794 			goto sleep;
795 		}
796 
797 		if (!cpu_exceeded(ips, 0) && cpu_busy)
798 			ips_cpu_raise(ips);
799 		else
800 			ips_cpu_lower(ips);
801 
802 		if (!mch_exceeded(ips) && gpu_busy)
803 			ips_gpu_raise(ips);
804 		else
805 			ips_gpu_lower(ips);
806 
807 sleep:
808 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
809 	} while (!kthread_should_stop());
810 
811 	dev_dbg(ips->dev, "ips-adjust thread stopped\n");
812 
813 	return 0;
814 }
815 
816 /*
817  * Helpers for reading out temp/power values and calculating their
818  * averages for the decision making and monitoring functions.
819  */
820 
821 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
822 {
823 	u64 total = 0;
824 	int i;
825 	u16 avg;
826 
827 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
828 		total += (u64)(array[i] * 100);
829 
830 	do_div(total, IPS_SAMPLE_COUNT);
831 
832 	avg = (u16)total;
833 
834 	return avg;
835 }
836 
837 static u16 read_mgtv(struct ips_driver *ips)
838 {
839 	u16 __maybe_unused ret;
840 	u64 slope, offset;
841 	u64 val;
842 
843 	val = thm_readq(THM_MGTV);
844 	val = (val & TV_MASK) >> TV_SHIFT;
845 
846 	slope = offset = thm_readw(THM_MGTA);
847 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
848 	offset = offset & MGTA_OFFSET_MASK;
849 
850 	ret = ((val * slope + 0x40) >> 7) + offset;
851 
852 	return 0; /* MCH temp reporting buggy */
853 }
854 
855 static u16 read_ptv(struct ips_driver *ips)
856 {
857 	u16 val;
858 
859 	val = thm_readw(THM_PTV) & PTV_MASK;
860 
861 	return val;
862 }
863 
864 static u16 read_ctv(struct ips_driver *ips, int cpu)
865 {
866 	int reg = cpu ? THM_CTV2 : THM_CTV1;
867 	u16 val;
868 
869 	val = thm_readw(reg);
870 	if (!(val & CTV_TEMP_ERROR))
871 		val = (val) >> 6; /* discard fractional component */
872 	else
873 		val = 0;
874 
875 	return val;
876 }
877 
878 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
879 {
880 	u32 val;
881 	u32 ret;
882 
883 	/*
884 	 * CEC is in joules/65535.  Take difference over time to
885 	 * get watts.
886 	 */
887 	val = thm_readl(THM_CEC);
888 
889 	/* period is in ms and we want mW */
890 	ret = (((val - *last) * 1000) / period);
891 	ret = (ret * 1000) / 65535;
892 	*last = val;
893 
894 	return 0;
895 }
896 
897 static const u16 temp_decay_factor = 2;
898 static u16 update_average_temp(u16 avg, u16 val)
899 {
900 	u16 ret;
901 
902 	/* Multiply by 100 for extra precision */
903 	ret = (val * 100 / temp_decay_factor) +
904 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
905 	return ret;
906 }
907 
908 static const u16 power_decay_factor = 2;
909 static u16 update_average_power(u32 avg, u32 val)
910 {
911 	u32 ret;
912 
913 	ret = (val / power_decay_factor) +
914 		(((power_decay_factor - 1) * avg) / power_decay_factor);
915 
916 	return ret;
917 }
918 
919 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
920 {
921 	u64 total = 0;
922 	u32 avg;
923 	int i;
924 
925 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
926 		total += array[i];
927 
928 	do_div(total, IPS_SAMPLE_COUNT);
929 	avg = (u32)total;
930 
931 	return avg;
932 }
933 
934 static void monitor_timeout(struct timer_list *t)
935 {
936 	struct ips_driver *ips = from_timer(ips, t, timer);
937 	wake_up_process(ips->monitor);
938 }
939 
940 /**
941  * ips_monitor - temp/power monitoring thread
942  * @data: ips driver structure
943  *
944  * This is the main function for the IPS driver.  It monitors power and
945  * temperature in the MCP and adjusts CPU and GPU power clamps accordingly.
946  *
947  * We keep a 5s moving average of power consumption and temperature.  Using
948  * that data, along with CPU vs GPU preference, we adjust the power clamps
949  * up or down.
950  *
951  * Returns: %0 on success or -errno on error
952  */
953 static int ips_monitor(void *data)
954 {
955 	struct ips_driver *ips = data;
956 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
957 	int i;
958 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
959 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
960 	u8 cur_seqno, last_seqno;
961 
962 	mcp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
963 	ctv1_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
964 	ctv2_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
965 	mch_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
966 	cpu_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
967 	mchp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
968 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
969 			!cpu_samples || !mchp_samples) {
970 		dev_err(ips->dev,
971 			"failed to allocate sample array, ips disabled\n");
972 		kfree(mcp_samples);
973 		kfree(ctv1_samples);
974 		kfree(ctv2_samples);
975 		kfree(mch_samples);
976 		kfree(cpu_samples);
977 		kfree(mchp_samples);
978 		return -ENOMEM;
979 	}
980 
981 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
982 		ITV_ME_SEQNO_SHIFT;
983 	seqno_timestamp = get_jiffies_64();
984 
985 	old_cpu_power = thm_readl(THM_CEC);
986 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
987 
988 	/* Collect an initial average */
989 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
990 		u32 mchp, cpu_power;
991 		u16 val;
992 
993 		mcp_samples[i] = read_ptv(ips);
994 
995 		val = read_ctv(ips, 0);
996 		ctv1_samples[i] = val;
997 
998 		val = read_ctv(ips, 1);
999 		ctv2_samples[i] = val;
1000 
1001 		val = read_mgtv(ips);
1002 		mch_samples[i] = val;
1003 
1004 		cpu_power = get_cpu_power(ips, &old_cpu_power,
1005 					  IPS_SAMPLE_PERIOD);
1006 		cpu_samples[i] = cpu_power;
1007 
1008 		if (ips->read_mch_val) {
1009 			mchp = ips->read_mch_val();
1010 			mchp_samples[i] = mchp;
1011 		}
1012 
1013 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1014 		if (kthread_should_stop())
1015 			break;
1016 	}
1017 
1018 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1019 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1020 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1021 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1022 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1023 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1024 	kfree(mcp_samples);
1025 	kfree(ctv1_samples);
1026 	kfree(ctv2_samples);
1027 	kfree(mch_samples);
1028 	kfree(cpu_samples);
1029 	kfree(mchp_samples);
1030 
1031 	/* Start the adjustment thread now that we have data */
1032 	wake_up_process(ips->adjust);
1033 
1034 	/*
1035 	 * Ok, now we have an initial avg.  From here on out, we track the
1036 	 * running avg using a decaying average calculation.  This allows
1037 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1038 	 */
1039 	old_cpu_power = thm_readl(THM_CEC);
1040 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1041 	last_sample_period = IPS_SAMPLE_PERIOD;
1042 
1043 	timer_setup(&ips->timer, monitor_timeout, TIMER_DEFERRABLE);
1044 	do {
1045 		u32 cpu_val, mch_val;
1046 		u16 val;
1047 
1048 		/* MCP itself */
1049 		val = read_ptv(ips);
1050 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1051 
1052 		/* Processor 0 */
1053 		val = read_ctv(ips, 0);
1054 		ips->ctv1_avg_temp =
1055 			update_average_temp(ips->ctv1_avg_temp, val);
1056 		/* Power */
1057 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1058 					last_sample_period);
1059 		ips->cpu_avg_power =
1060 			update_average_power(ips->cpu_avg_power, cpu_val);
1061 
1062 		if (ips->second_cpu) {
1063 			/* Processor 1 */
1064 			val = read_ctv(ips, 1);
1065 			ips->ctv2_avg_temp =
1066 				update_average_temp(ips->ctv2_avg_temp, val);
1067 		}
1068 
1069 		/* MCH */
1070 		val = read_mgtv(ips);
1071 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1072 		/* Power */
1073 		if (ips->read_mch_val) {
1074 			mch_val = ips->read_mch_val();
1075 			ips->mch_avg_power =
1076 				update_average_power(ips->mch_avg_power,
1077 						     mch_val);
1078 		}
1079 
1080 		/*
1081 		 * Make sure ME is updating thermal regs.
1082 		 * Note:
1083 		 * If it's been more than a second since the last update,
1084 		 * the ME is probably hung.
1085 		 */
1086 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1087 			ITV_ME_SEQNO_SHIFT;
1088 		if (cur_seqno == last_seqno &&
1089 		    time_after(jiffies, seqno_timestamp + HZ)) {
1090 			dev_warn(ips->dev,
1091 				 "ME failed to update for more than 1s, likely hung\n");
1092 		} else {
1093 			seqno_timestamp = get_jiffies_64();
1094 			last_seqno = cur_seqno;
1095 		}
1096 
1097 		last_msecs = jiffies_to_msecs(jiffies);
1098 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1099 
1100 		__set_current_state(TASK_INTERRUPTIBLE);
1101 		mod_timer(&ips->timer, expire);
1102 		schedule();
1103 
1104 		/* Calculate actual sample period for power averaging */
1105 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1106 		if (!last_sample_period)
1107 			last_sample_period = 1;
1108 	} while (!kthread_should_stop());
1109 
1110 	del_timer_sync(&ips->timer);
1111 
1112 	dev_dbg(ips->dev, "ips-monitor thread stopped\n");
1113 
1114 	return 0;
1115 }
1116 
1117 /**
1118  * ips_irq_handler - handle temperature triggers and other IPS events
1119  * @irq: irq number
1120  * @arg: unused
1121  *
1122  * Handle temperature limit trigger events, generally by lowering the clamps.
1123  * If we're at a critical limit, we clamp back to the lowest possible value
1124  * to prevent emergency shutdown.
1125  *
1126  * Returns: IRQ_NONE or IRQ_HANDLED
1127  */
1128 static irqreturn_t ips_irq_handler(int irq, void *arg)
1129 {
1130 	struct ips_driver *ips = arg;
1131 	u8 tses = thm_readb(THM_TSES);
1132 	u8 tes = thm_readb(THM_TES);
1133 
1134 	if (!tses && !tes)
1135 		return IRQ_NONE;
1136 
1137 	dev_info(ips->dev, "TSES: 0x%02x\n", tses);
1138 	dev_info(ips->dev, "TES: 0x%02x\n", tes);
1139 
1140 	/* STS update from EC? */
1141 	if (tes & 1) {
1142 		u32 sts, tc1;
1143 
1144 		sts = thm_readl(THM_STS);
1145 		tc1 = thm_readl(THM_TC1);
1146 
1147 		if (sts & STS_NVV) {
1148 			spin_lock(&ips->turbo_status_lock);
1149 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1150 				STS_PCPL_SHIFT;
1151 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1152 				STS_GPL_SHIFT;
1153 			/* ignore EC CPU vs GPU pref */
1154 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1155 			/*
1156 			 * Disable turbo for now, until we can figure
1157 			 * out why the power figures are wrong
1158 			 */
1159 			ips->cpu_turbo_enabled = false;
1160 			if (ips->gpu_busy)
1161 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1162 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1163 				STS_PTL_SHIFT;
1164 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1165 				STS_PPL_SHIFT;
1166 			verify_limits(ips);
1167 			spin_unlock(&ips->turbo_status_lock);
1168 
1169 			thm_writeb(THM_SEC, SEC_ACK);
1170 		}
1171 		thm_writeb(THM_TES, tes);
1172 	}
1173 
1174 	/* Thermal trip */
1175 	if (tses) {
1176 		dev_warn(ips->dev, "thermal trip occurred, tses: 0x%04x\n",
1177 			 tses);
1178 		thm_writeb(THM_TSES, tses);
1179 	}
1180 
1181 	return IRQ_HANDLED;
1182 }
1183 
1184 #ifndef CONFIG_DEBUG_FS
1185 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1186 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1187 #else
1188 
1189 /* Expose current state and limits in debugfs if possible */
1190 
1191 static int cpu_temp_show(struct seq_file *m, void *data)
1192 {
1193 	struct ips_driver *ips = m->private;
1194 
1195 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1196 		   ips->ctv1_avg_temp % 100);
1197 
1198 	return 0;
1199 }
1200 DEFINE_SHOW_ATTRIBUTE(cpu_temp);
1201 
1202 static int cpu_power_show(struct seq_file *m, void *data)
1203 {
1204 	struct ips_driver *ips = m->private;
1205 
1206 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1207 
1208 	return 0;
1209 }
1210 DEFINE_SHOW_ATTRIBUTE(cpu_power);
1211 
1212 static int cpu_clamp_show(struct seq_file *m, void *data)
1213 {
1214 	u64 turbo_override;
1215 	int tdp, tdc;
1216 
1217 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1218 
1219 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1220 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1221 
1222 	/* Convert to .1W/A units */
1223 	tdp = tdp * 10 / 8;
1224 	tdc = tdc * 10 / 8;
1225 
1226 	/* Watts Amperes */
1227 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1228 		   tdc / 10, tdc % 10);
1229 
1230 	return 0;
1231 }
1232 DEFINE_SHOW_ATTRIBUTE(cpu_clamp);
1233 
1234 static int mch_temp_show(struct seq_file *m, void *data)
1235 {
1236 	struct ips_driver *ips = m->private;
1237 
1238 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1239 		   ips->mch_avg_temp % 100);
1240 
1241 	return 0;
1242 }
1243 DEFINE_SHOW_ATTRIBUTE(mch_temp);
1244 
1245 static int mch_power_show(struct seq_file *m, void *data)
1246 {
1247 	struct ips_driver *ips = m->private;
1248 
1249 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1250 
1251 	return 0;
1252 }
1253 DEFINE_SHOW_ATTRIBUTE(mch_power);
1254 
1255 static void ips_debugfs_cleanup(struct ips_driver *ips)
1256 {
1257 	debugfs_remove_recursive(ips->debug_root);
1258 }
1259 
1260 static void ips_debugfs_init(struct ips_driver *ips)
1261 {
1262 	ips->debug_root = debugfs_create_dir("ips", NULL);
1263 
1264 	debugfs_create_file("cpu_temp", 0444, ips->debug_root, ips, &cpu_temp_fops);
1265 	debugfs_create_file("cpu_power", 0444, ips->debug_root, ips, &cpu_power_fops);
1266 	debugfs_create_file("cpu_clamp", 0444, ips->debug_root, ips, &cpu_clamp_fops);
1267 	debugfs_create_file("mch_temp", 0444, ips->debug_root, ips, &mch_temp_fops);
1268 	debugfs_create_file("mch_power", 0444, ips->debug_root, ips, &mch_power_fops);
1269 }
1270 #endif /* CONFIG_DEBUG_FS */
1271 
1272 /**
1273  * ips_detect_cpu - detect whether CPU supports IPS
1274  * @ips: IPS driver struct
1275  *
1276  * Walk our list and see if we're on a supported CPU.  If we find one,
1277  * return the limits for it.
1278  *
1279  * Returns: the &ips_mcp_limits struct that matches the boot CPU or %NULL
1280  */
1281 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1282 {
1283 	u64 turbo_power, misc_en;
1284 	struct ips_mcp_limits *limits = NULL;
1285 	u16 tdp;
1286 
1287 	if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1288 		dev_info(ips->dev, "Non-IPS CPU detected.\n");
1289 		return NULL;
1290 	}
1291 
1292 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1293 	/*
1294 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1295 	 * turbo manually or we'll get an illegal MSR access, even though
1296 	 * turbo will still be available.
1297 	 */
1298 	if (misc_en & IA32_MISC_TURBO_EN)
1299 		ips->turbo_toggle_allowed = true;
1300 	else
1301 		ips->turbo_toggle_allowed = false;
1302 
1303 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1304 		limits = &ips_sv_limits;
1305 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1306 		limits = &ips_lv_limits;
1307 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1308 		limits = &ips_ulv_limits;
1309 	else {
1310 		dev_info(ips->dev, "No CPUID match found.\n");
1311 		return NULL;
1312 	}
1313 
1314 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1315 	tdp = turbo_power & TURBO_TDP_MASK;
1316 
1317 	/* Sanity check TDP against CPU */
1318 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1319 		dev_info(ips->dev,
1320 			 "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1321 			 tdp / 8, limits->core_power_limit / 1000);
1322 		limits->core_power_limit = (tdp / 8) * 1000;
1323 	}
1324 
1325 	return limits;
1326 }
1327 
1328 /**
1329  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1330  * @ips: IPS driver
1331  *
1332  * The i915 driver exports several interfaces to allow the IPS driver to
1333  * monitor and control graphics turbo mode.  If we can find them, we can
1334  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1335  * thermal and power limits in the MCP.
1336  *
1337  * Returns: %true if the required symbols are found, else %false
1338  */
1339 static bool ips_get_i915_syms(struct ips_driver *ips)
1340 {
1341 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1342 	if (!ips->read_mch_val)
1343 		goto out_err;
1344 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1345 	if (!ips->gpu_raise)
1346 		goto out_put_mch;
1347 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1348 	if (!ips->gpu_lower)
1349 		goto out_put_raise;
1350 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1351 	if (!ips->gpu_busy)
1352 		goto out_put_lower;
1353 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1354 	if (!ips->gpu_turbo_disable)
1355 		goto out_put_busy;
1356 
1357 	return true;
1358 
1359 out_put_busy:
1360 	symbol_put(i915_gpu_busy);
1361 out_put_lower:
1362 	symbol_put(i915_gpu_lower);
1363 out_put_raise:
1364 	symbol_put(i915_gpu_raise);
1365 out_put_mch:
1366 	symbol_put(i915_read_mch_val);
1367 out_err:
1368 	return false;
1369 }
1370 
1371 static bool
1372 ips_gpu_turbo_enabled(struct ips_driver *ips)
1373 {
1374 	if (!ips->gpu_busy && late_i915_load) {
1375 		if (ips_get_i915_syms(ips)) {
1376 			dev_info(ips->dev,
1377 				 "i915 driver attached, reenabling gpu turbo\n");
1378 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1379 		}
1380 	}
1381 
1382 	return ips->gpu_turbo_enabled;
1383 }
1384 
1385 void
1386 ips_link_to_i915_driver(void)
1387 {
1388 	/* We can't cleanly get at the various ips_driver structs from
1389 	 * this caller (the i915 driver), so just set a flag saying
1390 	 * that it's time to try getting the symbols again.
1391 	 */
1392 	late_i915_load = true;
1393 }
1394 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1395 
1396 static const struct pci_device_id ips_id_table[] = {
1397 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1398 	{ 0, }
1399 };
1400 
1401 MODULE_DEVICE_TABLE(pci, ips_id_table);
1402 
1403 static int ips_blacklist_callback(const struct dmi_system_id *id)
1404 {
1405 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1406 	return 1;
1407 }
1408 
1409 static const struct dmi_system_id ips_blacklist[] = {
1410 	{
1411 		.callback = ips_blacklist_callback,
1412 		.ident = "HP ProBook",
1413 		.matches = {
1414 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1415 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1416 		},
1417 	},
1418 	{ }	/* terminating entry */
1419 };
1420 
1421 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1422 {
1423 	u64 platform_info;
1424 	struct ips_driver *ips;
1425 	u32 hts;
1426 	int ret = 0;
1427 	u16 htshi, trc, trc_required_mask;
1428 	u8 tse;
1429 
1430 	if (dmi_check_system(ips_blacklist))
1431 		return -ENODEV;
1432 
1433 	ips = devm_kzalloc(&dev->dev, sizeof(*ips), GFP_KERNEL);
1434 	if (!ips)
1435 		return -ENOMEM;
1436 
1437 	spin_lock_init(&ips->turbo_status_lock);
1438 	ips->dev = &dev->dev;
1439 
1440 	ips->limits = ips_detect_cpu(ips);
1441 	if (!ips->limits) {
1442 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1443 		return -ENXIO;
1444 	}
1445 
1446 	ret = pcim_enable_device(dev);
1447 	if (ret) {
1448 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1449 		return ret;
1450 	}
1451 
1452 	ret = pcim_iomap_regions(dev, 1 << 0, pci_name(dev));
1453 	if (ret) {
1454 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1455 		return ret;
1456 	}
1457 	ips->regmap = pcim_iomap_table(dev)[0];
1458 
1459 	pci_set_drvdata(dev, ips);
1460 
1461 	tse = thm_readb(THM_TSE);
1462 	if (tse != TSE_EN) {
1463 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1464 		return -ENXIO;
1465 	}
1466 
1467 	trc = thm_readw(THM_TRC);
1468 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1469 	if ((trc & trc_required_mask) != trc_required_mask) {
1470 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1471 		return -ENXIO;
1472 	}
1473 
1474 	if (trc & TRC_CORE2_EN)
1475 		ips->second_cpu = true;
1476 
1477 	update_turbo_limits(ips);
1478 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1479 		ips->mcp_power_limit / 10);
1480 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1481 		ips->core_power_limit / 10);
1482 	/* BIOS may update limits at runtime */
1483 	if (thm_readl(THM_PSC) & PSP_PBRT)
1484 		ips->poll_turbo_status = true;
1485 
1486 	if (!ips_get_i915_syms(ips)) {
1487 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1488 		ips->gpu_turbo_enabled = false;
1489 	} else {
1490 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1491 		ips->gpu_turbo_enabled = true;
1492 	}
1493 
1494 	/*
1495 	 * Check PLATFORM_INFO MSR to make sure this chip is
1496 	 * turbo capable.
1497 	 */
1498 	rdmsrl(PLATFORM_INFO, platform_info);
1499 	if (!(platform_info & PLATFORM_TDP)) {
1500 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1501 		return -ENODEV;
1502 	}
1503 
1504 	/*
1505 	 * IRQ handler for ME interaction
1506 	 * Note: don't use MSI here as the PCH has bugs.
1507 	 */
1508 	ret = pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_LEGACY);
1509 	if (ret < 0)
1510 		return ret;
1511 
1512 	ips->irq = pci_irq_vector(dev, 0);
1513 
1514 	ret = request_irq(ips->irq, ips_irq_handler, IRQF_SHARED, "ips", ips);
1515 	if (ret) {
1516 		dev_err(&dev->dev, "request irq failed, aborting\n");
1517 		return ret;
1518 	}
1519 
1520 	/* Enable aux, hot & critical interrupts */
1521 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1522 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1523 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1524 
1525 	/* Collect adjustment values */
1526 	ips->cta_val = thm_readw(THM_CTA);
1527 	ips->pta_val = thm_readw(THM_PTA);
1528 	ips->mgta_val = thm_readw(THM_MGTA);
1529 
1530 	/* Save turbo limits & ratios */
1531 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1532 
1533 	ips_disable_cpu_turbo(ips);
1534 	ips->cpu_turbo_enabled = false;
1535 
1536 	/* Create thermal adjust thread */
1537 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1538 	if (IS_ERR(ips->adjust)) {
1539 		dev_err(&dev->dev,
1540 			"failed to create thermal adjust thread, aborting\n");
1541 		ret = -ENOMEM;
1542 		goto error_free_irq;
1543 
1544 	}
1545 
1546 	/*
1547 	 * Set up the work queue and monitor thread. The monitor thread
1548 	 * will wake up ips_adjust thread.
1549 	 */
1550 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1551 	if (IS_ERR(ips->monitor)) {
1552 		dev_err(&dev->dev,
1553 			"failed to create thermal monitor thread, aborting\n");
1554 		ret = -ENOMEM;
1555 		goto error_thread_cleanup;
1556 	}
1557 
1558 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1559 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1560 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1561 
1562 	thm_writew(THM_HTSHI, htshi);
1563 	thm_writel(THM_HTS, hts);
1564 
1565 	ips_debugfs_init(ips);
1566 
1567 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1568 		 ips->mcp_temp_limit);
1569 	return ret;
1570 
1571 error_thread_cleanup:
1572 	kthread_stop(ips->adjust);
1573 error_free_irq:
1574 	free_irq(ips->irq, ips);
1575 	pci_free_irq_vectors(dev);
1576 	return ret;
1577 }
1578 
1579 static void ips_remove(struct pci_dev *dev)
1580 {
1581 	struct ips_driver *ips = pci_get_drvdata(dev);
1582 	u64 turbo_override;
1583 
1584 	ips_debugfs_cleanup(ips);
1585 
1586 	/* Release i915 driver */
1587 	if (ips->read_mch_val)
1588 		symbol_put(i915_read_mch_val);
1589 	if (ips->gpu_raise)
1590 		symbol_put(i915_gpu_raise);
1591 	if (ips->gpu_lower)
1592 		symbol_put(i915_gpu_lower);
1593 	if (ips->gpu_busy)
1594 		symbol_put(i915_gpu_busy);
1595 	if (ips->gpu_turbo_disable)
1596 		symbol_put(i915_gpu_turbo_disable);
1597 
1598 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1599 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1600 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1601 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1602 
1603 	free_irq(ips->irq, ips);
1604 	pci_free_irq_vectors(dev);
1605 	if (ips->adjust)
1606 		kthread_stop(ips->adjust);
1607 	if (ips->monitor)
1608 		kthread_stop(ips->monitor);
1609 	dev_dbg(&dev->dev, "IPS driver removed\n");
1610 }
1611 
1612 static struct pci_driver ips_pci_driver = {
1613 	.name = "intel ips",
1614 	.id_table = ips_id_table,
1615 	.probe = ips_probe,
1616 	.remove = ips_remove,
1617 };
1618 
1619 module_pci_driver(ips_pci_driver);
1620 
1621 MODULE_LICENSE("GPL v2");
1622 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1623 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1624