xref: /openbmc/linux/drivers/platform/x86/intel_ips.c (revision 63705da3)
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 static bool mcp_exceeded(struct ips_driver *ips)
595 {
596 	unsigned long flags;
597 	bool ret = false;
598 	u32 temp_limit;
599 	u32 avg_power;
600 
601 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
602 
603 	temp_limit = ips->mcp_temp_limit * 100;
604 	if (ips->mcp_avg_temp > temp_limit)
605 		ret = true;
606 
607 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
608 	if (avg_power > ips->mcp_power_limit)
609 		ret = true;
610 
611 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
612 
613 	return ret;
614 }
615 
616 /**
617  * cpu_exceeded - check whether a CPU core is outside its limits
618  * @ips: IPS driver struct
619  * @cpu: CPU number to check
620  *
621  * Check a given CPU's average temp or power is over its limit.
622  */
623 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
624 {
625 	unsigned long flags;
626 	int avg;
627 	bool ret = false;
628 
629 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
630 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
631 	if (avg > (ips->limits->core_temp_limit * 100))
632 		ret = true;
633 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
634 		ret = true;
635 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
636 
637 	if (ret)
638 		dev_info(ips->dev, "CPU power or thermal limit exceeded\n");
639 
640 	return ret;
641 }
642 
643 /**
644  * mch_exceeded - check whether the GPU is over budget
645  * @ips: IPS driver struct
646  *
647  * Check the MCH temp & power against their maximums.
648  */
649 static bool mch_exceeded(struct ips_driver *ips)
650 {
651 	unsigned long flags;
652 	bool ret = false;
653 
654 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
655 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
656 		ret = true;
657 	if (ips->mch_avg_power > ips->mch_power_limit)
658 		ret = true;
659 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
660 
661 	return ret;
662 }
663 
664 /**
665  * verify_limits - verify BIOS provided limits
666  * @ips: IPS structure
667  *
668  * BIOS can optionally provide non-default limits for power and temp.  Check
669  * them here and use the defaults if the BIOS values are not provided or
670  * are otherwise unusable.
671  */
672 static void verify_limits(struct ips_driver *ips)
673 {
674 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
675 	    ips->mcp_power_limit > 35000)
676 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
677 
678 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
679 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
680 	    ips->mcp_temp_limit > 150)
681 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
682 					  ips->limits->mch_temp_limit);
683 }
684 
685 /**
686  * update_turbo_limits - get various limits & settings from regs
687  * @ips: IPS driver struct
688  *
689  * Update the IPS power & temp limits, along with turbo enable flags,
690  * based on latest register contents.
691  *
692  * Used at init time and for runtime BIOS support, which requires polling
693  * the regs for updates (as a result of AC->DC transition for example).
694  *
695  * LOCKING:
696  * Caller must hold turbo_status_lock (outside of init)
697  */
698 static void update_turbo_limits(struct ips_driver *ips)
699 {
700 	u32 hts = thm_readl(THM_HTS);
701 
702 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
703 	/*
704 	 * Disable turbo for now, until we can figure out why the power figures
705 	 * are wrong
706 	 */
707 	ips->cpu_turbo_enabled = false;
708 
709 	if (ips->gpu_busy)
710 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
711 
712 	ips->core_power_limit = thm_readw(THM_MPCPC);
713 	ips->mch_power_limit = thm_readw(THM_MMGPC);
714 	ips->mcp_temp_limit = thm_readw(THM_PTL);
715 	ips->mcp_power_limit = thm_readw(THM_MPPC);
716 
717 	verify_limits(ips);
718 	/* Ignore BIOS CPU vs GPU pref */
719 }
720 
721 /**
722  * ips_adjust - adjust power clamp based on thermal state
723  * @data: ips driver structure
724  *
725  * Wake up every 5s or so and check whether we should adjust the power clamp.
726  * Check CPU and GPU load to determine which needs adjustment.  There are
727  * several things to consider here:
728  *   - do we need to adjust up or down?
729  *   - is CPU busy?
730  *   - is GPU busy?
731  *   - is CPU in turbo?
732  *   - is GPU in turbo?
733  *   - is CPU or GPU preferred? (CPU is default)
734  *
735  * So, given the above, we do the following:
736  *   - up (TDP available)
737  *     - CPU not busy, GPU not busy - nothing
738  *     - CPU busy, GPU not busy - adjust CPU up
739  *     - CPU not busy, GPU busy - adjust GPU up
740  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
741  *       non-preferred unit if necessary
742  *   - down (at TDP limit)
743  *     - adjust both CPU and GPU down if possible
744  *
745 		cpu+ gpu+	cpu+gpu-	cpu-gpu+	cpu-gpu-
746 cpu < gpu <	cpu+gpu+	cpu+		gpu+		nothing
747 cpu < gpu >=	cpu+gpu-(mcp<)	cpu+gpu-(mcp<)	gpu-		gpu-
748 cpu >= gpu <	cpu-gpu+(mcp<)	cpu-		cpu-gpu+(mcp<)	cpu-
749 cpu >= gpu >=	cpu-gpu-	cpu-gpu-	cpu-gpu-	cpu-gpu-
750  *
751  */
752 static int ips_adjust(void *data)
753 {
754 	struct ips_driver *ips = data;
755 	unsigned long flags;
756 
757 	dev_dbg(ips->dev, "starting ips-adjust thread\n");
758 
759 	/*
760 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
761 	 * often isn't recommended due to ME interaction.
762 	 */
763 	do {
764 		bool cpu_busy = ips_cpu_busy(ips);
765 		bool gpu_busy = ips_gpu_busy(ips);
766 
767 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
768 		if (ips->poll_turbo_status)
769 			update_turbo_limits(ips);
770 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
771 
772 		/* Update turbo status if necessary */
773 		if (ips->cpu_turbo_enabled)
774 			ips_enable_cpu_turbo(ips);
775 		else
776 			ips_disable_cpu_turbo(ips);
777 
778 		if (ips->gpu_turbo_enabled)
779 			ips_enable_gpu_turbo(ips);
780 		else
781 			ips_disable_gpu_turbo(ips);
782 
783 		/* We're outside our comfort zone, crank them down */
784 		if (mcp_exceeded(ips)) {
785 			ips_cpu_lower(ips);
786 			ips_gpu_lower(ips);
787 			goto sleep;
788 		}
789 
790 		if (!cpu_exceeded(ips, 0) && cpu_busy)
791 			ips_cpu_raise(ips);
792 		else
793 			ips_cpu_lower(ips);
794 
795 		if (!mch_exceeded(ips) && gpu_busy)
796 			ips_gpu_raise(ips);
797 		else
798 			ips_gpu_lower(ips);
799 
800 sleep:
801 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
802 	} while (!kthread_should_stop());
803 
804 	dev_dbg(ips->dev, "ips-adjust thread stopped\n");
805 
806 	return 0;
807 }
808 
809 /*
810  * Helpers for reading out temp/power values and calculating their
811  * averages for the decision making and monitoring functions.
812  */
813 
814 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
815 {
816 	u64 total = 0;
817 	int i;
818 	u16 avg;
819 
820 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
821 		total += (u64)(array[i] * 100);
822 
823 	do_div(total, IPS_SAMPLE_COUNT);
824 
825 	avg = (u16)total;
826 
827 	return avg;
828 }
829 
830 static u16 read_mgtv(struct ips_driver *ips)
831 {
832 	u16 __maybe_unused ret;
833 	u64 slope, offset;
834 	u64 val;
835 
836 	val = thm_readq(THM_MGTV);
837 	val = (val & TV_MASK) >> TV_SHIFT;
838 
839 	slope = offset = thm_readw(THM_MGTA);
840 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
841 	offset = offset & MGTA_OFFSET_MASK;
842 
843 	ret = ((val * slope + 0x40) >> 7) + offset;
844 
845 	return 0; /* MCH temp reporting buggy */
846 }
847 
848 static u16 read_ptv(struct ips_driver *ips)
849 {
850 	u16 val;
851 
852 	val = thm_readw(THM_PTV) & PTV_MASK;
853 
854 	return val;
855 }
856 
857 static u16 read_ctv(struct ips_driver *ips, int cpu)
858 {
859 	int reg = cpu ? THM_CTV2 : THM_CTV1;
860 	u16 val;
861 
862 	val = thm_readw(reg);
863 	if (!(val & CTV_TEMP_ERROR))
864 		val = (val) >> 6; /* discard fractional component */
865 	else
866 		val = 0;
867 
868 	return val;
869 }
870 
871 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
872 {
873 	u32 val;
874 	u32 ret;
875 
876 	/*
877 	 * CEC is in joules/65535.  Take difference over time to
878 	 * get watts.
879 	 */
880 	val = thm_readl(THM_CEC);
881 
882 	/* period is in ms and we want mW */
883 	ret = (((val - *last) * 1000) / period);
884 	ret = (ret * 1000) / 65535;
885 	*last = val;
886 
887 	return 0;
888 }
889 
890 static const u16 temp_decay_factor = 2;
891 static u16 update_average_temp(u16 avg, u16 val)
892 {
893 	u16 ret;
894 
895 	/* Multiply by 100 for extra precision */
896 	ret = (val * 100 / temp_decay_factor) +
897 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
898 	return ret;
899 }
900 
901 static const u16 power_decay_factor = 2;
902 static u16 update_average_power(u32 avg, u32 val)
903 {
904 	u32 ret;
905 
906 	ret = (val / power_decay_factor) +
907 		(((power_decay_factor - 1) * avg) / power_decay_factor);
908 
909 	return ret;
910 }
911 
912 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
913 {
914 	u64 total = 0;
915 	u32 avg;
916 	int i;
917 
918 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
919 		total += array[i];
920 
921 	do_div(total, IPS_SAMPLE_COUNT);
922 	avg = (u32)total;
923 
924 	return avg;
925 }
926 
927 static void monitor_timeout(struct timer_list *t)
928 {
929 	struct ips_driver *ips = from_timer(ips, t, timer);
930 	wake_up_process(ips->monitor);
931 }
932 
933 /**
934  * ips_monitor - temp/power monitoring thread
935  * @data: ips driver structure
936  *
937  * This is the main function for the IPS driver.  It monitors power and
938  * tempurature in the MCP and adjusts CPU and GPU power clams accordingly.
939  *
940  * We keep a 5s moving average of power consumption and tempurature.  Using
941  * that data, along with CPU vs GPU preference, we adjust the power clamps
942  * up or down.
943  */
944 static int ips_monitor(void *data)
945 {
946 	struct ips_driver *ips = data;
947 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
948 	int i;
949 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
950 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
951 	u8 cur_seqno, last_seqno;
952 
953 	mcp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
954 	ctv1_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
955 	ctv2_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
956 	mch_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
957 	cpu_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
958 	mchp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
959 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
960 			!cpu_samples || !mchp_samples) {
961 		dev_err(ips->dev,
962 			"failed to allocate sample array, ips disabled\n");
963 		kfree(mcp_samples);
964 		kfree(ctv1_samples);
965 		kfree(ctv2_samples);
966 		kfree(mch_samples);
967 		kfree(cpu_samples);
968 		kfree(mchp_samples);
969 		return -ENOMEM;
970 	}
971 
972 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
973 		ITV_ME_SEQNO_SHIFT;
974 	seqno_timestamp = get_jiffies_64();
975 
976 	old_cpu_power = thm_readl(THM_CEC);
977 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
978 
979 	/* Collect an initial average */
980 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
981 		u32 mchp, cpu_power;
982 		u16 val;
983 
984 		mcp_samples[i] = read_ptv(ips);
985 
986 		val = read_ctv(ips, 0);
987 		ctv1_samples[i] = val;
988 
989 		val = read_ctv(ips, 1);
990 		ctv2_samples[i] = val;
991 
992 		val = read_mgtv(ips);
993 		mch_samples[i] = val;
994 
995 		cpu_power = get_cpu_power(ips, &old_cpu_power,
996 					  IPS_SAMPLE_PERIOD);
997 		cpu_samples[i] = cpu_power;
998 
999 		if (ips->read_mch_val) {
1000 			mchp = ips->read_mch_val();
1001 			mchp_samples[i] = mchp;
1002 		}
1003 
1004 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1005 		if (kthread_should_stop())
1006 			break;
1007 	}
1008 
1009 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1010 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1011 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1012 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1013 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1014 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1015 	kfree(mcp_samples);
1016 	kfree(ctv1_samples);
1017 	kfree(ctv2_samples);
1018 	kfree(mch_samples);
1019 	kfree(cpu_samples);
1020 	kfree(mchp_samples);
1021 
1022 	/* Start the adjustment thread now that we have data */
1023 	wake_up_process(ips->adjust);
1024 
1025 	/*
1026 	 * Ok, now we have an initial avg.  From here on out, we track the
1027 	 * running avg using a decaying average calculation.  This allows
1028 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1029 	 */
1030 	old_cpu_power = thm_readl(THM_CEC);
1031 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1032 	last_sample_period = IPS_SAMPLE_PERIOD;
1033 
1034 	timer_setup(&ips->timer, monitor_timeout, TIMER_DEFERRABLE);
1035 	do {
1036 		u32 cpu_val, mch_val;
1037 		u16 val;
1038 
1039 		/* MCP itself */
1040 		val = read_ptv(ips);
1041 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1042 
1043 		/* Processor 0 */
1044 		val = read_ctv(ips, 0);
1045 		ips->ctv1_avg_temp =
1046 			update_average_temp(ips->ctv1_avg_temp, val);
1047 		/* Power */
1048 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1049 					last_sample_period);
1050 		ips->cpu_avg_power =
1051 			update_average_power(ips->cpu_avg_power, cpu_val);
1052 
1053 		if (ips->second_cpu) {
1054 			/* Processor 1 */
1055 			val = read_ctv(ips, 1);
1056 			ips->ctv2_avg_temp =
1057 				update_average_temp(ips->ctv2_avg_temp, val);
1058 		}
1059 
1060 		/* MCH */
1061 		val = read_mgtv(ips);
1062 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1063 		/* Power */
1064 		if (ips->read_mch_val) {
1065 			mch_val = ips->read_mch_val();
1066 			ips->mch_avg_power =
1067 				update_average_power(ips->mch_avg_power,
1068 						     mch_val);
1069 		}
1070 
1071 		/*
1072 		 * Make sure ME is updating thermal regs.
1073 		 * Note:
1074 		 * If it's been more than a second since the last update,
1075 		 * the ME is probably hung.
1076 		 */
1077 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1078 			ITV_ME_SEQNO_SHIFT;
1079 		if (cur_seqno == last_seqno &&
1080 		    time_after(jiffies, seqno_timestamp + HZ)) {
1081 			dev_warn(ips->dev,
1082 				 "ME failed to update for more than 1s, likely hung\n");
1083 		} else {
1084 			seqno_timestamp = get_jiffies_64();
1085 			last_seqno = cur_seqno;
1086 		}
1087 
1088 		last_msecs = jiffies_to_msecs(jiffies);
1089 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1090 
1091 		__set_current_state(TASK_INTERRUPTIBLE);
1092 		mod_timer(&ips->timer, expire);
1093 		schedule();
1094 
1095 		/* Calculate actual sample period for power averaging */
1096 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1097 		if (!last_sample_period)
1098 			last_sample_period = 1;
1099 	} while (!kthread_should_stop());
1100 
1101 	del_timer_sync(&ips->timer);
1102 
1103 	dev_dbg(ips->dev, "ips-monitor thread stopped\n");
1104 
1105 	return 0;
1106 }
1107 
1108 #if 0
1109 #define THM_DUMPW(reg) \
1110 	{ \
1111 	u16 val = thm_readw(reg); \
1112 	dev_dbg(ips->dev, #reg ": 0x%04x\n", val); \
1113 	}
1114 #define THM_DUMPL(reg) \
1115 	{ \
1116 	u32 val = thm_readl(reg); \
1117 	dev_dbg(ips->dev, #reg ": 0x%08x\n", val); \
1118 	}
1119 #define THM_DUMPQ(reg) \
1120 	{ \
1121 	u64 val = thm_readq(reg); \
1122 	dev_dbg(ips->dev, #reg ": 0x%016x\n", val); \
1123 	}
1124 
1125 static void dump_thermal_info(struct ips_driver *ips)
1126 {
1127 	u16 ptl;
1128 
1129 	ptl = thm_readw(THM_PTL);
1130 	dev_dbg(ips->dev, "Processor temp limit: %d\n", ptl);
1131 
1132 	THM_DUMPW(THM_CTA);
1133 	THM_DUMPW(THM_TRC);
1134 	THM_DUMPW(THM_CTV1);
1135 	THM_DUMPL(THM_STS);
1136 	THM_DUMPW(THM_PTV);
1137 	THM_DUMPQ(THM_MGTV);
1138 }
1139 #endif
1140 
1141 /**
1142  * ips_irq_handler - handle temperature triggers and other IPS events
1143  * @irq: irq number
1144  * @arg: unused
1145  *
1146  * Handle temperature limit trigger events, generally by lowering the clamps.
1147  * If we're at a critical limit, we clamp back to the lowest possible value
1148  * to prevent emergency shutdown.
1149  */
1150 static irqreturn_t ips_irq_handler(int irq, void *arg)
1151 {
1152 	struct ips_driver *ips = arg;
1153 	u8 tses = thm_readb(THM_TSES);
1154 	u8 tes = thm_readb(THM_TES);
1155 
1156 	if (!tses && !tes)
1157 		return IRQ_NONE;
1158 
1159 	dev_info(ips->dev, "TSES: 0x%02x\n", tses);
1160 	dev_info(ips->dev, "TES: 0x%02x\n", tes);
1161 
1162 	/* STS update from EC? */
1163 	if (tes & 1) {
1164 		u32 sts, tc1;
1165 
1166 		sts = thm_readl(THM_STS);
1167 		tc1 = thm_readl(THM_TC1);
1168 
1169 		if (sts & STS_NVV) {
1170 			spin_lock(&ips->turbo_status_lock);
1171 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1172 				STS_PCPL_SHIFT;
1173 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1174 				STS_GPL_SHIFT;
1175 			/* ignore EC CPU vs GPU pref */
1176 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1177 			/*
1178 			 * Disable turbo for now, until we can figure
1179 			 * out why the power figures are wrong
1180 			 */
1181 			ips->cpu_turbo_enabled = false;
1182 			if (ips->gpu_busy)
1183 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1184 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1185 				STS_PTL_SHIFT;
1186 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1187 				STS_PPL_SHIFT;
1188 			verify_limits(ips);
1189 			spin_unlock(&ips->turbo_status_lock);
1190 
1191 			thm_writeb(THM_SEC, SEC_ACK);
1192 		}
1193 		thm_writeb(THM_TES, tes);
1194 	}
1195 
1196 	/* Thermal trip */
1197 	if (tses) {
1198 		dev_warn(ips->dev, "thermal trip occurred, tses: 0x%04x\n",
1199 			 tses);
1200 		thm_writeb(THM_TSES, tses);
1201 	}
1202 
1203 	return IRQ_HANDLED;
1204 }
1205 
1206 #ifndef CONFIG_DEBUG_FS
1207 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1208 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1209 #else
1210 
1211 /* Expose current state and limits in debugfs if possible */
1212 
1213 static int cpu_temp_show(struct seq_file *m, void *data)
1214 {
1215 	struct ips_driver *ips = m->private;
1216 
1217 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1218 		   ips->ctv1_avg_temp % 100);
1219 
1220 	return 0;
1221 }
1222 DEFINE_SHOW_ATTRIBUTE(cpu_temp);
1223 
1224 static int cpu_power_show(struct seq_file *m, void *data)
1225 {
1226 	struct ips_driver *ips = m->private;
1227 
1228 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1229 
1230 	return 0;
1231 }
1232 DEFINE_SHOW_ATTRIBUTE(cpu_power);
1233 
1234 static int cpu_clamp_show(struct seq_file *m, void *data)
1235 {
1236 	u64 turbo_override;
1237 	int tdp, tdc;
1238 
1239 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1240 
1241 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1242 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1243 
1244 	/* Convert to .1W/A units */
1245 	tdp = tdp * 10 / 8;
1246 	tdc = tdc * 10 / 8;
1247 
1248 	/* Watts Amperes */
1249 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1250 		   tdc / 10, tdc % 10);
1251 
1252 	return 0;
1253 }
1254 DEFINE_SHOW_ATTRIBUTE(cpu_clamp);
1255 
1256 static int mch_temp_show(struct seq_file *m, void *data)
1257 {
1258 	struct ips_driver *ips = m->private;
1259 
1260 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1261 		   ips->mch_avg_temp % 100);
1262 
1263 	return 0;
1264 }
1265 DEFINE_SHOW_ATTRIBUTE(mch_temp);
1266 
1267 static int mch_power_show(struct seq_file *m, void *data)
1268 {
1269 	struct ips_driver *ips = m->private;
1270 
1271 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1272 
1273 	return 0;
1274 }
1275 DEFINE_SHOW_ATTRIBUTE(mch_power);
1276 
1277 static void ips_debugfs_cleanup(struct ips_driver *ips)
1278 {
1279 	debugfs_remove_recursive(ips->debug_root);
1280 }
1281 
1282 static void ips_debugfs_init(struct ips_driver *ips)
1283 {
1284 	ips->debug_root = debugfs_create_dir("ips", NULL);
1285 
1286 	debugfs_create_file("cpu_temp", 0444, ips->debug_root, ips, &cpu_temp_fops);
1287 	debugfs_create_file("cpu_power", 0444, ips->debug_root, ips, &cpu_power_fops);
1288 	debugfs_create_file("cpu_clamp", 0444, ips->debug_root, ips, &cpu_clamp_fops);
1289 	debugfs_create_file("mch_temp", 0444, ips->debug_root, ips, &mch_temp_fops);
1290 	debugfs_create_file("mch_power", 0444, ips->debug_root, ips, &mch_power_fops);
1291 }
1292 #endif /* CONFIG_DEBUG_FS */
1293 
1294 /**
1295  * ips_detect_cpu - detect whether CPU supports IPS
1296  *
1297  * Walk our list and see if we're on a supported CPU.  If we find one,
1298  * return the limits for it.
1299  */
1300 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1301 {
1302 	u64 turbo_power, misc_en;
1303 	struct ips_mcp_limits *limits = NULL;
1304 	u16 tdp;
1305 
1306 	if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1307 		dev_info(ips->dev, "Non-IPS CPU detected.\n");
1308 		return NULL;
1309 	}
1310 
1311 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1312 	/*
1313 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1314 	 * turbo manually or we'll get an illegal MSR access, even though
1315 	 * turbo will still be available.
1316 	 */
1317 	if (misc_en & IA32_MISC_TURBO_EN)
1318 		ips->turbo_toggle_allowed = true;
1319 	else
1320 		ips->turbo_toggle_allowed = false;
1321 
1322 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1323 		limits = &ips_sv_limits;
1324 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1325 		limits = &ips_lv_limits;
1326 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1327 		limits = &ips_ulv_limits;
1328 	else {
1329 		dev_info(ips->dev, "No CPUID match found.\n");
1330 		return NULL;
1331 	}
1332 
1333 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1334 	tdp = turbo_power & TURBO_TDP_MASK;
1335 
1336 	/* Sanity check TDP against CPU */
1337 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1338 		dev_info(ips->dev,
1339 			 "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1340 			 tdp / 8, limits->core_power_limit / 1000);
1341 		limits->core_power_limit = (tdp / 8) * 1000;
1342 	}
1343 
1344 	return limits;
1345 }
1346 
1347 /**
1348  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1349  * @ips: IPS driver
1350  *
1351  * The i915 driver exports several interfaces to allow the IPS driver to
1352  * monitor and control graphics turbo mode.  If we can find them, we can
1353  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1354  * thermal and power limits in the MCP.
1355  */
1356 static bool ips_get_i915_syms(struct ips_driver *ips)
1357 {
1358 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1359 	if (!ips->read_mch_val)
1360 		goto out_err;
1361 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1362 	if (!ips->gpu_raise)
1363 		goto out_put_mch;
1364 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1365 	if (!ips->gpu_lower)
1366 		goto out_put_raise;
1367 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1368 	if (!ips->gpu_busy)
1369 		goto out_put_lower;
1370 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1371 	if (!ips->gpu_turbo_disable)
1372 		goto out_put_busy;
1373 
1374 	return true;
1375 
1376 out_put_busy:
1377 	symbol_put(i915_gpu_busy);
1378 out_put_lower:
1379 	symbol_put(i915_gpu_lower);
1380 out_put_raise:
1381 	symbol_put(i915_gpu_raise);
1382 out_put_mch:
1383 	symbol_put(i915_read_mch_val);
1384 out_err:
1385 	return false;
1386 }
1387 
1388 static bool
1389 ips_gpu_turbo_enabled(struct ips_driver *ips)
1390 {
1391 	if (!ips->gpu_busy && late_i915_load) {
1392 		if (ips_get_i915_syms(ips)) {
1393 			dev_info(ips->dev,
1394 				 "i915 driver attached, reenabling gpu turbo\n");
1395 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1396 		}
1397 	}
1398 
1399 	return ips->gpu_turbo_enabled;
1400 }
1401 
1402 void
1403 ips_link_to_i915_driver(void)
1404 {
1405 	/* We can't cleanly get at the various ips_driver structs from
1406 	 * this caller (the i915 driver), so just set a flag saying
1407 	 * that it's time to try getting the symbols again.
1408 	 */
1409 	late_i915_load = true;
1410 }
1411 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1412 
1413 static const struct pci_device_id ips_id_table[] = {
1414 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1415 	{ 0, }
1416 };
1417 
1418 MODULE_DEVICE_TABLE(pci, ips_id_table);
1419 
1420 static int ips_blacklist_callback(const struct dmi_system_id *id)
1421 {
1422 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1423 	return 1;
1424 }
1425 
1426 static const struct dmi_system_id ips_blacklist[] = {
1427 	{
1428 		.callback = ips_blacklist_callback,
1429 		.ident = "HP ProBook",
1430 		.matches = {
1431 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1432 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1433 		},
1434 	},
1435 	{ }	/* terminating entry */
1436 };
1437 
1438 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1439 {
1440 	u64 platform_info;
1441 	struct ips_driver *ips;
1442 	u32 hts;
1443 	int ret = 0;
1444 	u16 htshi, trc, trc_required_mask;
1445 	u8 tse;
1446 
1447 	if (dmi_check_system(ips_blacklist))
1448 		return -ENODEV;
1449 
1450 	ips = devm_kzalloc(&dev->dev, sizeof(*ips), GFP_KERNEL);
1451 	if (!ips)
1452 		return -ENOMEM;
1453 
1454 	spin_lock_init(&ips->turbo_status_lock);
1455 	ips->dev = &dev->dev;
1456 
1457 	ips->limits = ips_detect_cpu(ips);
1458 	if (!ips->limits) {
1459 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1460 		return -ENXIO;
1461 	}
1462 
1463 	ret = pcim_enable_device(dev);
1464 	if (ret) {
1465 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1466 		return ret;
1467 	}
1468 
1469 	ret = pcim_iomap_regions(dev, 1 << 0, pci_name(dev));
1470 	if (ret) {
1471 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1472 		return ret;
1473 	}
1474 	ips->regmap = pcim_iomap_table(dev)[0];
1475 
1476 	pci_set_drvdata(dev, ips);
1477 
1478 	tse = thm_readb(THM_TSE);
1479 	if (tse != TSE_EN) {
1480 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1481 		return -ENXIO;
1482 	}
1483 
1484 	trc = thm_readw(THM_TRC);
1485 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1486 	if ((trc & trc_required_mask) != trc_required_mask) {
1487 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1488 		return -ENXIO;
1489 	}
1490 
1491 	if (trc & TRC_CORE2_EN)
1492 		ips->second_cpu = true;
1493 
1494 	update_turbo_limits(ips);
1495 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1496 		ips->mcp_power_limit / 10);
1497 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1498 		ips->core_power_limit / 10);
1499 	/* BIOS may update limits at runtime */
1500 	if (thm_readl(THM_PSC) & PSP_PBRT)
1501 		ips->poll_turbo_status = true;
1502 
1503 	if (!ips_get_i915_syms(ips)) {
1504 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1505 		ips->gpu_turbo_enabled = false;
1506 	} else {
1507 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1508 		ips->gpu_turbo_enabled = true;
1509 	}
1510 
1511 	/*
1512 	 * Check PLATFORM_INFO MSR to make sure this chip is
1513 	 * turbo capable.
1514 	 */
1515 	rdmsrl(PLATFORM_INFO, platform_info);
1516 	if (!(platform_info & PLATFORM_TDP)) {
1517 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1518 		return -ENODEV;
1519 	}
1520 
1521 	/*
1522 	 * IRQ handler for ME interaction
1523 	 * Note: don't use MSI here as the PCH has bugs.
1524 	 */
1525 	ret = pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_LEGACY);
1526 	if (ret < 0)
1527 		return ret;
1528 
1529 	ips->irq = pci_irq_vector(dev, 0);
1530 
1531 	ret = request_irq(ips->irq, ips_irq_handler, IRQF_SHARED, "ips", ips);
1532 	if (ret) {
1533 		dev_err(&dev->dev, "request irq failed, aborting\n");
1534 		return ret;
1535 	}
1536 
1537 	/* Enable aux, hot & critical interrupts */
1538 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1539 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1540 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1541 
1542 	/* Collect adjustment values */
1543 	ips->cta_val = thm_readw(THM_CTA);
1544 	ips->pta_val = thm_readw(THM_PTA);
1545 	ips->mgta_val = thm_readw(THM_MGTA);
1546 
1547 	/* Save turbo limits & ratios */
1548 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1549 
1550 	ips_disable_cpu_turbo(ips);
1551 	ips->cpu_turbo_enabled = false;
1552 
1553 	/* Create thermal adjust thread */
1554 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1555 	if (IS_ERR(ips->adjust)) {
1556 		dev_err(&dev->dev,
1557 			"failed to create thermal adjust thread, aborting\n");
1558 		ret = -ENOMEM;
1559 		goto error_free_irq;
1560 
1561 	}
1562 
1563 	/*
1564 	 * Set up the work queue and monitor thread. The monitor thread
1565 	 * will wake up ips_adjust thread.
1566 	 */
1567 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1568 	if (IS_ERR(ips->monitor)) {
1569 		dev_err(&dev->dev,
1570 			"failed to create thermal monitor thread, aborting\n");
1571 		ret = -ENOMEM;
1572 		goto error_thread_cleanup;
1573 	}
1574 
1575 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1576 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1577 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1578 
1579 	thm_writew(THM_HTSHI, htshi);
1580 	thm_writel(THM_HTS, hts);
1581 
1582 	ips_debugfs_init(ips);
1583 
1584 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1585 		 ips->mcp_temp_limit);
1586 	return ret;
1587 
1588 error_thread_cleanup:
1589 	kthread_stop(ips->adjust);
1590 error_free_irq:
1591 	free_irq(ips->irq, ips);
1592 	pci_free_irq_vectors(dev);
1593 	return ret;
1594 }
1595 
1596 static void ips_remove(struct pci_dev *dev)
1597 {
1598 	struct ips_driver *ips = pci_get_drvdata(dev);
1599 	u64 turbo_override;
1600 
1601 	ips_debugfs_cleanup(ips);
1602 
1603 	/* Release i915 driver */
1604 	if (ips->read_mch_val)
1605 		symbol_put(i915_read_mch_val);
1606 	if (ips->gpu_raise)
1607 		symbol_put(i915_gpu_raise);
1608 	if (ips->gpu_lower)
1609 		symbol_put(i915_gpu_lower);
1610 	if (ips->gpu_busy)
1611 		symbol_put(i915_gpu_busy);
1612 	if (ips->gpu_turbo_disable)
1613 		symbol_put(i915_gpu_turbo_disable);
1614 
1615 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1616 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1617 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1618 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1619 
1620 	free_irq(ips->irq, ips);
1621 	pci_free_irq_vectors(dev);
1622 	if (ips->adjust)
1623 		kthread_stop(ips->adjust);
1624 	if (ips->monitor)
1625 		kthread_stop(ips->monitor);
1626 	dev_dbg(&dev->dev, "IPS driver removed\n");
1627 }
1628 
1629 static struct pci_driver ips_pci_driver = {
1630 	.name = "intel ips",
1631 	.id_table = ips_id_table,
1632 	.probe = ips_probe,
1633 	.remove = ips_remove,
1634 };
1635 
1636 module_pci_driver(ips_pci_driver);
1637 
1638 MODULE_LICENSE("GPL v2");
1639 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1640 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1641