xref: /openbmc/linux/drivers/platform/x86/intel_ips.c (revision a17922de)
1 /*
2  * Copyright (c) 2009-2010 Intel Corporation
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * The full GNU General Public License is included in this distribution in
14  * the file called "COPYING".
15  *
16  * Authors:
17  *	Jesse Barnes <jbarnes@virtuousgeek.org>
18  */
19 
20 /*
21  * Some Intel Ibex Peak based platforms support so-called "intelligent
22  * power sharing", which allows the CPU and GPU to cooperate to maximize
23  * performance within a given TDP (thermal design point).  This driver
24  * performs the coordination between the CPU and GPU, monitors thermal and
25  * power statistics in the platform, and initializes power monitoring
26  * hardware.  It also provides a few tunables to control behavior.  Its
27  * primary purpose is to safely allow CPU and GPU turbo modes to be enabled
28  * by tracking power and thermal budget; secondarily it can boost turbo
29  * performance by allocating more power or thermal budget to the CPU or GPU
30  * based on available headroom and activity.
31  *
32  * The basic algorithm is driven by a 5s moving average of temperature.  If
33  * thermal headroom is available, the CPU and/or GPU power clamps may be
34  * adjusted upwards.  If we hit the thermal ceiling or a thermal trigger,
35  * we scale back the clamp.  Aside from trigger events (when we're critically
36  * close or over our TDP) we don't adjust the clamps more than once every
37  * five seconds.
38  *
39  * The thermal device (device 31, function 6) has a set of registers that
40  * are updated by the ME firmware.  The ME should also take the clamp values
41  * written to those registers and write them to the CPU, but we currently
42  * bypass that functionality and write the CPU MSR directly.
43  *
44  * UNSUPPORTED:
45  *   - dual MCP configs
46  *
47  * TODO:
48  *   - handle CPU hotplug
49  *   - provide turbo enable/disable api
50  *
51  * Related documents:
52  *   - CDI 403777, 403778 - Auburndale EDS vol 1 & 2
53  *   - CDI 401376 - Ibex Peak EDS
54  *   - ref 26037, 26641 - IPS BIOS spec
55  *   - ref 26489 - Nehalem BIOS writer's guide
56  *   - ref 26921 - Ibex Peak BIOS Specification
57  */
58 
59 #include <linux/debugfs.h>
60 #include <linux/delay.h>
61 #include <linux/interrupt.h>
62 #include <linux/kernel.h>
63 #include <linux/kthread.h>
64 #include <linux/module.h>
65 #include <linux/pci.h>
66 #include <linux/sched.h>
67 #include <linux/sched/loadavg.h>
68 #include <linux/seq_file.h>
69 #include <linux/string.h>
70 #include <linux/tick.h>
71 #include <linux/timer.h>
72 #include <linux/dmi.h>
73 #include <drm/i915_drm.h>
74 #include <asm/msr.h>
75 #include <asm/processor.h>
76 #include "intel_ips.h"
77 
78 #include <linux/io-64-nonatomic-lo-hi.h>
79 
80 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
81 
82 /*
83  * Package level MSRs for monitor/control
84  */
85 #define PLATFORM_INFO	0xce
86 #define   PLATFORM_TDP		(1<<29)
87 #define   PLATFORM_RATIO	(1<<28)
88 
89 #define IA32_MISC_ENABLE	0x1a0
90 #define   IA32_MISC_TURBO_EN	(1ULL<<38)
91 
92 #define TURBO_POWER_CURRENT_LIMIT	0x1ac
93 #define   TURBO_TDC_OVR_EN	(1UL<<31)
94 #define   TURBO_TDC_MASK	(0x000000007fff0000UL)
95 #define   TURBO_TDC_SHIFT	(16)
96 #define   TURBO_TDP_OVR_EN	(1UL<<15)
97 #define   TURBO_TDP_MASK	(0x0000000000003fffUL)
98 
99 /*
100  * Core/thread MSRs for monitoring
101  */
102 #define IA32_PERF_CTL		0x199
103 #define   IA32_PERF_TURBO_DIS	(1ULL<<32)
104 
105 /*
106  * Thermal PCI device regs
107  */
108 #define THM_CFG_TBAR	0x10
109 #define THM_CFG_TBAR_HI	0x14
110 
111 #define THM_TSIU	0x00
112 #define THM_TSE		0x01
113 #define   TSE_EN	0xb8
114 #define THM_TSS		0x02
115 #define THM_TSTR	0x03
116 #define THM_TSTTP	0x04
117 #define THM_TSCO	0x08
118 #define THM_TSES	0x0c
119 #define THM_TSGPEN	0x0d
120 #define   TSGPEN_HOT_LOHI	(1<<1)
121 #define   TSGPEN_CRIT_LOHI	(1<<2)
122 #define THM_TSPC	0x0e
123 #define THM_PPEC	0x10
124 #define THM_CTA		0x12
125 #define THM_PTA		0x14
126 #define   PTA_SLOPE_MASK	(0xff00)
127 #define   PTA_SLOPE_SHIFT	8
128 #define   PTA_OFFSET_MASK	(0x00ff)
129 #define THM_MGTA	0x16
130 #define   MGTA_SLOPE_MASK	(0xff00)
131 #define   MGTA_SLOPE_SHIFT	8
132 #define   MGTA_OFFSET_MASK	(0x00ff)
133 #define THM_TRC		0x1a
134 #define   TRC_CORE2_EN	(1<<15)
135 #define   TRC_THM_EN	(1<<12)
136 #define   TRC_C6_WAR	(1<<8)
137 #define   TRC_CORE1_EN	(1<<7)
138 #define   TRC_CORE_PWR	(1<<6)
139 #define   TRC_PCH_EN	(1<<5)
140 #define   TRC_MCH_EN	(1<<4)
141 #define   TRC_DIMM4	(1<<3)
142 #define   TRC_DIMM3	(1<<2)
143 #define   TRC_DIMM2	(1<<1)
144 #define   TRC_DIMM1	(1<<0)
145 #define THM_TES		0x20
146 #define THM_TEN		0x21
147 #define   TEN_UPDATE_EN	1
148 #define THM_PSC		0x24
149 #define   PSC_NTG	(1<<0) /* No GFX turbo support */
150 #define   PSC_NTPC	(1<<1) /* No CPU turbo support */
151 #define   PSC_PP_DEF	(0<<2) /* Perf policy up to driver */
152 #define   PSP_PP_PC	(1<<2) /* BIOS prefers CPU perf */
153 #define   PSP_PP_BAL	(2<<2) /* BIOS wants balanced perf */
154 #define   PSP_PP_GFX	(3<<2) /* BIOS prefers GFX perf */
155 #define   PSP_PBRT	(1<<4) /* BIOS run time support */
156 #define THM_CTV1	0x30
157 #define   CTV_TEMP_ERROR (1<<15)
158 #define   CTV_TEMP_MASK	0x3f
159 #define   CTV_
160 #define THM_CTV2	0x32
161 #define THM_CEC		0x34 /* undocumented power accumulator in joules */
162 #define THM_AE		0x3f
163 #define THM_HTS		0x50 /* 32 bits */
164 #define   HTS_PCPL_MASK	(0x7fe00000)
165 #define   HTS_PCPL_SHIFT 21
166 #define   HTS_GPL_MASK  (0x001ff000)
167 #define   HTS_GPL_SHIFT 12
168 #define   HTS_PP_MASK	(0x00000c00)
169 #define   HTS_PP_SHIFT  10
170 #define   HTS_PP_DEF	0
171 #define   HTS_PP_PROC	1
172 #define   HTS_PP_BAL	2
173 #define   HTS_PP_GFX	3
174 #define   HTS_PCTD_DIS	(1<<9)
175 #define   HTS_GTD_DIS	(1<<8)
176 #define   HTS_PTL_MASK  (0x000000fe)
177 #define   HTS_PTL_SHIFT 1
178 #define   HTS_NVV	(1<<0)
179 #define THM_HTSHI	0x54 /* 16 bits */
180 #define   HTS2_PPL_MASK		(0x03ff)
181 #define   HTS2_PRST_MASK	(0x3c00)
182 #define   HTS2_PRST_SHIFT	10
183 #define   HTS2_PRST_UNLOADED	0
184 #define   HTS2_PRST_RUNNING	1
185 #define   HTS2_PRST_TDISOP	2 /* turbo disabled due to power */
186 #define   HTS2_PRST_TDISHT	3 /* turbo disabled due to high temp */
187 #define   HTS2_PRST_TDISUSR	4 /* user disabled turbo */
188 #define   HTS2_PRST_TDISPLAT	5 /* platform disabled turbo */
189 #define   HTS2_PRST_TDISPM	6 /* power management disabled turbo */
190 #define   HTS2_PRST_TDISERR	7 /* some kind of error disabled turbo */
191 #define THM_PTL		0x56
192 #define THM_MGTV	0x58
193 #define   TV_MASK	0x000000000000ff00
194 #define   TV_SHIFT	8
195 #define THM_PTV		0x60
196 #define   PTV_MASK	0x00ff
197 #define THM_MMGPC	0x64
198 #define THM_MPPC	0x66
199 #define THM_MPCPC	0x68
200 #define THM_TSPIEN	0x82
201 #define   TSPIEN_AUX_LOHI	(1<<0)
202 #define   TSPIEN_HOT_LOHI	(1<<1)
203 #define   TSPIEN_CRIT_LOHI	(1<<2)
204 #define   TSPIEN_AUX2_LOHI	(1<<3)
205 #define THM_TSLOCK	0x83
206 #define THM_ATR		0x84
207 #define THM_TOF		0x87
208 #define THM_STS		0x98
209 #define   STS_PCPL_MASK		(0x7fe00000)
210 #define   STS_PCPL_SHIFT	21
211 #define   STS_GPL_MASK		(0x001ff000)
212 #define   STS_GPL_SHIFT		12
213 #define   STS_PP_MASK		(0x00000c00)
214 #define   STS_PP_SHIFT		10
215 #define   STS_PP_DEF		0
216 #define   STS_PP_PROC		1
217 #define   STS_PP_BAL		2
218 #define   STS_PP_GFX		3
219 #define   STS_PCTD_DIS		(1<<9)
220 #define   STS_GTD_DIS		(1<<8)
221 #define   STS_PTL_MASK		(0x000000fe)
222 #define   STS_PTL_SHIFT		1
223 #define   STS_NVV		(1<<0)
224 #define THM_SEC		0x9c
225 #define   SEC_ACK	(1<<0)
226 #define THM_TC3		0xa4
227 #define THM_TC1		0xa8
228 #define   STS_PPL_MASK		(0x0003ff00)
229 #define   STS_PPL_SHIFT		16
230 #define THM_TC2		0xac
231 #define THM_DTV		0xb0
232 #define THM_ITV		0xd8
233 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
234 #define   ITV_ME_SEQNO_SHIFT (16)
235 #define   ITV_MCH_TEMP_MASK 0x0000ff00
236 #define   ITV_MCH_TEMP_SHIFT (8)
237 #define   ITV_PCH_TEMP_MASK 0x000000ff
238 
239 #define thm_readb(off) readb(ips->regmap + (off))
240 #define thm_readw(off) readw(ips->regmap + (off))
241 #define thm_readl(off) readl(ips->regmap + (off))
242 #define thm_readq(off) readq(ips->regmap + (off))
243 
244 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
245 #define thm_writew(off, val) writew((val), ips->regmap + (off))
246 #define thm_writel(off, val) writel((val), ips->regmap + (off))
247 
248 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
249 static bool late_i915_load = false;
250 
251 /* For initial average collection */
252 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
253 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
254 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
255 
256 /* Per-SKU limits */
257 struct ips_mcp_limits {
258 	int mcp_power_limit; /* mW units */
259 	int core_power_limit;
260 	int mch_power_limit;
261 	int core_temp_limit; /* degrees C */
262 	int mch_temp_limit;
263 };
264 
265 /* Max temps are -10 degrees C to avoid PROCHOT# */
266 
267 static struct ips_mcp_limits ips_sv_limits = {
268 	.mcp_power_limit = 35000,
269 	.core_power_limit = 29000,
270 	.mch_power_limit = 20000,
271 	.core_temp_limit = 95,
272 	.mch_temp_limit = 90
273 };
274 
275 static struct ips_mcp_limits ips_lv_limits = {
276 	.mcp_power_limit = 25000,
277 	.core_power_limit = 21000,
278 	.mch_power_limit = 13000,
279 	.core_temp_limit = 95,
280 	.mch_temp_limit = 90
281 };
282 
283 static struct ips_mcp_limits ips_ulv_limits = {
284 	.mcp_power_limit = 18000,
285 	.core_power_limit = 14000,
286 	.mch_power_limit = 11000,
287 	.core_temp_limit = 95,
288 	.mch_temp_limit = 90
289 };
290 
291 struct ips_driver {
292 	struct device *dev;
293 	void __iomem *regmap;
294 	int irq;
295 
296 	struct task_struct *monitor;
297 	struct task_struct *adjust;
298 	struct dentry *debug_root;
299 	struct timer_list timer;
300 
301 	/* Average CPU core temps (all averages in .01 degrees C for precision) */
302 	u16 ctv1_avg_temp;
303 	u16 ctv2_avg_temp;
304 	/* GMCH average */
305 	u16 mch_avg_temp;
306 	/* Average for the CPU (both cores?) */
307 	u16 mcp_avg_temp;
308 	/* Average power consumption (in mW) */
309 	u32 cpu_avg_power;
310 	u32 mch_avg_power;
311 
312 	/* Offset values */
313 	u16 cta_val;
314 	u16 pta_val;
315 	u16 mgta_val;
316 
317 	/* Maximums & prefs, protected by turbo status lock */
318 	spinlock_t turbo_status_lock;
319 	u16 mcp_temp_limit;
320 	u16 mcp_power_limit;
321 	u16 core_power_limit;
322 	u16 mch_power_limit;
323 	bool cpu_turbo_enabled;
324 	bool __cpu_turbo_on;
325 	bool gpu_turbo_enabled;
326 	bool __gpu_turbo_on;
327 	bool gpu_preferred;
328 	bool poll_turbo_status;
329 	bool second_cpu;
330 	bool turbo_toggle_allowed;
331 	struct ips_mcp_limits *limits;
332 
333 	/* Optional MCH interfaces for if i915 is in use */
334 	unsigned long (*read_mch_val)(void);
335 	bool (*gpu_raise)(void);
336 	bool (*gpu_lower)(void);
337 	bool (*gpu_busy)(void);
338 	bool (*gpu_turbo_disable)(void);
339 
340 	/* For restoration at unload */
341 	u64 orig_turbo_limit;
342 	u64 orig_turbo_ratios;
343 };
344 
345 static bool
346 ips_gpu_turbo_enabled(struct ips_driver *ips);
347 
348 /**
349  * ips_cpu_busy - is CPU busy?
350  * @ips: IPS driver struct
351  *
352  * Check CPU for load to see whether we should increase its thermal budget.
353  *
354  * RETURNS:
355  * True if the CPU could use more power, false otherwise.
356  */
357 static bool ips_cpu_busy(struct ips_driver *ips)
358 {
359 	if ((avenrun[0] >> FSHIFT) > 1)
360 		return true;
361 
362 	return false;
363 }
364 
365 /**
366  * ips_cpu_raise - raise CPU power clamp
367  * @ips: IPS driver struct
368  *
369  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
370  * this platform.
371  *
372  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
373  * long as we haven't hit the TDP limit for the SKU).
374  */
375 static void ips_cpu_raise(struct ips_driver *ips)
376 {
377 	u64 turbo_override;
378 	u16 cur_tdp_limit, new_tdp_limit;
379 
380 	if (!ips->cpu_turbo_enabled)
381 		return;
382 
383 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
384 
385 	cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
386 	new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
387 
388 	/* Clamp to SKU TDP limit */
389 	if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
390 		new_tdp_limit = cur_tdp_limit;
391 
392 	thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
393 
394 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
395 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
396 
397 	turbo_override &= ~TURBO_TDP_MASK;
398 	turbo_override |= new_tdp_limit;
399 
400 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
401 }
402 
403 /**
404  * ips_cpu_lower - lower CPU power clamp
405  * @ips: IPS driver struct
406  *
407  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
408  *
409  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
410  * as low as the platform limits will allow (though we could go lower there
411  * wouldn't be much point).
412  */
413 static void ips_cpu_lower(struct ips_driver *ips)
414 {
415 	u64 turbo_override;
416 	u16 cur_limit, new_limit;
417 
418 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
419 
420 	cur_limit = turbo_override & TURBO_TDP_MASK;
421 	new_limit = cur_limit - 8; /* 1W decrease */
422 
423 	/* Clamp to SKU TDP limit */
424 	if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
425 		new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
426 
427 	thm_writew(THM_MPCPC, (new_limit * 10) / 8);
428 
429 	turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
430 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
431 
432 	turbo_override &= ~TURBO_TDP_MASK;
433 	turbo_override |= new_limit;
434 
435 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
436 }
437 
438 /**
439  * do_enable_cpu_turbo - internal turbo enable function
440  * @data: unused
441  *
442  * Internal function for actually updating MSRs.  When we enable/disable
443  * turbo, we need to do it on each CPU; this function is the one called
444  * by on_each_cpu() when needed.
445  */
446 static void do_enable_cpu_turbo(void *data)
447 {
448 	u64 perf_ctl;
449 
450 	rdmsrl(IA32_PERF_CTL, perf_ctl);
451 	if (perf_ctl & IA32_PERF_TURBO_DIS) {
452 		perf_ctl &= ~IA32_PERF_TURBO_DIS;
453 		wrmsrl(IA32_PERF_CTL, perf_ctl);
454 	}
455 }
456 
457 /**
458  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
459  * @ips: IPS driver struct
460  *
461  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
462  * all logical threads.
463  */
464 static void ips_enable_cpu_turbo(struct ips_driver *ips)
465 {
466 	/* Already on, no need to mess with MSRs */
467 	if (ips->__cpu_turbo_on)
468 		return;
469 
470 	if (ips->turbo_toggle_allowed)
471 		on_each_cpu(do_enable_cpu_turbo, ips, 1);
472 
473 	ips->__cpu_turbo_on = true;
474 }
475 
476 /**
477  * do_disable_cpu_turbo - internal turbo disable function
478  * @data: unused
479  *
480  * Internal function for actually updating MSRs.  When we enable/disable
481  * turbo, we need to do it on each CPU; this function is the one called
482  * by on_each_cpu() when needed.
483  */
484 static void do_disable_cpu_turbo(void *data)
485 {
486 	u64 perf_ctl;
487 
488 	rdmsrl(IA32_PERF_CTL, perf_ctl);
489 	if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
490 		perf_ctl |= IA32_PERF_TURBO_DIS;
491 		wrmsrl(IA32_PERF_CTL, perf_ctl);
492 	}
493 }
494 
495 /**
496  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
497  * @ips: IPS driver struct
498  *
499  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
500  * all logical threads.
501  */
502 static void ips_disable_cpu_turbo(struct ips_driver *ips)
503 {
504 	/* Already off, leave it */
505 	if (!ips->__cpu_turbo_on)
506 		return;
507 
508 	if (ips->turbo_toggle_allowed)
509 		on_each_cpu(do_disable_cpu_turbo, ips, 1);
510 
511 	ips->__cpu_turbo_on = false;
512 }
513 
514 /**
515  * ips_gpu_busy - is GPU busy?
516  * @ips: IPS driver struct
517  *
518  * Check GPU for load to see whether we should increase its thermal budget.
519  * We need to call into the i915 driver in this case.
520  *
521  * RETURNS:
522  * True if the GPU could use more power, false otherwise.
523  */
524 static bool ips_gpu_busy(struct ips_driver *ips)
525 {
526 	if (!ips_gpu_turbo_enabled(ips))
527 		return false;
528 
529 	return ips->gpu_busy();
530 }
531 
532 /**
533  * ips_gpu_raise - raise GPU power clamp
534  * @ips: IPS driver struct
535  *
536  * Raise the GPU frequency/power if possible.  We need to call into the
537  * i915 driver in this case.
538  */
539 static void ips_gpu_raise(struct ips_driver *ips)
540 {
541 	if (!ips_gpu_turbo_enabled(ips))
542 		return;
543 
544 	if (!ips->gpu_raise())
545 		ips->gpu_turbo_enabled = false;
546 
547 	return;
548 }
549 
550 /**
551  * ips_gpu_lower - lower GPU power clamp
552  * @ips: IPS driver struct
553  *
554  * Lower GPU frequency/power if possible.  Need to call i915.
555  */
556 static void ips_gpu_lower(struct ips_driver *ips)
557 {
558 	if (!ips_gpu_turbo_enabled(ips))
559 		return;
560 
561 	if (!ips->gpu_lower())
562 		ips->gpu_turbo_enabled = false;
563 
564 	return;
565 }
566 
567 /**
568  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
569  * @ips: IPS driver struct
570  *
571  * Call into the graphics driver indicating that it can safely use
572  * turbo mode.
573  */
574 static void ips_enable_gpu_turbo(struct ips_driver *ips)
575 {
576 	if (ips->__gpu_turbo_on)
577 		return;
578 	ips->__gpu_turbo_on = true;
579 }
580 
581 /**
582  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
583  * @ips: IPS driver struct
584  *
585  * Request that the graphics driver disable turbo mode.
586  */
587 static void ips_disable_gpu_turbo(struct ips_driver *ips)
588 {
589 	/* Avoid calling i915 if turbo is already disabled */
590 	if (!ips->__gpu_turbo_on)
591 		return;
592 
593 	if (!ips->gpu_turbo_disable())
594 		dev_err(ips->dev, "failed to disable graphics turbo\n");
595 	else
596 		ips->__gpu_turbo_on = false;
597 }
598 
599 /**
600  * mcp_exceeded - check whether we're outside our thermal & power limits
601  * @ips: IPS driver struct
602  *
603  * Check whether the MCP is over its thermal or power budget.
604  */
605 static bool mcp_exceeded(struct ips_driver *ips)
606 {
607 	unsigned long flags;
608 	bool ret = false;
609 	u32 temp_limit;
610 	u32 avg_power;
611 
612 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
613 
614 	temp_limit = ips->mcp_temp_limit * 100;
615 	if (ips->mcp_avg_temp > temp_limit)
616 		ret = true;
617 
618 	avg_power = ips->cpu_avg_power + ips->mch_avg_power;
619 	if (avg_power > ips->mcp_power_limit)
620 		ret = true;
621 
622 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
623 
624 	return ret;
625 }
626 
627 /**
628  * cpu_exceeded - check whether a CPU core is outside its limits
629  * @ips: IPS driver struct
630  * @cpu: CPU number to check
631  *
632  * Check a given CPU's average temp or power is over its limit.
633  */
634 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
635 {
636 	unsigned long flags;
637 	int avg;
638 	bool ret = false;
639 
640 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
641 	avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
642 	if (avg > (ips->limits->core_temp_limit * 100))
643 		ret = true;
644 	if (ips->cpu_avg_power > ips->core_power_limit * 100)
645 		ret = true;
646 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
647 
648 	if (ret)
649 		dev_info(ips->dev, "CPU power or thermal limit exceeded\n");
650 
651 	return ret;
652 }
653 
654 /**
655  * mch_exceeded - check whether the GPU is over budget
656  * @ips: IPS driver struct
657  *
658  * Check the MCH temp & power against their maximums.
659  */
660 static bool mch_exceeded(struct ips_driver *ips)
661 {
662 	unsigned long flags;
663 	bool ret = false;
664 
665 	spin_lock_irqsave(&ips->turbo_status_lock, flags);
666 	if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
667 		ret = true;
668 	if (ips->mch_avg_power > ips->mch_power_limit)
669 		ret = true;
670 	spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
671 
672 	return ret;
673 }
674 
675 /**
676  * verify_limits - verify BIOS provided limits
677  * @ips: IPS structure
678  *
679  * BIOS can optionally provide non-default limits for power and temp.  Check
680  * them here and use the defaults if the BIOS values are not provided or
681  * are otherwise unusable.
682  */
683 static void verify_limits(struct ips_driver *ips)
684 {
685 	if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
686 	    ips->mcp_power_limit > 35000)
687 		ips->mcp_power_limit = ips->limits->mcp_power_limit;
688 
689 	if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
690 	    ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
691 	    ips->mcp_temp_limit > 150)
692 		ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
693 					  ips->limits->mch_temp_limit);
694 }
695 
696 /**
697  * update_turbo_limits - get various limits & settings from regs
698  * @ips: IPS driver struct
699  *
700  * Update the IPS power & temp limits, along with turbo enable flags,
701  * based on latest register contents.
702  *
703  * Used at init time and for runtime BIOS support, which requires polling
704  * the regs for updates (as a result of AC->DC transition for example).
705  *
706  * LOCKING:
707  * Caller must hold turbo_status_lock (outside of init)
708  */
709 static void update_turbo_limits(struct ips_driver *ips)
710 {
711 	u32 hts = thm_readl(THM_HTS);
712 
713 	ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
714 	/*
715 	 * Disable turbo for now, until we can figure out why the power figures
716 	 * are wrong
717 	 */
718 	ips->cpu_turbo_enabled = false;
719 
720 	if (ips->gpu_busy)
721 		ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
722 
723 	ips->core_power_limit = thm_readw(THM_MPCPC);
724 	ips->mch_power_limit = thm_readw(THM_MMGPC);
725 	ips->mcp_temp_limit = thm_readw(THM_PTL);
726 	ips->mcp_power_limit = thm_readw(THM_MPPC);
727 
728 	verify_limits(ips);
729 	/* Ignore BIOS CPU vs GPU pref */
730 }
731 
732 /**
733  * ips_adjust - adjust power clamp based on thermal state
734  * @data: ips driver structure
735  *
736  * Wake up every 5s or so and check whether we should adjust the power clamp.
737  * Check CPU and GPU load to determine which needs adjustment.  There are
738  * several things to consider here:
739  *   - do we need to adjust up or down?
740  *   - is CPU busy?
741  *   - is GPU busy?
742  *   - is CPU in turbo?
743  *   - is GPU in turbo?
744  *   - is CPU or GPU preferred? (CPU is default)
745  *
746  * So, given the above, we do the following:
747  *   - up (TDP available)
748  *     - CPU not busy, GPU not busy - nothing
749  *     - CPU busy, GPU not busy - adjust CPU up
750  *     - CPU not busy, GPU busy - adjust GPU up
751  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
752  *       non-preferred unit if necessary
753  *   - down (at TDP limit)
754  *     - adjust both CPU and GPU down if possible
755  *
756 		cpu+ gpu+	cpu+gpu-	cpu-gpu+	cpu-gpu-
757 cpu < gpu <	cpu+gpu+	cpu+		gpu+		nothing
758 cpu < gpu >=	cpu+gpu-(mcp<)	cpu+gpu-(mcp<)	gpu-		gpu-
759 cpu >= gpu <	cpu-gpu+(mcp<)	cpu-		cpu-gpu+(mcp<)	cpu-
760 cpu >= gpu >=	cpu-gpu-	cpu-gpu-	cpu-gpu-	cpu-gpu-
761  *
762  */
763 static int ips_adjust(void *data)
764 {
765 	struct ips_driver *ips = data;
766 	unsigned long flags;
767 
768 	dev_dbg(ips->dev, "starting ips-adjust thread\n");
769 
770 	/*
771 	 * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
772 	 * often isn't recommended due to ME interaction.
773 	 */
774 	do {
775 		bool cpu_busy = ips_cpu_busy(ips);
776 		bool gpu_busy = ips_gpu_busy(ips);
777 
778 		spin_lock_irqsave(&ips->turbo_status_lock, flags);
779 		if (ips->poll_turbo_status)
780 			update_turbo_limits(ips);
781 		spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
782 
783 		/* Update turbo status if necessary */
784 		if (ips->cpu_turbo_enabled)
785 			ips_enable_cpu_turbo(ips);
786 		else
787 			ips_disable_cpu_turbo(ips);
788 
789 		if (ips->gpu_turbo_enabled)
790 			ips_enable_gpu_turbo(ips);
791 		else
792 			ips_disable_gpu_turbo(ips);
793 
794 		/* We're outside our comfort zone, crank them down */
795 		if (mcp_exceeded(ips)) {
796 			ips_cpu_lower(ips);
797 			ips_gpu_lower(ips);
798 			goto sleep;
799 		}
800 
801 		if (!cpu_exceeded(ips, 0) && cpu_busy)
802 			ips_cpu_raise(ips);
803 		else
804 			ips_cpu_lower(ips);
805 
806 		if (!mch_exceeded(ips) && gpu_busy)
807 			ips_gpu_raise(ips);
808 		else
809 			ips_gpu_lower(ips);
810 
811 sleep:
812 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
813 	} while (!kthread_should_stop());
814 
815 	dev_dbg(ips->dev, "ips-adjust thread stopped\n");
816 
817 	return 0;
818 }
819 
820 /*
821  * Helpers for reading out temp/power values and calculating their
822  * averages for the decision making and monitoring functions.
823  */
824 
825 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
826 {
827 	u64 total = 0;
828 	int i;
829 	u16 avg;
830 
831 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
832 		total += (u64)(array[i] * 100);
833 
834 	do_div(total, IPS_SAMPLE_COUNT);
835 
836 	avg = (u16)total;
837 
838 	return avg;
839 }
840 
841 static u16 read_mgtv(struct ips_driver *ips)
842 {
843 	u16 ret;
844 	u64 slope, offset;
845 	u64 val;
846 
847 	val = thm_readq(THM_MGTV);
848 	val = (val & TV_MASK) >> TV_SHIFT;
849 
850 	slope = offset = thm_readw(THM_MGTA);
851 	slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
852 	offset = offset & MGTA_OFFSET_MASK;
853 
854 	ret = ((val * slope + 0x40) >> 7) + offset;
855 
856 	return 0; /* MCH temp reporting buggy */
857 }
858 
859 static u16 read_ptv(struct ips_driver *ips)
860 {
861 	u16 val, slope, offset;
862 
863 	slope = (ips->pta_val & PTA_SLOPE_MASK) >> PTA_SLOPE_SHIFT;
864 	offset = ips->pta_val & PTA_OFFSET_MASK;
865 
866 	val = thm_readw(THM_PTV) & PTV_MASK;
867 
868 	return val;
869 }
870 
871 static u16 read_ctv(struct ips_driver *ips, int cpu)
872 {
873 	int reg = cpu ? THM_CTV2 : THM_CTV1;
874 	u16 val;
875 
876 	val = thm_readw(reg);
877 	if (!(val & CTV_TEMP_ERROR))
878 		val = (val) >> 6; /* discard fractional component */
879 	else
880 		val = 0;
881 
882 	return val;
883 }
884 
885 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
886 {
887 	u32 val;
888 	u32 ret;
889 
890 	/*
891 	 * CEC is in joules/65535.  Take difference over time to
892 	 * get watts.
893 	 */
894 	val = thm_readl(THM_CEC);
895 
896 	/* period is in ms and we want mW */
897 	ret = (((val - *last) * 1000) / period);
898 	ret = (ret * 1000) / 65535;
899 	*last = val;
900 
901 	return 0;
902 }
903 
904 static const u16 temp_decay_factor = 2;
905 static u16 update_average_temp(u16 avg, u16 val)
906 {
907 	u16 ret;
908 
909 	/* Multiply by 100 for extra precision */
910 	ret = (val * 100 / temp_decay_factor) +
911 		(((temp_decay_factor - 1) * avg) / temp_decay_factor);
912 	return ret;
913 }
914 
915 static const u16 power_decay_factor = 2;
916 static u16 update_average_power(u32 avg, u32 val)
917 {
918 	u32 ret;
919 
920 	ret = (val / power_decay_factor) +
921 		(((power_decay_factor - 1) * avg) / power_decay_factor);
922 
923 	return ret;
924 }
925 
926 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
927 {
928 	u64 total = 0;
929 	u32 avg;
930 	int i;
931 
932 	for (i = 0; i < IPS_SAMPLE_COUNT; i++)
933 		total += array[i];
934 
935 	do_div(total, IPS_SAMPLE_COUNT);
936 	avg = (u32)total;
937 
938 	return avg;
939 }
940 
941 static void monitor_timeout(struct timer_list *t)
942 {
943 	struct ips_driver *ips = from_timer(ips, t, timer);
944 	wake_up_process(ips->monitor);
945 }
946 
947 /**
948  * ips_monitor - temp/power monitoring thread
949  * @data: ips driver structure
950  *
951  * This is the main function for the IPS driver.  It monitors power and
952  * tempurature in the MCP and adjusts CPU and GPU power clams accordingly.
953  *
954  * We keep a 5s moving average of power consumption and tempurature.  Using
955  * that data, along with CPU vs GPU preference, we adjust the power clamps
956  * up or down.
957  */
958 static int ips_monitor(void *data)
959 {
960 	struct ips_driver *ips = data;
961 	unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
962 	int i;
963 	u32 *cpu_samples, *mchp_samples, old_cpu_power;
964 	u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
965 	u8 cur_seqno, last_seqno;
966 
967 	mcp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
968 	ctv1_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
969 	ctv2_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
970 	mch_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u16), GFP_KERNEL);
971 	cpu_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
972 	mchp_samples = kcalloc(IPS_SAMPLE_COUNT, sizeof(u32), GFP_KERNEL);
973 	if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
974 			!cpu_samples || !mchp_samples) {
975 		dev_err(ips->dev,
976 			"failed to allocate sample array, ips disabled\n");
977 		kfree(mcp_samples);
978 		kfree(ctv1_samples);
979 		kfree(ctv2_samples);
980 		kfree(mch_samples);
981 		kfree(cpu_samples);
982 		kfree(mchp_samples);
983 		return -ENOMEM;
984 	}
985 
986 	last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
987 		ITV_ME_SEQNO_SHIFT;
988 	seqno_timestamp = get_jiffies_64();
989 
990 	old_cpu_power = thm_readl(THM_CEC);
991 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
992 
993 	/* Collect an initial average */
994 	for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
995 		u32 mchp, cpu_power;
996 		u16 val;
997 
998 		mcp_samples[i] = read_ptv(ips);
999 
1000 		val = read_ctv(ips, 0);
1001 		ctv1_samples[i] = val;
1002 
1003 		val = read_ctv(ips, 1);
1004 		ctv2_samples[i] = val;
1005 
1006 		val = read_mgtv(ips);
1007 		mch_samples[i] = val;
1008 
1009 		cpu_power = get_cpu_power(ips, &old_cpu_power,
1010 					  IPS_SAMPLE_PERIOD);
1011 		cpu_samples[i] = cpu_power;
1012 
1013 		if (ips->read_mch_val) {
1014 			mchp = ips->read_mch_val();
1015 			mchp_samples[i] = mchp;
1016 		}
1017 
1018 		schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1019 		if (kthread_should_stop())
1020 			break;
1021 	}
1022 
1023 	ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1024 	ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1025 	ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1026 	ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1027 	ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1028 	ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1029 	kfree(mcp_samples);
1030 	kfree(ctv1_samples);
1031 	kfree(ctv2_samples);
1032 	kfree(mch_samples);
1033 	kfree(cpu_samples);
1034 	kfree(mchp_samples);
1035 
1036 	/* Start the adjustment thread now that we have data */
1037 	wake_up_process(ips->adjust);
1038 
1039 	/*
1040 	 * Ok, now we have an initial avg.  From here on out, we track the
1041 	 * running avg using a decaying average calculation.  This allows
1042 	 * us to reduce the sample frequency if the CPU and GPU are idle.
1043 	 */
1044 	old_cpu_power = thm_readl(THM_CEC);
1045 	schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1046 	last_sample_period = IPS_SAMPLE_PERIOD;
1047 
1048 	timer_setup(&ips->timer, monitor_timeout, TIMER_DEFERRABLE);
1049 	do {
1050 		u32 cpu_val, mch_val;
1051 		u16 val;
1052 
1053 		/* MCP itself */
1054 		val = read_ptv(ips);
1055 		ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1056 
1057 		/* Processor 0 */
1058 		val = read_ctv(ips, 0);
1059 		ips->ctv1_avg_temp =
1060 			update_average_temp(ips->ctv1_avg_temp, val);
1061 		/* Power */
1062 		cpu_val = get_cpu_power(ips, &old_cpu_power,
1063 					last_sample_period);
1064 		ips->cpu_avg_power =
1065 			update_average_power(ips->cpu_avg_power, cpu_val);
1066 
1067 		if (ips->second_cpu) {
1068 			/* Processor 1 */
1069 			val = read_ctv(ips, 1);
1070 			ips->ctv2_avg_temp =
1071 				update_average_temp(ips->ctv2_avg_temp, val);
1072 		}
1073 
1074 		/* MCH */
1075 		val = read_mgtv(ips);
1076 		ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1077 		/* Power */
1078 		if (ips->read_mch_val) {
1079 			mch_val = ips->read_mch_val();
1080 			ips->mch_avg_power =
1081 				update_average_power(ips->mch_avg_power,
1082 						     mch_val);
1083 		}
1084 
1085 		/*
1086 		 * Make sure ME is updating thermal regs.
1087 		 * Note:
1088 		 * If it's been more than a second since the last update,
1089 		 * the ME is probably hung.
1090 		 */
1091 		cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1092 			ITV_ME_SEQNO_SHIFT;
1093 		if (cur_seqno == last_seqno &&
1094 		    time_after(jiffies, seqno_timestamp + HZ)) {
1095 			dev_warn(ips->dev,
1096 				 "ME failed to update for more than 1s, likely hung\n");
1097 		} else {
1098 			seqno_timestamp = get_jiffies_64();
1099 			last_seqno = cur_seqno;
1100 		}
1101 
1102 		last_msecs = jiffies_to_msecs(jiffies);
1103 		expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1104 
1105 		__set_current_state(TASK_INTERRUPTIBLE);
1106 		mod_timer(&ips->timer, expire);
1107 		schedule();
1108 
1109 		/* Calculate actual sample period for power averaging */
1110 		last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1111 		if (!last_sample_period)
1112 			last_sample_period = 1;
1113 	} while (!kthread_should_stop());
1114 
1115 	del_timer_sync(&ips->timer);
1116 
1117 	dev_dbg(ips->dev, "ips-monitor thread stopped\n");
1118 
1119 	return 0;
1120 }
1121 
1122 #if 0
1123 #define THM_DUMPW(reg) \
1124 	{ \
1125 	u16 val = thm_readw(reg); \
1126 	dev_dbg(ips->dev, #reg ": 0x%04x\n", val); \
1127 	}
1128 #define THM_DUMPL(reg) \
1129 	{ \
1130 	u32 val = thm_readl(reg); \
1131 	dev_dbg(ips->dev, #reg ": 0x%08x\n", val); \
1132 	}
1133 #define THM_DUMPQ(reg) \
1134 	{ \
1135 	u64 val = thm_readq(reg); \
1136 	dev_dbg(ips->dev, #reg ": 0x%016x\n", val); \
1137 	}
1138 
1139 static void dump_thermal_info(struct ips_driver *ips)
1140 {
1141 	u16 ptl;
1142 
1143 	ptl = thm_readw(THM_PTL);
1144 	dev_dbg(ips->dev, "Processor temp limit: %d\n", ptl);
1145 
1146 	THM_DUMPW(THM_CTA);
1147 	THM_DUMPW(THM_TRC);
1148 	THM_DUMPW(THM_CTV1);
1149 	THM_DUMPL(THM_STS);
1150 	THM_DUMPW(THM_PTV);
1151 	THM_DUMPQ(THM_MGTV);
1152 }
1153 #endif
1154 
1155 /**
1156  * ips_irq_handler - handle temperature triggers and other IPS events
1157  * @irq: irq number
1158  * @arg: unused
1159  *
1160  * Handle temperature limit trigger events, generally by lowering the clamps.
1161  * If we're at a critical limit, we clamp back to the lowest possible value
1162  * to prevent emergency shutdown.
1163  */
1164 static irqreturn_t ips_irq_handler(int irq, void *arg)
1165 {
1166 	struct ips_driver *ips = arg;
1167 	u8 tses = thm_readb(THM_TSES);
1168 	u8 tes = thm_readb(THM_TES);
1169 
1170 	if (!tses && !tes)
1171 		return IRQ_NONE;
1172 
1173 	dev_info(ips->dev, "TSES: 0x%02x\n", tses);
1174 	dev_info(ips->dev, "TES: 0x%02x\n", tes);
1175 
1176 	/* STS update from EC? */
1177 	if (tes & 1) {
1178 		u32 sts, tc1;
1179 
1180 		sts = thm_readl(THM_STS);
1181 		tc1 = thm_readl(THM_TC1);
1182 
1183 		if (sts & STS_NVV) {
1184 			spin_lock(&ips->turbo_status_lock);
1185 			ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1186 				STS_PCPL_SHIFT;
1187 			ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1188 				STS_GPL_SHIFT;
1189 			/* ignore EC CPU vs GPU pref */
1190 			ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1191 			/*
1192 			 * Disable turbo for now, until we can figure
1193 			 * out why the power figures are wrong
1194 			 */
1195 			ips->cpu_turbo_enabled = false;
1196 			if (ips->gpu_busy)
1197 				ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1198 			ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1199 				STS_PTL_SHIFT;
1200 			ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1201 				STS_PPL_SHIFT;
1202 			verify_limits(ips);
1203 			spin_unlock(&ips->turbo_status_lock);
1204 
1205 			thm_writeb(THM_SEC, SEC_ACK);
1206 		}
1207 		thm_writeb(THM_TES, tes);
1208 	}
1209 
1210 	/* Thermal trip */
1211 	if (tses) {
1212 		dev_warn(ips->dev, "thermal trip occurred, tses: 0x%04x\n",
1213 			 tses);
1214 		thm_writeb(THM_TSES, tses);
1215 	}
1216 
1217 	return IRQ_HANDLED;
1218 }
1219 
1220 #ifndef CONFIG_DEBUG_FS
1221 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1222 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1223 #else
1224 
1225 /* Expose current state and limits in debugfs if possible */
1226 
1227 struct ips_debugfs_node {
1228 	struct ips_driver *ips;
1229 	char *name;
1230 	int (*show)(struct seq_file *m, void *data);
1231 };
1232 
1233 static int show_cpu_temp(struct seq_file *m, void *data)
1234 {
1235 	struct ips_driver *ips = m->private;
1236 
1237 	seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1238 		   ips->ctv1_avg_temp % 100);
1239 
1240 	return 0;
1241 }
1242 
1243 static int show_cpu_power(struct seq_file *m, void *data)
1244 {
1245 	struct ips_driver *ips = m->private;
1246 
1247 	seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1248 
1249 	return 0;
1250 }
1251 
1252 static int show_cpu_clamp(struct seq_file *m, void *data)
1253 {
1254 	u64 turbo_override;
1255 	int tdp, tdc;
1256 
1257 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1258 
1259 	tdp = (int)(turbo_override & TURBO_TDP_MASK);
1260 	tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1261 
1262 	/* Convert to .1W/A units */
1263 	tdp = tdp * 10 / 8;
1264 	tdc = tdc * 10 / 8;
1265 
1266 	/* Watts Amperes */
1267 	seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1268 		   tdc / 10, tdc % 10);
1269 
1270 	return 0;
1271 }
1272 
1273 static int show_mch_temp(struct seq_file *m, void *data)
1274 {
1275 	struct ips_driver *ips = m->private;
1276 
1277 	seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1278 		   ips->mch_avg_temp % 100);
1279 
1280 	return 0;
1281 }
1282 
1283 static int show_mch_power(struct seq_file *m, void *data)
1284 {
1285 	struct ips_driver *ips = m->private;
1286 
1287 	seq_printf(m, "%dmW\n", ips->mch_avg_power);
1288 
1289 	return 0;
1290 }
1291 
1292 static struct ips_debugfs_node ips_debug_files[] = {
1293 	{ NULL, "cpu_temp", show_cpu_temp },
1294 	{ NULL, "cpu_power", show_cpu_power },
1295 	{ NULL, "cpu_clamp", show_cpu_clamp },
1296 	{ NULL, "mch_temp", show_mch_temp },
1297 	{ NULL, "mch_power", show_mch_power },
1298 };
1299 
1300 static int ips_debugfs_open(struct inode *inode, struct file *file)
1301 {
1302 	struct ips_debugfs_node *node = inode->i_private;
1303 
1304 	return single_open(file, node->show, node->ips);
1305 }
1306 
1307 static const struct file_operations ips_debugfs_ops = {
1308 	.owner = THIS_MODULE,
1309 	.open = ips_debugfs_open,
1310 	.read = seq_read,
1311 	.llseek = seq_lseek,
1312 	.release = single_release,
1313 };
1314 
1315 static void ips_debugfs_cleanup(struct ips_driver *ips)
1316 {
1317 	if (ips->debug_root)
1318 		debugfs_remove_recursive(ips->debug_root);
1319 	return;
1320 }
1321 
1322 static void ips_debugfs_init(struct ips_driver *ips)
1323 {
1324 	int i;
1325 
1326 	ips->debug_root = debugfs_create_dir("ips", NULL);
1327 	if (!ips->debug_root) {
1328 		dev_err(ips->dev, "failed to create debugfs entries: %ld\n",
1329 			PTR_ERR(ips->debug_root));
1330 		return;
1331 	}
1332 
1333 	for (i = 0; i < ARRAY_SIZE(ips_debug_files); i++) {
1334 		struct dentry *ent;
1335 		struct ips_debugfs_node *node = &ips_debug_files[i];
1336 
1337 		node->ips = ips;
1338 		ent = debugfs_create_file(node->name, S_IFREG | S_IRUGO,
1339 					  ips->debug_root, node,
1340 					  &ips_debugfs_ops);
1341 		if (!ent) {
1342 			dev_err(ips->dev, "failed to create debug file: %ld\n",
1343 				PTR_ERR(ent));
1344 			goto err_cleanup;
1345 		}
1346 	}
1347 
1348 	return;
1349 
1350 err_cleanup:
1351 	ips_debugfs_cleanup(ips);
1352 	return;
1353 }
1354 #endif /* CONFIG_DEBUG_FS */
1355 
1356 /**
1357  * ips_detect_cpu - detect whether CPU supports IPS
1358  *
1359  * Walk our list and see if we're on a supported CPU.  If we find one,
1360  * return the limits for it.
1361  */
1362 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1363 {
1364 	u64 turbo_power, misc_en;
1365 	struct ips_mcp_limits *limits = NULL;
1366 	u16 tdp;
1367 
1368 	if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1369 		dev_info(ips->dev, "Non-IPS CPU detected.\n");
1370 		return NULL;
1371 	}
1372 
1373 	rdmsrl(IA32_MISC_ENABLE, misc_en);
1374 	/*
1375 	 * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1376 	 * turbo manually or we'll get an illegal MSR access, even though
1377 	 * turbo will still be available.
1378 	 */
1379 	if (misc_en & IA32_MISC_TURBO_EN)
1380 		ips->turbo_toggle_allowed = true;
1381 	else
1382 		ips->turbo_toggle_allowed = false;
1383 
1384 	if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1385 		limits = &ips_sv_limits;
1386 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1387 		limits = &ips_lv_limits;
1388 	else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1389 		limits = &ips_ulv_limits;
1390 	else {
1391 		dev_info(ips->dev, "No CPUID match found.\n");
1392 		return NULL;
1393 	}
1394 
1395 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1396 	tdp = turbo_power & TURBO_TDP_MASK;
1397 
1398 	/* Sanity check TDP against CPU */
1399 	if (limits->core_power_limit != (tdp / 8) * 1000) {
1400 		dev_info(ips->dev,
1401 			 "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1402 			 tdp / 8, limits->core_power_limit / 1000);
1403 		limits->core_power_limit = (tdp / 8) * 1000;
1404 	}
1405 
1406 	return limits;
1407 }
1408 
1409 /**
1410  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1411  * @ips: IPS driver
1412  *
1413  * The i915 driver exports several interfaces to allow the IPS driver to
1414  * monitor and control graphics turbo mode.  If we can find them, we can
1415  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1416  * thermal and power limits in the MCP.
1417  */
1418 static bool ips_get_i915_syms(struct ips_driver *ips)
1419 {
1420 	ips->read_mch_val = symbol_get(i915_read_mch_val);
1421 	if (!ips->read_mch_val)
1422 		goto out_err;
1423 	ips->gpu_raise = symbol_get(i915_gpu_raise);
1424 	if (!ips->gpu_raise)
1425 		goto out_put_mch;
1426 	ips->gpu_lower = symbol_get(i915_gpu_lower);
1427 	if (!ips->gpu_lower)
1428 		goto out_put_raise;
1429 	ips->gpu_busy = symbol_get(i915_gpu_busy);
1430 	if (!ips->gpu_busy)
1431 		goto out_put_lower;
1432 	ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1433 	if (!ips->gpu_turbo_disable)
1434 		goto out_put_busy;
1435 
1436 	return true;
1437 
1438 out_put_busy:
1439 	symbol_put(i915_gpu_busy);
1440 out_put_lower:
1441 	symbol_put(i915_gpu_lower);
1442 out_put_raise:
1443 	symbol_put(i915_gpu_raise);
1444 out_put_mch:
1445 	symbol_put(i915_read_mch_val);
1446 out_err:
1447 	return false;
1448 }
1449 
1450 static bool
1451 ips_gpu_turbo_enabled(struct ips_driver *ips)
1452 {
1453 	if (!ips->gpu_busy && late_i915_load) {
1454 		if (ips_get_i915_syms(ips)) {
1455 			dev_info(ips->dev,
1456 				 "i915 driver attached, reenabling gpu turbo\n");
1457 			ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1458 		}
1459 	}
1460 
1461 	return ips->gpu_turbo_enabled;
1462 }
1463 
1464 void
1465 ips_link_to_i915_driver(void)
1466 {
1467 	/* We can't cleanly get at the various ips_driver structs from
1468 	 * this caller (the i915 driver), so just set a flag saying
1469 	 * that it's time to try getting the symbols again.
1470 	 */
1471 	late_i915_load = true;
1472 }
1473 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1474 
1475 static const struct pci_device_id ips_id_table[] = {
1476 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1477 	{ 0, }
1478 };
1479 
1480 MODULE_DEVICE_TABLE(pci, ips_id_table);
1481 
1482 static int ips_blacklist_callback(const struct dmi_system_id *id)
1483 {
1484 	pr_info("Blacklisted intel_ips for %s\n", id->ident);
1485 	return 1;
1486 }
1487 
1488 static const struct dmi_system_id ips_blacklist[] = {
1489 	{
1490 		.callback = ips_blacklist_callback,
1491 		.ident = "HP ProBook",
1492 		.matches = {
1493 			DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1494 			DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1495 		},
1496 	},
1497 	{ }	/* terminating entry */
1498 };
1499 
1500 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1501 {
1502 	u64 platform_info;
1503 	struct ips_driver *ips;
1504 	u32 hts;
1505 	int ret = 0;
1506 	u16 htshi, trc, trc_required_mask;
1507 	u8 tse;
1508 
1509 	if (dmi_check_system(ips_blacklist))
1510 		return -ENODEV;
1511 
1512 	ips = devm_kzalloc(&dev->dev, sizeof(*ips), GFP_KERNEL);
1513 	if (!ips)
1514 		return -ENOMEM;
1515 
1516 	spin_lock_init(&ips->turbo_status_lock);
1517 	ips->dev = &dev->dev;
1518 
1519 	ips->limits = ips_detect_cpu(ips);
1520 	if (!ips->limits) {
1521 		dev_info(&dev->dev, "IPS not supported on this CPU\n");
1522 		return -ENXIO;
1523 	}
1524 
1525 	ret = pcim_enable_device(dev);
1526 	if (ret) {
1527 		dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1528 		return ret;
1529 	}
1530 
1531 	ret = pcim_iomap_regions(dev, 1 << 0, pci_name(dev));
1532 	if (ret) {
1533 		dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1534 		return ret;
1535 	}
1536 	ips->regmap = pcim_iomap_table(dev)[0];
1537 
1538 	pci_set_drvdata(dev, ips);
1539 
1540 	tse = thm_readb(THM_TSE);
1541 	if (tse != TSE_EN) {
1542 		dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1543 		return -ENXIO;
1544 	}
1545 
1546 	trc = thm_readw(THM_TRC);
1547 	trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1548 	if ((trc & trc_required_mask) != trc_required_mask) {
1549 		dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1550 		return -ENXIO;
1551 	}
1552 
1553 	if (trc & TRC_CORE2_EN)
1554 		ips->second_cpu = true;
1555 
1556 	update_turbo_limits(ips);
1557 	dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1558 		ips->mcp_power_limit / 10);
1559 	dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1560 		ips->core_power_limit / 10);
1561 	/* BIOS may update limits at runtime */
1562 	if (thm_readl(THM_PSC) & PSP_PBRT)
1563 		ips->poll_turbo_status = true;
1564 
1565 	if (!ips_get_i915_syms(ips)) {
1566 		dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1567 		ips->gpu_turbo_enabled = false;
1568 	} else {
1569 		dev_dbg(&dev->dev, "graphics turbo enabled\n");
1570 		ips->gpu_turbo_enabled = true;
1571 	}
1572 
1573 	/*
1574 	 * Check PLATFORM_INFO MSR to make sure this chip is
1575 	 * turbo capable.
1576 	 */
1577 	rdmsrl(PLATFORM_INFO, platform_info);
1578 	if (!(platform_info & PLATFORM_TDP)) {
1579 		dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1580 		return -ENODEV;
1581 	}
1582 
1583 	/*
1584 	 * IRQ handler for ME interaction
1585 	 * Note: don't use MSI here as the PCH has bugs.
1586 	 */
1587 	ret = pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_LEGACY);
1588 	if (ret < 0)
1589 		return ret;
1590 
1591 	ips->irq = pci_irq_vector(dev, 0);
1592 
1593 	ret = request_irq(ips->irq, ips_irq_handler, IRQF_SHARED, "ips", ips);
1594 	if (ret) {
1595 		dev_err(&dev->dev, "request irq failed, aborting\n");
1596 		return ret;
1597 	}
1598 
1599 	/* Enable aux, hot & critical interrupts */
1600 	thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1601 		   TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1602 	thm_writeb(THM_TEN, TEN_UPDATE_EN);
1603 
1604 	/* Collect adjustment values */
1605 	ips->cta_val = thm_readw(THM_CTA);
1606 	ips->pta_val = thm_readw(THM_PTA);
1607 	ips->mgta_val = thm_readw(THM_MGTA);
1608 
1609 	/* Save turbo limits & ratios */
1610 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1611 
1612 	ips_disable_cpu_turbo(ips);
1613 	ips->cpu_turbo_enabled = false;
1614 
1615 	/* Create thermal adjust thread */
1616 	ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1617 	if (IS_ERR(ips->adjust)) {
1618 		dev_err(&dev->dev,
1619 			"failed to create thermal adjust thread, aborting\n");
1620 		ret = -ENOMEM;
1621 		goto error_free_irq;
1622 
1623 	}
1624 
1625 	/*
1626 	 * Set up the work queue and monitor thread. The monitor thread
1627 	 * will wake up ips_adjust thread.
1628 	 */
1629 	ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1630 	if (IS_ERR(ips->monitor)) {
1631 		dev_err(&dev->dev,
1632 			"failed to create thermal monitor thread, aborting\n");
1633 		ret = -ENOMEM;
1634 		goto error_thread_cleanup;
1635 	}
1636 
1637 	hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1638 		(ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1639 	htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1640 
1641 	thm_writew(THM_HTSHI, htshi);
1642 	thm_writel(THM_HTS, hts);
1643 
1644 	ips_debugfs_init(ips);
1645 
1646 	dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1647 		 ips->mcp_temp_limit);
1648 	return ret;
1649 
1650 error_thread_cleanup:
1651 	kthread_stop(ips->adjust);
1652 error_free_irq:
1653 	free_irq(ips->irq, ips);
1654 	pci_free_irq_vectors(dev);
1655 	return ret;
1656 }
1657 
1658 static void ips_remove(struct pci_dev *dev)
1659 {
1660 	struct ips_driver *ips = pci_get_drvdata(dev);
1661 	u64 turbo_override;
1662 
1663 	if (!ips)
1664 		return;
1665 
1666 	ips_debugfs_cleanup(ips);
1667 
1668 	/* Release i915 driver */
1669 	if (ips->read_mch_val)
1670 		symbol_put(i915_read_mch_val);
1671 	if (ips->gpu_raise)
1672 		symbol_put(i915_gpu_raise);
1673 	if (ips->gpu_lower)
1674 		symbol_put(i915_gpu_lower);
1675 	if (ips->gpu_busy)
1676 		symbol_put(i915_gpu_busy);
1677 	if (ips->gpu_turbo_disable)
1678 		symbol_put(i915_gpu_turbo_disable);
1679 
1680 	rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1681 	turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1682 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1683 	wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1684 
1685 	free_irq(ips->irq, ips);
1686 	pci_free_irq_vectors(dev);
1687 	if (ips->adjust)
1688 		kthread_stop(ips->adjust);
1689 	if (ips->monitor)
1690 		kthread_stop(ips->monitor);
1691 	dev_dbg(&dev->dev, "IPS driver removed\n");
1692 }
1693 
1694 static struct pci_driver ips_pci_driver = {
1695 	.name = "intel ips",
1696 	.id_table = ips_id_table,
1697 	.probe = ips_probe,
1698 	.remove = ips_remove,
1699 };
1700 
1701 module_pci_driver(ips_pci_driver);
1702 
1703 MODULE_LICENSE("GPL");
1704 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1705 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");
1706