drivers/platform/x86/intel_ips.c

   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  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  16  *
  17  * The full GNU General Public License is included in this distribution in
  18  * the file called "COPYING".
  19  *
  20  * Authors:
  21  *      Jesse Barnes <jbarnes@virtuousgeek.org>
  22  */
  23
  24 /*
  25  * Some Intel Ibex Peak based platforms support so-called "intelligent
  26  * power sharing", which allows the CPU and GPU to cooperate to maximize
  27  * performance within a given TDP (thermal design point).  This driver
  28  * performs the coordination between the CPU and GPU, monitors thermal and
  29  * power statistics in the platform, and initializes power monitoring
  30  * hardware.  It also provides a few tunables to control behavior.  Its
  31  * primary purpose is to safely allow CPU and GPU turbo modes to be enabled
  32  * by tracking power and thermal budget; secondarily it can boost turbo
  33  * performance by allocating more power or thermal budget to the CPU or GPU
  34  * based on available headroom and activity.
  35  *
  36  * The basic algorithm is driven by a 5s moving average of temperature.  If
  37  * thermal headroom is available, the CPU and/or GPU power clamps may be
  38  * adjusted upwards.  If we hit the thermal ceiling or a thermal trigger,
  39  * we scale back the clamp.  Aside from trigger events (when we're critically
  40  * close or over our TDP) we don't adjust the clamps more than once every
  41  * five seconds.
  42  *
  43  * The thermal device (device 31, function 6) has a set of registers that
  44  * are updated by the ME firmware.  The ME should also take the clamp values
  45  * written to those registers and write them to the CPU, but we currently
  46  * bypass that functionality and write the CPU MSR directly.
  47  *
  48  * UNSUPPORTED:
  49  *   - dual MCP configs
  50  *
  51  * TODO:
  52  *   - handle CPU hotplug
  53  *   - provide turbo enable/disable api
  54  *
  55  * Related documents:
  56  *   - CDI 403777, 403778 - Auburndale EDS vol 1 & 2
  57  *   - CDI 401376 - Ibex Peak EDS
  58  *   - ref 26037, 26641 - IPS BIOS spec
  59  *   - ref 26489 - Nehalem BIOS writer's guide
  60  *   - ref 26921 - Ibex Peak BIOS Specification
  61  */
  62
  63 #include <linux/debugfs.h>
  64 #include <linux/delay.h>
  65 #include <linux/interrupt.h>
  66 #include <linux/kernel.h>
  67 #include <linux/kthread.h>
  68 #include <linux/module.h>
  69 #include <linux/pci.h>
  70 #include <linux/sched.h>
  71 #include <linux/sched/loadavg.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/string.h>
  74 #include <linux/tick.h>
  75 #include <linux/timer.h>
  76 #include <linux/dmi.h>
  77 #include <drm/i915_drm.h>
  78 #include <asm/msr.h>
  79 #include <asm/processor.h>
  80 #include "intel_ips.h"
  81
  82 #include <linux/io-64-nonatomic-lo-hi.h>
  83
  84 #define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32
  85
  86 /*
  87  * Package level MSRs for monitor/control
  88  */
  89 #define PLATFORM_INFO   0xce
  90 #define   PLATFORM_TDP          (1<<29)
  91 #define   PLATFORM_RATIO        (1<<28)
  92
  93 #define IA32_MISC_ENABLE        0x1a0
  94 #define   IA32_MISC_TURBO_EN    (1ULL<<38)
  95
  96 #define TURBO_POWER_CURRENT_LIMIT       0x1ac
  97 #define   TURBO_TDC_OVR_EN      (1UL<<31)
  98 #define   TURBO_TDC_MASK        (0x000000007fff0000UL)
  99 #define   TURBO_TDC_SHIFT       (16)
 100 #define   TURBO_TDP_OVR_EN      (1UL<<15)
 101 #define   TURBO_TDP_MASK        (0x0000000000003fffUL)
 102
 103 /*
 104  * Core/thread MSRs for monitoring
 105  */
 106 #define IA32_PERF_CTL           0x199
 107 #define   IA32_PERF_TURBO_DIS   (1ULL<<32)
 108
 109 /*
 110  * Thermal PCI device regs
 111  */
 112 #define THM_CFG_TBAR    0x10
 113 #define THM_CFG_TBAR_HI 0x14
 114
 115 #define THM_TSIU        0x00
 116 #define THM_TSE         0x01
 117 #define   TSE_EN        0xb8
 118 #define THM_TSS         0x02
 119 #define THM_TSTR        0x03
 120 #define THM_TSTTP       0x04
 121 #define THM_TSCO        0x08
 122 #define THM_TSES        0x0c
 123 #define THM_TSGPEN      0x0d
 124 #define   TSGPEN_HOT_LOHI       (1<<1)
 125 #define   TSGPEN_CRIT_LOHI      (1<<2)
 126 #define THM_TSPC        0x0e
 127 #define THM_PPEC        0x10
 128 #define THM_CTA         0x12
 129 #define THM_PTA         0x14
 130 #define   PTA_SLOPE_MASK        (0xff00)
 131 #define   PTA_SLOPE_SHIFT       8
 132 #define   PTA_OFFSET_MASK       (0x00ff)
 133 #define THM_MGTA        0x16
 134 #define   MGTA_SLOPE_MASK       (0xff00)
 135 #define   MGTA_SLOPE_SHIFT      8
 136 #define   MGTA_OFFSET_MASK      (0x00ff)
 137 #define THM_TRC         0x1a
 138 #define   TRC_CORE2_EN  (1<<15)
 139 #define   TRC_THM_EN    (1<<12)
 140 #define   TRC_C6_WAR    (1<<8)
 141 #define   TRC_CORE1_EN  (1<<7)
 142 #define   TRC_CORE_PWR  (1<<6)
 143 #define   TRC_PCH_EN    (1<<5)
 144 #define   TRC_MCH_EN    (1<<4)
 145 #define   TRC_DIMM4     (1<<3)
 146 #define   TRC_DIMM3     (1<<2)
 147 #define   TRC_DIMM2     (1<<1)
 148 #define   TRC_DIMM1     (1<<0)
 149 #define THM_TES         0x20
 150 #define THM_TEN         0x21
 151 #define   TEN_UPDATE_EN 1
 152 #define THM_PSC         0x24
 153 #define   PSC_NTG       (1<<0) /* No GFX turbo support */
 154 #define   PSC_NTPC      (1<<1) /* No CPU turbo support */
 155 #define   PSC_PP_DEF    (0<<2) /* Perf policy up to driver */
 156 #define   PSP_PP_PC     (1<<2) /* BIOS prefers CPU perf */
 157 #define   PSP_PP_BAL    (2<<2) /* BIOS wants balanced perf */
 158 #define   PSP_PP_GFX    (3<<2) /* BIOS prefers GFX perf */
 159 #define   PSP_PBRT      (1<<4) /* BIOS run time support */
 160 #define THM_CTV1        0x30
 161 #define   CTV_TEMP_ERROR (1<<15)
 162 #define   CTV_TEMP_MASK 0x3f
 163 #define   CTV_
 164 #define THM_CTV2        0x32
 165 #define THM_CEC         0x34 /* undocumented power accumulator in joules */
 166 #define THM_AE          0x3f
 167 #define THM_HTS         0x50 /* 32 bits */
 168 #define   HTS_PCPL_MASK (0x7fe00000)
 169 #define   HTS_PCPL_SHIFT 21
 170 #define   HTS_GPL_MASK  (0x001ff000)
 171 #define   HTS_GPL_SHIFT 12
 172 #define   HTS_PP_MASK   (0x00000c00)
 173 #define   HTS_PP_SHIFT  10
 174 #define   HTS_PP_DEF    0
 175 #define   HTS_PP_PROC   1
 176 #define   HTS_PP_BAL    2
 177 #define   HTS_PP_GFX    3
 178 #define   HTS_PCTD_DIS  (1<<9)
 179 #define   HTS_GTD_DIS   (1<<8)
 180 #define   HTS_PTL_MASK  (0x000000fe)
 181 #define   HTS_PTL_SHIFT 1
 182 #define   HTS_NVV       (1<<0)
 183 #define THM_HTSHI       0x54 /* 16 bits */
 184 #define   HTS2_PPL_MASK         (0x03ff)
 185 #define   HTS2_PRST_MASK        (0x3c00)
 186 #define   HTS2_PRST_SHIFT       10
 187 #define   HTS2_PRST_UNLOADED    0
 188 #define   HTS2_PRST_RUNNING     1
 189 #define   HTS2_PRST_TDISOP      2 /* turbo disabled due to power */
 190 #define   HTS2_PRST_TDISHT      3 /* turbo disabled due to high temp */
 191 #define   HTS2_PRST_TDISUSR     4 /* user disabled turbo */
 192 #define   HTS2_PRST_TDISPLAT    5 /* platform disabled turbo */
 193 #define   HTS2_PRST_TDISPM      6 /* power management disabled turbo */
 194 #define   HTS2_PRST_TDISERR     7 /* some kind of error disabled turbo */
 195 #define THM_PTL         0x56
 196 #define THM_MGTV        0x58
 197 #define   TV_MASK       0x000000000000ff00
 198 #define   TV_SHIFT      8
 199 #define THM_PTV         0x60
 200 #define   PTV_MASK      0x00ff
 201 #define THM_MMGPC       0x64
 202 #define THM_MPPC        0x66
 203 #define THM_MPCPC       0x68
 204 #define THM_TSPIEN      0x82
 205 #define   TSPIEN_AUX_LOHI       (1<<0)
 206 #define   TSPIEN_HOT_LOHI       (1<<1)
 207 #define   TSPIEN_CRIT_LOHI      (1<<2)
 208 #define   TSPIEN_AUX2_LOHI      (1<<3)
 209 #define THM_TSLOCK      0x83
 210 #define THM_ATR         0x84
 211 #define THM_TOF         0x87
 212 #define THM_STS         0x98
 213 #define   STS_PCPL_MASK         (0x7fe00000)
 214 #define   STS_PCPL_SHIFT        21
 215 #define   STS_GPL_MASK          (0x001ff000)
 216 #define   STS_GPL_SHIFT         12
 217 #define   STS_PP_MASK           (0x00000c00)
 218 #define   STS_PP_SHIFT          10
 219 #define   STS_PP_DEF            0
 220 #define   STS_PP_PROC           1
 221 #define   STS_PP_BAL            2
 222 #define   STS_PP_GFX            3
 223 #define   STS_PCTD_DIS          (1<<9)
 224 #define   STS_GTD_DIS           (1<<8)
 225 #define   STS_PTL_MASK          (0x000000fe)
 226 #define   STS_PTL_SHIFT         1
 227 #define   STS_NVV               (1<<0)
 228 #define THM_SEC         0x9c
 229 #define   SEC_ACK       (1<<0)
 230 #define THM_TC3         0xa4
 231 #define THM_TC1         0xa8
 232 #define   STS_PPL_MASK          (0x0003ff00)
 233 #define   STS_PPL_SHIFT         16
 234 #define THM_TC2         0xac
 235 #define THM_DTV         0xb0
 236 #define THM_ITV         0xd8
 237 #define   ITV_ME_SEQNO_MASK 0x00ff0000 /* ME should update every ~200ms */
 238 #define   ITV_ME_SEQNO_SHIFT (16)
 239 #define   ITV_MCH_TEMP_MASK 0x0000ff00
 240 #define   ITV_MCH_TEMP_SHIFT (8)
 241 #define   ITV_PCH_TEMP_MASK 0x000000ff
 242
 243 #define thm_readb(off) readb(ips->regmap + (off))
 244 #define thm_readw(off) readw(ips->regmap + (off))
 245 #define thm_readl(off) readl(ips->regmap + (off))
 246 #define thm_readq(off) readq(ips->regmap + (off))
 247
 248 #define thm_writeb(off, val) writeb((val), ips->regmap + (off))
 249 #define thm_writew(off, val) writew((val), ips->regmap + (off))
 250 #define thm_writel(off, val) writel((val), ips->regmap + (off))
 251
 252 static const int IPS_ADJUST_PERIOD = 5000; /* ms */
 253 static bool late_i915_load = false;
 254
 255 /* For initial average collection */
 256 static const int IPS_SAMPLE_PERIOD = 200; /* ms */
 257 static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */
 258 #define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD)
 259
 260 /* Per-SKU limits */
 261 struct ips_mcp_limits {
 262         int cpu_family;
 263         int cpu_model; /* includes extended model... */
 264         int mcp_power_limit; /* mW units */
 265         int core_power_limit;
 266         int mch_power_limit;
 267         int core_temp_limit; /* degrees C */
 268         int mch_temp_limit;
 269 };
 270
 271 /* Max temps are -10 degrees C to avoid PROCHOT# */
 272
 273 static struct ips_mcp_limits ips_sv_limits = {
 274         .mcp_power_limit = 35000,
 275         .core_power_limit = 29000,
 276         .mch_power_limit = 20000,
 277         .core_temp_limit = 95,
 278         .mch_temp_limit = 90
 279 };
 280
 281 static struct ips_mcp_limits ips_lv_limits = {
 282         .mcp_power_limit = 25000,
 283         .core_power_limit = 21000,
 284         .mch_power_limit = 13000,
 285         .core_temp_limit = 95,
 286         .mch_temp_limit = 90
 287 };
 288
 289 static struct ips_mcp_limits ips_ulv_limits = {
 290         .mcp_power_limit = 18000,
 291         .core_power_limit = 14000,
 292         .mch_power_limit = 11000,
 293         .core_temp_limit = 95,
 294         .mch_temp_limit = 90
 295 };
 296
 297 struct ips_driver {
 298         struct pci_dev *dev;
 299         void *regmap;
 300         struct task_struct *monitor;
 301         struct task_struct *adjust;
 302         struct dentry *debug_root;
 303
 304         /* Average CPU core temps (all averages in .01 degrees C for precision) */
 305         u16 ctv1_avg_temp;
 306         u16 ctv2_avg_temp;
 307         /* GMCH average */
 308         u16 mch_avg_temp;
 309         /* Average for the CPU (both cores?) */
 310         u16 mcp_avg_temp;
 311         /* Average power consumption (in mW) */
 312         u32 cpu_avg_power;
 313         u32 mch_avg_power;
 314
 315         /* Offset values */
 316         u16 cta_val;
 317         u16 pta_val;
 318         u16 mgta_val;
 319
 320         /* Maximums & prefs, protected by turbo status lock */
 321         spinlock_t turbo_status_lock;
 322         u16 mcp_temp_limit;
 323         u16 mcp_power_limit;
 324         u16 core_power_limit;
 325         u16 mch_power_limit;
 326         bool cpu_turbo_enabled;
 327         bool __cpu_turbo_on;
 328         bool gpu_turbo_enabled;
 329         bool __gpu_turbo_on;
 330         bool gpu_preferred;
 331         bool poll_turbo_status;
 332         bool second_cpu;
 333         bool turbo_toggle_allowed;
 334         struct ips_mcp_limits *limits;
 335
 336         /* Optional MCH interfaces for if i915 is in use */
 337         unsigned long (*read_mch_val)(void);
 338         bool (*gpu_raise)(void);
 339         bool (*gpu_lower)(void);
 340         bool (*gpu_busy)(void);
 341         bool (*gpu_turbo_disable)(void);
 342
 343         /* For restoration at unload */
 344         u64 orig_turbo_limit;
 345         u64 orig_turbo_ratios;
 346 };
 347
 348 static bool
 349 ips_gpu_turbo_enabled(struct ips_driver *ips);
 350
 351 /**
 352  * ips_cpu_busy - is CPU busy?
 353  * @ips: IPS driver struct
 354  *
 355  * Check CPU for load to see whether we should increase its thermal budget.
 356  *
 357  * RETURNS:
 358  * True if the CPU could use more power, false otherwise.
 359  */
 360 static bool ips_cpu_busy(struct ips_driver *ips)
 361 {
 362         if ((avenrun[0] >> FSHIFT) > 1)
 363                 return true;
 364
 365         return false;
 366 }
 367
 368 /**
 369  * ips_cpu_raise - raise CPU power clamp
 370  * @ips: IPS driver struct
 371  *
 372  * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for
 373  * this platform.
 374  *
 375  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as
 376  * long as we haven't hit the TDP limit for the SKU).
 377  */
 378 static void ips_cpu_raise(struct ips_driver *ips)
 379 {
 380         u64 turbo_override;
 381         u16 cur_tdp_limit, new_tdp_limit;
 382
 383         if (!ips->cpu_turbo_enabled)
 384                 return;
 385
 386         rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 387
 388         cur_tdp_limit = turbo_override & TURBO_TDP_MASK;
 389         new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */
 390
 391         /* Clamp to SKU TDP limit */
 392         if (((new_tdp_limit * 10) / 8) > ips->core_power_limit)
 393                 new_tdp_limit = cur_tdp_limit;
 394
 395         thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8);
 396
 397         turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
 398         wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 399
 400         turbo_override &= ~TURBO_TDP_MASK;
 401         turbo_override |= new_tdp_limit;
 402
 403         wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 404 }
 405
 406 /**
 407  * ips_cpu_lower - lower CPU power clamp
 408  * @ips: IPS driver struct
 409  *
 410  * Lower CPU power clamp b %IPS_CPU_STEP if possible.
 411  *
 412  * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going
 413  * as low as the platform limits will allow (though we could go lower there
 414  * wouldn't be much point).
 415  */
 416 static void ips_cpu_lower(struct ips_driver *ips)
 417 {
 418         u64 turbo_override;
 419         u16 cur_limit, new_limit;
 420
 421         rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 422
 423         cur_limit = turbo_override & TURBO_TDP_MASK;
 424         new_limit = cur_limit - 8; /* 1W decrease */
 425
 426         /* Clamp to SKU TDP limit */
 427         if (new_limit  < (ips->orig_turbo_limit & TURBO_TDP_MASK))
 428                 new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK;
 429
 430         thm_writew(THM_MPCPC, (new_limit * 10) / 8);
 431
 432         turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN;
 433         wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 434
 435         turbo_override &= ~TURBO_TDP_MASK;
 436         turbo_override |= new_limit;
 437
 438         wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
 439 }
 440
 441 /**
 442  * do_enable_cpu_turbo - internal turbo enable function
 443  * @data: unused
 444  *
 445  * Internal function for actually updating MSRs.  When we enable/disable
 446  * turbo, we need to do it on each CPU; this function is the one called
 447  * by on_each_cpu() when needed.
 448  */
 449 static void do_enable_cpu_turbo(void *data)
 450 {
 451         u64 perf_ctl;
 452
 453         rdmsrl(IA32_PERF_CTL, perf_ctl);
 454         if (perf_ctl & IA32_PERF_TURBO_DIS) {
 455                 perf_ctl &= ~IA32_PERF_TURBO_DIS;
 456                 wrmsrl(IA32_PERF_CTL, perf_ctl);
 457         }
 458 }
 459
 460 /**
 461  * ips_enable_cpu_turbo - enable turbo mode on all CPUs
 462  * @ips: IPS driver struct
 463  *
 464  * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on
 465  * all logical threads.
 466  */
 467 static void ips_enable_cpu_turbo(struct ips_driver *ips)
 468 {
 469         /* Already on, no need to mess with MSRs */
 470         if (ips->__cpu_turbo_on)
 471                 return;
 472
 473         if (ips->turbo_toggle_allowed)
 474                 on_each_cpu(do_enable_cpu_turbo, ips, 1);
 475
 476         ips->__cpu_turbo_on = true;
 477 }
 478
 479 /**
 480  * do_disable_cpu_turbo - internal turbo disable function
 481  * @data: unused
 482  *
 483  * Internal function for actually updating MSRs.  When we enable/disable
 484  * turbo, we need to do it on each CPU; this function is the one called
 485  * by on_each_cpu() when needed.
 486  */
 487 static void do_disable_cpu_turbo(void *data)
 488 {
 489         u64 perf_ctl;
 490
 491         rdmsrl(IA32_PERF_CTL, perf_ctl);
 492         if (!(perf_ctl & IA32_PERF_TURBO_DIS)) {
 493                 perf_ctl |= IA32_PERF_TURBO_DIS;
 494                 wrmsrl(IA32_PERF_CTL, perf_ctl);
 495         }
 496 }
 497
 498 /**
 499  * ips_disable_cpu_turbo - disable turbo mode on all CPUs
 500  * @ips: IPS driver struct
 501  *
 502  * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on
 503  * all logical threads.
 504  */
 505 static void ips_disable_cpu_turbo(struct ips_driver *ips)
 506 {
 507         /* Already off, leave it */
 508         if (!ips->__cpu_turbo_on)
 509                 return;
 510
 511         if (ips->turbo_toggle_allowed)
 512                 on_each_cpu(do_disable_cpu_turbo, ips, 1);
 513
 514         ips->__cpu_turbo_on = false;
 515 }
 516
 517 /**
 518  * ips_gpu_busy - is GPU busy?
 519  * @ips: IPS driver struct
 520  *
 521  * Check GPU for load to see whether we should increase its thermal budget.
 522  * We need to call into the i915 driver in this case.
 523  *
 524  * RETURNS:
 525  * True if the GPU could use more power, false otherwise.
 526  */
 527 static bool ips_gpu_busy(struct ips_driver *ips)
 528 {
 529         if (!ips_gpu_turbo_enabled(ips))
 530                 return false;
 531
 532         return ips->gpu_busy();
 533 }
 534
 535 /**
 536  * ips_gpu_raise - raise GPU power clamp
 537  * @ips: IPS driver struct
 538  *
 539  * Raise the GPU frequency/power if possible.  We need to call into the
 540  * i915 driver in this case.
 541  */
 542 static void ips_gpu_raise(struct ips_driver *ips)
 543 {
 544         if (!ips_gpu_turbo_enabled(ips))
 545                 return;
 546
 547         if (!ips->gpu_raise())
 548                 ips->gpu_turbo_enabled = false;
 549
 550         return;
 551 }
 552
 553 /**
 554  * ips_gpu_lower - lower GPU power clamp
 555  * @ips: IPS driver struct
 556  *
 557  * Lower GPU frequency/power if possible.  Need to call i915.
 558  */
 559 static void ips_gpu_lower(struct ips_driver *ips)
 560 {
 561         if (!ips_gpu_turbo_enabled(ips))
 562                 return;
 563
 564         if (!ips->gpu_lower())
 565                 ips->gpu_turbo_enabled = false;
 566
 567         return;
 568 }
 569
 570 /**
 571  * ips_enable_gpu_turbo - notify the gfx driver turbo is available
 572  * @ips: IPS driver struct
 573  *
 574  * Call into the graphics driver indicating that it can safely use
 575  * turbo mode.
 576  */
 577 static void ips_enable_gpu_turbo(struct ips_driver *ips)
 578 {
 579         if (ips->__gpu_turbo_on)
 580                 return;
 581         ips->__gpu_turbo_on = true;
 582 }
 583
 584 /**
 585  * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode
 586  * @ips: IPS driver struct
 587  *
 588  * Request that the graphics driver disable turbo mode.
 589  */
 590 static void ips_disable_gpu_turbo(struct ips_driver *ips)
 591 {
 592         /* Avoid calling i915 if turbo is already disabled */
 593         if (!ips->__gpu_turbo_on)
 594                 return;
 595
 596         if (!ips->gpu_turbo_disable())
 597                 dev_err(&ips->dev->dev, "failed to disable graphics turbo\n");
 598         else
 599                 ips->__gpu_turbo_on = false;
 600 }
 601
 602 /**
 603  * mcp_exceeded - check whether we're outside our thermal & power limits
 604  * @ips: IPS driver struct
 605  *
 606  * Check whether the MCP is over its thermal or power budget.
 607  */
 608 static bool mcp_exceeded(struct ips_driver *ips)
 609 {
 610         unsigned long flags;
 611         bool ret = false;
 612         u32 temp_limit;
 613         u32 avg_power;
 614
 615         spin_lock_irqsave(&ips->turbo_status_lock, flags);
 616
 617         temp_limit = ips->mcp_temp_limit * 100;
 618         if (ips->mcp_avg_temp > temp_limit)
 619                 ret = true;
 620
 621         avg_power = ips->cpu_avg_power + ips->mch_avg_power;
 622         if (avg_power > ips->mcp_power_limit)
 623                 ret = true;
 624
 625         spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
 626
 627         return ret;
 628 }
 629
 630 /**
 631  * cpu_exceeded - check whether a CPU core is outside its limits
 632  * @ips: IPS driver struct
 633  * @cpu: CPU number to check
 634  *
 635  * Check a given CPU's average temp or power is over its limit.
 636  */
 637 static bool cpu_exceeded(struct ips_driver *ips, int cpu)
 638 {
 639         unsigned long flags;
 640         int avg;
 641         bool ret = false;
 642
 643         spin_lock_irqsave(&ips->turbo_status_lock, flags);
 644         avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp;
 645         if (avg > (ips->limits->core_temp_limit * 100))
 646                 ret = true;
 647         if (ips->cpu_avg_power > ips->core_power_limit * 100)
 648                 ret = true;
 649         spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
 650
 651         if (ret)
 652                 dev_info(&ips->dev->dev,
 653                          "CPU power or thermal limit exceeded\n");
 654
 655         return ret;
 656 }
 657
 658 /**
 659  * mch_exceeded - check whether the GPU is over budget
 660  * @ips: IPS driver struct
 661  *
 662  * Check the MCH temp & power against their maximums.
 663  */
 664 static bool mch_exceeded(struct ips_driver *ips)
 665 {
 666         unsigned long flags;
 667         bool ret = false;
 668
 669         spin_lock_irqsave(&ips->turbo_status_lock, flags);
 670         if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100))
 671                 ret = true;
 672         if (ips->mch_avg_power > ips->mch_power_limit)
 673                 ret = true;
 674         spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
 675
 676         return ret;
 677 }
 678
 679 /**
 680  * verify_limits - verify BIOS provided limits
 681  * @ips: IPS structure
 682  *
 683  * BIOS can optionally provide non-default limits for power and temp.  Check
 684  * them here and use the defaults if the BIOS values are not provided or
 685  * are otherwise unusable.
 686  */
 687 static void verify_limits(struct ips_driver *ips)
 688 {
 689         if (ips->mcp_power_limit < ips->limits->mcp_power_limit ||
 690             ips->mcp_power_limit > 35000)
 691                 ips->mcp_power_limit = ips->limits->mcp_power_limit;
 692
 693         if (ips->mcp_temp_limit < ips->limits->core_temp_limit ||
 694             ips->mcp_temp_limit < ips->limits->mch_temp_limit ||
 695             ips->mcp_temp_limit > 150)
 696                 ips->mcp_temp_limit = min(ips->limits->core_temp_limit,
 697                                           ips->limits->mch_temp_limit);
 698 }
 699
 700 /**
 701  * update_turbo_limits - get various limits & settings from regs
 702  * @ips: IPS driver struct
 703  *
 704  * Update the IPS power & temp limits, along with turbo enable flags,
 705  * based on latest register contents.
 706  *
 707  * Used at init time and for runtime BIOS support, which requires polling
 708  * the regs for updates (as a result of AC->DC transition for example).
 709  *
 710  * LOCKING:
 711  * Caller must hold turbo_status_lock (outside of init)
 712  */
 713 static void update_turbo_limits(struct ips_driver *ips)
 714 {
 715         u32 hts = thm_readl(THM_HTS);
 716
 717         ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS);
 718         /*
 719          * Disable turbo for now, until we can figure out why the power figures
 720          * are wrong
 721          */
 722         ips->cpu_turbo_enabled = false;
 723
 724         if (ips->gpu_busy)
 725                 ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS);
 726
 727         ips->core_power_limit = thm_readw(THM_MPCPC);
 728         ips->mch_power_limit = thm_readw(THM_MMGPC);
 729         ips->mcp_temp_limit = thm_readw(THM_PTL);
 730         ips->mcp_power_limit = thm_readw(THM_MPPC);
 731
 732         verify_limits(ips);
 733         /* Ignore BIOS CPU vs GPU pref */
 734 }
 735
 736 /**
 737  * ips_adjust - adjust power clamp based on thermal state
 738  * @data: ips driver structure
 739  *
 740  * Wake up every 5s or so and check whether we should adjust the power clamp.
 741  * Check CPU and GPU load to determine which needs adjustment.  There are
 742  * several things to consider here:
 743  *   - do we need to adjust up or down?
 744  *   - is CPU busy?
 745  *   - is GPU busy?
 746  *   - is CPU in turbo?
 747  *   - is GPU in turbo?
 748  *   - is CPU or GPU preferred? (CPU is default)
 749  *
 750  * So, given the above, we do the following:
 751  *   - up (TDP available)
 752  *     - CPU not busy, GPU not busy - nothing
 753  *     - CPU busy, GPU not busy - adjust CPU up
 754  *     - CPU not busy, GPU busy - adjust GPU up
 755  *     - CPU busy, GPU busy - adjust preferred unit up, taking headroom from
 756  *       non-preferred unit if necessary
 757  *   - down (at TDP limit)
 758  *     - adjust both CPU and GPU down if possible
 759  *
 760                 cpu+ gpu+       cpu+gpu-        cpu-gpu+        cpu-gpu-
 761 cpu < gpu <     cpu+gpu+        cpu+            gpu+            nothing
 762 cpu < gpu >=    cpu+gpu-(mcp<)  cpu+gpu-(mcp<)  gpu-            gpu-
 763 cpu >= gpu <    cpu-gpu+(mcp<)  cpu-            cpu-gpu+(mcp<)  cpu-
 764 cpu >= gpu >=   cpu-gpu-        cpu-gpu-        cpu-gpu-        cpu-gpu-
 765  *
 766  */
 767 static int ips_adjust(void *data)
 768 {
 769         struct ips_driver *ips = data;
 770         unsigned long flags;
 771
 772         dev_dbg(&ips->dev->dev, "starting ips-adjust thread\n");
 773
 774         /*
 775          * Adjust CPU and GPU clamps every 5s if needed.  Doing it more
 776          * often isn't recommended due to ME interaction.
 777          */
 778         do {
 779                 bool cpu_busy = ips_cpu_busy(ips);
 780                 bool gpu_busy = ips_gpu_busy(ips);
 781
 782                 spin_lock_irqsave(&ips->turbo_status_lock, flags);
 783                 if (ips->poll_turbo_status)
 784                         update_turbo_limits(ips);
 785                 spin_unlock_irqrestore(&ips->turbo_status_lock, flags);
 786
 787                 /* Update turbo status if necessary */
 788                 if (ips->cpu_turbo_enabled)
 789                         ips_enable_cpu_turbo(ips);
 790                 else
 791                         ips_disable_cpu_turbo(ips);
 792
 793                 if (ips->gpu_turbo_enabled)
 794                         ips_enable_gpu_turbo(ips);
 795                 else
 796                         ips_disable_gpu_turbo(ips);
 797
 798                 /* We're outside our comfort zone, crank them down */
 799                 if (mcp_exceeded(ips)) {
 800                         ips_cpu_lower(ips);
 801                         ips_gpu_lower(ips);
 802                         goto sleep;
 803                 }
 804
 805                 if (!cpu_exceeded(ips, 0) && cpu_busy)
 806                         ips_cpu_raise(ips);
 807                 else
 808                         ips_cpu_lower(ips);
 809
 810                 if (!mch_exceeded(ips) && gpu_busy)
 811                         ips_gpu_raise(ips);
 812                 else
 813                         ips_gpu_lower(ips);
 814
 815 sleep:
 816                 schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD));
 817         } while (!kthread_should_stop());
 818
 819         dev_dbg(&ips->dev->dev, "ips-adjust thread stopped\n");
 820
 821         return 0;
 822 }
 823
 824 /*
 825  * Helpers for reading out temp/power values and calculating their
 826  * averages for the decision making and monitoring functions.
 827  */
 828
 829 static u16 calc_avg_temp(struct ips_driver *ips, u16 *array)
 830 {
 831         u64 total = 0;
 832         int i;
 833         u16 avg;
 834
 835         for (i = 0; i < IPS_SAMPLE_COUNT; i++)
 836                 total += (u64)(array[i] * 100);
 837
 838         do_div(total, IPS_SAMPLE_COUNT);
 839
 840         avg = (u16)total;
 841
 842         return avg;
 843 }
 844
 845 static u16 read_mgtv(struct ips_driver *ips)
 846 {
 847         u16 ret;
 848         u64 slope, offset;
 849         u64 val;
 850
 851         val = thm_readq(THM_MGTV);
 852         val = (val & TV_MASK) >> TV_SHIFT;
 853
 854         slope = offset = thm_readw(THM_MGTA);
 855         slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT;
 856         offset = offset & MGTA_OFFSET_MASK;
 857
 858         ret = ((val * slope + 0x40) >> 7) + offset;
 859
 860         return 0; /* MCH temp reporting buggy */
 861 }
 862
 863 static u16 read_ptv(struct ips_driver *ips)
 864 {
 865         u16 val, slope, offset;
 866
 867         slope = (ips->pta_val & PTA_SLOPE_MASK) >> PTA_SLOPE_SHIFT;
 868         offset = ips->pta_val & PTA_OFFSET_MASK;
 869
 870         val = thm_readw(THM_PTV) & PTV_MASK;
 871
 872         return val;
 873 }
 874
 875 static u16 read_ctv(struct ips_driver *ips, int cpu)
 876 {
 877         int reg = cpu ? THM_CTV2 : THM_CTV1;
 878         u16 val;
 879
 880         val = thm_readw(reg);
 881         if (!(val & CTV_TEMP_ERROR))
 882                 val = (val) >> 6; /* discard fractional component */
 883         else
 884                 val = 0;
 885
 886         return val;
 887 }
 888
 889 static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period)
 890 {
 891         u32 val;
 892         u32 ret;
 893
 894         /*
 895          * CEC is in joules/65535.  Take difference over time to
 896          * get watts.
 897          */
 898         val = thm_readl(THM_CEC);
 899
 900         /* period is in ms and we want mW */
 901         ret = (((val - *last) * 1000) / period);
 902         ret = (ret * 1000) / 65535;
 903         *last = val;
 904
 905         return 0;
 906 }
 907
 908 static const u16 temp_decay_factor = 2;
 909 static u16 update_average_temp(u16 avg, u16 val)
 910 {
 911         u16 ret;
 912
 913         /* Multiply by 100 for extra precision */
 914         ret = (val * 100 / temp_decay_factor) +
 915                 (((temp_decay_factor - 1) * avg) / temp_decay_factor);
 916         return ret;
 917 }
 918
 919 static const u16 power_decay_factor = 2;
 920 static u16 update_average_power(u32 avg, u32 val)
 921 {
 922         u32 ret;
 923
 924         ret = (val / power_decay_factor) +
 925                 (((power_decay_factor - 1) * avg) / power_decay_factor);
 926
 927         return ret;
 928 }
 929
 930 static u32 calc_avg_power(struct ips_driver *ips, u32 *array)
 931 {
 932         u64 total = 0;
 933         u32 avg;
 934         int i;
 935
 936         for (i = 0; i < IPS_SAMPLE_COUNT; i++)
 937                 total += array[i];
 938
 939         do_div(total, IPS_SAMPLE_COUNT);
 940         avg = (u32)total;
 941
 942         return avg;
 943 }
 944
 945 static void monitor_timeout(unsigned long arg)
 946 {
 947         wake_up_process((struct task_struct *)arg);
 948 }
 949
 950 /**
 951  * ips_monitor - temp/power monitoring thread
 952  * @data: ips driver structure
 953  *
 954  * This is the main function for the IPS driver.  It monitors power and
 955  * tempurature in the MCP and adjusts CPU and GPU power clams accordingly.
 956  *
 957  * We keep a 5s moving average of power consumption and tempurature.  Using
 958  * that data, along with CPU vs GPU preference, we adjust the power clamps
 959  * up or down.
 960  */
 961 static int ips_monitor(void *data)
 962 {
 963         struct ips_driver *ips = data;
 964         struct timer_list timer;
 965         unsigned long seqno_timestamp, expire, last_msecs, last_sample_period;
 966         int i;
 967         u32 *cpu_samples, *mchp_samples, old_cpu_power;
 968         u16 *mcp_samples, *ctv1_samples, *ctv2_samples, *mch_samples;
 969         u8 cur_seqno, last_seqno;
 970
 971         mcp_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 972         ctv1_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 973         ctv2_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 974         mch_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 975         cpu_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 976         mchp_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL);
 977         if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples ||
 978                         !cpu_samples || !mchp_samples) {
 979                 dev_err(&ips->dev->dev,
 980                         "failed to allocate sample array, ips disabled\n");
 981                 kfree(mcp_samples);
 982                 kfree(ctv1_samples);
 983                 kfree(ctv2_samples);
 984                 kfree(mch_samples);
 985                 kfree(cpu_samples);
 986                 kfree(mchp_samples);
 987                 return -ENOMEM;
 988         }
 989
 990         last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
 991                 ITV_ME_SEQNO_SHIFT;
 992         seqno_timestamp = get_jiffies_64();
 993
 994         old_cpu_power = thm_readl(THM_CEC);
 995         schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
 996
 997         /* Collect an initial average */
 998         for (i = 0; i < IPS_SAMPLE_COUNT; i++) {
 999                 u32 mchp, cpu_power;
1000                 u16 val;
1001
1002                 mcp_samples[i] = read_ptv(ips);
1003
1004                 val = read_ctv(ips, 0);
1005                 ctv1_samples[i] = val;
1006
1007                 val = read_ctv(ips, 1);
1008                 ctv2_samples[i] = val;
1009
1010                 val = read_mgtv(ips);
1011                 mch_samples[i] = val;
1012
1013                 cpu_power = get_cpu_power(ips, &old_cpu_power,
1014                                           IPS_SAMPLE_PERIOD);
1015                 cpu_samples[i] = cpu_power;
1016
1017                 if (ips->read_mch_val) {
1018                         mchp = ips->read_mch_val();
1019                         mchp_samples[i] = mchp;
1020                 }
1021
1022                 schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1023                 if (kthread_should_stop())
1024                         break;
1025         }
1026
1027         ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples);
1028         ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples);
1029         ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples);
1030         ips->mch_avg_temp = calc_avg_temp(ips, mch_samples);
1031         ips->cpu_avg_power = calc_avg_power(ips, cpu_samples);
1032         ips->mch_avg_power = calc_avg_power(ips, mchp_samples);
1033         kfree(mcp_samples);
1034         kfree(ctv1_samples);
1035         kfree(ctv2_samples);
1036         kfree(mch_samples);
1037         kfree(cpu_samples);
1038         kfree(mchp_samples);
1039
1040         /* Start the adjustment thread now that we have data */
1041         wake_up_process(ips->adjust);
1042
1043         /*
1044          * Ok, now we have an initial avg.  From here on out, we track the
1045          * running avg using a decaying average calculation.  This allows
1046          * us to reduce the sample frequency if the CPU and GPU are idle.
1047          */
1048         old_cpu_power = thm_readl(THM_CEC);
1049         schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD));
1050         last_sample_period = IPS_SAMPLE_PERIOD;
1051
1052         setup_deferrable_timer_on_stack(&timer, monitor_timeout,
1053                                         (unsigned long)current);
1054         do {
1055                 u32 cpu_val, mch_val;
1056                 u16 val;
1057
1058                 /* MCP itself */
1059                 val = read_ptv(ips);
1060                 ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val);
1061
1062                 /* Processor 0 */
1063                 val = read_ctv(ips, 0);
1064                 ips->ctv1_avg_temp =
1065                         update_average_temp(ips->ctv1_avg_temp, val);
1066                 /* Power */
1067                 cpu_val = get_cpu_power(ips, &old_cpu_power,
1068                                         last_sample_period);
1069                 ips->cpu_avg_power =
1070                         update_average_power(ips->cpu_avg_power, cpu_val);
1071
1072                 if (ips->second_cpu) {
1073                         /* Processor 1 */
1074                         val = read_ctv(ips, 1);
1075                         ips->ctv2_avg_temp =
1076                                 update_average_temp(ips->ctv2_avg_temp, val);
1077                 }
1078
1079                 /* MCH */
1080                 val = read_mgtv(ips);
1081                 ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val);
1082                 /* Power */
1083                 if (ips->read_mch_val) {
1084                         mch_val = ips->read_mch_val();
1085                         ips->mch_avg_power =
1086                                 update_average_power(ips->mch_avg_power,
1087                                                      mch_val);
1088                 }
1089
1090                 /*
1091                  * Make sure ME is updating thermal regs.
1092                  * Note:
1093                  * If it's been more than a second since the last update,
1094                  * the ME is probably hung.
1095                  */
1096                 cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >>
1097                         ITV_ME_SEQNO_SHIFT;
1098                 if (cur_seqno == last_seqno &&
1099                     time_after(jiffies, seqno_timestamp + HZ)) {
1100                         dev_warn(&ips->dev->dev, "ME failed to update for more than 1s, likely hung\n");
1101                 } else {
1102                         seqno_timestamp = get_jiffies_64();
1103                         last_seqno = cur_seqno;
1104                 }
1105
1106                 last_msecs = jiffies_to_msecs(jiffies);
1107                 expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD);
1108
1109                 __set_current_state(TASK_INTERRUPTIBLE);
1110                 mod_timer(&timer, expire);
1111                 schedule();
1112
1113                 /* Calculate actual sample period for power averaging */
1114                 last_sample_period = jiffies_to_msecs(jiffies) - last_msecs;
1115                 if (!last_sample_period)
1116                         last_sample_period = 1;
1117         } while (!kthread_should_stop());
1118
1119         del_timer_sync(&timer);
1120         destroy_timer_on_stack(&timer);
1121
1122         dev_dbg(&ips->dev->dev, "ips-monitor thread stopped\n");
1123
1124         return 0;
1125 }
1126
1127 #if 0
1128 #define THM_DUMPW(reg) \
1129         { \
1130         u16 val = thm_readw(reg); \
1131         dev_dbg(&ips->dev->dev, #reg ": 0x%04x\n", val); \
1132         }
1133 #define THM_DUMPL(reg) \
1134         { \
1135         u32 val = thm_readl(reg); \
1136         dev_dbg(&ips->dev->dev, #reg ": 0x%08x\n", val); \
1137         }
1138 #define THM_DUMPQ(reg) \
1139         { \
1140         u64 val = thm_readq(reg); \
1141         dev_dbg(&ips->dev->dev, #reg ": 0x%016x\n", val); \
1142         }
1143
1144 static void dump_thermal_info(struct ips_driver *ips)
1145 {
1146         u16 ptl;
1147
1148         ptl = thm_readw(THM_PTL);
1149         dev_dbg(&ips->dev->dev, "Processor temp limit: %d\n", ptl);
1150
1151         THM_DUMPW(THM_CTA);
1152         THM_DUMPW(THM_TRC);
1153         THM_DUMPW(THM_CTV1);
1154         THM_DUMPL(THM_STS);
1155         THM_DUMPW(THM_PTV);
1156         THM_DUMPQ(THM_MGTV);
1157 }
1158 #endif
1159
1160 /**
1161  * ips_irq_handler - handle temperature triggers and other IPS events
1162  * @irq: irq number
1163  * @arg: unused
1164  *
1165  * Handle temperature limit trigger events, generally by lowering the clamps.
1166  * If we're at a critical limit, we clamp back to the lowest possible value
1167  * to prevent emergency shutdown.
1168  */
1169 static irqreturn_t ips_irq_handler(int irq, void *arg)
1170 {
1171         struct ips_driver *ips = arg;
1172         u8 tses = thm_readb(THM_TSES);
1173         u8 tes = thm_readb(THM_TES);
1174
1175         if (!tses && !tes)
1176                 return IRQ_NONE;
1177
1178         dev_info(&ips->dev->dev, "TSES: 0x%02x\n", tses);
1179         dev_info(&ips->dev->dev, "TES: 0x%02x\n", tes);
1180
1181         /* STS update from EC? */
1182         if (tes & 1) {
1183                 u32 sts, tc1;
1184
1185                 sts = thm_readl(THM_STS);
1186                 tc1 = thm_readl(THM_TC1);
1187
1188                 if (sts & STS_NVV) {
1189                         spin_lock(&ips->turbo_status_lock);
1190                         ips->core_power_limit = (sts & STS_PCPL_MASK) >>
1191                                 STS_PCPL_SHIFT;
1192                         ips->mch_power_limit = (sts & STS_GPL_MASK) >>
1193                                 STS_GPL_SHIFT;
1194                         /* ignore EC CPU vs GPU pref */
1195                         ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS);
1196                         /*
1197                          * Disable turbo for now, until we can figure
1198                          * out why the power figures are wrong
1199                          */
1200                         ips->cpu_turbo_enabled = false;
1201                         if (ips->gpu_busy)
1202                                 ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS);
1203                         ips->mcp_temp_limit = (sts & STS_PTL_MASK) >>
1204                                 STS_PTL_SHIFT;
1205                         ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >>
1206                                 STS_PPL_SHIFT;
1207                         verify_limits(ips);
1208                         spin_unlock(&ips->turbo_status_lock);
1209
1210                         thm_writeb(THM_SEC, SEC_ACK);
1211                 }
1212                 thm_writeb(THM_TES, tes);
1213         }
1214
1215         /* Thermal trip */
1216         if (tses) {
1217                 dev_warn(&ips->dev->dev,
1218                          "thermal trip occurred, tses: 0x%04x\n", tses);
1219                 thm_writeb(THM_TSES, tses);
1220         }
1221
1222         return IRQ_HANDLED;
1223 }
1224
1225 #ifndef CONFIG_DEBUG_FS
1226 static void ips_debugfs_init(struct ips_driver *ips) { return; }
1227 static void ips_debugfs_cleanup(struct ips_driver *ips) { return; }
1228 #else
1229
1230 /* Expose current state and limits in debugfs if possible */
1231
1232 struct ips_debugfs_node {
1233         struct ips_driver *ips;
1234         char *name;
1235         int (*show)(struct seq_file *m, void *data);
1236 };
1237
1238 static int show_cpu_temp(struct seq_file *m, void *data)
1239 {
1240         struct ips_driver *ips = m->private;
1241
1242         seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100,
1243                    ips->ctv1_avg_temp % 100);
1244
1245         return 0;
1246 }
1247
1248 static int show_cpu_power(struct seq_file *m, void *data)
1249 {
1250         struct ips_driver *ips = m->private;
1251
1252         seq_printf(m, "%dmW\n", ips->cpu_avg_power);
1253
1254         return 0;
1255 }
1256
1257 static int show_cpu_clamp(struct seq_file *m, void *data)
1258 {
1259         u64 turbo_override;
1260         int tdp, tdc;
1261
1262         rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1263
1264         tdp = (int)(turbo_override & TURBO_TDP_MASK);
1265         tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT);
1266
1267         /* Convert to .1W/A units */
1268         tdp = tdp * 10 / 8;
1269         tdc = tdc * 10 / 8;
1270
1271         /* Watts Amperes */
1272         seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10,
1273                    tdc / 10, tdc % 10);
1274
1275         return 0;
1276 }
1277
1278 static int show_mch_temp(struct seq_file *m, void *data)
1279 {
1280         struct ips_driver *ips = m->private;
1281
1282         seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100,
1283                    ips->mch_avg_temp % 100);
1284
1285         return 0;
1286 }
1287
1288 static int show_mch_power(struct seq_file *m, void *data)
1289 {
1290         struct ips_driver *ips = m->private;
1291
1292         seq_printf(m, "%dmW\n", ips->mch_avg_power);
1293
1294         return 0;
1295 }
1296
1297 static struct ips_debugfs_node ips_debug_files[] = {
1298         { NULL, "cpu_temp", show_cpu_temp },
1299         { NULL, "cpu_power", show_cpu_power },
1300         { NULL, "cpu_clamp", show_cpu_clamp },
1301         { NULL, "mch_temp", show_mch_temp },
1302         { NULL, "mch_power", show_mch_power },
1303 };
1304
1305 static int ips_debugfs_open(struct inode *inode, struct file *file)
1306 {
1307         struct ips_debugfs_node *node = inode->i_private;
1308
1309         return single_open(file, node->show, node->ips);
1310 }
1311
1312 static const struct file_operations ips_debugfs_ops = {
1313         .owner = THIS_MODULE,
1314         .open = ips_debugfs_open,
1315         .read = seq_read,
1316         .llseek = seq_lseek,
1317         .release = single_release,
1318 };
1319
1320 static void ips_debugfs_cleanup(struct ips_driver *ips)
1321 {
1322         if (ips->debug_root)
1323                 debugfs_remove_recursive(ips->debug_root);
1324         return;
1325 }
1326
1327 static void ips_debugfs_init(struct ips_driver *ips)
1328 {
1329         int i;
1330
1331         ips->debug_root = debugfs_create_dir("ips", NULL);
1332         if (!ips->debug_root) {
1333                 dev_err(&ips->dev->dev,
1334                         "failed to create debugfs entries: %ld\n",
1335                         PTR_ERR(ips->debug_root));
1336                 return;
1337         }
1338
1339         for (i = 0; i < ARRAY_SIZE(ips_debug_files); i++) {
1340                 struct dentry *ent;
1341                 struct ips_debugfs_node *node = &ips_debug_files[i];
1342
1343                 node->ips = ips;
1344                 ent = debugfs_create_file(node->name, S_IFREG | S_IRUGO,
1345                                           ips->debug_root, node,
1346                                           &ips_debugfs_ops);
1347                 if (!ent) {
1348                         dev_err(&ips->dev->dev,
1349                                 "failed to create debug file: %ld\n",
1350                                 PTR_ERR(ent));
1351                         goto err_cleanup;
1352                 }
1353         }
1354
1355         return;
1356
1357 err_cleanup:
1358         ips_debugfs_cleanup(ips);
1359         return;
1360 }
1361 #endif /* CONFIG_DEBUG_FS */
1362
1363 /**
1364  * ips_detect_cpu - detect whether CPU supports IPS
1365  *
1366  * Walk our list and see if we're on a supported CPU.  If we find one,
1367  * return the limits for it.
1368  */
1369 static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips)
1370 {
1371         u64 turbo_power, misc_en;
1372         struct ips_mcp_limits *limits = NULL;
1373         u16 tdp;
1374
1375         if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) {
1376                 dev_info(&ips->dev->dev, "Non-IPS CPU detected.\n");
1377                 goto out;
1378         }
1379
1380         rdmsrl(IA32_MISC_ENABLE, misc_en);
1381         /*
1382          * If the turbo enable bit isn't set, we shouldn't try to enable/disable
1383          * turbo manually or we'll get an illegal MSR access, even though
1384          * turbo will still be available.
1385          */
1386         if (misc_en & IA32_MISC_TURBO_EN)
1387                 ips->turbo_toggle_allowed = true;
1388         else
1389                 ips->turbo_toggle_allowed = false;
1390
1391         if (strstr(boot_cpu_data.x86_model_id, "CPU       M"))
1392                 limits = &ips_sv_limits;
1393         else if (strstr(boot_cpu_data.x86_model_id, "CPU       L"))
1394                 limits = &ips_lv_limits;
1395         else if (strstr(boot_cpu_data.x86_model_id, "CPU       U"))
1396                 limits = &ips_ulv_limits;
1397         else {
1398                 dev_info(&ips->dev->dev, "No CPUID match found.\n");
1399                 goto out;
1400         }
1401
1402         rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power);
1403         tdp = turbo_power & TURBO_TDP_MASK;
1404
1405         /* Sanity check TDP against CPU */
1406         if (limits->core_power_limit != (tdp / 8) * 1000) {
1407                 dev_info(&ips->dev->dev, "CPU TDP doesn't match expected value (found %d, expected %d)\n",
1408                          tdp / 8, limits->core_power_limit / 1000);
1409                 limits->core_power_limit = (tdp / 8) * 1000;
1410         }
1411
1412 out:
1413         return limits;
1414 }
1415
1416 /**
1417  * ips_get_i915_syms - try to get GPU control methods from i915 driver
1418  * @ips: IPS driver
1419  *
1420  * The i915 driver exports several interfaces to allow the IPS driver to
1421  * monitor and control graphics turbo mode.  If we can find them, we can
1422  * enable graphics turbo, otherwise we must disable it to avoid exceeding
1423  * thermal and power limits in the MCP.
1424  */
1425 static bool ips_get_i915_syms(struct ips_driver *ips)
1426 {
1427         ips->read_mch_val = symbol_get(i915_read_mch_val);
1428         if (!ips->read_mch_val)
1429                 goto out_err;
1430         ips->gpu_raise = symbol_get(i915_gpu_raise);
1431         if (!ips->gpu_raise)
1432                 goto out_put_mch;
1433         ips->gpu_lower = symbol_get(i915_gpu_lower);
1434         if (!ips->gpu_lower)
1435                 goto out_put_raise;
1436         ips->gpu_busy = symbol_get(i915_gpu_busy);
1437         if (!ips->gpu_busy)
1438                 goto out_put_lower;
1439         ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable);
1440         if (!ips->gpu_turbo_disable)
1441                 goto out_put_busy;
1442
1443         return true;
1444
1445 out_put_busy:
1446         symbol_put(i915_gpu_busy);
1447 out_put_lower:
1448         symbol_put(i915_gpu_lower);
1449 out_put_raise:
1450         symbol_put(i915_gpu_raise);
1451 out_put_mch:
1452         symbol_put(i915_read_mch_val);
1453 out_err:
1454         return false;
1455 }
1456
1457 static bool
1458 ips_gpu_turbo_enabled(struct ips_driver *ips)
1459 {
1460         if (!ips->gpu_busy && late_i915_load) {
1461                 if (ips_get_i915_syms(ips)) {
1462                         dev_info(&ips->dev->dev,
1463                                  "i915 driver attached, reenabling gpu turbo\n");
1464                         ips->gpu_turbo_enabled = !(thm_readl(THM_HTS) & HTS_GTD_DIS);
1465                 }
1466         }
1467
1468         return ips->gpu_turbo_enabled;
1469 }
1470
1471 void
1472 ips_link_to_i915_driver(void)
1473 {
1474         /* We can't cleanly get at the various ips_driver structs from
1475          * this caller (the i915 driver), so just set a flag saying
1476          * that it's time to try getting the symbols again.
1477          */
1478         late_i915_load = true;
1479 }
1480 EXPORT_SYMBOL_GPL(ips_link_to_i915_driver);
1481
1482 static const struct pci_device_id ips_id_table[] = {
1483         { PCI_DEVICE(PCI_VENDOR_ID_INTEL,
1484                      PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), },
1485         { 0, }
1486 };
1487
1488 MODULE_DEVICE_TABLE(pci, ips_id_table);
1489
1490 static int ips_blacklist_callback(const struct dmi_system_id *id)
1491 {
1492         pr_info("Blacklisted intel_ips for %s\n", id->ident);
1493         return 1;
1494 }
1495
1496 static const struct dmi_system_id ips_blacklist[] = {
1497         {
1498                 .callback = ips_blacklist_callback,
1499                 .ident = "HP ProBook",
1500                 .matches = {
1501                         DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1502                         DMI_MATCH(DMI_PRODUCT_NAME, "HP ProBook"),
1503                 },
1504         },
1505         { }     /* terminating entry */
1506 };
1507
1508 static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id)
1509 {
1510         u64 platform_info;
1511         struct ips_driver *ips;
1512         u32 hts;
1513         int ret = 0;
1514         u16 htshi, trc, trc_required_mask;
1515         u8 tse;
1516
1517         if (dmi_check_system(ips_blacklist))
1518                 return -ENODEV;
1519
1520         ips = kzalloc(sizeof(struct ips_driver), GFP_KERNEL);
1521         if (!ips)
1522                 return -ENOMEM;
1523
1524         pci_set_drvdata(dev, ips);
1525         ips->dev = dev;
1526
1527         ips->limits = ips_detect_cpu(ips);
1528         if (!ips->limits) {
1529                 dev_info(&dev->dev, "IPS not supported on this CPU\n");
1530                 ret = -ENXIO;
1531                 goto error_free;
1532         }
1533
1534         spin_lock_init(&ips->turbo_status_lock);
1535
1536         ret = pci_enable_device(dev);
1537         if (ret) {
1538                 dev_err(&dev->dev, "can't enable PCI device, aborting\n");
1539                 goto error_free;
1540         }
1541
1542         if (!pci_resource_start(dev, 0)) {
1543                 dev_err(&dev->dev, "TBAR not assigned, aborting\n");
1544                 ret = -ENXIO;
1545                 goto error_free;
1546         }
1547
1548         ret = pci_request_regions(dev, "ips thermal sensor");
1549         if (ret) {
1550                 dev_err(&dev->dev, "thermal resource busy, aborting\n");
1551                 goto error_free;
1552         }
1553
1554
1555         ips->regmap = ioremap(pci_resource_start(dev, 0),
1556                               pci_resource_len(dev, 0));
1557         if (!ips->regmap) {
1558                 dev_err(&dev->dev, "failed to map thermal regs, aborting\n");
1559                 ret = -EBUSY;
1560                 goto error_release;
1561         }
1562
1563         tse = thm_readb(THM_TSE);
1564         if (tse != TSE_EN) {
1565                 dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse);
1566                 ret = -ENXIO;
1567                 goto error_unmap;
1568         }
1569
1570         trc = thm_readw(THM_TRC);
1571         trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN;
1572         if ((trc & trc_required_mask) != trc_required_mask) {
1573                 dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n");
1574                 ret = -ENXIO;
1575                 goto error_unmap;
1576         }
1577
1578         if (trc & TRC_CORE2_EN)
1579                 ips->second_cpu = true;
1580
1581         update_turbo_limits(ips);
1582         dev_dbg(&dev->dev, "max cpu power clamp: %dW\n",
1583                 ips->mcp_power_limit / 10);
1584         dev_dbg(&dev->dev, "max core power clamp: %dW\n",
1585                 ips->core_power_limit / 10);
1586         /* BIOS may update limits at runtime */
1587         if (thm_readl(THM_PSC) & PSP_PBRT)
1588                 ips->poll_turbo_status = true;
1589
1590         if (!ips_get_i915_syms(ips)) {
1591                 dev_info(&dev->dev, "failed to get i915 symbols, graphics turbo disabled until i915 loads\n");
1592                 ips->gpu_turbo_enabled = false;
1593         } else {
1594                 dev_dbg(&dev->dev, "graphics turbo enabled\n");
1595                 ips->gpu_turbo_enabled = true;
1596         }
1597
1598         /*
1599          * Check PLATFORM_INFO MSR to make sure this chip is
1600          * turbo capable.
1601          */
1602         rdmsrl(PLATFORM_INFO, platform_info);
1603         if (!(platform_info & PLATFORM_TDP)) {
1604                 dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n");
1605                 ret = -ENODEV;
1606                 goto error_unmap;
1607         }
1608
1609         /*
1610          * IRQ handler for ME interaction
1611          * Note: don't use MSI here as the PCH has bugs.
1612          */
1613         pci_disable_msi(dev);
1614         ret = request_irq(dev->irq, ips_irq_handler, IRQF_SHARED, "ips",
1615                           ips);
1616         if (ret) {
1617                 dev_err(&dev->dev, "request irq failed, aborting\n");
1618                 goto error_unmap;
1619         }
1620
1621         /* Enable aux, hot & critical interrupts */
1622         thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI |
1623                    TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI);
1624         thm_writeb(THM_TEN, TEN_UPDATE_EN);
1625
1626         /* Collect adjustment values */
1627         ips->cta_val = thm_readw(THM_CTA);
1628         ips->pta_val = thm_readw(THM_PTA);
1629         ips->mgta_val = thm_readw(THM_MGTA);
1630
1631         /* Save turbo limits & ratios */
1632         rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1633
1634         ips_disable_cpu_turbo(ips);
1635         ips->cpu_turbo_enabled = false;
1636
1637         /* Create thermal adjust thread */
1638         ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust");
1639         if (IS_ERR(ips->adjust)) {
1640                 dev_err(&dev->dev,
1641                         "failed to create thermal adjust thread, aborting\n");
1642                 ret = -ENOMEM;
1643                 goto error_free_irq;
1644
1645         }
1646
1647         /*
1648          * Set up the work queue and monitor thread. The monitor thread
1649          * will wake up ips_adjust thread.
1650          */
1651         ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor");
1652         if (IS_ERR(ips->monitor)) {
1653                 dev_err(&dev->dev,
1654                         "failed to create thermal monitor thread, aborting\n");
1655                 ret = -ENOMEM;
1656                 goto error_thread_cleanup;
1657         }
1658
1659         hts = (ips->core_power_limit << HTS_PCPL_SHIFT) |
1660                 (ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV;
1661         htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT;
1662
1663         thm_writew(THM_HTSHI, htshi);
1664         thm_writel(THM_HTS, hts);
1665
1666         ips_debugfs_init(ips);
1667
1668         dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n",
1669                  ips->mcp_temp_limit);
1670         return ret;
1671
1672 error_thread_cleanup:
1673         kthread_stop(ips->adjust);
1674 error_free_irq:
1675         free_irq(ips->dev->irq, ips);
1676 error_unmap:
1677         iounmap(ips->regmap);
1678 error_release:
1679         pci_release_regions(dev);
1680 error_free:
1681         kfree(ips);
1682         return ret;
1683 }
1684
1685 static void ips_remove(struct pci_dev *dev)
1686 {
1687         struct ips_driver *ips = pci_get_drvdata(dev);
1688         u64 turbo_override;
1689
1690         if (!ips)
1691                 return;
1692
1693         ips_debugfs_cleanup(ips);
1694
1695         /* Release i915 driver */
1696         if (ips->read_mch_val)
1697                 symbol_put(i915_read_mch_val);
1698         if (ips->gpu_raise)
1699                 symbol_put(i915_gpu_raise);
1700         if (ips->gpu_lower)
1701                 symbol_put(i915_gpu_lower);
1702         if (ips->gpu_busy)
1703                 symbol_put(i915_gpu_busy);
1704         if (ips->gpu_turbo_disable)
1705                 symbol_put(i915_gpu_turbo_disable);
1706
1707         rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1708         turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN);
1709         wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override);
1710         wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit);
1711
1712         free_irq(ips->dev->irq, ips);
1713         if (ips->adjust)
1714                 kthread_stop(ips->adjust);
1715         if (ips->monitor)
1716                 kthread_stop(ips->monitor);
1717         iounmap(ips->regmap);
1718         pci_release_regions(dev);
1719         kfree(ips);
1720         dev_dbg(&dev->dev, "IPS driver removed\n");
1721 }
1722
1723 static void ips_shutdown(struct pci_dev *dev)
1724 {
1725 }
1726
1727 static struct pci_driver ips_pci_driver = {
1728         .name = "intel ips",
1729         .id_table = ips_id_table,
1730         .probe = ips_probe,
1731         .remove = ips_remove,
1732         .shutdown = ips_shutdown,
1733 };
1734
1735 module_pci_driver(ips_pci_driver);
1736
1737 MODULE_LICENSE("GPL");
1738 MODULE_AUTHOR("Jesse Barnes <jbarnes@virtuousgeek.org>");
1739 MODULE_DESCRIPTION("Intelligent Power Sharing Driver");