xref: /linux/drivers/acpi/cppc_acpi.c (revision 1e525507)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * CPPC (Collaborative Processor Performance Control) methods used by CPUfreq drivers.
4  *
5  * (C) Copyright 2014, 2015 Linaro Ltd.
6  * Author: Ashwin Chaugule <ashwin.chaugule@linaro.org>
7  *
8  * CPPC describes a few methods for controlling CPU performance using
9  * information from a per CPU table called CPC. This table is described in
10  * the ACPI v5.0+ specification. The table consists of a list of
11  * registers which may be memory mapped or hardware registers and also may
12  * include some static integer values.
13  *
14  * CPU performance is on an abstract continuous scale as against a discretized
15  * P-state scale which is tied to CPU frequency only. In brief, the basic
16  * operation involves:
17  *
18  * - OS makes a CPU performance request. (Can provide min and max bounds)
19  *
20  * - Platform (such as BMC) is free to optimize request within requested bounds
21  *   depending on power/thermal budgets etc.
22  *
23  * - Platform conveys its decision back to OS
24  *
25  * The communication between OS and platform occurs through another medium
26  * called (PCC) Platform Communication Channel. This is a generic mailbox like
27  * mechanism which includes doorbell semantics to indicate register updates.
28  * See drivers/mailbox/pcc.c for details on PCC.
29  *
30  * Finer details about the PCC and CPPC spec are available in the ACPI v5.1 and
31  * above specifications.
32  */
33 
34 #define pr_fmt(fmt)	"ACPI CPPC: " fmt
35 
36 #include <linux/delay.h>
37 #include <linux/iopoll.h>
38 #include <linux/ktime.h>
39 #include <linux/rwsem.h>
40 #include <linux/wait.h>
41 #include <linux/topology.h>
42 #include <linux/dmi.h>
43 #include <linux/units.h>
44 #include <asm/unaligned.h>
45 
46 #include <acpi/cppc_acpi.h>
47 
48 struct cppc_pcc_data {
49 	struct pcc_mbox_chan *pcc_channel;
50 	void __iomem *pcc_comm_addr;
51 	bool pcc_channel_acquired;
52 	unsigned int deadline_us;
53 	unsigned int pcc_mpar, pcc_mrtt, pcc_nominal;
54 
55 	bool pending_pcc_write_cmd;	/* Any pending/batched PCC write cmds? */
56 	bool platform_owns_pcc;		/* Ownership of PCC subspace */
57 	unsigned int pcc_write_cnt;	/* Running count of PCC write commands */
58 
59 	/*
60 	 * Lock to provide controlled access to the PCC channel.
61 	 *
62 	 * For performance critical usecases(currently cppc_set_perf)
63 	 *	We need to take read_lock and check if channel belongs to OSPM
64 	 * before reading or writing to PCC subspace
65 	 *	We need to take write_lock before transferring the channel
66 	 * ownership to the platform via a Doorbell
67 	 *	This allows us to batch a number of CPPC requests if they happen
68 	 * to originate in about the same time
69 	 *
70 	 * For non-performance critical usecases(init)
71 	 *	Take write_lock for all purposes which gives exclusive access
72 	 */
73 	struct rw_semaphore pcc_lock;
74 
75 	/* Wait queue for CPUs whose requests were batched */
76 	wait_queue_head_t pcc_write_wait_q;
77 	ktime_t last_cmd_cmpl_time;
78 	ktime_t last_mpar_reset;
79 	int mpar_count;
80 	int refcount;
81 };
82 
83 /* Array to represent the PCC channel per subspace ID */
84 static struct cppc_pcc_data *pcc_data[MAX_PCC_SUBSPACES];
85 /* The cpu_pcc_subspace_idx contains per CPU subspace ID */
86 static DEFINE_PER_CPU(int, cpu_pcc_subspace_idx);
87 
88 /*
89  * The cpc_desc structure contains the ACPI register details
90  * as described in the per CPU _CPC tables. The details
91  * include the type of register (e.g. PCC, System IO, FFH etc.)
92  * and destination addresses which lets us READ/WRITE CPU performance
93  * information using the appropriate I/O methods.
94  */
95 static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
96 
97 /* pcc mapped address + header size + offset within PCC subspace */
98 #define GET_PCC_VADDR(offs, pcc_ss_id) (pcc_data[pcc_ss_id]->pcc_comm_addr + \
99 						0x8 + (offs))
100 
101 /* Check if a CPC register is in PCC */
102 #define CPC_IN_PCC(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&		\
103 				(cpc)->cpc_entry.reg.space_id ==	\
104 				ACPI_ADR_SPACE_PLATFORM_COMM)
105 
106 /* Check if a CPC register is in SystemMemory */
107 #define CPC_IN_SYSTEM_MEMORY(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&	\
108 				(cpc)->cpc_entry.reg.space_id ==	\
109 				ACPI_ADR_SPACE_SYSTEM_MEMORY)
110 
111 /* Check if a CPC register is in SystemIo */
112 #define CPC_IN_SYSTEM_IO(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&	\
113 				(cpc)->cpc_entry.reg.space_id ==	\
114 				ACPI_ADR_SPACE_SYSTEM_IO)
115 
116 /* Evaluates to True if reg is a NULL register descriptor */
117 #define IS_NULL_REG(reg) ((reg)->space_id ==  ACPI_ADR_SPACE_SYSTEM_MEMORY && \
118 				(reg)->address == 0 &&			\
119 				(reg)->bit_width == 0 &&		\
120 				(reg)->bit_offset == 0 &&		\
121 				(reg)->access_width == 0)
122 
123 /* Evaluates to True if an optional cpc field is supported */
124 #define CPC_SUPPORTED(cpc) ((cpc)->type == ACPI_TYPE_INTEGER ?		\
125 				!!(cpc)->cpc_entry.int_value :		\
126 				!IS_NULL_REG(&(cpc)->cpc_entry.reg))
127 /*
128  * Arbitrary Retries in case the remote processor is slow to respond
129  * to PCC commands. Keeping it high enough to cover emulators where
130  * the processors run painfully slow.
131  */
132 #define NUM_RETRIES 500ULL
133 
134 #define OVER_16BTS_MASK ~0xFFFFULL
135 
136 #define define_one_cppc_ro(_name)		\
137 static struct kobj_attribute _name =		\
138 __ATTR(_name, 0444, show_##_name, NULL)
139 
140 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
141 
142 #define show_cppc_data(access_fn, struct_name, member_name)		\
143 	static ssize_t show_##member_name(struct kobject *kobj,		\
144 				struct kobj_attribute *attr, char *buf)	\
145 	{								\
146 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
147 		struct struct_name st_name = {0};			\
148 		int ret;						\
149 									\
150 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
151 		if (ret)						\
152 			return ret;					\
153 									\
154 		return sysfs_emit(buf, "%llu\n",		\
155 				(u64)st_name.member_name);		\
156 	}								\
157 	define_one_cppc_ro(member_name)
158 
159 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
160 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
161 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
162 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
163 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_freq);
164 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_freq);
165 
166 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, reference_perf);
167 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
168 
169 /* Check for valid access_width, otherwise, fallback to using bit_width */
170 #define GET_BIT_WIDTH(reg) ((reg)->access_width ? (8 << ((reg)->access_width - 1)) : (reg)->bit_width)
171 
172 /* Shift and apply the mask for CPC reads/writes */
173 #define MASK_VAL(reg, val) ((val) >> ((reg)->bit_offset & 			\
174 					GENMASK(((reg)->bit_width), 0)))
175 
176 static ssize_t show_feedback_ctrs(struct kobject *kobj,
177 		struct kobj_attribute *attr, char *buf)
178 {
179 	struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);
180 	struct cppc_perf_fb_ctrs fb_ctrs = {0};
181 	int ret;
182 
183 	ret = cppc_get_perf_ctrs(cpc_ptr->cpu_id, &fb_ctrs);
184 	if (ret)
185 		return ret;
186 
187 	return sysfs_emit(buf, "ref:%llu del:%llu\n",
188 			fb_ctrs.reference, fb_ctrs.delivered);
189 }
190 define_one_cppc_ro(feedback_ctrs);
191 
192 static struct attribute *cppc_attrs[] = {
193 	&feedback_ctrs.attr,
194 	&reference_perf.attr,
195 	&wraparound_time.attr,
196 	&highest_perf.attr,
197 	&lowest_perf.attr,
198 	&lowest_nonlinear_perf.attr,
199 	&nominal_perf.attr,
200 	&nominal_freq.attr,
201 	&lowest_freq.attr,
202 	NULL
203 };
204 ATTRIBUTE_GROUPS(cppc);
205 
206 static const struct kobj_type cppc_ktype = {
207 	.sysfs_ops = &kobj_sysfs_ops,
208 	.default_groups = cppc_groups,
209 };
210 
211 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
212 {
213 	int ret, status;
214 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
215 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
216 		pcc_ss_data->pcc_comm_addr;
217 
218 	if (!pcc_ss_data->platform_owns_pcc)
219 		return 0;
220 
221 	/*
222 	 * Poll PCC status register every 3us(delay_us) for maximum of
223 	 * deadline_us(timeout_us) until PCC command complete bit is set(cond)
224 	 */
225 	ret = readw_relaxed_poll_timeout(&generic_comm_base->status, status,
226 					status & PCC_CMD_COMPLETE_MASK, 3,
227 					pcc_ss_data->deadline_us);
228 
229 	if (likely(!ret)) {
230 		pcc_ss_data->platform_owns_pcc = false;
231 		if (chk_err_bit && (status & PCC_ERROR_MASK))
232 			ret = -EIO;
233 	}
234 
235 	if (unlikely(ret))
236 		pr_err("PCC check channel failed for ss: %d. ret=%d\n",
237 		       pcc_ss_id, ret);
238 
239 	return ret;
240 }
241 
242 /*
243  * This function transfers the ownership of the PCC to the platform
244  * So it must be called while holding write_lock(pcc_lock)
245  */
246 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
247 {
248 	int ret = -EIO, i;
249 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
250 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
251 		pcc_ss_data->pcc_comm_addr;
252 	unsigned int time_delta;
253 
254 	/*
255 	 * For CMD_WRITE we know for a fact the caller should have checked
256 	 * the channel before writing to PCC space
257 	 */
258 	if (cmd == CMD_READ) {
259 		/*
260 		 * If there are pending cpc_writes, then we stole the channel
261 		 * before write completion, so first send a WRITE command to
262 		 * platform
263 		 */
264 		if (pcc_ss_data->pending_pcc_write_cmd)
265 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
266 
267 		ret = check_pcc_chan(pcc_ss_id, false);
268 		if (ret)
269 			goto end;
270 	} else /* CMD_WRITE */
271 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
272 
273 	/*
274 	 * Handle the Minimum Request Turnaround Time(MRTT)
275 	 * "The minimum amount of time that OSPM must wait after the completion
276 	 * of a command before issuing the next command, in microseconds"
277 	 */
278 	if (pcc_ss_data->pcc_mrtt) {
279 		time_delta = ktime_us_delta(ktime_get(),
280 					    pcc_ss_data->last_cmd_cmpl_time);
281 		if (pcc_ss_data->pcc_mrtt > time_delta)
282 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
283 	}
284 
285 	/*
286 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
287 	 * "The maximum number of periodic requests that the subspace channel can
288 	 * support, reported in commands per minute. 0 indicates no limitation."
289 	 *
290 	 * This parameter should be ideally zero or large enough so that it can
291 	 * handle maximum number of requests that all the cores in the system can
292 	 * collectively generate. If it is not, we will follow the spec and just
293 	 * not send the request to the platform after hitting the MPAR limit in
294 	 * any 60s window
295 	 */
296 	if (pcc_ss_data->pcc_mpar) {
297 		if (pcc_ss_data->mpar_count == 0) {
298 			time_delta = ktime_ms_delta(ktime_get(),
299 						    pcc_ss_data->last_mpar_reset);
300 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
301 				pr_debug("PCC cmd for subspace %d not sent due to MPAR limit",
302 					 pcc_ss_id);
303 				ret = -EIO;
304 				goto end;
305 			}
306 			pcc_ss_data->last_mpar_reset = ktime_get();
307 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
308 		}
309 		pcc_ss_data->mpar_count--;
310 	}
311 
312 	/* Write to the shared comm region. */
313 	writew_relaxed(cmd, &generic_comm_base->command);
314 
315 	/* Flip CMD COMPLETE bit */
316 	writew_relaxed(0, &generic_comm_base->status);
317 
318 	pcc_ss_data->platform_owns_pcc = true;
319 
320 	/* Ring doorbell */
321 	ret = mbox_send_message(pcc_ss_data->pcc_channel->mchan, &cmd);
322 	if (ret < 0) {
323 		pr_err("Err sending PCC mbox message. ss: %d cmd:%d, ret:%d\n",
324 		       pcc_ss_id, cmd, ret);
325 		goto end;
326 	}
327 
328 	/* wait for completion and check for PCC error bit */
329 	ret = check_pcc_chan(pcc_ss_id, true);
330 
331 	if (pcc_ss_data->pcc_mrtt)
332 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
333 
334 	if (pcc_ss_data->pcc_channel->mchan->mbox->txdone_irq)
335 		mbox_chan_txdone(pcc_ss_data->pcc_channel->mchan, ret);
336 	else
337 		mbox_client_txdone(pcc_ss_data->pcc_channel->mchan, ret);
338 
339 end:
340 	if (cmd == CMD_WRITE) {
341 		if (unlikely(ret)) {
342 			for_each_possible_cpu(i) {
343 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
344 
345 				if (!desc)
346 					continue;
347 
348 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
349 					desc->write_cmd_status = ret;
350 			}
351 		}
352 		pcc_ss_data->pcc_write_cnt++;
353 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
354 	}
355 
356 	return ret;
357 }
358 
359 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
360 {
361 	if (ret < 0)
362 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
363 				*(u16 *)msg, ret);
364 	else
365 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
366 				*(u16 *)msg, ret);
367 }
368 
369 static struct mbox_client cppc_mbox_cl = {
370 	.tx_done = cppc_chan_tx_done,
371 	.knows_txdone = true,
372 };
373 
374 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
375 {
376 	int result = -EFAULT;
377 	acpi_status status = AE_OK;
378 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
379 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
380 	struct acpi_buffer state = {0, NULL};
381 	union acpi_object  *psd = NULL;
382 	struct acpi_psd_package *pdomain;
383 
384 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL,
385 					    &buffer, ACPI_TYPE_PACKAGE);
386 	if (status == AE_NOT_FOUND)	/* _PSD is optional */
387 		return 0;
388 	if (ACPI_FAILURE(status))
389 		return -ENODEV;
390 
391 	psd = buffer.pointer;
392 	if (!psd || psd->package.count != 1) {
393 		pr_debug("Invalid _PSD data\n");
394 		goto end;
395 	}
396 
397 	pdomain = &(cpc_ptr->domain_info);
398 
399 	state.length = sizeof(struct acpi_psd_package);
400 	state.pointer = pdomain;
401 
402 	status = acpi_extract_package(&(psd->package.elements[0]),
403 		&format, &state);
404 	if (ACPI_FAILURE(status)) {
405 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
406 		goto end;
407 	}
408 
409 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
410 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
411 		goto end;
412 	}
413 
414 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
415 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
416 		goto end;
417 	}
418 
419 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
420 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
421 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
422 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
423 		goto end;
424 	}
425 
426 	result = 0;
427 end:
428 	kfree(buffer.pointer);
429 	return result;
430 }
431 
432 bool acpi_cpc_valid(void)
433 {
434 	struct cpc_desc *cpc_ptr;
435 	int cpu;
436 
437 	if (acpi_disabled)
438 		return false;
439 
440 	for_each_present_cpu(cpu) {
441 		cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
442 		if (!cpc_ptr)
443 			return false;
444 	}
445 
446 	return true;
447 }
448 EXPORT_SYMBOL_GPL(acpi_cpc_valid);
449 
450 bool cppc_allow_fast_switch(void)
451 {
452 	struct cpc_register_resource *desired_reg;
453 	struct cpc_desc *cpc_ptr;
454 	int cpu;
455 
456 	for_each_possible_cpu(cpu) {
457 		cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
458 		desired_reg = &cpc_ptr->cpc_regs[DESIRED_PERF];
459 		if (!CPC_IN_SYSTEM_MEMORY(desired_reg) &&
460 				!CPC_IN_SYSTEM_IO(desired_reg))
461 			return false;
462 	}
463 
464 	return true;
465 }
466 EXPORT_SYMBOL_GPL(cppc_allow_fast_switch);
467 
468 /**
469  * acpi_get_psd_map - Map the CPUs in the freq domain of a given cpu
470  * @cpu: Find all CPUs that share a domain with cpu.
471  * @cpu_data: Pointer to CPU specific CPPC data including PSD info.
472  *
473  *	Return: 0 for success or negative value for err.
474  */
475 int acpi_get_psd_map(unsigned int cpu, struct cppc_cpudata *cpu_data)
476 {
477 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
478 	struct acpi_psd_package *match_pdomain;
479 	struct acpi_psd_package *pdomain;
480 	int count_target, i;
481 
482 	/*
483 	 * Now that we have _PSD data from all CPUs, let's setup P-state
484 	 * domain info.
485 	 */
486 	cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
487 	if (!cpc_ptr)
488 		return -EFAULT;
489 
490 	pdomain = &(cpc_ptr->domain_info);
491 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
492 	if (pdomain->num_processors <= 1)
493 		return 0;
494 
495 	/* Validate the Domain info */
496 	count_target = pdomain->num_processors;
497 	if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
498 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ALL;
499 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
500 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_HW;
501 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
502 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ANY;
503 
504 	for_each_possible_cpu(i) {
505 		if (i == cpu)
506 			continue;
507 
508 		match_cpc_ptr = per_cpu(cpc_desc_ptr, i);
509 		if (!match_cpc_ptr)
510 			goto err_fault;
511 
512 		match_pdomain = &(match_cpc_ptr->domain_info);
513 		if (match_pdomain->domain != pdomain->domain)
514 			continue;
515 
516 		/* Here i and cpu are in the same domain */
517 		if (match_pdomain->num_processors != count_target)
518 			goto err_fault;
519 
520 		if (pdomain->coord_type != match_pdomain->coord_type)
521 			goto err_fault;
522 
523 		cpumask_set_cpu(i, cpu_data->shared_cpu_map);
524 	}
525 
526 	return 0;
527 
528 err_fault:
529 	/* Assume no coordination on any error parsing domain info */
530 	cpumask_clear(cpu_data->shared_cpu_map);
531 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
532 	cpu_data->shared_type = CPUFREQ_SHARED_TYPE_NONE;
533 
534 	return -EFAULT;
535 }
536 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
537 
538 static int register_pcc_channel(int pcc_ss_idx)
539 {
540 	struct pcc_mbox_chan *pcc_chan;
541 	u64 usecs_lat;
542 
543 	if (pcc_ss_idx >= 0) {
544 		pcc_chan = pcc_mbox_request_channel(&cppc_mbox_cl, pcc_ss_idx);
545 
546 		if (IS_ERR(pcc_chan)) {
547 			pr_err("Failed to find PCC channel for subspace %d\n",
548 			       pcc_ss_idx);
549 			return -ENODEV;
550 		}
551 
552 		pcc_data[pcc_ss_idx]->pcc_channel = pcc_chan;
553 		/*
554 		 * cppc_ss->latency is just a Nominal value. In reality
555 		 * the remote processor could be much slower to reply.
556 		 * So add an arbitrary amount of wait on top of Nominal.
557 		 */
558 		usecs_lat = NUM_RETRIES * pcc_chan->latency;
559 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
560 		pcc_data[pcc_ss_idx]->pcc_mrtt = pcc_chan->min_turnaround_time;
561 		pcc_data[pcc_ss_idx]->pcc_mpar = pcc_chan->max_access_rate;
562 		pcc_data[pcc_ss_idx]->pcc_nominal = pcc_chan->latency;
563 
564 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
565 			acpi_os_ioremap(pcc_chan->shmem_base_addr,
566 					pcc_chan->shmem_size);
567 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
568 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
569 			       pcc_ss_idx);
570 			return -ENOMEM;
571 		}
572 
573 		/* Set flag so that we don't come here for each CPU. */
574 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
575 	}
576 
577 	return 0;
578 }
579 
580 /**
581  * cpc_ffh_supported() - check if FFH reading supported
582  *
583  * Check if the architecture has support for functional fixed hardware
584  * read/write capability.
585  *
586  * Return: true for supported, false for not supported
587  */
588 bool __weak cpc_ffh_supported(void)
589 {
590 	return false;
591 }
592 
593 /**
594  * cpc_supported_by_cpu() - check if CPPC is supported by CPU
595  *
596  * Check if the architectural support for CPPC is present even
597  * if the _OSC hasn't prescribed it
598  *
599  * Return: true for supported, false for not supported
600  */
601 bool __weak cpc_supported_by_cpu(void)
602 {
603 	return false;
604 }
605 
606 /**
607  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
608  * @pcc_ss_id: PCC Subspace index as in the PCC client ACPI package.
609  *
610  * Check and allocate the cppc_pcc_data memory.
611  * In some processor configurations it is possible that same subspace
612  * is shared between multiple CPUs. This is seen especially in CPUs
613  * with hardware multi-threading support.
614  *
615  * Return: 0 for success, errno for failure
616  */
617 static int pcc_data_alloc(int pcc_ss_id)
618 {
619 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
620 		return -EINVAL;
621 
622 	if (pcc_data[pcc_ss_id]) {
623 		pcc_data[pcc_ss_id]->refcount++;
624 	} else {
625 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
626 					      GFP_KERNEL);
627 		if (!pcc_data[pcc_ss_id])
628 			return -ENOMEM;
629 		pcc_data[pcc_ss_id]->refcount++;
630 	}
631 
632 	return 0;
633 }
634 
635 /*
636  * An example CPC table looks like the following.
637  *
638  *  Name (_CPC, Package() {
639  *      17,							// NumEntries
640  *      1,							// Revision
641  *      ResourceTemplate() {Register(PCC, 32, 0, 0x120, 2)},	// Highest Performance
642  *      ResourceTemplate() {Register(PCC, 32, 0, 0x124, 2)},	// Nominal Performance
643  *      ResourceTemplate() {Register(PCC, 32, 0, 0x128, 2)},	// Lowest Nonlinear Performance
644  *      ResourceTemplate() {Register(PCC, 32, 0, 0x12C, 2)},	// Lowest Performance
645  *      ResourceTemplate() {Register(PCC, 32, 0, 0x130, 2)},	// Guaranteed Performance Register
646  *      ResourceTemplate() {Register(PCC, 32, 0, 0x110, 2)},	// Desired Performance Register
647  *      ResourceTemplate() {Register(SystemMemory, 0, 0, 0, 0)},
648  *      ...
649  *      ...
650  *      ...
651  *  }
652  * Each Register() encodes how to access that specific register.
653  * e.g. a sample PCC entry has the following encoding:
654  *
655  *  Register (
656  *      PCC,	// AddressSpaceKeyword
657  *      8,	// RegisterBitWidth
658  *      8,	// RegisterBitOffset
659  *      0x30,	// RegisterAddress
660  *      9,	// AccessSize (subspace ID)
661  *  )
662  */
663 
664 #ifndef arch_init_invariance_cppc
665 static inline void arch_init_invariance_cppc(void) { }
666 #endif
667 
668 /**
669  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
670  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
671  *
672  *	Return: 0 for success or negative value for err.
673  */
674 int acpi_cppc_processor_probe(struct acpi_processor *pr)
675 {
676 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
677 	union acpi_object *out_obj, *cpc_obj;
678 	struct cpc_desc *cpc_ptr;
679 	struct cpc_reg *gas_t;
680 	struct device *cpu_dev;
681 	acpi_handle handle = pr->handle;
682 	unsigned int num_ent, i, cpc_rev;
683 	int pcc_subspace_id = -1;
684 	acpi_status status;
685 	int ret = -ENODATA;
686 
687 	if (!osc_sb_cppc2_support_acked) {
688 		pr_debug("CPPC v2 _OSC not acked\n");
689 		if (!cpc_supported_by_cpu())
690 			return -ENODEV;
691 	}
692 
693 	/* Parse the ACPI _CPC table for this CPU. */
694 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
695 			ACPI_TYPE_PACKAGE);
696 	if (ACPI_FAILURE(status)) {
697 		ret = -ENODEV;
698 		goto out_buf_free;
699 	}
700 
701 	out_obj = (union acpi_object *) output.pointer;
702 
703 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
704 	if (!cpc_ptr) {
705 		ret = -ENOMEM;
706 		goto out_buf_free;
707 	}
708 
709 	/* First entry is NumEntries. */
710 	cpc_obj = &out_obj->package.elements[0];
711 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
712 		num_ent = cpc_obj->integer.value;
713 		if (num_ent <= 1) {
714 			pr_debug("Unexpected _CPC NumEntries value (%d) for CPU:%d\n",
715 				 num_ent, pr->id);
716 			goto out_free;
717 		}
718 	} else {
719 		pr_debug("Unexpected _CPC NumEntries entry type (%d) for CPU:%d\n",
720 			 cpc_obj->type, pr->id);
721 		goto out_free;
722 	}
723 
724 	/* Second entry should be revision. */
725 	cpc_obj = &out_obj->package.elements[1];
726 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
727 		cpc_rev = cpc_obj->integer.value;
728 	} else {
729 		pr_debug("Unexpected _CPC Revision entry type (%d) for CPU:%d\n",
730 			 cpc_obj->type, pr->id);
731 		goto out_free;
732 	}
733 
734 	if (cpc_rev < CPPC_V2_REV) {
735 		pr_debug("Unsupported _CPC Revision (%d) for CPU:%d\n", cpc_rev,
736 			 pr->id);
737 		goto out_free;
738 	}
739 
740 	/*
741 	 * Disregard _CPC if the number of entries in the return pachage is not
742 	 * as expected, but support future revisions being proper supersets of
743 	 * the v3 and only causing more entries to be returned by _CPC.
744 	 */
745 	if ((cpc_rev == CPPC_V2_REV && num_ent != CPPC_V2_NUM_ENT) ||
746 	    (cpc_rev == CPPC_V3_REV && num_ent != CPPC_V3_NUM_ENT) ||
747 	    (cpc_rev > CPPC_V3_REV && num_ent <= CPPC_V3_NUM_ENT)) {
748 		pr_debug("Unexpected number of _CPC return package entries (%d) for CPU:%d\n",
749 			 num_ent, pr->id);
750 		goto out_free;
751 	}
752 	if (cpc_rev > CPPC_V3_REV) {
753 		num_ent = CPPC_V3_NUM_ENT;
754 		cpc_rev = CPPC_V3_REV;
755 	}
756 
757 	cpc_ptr->num_entries = num_ent;
758 	cpc_ptr->version = cpc_rev;
759 
760 	/* Iterate through remaining entries in _CPC */
761 	for (i = 2; i < num_ent; i++) {
762 		cpc_obj = &out_obj->package.elements[i];
763 
764 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
765 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
766 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
767 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
768 			gas_t = (struct cpc_reg *)
769 				cpc_obj->buffer.pointer;
770 
771 			/*
772 			 * The PCC Subspace index is encoded inside
773 			 * the CPC table entries. The same PCC index
774 			 * will be used for all the PCC entries,
775 			 * so extract it only once.
776 			 */
777 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
778 				if (pcc_subspace_id < 0) {
779 					pcc_subspace_id = gas_t->access_width;
780 					if (pcc_data_alloc(pcc_subspace_id))
781 						goto out_free;
782 				} else if (pcc_subspace_id != gas_t->access_width) {
783 					pr_debug("Mismatched PCC ids in _CPC for CPU:%d\n",
784 						 pr->id);
785 					goto out_free;
786 				}
787 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
788 				if (gas_t->address) {
789 					void __iomem *addr;
790 					size_t access_width;
791 
792 					if (!osc_cpc_flexible_adr_space_confirmed) {
793 						pr_debug("Flexible address space capability not supported\n");
794 						if (!cpc_supported_by_cpu())
795 							goto out_free;
796 					}
797 
798 					access_width = GET_BIT_WIDTH(gas_t) / 8;
799 					addr = ioremap(gas_t->address, access_width);
800 					if (!addr)
801 						goto out_free;
802 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
803 				}
804 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
805 				if (gas_t->access_width < 1 || gas_t->access_width > 3) {
806 					/*
807 					 * 1 = 8-bit, 2 = 16-bit, and 3 = 32-bit.
808 					 * SystemIO doesn't implement 64-bit
809 					 * registers.
810 					 */
811 					pr_debug("Invalid access width %d for SystemIO register in _CPC\n",
812 						 gas_t->access_width);
813 					goto out_free;
814 				}
815 				if (gas_t->address & OVER_16BTS_MASK) {
816 					/* SystemIO registers use 16-bit integer addresses */
817 					pr_debug("Invalid IO port %llu for SystemIO register in _CPC\n",
818 						 gas_t->address);
819 					goto out_free;
820 				}
821 				if (!osc_cpc_flexible_adr_space_confirmed) {
822 					pr_debug("Flexible address space capability not supported\n");
823 					if (!cpc_supported_by_cpu())
824 						goto out_free;
825 				}
826 			} else {
827 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
828 					/* Support only PCC, SystemMemory, SystemIO, and FFH type regs. */
829 					pr_debug("Unsupported register type (%d) in _CPC\n",
830 						 gas_t->space_id);
831 					goto out_free;
832 				}
833 			}
834 
835 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
836 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
837 		} else {
838 			pr_debug("Invalid entry type (%d) in _CPC for CPU:%d\n",
839 				 i, pr->id);
840 			goto out_free;
841 		}
842 	}
843 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
844 
845 	/*
846 	 * Initialize the remaining cpc_regs as unsupported.
847 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
848 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
849 	 */
850 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
851 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
852 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
853 	}
854 
855 
856 	/* Store CPU Logical ID */
857 	cpc_ptr->cpu_id = pr->id;
858 
859 	/* Parse PSD data for this CPU */
860 	ret = acpi_get_psd(cpc_ptr, handle);
861 	if (ret)
862 		goto out_free;
863 
864 	/* Register PCC channel once for all PCC subspace ID. */
865 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
866 		ret = register_pcc_channel(pcc_subspace_id);
867 		if (ret)
868 			goto out_free;
869 
870 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
871 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
872 	}
873 
874 	/* Everything looks okay */
875 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
876 
877 	/* Add per logical CPU nodes for reading its feedback counters. */
878 	cpu_dev = get_cpu_device(pr->id);
879 	if (!cpu_dev) {
880 		ret = -EINVAL;
881 		goto out_free;
882 	}
883 
884 	/* Plug PSD data into this CPU's CPC descriptor. */
885 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
886 
887 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
888 			"acpi_cppc");
889 	if (ret) {
890 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
891 		kobject_put(&cpc_ptr->kobj);
892 		goto out_free;
893 	}
894 
895 	arch_init_invariance_cppc();
896 
897 	kfree(output.pointer);
898 	return 0;
899 
900 out_free:
901 	/* Free all the mapped sys mem areas for this CPU */
902 	for (i = 2; i < cpc_ptr->num_entries; i++) {
903 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
904 
905 		if (addr)
906 			iounmap(addr);
907 	}
908 	kfree(cpc_ptr);
909 
910 out_buf_free:
911 	kfree(output.pointer);
912 	return ret;
913 }
914 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
915 
916 /**
917  * acpi_cppc_processor_exit - Cleanup CPC structs.
918  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
919  *
920  * Return: Void
921  */
922 void acpi_cppc_processor_exit(struct acpi_processor *pr)
923 {
924 	struct cpc_desc *cpc_ptr;
925 	unsigned int i;
926 	void __iomem *addr;
927 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
928 
929 	if (pcc_ss_id >= 0 && pcc_data[pcc_ss_id]) {
930 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
931 			pcc_data[pcc_ss_id]->refcount--;
932 			if (!pcc_data[pcc_ss_id]->refcount) {
933 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
934 				kfree(pcc_data[pcc_ss_id]);
935 				pcc_data[pcc_ss_id] = NULL;
936 			}
937 		}
938 	}
939 
940 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
941 	if (!cpc_ptr)
942 		return;
943 
944 	/* Free all the mapped sys mem areas for this CPU */
945 	for (i = 2; i < cpc_ptr->num_entries; i++) {
946 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
947 		if (addr)
948 			iounmap(addr);
949 	}
950 
951 	kobject_put(&cpc_ptr->kobj);
952 	kfree(cpc_ptr);
953 }
954 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
955 
956 /**
957  * cpc_read_ffh() - Read FFH register
958  * @cpunum:	CPU number to read
959  * @reg:	cppc register information
960  * @val:	place holder for return value
961  *
962  * Read bit_width bits from a specified address and bit_offset
963  *
964  * Return: 0 for success and error code
965  */
966 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
967 {
968 	return -ENOTSUPP;
969 }
970 
971 /**
972  * cpc_write_ffh() - Write FFH register
973  * @cpunum:	CPU number to write
974  * @reg:	cppc register information
975  * @val:	value to write
976  *
977  * Write value of bit_width bits to a specified address and bit_offset
978  *
979  * Return: 0 for success and error code
980  */
981 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
982 {
983 	return -ENOTSUPP;
984 }
985 
986 /*
987  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
988  * as fast as possible. We have already mapped the PCC subspace during init, so
989  * we can directly write to it.
990  */
991 
992 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
993 {
994 	void __iomem *vaddr = NULL;
995 	int size;
996 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
997 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
998 
999 	if (reg_res->type == ACPI_TYPE_INTEGER) {
1000 		*val = reg_res->cpc_entry.int_value;
1001 		return 0;
1002 	}
1003 
1004 	*val = 0;
1005 
1006 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
1007 		u32 width = GET_BIT_WIDTH(reg);
1008 		u32 val_u32;
1009 		acpi_status status;
1010 
1011 		status = acpi_os_read_port((acpi_io_address)reg->address,
1012 					   &val_u32, width);
1013 		if (ACPI_FAILURE(status)) {
1014 			pr_debug("Error: Failed to read SystemIO port %llx\n",
1015 				 reg->address);
1016 			return -EFAULT;
1017 		}
1018 
1019 		*val = val_u32;
1020 		return 0;
1021 	} else if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
1022 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
1023 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1024 		vaddr = reg_res->sys_mem_vaddr;
1025 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1026 		return cpc_read_ffh(cpu, reg, val);
1027 	else
1028 		return acpi_os_read_memory((acpi_physical_address)reg->address,
1029 				val, reg->bit_width);
1030 
1031 	size = GET_BIT_WIDTH(reg);
1032 
1033 	switch (size) {
1034 	case 8:
1035 		*val = readb_relaxed(vaddr);
1036 		break;
1037 	case 16:
1038 		*val = readw_relaxed(vaddr);
1039 		break;
1040 	case 32:
1041 		*val = readl_relaxed(vaddr);
1042 		break;
1043 	case 64:
1044 		*val = readq_relaxed(vaddr);
1045 		break;
1046 	default:
1047 		pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
1048 			 reg->bit_width, pcc_ss_id);
1049 		return -EFAULT;
1050 	}
1051 
1052 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1053 		*val = MASK_VAL(reg, *val);
1054 
1055 	return 0;
1056 }
1057 
1058 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
1059 {
1060 	int ret_val = 0;
1061 	int size;
1062 	void __iomem *vaddr = NULL;
1063 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1064 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
1065 
1066 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
1067 		u32 width = GET_BIT_WIDTH(reg);
1068 		acpi_status status;
1069 
1070 		status = acpi_os_write_port((acpi_io_address)reg->address,
1071 					    (u32)val, width);
1072 		if (ACPI_FAILURE(status)) {
1073 			pr_debug("Error: Failed to write SystemIO port %llx\n",
1074 				 reg->address);
1075 			return -EFAULT;
1076 		}
1077 
1078 		return 0;
1079 	} else if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
1080 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
1081 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1082 		vaddr = reg_res->sys_mem_vaddr;
1083 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1084 		return cpc_write_ffh(cpu, reg, val);
1085 	else
1086 		return acpi_os_write_memory((acpi_physical_address)reg->address,
1087 				val, reg->bit_width);
1088 
1089 	size = GET_BIT_WIDTH(reg);
1090 
1091 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1092 		val = MASK_VAL(reg, val);
1093 
1094 	switch (size) {
1095 	case 8:
1096 		writeb_relaxed(val, vaddr);
1097 		break;
1098 	case 16:
1099 		writew_relaxed(val, vaddr);
1100 		break;
1101 	case 32:
1102 		writel_relaxed(val, vaddr);
1103 		break;
1104 	case 64:
1105 		writeq_relaxed(val, vaddr);
1106 		break;
1107 	default:
1108 		pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1109 			 reg->bit_width, pcc_ss_id);
1110 		ret_val = -EFAULT;
1111 		break;
1112 	}
1113 
1114 	return ret_val;
1115 }
1116 
1117 static int cppc_get_perf(int cpunum, enum cppc_regs reg_idx, u64 *perf)
1118 {
1119 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1120 	struct cpc_register_resource *reg;
1121 
1122 	if (!cpc_desc) {
1123 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1124 		return -ENODEV;
1125 	}
1126 
1127 	reg = &cpc_desc->cpc_regs[reg_idx];
1128 
1129 	if (CPC_IN_PCC(reg)) {
1130 		int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1131 		struct cppc_pcc_data *pcc_ss_data = NULL;
1132 		int ret = 0;
1133 
1134 		if (pcc_ss_id < 0)
1135 			return -EIO;
1136 
1137 		pcc_ss_data = pcc_data[pcc_ss_id];
1138 
1139 		down_write(&pcc_ss_data->pcc_lock);
1140 
1141 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1142 			cpc_read(cpunum, reg, perf);
1143 		else
1144 			ret = -EIO;
1145 
1146 		up_write(&pcc_ss_data->pcc_lock);
1147 
1148 		return ret;
1149 	}
1150 
1151 	cpc_read(cpunum, reg, perf);
1152 
1153 	return 0;
1154 }
1155 
1156 /**
1157  * cppc_get_desired_perf - Get the desired performance register value.
1158  * @cpunum: CPU from which to get desired performance.
1159  * @desired_perf: Return address.
1160  *
1161  * Return: 0 for success, -EIO otherwise.
1162  */
1163 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1164 {
1165 	return cppc_get_perf(cpunum, DESIRED_PERF, desired_perf);
1166 }
1167 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1168 
1169 /**
1170  * cppc_get_nominal_perf - Get the nominal performance register value.
1171  * @cpunum: CPU from which to get nominal performance.
1172  * @nominal_perf: Return address.
1173  *
1174  * Return: 0 for success, -EIO otherwise.
1175  */
1176 int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
1177 {
1178 	return cppc_get_perf(cpunum, NOMINAL_PERF, nominal_perf);
1179 }
1180 
1181 /**
1182  * cppc_get_highest_perf - Get the highest performance register value.
1183  * @cpunum: CPU from which to get highest performance.
1184  * @highest_perf: Return address.
1185  *
1186  * Return: 0 for success, -EIO otherwise.
1187  */
1188 int cppc_get_highest_perf(int cpunum, u64 *highest_perf)
1189 {
1190 	return cppc_get_perf(cpunum, HIGHEST_PERF, highest_perf);
1191 }
1192 EXPORT_SYMBOL_GPL(cppc_get_highest_perf);
1193 
1194 /**
1195  * cppc_get_epp_perf - Get the epp register value.
1196  * @cpunum: CPU from which to get epp preference value.
1197  * @epp_perf: Return address.
1198  *
1199  * Return: 0 for success, -EIO otherwise.
1200  */
1201 int cppc_get_epp_perf(int cpunum, u64 *epp_perf)
1202 {
1203 	return cppc_get_perf(cpunum, ENERGY_PERF, epp_perf);
1204 }
1205 EXPORT_SYMBOL_GPL(cppc_get_epp_perf);
1206 
1207 /**
1208  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1209  * @cpunum: CPU from which to get capabilities info.
1210  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1211  *
1212  * Return: 0 for success with perf_caps populated else -ERRNO.
1213  */
1214 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1215 {
1216 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1217 	struct cpc_register_resource *highest_reg, *lowest_reg,
1218 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1219 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1220 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1221 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1222 	struct cppc_pcc_data *pcc_ss_data = NULL;
1223 	int ret = 0, regs_in_pcc = 0;
1224 
1225 	if (!cpc_desc) {
1226 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1227 		return -ENODEV;
1228 	}
1229 
1230 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1231 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1232 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1233 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1234 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1235 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1236 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1237 
1238 	/* Are any of the regs PCC ?*/
1239 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1240 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1241 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1242 		if (pcc_ss_id < 0) {
1243 			pr_debug("Invalid pcc_ss_id\n");
1244 			return -ENODEV;
1245 		}
1246 		pcc_ss_data = pcc_data[pcc_ss_id];
1247 		regs_in_pcc = 1;
1248 		down_write(&pcc_ss_data->pcc_lock);
1249 		/* Ring doorbell once to update PCC subspace */
1250 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1251 			ret = -EIO;
1252 			goto out_err;
1253 		}
1254 	}
1255 
1256 	cpc_read(cpunum, highest_reg, &high);
1257 	perf_caps->highest_perf = high;
1258 
1259 	cpc_read(cpunum, lowest_reg, &low);
1260 	perf_caps->lowest_perf = low;
1261 
1262 	cpc_read(cpunum, nominal_reg, &nom);
1263 	perf_caps->nominal_perf = nom;
1264 
1265 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1266 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1267 		perf_caps->guaranteed_perf = 0;
1268 	} else {
1269 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1270 		perf_caps->guaranteed_perf = guaranteed;
1271 	}
1272 
1273 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1274 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1275 
1276 	if (!high || !low || !nom || !min_nonlinear)
1277 		ret = -EFAULT;
1278 
1279 	/* Read optional lowest and nominal frequencies if present */
1280 	if (CPC_SUPPORTED(low_freq_reg))
1281 		cpc_read(cpunum, low_freq_reg, &low_f);
1282 
1283 	if (CPC_SUPPORTED(nom_freq_reg))
1284 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1285 
1286 	perf_caps->lowest_freq = low_f;
1287 	perf_caps->nominal_freq = nom_f;
1288 
1289 
1290 out_err:
1291 	if (regs_in_pcc)
1292 		up_write(&pcc_ss_data->pcc_lock);
1293 	return ret;
1294 }
1295 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1296 
1297 /**
1298  * cppc_perf_ctrs_in_pcc - Check if any perf counters are in a PCC region.
1299  *
1300  * CPPC has flexibility about how CPU performance counters are accessed.
1301  * One of the choices is PCC regions, which can have a high access latency. This
1302  * routine allows callers of cppc_get_perf_ctrs() to know this ahead of time.
1303  *
1304  * Return: true if any of the counters are in PCC regions, false otherwise
1305  */
1306 bool cppc_perf_ctrs_in_pcc(void)
1307 {
1308 	int cpu;
1309 
1310 	for_each_present_cpu(cpu) {
1311 		struct cpc_register_resource *ref_perf_reg;
1312 		struct cpc_desc *cpc_desc;
1313 
1314 		cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1315 
1316 		if (CPC_IN_PCC(&cpc_desc->cpc_regs[DELIVERED_CTR]) ||
1317 		    CPC_IN_PCC(&cpc_desc->cpc_regs[REFERENCE_CTR]) ||
1318 		    CPC_IN_PCC(&cpc_desc->cpc_regs[CTR_WRAP_TIME]))
1319 			return true;
1320 
1321 
1322 		ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1323 
1324 		/*
1325 		 * If reference perf register is not supported then we should
1326 		 * use the nominal perf value
1327 		 */
1328 		if (!CPC_SUPPORTED(ref_perf_reg))
1329 			ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1330 
1331 		if (CPC_IN_PCC(ref_perf_reg))
1332 			return true;
1333 	}
1334 
1335 	return false;
1336 }
1337 EXPORT_SYMBOL_GPL(cppc_perf_ctrs_in_pcc);
1338 
1339 /**
1340  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1341  * @cpunum: CPU from which to read counters.
1342  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1343  *
1344  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1345  */
1346 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1347 {
1348 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1349 	struct cpc_register_resource *delivered_reg, *reference_reg,
1350 		*ref_perf_reg, *ctr_wrap_reg;
1351 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1352 	struct cppc_pcc_data *pcc_ss_data = NULL;
1353 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1354 	int ret = 0, regs_in_pcc = 0;
1355 
1356 	if (!cpc_desc) {
1357 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1358 		return -ENODEV;
1359 	}
1360 
1361 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1362 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1363 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1364 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1365 
1366 	/*
1367 	 * If reference perf register is not supported then we should
1368 	 * use the nominal perf value
1369 	 */
1370 	if (!CPC_SUPPORTED(ref_perf_reg))
1371 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1372 
1373 	/* Are any of the regs PCC ?*/
1374 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1375 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1376 		if (pcc_ss_id < 0) {
1377 			pr_debug("Invalid pcc_ss_id\n");
1378 			return -ENODEV;
1379 		}
1380 		pcc_ss_data = pcc_data[pcc_ss_id];
1381 		down_write(&pcc_ss_data->pcc_lock);
1382 		regs_in_pcc = 1;
1383 		/* Ring doorbell once to update PCC subspace */
1384 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1385 			ret = -EIO;
1386 			goto out_err;
1387 		}
1388 	}
1389 
1390 	cpc_read(cpunum, delivered_reg, &delivered);
1391 	cpc_read(cpunum, reference_reg, &reference);
1392 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1393 
1394 	/*
1395 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1396 	 * performance counters are assumed to never wrap during the lifetime of
1397 	 * platform
1398 	 */
1399 	ctr_wrap_time = (u64)(~((u64)0));
1400 	if (CPC_SUPPORTED(ctr_wrap_reg))
1401 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1402 
1403 	if (!delivered || !reference ||	!ref_perf) {
1404 		ret = -EFAULT;
1405 		goto out_err;
1406 	}
1407 
1408 	perf_fb_ctrs->delivered = delivered;
1409 	perf_fb_ctrs->reference = reference;
1410 	perf_fb_ctrs->reference_perf = ref_perf;
1411 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1412 out_err:
1413 	if (regs_in_pcc)
1414 		up_write(&pcc_ss_data->pcc_lock);
1415 	return ret;
1416 }
1417 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1418 
1419 /*
1420  * Set Energy Performance Preference Register value through
1421  * Performance Controls Interface
1422  */
1423 int cppc_set_epp_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls, bool enable)
1424 {
1425 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1426 	struct cpc_register_resource *epp_set_reg;
1427 	struct cpc_register_resource *auto_sel_reg;
1428 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1429 	struct cppc_pcc_data *pcc_ss_data = NULL;
1430 	int ret;
1431 
1432 	if (!cpc_desc) {
1433 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1434 		return -ENODEV;
1435 	}
1436 
1437 	auto_sel_reg = &cpc_desc->cpc_regs[AUTO_SEL_ENABLE];
1438 	epp_set_reg = &cpc_desc->cpc_regs[ENERGY_PERF];
1439 
1440 	if (CPC_IN_PCC(epp_set_reg) || CPC_IN_PCC(auto_sel_reg)) {
1441 		if (pcc_ss_id < 0) {
1442 			pr_debug("Invalid pcc_ss_id for CPU:%d\n", cpu);
1443 			return -ENODEV;
1444 		}
1445 
1446 		if (CPC_SUPPORTED(auto_sel_reg)) {
1447 			ret = cpc_write(cpu, auto_sel_reg, enable);
1448 			if (ret)
1449 				return ret;
1450 		}
1451 
1452 		if (CPC_SUPPORTED(epp_set_reg)) {
1453 			ret = cpc_write(cpu, epp_set_reg, perf_ctrls->energy_perf);
1454 			if (ret)
1455 				return ret;
1456 		}
1457 
1458 		pcc_ss_data = pcc_data[pcc_ss_id];
1459 
1460 		down_write(&pcc_ss_data->pcc_lock);
1461 		/* after writing CPC, transfer the ownership of PCC to platform */
1462 		ret = send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1463 		up_write(&pcc_ss_data->pcc_lock);
1464 	} else {
1465 		ret = -ENOTSUPP;
1466 		pr_debug("_CPC in PCC is not supported\n");
1467 	}
1468 
1469 	return ret;
1470 }
1471 EXPORT_SYMBOL_GPL(cppc_set_epp_perf);
1472 
1473 /**
1474  * cppc_get_auto_sel_caps - Read autonomous selection register.
1475  * @cpunum : CPU from which to read register.
1476  * @perf_caps : struct where autonomous selection register value is updated.
1477  */
1478 int cppc_get_auto_sel_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1479 {
1480 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1481 	struct cpc_register_resource *auto_sel_reg;
1482 	u64  auto_sel;
1483 
1484 	if (!cpc_desc) {
1485 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1486 		return -ENODEV;
1487 	}
1488 
1489 	auto_sel_reg = &cpc_desc->cpc_regs[AUTO_SEL_ENABLE];
1490 
1491 	if (!CPC_SUPPORTED(auto_sel_reg))
1492 		pr_warn_once("Autonomous mode is not unsupported!\n");
1493 
1494 	if (CPC_IN_PCC(auto_sel_reg)) {
1495 		int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1496 		struct cppc_pcc_data *pcc_ss_data = NULL;
1497 		int ret = 0;
1498 
1499 		if (pcc_ss_id < 0)
1500 			return -ENODEV;
1501 
1502 		pcc_ss_data = pcc_data[pcc_ss_id];
1503 
1504 		down_write(&pcc_ss_data->pcc_lock);
1505 
1506 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0) {
1507 			cpc_read(cpunum, auto_sel_reg, &auto_sel);
1508 			perf_caps->auto_sel = (bool)auto_sel;
1509 		} else {
1510 			ret = -EIO;
1511 		}
1512 
1513 		up_write(&pcc_ss_data->pcc_lock);
1514 
1515 		return ret;
1516 	}
1517 
1518 	return 0;
1519 }
1520 EXPORT_SYMBOL_GPL(cppc_get_auto_sel_caps);
1521 
1522 /**
1523  * cppc_set_auto_sel - Write autonomous selection register.
1524  * @cpu    : CPU to which to write register.
1525  * @enable : the desired value of autonomous selection resiter to be updated.
1526  */
1527 int cppc_set_auto_sel(int cpu, bool enable)
1528 {
1529 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1530 	struct cpc_register_resource *auto_sel_reg;
1531 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1532 	struct cppc_pcc_data *pcc_ss_data = NULL;
1533 	int ret = -EINVAL;
1534 
1535 	if (!cpc_desc) {
1536 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1537 		return -ENODEV;
1538 	}
1539 
1540 	auto_sel_reg = &cpc_desc->cpc_regs[AUTO_SEL_ENABLE];
1541 
1542 	if (CPC_IN_PCC(auto_sel_reg)) {
1543 		if (pcc_ss_id < 0) {
1544 			pr_debug("Invalid pcc_ss_id\n");
1545 			return -ENODEV;
1546 		}
1547 
1548 		if (CPC_SUPPORTED(auto_sel_reg)) {
1549 			ret = cpc_write(cpu, auto_sel_reg, enable);
1550 			if (ret)
1551 				return ret;
1552 		}
1553 
1554 		pcc_ss_data = pcc_data[pcc_ss_id];
1555 
1556 		down_write(&pcc_ss_data->pcc_lock);
1557 		/* after writing CPC, transfer the ownership of PCC to platform */
1558 		ret = send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1559 		up_write(&pcc_ss_data->pcc_lock);
1560 	} else {
1561 		ret = -ENOTSUPP;
1562 		pr_debug("_CPC in PCC is not supported\n");
1563 	}
1564 
1565 	return ret;
1566 }
1567 EXPORT_SYMBOL_GPL(cppc_set_auto_sel);
1568 
1569 /**
1570  * cppc_set_enable - Set to enable CPPC on the processor by writing the
1571  * Continuous Performance Control package EnableRegister field.
1572  * @cpu: CPU for which to enable CPPC register.
1573  * @enable: 0 - disable, 1 - enable CPPC feature on the processor.
1574  *
1575  * Return: 0 for success, -ERRNO or -EIO otherwise.
1576  */
1577 int cppc_set_enable(int cpu, bool enable)
1578 {
1579 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1580 	struct cpc_register_resource *enable_reg;
1581 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1582 	struct cppc_pcc_data *pcc_ss_data = NULL;
1583 	int ret = -EINVAL;
1584 
1585 	if (!cpc_desc) {
1586 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1587 		return -EINVAL;
1588 	}
1589 
1590 	enable_reg = &cpc_desc->cpc_regs[ENABLE];
1591 
1592 	if (CPC_IN_PCC(enable_reg)) {
1593 
1594 		if (pcc_ss_id < 0)
1595 			return -EIO;
1596 
1597 		ret = cpc_write(cpu, enable_reg, enable);
1598 		if (ret)
1599 			return ret;
1600 
1601 		pcc_ss_data = pcc_data[pcc_ss_id];
1602 
1603 		down_write(&pcc_ss_data->pcc_lock);
1604 		/* after writing CPC, transfer the ownership of PCC to platfrom */
1605 		ret = send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1606 		up_write(&pcc_ss_data->pcc_lock);
1607 		return ret;
1608 	}
1609 
1610 	return cpc_write(cpu, enable_reg, enable);
1611 }
1612 EXPORT_SYMBOL_GPL(cppc_set_enable);
1613 
1614 /**
1615  * cppc_set_perf - Set a CPU's performance controls.
1616  * @cpu: CPU for which to set performance controls.
1617  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1618  *
1619  * Return: 0 for success, -ERRNO otherwise.
1620  */
1621 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1622 {
1623 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1624 	struct cpc_register_resource *desired_reg, *min_perf_reg, *max_perf_reg;
1625 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1626 	struct cppc_pcc_data *pcc_ss_data = NULL;
1627 	int ret = 0;
1628 
1629 	if (!cpc_desc) {
1630 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1631 		return -ENODEV;
1632 	}
1633 
1634 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1635 	min_perf_reg = &cpc_desc->cpc_regs[MIN_PERF];
1636 	max_perf_reg = &cpc_desc->cpc_regs[MAX_PERF];
1637 
1638 	/*
1639 	 * This is Phase-I where we want to write to CPC registers
1640 	 * -> We want all CPUs to be able to execute this phase in parallel
1641 	 *
1642 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1643 	 * achieve that goal here
1644 	 */
1645 	if (CPC_IN_PCC(desired_reg) || CPC_IN_PCC(min_perf_reg) || CPC_IN_PCC(max_perf_reg)) {
1646 		if (pcc_ss_id < 0) {
1647 			pr_debug("Invalid pcc_ss_id\n");
1648 			return -ENODEV;
1649 		}
1650 		pcc_ss_data = pcc_data[pcc_ss_id];
1651 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1652 		if (pcc_ss_data->platform_owns_pcc) {
1653 			ret = check_pcc_chan(pcc_ss_id, false);
1654 			if (ret) {
1655 				up_read(&pcc_ss_data->pcc_lock);
1656 				return ret;
1657 			}
1658 		}
1659 		/*
1660 		 * Update the pending_write to make sure a PCC CMD_READ will not
1661 		 * arrive and steal the channel during the switch to write lock
1662 		 */
1663 		pcc_ss_data->pending_pcc_write_cmd = true;
1664 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1665 		cpc_desc->write_cmd_status = 0;
1666 	}
1667 
1668 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1669 
1670 	/*
1671 	 * Only write if min_perf and max_perf not zero. Some drivers pass zero
1672 	 * value to min and max perf, but they don't mean to set the zero value,
1673 	 * they just don't want to write to those registers.
1674 	 */
1675 	if (perf_ctrls->min_perf)
1676 		cpc_write(cpu, min_perf_reg, perf_ctrls->min_perf);
1677 	if (perf_ctrls->max_perf)
1678 		cpc_write(cpu, max_perf_reg, perf_ctrls->max_perf);
1679 
1680 	if (CPC_IN_PCC(desired_reg) || CPC_IN_PCC(min_perf_reg) || CPC_IN_PCC(max_perf_reg))
1681 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1682 	/*
1683 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1684 	 *
1685 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1686 	 * requests that happened in short overlapping interval. The last CPU to
1687 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1688 	 *
1689 	 * We have the following requirements for Phase-II:
1690 	 *     1. We want to execute Phase-II only when there are no CPUs
1691 	 * currently executing in Phase-I
1692 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1693 	 * entering Phase-I.
1694 	 *     3. We want only one CPU among all those who went through Phase-I
1695 	 * to run phase-II
1696 	 *
1697 	 * If write_trylock fails to get the lock and doesn't transfer the
1698 	 * PCC ownership to the platform, then one of the following will be TRUE
1699 	 *     1. There is at-least one CPU in Phase-I which will later execute
1700 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1701 	 * executing the Phase-II.
1702 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1703 	 * write_trylock and has already acquired the write_lock. We know for a
1704 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1705 	 * before this CPU's Phase-I as we held the read_lock.
1706 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1707 	 * down_write, in which case, send_pcc_cmd will check for pending
1708 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1709 	 * So this CPU can be certain that its request will be delivered
1710 	 *    So in all cases, this CPU knows that its request will be delivered
1711 	 * by another CPU and can return
1712 	 *
1713 	 * After getting the down_write we still need to check for
1714 	 * pending_pcc_write_cmd to take care of the following scenario
1715 	 *    The thread running this code could be scheduled out between
1716 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1717 	 * could have delivered the request to Platform by triggering the
1718 	 * doorbell and transferred the ownership of PCC to platform. So this
1719 	 * avoids triggering an unnecessary doorbell and more importantly before
1720 	 * triggering the doorbell it makes sure that the PCC channel ownership
1721 	 * is still with OSPM.
1722 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1723 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1724 	 * before the pcc CMD_WRITE is completed. send_pcc_cmd checks for this
1725 	 * case during a CMD_READ and if there are pending writes it delivers
1726 	 * the write command before servicing the read command
1727 	 */
1728 	if (CPC_IN_PCC(desired_reg) || CPC_IN_PCC(min_perf_reg) || CPC_IN_PCC(max_perf_reg)) {
1729 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1730 			/* Update only if there are pending write commands */
1731 			if (pcc_ss_data->pending_pcc_write_cmd)
1732 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1733 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1734 		} else
1735 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1736 			wait_event(pcc_ss_data->pcc_write_wait_q,
1737 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1738 
1739 		/* send_pcc_cmd updates the status in case of failure */
1740 		ret = cpc_desc->write_cmd_status;
1741 	}
1742 	return ret;
1743 }
1744 EXPORT_SYMBOL_GPL(cppc_set_perf);
1745 
1746 /**
1747  * cppc_get_transition_latency - returns frequency transition latency in ns
1748  * @cpu_num: CPU number for per_cpu().
1749  *
1750  * ACPI CPPC does not explicitly specify how a platform can specify the
1751  * transition latency for performance change requests. The closest we have
1752  * is the timing information from the PCCT tables which provides the info
1753  * on the number and frequency of PCC commands the platform can handle.
1754  *
1755  * If desired_reg is in the SystemMemory or SystemIo ACPI address space,
1756  * then assume there is no latency.
1757  */
1758 unsigned int cppc_get_transition_latency(int cpu_num)
1759 {
1760 	/*
1761 	 * Expected transition latency is based on the PCCT timing values
1762 	 * Below are definition from ACPI spec:
1763 	 * pcc_nominal- Expected latency to process a command, in microseconds
1764 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1765 	 *              channel can support, reported in commands per minute. 0
1766 	 *              indicates no limitation.
1767 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1768 	 *              completion of a command before issuing the next command,
1769 	 *              in microseconds.
1770 	 */
1771 	unsigned int latency_ns = 0;
1772 	struct cpc_desc *cpc_desc;
1773 	struct cpc_register_resource *desired_reg;
1774 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1775 	struct cppc_pcc_data *pcc_ss_data;
1776 
1777 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1778 	if (!cpc_desc)
1779 		return CPUFREQ_ETERNAL;
1780 
1781 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1782 	if (CPC_IN_SYSTEM_MEMORY(desired_reg) || CPC_IN_SYSTEM_IO(desired_reg))
1783 		return 0;
1784 	else if (!CPC_IN_PCC(desired_reg))
1785 		return CPUFREQ_ETERNAL;
1786 
1787 	if (pcc_ss_id < 0)
1788 		return CPUFREQ_ETERNAL;
1789 
1790 	pcc_ss_data = pcc_data[pcc_ss_id];
1791 	if (pcc_ss_data->pcc_mpar)
1792 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1793 
1794 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1795 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1796 
1797 	return latency_ns;
1798 }
1799 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1800 
1801 /* Minimum struct length needed for the DMI processor entry we want */
1802 #define DMI_ENTRY_PROCESSOR_MIN_LENGTH	48
1803 
1804 /* Offset in the DMI processor structure for the max frequency */
1805 #define DMI_PROCESSOR_MAX_SPEED		0x14
1806 
1807 /* Callback function used to retrieve the max frequency from DMI */
1808 static void cppc_find_dmi_mhz(const struct dmi_header *dm, void *private)
1809 {
1810 	const u8 *dmi_data = (const u8 *)dm;
1811 	u16 *mhz = (u16 *)private;
1812 
1813 	if (dm->type == DMI_ENTRY_PROCESSOR &&
1814 	    dm->length >= DMI_ENTRY_PROCESSOR_MIN_LENGTH) {
1815 		u16 val = (u16)get_unaligned((const u16 *)
1816 				(dmi_data + DMI_PROCESSOR_MAX_SPEED));
1817 		*mhz = val > *mhz ? val : *mhz;
1818 	}
1819 }
1820 
1821 /* Look up the max frequency in DMI */
1822 static u64 cppc_get_dmi_max_khz(void)
1823 {
1824 	u16 mhz = 0;
1825 
1826 	dmi_walk(cppc_find_dmi_mhz, &mhz);
1827 
1828 	/*
1829 	 * Real stupid fallback value, just in case there is no
1830 	 * actual value set.
1831 	 */
1832 	mhz = mhz ? mhz : 1;
1833 
1834 	return KHZ_PER_MHZ * mhz;
1835 }
1836 
1837 /*
1838  * If CPPC lowest_freq and nominal_freq registers are exposed then we can
1839  * use them to convert perf to freq and vice versa. The conversion is
1840  * extrapolated as an affine function passing by the 2 points:
1841  *  - (Low perf, Low freq)
1842  *  - (Nominal perf, Nominal freq)
1843  */
1844 unsigned int cppc_perf_to_khz(struct cppc_perf_caps *caps, unsigned int perf)
1845 {
1846 	s64 retval, offset = 0;
1847 	static u64 max_khz;
1848 	u64 mul, div;
1849 
1850 	if (caps->lowest_freq && caps->nominal_freq) {
1851 		mul = caps->nominal_freq - caps->lowest_freq;
1852 		mul *= KHZ_PER_MHZ;
1853 		div = caps->nominal_perf - caps->lowest_perf;
1854 		offset = caps->nominal_freq * KHZ_PER_MHZ -
1855 			 div64_u64(caps->nominal_perf * mul, div);
1856 	} else {
1857 		if (!max_khz)
1858 			max_khz = cppc_get_dmi_max_khz();
1859 		mul = max_khz;
1860 		div = caps->highest_perf;
1861 	}
1862 
1863 	retval = offset + div64_u64(perf * mul, div);
1864 	if (retval >= 0)
1865 		return retval;
1866 	return 0;
1867 }
1868 EXPORT_SYMBOL_GPL(cppc_perf_to_khz);
1869 
1870 unsigned int cppc_khz_to_perf(struct cppc_perf_caps *caps, unsigned int freq)
1871 {
1872 	s64 retval, offset = 0;
1873 	static u64 max_khz;
1874 	u64  mul, div;
1875 
1876 	if (caps->lowest_freq && caps->nominal_freq) {
1877 		mul = caps->nominal_perf - caps->lowest_perf;
1878 		div = caps->nominal_freq - caps->lowest_freq;
1879 		/*
1880 		 * We don't need to convert to kHz for computing offset and can
1881 		 * directly use nominal_freq and lowest_freq as the div64_u64
1882 		 * will remove the frequency unit.
1883 		 */
1884 		offset = caps->nominal_perf -
1885 			 div64_u64(caps->nominal_freq * mul, div);
1886 		/* But we need it for computing the perf level. */
1887 		div *= KHZ_PER_MHZ;
1888 	} else {
1889 		if (!max_khz)
1890 			max_khz = cppc_get_dmi_max_khz();
1891 		mul = caps->highest_perf;
1892 		div = max_khz;
1893 	}
1894 
1895 	retval = offset + div64_u64(freq * mul, div);
1896 	if (retval >= 0)
1897 		return retval;
1898 	return 0;
1899 }
1900 EXPORT_SYMBOL_GPL(cppc_khz_to_perf);
1901