xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision fe7498ef)
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 
43 #include <acpi/cppc_acpi.h>
44 
45 struct cppc_pcc_data {
46 	struct pcc_mbox_chan *pcc_channel;
47 	void __iomem *pcc_comm_addr;
48 	bool pcc_channel_acquired;
49 	unsigned int deadline_us;
50 	unsigned int pcc_mpar, pcc_mrtt, pcc_nominal;
51 
52 	bool pending_pcc_write_cmd;	/* Any pending/batched PCC write cmds? */
53 	bool platform_owns_pcc;		/* Ownership of PCC subspace */
54 	unsigned int pcc_write_cnt;	/* Running count of PCC write commands */
55 
56 	/*
57 	 * Lock to provide controlled access to the PCC channel.
58 	 *
59 	 * For performance critical usecases(currently cppc_set_perf)
60 	 *	We need to take read_lock and check if channel belongs to OSPM
61 	 * before reading or writing to PCC subspace
62 	 *	We need to take write_lock before transferring the channel
63 	 * ownership to the platform via a Doorbell
64 	 *	This allows us to batch a number of CPPC requests if they happen
65 	 * to originate in about the same time
66 	 *
67 	 * For non-performance critical usecases(init)
68 	 *	Take write_lock for all purposes which gives exclusive access
69 	 */
70 	struct rw_semaphore pcc_lock;
71 
72 	/* Wait queue for CPUs whose requests were batched */
73 	wait_queue_head_t pcc_write_wait_q;
74 	ktime_t last_cmd_cmpl_time;
75 	ktime_t last_mpar_reset;
76 	int mpar_count;
77 	int refcount;
78 };
79 
80 /* Array to represent the PCC channel per subspace ID */
81 static struct cppc_pcc_data *pcc_data[MAX_PCC_SUBSPACES];
82 /* The cpu_pcc_subspace_idx contains per CPU subspace ID */
83 static DEFINE_PER_CPU(int, cpu_pcc_subspace_idx);
84 
85 /*
86  * The cpc_desc structure contains the ACPI register details
87  * as described in the per CPU _CPC tables. The details
88  * include the type of register (e.g. PCC, System IO, FFH etc.)
89  * and destination addresses which lets us READ/WRITE CPU performance
90  * information using the appropriate I/O methods.
91  */
92 static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
93 
94 /* pcc mapped address + header size + offset within PCC subspace */
95 #define GET_PCC_VADDR(offs, pcc_ss_id) (pcc_data[pcc_ss_id]->pcc_comm_addr + \
96 						0x8 + (offs))
97 
98 /* Check if a CPC register is in PCC */
99 #define CPC_IN_PCC(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&		\
100 				(cpc)->cpc_entry.reg.space_id ==	\
101 				ACPI_ADR_SPACE_PLATFORM_COMM)
102 
103 /* Evaluates to True if reg is a NULL register descriptor */
104 #define IS_NULL_REG(reg) ((reg)->space_id ==  ACPI_ADR_SPACE_SYSTEM_MEMORY && \
105 				(reg)->address == 0 &&			\
106 				(reg)->bit_width == 0 &&		\
107 				(reg)->bit_offset == 0 &&		\
108 				(reg)->access_width == 0)
109 
110 /* Evaluates to True if an optional cpc field is supported */
111 #define CPC_SUPPORTED(cpc) ((cpc)->type == ACPI_TYPE_INTEGER ?		\
112 				!!(cpc)->cpc_entry.int_value :		\
113 				!IS_NULL_REG(&(cpc)->cpc_entry.reg))
114 /*
115  * Arbitrary Retries in case the remote processor is slow to respond
116  * to PCC commands. Keeping it high enough to cover emulators where
117  * the processors run painfully slow.
118  */
119 #define NUM_RETRIES 500ULL
120 
121 #define define_one_cppc_ro(_name)		\
122 static struct kobj_attribute _name =		\
123 __ATTR(_name, 0444, show_##_name, NULL)
124 
125 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
126 
127 #define show_cppc_data(access_fn, struct_name, member_name)		\
128 	static ssize_t show_##member_name(struct kobject *kobj,		\
129 				struct kobj_attribute *attr, char *buf)	\
130 	{								\
131 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
132 		struct struct_name st_name = {0};			\
133 		int ret;						\
134 									\
135 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
136 		if (ret)						\
137 			return ret;					\
138 									\
139 		return scnprintf(buf, PAGE_SIZE, "%llu\n",		\
140 				(u64)st_name.member_name);		\
141 	}								\
142 	define_one_cppc_ro(member_name)
143 
144 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
145 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
146 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
147 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
148 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_freq);
149 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_freq);
150 
151 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, reference_perf);
152 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
153 
154 static ssize_t show_feedback_ctrs(struct kobject *kobj,
155 		struct kobj_attribute *attr, char *buf)
156 {
157 	struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);
158 	struct cppc_perf_fb_ctrs fb_ctrs = {0};
159 	int ret;
160 
161 	ret = cppc_get_perf_ctrs(cpc_ptr->cpu_id, &fb_ctrs);
162 	if (ret)
163 		return ret;
164 
165 	return scnprintf(buf, PAGE_SIZE, "ref:%llu del:%llu\n",
166 			fb_ctrs.reference, fb_ctrs.delivered);
167 }
168 define_one_cppc_ro(feedback_ctrs);
169 
170 static struct attribute *cppc_attrs[] = {
171 	&feedback_ctrs.attr,
172 	&reference_perf.attr,
173 	&wraparound_time.attr,
174 	&highest_perf.attr,
175 	&lowest_perf.attr,
176 	&lowest_nonlinear_perf.attr,
177 	&nominal_perf.attr,
178 	&nominal_freq.attr,
179 	&lowest_freq.attr,
180 	NULL
181 };
182 
183 static struct kobj_type cppc_ktype = {
184 	.sysfs_ops = &kobj_sysfs_ops,
185 	.default_attrs = cppc_attrs,
186 };
187 
188 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
189 {
190 	int ret, status;
191 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
192 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
193 		pcc_ss_data->pcc_comm_addr;
194 
195 	if (!pcc_ss_data->platform_owns_pcc)
196 		return 0;
197 
198 	/*
199 	 * Poll PCC status register every 3us(delay_us) for maximum of
200 	 * deadline_us(timeout_us) until PCC command complete bit is set(cond)
201 	 */
202 	ret = readw_relaxed_poll_timeout(&generic_comm_base->status, status,
203 					status & PCC_CMD_COMPLETE_MASK, 3,
204 					pcc_ss_data->deadline_us);
205 
206 	if (likely(!ret)) {
207 		pcc_ss_data->platform_owns_pcc = false;
208 		if (chk_err_bit && (status & PCC_ERROR_MASK))
209 			ret = -EIO;
210 	}
211 
212 	if (unlikely(ret))
213 		pr_err("PCC check channel failed for ss: %d. ret=%d\n",
214 		       pcc_ss_id, ret);
215 
216 	return ret;
217 }
218 
219 /*
220  * This function transfers the ownership of the PCC to the platform
221  * So it must be called while holding write_lock(pcc_lock)
222  */
223 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
224 {
225 	int ret = -EIO, i;
226 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
227 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
228 		pcc_ss_data->pcc_comm_addr;
229 	unsigned int time_delta;
230 
231 	/*
232 	 * For CMD_WRITE we know for a fact the caller should have checked
233 	 * the channel before writing to PCC space
234 	 */
235 	if (cmd == CMD_READ) {
236 		/*
237 		 * If there are pending cpc_writes, then we stole the channel
238 		 * before write completion, so first send a WRITE command to
239 		 * platform
240 		 */
241 		if (pcc_ss_data->pending_pcc_write_cmd)
242 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
243 
244 		ret = check_pcc_chan(pcc_ss_id, false);
245 		if (ret)
246 			goto end;
247 	} else /* CMD_WRITE */
248 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
249 
250 	/*
251 	 * Handle the Minimum Request Turnaround Time(MRTT)
252 	 * "The minimum amount of time that OSPM must wait after the completion
253 	 * of a command before issuing the next command, in microseconds"
254 	 */
255 	if (pcc_ss_data->pcc_mrtt) {
256 		time_delta = ktime_us_delta(ktime_get(),
257 					    pcc_ss_data->last_cmd_cmpl_time);
258 		if (pcc_ss_data->pcc_mrtt > time_delta)
259 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
260 	}
261 
262 	/*
263 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
264 	 * "The maximum number of periodic requests that the subspace channel can
265 	 * support, reported in commands per minute. 0 indicates no limitation."
266 	 *
267 	 * This parameter should be ideally zero or large enough so that it can
268 	 * handle maximum number of requests that all the cores in the system can
269 	 * collectively generate. If it is not, we will follow the spec and just
270 	 * not send the request to the platform after hitting the MPAR limit in
271 	 * any 60s window
272 	 */
273 	if (pcc_ss_data->pcc_mpar) {
274 		if (pcc_ss_data->mpar_count == 0) {
275 			time_delta = ktime_ms_delta(ktime_get(),
276 						    pcc_ss_data->last_mpar_reset);
277 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
278 				pr_debug("PCC cmd for subspace %d not sent due to MPAR limit",
279 					 pcc_ss_id);
280 				ret = -EIO;
281 				goto end;
282 			}
283 			pcc_ss_data->last_mpar_reset = ktime_get();
284 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
285 		}
286 		pcc_ss_data->mpar_count--;
287 	}
288 
289 	/* Write to the shared comm region. */
290 	writew_relaxed(cmd, &generic_comm_base->command);
291 
292 	/* Flip CMD COMPLETE bit */
293 	writew_relaxed(0, &generic_comm_base->status);
294 
295 	pcc_ss_data->platform_owns_pcc = true;
296 
297 	/* Ring doorbell */
298 	ret = mbox_send_message(pcc_ss_data->pcc_channel->mchan, &cmd);
299 	if (ret < 0) {
300 		pr_err("Err sending PCC mbox message. ss: %d cmd:%d, ret:%d\n",
301 		       pcc_ss_id, cmd, ret);
302 		goto end;
303 	}
304 
305 	/* wait for completion and check for PCC errro bit */
306 	ret = check_pcc_chan(pcc_ss_id, true);
307 
308 	if (pcc_ss_data->pcc_mrtt)
309 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
310 
311 	if (pcc_ss_data->pcc_channel->mchan->mbox->txdone_irq)
312 		mbox_chan_txdone(pcc_ss_data->pcc_channel->mchan, ret);
313 	else
314 		mbox_client_txdone(pcc_ss_data->pcc_channel->mchan, ret);
315 
316 end:
317 	if (cmd == CMD_WRITE) {
318 		if (unlikely(ret)) {
319 			for_each_possible_cpu(i) {
320 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
321 
322 				if (!desc)
323 					continue;
324 
325 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
326 					desc->write_cmd_status = ret;
327 			}
328 		}
329 		pcc_ss_data->pcc_write_cnt++;
330 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
331 	}
332 
333 	return ret;
334 }
335 
336 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
337 {
338 	if (ret < 0)
339 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
340 				*(u16 *)msg, ret);
341 	else
342 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
343 				*(u16 *)msg, ret);
344 }
345 
346 static struct mbox_client cppc_mbox_cl = {
347 	.tx_done = cppc_chan_tx_done,
348 	.knows_txdone = true,
349 };
350 
351 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
352 {
353 	int result = -EFAULT;
354 	acpi_status status = AE_OK;
355 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
356 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
357 	struct acpi_buffer state = {0, NULL};
358 	union acpi_object  *psd = NULL;
359 	struct acpi_psd_package *pdomain;
360 
361 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL,
362 					    &buffer, ACPI_TYPE_PACKAGE);
363 	if (status == AE_NOT_FOUND)	/* _PSD is optional */
364 		return 0;
365 	if (ACPI_FAILURE(status))
366 		return -ENODEV;
367 
368 	psd = buffer.pointer;
369 	if (!psd || psd->package.count != 1) {
370 		pr_debug("Invalid _PSD data\n");
371 		goto end;
372 	}
373 
374 	pdomain = &(cpc_ptr->domain_info);
375 
376 	state.length = sizeof(struct acpi_psd_package);
377 	state.pointer = pdomain;
378 
379 	status = acpi_extract_package(&(psd->package.elements[0]),
380 		&format, &state);
381 	if (ACPI_FAILURE(status)) {
382 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
383 		goto end;
384 	}
385 
386 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
387 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
388 		goto end;
389 	}
390 
391 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
392 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
393 		goto end;
394 	}
395 
396 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
397 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
398 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
399 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
400 		goto end;
401 	}
402 
403 	result = 0;
404 end:
405 	kfree(buffer.pointer);
406 	return result;
407 }
408 
409 bool acpi_cpc_valid(void)
410 {
411 	struct cpc_desc *cpc_ptr;
412 	int cpu;
413 
414 	for_each_possible_cpu(cpu) {
415 		cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
416 		if (!cpc_ptr)
417 			return false;
418 	}
419 
420 	return true;
421 }
422 EXPORT_SYMBOL_GPL(acpi_cpc_valid);
423 
424 /**
425  * acpi_get_psd_map - Map the CPUs in the freq domain of a given cpu
426  * @cpu: Find all CPUs that share a domain with cpu.
427  * @cpu_data: Pointer to CPU specific CPPC data including PSD info.
428  *
429  *	Return: 0 for success or negative value for err.
430  */
431 int acpi_get_psd_map(unsigned int cpu, struct cppc_cpudata *cpu_data)
432 {
433 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
434 	struct acpi_psd_package *match_pdomain;
435 	struct acpi_psd_package *pdomain;
436 	int count_target, i;
437 
438 	/*
439 	 * Now that we have _PSD data from all CPUs, let's setup P-state
440 	 * domain info.
441 	 */
442 	cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
443 	if (!cpc_ptr)
444 		return -EFAULT;
445 
446 	pdomain = &(cpc_ptr->domain_info);
447 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
448 	if (pdomain->num_processors <= 1)
449 		return 0;
450 
451 	/* Validate the Domain info */
452 	count_target = pdomain->num_processors;
453 	if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
454 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ALL;
455 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
456 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_HW;
457 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
458 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ANY;
459 
460 	for_each_possible_cpu(i) {
461 		if (i == cpu)
462 			continue;
463 
464 		match_cpc_ptr = per_cpu(cpc_desc_ptr, i);
465 		if (!match_cpc_ptr)
466 			goto err_fault;
467 
468 		match_pdomain = &(match_cpc_ptr->domain_info);
469 		if (match_pdomain->domain != pdomain->domain)
470 			continue;
471 
472 		/* Here i and cpu are in the same domain */
473 		if (match_pdomain->num_processors != count_target)
474 			goto err_fault;
475 
476 		if (pdomain->coord_type != match_pdomain->coord_type)
477 			goto err_fault;
478 
479 		cpumask_set_cpu(i, cpu_data->shared_cpu_map);
480 	}
481 
482 	return 0;
483 
484 err_fault:
485 	/* Assume no coordination on any error parsing domain info */
486 	cpumask_clear(cpu_data->shared_cpu_map);
487 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
488 	cpu_data->shared_type = CPUFREQ_SHARED_TYPE_NONE;
489 
490 	return -EFAULT;
491 }
492 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
493 
494 static int register_pcc_channel(int pcc_ss_idx)
495 {
496 	struct pcc_mbox_chan *pcc_chan;
497 	u64 usecs_lat;
498 
499 	if (pcc_ss_idx >= 0) {
500 		pcc_chan = pcc_mbox_request_channel(&cppc_mbox_cl, pcc_ss_idx);
501 
502 		if (IS_ERR(pcc_chan)) {
503 			pr_err("Failed to find PCC channel for subspace %d\n",
504 			       pcc_ss_idx);
505 			return -ENODEV;
506 		}
507 
508 		pcc_data[pcc_ss_idx]->pcc_channel = pcc_chan;
509 		/*
510 		 * cppc_ss->latency is just a Nominal value. In reality
511 		 * the remote processor could be much slower to reply.
512 		 * So add an arbitrary amount of wait on top of Nominal.
513 		 */
514 		usecs_lat = NUM_RETRIES * pcc_chan->latency;
515 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
516 		pcc_data[pcc_ss_idx]->pcc_mrtt = pcc_chan->min_turnaround_time;
517 		pcc_data[pcc_ss_idx]->pcc_mpar = pcc_chan->max_access_rate;
518 		pcc_data[pcc_ss_idx]->pcc_nominal = pcc_chan->latency;
519 
520 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
521 			acpi_os_ioremap(pcc_chan->shmem_base_addr,
522 					pcc_chan->shmem_size);
523 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
524 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
525 			       pcc_ss_idx);
526 			return -ENOMEM;
527 		}
528 
529 		/* Set flag so that we don't come here for each CPU. */
530 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
531 	}
532 
533 	return 0;
534 }
535 
536 /**
537  * cpc_ffh_supported() - check if FFH reading supported
538  *
539  * Check if the architecture has support for functional fixed hardware
540  * read/write capability.
541  *
542  * Return: true for supported, false for not supported
543  */
544 bool __weak cpc_ffh_supported(void)
545 {
546 	return false;
547 }
548 
549 /**
550  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
551  *
552  * Check and allocate the cppc_pcc_data memory.
553  * In some processor configurations it is possible that same subspace
554  * is shared between multiple CPUs. This is seen especially in CPUs
555  * with hardware multi-threading support.
556  *
557  * Return: 0 for success, errno for failure
558  */
559 static int pcc_data_alloc(int pcc_ss_id)
560 {
561 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
562 		return -EINVAL;
563 
564 	if (pcc_data[pcc_ss_id]) {
565 		pcc_data[pcc_ss_id]->refcount++;
566 	} else {
567 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
568 					      GFP_KERNEL);
569 		if (!pcc_data[pcc_ss_id])
570 			return -ENOMEM;
571 		pcc_data[pcc_ss_id]->refcount++;
572 	}
573 
574 	return 0;
575 }
576 
577 /* Check if CPPC revision + num_ent combination is supported */
578 static bool is_cppc_supported(int revision, int num_ent)
579 {
580 	int expected_num_ent;
581 
582 	switch (revision) {
583 	case CPPC_V2_REV:
584 		expected_num_ent = CPPC_V2_NUM_ENT;
585 		break;
586 	case CPPC_V3_REV:
587 		expected_num_ent = CPPC_V3_NUM_ENT;
588 		break;
589 	default:
590 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
591 			revision);
592 		return false;
593 	}
594 
595 	if (expected_num_ent != num_ent) {
596 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
597 			num_ent, expected_num_ent, revision);
598 		return false;
599 	}
600 
601 	return true;
602 }
603 
604 /*
605  * An example CPC table looks like the following.
606  *
607  *	Name(_CPC, Package()
608  *			{
609  *			17,
610  *			NumEntries
611  *			1,
612  *			// Revision
613  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
614  *			// Highest Performance
615  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
616  *			// Nominal Performance
617  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
618  *			// Lowest Nonlinear Performance
619  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
620  *			// Lowest Performance
621  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
622  *			// Guaranteed Performance Register
623  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
624  *			// Desired Performance Register
625  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
626  *			..
627  *			..
628  *			..
629  *
630  *		}
631  * Each Register() encodes how to access that specific register.
632  * e.g. a sample PCC entry has the following encoding:
633  *
634  *	Register (
635  *		PCC,
636  *		AddressSpaceKeyword
637  *		8,
638  *		//RegisterBitWidth
639  *		8,
640  *		//RegisterBitOffset
641  *		0x30,
642  *		//RegisterAddress
643  *		9
644  *		//AccessSize (subspace ID)
645  *		0
646  *		)
647  *	}
648  */
649 
650 #ifndef init_freq_invariance_cppc
651 static inline void init_freq_invariance_cppc(void) { }
652 #endif
653 
654 /**
655  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
656  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
657  *
658  *	Return: 0 for success or negative value for err.
659  */
660 int acpi_cppc_processor_probe(struct acpi_processor *pr)
661 {
662 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
663 	union acpi_object *out_obj, *cpc_obj;
664 	struct cpc_desc *cpc_ptr;
665 	struct cpc_reg *gas_t;
666 	struct device *cpu_dev;
667 	acpi_handle handle = pr->handle;
668 	unsigned int num_ent, i, cpc_rev;
669 	int pcc_subspace_id = -1;
670 	acpi_status status;
671 	int ret = -EFAULT;
672 
673 	/* Parse the ACPI _CPC table for this CPU. */
674 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
675 			ACPI_TYPE_PACKAGE);
676 	if (ACPI_FAILURE(status)) {
677 		ret = -ENODEV;
678 		goto out_buf_free;
679 	}
680 
681 	out_obj = (union acpi_object *) output.pointer;
682 
683 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
684 	if (!cpc_ptr) {
685 		ret = -ENOMEM;
686 		goto out_buf_free;
687 	}
688 
689 	/* First entry is NumEntries. */
690 	cpc_obj = &out_obj->package.elements[0];
691 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
692 		num_ent = cpc_obj->integer.value;
693 	} else {
694 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
695 				cpc_obj->type);
696 		goto out_free;
697 	}
698 	cpc_ptr->num_entries = num_ent;
699 
700 	/* Second entry should be revision. */
701 	cpc_obj = &out_obj->package.elements[1];
702 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
703 		cpc_rev = cpc_obj->integer.value;
704 	} else {
705 		pr_debug("Unexpected entry type(%d) for Revision\n",
706 				cpc_obj->type);
707 		goto out_free;
708 	}
709 	cpc_ptr->version = cpc_rev;
710 
711 	if (!is_cppc_supported(cpc_rev, num_ent))
712 		goto out_free;
713 
714 	/* Iterate through remaining entries in _CPC */
715 	for (i = 2; i < num_ent; i++) {
716 		cpc_obj = &out_obj->package.elements[i];
717 
718 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
719 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
720 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
721 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
722 			gas_t = (struct cpc_reg *)
723 				cpc_obj->buffer.pointer;
724 
725 			/*
726 			 * The PCC Subspace index is encoded inside
727 			 * the CPC table entries. The same PCC index
728 			 * will be used for all the PCC entries,
729 			 * so extract it only once.
730 			 */
731 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
732 				if (pcc_subspace_id < 0) {
733 					pcc_subspace_id = gas_t->access_width;
734 					if (pcc_data_alloc(pcc_subspace_id))
735 						goto out_free;
736 				} else if (pcc_subspace_id != gas_t->access_width) {
737 					pr_debug("Mismatched PCC ids.\n");
738 					goto out_free;
739 				}
740 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
741 				if (gas_t->address) {
742 					void __iomem *addr;
743 
744 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
745 					if (!addr)
746 						goto out_free;
747 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
748 				}
749 			} else {
750 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
751 					/* Support only PCC ,SYS MEM and FFH type regs */
752 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
753 					goto out_free;
754 				}
755 			}
756 
757 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
758 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
759 		} else {
760 			pr_debug("Err in entry:%d in CPC table of CPU:%d\n", i, pr->id);
761 			goto out_free;
762 		}
763 	}
764 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
765 
766 	/*
767 	 * Initialize the remaining cpc_regs as unsupported.
768 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
769 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
770 	 */
771 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
772 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
773 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
774 	}
775 
776 
777 	/* Store CPU Logical ID */
778 	cpc_ptr->cpu_id = pr->id;
779 
780 	/* Parse PSD data for this CPU */
781 	ret = acpi_get_psd(cpc_ptr, handle);
782 	if (ret)
783 		goto out_free;
784 
785 	/* Register PCC channel once for all PCC subspace ID. */
786 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
787 		ret = register_pcc_channel(pcc_subspace_id);
788 		if (ret)
789 			goto out_free;
790 
791 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
792 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
793 	}
794 
795 	/* Everything looks okay */
796 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
797 
798 	/* Add per logical CPU nodes for reading its feedback counters. */
799 	cpu_dev = get_cpu_device(pr->id);
800 	if (!cpu_dev) {
801 		ret = -EINVAL;
802 		goto out_free;
803 	}
804 
805 	/* Plug PSD data into this CPU's CPC descriptor. */
806 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
807 
808 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
809 			"acpi_cppc");
810 	if (ret) {
811 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
812 		kobject_put(&cpc_ptr->kobj);
813 		goto out_free;
814 	}
815 
816 	init_freq_invariance_cppc();
817 
818 	kfree(output.pointer);
819 	return 0;
820 
821 out_free:
822 	/* Free all the mapped sys mem areas for this CPU */
823 	for (i = 2; i < cpc_ptr->num_entries; i++) {
824 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
825 
826 		if (addr)
827 			iounmap(addr);
828 	}
829 	kfree(cpc_ptr);
830 
831 out_buf_free:
832 	kfree(output.pointer);
833 	return ret;
834 }
835 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
836 
837 /**
838  * acpi_cppc_processor_exit - Cleanup CPC structs.
839  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
840  *
841  * Return: Void
842  */
843 void acpi_cppc_processor_exit(struct acpi_processor *pr)
844 {
845 	struct cpc_desc *cpc_ptr;
846 	unsigned int i;
847 	void __iomem *addr;
848 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
849 
850 	if (pcc_ss_id >= 0 && pcc_data[pcc_ss_id]) {
851 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
852 			pcc_data[pcc_ss_id]->refcount--;
853 			if (!pcc_data[pcc_ss_id]->refcount) {
854 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
855 				kfree(pcc_data[pcc_ss_id]);
856 				pcc_data[pcc_ss_id] = NULL;
857 			}
858 		}
859 	}
860 
861 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
862 	if (!cpc_ptr)
863 		return;
864 
865 	/* Free all the mapped sys mem areas for this CPU */
866 	for (i = 2; i < cpc_ptr->num_entries; i++) {
867 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
868 		if (addr)
869 			iounmap(addr);
870 	}
871 
872 	kobject_put(&cpc_ptr->kobj);
873 	kfree(cpc_ptr);
874 }
875 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
876 
877 /**
878  * cpc_read_ffh() - Read FFH register
879  * @cpunum:	CPU number to read
880  * @reg:	cppc register information
881  * @val:	place holder for return value
882  *
883  * Read bit_width bits from a specified address and bit_offset
884  *
885  * Return: 0 for success and error code
886  */
887 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
888 {
889 	return -ENOTSUPP;
890 }
891 
892 /**
893  * cpc_write_ffh() - Write FFH register
894  * @cpunum:	CPU number to write
895  * @reg:	cppc register information
896  * @val:	value to write
897  *
898  * Write value of bit_width bits to a specified address and bit_offset
899  *
900  * Return: 0 for success and error code
901  */
902 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
903 {
904 	return -ENOTSUPP;
905 }
906 
907 /*
908  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
909  * as fast as possible. We have already mapped the PCC subspace during init, so
910  * we can directly write to it.
911  */
912 
913 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
914 {
915 	int ret_val = 0;
916 	void __iomem *vaddr = NULL;
917 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
918 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
919 
920 	if (reg_res->type == ACPI_TYPE_INTEGER) {
921 		*val = reg_res->cpc_entry.int_value;
922 		return ret_val;
923 	}
924 
925 	*val = 0;
926 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
927 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
928 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
929 		vaddr = reg_res->sys_mem_vaddr;
930 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
931 		return cpc_read_ffh(cpu, reg, val);
932 	else
933 		return acpi_os_read_memory((acpi_physical_address)reg->address,
934 				val, reg->bit_width);
935 
936 	switch (reg->bit_width) {
937 	case 8:
938 		*val = readb_relaxed(vaddr);
939 		break;
940 	case 16:
941 		*val = readw_relaxed(vaddr);
942 		break;
943 	case 32:
944 		*val = readl_relaxed(vaddr);
945 		break;
946 	case 64:
947 		*val = readq_relaxed(vaddr);
948 		break;
949 	default:
950 		pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
951 			 reg->bit_width, pcc_ss_id);
952 		ret_val = -EFAULT;
953 	}
954 
955 	return ret_val;
956 }
957 
958 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
959 {
960 	int ret_val = 0;
961 	void __iomem *vaddr = NULL;
962 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
963 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
964 
965 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
966 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
967 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
968 		vaddr = reg_res->sys_mem_vaddr;
969 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
970 		return cpc_write_ffh(cpu, reg, val);
971 	else
972 		return acpi_os_write_memory((acpi_physical_address)reg->address,
973 				val, reg->bit_width);
974 
975 	switch (reg->bit_width) {
976 	case 8:
977 		writeb_relaxed(val, vaddr);
978 		break;
979 	case 16:
980 		writew_relaxed(val, vaddr);
981 		break;
982 	case 32:
983 		writel_relaxed(val, vaddr);
984 		break;
985 	case 64:
986 		writeq_relaxed(val, vaddr);
987 		break;
988 	default:
989 		pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
990 			 reg->bit_width, pcc_ss_id);
991 		ret_val = -EFAULT;
992 		break;
993 	}
994 
995 	return ret_val;
996 }
997 
998 static int cppc_get_perf(int cpunum, enum cppc_regs reg_idx, u64 *perf)
999 {
1000 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1001 	struct cpc_register_resource *reg = &cpc_desc->cpc_regs[reg_idx];
1002 
1003 	if (CPC_IN_PCC(reg)) {
1004 		int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1005 		struct cppc_pcc_data *pcc_ss_data = NULL;
1006 		int ret = 0;
1007 
1008 		if (pcc_ss_id < 0)
1009 			return -EIO;
1010 
1011 		pcc_ss_data = pcc_data[pcc_ss_id];
1012 
1013 		down_write(&pcc_ss_data->pcc_lock);
1014 
1015 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1016 			cpc_read(cpunum, reg, perf);
1017 		else
1018 			ret = -EIO;
1019 
1020 		up_write(&pcc_ss_data->pcc_lock);
1021 
1022 		return ret;
1023 	}
1024 
1025 	cpc_read(cpunum, reg, perf);
1026 
1027 	return 0;
1028 }
1029 
1030 /**
1031  * cppc_get_desired_perf - Get the desired performance register value.
1032  * @cpunum: CPU from which to get desired performance.
1033  * @desired_perf: Return address.
1034  *
1035  * Return: 0 for success, -EIO otherwise.
1036  */
1037 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1038 {
1039 	return cppc_get_perf(cpunum, DESIRED_PERF, desired_perf);
1040 }
1041 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1042 
1043 /**
1044  * cppc_get_nominal_perf - Get the nominal performance register value.
1045  * @cpunum: CPU from which to get nominal performance.
1046  * @nominal_perf: Return address.
1047  *
1048  * Return: 0 for success, -EIO otherwise.
1049  */
1050 int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
1051 {
1052 	return cppc_get_perf(cpunum, NOMINAL_PERF, nominal_perf);
1053 }
1054 
1055 /**
1056  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1057  * @cpunum: CPU from which to get capabilities info.
1058  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1059  *
1060  * Return: 0 for success with perf_caps populated else -ERRNO.
1061  */
1062 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1063 {
1064 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1065 	struct cpc_register_resource *highest_reg, *lowest_reg,
1066 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1067 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1068 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1069 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1070 	struct cppc_pcc_data *pcc_ss_data = NULL;
1071 	int ret = 0, regs_in_pcc = 0;
1072 
1073 	if (!cpc_desc) {
1074 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1075 		return -ENODEV;
1076 	}
1077 
1078 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1079 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1080 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1081 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1082 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1083 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1084 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1085 
1086 	/* Are any of the regs PCC ?*/
1087 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1088 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1089 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1090 		if (pcc_ss_id < 0) {
1091 			pr_debug("Invalid pcc_ss_id\n");
1092 			return -ENODEV;
1093 		}
1094 		pcc_ss_data = pcc_data[pcc_ss_id];
1095 		regs_in_pcc = 1;
1096 		down_write(&pcc_ss_data->pcc_lock);
1097 		/* Ring doorbell once to update PCC subspace */
1098 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1099 			ret = -EIO;
1100 			goto out_err;
1101 		}
1102 	}
1103 
1104 	cpc_read(cpunum, highest_reg, &high);
1105 	perf_caps->highest_perf = high;
1106 
1107 	cpc_read(cpunum, lowest_reg, &low);
1108 	perf_caps->lowest_perf = low;
1109 
1110 	cpc_read(cpunum, nominal_reg, &nom);
1111 	perf_caps->nominal_perf = nom;
1112 
1113 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1114 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1115 		perf_caps->guaranteed_perf = 0;
1116 	} else {
1117 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1118 		perf_caps->guaranteed_perf = guaranteed;
1119 	}
1120 
1121 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1122 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1123 
1124 	if (!high || !low || !nom || !min_nonlinear)
1125 		ret = -EFAULT;
1126 
1127 	/* Read optional lowest and nominal frequencies if present */
1128 	if (CPC_SUPPORTED(low_freq_reg))
1129 		cpc_read(cpunum, low_freq_reg, &low_f);
1130 
1131 	if (CPC_SUPPORTED(nom_freq_reg))
1132 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1133 
1134 	perf_caps->lowest_freq = low_f;
1135 	perf_caps->nominal_freq = nom_f;
1136 
1137 
1138 out_err:
1139 	if (regs_in_pcc)
1140 		up_write(&pcc_ss_data->pcc_lock);
1141 	return ret;
1142 }
1143 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1144 
1145 /**
1146  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1147  * @cpunum: CPU from which to read counters.
1148  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1149  *
1150  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1151  */
1152 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1153 {
1154 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1155 	struct cpc_register_resource *delivered_reg, *reference_reg,
1156 		*ref_perf_reg, *ctr_wrap_reg;
1157 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1158 	struct cppc_pcc_data *pcc_ss_data = NULL;
1159 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1160 	int ret = 0, regs_in_pcc = 0;
1161 
1162 	if (!cpc_desc) {
1163 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1164 		return -ENODEV;
1165 	}
1166 
1167 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1168 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1169 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1170 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1171 
1172 	/*
1173 	 * If reference perf register is not supported then we should
1174 	 * use the nominal perf value
1175 	 */
1176 	if (!CPC_SUPPORTED(ref_perf_reg))
1177 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1178 
1179 	/* Are any of the regs PCC ?*/
1180 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1181 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1182 		if (pcc_ss_id < 0) {
1183 			pr_debug("Invalid pcc_ss_id\n");
1184 			return -ENODEV;
1185 		}
1186 		pcc_ss_data = pcc_data[pcc_ss_id];
1187 		down_write(&pcc_ss_data->pcc_lock);
1188 		regs_in_pcc = 1;
1189 		/* Ring doorbell once to update PCC subspace */
1190 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1191 			ret = -EIO;
1192 			goto out_err;
1193 		}
1194 	}
1195 
1196 	cpc_read(cpunum, delivered_reg, &delivered);
1197 	cpc_read(cpunum, reference_reg, &reference);
1198 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1199 
1200 	/*
1201 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1202 	 * performance counters are assumed to never wrap during the lifetime of
1203 	 * platform
1204 	 */
1205 	ctr_wrap_time = (u64)(~((u64)0));
1206 	if (CPC_SUPPORTED(ctr_wrap_reg))
1207 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1208 
1209 	if (!delivered || !reference ||	!ref_perf) {
1210 		ret = -EFAULT;
1211 		goto out_err;
1212 	}
1213 
1214 	perf_fb_ctrs->delivered = delivered;
1215 	perf_fb_ctrs->reference = reference;
1216 	perf_fb_ctrs->reference_perf = ref_perf;
1217 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1218 out_err:
1219 	if (regs_in_pcc)
1220 		up_write(&pcc_ss_data->pcc_lock);
1221 	return ret;
1222 }
1223 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1224 
1225 /**
1226  * cppc_set_perf - Set a CPU's performance controls.
1227  * @cpu: CPU for which to set performance controls.
1228  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1229  *
1230  * Return: 0 for success, -ERRNO otherwise.
1231  */
1232 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1233 {
1234 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1235 	struct cpc_register_resource *desired_reg;
1236 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1237 	struct cppc_pcc_data *pcc_ss_data = NULL;
1238 	int ret = 0;
1239 
1240 	if (!cpc_desc) {
1241 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1242 		return -ENODEV;
1243 	}
1244 
1245 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1246 
1247 	/*
1248 	 * This is Phase-I where we want to write to CPC registers
1249 	 * -> We want all CPUs to be able to execute this phase in parallel
1250 	 *
1251 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1252 	 * achieve that goal here
1253 	 */
1254 	if (CPC_IN_PCC(desired_reg)) {
1255 		if (pcc_ss_id < 0) {
1256 			pr_debug("Invalid pcc_ss_id\n");
1257 			return -ENODEV;
1258 		}
1259 		pcc_ss_data = pcc_data[pcc_ss_id];
1260 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1261 		if (pcc_ss_data->platform_owns_pcc) {
1262 			ret = check_pcc_chan(pcc_ss_id, false);
1263 			if (ret) {
1264 				up_read(&pcc_ss_data->pcc_lock);
1265 				return ret;
1266 			}
1267 		}
1268 		/*
1269 		 * Update the pending_write to make sure a PCC CMD_READ will not
1270 		 * arrive and steal the channel during the switch to write lock
1271 		 */
1272 		pcc_ss_data->pending_pcc_write_cmd = true;
1273 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1274 		cpc_desc->write_cmd_status = 0;
1275 	}
1276 
1277 	/*
1278 	 * Skip writing MIN/MAX until Linux knows how to come up with
1279 	 * useful values.
1280 	 */
1281 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1282 
1283 	if (CPC_IN_PCC(desired_reg))
1284 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1285 	/*
1286 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1287 	 *
1288 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1289 	 * requests that happened in short overlapping interval. The last CPU to
1290 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1291 	 *
1292 	 * We have the following requirements for Phase-II:
1293 	 *     1. We want to execute Phase-II only when there are no CPUs
1294 	 * currently executing in Phase-I
1295 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1296 	 * entering Phase-I.
1297 	 *     3. We want only one CPU among all those who went through Phase-I
1298 	 * to run phase-II
1299 	 *
1300 	 * If write_trylock fails to get the lock and doesn't transfer the
1301 	 * PCC ownership to the platform, then one of the following will be TRUE
1302 	 *     1. There is at-least one CPU in Phase-I which will later execute
1303 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1304 	 * executing the Phase-II.
1305 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1306 	 * write_trylock and has already acquired the write_lock. We know for a
1307 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1308 	 * before this CPU's Phase-I as we held the read_lock.
1309 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1310 	 * down_write, in which case, send_pcc_cmd will check for pending
1311 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1312 	 * So this CPU can be certain that its request will be delivered
1313 	 *    So in all cases, this CPU knows that its request will be delivered
1314 	 * by another CPU and can return
1315 	 *
1316 	 * After getting the down_write we still need to check for
1317 	 * pending_pcc_write_cmd to take care of the following scenario
1318 	 *    The thread running this code could be scheduled out between
1319 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1320 	 * could have delivered the request to Platform by triggering the
1321 	 * doorbell and transferred the ownership of PCC to platform. So this
1322 	 * avoids triggering an unnecessary doorbell and more importantly before
1323 	 * triggering the doorbell it makes sure that the PCC channel ownership
1324 	 * is still with OSPM.
1325 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1326 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1327 	 * before the pcc CMD_WRITE is completed. send_pcc_cmd checks for this
1328 	 * case during a CMD_READ and if there are pending writes it delivers
1329 	 * the write command before servicing the read command
1330 	 */
1331 	if (CPC_IN_PCC(desired_reg)) {
1332 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1333 			/* Update only if there are pending write commands */
1334 			if (pcc_ss_data->pending_pcc_write_cmd)
1335 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1336 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1337 		} else
1338 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1339 			wait_event(pcc_ss_data->pcc_write_wait_q,
1340 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1341 
1342 		/* send_pcc_cmd updates the status in case of failure */
1343 		ret = cpc_desc->write_cmd_status;
1344 	}
1345 	return ret;
1346 }
1347 EXPORT_SYMBOL_GPL(cppc_set_perf);
1348 
1349 /**
1350  * cppc_get_transition_latency - returns frequency transition latency in ns
1351  *
1352  * ACPI CPPC does not explicitly specify how a platform can specify the
1353  * transition latency for performance change requests. The closest we have
1354  * is the timing information from the PCCT tables which provides the info
1355  * on the number and frequency of PCC commands the platform can handle.
1356  */
1357 unsigned int cppc_get_transition_latency(int cpu_num)
1358 {
1359 	/*
1360 	 * Expected transition latency is based on the PCCT timing values
1361 	 * Below are definition from ACPI spec:
1362 	 * pcc_nominal- Expected latency to process a command, in microseconds
1363 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1364 	 *              channel can support, reported in commands per minute. 0
1365 	 *              indicates no limitation.
1366 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1367 	 *              completion of a command before issuing the next command,
1368 	 *              in microseconds.
1369 	 */
1370 	unsigned int latency_ns = 0;
1371 	struct cpc_desc *cpc_desc;
1372 	struct cpc_register_resource *desired_reg;
1373 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1374 	struct cppc_pcc_data *pcc_ss_data;
1375 
1376 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1377 	if (!cpc_desc)
1378 		return CPUFREQ_ETERNAL;
1379 
1380 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1381 	if (!CPC_IN_PCC(desired_reg))
1382 		return CPUFREQ_ETERNAL;
1383 
1384 	if (pcc_ss_id < 0)
1385 		return CPUFREQ_ETERNAL;
1386 
1387 	pcc_ss_data = pcc_data[pcc_ss_id];
1388 	if (pcc_ss_data->pcc_mpar)
1389 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1390 
1391 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1392 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1393 
1394 	return latency_ns;
1395 }
1396 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1397