xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 7a010c3c)
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 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, &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->mbox->txdone_irq)
312 		mbox_chan_txdone(pcc_ss_data->pcc_channel, ret);
313 	else
314 		mbox_client_txdone(pcc_ss_data->pcc_channel, 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 acpi_pcct_hw_reduced *cppc_ss;
497 	u64 usecs_lat;
498 
499 	if (pcc_ss_idx >= 0) {
500 		pcc_data[pcc_ss_idx]->pcc_channel =
501 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
502 
503 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
504 			pr_err("Failed to find PCC channel for subspace %d\n",
505 			       pcc_ss_idx);
506 			return -ENODEV;
507 		}
508 
509 		/*
510 		 * The PCC mailbox controller driver should
511 		 * have parsed the PCCT (global table of all
512 		 * PCC channels) and stored pointers to the
513 		 * subspace communication region in con_priv.
514 		 */
515 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
516 
517 		if (!cppc_ss) {
518 			pr_err("No PCC subspace found for %d CPPC\n",
519 			       pcc_ss_idx);
520 			return -ENODEV;
521 		}
522 
523 		/*
524 		 * cppc_ss->latency is just a Nominal value. In reality
525 		 * the remote processor could be much slower to reply.
526 		 * So add an arbitrary amount of wait on top of Nominal.
527 		 */
528 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
529 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
530 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
531 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
532 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
533 
534 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
535 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
536 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
537 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
538 			       pcc_ss_idx);
539 			return -ENOMEM;
540 		}
541 
542 		/* Set flag so that we don't come here for each CPU. */
543 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
544 	}
545 
546 	return 0;
547 }
548 
549 /**
550  * cpc_ffh_supported() - check if FFH reading supported
551  *
552  * Check if the architecture has support for functional fixed hardware
553  * read/write capability.
554  *
555  * Return: true for supported, false for not supported
556  */
557 bool __weak cpc_ffh_supported(void)
558 {
559 	return false;
560 }
561 
562 /**
563  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
564  *
565  * Check and allocate the cppc_pcc_data memory.
566  * In some processor configurations it is possible that same subspace
567  * is shared between multiple CPUs. This is seen especially in CPUs
568  * with hardware multi-threading support.
569  *
570  * Return: 0 for success, errno for failure
571  */
572 static int pcc_data_alloc(int pcc_ss_id)
573 {
574 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
575 		return -EINVAL;
576 
577 	if (pcc_data[pcc_ss_id]) {
578 		pcc_data[pcc_ss_id]->refcount++;
579 	} else {
580 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
581 					      GFP_KERNEL);
582 		if (!pcc_data[pcc_ss_id])
583 			return -ENOMEM;
584 		pcc_data[pcc_ss_id]->refcount++;
585 	}
586 
587 	return 0;
588 }
589 
590 /* Check if CPPC revision + num_ent combination is supported */
591 static bool is_cppc_supported(int revision, int num_ent)
592 {
593 	int expected_num_ent;
594 
595 	switch (revision) {
596 	case CPPC_V2_REV:
597 		expected_num_ent = CPPC_V2_NUM_ENT;
598 		break;
599 	case CPPC_V3_REV:
600 		expected_num_ent = CPPC_V3_NUM_ENT;
601 		break;
602 	default:
603 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
604 			revision);
605 		return false;
606 	}
607 
608 	if (expected_num_ent != num_ent) {
609 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
610 			num_ent, expected_num_ent, revision);
611 		return false;
612 	}
613 
614 	return true;
615 }
616 
617 /*
618  * An example CPC table looks like the following.
619  *
620  *	Name(_CPC, Package()
621  *			{
622  *			17,
623  *			NumEntries
624  *			1,
625  *			// Revision
626  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
627  *			// Highest Performance
628  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
629  *			// Nominal Performance
630  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
631  *			// Lowest Nonlinear Performance
632  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
633  *			// Lowest Performance
634  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
635  *			// Guaranteed Performance Register
636  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
637  *			// Desired Performance Register
638  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
639  *			..
640  *			..
641  *			..
642  *
643  *		}
644  * Each Register() encodes how to access that specific register.
645  * e.g. a sample PCC entry has the following encoding:
646  *
647  *	Register (
648  *		PCC,
649  *		AddressSpaceKeyword
650  *		8,
651  *		//RegisterBitWidth
652  *		8,
653  *		//RegisterBitOffset
654  *		0x30,
655  *		//RegisterAddress
656  *		9
657  *		//AccessSize (subspace ID)
658  *		0
659  *		)
660  *	}
661  */
662 
663 #ifndef init_freq_invariance_cppc
664 static inline void init_freq_invariance_cppc(void) { }
665 #endif
666 
667 /**
668  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
669  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
670  *
671  *	Return: 0 for success or negative value for err.
672  */
673 int acpi_cppc_processor_probe(struct acpi_processor *pr)
674 {
675 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
676 	union acpi_object *out_obj, *cpc_obj;
677 	struct cpc_desc *cpc_ptr;
678 	struct cpc_reg *gas_t;
679 	struct device *cpu_dev;
680 	acpi_handle handle = pr->handle;
681 	unsigned int num_ent, i, cpc_rev;
682 	int pcc_subspace_id = -1;
683 	acpi_status status;
684 	int ret = -EFAULT;
685 
686 	/* Parse the ACPI _CPC table for this CPU. */
687 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
688 			ACPI_TYPE_PACKAGE);
689 	if (ACPI_FAILURE(status)) {
690 		ret = -ENODEV;
691 		goto out_buf_free;
692 	}
693 
694 	out_obj = (union acpi_object *) output.pointer;
695 
696 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
697 	if (!cpc_ptr) {
698 		ret = -ENOMEM;
699 		goto out_buf_free;
700 	}
701 
702 	/* First entry is NumEntries. */
703 	cpc_obj = &out_obj->package.elements[0];
704 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
705 		num_ent = cpc_obj->integer.value;
706 	} else {
707 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
708 				cpc_obj->type);
709 		goto out_free;
710 	}
711 	cpc_ptr->num_entries = num_ent;
712 
713 	/* Second entry should be revision. */
714 	cpc_obj = &out_obj->package.elements[1];
715 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
716 		cpc_rev = cpc_obj->integer.value;
717 	} else {
718 		pr_debug("Unexpected entry type(%d) for Revision\n",
719 				cpc_obj->type);
720 		goto out_free;
721 	}
722 	cpc_ptr->version = cpc_rev;
723 
724 	if (!is_cppc_supported(cpc_rev, num_ent))
725 		goto out_free;
726 
727 	/* Iterate through remaining entries in _CPC */
728 	for (i = 2; i < num_ent; i++) {
729 		cpc_obj = &out_obj->package.elements[i];
730 
731 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
732 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
733 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
734 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
735 			gas_t = (struct cpc_reg *)
736 				cpc_obj->buffer.pointer;
737 
738 			/*
739 			 * The PCC Subspace index is encoded inside
740 			 * the CPC table entries. The same PCC index
741 			 * will be used for all the PCC entries,
742 			 * so extract it only once.
743 			 */
744 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
745 				if (pcc_subspace_id < 0) {
746 					pcc_subspace_id = gas_t->access_width;
747 					if (pcc_data_alloc(pcc_subspace_id))
748 						goto out_free;
749 				} else if (pcc_subspace_id != gas_t->access_width) {
750 					pr_debug("Mismatched PCC ids.\n");
751 					goto out_free;
752 				}
753 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
754 				if (gas_t->address) {
755 					void __iomem *addr;
756 
757 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
758 					if (!addr)
759 						goto out_free;
760 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
761 				}
762 			} else {
763 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
764 					/* Support only PCC ,SYS MEM and FFH type regs */
765 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
766 					goto out_free;
767 				}
768 			}
769 
770 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
771 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
772 		} else {
773 			pr_debug("Err in entry:%d in CPC table of CPU:%d\n", i, pr->id);
774 			goto out_free;
775 		}
776 	}
777 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
778 
779 	/*
780 	 * Initialize the remaining cpc_regs as unsupported.
781 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
782 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
783 	 */
784 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
785 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
786 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
787 	}
788 
789 
790 	/* Store CPU Logical ID */
791 	cpc_ptr->cpu_id = pr->id;
792 
793 	/* Parse PSD data for this CPU */
794 	ret = acpi_get_psd(cpc_ptr, handle);
795 	if (ret)
796 		goto out_free;
797 
798 	/* Register PCC channel once for all PCC subspace ID. */
799 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
800 		ret = register_pcc_channel(pcc_subspace_id);
801 		if (ret)
802 			goto out_free;
803 
804 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
805 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
806 	}
807 
808 	/* Everything looks okay */
809 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
810 
811 	/* Add per logical CPU nodes for reading its feedback counters. */
812 	cpu_dev = get_cpu_device(pr->id);
813 	if (!cpu_dev) {
814 		ret = -EINVAL;
815 		goto out_free;
816 	}
817 
818 	/* Plug PSD data into this CPU's CPC descriptor. */
819 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
820 
821 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
822 			"acpi_cppc");
823 	if (ret) {
824 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
825 		kobject_put(&cpc_ptr->kobj);
826 		goto out_free;
827 	}
828 
829 	init_freq_invariance_cppc();
830 
831 	kfree(output.pointer);
832 	return 0;
833 
834 out_free:
835 	/* Free all the mapped sys mem areas for this CPU */
836 	for (i = 2; i < cpc_ptr->num_entries; i++) {
837 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
838 
839 		if (addr)
840 			iounmap(addr);
841 	}
842 	kfree(cpc_ptr);
843 
844 out_buf_free:
845 	kfree(output.pointer);
846 	return ret;
847 }
848 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
849 
850 /**
851  * acpi_cppc_processor_exit - Cleanup CPC structs.
852  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
853  *
854  * Return: Void
855  */
856 void acpi_cppc_processor_exit(struct acpi_processor *pr)
857 {
858 	struct cpc_desc *cpc_ptr;
859 	unsigned int i;
860 	void __iomem *addr;
861 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
862 
863 	if (pcc_ss_id >= 0 && pcc_data[pcc_ss_id]) {
864 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
865 			pcc_data[pcc_ss_id]->refcount--;
866 			if (!pcc_data[pcc_ss_id]->refcount) {
867 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
868 				kfree(pcc_data[pcc_ss_id]);
869 				pcc_data[pcc_ss_id] = NULL;
870 			}
871 		}
872 	}
873 
874 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
875 	if (!cpc_ptr)
876 		return;
877 
878 	/* Free all the mapped sys mem areas for this CPU */
879 	for (i = 2; i < cpc_ptr->num_entries; i++) {
880 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
881 		if (addr)
882 			iounmap(addr);
883 	}
884 
885 	kobject_put(&cpc_ptr->kobj);
886 	kfree(cpc_ptr);
887 }
888 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
889 
890 /**
891  * cpc_read_ffh() - Read FFH register
892  * @cpunum:	CPU number to read
893  * @reg:	cppc register information
894  * @val:	place holder for return value
895  *
896  * Read bit_width bits from a specified address and bit_offset
897  *
898  * Return: 0 for success and error code
899  */
900 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
901 {
902 	return -ENOTSUPP;
903 }
904 
905 /**
906  * cpc_write_ffh() - Write FFH register
907  * @cpunum:	CPU number to write
908  * @reg:	cppc register information
909  * @val:	value to write
910  *
911  * Write value of bit_width bits to a specified address and bit_offset
912  *
913  * Return: 0 for success and error code
914  */
915 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
916 {
917 	return -ENOTSUPP;
918 }
919 
920 /*
921  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
922  * as fast as possible. We have already mapped the PCC subspace during init, so
923  * we can directly write to it.
924  */
925 
926 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
927 {
928 	int ret_val = 0;
929 	void __iomem *vaddr = NULL;
930 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
931 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
932 
933 	if (reg_res->type == ACPI_TYPE_INTEGER) {
934 		*val = reg_res->cpc_entry.int_value;
935 		return ret_val;
936 	}
937 
938 	*val = 0;
939 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
940 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
941 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
942 		vaddr = reg_res->sys_mem_vaddr;
943 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
944 		return cpc_read_ffh(cpu, reg, val);
945 	else
946 		return acpi_os_read_memory((acpi_physical_address)reg->address,
947 				val, reg->bit_width);
948 
949 	switch (reg->bit_width) {
950 	case 8:
951 		*val = readb_relaxed(vaddr);
952 		break;
953 	case 16:
954 		*val = readw_relaxed(vaddr);
955 		break;
956 	case 32:
957 		*val = readl_relaxed(vaddr);
958 		break;
959 	case 64:
960 		*val = readq_relaxed(vaddr);
961 		break;
962 	default:
963 		pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
964 			 reg->bit_width, pcc_ss_id);
965 		ret_val = -EFAULT;
966 	}
967 
968 	return ret_val;
969 }
970 
971 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
972 {
973 	int ret_val = 0;
974 	void __iomem *vaddr = NULL;
975 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
976 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
977 
978 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
979 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
980 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
981 		vaddr = reg_res->sys_mem_vaddr;
982 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
983 		return cpc_write_ffh(cpu, reg, val);
984 	else
985 		return acpi_os_write_memory((acpi_physical_address)reg->address,
986 				val, reg->bit_width);
987 
988 	switch (reg->bit_width) {
989 	case 8:
990 		writeb_relaxed(val, vaddr);
991 		break;
992 	case 16:
993 		writew_relaxed(val, vaddr);
994 		break;
995 	case 32:
996 		writel_relaxed(val, vaddr);
997 		break;
998 	case 64:
999 		writeq_relaxed(val, vaddr);
1000 		break;
1001 	default:
1002 		pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1003 			 reg->bit_width, pcc_ss_id);
1004 		ret_val = -EFAULT;
1005 		break;
1006 	}
1007 
1008 	return ret_val;
1009 }
1010 
1011 static int cppc_get_perf(int cpunum, enum cppc_regs reg_idx, u64 *perf)
1012 {
1013 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1014 	struct cpc_register_resource *reg = &cpc_desc->cpc_regs[reg_idx];
1015 
1016 	if (CPC_IN_PCC(reg)) {
1017 		int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1018 		struct cppc_pcc_data *pcc_ss_data = NULL;
1019 		int ret = 0;
1020 
1021 		if (pcc_ss_id < 0)
1022 			return -EIO;
1023 
1024 		pcc_ss_data = pcc_data[pcc_ss_id];
1025 
1026 		down_write(&pcc_ss_data->pcc_lock);
1027 
1028 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1029 			cpc_read(cpunum, reg, perf);
1030 		else
1031 			ret = -EIO;
1032 
1033 		up_write(&pcc_ss_data->pcc_lock);
1034 
1035 		return ret;
1036 	}
1037 
1038 	cpc_read(cpunum, reg, perf);
1039 
1040 	return 0;
1041 }
1042 
1043 /**
1044  * cppc_get_desired_perf - Get the desired performance register value.
1045  * @cpunum: CPU from which to get desired performance.
1046  * @desired_perf: Return address.
1047  *
1048  * Return: 0 for success, -EIO otherwise.
1049  */
1050 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1051 {
1052 	return cppc_get_perf(cpunum, DESIRED_PERF, desired_perf);
1053 }
1054 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1055 
1056 /**
1057  * cppc_get_nominal_perf - Get the nominal performance register value.
1058  * @cpunum: CPU from which to get nominal performance.
1059  * @nominal_perf: Return address.
1060  *
1061  * Return: 0 for success, -EIO otherwise.
1062  */
1063 int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
1064 {
1065 	return cppc_get_perf(cpunum, NOMINAL_PERF, nominal_perf);
1066 }
1067 
1068 /**
1069  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1070  * @cpunum: CPU from which to get capabilities info.
1071  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1072  *
1073  * Return: 0 for success with perf_caps populated else -ERRNO.
1074  */
1075 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1076 {
1077 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1078 	struct cpc_register_resource *highest_reg, *lowest_reg,
1079 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1080 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1081 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1082 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1083 	struct cppc_pcc_data *pcc_ss_data = NULL;
1084 	int ret = 0, regs_in_pcc = 0;
1085 
1086 	if (!cpc_desc) {
1087 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1088 		return -ENODEV;
1089 	}
1090 
1091 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1092 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1093 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1094 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1095 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1096 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1097 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1098 
1099 	/* Are any of the regs PCC ?*/
1100 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1101 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1102 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1103 		if (pcc_ss_id < 0) {
1104 			pr_debug("Invalid pcc_ss_id\n");
1105 			return -ENODEV;
1106 		}
1107 		pcc_ss_data = pcc_data[pcc_ss_id];
1108 		regs_in_pcc = 1;
1109 		down_write(&pcc_ss_data->pcc_lock);
1110 		/* Ring doorbell once to update PCC subspace */
1111 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1112 			ret = -EIO;
1113 			goto out_err;
1114 		}
1115 	}
1116 
1117 	cpc_read(cpunum, highest_reg, &high);
1118 	perf_caps->highest_perf = high;
1119 
1120 	cpc_read(cpunum, lowest_reg, &low);
1121 	perf_caps->lowest_perf = low;
1122 
1123 	cpc_read(cpunum, nominal_reg, &nom);
1124 	perf_caps->nominal_perf = nom;
1125 
1126 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1127 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1128 		perf_caps->guaranteed_perf = 0;
1129 	} else {
1130 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1131 		perf_caps->guaranteed_perf = guaranteed;
1132 	}
1133 
1134 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1135 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1136 
1137 	if (!high || !low || !nom || !min_nonlinear)
1138 		ret = -EFAULT;
1139 
1140 	/* Read optional lowest and nominal frequencies if present */
1141 	if (CPC_SUPPORTED(low_freq_reg))
1142 		cpc_read(cpunum, low_freq_reg, &low_f);
1143 
1144 	if (CPC_SUPPORTED(nom_freq_reg))
1145 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1146 
1147 	perf_caps->lowest_freq = low_f;
1148 	perf_caps->nominal_freq = nom_f;
1149 
1150 
1151 out_err:
1152 	if (regs_in_pcc)
1153 		up_write(&pcc_ss_data->pcc_lock);
1154 	return ret;
1155 }
1156 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1157 
1158 /**
1159  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1160  * @cpunum: CPU from which to read counters.
1161  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1162  *
1163  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1164  */
1165 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1166 {
1167 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1168 	struct cpc_register_resource *delivered_reg, *reference_reg,
1169 		*ref_perf_reg, *ctr_wrap_reg;
1170 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1171 	struct cppc_pcc_data *pcc_ss_data = NULL;
1172 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1173 	int ret = 0, regs_in_pcc = 0;
1174 
1175 	if (!cpc_desc) {
1176 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1177 		return -ENODEV;
1178 	}
1179 
1180 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1181 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1182 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1183 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1184 
1185 	/*
1186 	 * If reference perf register is not supported then we should
1187 	 * use the nominal perf value
1188 	 */
1189 	if (!CPC_SUPPORTED(ref_perf_reg))
1190 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1191 
1192 	/* Are any of the regs PCC ?*/
1193 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1194 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1195 		if (pcc_ss_id < 0) {
1196 			pr_debug("Invalid pcc_ss_id\n");
1197 			return -ENODEV;
1198 		}
1199 		pcc_ss_data = pcc_data[pcc_ss_id];
1200 		down_write(&pcc_ss_data->pcc_lock);
1201 		regs_in_pcc = 1;
1202 		/* Ring doorbell once to update PCC subspace */
1203 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1204 			ret = -EIO;
1205 			goto out_err;
1206 		}
1207 	}
1208 
1209 	cpc_read(cpunum, delivered_reg, &delivered);
1210 	cpc_read(cpunum, reference_reg, &reference);
1211 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1212 
1213 	/*
1214 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1215 	 * performance counters are assumed to never wrap during the lifetime of
1216 	 * platform
1217 	 */
1218 	ctr_wrap_time = (u64)(~((u64)0));
1219 	if (CPC_SUPPORTED(ctr_wrap_reg))
1220 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1221 
1222 	if (!delivered || !reference ||	!ref_perf) {
1223 		ret = -EFAULT;
1224 		goto out_err;
1225 	}
1226 
1227 	perf_fb_ctrs->delivered = delivered;
1228 	perf_fb_ctrs->reference = reference;
1229 	perf_fb_ctrs->reference_perf = ref_perf;
1230 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1231 out_err:
1232 	if (regs_in_pcc)
1233 		up_write(&pcc_ss_data->pcc_lock);
1234 	return ret;
1235 }
1236 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1237 
1238 /**
1239  * cppc_set_perf - Set a CPU's performance controls.
1240  * @cpu: CPU for which to set performance controls.
1241  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1242  *
1243  * Return: 0 for success, -ERRNO otherwise.
1244  */
1245 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1246 {
1247 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1248 	struct cpc_register_resource *desired_reg;
1249 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1250 	struct cppc_pcc_data *pcc_ss_data = NULL;
1251 	int ret = 0;
1252 
1253 	if (!cpc_desc) {
1254 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1255 		return -ENODEV;
1256 	}
1257 
1258 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1259 
1260 	/*
1261 	 * This is Phase-I where we want to write to CPC registers
1262 	 * -> We want all CPUs to be able to execute this phase in parallel
1263 	 *
1264 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1265 	 * achieve that goal here
1266 	 */
1267 	if (CPC_IN_PCC(desired_reg)) {
1268 		if (pcc_ss_id < 0) {
1269 			pr_debug("Invalid pcc_ss_id\n");
1270 			return -ENODEV;
1271 		}
1272 		pcc_ss_data = pcc_data[pcc_ss_id];
1273 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1274 		if (pcc_ss_data->platform_owns_pcc) {
1275 			ret = check_pcc_chan(pcc_ss_id, false);
1276 			if (ret) {
1277 				up_read(&pcc_ss_data->pcc_lock);
1278 				return ret;
1279 			}
1280 		}
1281 		/*
1282 		 * Update the pending_write to make sure a PCC CMD_READ will not
1283 		 * arrive and steal the channel during the switch to write lock
1284 		 */
1285 		pcc_ss_data->pending_pcc_write_cmd = true;
1286 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1287 		cpc_desc->write_cmd_status = 0;
1288 	}
1289 
1290 	/*
1291 	 * Skip writing MIN/MAX until Linux knows how to come up with
1292 	 * useful values.
1293 	 */
1294 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1295 
1296 	if (CPC_IN_PCC(desired_reg))
1297 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1298 	/*
1299 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1300 	 *
1301 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1302 	 * requests that happened in short overlapping interval. The last CPU to
1303 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1304 	 *
1305 	 * We have the following requirements for Phase-II:
1306 	 *     1. We want to execute Phase-II only when there are no CPUs
1307 	 * currently executing in Phase-I
1308 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1309 	 * entering Phase-I.
1310 	 *     3. We want only one CPU among all those who went through Phase-I
1311 	 * to run phase-II
1312 	 *
1313 	 * If write_trylock fails to get the lock and doesn't transfer the
1314 	 * PCC ownership to the platform, then one of the following will be TRUE
1315 	 *     1. There is at-least one CPU in Phase-I which will later execute
1316 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1317 	 * executing the Phase-II.
1318 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1319 	 * write_trylock and has already acquired the write_lock. We know for a
1320 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1321 	 * before this CPU's Phase-I as we held the read_lock.
1322 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1323 	 * down_write, in which case, send_pcc_cmd will check for pending
1324 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1325 	 * So this CPU can be certain that its request will be delivered
1326 	 *    So in all cases, this CPU knows that its request will be delivered
1327 	 * by another CPU and can return
1328 	 *
1329 	 * After getting the down_write we still need to check for
1330 	 * pending_pcc_write_cmd to take care of the following scenario
1331 	 *    The thread running this code could be scheduled out between
1332 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1333 	 * could have delivered the request to Platform by triggering the
1334 	 * doorbell and transferred the ownership of PCC to platform. So this
1335 	 * avoids triggering an unnecessary doorbell and more importantly before
1336 	 * triggering the doorbell it makes sure that the PCC channel ownership
1337 	 * is still with OSPM.
1338 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1339 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1340 	 * before the pcc CMD_WRITE is completed. send_pcc_cmd checks for this
1341 	 * case during a CMD_READ and if there are pending writes it delivers
1342 	 * the write command before servicing the read command
1343 	 */
1344 	if (CPC_IN_PCC(desired_reg)) {
1345 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1346 			/* Update only if there are pending write commands */
1347 			if (pcc_ss_data->pending_pcc_write_cmd)
1348 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1349 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1350 		} else
1351 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1352 			wait_event(pcc_ss_data->pcc_write_wait_q,
1353 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1354 
1355 		/* send_pcc_cmd updates the status in case of failure */
1356 		ret = cpc_desc->write_cmd_status;
1357 	}
1358 	return ret;
1359 }
1360 EXPORT_SYMBOL_GPL(cppc_set_perf);
1361 
1362 /**
1363  * cppc_get_transition_latency - returns frequency transition latency in ns
1364  *
1365  * ACPI CPPC does not explicitly specify how a platform can specify the
1366  * transition latency for performance change requests. The closest we have
1367  * is the timing information from the PCCT tables which provides the info
1368  * on the number and frequency of PCC commands the platform can handle.
1369  */
1370 unsigned int cppc_get_transition_latency(int cpu_num)
1371 {
1372 	/*
1373 	 * Expected transition latency is based on the PCCT timing values
1374 	 * Below are definition from ACPI spec:
1375 	 * pcc_nominal- Expected latency to process a command, in microseconds
1376 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1377 	 *              channel can support, reported in commands per minute. 0
1378 	 *              indicates no limitation.
1379 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1380 	 *              completion of a command before issuing the next command,
1381 	 *              in microseconds.
1382 	 */
1383 	unsigned int latency_ns = 0;
1384 	struct cpc_desc *cpc_desc;
1385 	struct cpc_register_resource *desired_reg;
1386 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1387 	struct cppc_pcc_data *pcc_ss_data;
1388 
1389 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1390 	if (!cpc_desc)
1391 		return CPUFREQ_ETERNAL;
1392 
1393 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1394 	if (!CPC_IN_PCC(desired_reg))
1395 		return CPUFREQ_ETERNAL;
1396 
1397 	if (pcc_ss_id < 0)
1398 		return CPUFREQ_ETERNAL;
1399 
1400 	pcc_ss_data = pcc_data[pcc_ss_id];
1401 	if (pcc_ss_data->pcc_mpar)
1402 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1403 
1404 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1405 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1406 
1407 	return latency_ns;
1408 }
1409 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1410