xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 31e67366)
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/cpufreq.h>
37 #include <linux/delay.h>
38 #include <linux/iopoll.h>
39 #include <linux/ktime.h>
40 #include <linux/rwsem.h>
41 #include <linux/wait.h>
42 #include <linux/topology.h>
43 
44 #include <acpi/cppc_acpi.h>
45 
46 struct cppc_pcc_data {
47 	struct mbox_chan *pcc_channel;
48 	void __iomem *pcc_comm_addr;
49 	bool pcc_channel_acquired;
50 	unsigned int deadline_us;
51 	unsigned int pcc_mpar, pcc_mrtt, pcc_nominal;
52 
53 	bool pending_pcc_write_cmd;	/* Any pending/batched PCC write cmds? */
54 	bool platform_owns_pcc;		/* Ownership of PCC subspace */
55 	unsigned int pcc_write_cnt;	/* Running count of PCC write commands */
56 
57 	/*
58 	 * Lock to provide controlled access to the PCC channel.
59 	 *
60 	 * For performance critical usecases(currently cppc_set_perf)
61 	 *	We need to take read_lock and check if channel belongs to OSPM
62 	 * before reading or writing to PCC subspace
63 	 *	We need to take write_lock before transferring the channel
64 	 * ownership to the platform via a Doorbell
65 	 *	This allows us to batch a number of CPPC requests if they happen
66 	 * to originate in about the same time
67 	 *
68 	 * For non-performance critical usecases(init)
69 	 *	Take write_lock for all purposes which gives exclusive access
70 	 */
71 	struct rw_semaphore pcc_lock;
72 
73 	/* Wait queue for CPUs whose requests were batched */
74 	wait_queue_head_t pcc_write_wait_q;
75 	ktime_t last_cmd_cmpl_time;
76 	ktime_t last_mpar_reset;
77 	int mpar_count;
78 	int refcount;
79 };
80 
81 /* Array to represent the PCC channel per subspace ID */
82 static struct cppc_pcc_data *pcc_data[MAX_PCC_SUBSPACES];
83 /* The cpu_pcc_subspace_idx contains per CPU subspace ID */
84 static DEFINE_PER_CPU(int, cpu_pcc_subspace_idx);
85 
86 /*
87  * The cpc_desc structure contains the ACPI register details
88  * as described in the per CPU _CPC tables. The details
89  * include the type of register (e.g. PCC, System IO, FFH etc.)
90  * and destination addresses which lets us READ/WRITE CPU performance
91  * information using the appropriate I/O methods.
92  */
93 static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
94 
95 /* pcc mapped address + header size + offset within PCC subspace */
96 #define GET_PCC_VADDR(offs, pcc_ss_id) (pcc_data[pcc_ss_id]->pcc_comm_addr + \
97 						0x8 + (offs))
98 
99 /* Check if a CPC register is in PCC */
100 #define CPC_IN_PCC(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&		\
101 				(cpc)->cpc_entry.reg.space_id ==	\
102 				ACPI_ADR_SPACE_PLATFORM_COMM)
103 
104 /* Evalutes to True if reg is a NULL register descriptor */
105 #define IS_NULL_REG(reg) ((reg)->space_id ==  ACPI_ADR_SPACE_SYSTEM_MEMORY && \
106 				(reg)->address == 0 &&			\
107 				(reg)->bit_width == 0 &&		\
108 				(reg)->bit_offset == 0 &&		\
109 				(reg)->access_width == 0)
110 
111 /* Evalutes to True if an optional cpc field is supported */
112 #define CPC_SUPPORTED(cpc) ((cpc)->type == ACPI_TYPE_INTEGER ?		\
113 				!!(cpc)->cpc_entry.int_value :		\
114 				!IS_NULL_REG(&(cpc)->cpc_entry.reg))
115 /*
116  * Arbitrary Retries in case the remote processor is slow to respond
117  * to PCC commands. Keeping it high enough to cover emulators where
118  * the processors run painfully slow.
119  */
120 #define NUM_RETRIES 500ULL
121 
122 struct cppc_attr {
123 	struct attribute attr;
124 	ssize_t (*show)(struct kobject *kobj,
125 			struct attribute *attr, char *buf);
126 	ssize_t (*store)(struct kobject *kobj,
127 			struct attribute *attr, const char *c, ssize_t count);
128 };
129 
130 #define define_one_cppc_ro(_name)		\
131 static struct cppc_attr _name =			\
132 __ATTR(_name, 0444, show_##_name, NULL)
133 
134 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
135 
136 #define show_cppc_data(access_fn, struct_name, member_name)		\
137 	static ssize_t show_##member_name(struct kobject *kobj,		\
138 					struct attribute *attr,	char *buf) \
139 	{								\
140 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
141 		struct struct_name st_name = {0};			\
142 		int ret;						\
143 									\
144 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
145 		if (ret)						\
146 			return ret;					\
147 									\
148 		return scnprintf(buf, PAGE_SIZE, "%llu\n",		\
149 				(u64)st_name.member_name);		\
150 	}								\
151 	define_one_cppc_ro(member_name)
152 
153 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
154 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
155 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
156 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
157 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_freq);
158 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_freq);
159 
160 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, reference_perf);
161 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
162 
163 static ssize_t show_feedback_ctrs(struct kobject *kobj,
164 		struct attribute *attr, char *buf)
165 {
166 	struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);
167 	struct cppc_perf_fb_ctrs fb_ctrs = {0};
168 	int ret;
169 
170 	ret = cppc_get_perf_ctrs(cpc_ptr->cpu_id, &fb_ctrs);
171 	if (ret)
172 		return ret;
173 
174 	return scnprintf(buf, PAGE_SIZE, "ref:%llu del:%llu\n",
175 			fb_ctrs.reference, fb_ctrs.delivered);
176 }
177 define_one_cppc_ro(feedback_ctrs);
178 
179 static struct attribute *cppc_attrs[] = {
180 	&feedback_ctrs.attr,
181 	&reference_perf.attr,
182 	&wraparound_time.attr,
183 	&highest_perf.attr,
184 	&lowest_perf.attr,
185 	&lowest_nonlinear_perf.attr,
186 	&nominal_perf.attr,
187 	&nominal_freq.attr,
188 	&lowest_freq.attr,
189 	NULL
190 };
191 
192 static struct kobj_type cppc_ktype = {
193 	.sysfs_ops = &kobj_sysfs_ops,
194 	.default_attrs = cppc_attrs,
195 };
196 
197 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
198 {
199 	int ret, status;
200 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
201 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
202 		pcc_ss_data->pcc_comm_addr;
203 
204 	if (!pcc_ss_data->platform_owns_pcc)
205 		return 0;
206 
207 	/*
208 	 * Poll PCC status register every 3us(delay_us) for maximum of
209 	 * deadline_us(timeout_us) until PCC command complete bit is set(cond)
210 	 */
211 	ret = readw_relaxed_poll_timeout(&generic_comm_base->status, status,
212 					status & PCC_CMD_COMPLETE_MASK, 3,
213 					pcc_ss_data->deadline_us);
214 
215 	if (likely(!ret)) {
216 		pcc_ss_data->platform_owns_pcc = false;
217 		if (chk_err_bit && (status & PCC_ERROR_MASK))
218 			ret = -EIO;
219 	}
220 
221 	if (unlikely(ret))
222 		pr_err("PCC check channel failed for ss: %d. ret=%d\n",
223 		       pcc_ss_id, ret);
224 
225 	return ret;
226 }
227 
228 /*
229  * This function transfers the ownership of the PCC to the platform
230  * So it must be called while holding write_lock(pcc_lock)
231  */
232 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
233 {
234 	int ret = -EIO, i;
235 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
236 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
237 		pcc_ss_data->pcc_comm_addr;
238 	unsigned int time_delta;
239 
240 	/*
241 	 * For CMD_WRITE we know for a fact the caller should have checked
242 	 * the channel before writing to PCC space
243 	 */
244 	if (cmd == CMD_READ) {
245 		/*
246 		 * If there are pending cpc_writes, then we stole the channel
247 		 * before write completion, so first send a WRITE command to
248 		 * platform
249 		 */
250 		if (pcc_ss_data->pending_pcc_write_cmd)
251 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
252 
253 		ret = check_pcc_chan(pcc_ss_id, false);
254 		if (ret)
255 			goto end;
256 	} else /* CMD_WRITE */
257 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
258 
259 	/*
260 	 * Handle the Minimum Request Turnaround Time(MRTT)
261 	 * "The minimum amount of time that OSPM must wait after the completion
262 	 * of a command before issuing the next command, in microseconds"
263 	 */
264 	if (pcc_ss_data->pcc_mrtt) {
265 		time_delta = ktime_us_delta(ktime_get(),
266 					    pcc_ss_data->last_cmd_cmpl_time);
267 		if (pcc_ss_data->pcc_mrtt > time_delta)
268 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
269 	}
270 
271 	/*
272 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
273 	 * "The maximum number of periodic requests that the subspace channel can
274 	 * support, reported in commands per minute. 0 indicates no limitation."
275 	 *
276 	 * This parameter should be ideally zero or large enough so that it can
277 	 * handle maximum number of requests that all the cores in the system can
278 	 * collectively generate. If it is not, we will follow the spec and just
279 	 * not send the request to the platform after hitting the MPAR limit in
280 	 * any 60s window
281 	 */
282 	if (pcc_ss_data->pcc_mpar) {
283 		if (pcc_ss_data->mpar_count == 0) {
284 			time_delta = ktime_ms_delta(ktime_get(),
285 						    pcc_ss_data->last_mpar_reset);
286 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
287 				pr_debug("PCC cmd for subspace %d not sent due to MPAR limit",
288 					 pcc_ss_id);
289 				ret = -EIO;
290 				goto end;
291 			}
292 			pcc_ss_data->last_mpar_reset = ktime_get();
293 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
294 		}
295 		pcc_ss_data->mpar_count--;
296 	}
297 
298 	/* Write to the shared comm region. */
299 	writew_relaxed(cmd, &generic_comm_base->command);
300 
301 	/* Flip CMD COMPLETE bit */
302 	writew_relaxed(0, &generic_comm_base->status);
303 
304 	pcc_ss_data->platform_owns_pcc = true;
305 
306 	/* Ring doorbell */
307 	ret = mbox_send_message(pcc_ss_data->pcc_channel, &cmd);
308 	if (ret < 0) {
309 		pr_err("Err sending PCC mbox message. ss: %d cmd:%d, ret:%d\n",
310 		       pcc_ss_id, cmd, ret);
311 		goto end;
312 	}
313 
314 	/* wait for completion and check for PCC errro bit */
315 	ret = check_pcc_chan(pcc_ss_id, true);
316 
317 	if (pcc_ss_data->pcc_mrtt)
318 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
319 
320 	if (pcc_ss_data->pcc_channel->mbox->txdone_irq)
321 		mbox_chan_txdone(pcc_ss_data->pcc_channel, ret);
322 	else
323 		mbox_client_txdone(pcc_ss_data->pcc_channel, ret);
324 
325 end:
326 	if (cmd == CMD_WRITE) {
327 		if (unlikely(ret)) {
328 			for_each_possible_cpu(i) {
329 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
330 				if (!desc)
331 					continue;
332 
333 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
334 					desc->write_cmd_status = ret;
335 			}
336 		}
337 		pcc_ss_data->pcc_write_cnt++;
338 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
339 	}
340 
341 	return ret;
342 }
343 
344 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
345 {
346 	if (ret < 0)
347 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
348 				*(u16 *)msg, ret);
349 	else
350 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
351 				*(u16 *)msg, ret);
352 }
353 
354 static struct mbox_client cppc_mbox_cl = {
355 	.tx_done = cppc_chan_tx_done,
356 	.knows_txdone = true,
357 };
358 
359 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
360 {
361 	int result = -EFAULT;
362 	acpi_status status = AE_OK;
363 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
364 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
365 	struct acpi_buffer state = {0, NULL};
366 	union acpi_object  *psd = NULL;
367 	struct acpi_psd_package *pdomain;
368 
369 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL,
370 					    &buffer, ACPI_TYPE_PACKAGE);
371 	if (status == AE_NOT_FOUND)	/* _PSD is optional */
372 		return 0;
373 	if (ACPI_FAILURE(status))
374 		return -ENODEV;
375 
376 	psd = buffer.pointer;
377 	if (!psd || psd->package.count != 1) {
378 		pr_debug("Invalid _PSD data\n");
379 		goto end;
380 	}
381 
382 	pdomain = &(cpc_ptr->domain_info);
383 
384 	state.length = sizeof(struct acpi_psd_package);
385 	state.pointer = pdomain;
386 
387 	status = acpi_extract_package(&(psd->package.elements[0]),
388 		&format, &state);
389 	if (ACPI_FAILURE(status)) {
390 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
391 		goto end;
392 	}
393 
394 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
395 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
396 		goto end;
397 	}
398 
399 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
400 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
401 		goto end;
402 	}
403 
404 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
405 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
406 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
407 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
408 		goto end;
409 	}
410 
411 	result = 0;
412 end:
413 	kfree(buffer.pointer);
414 	return result;
415 }
416 
417 bool acpi_cpc_valid(void)
418 {
419 	struct cpc_desc *cpc_ptr;
420 	int cpu;
421 
422 	for_each_possible_cpu(cpu) {
423 		cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
424 		if (!cpc_ptr)
425 			return false;
426 	}
427 
428 	return true;
429 }
430 EXPORT_SYMBOL_GPL(acpi_cpc_valid);
431 
432 /**
433  * acpi_get_psd_map - Map the CPUs in the freq domain of a given cpu
434  * @cpu: Find all CPUs that share a domain with cpu.
435  * @cpu_data: Pointer to CPU specific CPPC data including PSD info.
436  *
437  *	Return: 0 for success or negative value for err.
438  */
439 int acpi_get_psd_map(unsigned int cpu, struct cppc_cpudata *cpu_data)
440 {
441 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
442 	struct acpi_psd_package *match_pdomain;
443 	struct acpi_psd_package *pdomain;
444 	int count_target, i;
445 
446 	/*
447 	 * Now that we have _PSD data from all CPUs, let's setup P-state
448 	 * domain info.
449 	 */
450 	cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
451 	if (!cpc_ptr)
452 		return -EFAULT;
453 
454 	pdomain = &(cpc_ptr->domain_info);
455 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
456 	if (pdomain->num_processors <= 1)
457 		return 0;
458 
459 	/* Validate the Domain info */
460 	count_target = pdomain->num_processors;
461 	if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
462 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ALL;
463 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
464 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_HW;
465 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
466 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ANY;
467 
468 	for_each_possible_cpu(i) {
469 		if (i == cpu)
470 			continue;
471 
472 		match_cpc_ptr = per_cpu(cpc_desc_ptr, i);
473 		if (!match_cpc_ptr)
474 			goto err_fault;
475 
476 		match_pdomain = &(match_cpc_ptr->domain_info);
477 		if (match_pdomain->domain != pdomain->domain)
478 			continue;
479 
480 		/* Here i and cpu are in the same domain */
481 		if (match_pdomain->num_processors != count_target)
482 			goto err_fault;
483 
484 		if (pdomain->coord_type != match_pdomain->coord_type)
485 			goto err_fault;
486 
487 		cpumask_set_cpu(i, cpu_data->shared_cpu_map);
488 	}
489 
490 	return 0;
491 
492 err_fault:
493 	/* Assume no coordination on any error parsing domain info */
494 	cpumask_clear(cpu_data->shared_cpu_map);
495 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
496 	cpu_data->shared_type = CPUFREQ_SHARED_TYPE_NONE;
497 
498 	return -EFAULT;
499 }
500 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
501 
502 static int register_pcc_channel(int pcc_ss_idx)
503 {
504 	struct acpi_pcct_hw_reduced *cppc_ss;
505 	u64 usecs_lat;
506 
507 	if (pcc_ss_idx >= 0) {
508 		pcc_data[pcc_ss_idx]->pcc_channel =
509 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
510 
511 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
512 			pr_err("Failed to find PCC channel for subspace %d\n",
513 			       pcc_ss_idx);
514 			return -ENODEV;
515 		}
516 
517 		/*
518 		 * The PCC mailbox controller driver should
519 		 * have parsed the PCCT (global table of all
520 		 * PCC channels) and stored pointers to the
521 		 * subspace communication region in con_priv.
522 		 */
523 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
524 
525 		if (!cppc_ss) {
526 			pr_err("No PCC subspace found for %d CPPC\n",
527 			       pcc_ss_idx);
528 			return -ENODEV;
529 		}
530 
531 		/*
532 		 * cppc_ss->latency is just a Nominal value. In reality
533 		 * the remote processor could be much slower to reply.
534 		 * So add an arbitrary amount of wait on top of Nominal.
535 		 */
536 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
537 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
538 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
539 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
540 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
541 
542 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
543 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
544 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
545 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
546 			       pcc_ss_idx);
547 			return -ENOMEM;
548 		}
549 
550 		/* Set flag so that we don't come here for each CPU. */
551 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
552 	}
553 
554 	return 0;
555 }
556 
557 /**
558  * cpc_ffh_supported() - check if FFH reading supported
559  *
560  * Check if the architecture has support for functional fixed hardware
561  * read/write capability.
562  *
563  * Return: true for supported, false for not supported
564  */
565 bool __weak cpc_ffh_supported(void)
566 {
567 	return false;
568 }
569 
570 /**
571  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
572  *
573  * Check and allocate the cppc_pcc_data memory.
574  * In some processor configurations it is possible that same subspace
575  * is shared between multiple CPUs. This is seen especially in CPUs
576  * with hardware multi-threading support.
577  *
578  * Return: 0 for success, errno for failure
579  */
580 static int pcc_data_alloc(int pcc_ss_id)
581 {
582 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
583 		return -EINVAL;
584 
585 	if (pcc_data[pcc_ss_id]) {
586 		pcc_data[pcc_ss_id]->refcount++;
587 	} else {
588 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
589 					      GFP_KERNEL);
590 		if (!pcc_data[pcc_ss_id])
591 			return -ENOMEM;
592 		pcc_data[pcc_ss_id]->refcount++;
593 	}
594 
595 	return 0;
596 }
597 
598 /* Check if CPPC revision + num_ent combination is supported */
599 static bool is_cppc_supported(int revision, int num_ent)
600 {
601 	int expected_num_ent;
602 
603 	switch (revision) {
604 	case CPPC_V2_REV:
605 		expected_num_ent = CPPC_V2_NUM_ENT;
606 		break;
607 	case CPPC_V3_REV:
608 		expected_num_ent = CPPC_V3_NUM_ENT;
609 		break;
610 	default:
611 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
612 			revision);
613 		return false;
614 	}
615 
616 	if (expected_num_ent != num_ent) {
617 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
618 			num_ent, expected_num_ent, revision);
619 		return false;
620 	}
621 
622 	return true;
623 }
624 
625 /*
626  * An example CPC table looks like the following.
627  *
628  *	Name(_CPC, Package()
629  *			{
630  *			17,
631  *			NumEntries
632  *			1,
633  *			// Revision
634  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
635  *			// Highest Performance
636  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
637  *			// Nominal Performance
638  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
639  *			// Lowest Nonlinear Performance
640  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
641  *			// Lowest Performance
642  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
643  *			// Guaranteed Performance Register
644  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
645  *			// Desired Performance Register
646  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
647  *			..
648  *			..
649  *			..
650  *
651  *		}
652  * Each Register() encodes how to access that specific register.
653  * e.g. a sample PCC entry has the following encoding:
654  *
655  *	Register (
656  *		PCC,
657  *		AddressSpaceKeyword
658  *		8,
659  *		//RegisterBitWidth
660  *		8,
661  *		//RegisterBitOffset
662  *		0x30,
663  *		//RegisterAddress
664  *		9
665  *		//AccessSize (subspace ID)
666  *		0
667  *		)
668  *	}
669  */
670 
671 #ifndef init_freq_invariance_cppc
672 static inline void init_freq_invariance_cppc(void) { }
673 #endif
674 
675 /**
676  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
677  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
678  *
679  *	Return: 0 for success or negative value for err.
680  */
681 int acpi_cppc_processor_probe(struct acpi_processor *pr)
682 {
683 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
684 	union acpi_object *out_obj, *cpc_obj;
685 	struct cpc_desc *cpc_ptr;
686 	struct cpc_reg *gas_t;
687 	struct device *cpu_dev;
688 	acpi_handle handle = pr->handle;
689 	unsigned int num_ent, i, cpc_rev;
690 	int pcc_subspace_id = -1;
691 	acpi_status status;
692 	int ret = -EFAULT;
693 
694 	/* Parse the ACPI _CPC table for this CPU. */
695 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
696 			ACPI_TYPE_PACKAGE);
697 	if (ACPI_FAILURE(status)) {
698 		ret = -ENODEV;
699 		goto out_buf_free;
700 	}
701 
702 	out_obj = (union acpi_object *) output.pointer;
703 
704 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
705 	if (!cpc_ptr) {
706 		ret = -ENOMEM;
707 		goto out_buf_free;
708 	}
709 
710 	/* First entry is NumEntries. */
711 	cpc_obj = &out_obj->package.elements[0];
712 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
713 		num_ent = cpc_obj->integer.value;
714 	} else {
715 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
716 				cpc_obj->type);
717 		goto out_free;
718 	}
719 	cpc_ptr->num_entries = num_ent;
720 
721 	/* Second entry should be revision. */
722 	cpc_obj = &out_obj->package.elements[1];
723 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
724 		cpc_rev = cpc_obj->integer.value;
725 	} else {
726 		pr_debug("Unexpected entry type(%d) for Revision\n",
727 				cpc_obj->type);
728 		goto out_free;
729 	}
730 	cpc_ptr->version = cpc_rev;
731 
732 	if (!is_cppc_supported(cpc_rev, num_ent))
733 		goto out_free;
734 
735 	/* Iterate through remaining entries in _CPC */
736 	for (i = 2; i < num_ent; i++) {
737 		cpc_obj = &out_obj->package.elements[i];
738 
739 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
740 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
741 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
742 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
743 			gas_t = (struct cpc_reg *)
744 				cpc_obj->buffer.pointer;
745 
746 			/*
747 			 * The PCC Subspace index is encoded inside
748 			 * the CPC table entries. The same PCC index
749 			 * will be used for all the PCC entries,
750 			 * so extract it only once.
751 			 */
752 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
753 				if (pcc_subspace_id < 0) {
754 					pcc_subspace_id = gas_t->access_width;
755 					if (pcc_data_alloc(pcc_subspace_id))
756 						goto out_free;
757 				} else if (pcc_subspace_id != gas_t->access_width) {
758 					pr_debug("Mismatched PCC ids.\n");
759 					goto out_free;
760 				}
761 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
762 				if (gas_t->address) {
763 					void __iomem *addr;
764 
765 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
766 					if (!addr)
767 						goto out_free;
768 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
769 				}
770 			} else {
771 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
772 					/* Support only PCC ,SYS MEM and FFH type regs */
773 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
774 					goto out_free;
775 				}
776 			}
777 
778 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
779 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
780 		} else {
781 			pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
782 			goto out_free;
783 		}
784 	}
785 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
786 
787 	/*
788 	 * Initialize the remaining cpc_regs as unsupported.
789 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
790 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
791 	 */
792 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
793 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
794 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
795 	}
796 
797 
798 	/* Store CPU Logical ID */
799 	cpc_ptr->cpu_id = pr->id;
800 
801 	/* Parse PSD data for this CPU */
802 	ret = acpi_get_psd(cpc_ptr, handle);
803 	if (ret)
804 		goto out_free;
805 
806 	/* Register PCC channel once for all PCC subspace ID. */
807 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
808 		ret = register_pcc_channel(pcc_subspace_id);
809 		if (ret)
810 			goto out_free;
811 
812 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
813 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
814 	}
815 
816 	/* Everything looks okay */
817 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
818 
819 	/* Add per logical CPU nodes for reading its feedback counters. */
820 	cpu_dev = get_cpu_device(pr->id);
821 	if (!cpu_dev) {
822 		ret = -EINVAL;
823 		goto out_free;
824 	}
825 
826 	/* Plug PSD data into this CPU's CPC descriptor. */
827 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
828 
829 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
830 			"acpi_cppc");
831 	if (ret) {
832 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
833 		kobject_put(&cpc_ptr->kobj);
834 		goto out_free;
835 	}
836 
837 	init_freq_invariance_cppc();
838 
839 	kfree(output.pointer);
840 	return 0;
841 
842 out_free:
843 	/* Free all the mapped sys mem areas for this CPU */
844 	for (i = 2; i < cpc_ptr->num_entries; i++) {
845 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
846 
847 		if (addr)
848 			iounmap(addr);
849 	}
850 	kfree(cpc_ptr);
851 
852 out_buf_free:
853 	kfree(output.pointer);
854 	return ret;
855 }
856 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
857 
858 /**
859  * acpi_cppc_processor_exit - Cleanup CPC structs.
860  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
861  *
862  * Return: Void
863  */
864 void acpi_cppc_processor_exit(struct acpi_processor *pr)
865 {
866 	struct cpc_desc *cpc_ptr;
867 	unsigned int i;
868 	void __iomem *addr;
869 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
870 
871 	if (pcc_ss_id >=0 && pcc_data[pcc_ss_id]) {
872 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
873 			pcc_data[pcc_ss_id]->refcount--;
874 			if (!pcc_data[pcc_ss_id]->refcount) {
875 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
876 				kfree(pcc_data[pcc_ss_id]);
877 				pcc_data[pcc_ss_id] = NULL;
878 			}
879 		}
880 	}
881 
882 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
883 	if (!cpc_ptr)
884 		return;
885 
886 	/* Free all the mapped sys mem areas for this CPU */
887 	for (i = 2; i < cpc_ptr->num_entries; i++) {
888 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
889 		if (addr)
890 			iounmap(addr);
891 	}
892 
893 	kobject_put(&cpc_ptr->kobj);
894 	kfree(cpc_ptr);
895 }
896 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
897 
898 /**
899  * cpc_read_ffh() - Read FFH register
900  * @cpunum:	CPU number to read
901  * @reg:	cppc register information
902  * @val:	place holder for return value
903  *
904  * Read bit_width bits from a specified address and bit_offset
905  *
906  * Return: 0 for success and error code
907  */
908 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
909 {
910 	return -ENOTSUPP;
911 }
912 
913 /**
914  * cpc_write_ffh() - Write FFH register
915  * @cpunum:	CPU number to write
916  * @reg:	cppc register information
917  * @val:	value to write
918  *
919  * Write value of bit_width bits to a specified address and bit_offset
920  *
921  * Return: 0 for success and error code
922  */
923 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
924 {
925 	return -ENOTSUPP;
926 }
927 
928 /*
929  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
930  * as fast as possible. We have already mapped the PCC subspace during init, so
931  * we can directly write to it.
932  */
933 
934 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
935 {
936 	int ret_val = 0;
937 	void __iomem *vaddr = NULL;
938 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
939 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
940 
941 	if (reg_res->type == ACPI_TYPE_INTEGER) {
942 		*val = reg_res->cpc_entry.int_value;
943 		return ret_val;
944 	}
945 
946 	*val = 0;
947 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
948 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
949 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
950 		vaddr = reg_res->sys_mem_vaddr;
951 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
952 		return cpc_read_ffh(cpu, reg, val);
953 	else
954 		return acpi_os_read_memory((acpi_physical_address)reg->address,
955 				val, reg->bit_width);
956 
957 	switch (reg->bit_width) {
958 		case 8:
959 			*val = readb_relaxed(vaddr);
960 			break;
961 		case 16:
962 			*val = readw_relaxed(vaddr);
963 			break;
964 		case 32:
965 			*val = readl_relaxed(vaddr);
966 			break;
967 		case 64:
968 			*val = readq_relaxed(vaddr);
969 			break;
970 		default:
971 			pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
972 				 reg->bit_width, pcc_ss_id);
973 			ret_val = -EFAULT;
974 	}
975 
976 	return ret_val;
977 }
978 
979 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
980 {
981 	int ret_val = 0;
982 	void __iomem *vaddr = NULL;
983 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
984 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
985 
986 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
987 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
988 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
989 		vaddr = reg_res->sys_mem_vaddr;
990 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
991 		return cpc_write_ffh(cpu, reg, val);
992 	else
993 		return acpi_os_write_memory((acpi_physical_address)reg->address,
994 				val, reg->bit_width);
995 
996 	switch (reg->bit_width) {
997 		case 8:
998 			writeb_relaxed(val, vaddr);
999 			break;
1000 		case 16:
1001 			writew_relaxed(val, vaddr);
1002 			break;
1003 		case 32:
1004 			writel_relaxed(val, vaddr);
1005 			break;
1006 		case 64:
1007 			writeq_relaxed(val, vaddr);
1008 			break;
1009 		default:
1010 			pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1011 				 reg->bit_width, pcc_ss_id);
1012 			ret_val = -EFAULT;
1013 			break;
1014 	}
1015 
1016 	return ret_val;
1017 }
1018 
1019 /**
1020  * cppc_get_desired_perf - Get the value of desired performance register.
1021  * @cpunum: CPU from which to get desired performance.
1022  * @desired_perf: address of a variable to store the returned desired performance
1023  *
1024  * Return: 0 for success, -EIO otherwise.
1025  */
1026 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1027 {
1028 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1029 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1030 	struct cpc_register_resource *desired_reg;
1031 	struct cppc_pcc_data *pcc_ss_data = NULL;
1032 
1033 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1034 
1035 	if (CPC_IN_PCC(desired_reg)) {
1036 		int ret = 0;
1037 
1038 		if (pcc_ss_id < 0)
1039 			return -EIO;
1040 
1041 		pcc_ss_data = pcc_data[pcc_ss_id];
1042 
1043 		down_write(&pcc_ss_data->pcc_lock);
1044 
1045 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1046 			cpc_read(cpunum, desired_reg, desired_perf);
1047 		else
1048 			ret = -EIO;
1049 
1050 		up_write(&pcc_ss_data->pcc_lock);
1051 
1052 		return ret;
1053 	}
1054 
1055 	cpc_read(cpunum, desired_reg, desired_perf);
1056 
1057 	return 0;
1058 }
1059 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1060 
1061 /**
1062  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1063  * @cpunum: CPU from which to get capabilities info.
1064  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1065  *
1066  * Return: 0 for success with perf_caps populated else -ERRNO.
1067  */
1068 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1069 {
1070 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1071 	struct cpc_register_resource *highest_reg, *lowest_reg,
1072 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1073 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1074 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1075 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1076 	struct cppc_pcc_data *pcc_ss_data = NULL;
1077 	int ret = 0, regs_in_pcc = 0;
1078 
1079 	if (!cpc_desc) {
1080 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1081 		return -ENODEV;
1082 	}
1083 
1084 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1085 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1086 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1087 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1088 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1089 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1090 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1091 
1092 	/* Are any of the regs PCC ?*/
1093 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1094 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1095 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1096 		if (pcc_ss_id < 0) {
1097 			pr_debug("Invalid pcc_ss_id\n");
1098 			return -ENODEV;
1099 		}
1100 		pcc_ss_data = pcc_data[pcc_ss_id];
1101 		regs_in_pcc = 1;
1102 		down_write(&pcc_ss_data->pcc_lock);
1103 		/* Ring doorbell once to update PCC subspace */
1104 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1105 			ret = -EIO;
1106 			goto out_err;
1107 		}
1108 	}
1109 
1110 	cpc_read(cpunum, highest_reg, &high);
1111 	perf_caps->highest_perf = high;
1112 
1113 	cpc_read(cpunum, lowest_reg, &low);
1114 	perf_caps->lowest_perf = low;
1115 
1116 	cpc_read(cpunum, nominal_reg, &nom);
1117 	perf_caps->nominal_perf = nom;
1118 
1119 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1120 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1121 		perf_caps->guaranteed_perf = 0;
1122 	} else {
1123 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1124 		perf_caps->guaranteed_perf = guaranteed;
1125 	}
1126 
1127 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1128 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1129 
1130 	if (!high || !low || !nom || !min_nonlinear)
1131 		ret = -EFAULT;
1132 
1133 	/* Read optional lowest and nominal frequencies if present */
1134 	if (CPC_SUPPORTED(low_freq_reg))
1135 		cpc_read(cpunum, low_freq_reg, &low_f);
1136 
1137 	if (CPC_SUPPORTED(nom_freq_reg))
1138 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1139 
1140 	perf_caps->lowest_freq = low_f;
1141 	perf_caps->nominal_freq = nom_f;
1142 
1143 
1144 out_err:
1145 	if (regs_in_pcc)
1146 		up_write(&pcc_ss_data->pcc_lock);
1147 	return ret;
1148 }
1149 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1150 
1151 /**
1152  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1153  * @cpunum: CPU from which to read counters.
1154  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1155  *
1156  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1157  */
1158 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1159 {
1160 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1161 	struct cpc_register_resource *delivered_reg, *reference_reg,
1162 		*ref_perf_reg, *ctr_wrap_reg;
1163 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1164 	struct cppc_pcc_data *pcc_ss_data = NULL;
1165 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1166 	int ret = 0, regs_in_pcc = 0;
1167 
1168 	if (!cpc_desc) {
1169 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1170 		return -ENODEV;
1171 	}
1172 
1173 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1174 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1175 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1176 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1177 
1178 	/*
1179 	 * If reference perf register is not supported then we should
1180 	 * use the nominal perf value
1181 	 */
1182 	if (!CPC_SUPPORTED(ref_perf_reg))
1183 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1184 
1185 	/* Are any of the regs PCC ?*/
1186 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1187 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1188 		if (pcc_ss_id < 0) {
1189 			pr_debug("Invalid pcc_ss_id\n");
1190 			return -ENODEV;
1191 		}
1192 		pcc_ss_data = pcc_data[pcc_ss_id];
1193 		down_write(&pcc_ss_data->pcc_lock);
1194 		regs_in_pcc = 1;
1195 		/* Ring doorbell once to update PCC subspace */
1196 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1197 			ret = -EIO;
1198 			goto out_err;
1199 		}
1200 	}
1201 
1202 	cpc_read(cpunum, delivered_reg, &delivered);
1203 	cpc_read(cpunum, reference_reg, &reference);
1204 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1205 
1206 	/*
1207 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1208 	 * performance counters are assumed to never wrap during the lifetime of
1209 	 * platform
1210 	 */
1211 	ctr_wrap_time = (u64)(~((u64)0));
1212 	if (CPC_SUPPORTED(ctr_wrap_reg))
1213 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1214 
1215 	if (!delivered || !reference ||	!ref_perf) {
1216 		ret = -EFAULT;
1217 		goto out_err;
1218 	}
1219 
1220 	perf_fb_ctrs->delivered = delivered;
1221 	perf_fb_ctrs->reference = reference;
1222 	perf_fb_ctrs->reference_perf = ref_perf;
1223 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1224 out_err:
1225 	if (regs_in_pcc)
1226 		up_write(&pcc_ss_data->pcc_lock);
1227 	return ret;
1228 }
1229 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1230 
1231 /**
1232  * cppc_set_perf - Set a CPU's performance controls.
1233  * @cpu: CPU for which to set performance controls.
1234  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1235  *
1236  * Return: 0 for success, -ERRNO otherwise.
1237  */
1238 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1239 {
1240 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1241 	struct cpc_register_resource *desired_reg;
1242 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1243 	struct cppc_pcc_data *pcc_ss_data = NULL;
1244 	int ret = 0;
1245 
1246 	if (!cpc_desc) {
1247 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1248 		return -ENODEV;
1249 	}
1250 
1251 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1252 
1253 	/*
1254 	 * This is Phase-I where we want to write to CPC registers
1255 	 * -> We want all CPUs to be able to execute this phase in parallel
1256 	 *
1257 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1258 	 * achieve that goal here
1259 	 */
1260 	if (CPC_IN_PCC(desired_reg)) {
1261 		if (pcc_ss_id < 0) {
1262 			pr_debug("Invalid pcc_ss_id\n");
1263 			return -ENODEV;
1264 		}
1265 		pcc_ss_data = pcc_data[pcc_ss_id];
1266 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1267 		if (pcc_ss_data->platform_owns_pcc) {
1268 			ret = check_pcc_chan(pcc_ss_id, false);
1269 			if (ret) {
1270 				up_read(&pcc_ss_data->pcc_lock);
1271 				return ret;
1272 			}
1273 		}
1274 		/*
1275 		 * Update the pending_write to make sure a PCC CMD_READ will not
1276 		 * arrive and steal the channel during the switch to write lock
1277 		 */
1278 		pcc_ss_data->pending_pcc_write_cmd = true;
1279 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1280 		cpc_desc->write_cmd_status = 0;
1281 	}
1282 
1283 	/*
1284 	 * Skip writing MIN/MAX until Linux knows how to come up with
1285 	 * useful values.
1286 	 */
1287 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1288 
1289 	if (CPC_IN_PCC(desired_reg))
1290 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1291 	/*
1292 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1293 	 *
1294 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1295 	 * requests that happened in short overlapping interval. The last CPU to
1296 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1297 	 *
1298 	 * We have the following requirements for Phase-II:
1299 	 *     1. We want to execute Phase-II only when there are no CPUs
1300 	 * currently executing in Phase-I
1301 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1302 	 * entering Phase-I.
1303 	 *     3. We want only one CPU among all those who went through Phase-I
1304 	 * to run phase-II
1305 	 *
1306 	 * If write_trylock fails to get the lock and doesn't transfer the
1307 	 * PCC ownership to the platform, then one of the following will be TRUE
1308 	 *     1. There is at-least one CPU in Phase-I which will later execute
1309 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1310 	 * executing the Phase-II.
1311 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1312 	 * write_trylock and has already acquired the write_lock. We know for a
1313 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1314 	 * before this CPU's Phase-I as we held the read_lock.
1315 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1316 	 * down_write, in which case, send_pcc_cmd will check for pending
1317 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1318 	 * So this CPU can be certain that its request will be delivered
1319 	 *    So in all cases, this CPU knows that its request will be delivered
1320 	 * by another CPU and can return
1321 	 *
1322 	 * After getting the down_write we still need to check for
1323 	 * pending_pcc_write_cmd to take care of the following scenario
1324 	 *    The thread running this code could be scheduled out between
1325 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1326 	 * could have delivered the request to Platform by triggering the
1327 	 * doorbell and transferred the ownership of PCC to platform. So this
1328 	 * avoids triggering an unnecessary doorbell and more importantly before
1329 	 * triggering the doorbell it makes sure that the PCC channel ownership
1330 	 * is still with OSPM.
1331 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1332 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1333 	 * before the pcc CMD_WRITE is completed. pcc_send_cmd checks for this
1334 	 * case during a CMD_READ and if there are pending writes it delivers
1335 	 * the write command before servicing the read command
1336 	 */
1337 	if (CPC_IN_PCC(desired_reg)) {
1338 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1339 			/* Update only if there are pending write commands */
1340 			if (pcc_ss_data->pending_pcc_write_cmd)
1341 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1342 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1343 		} else
1344 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1345 			wait_event(pcc_ss_data->pcc_write_wait_q,
1346 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1347 
1348 		/* send_pcc_cmd updates the status in case of failure */
1349 		ret = cpc_desc->write_cmd_status;
1350 	}
1351 	return ret;
1352 }
1353 EXPORT_SYMBOL_GPL(cppc_set_perf);
1354 
1355 /**
1356  * cppc_get_transition_latency - returns frequency transition latency in ns
1357  *
1358  * ACPI CPPC does not explicitly specifiy how a platform can specify the
1359  * transition latency for perfromance change requests. The closest we have
1360  * is the timing information from the PCCT tables which provides the info
1361  * on the number and frequency of PCC commands the platform can handle.
1362  */
1363 unsigned int cppc_get_transition_latency(int cpu_num)
1364 {
1365 	/*
1366 	 * Expected transition latency is based on the PCCT timing values
1367 	 * Below are definition from ACPI spec:
1368 	 * pcc_nominal- Expected latency to process a command, in microseconds
1369 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1370 	 *              channel can support, reported in commands per minute. 0
1371 	 *              indicates no limitation.
1372 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1373 	 *              completion of a command before issuing the next command,
1374 	 *              in microseconds.
1375 	 */
1376 	unsigned int latency_ns = 0;
1377 	struct cpc_desc *cpc_desc;
1378 	struct cpc_register_resource *desired_reg;
1379 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1380 	struct cppc_pcc_data *pcc_ss_data;
1381 
1382 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1383 	if (!cpc_desc)
1384 		return CPUFREQ_ETERNAL;
1385 
1386 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1387 	if (!CPC_IN_PCC(desired_reg))
1388 		return CPUFREQ_ETERNAL;
1389 
1390 	if (pcc_ss_id < 0)
1391 		return CPUFREQ_ETERNAL;
1392 
1393 	pcc_ss_data = pcc_data[pcc_ss_id];
1394 	if (pcc_ss_data->pcc_mpar)
1395 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1396 
1397 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1398 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1399 
1400 	return latency_ns;
1401 }
1402 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1403