xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 82e6fdd6)
1 /*
2  * CPPC (Collaborative Processor Performance Control) methods used by CPUfreq drivers.
3  *
4  * (C) Copyright 2014, 2015 Linaro Ltd.
5  * Author: Ashwin Chaugule <ashwin.chaugule@linaro.org>
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; version 2
10  * of the License.
11  *
12  * CPPC describes a few methods for controlling CPU performance using
13  * information from a per CPU table called CPC. This table is described in
14  * the ACPI v5.0+ specification. The table consists of a list of
15  * registers which may be memory mapped or hardware registers and also may
16  * include some static integer values.
17  *
18  * CPU performance is on an abstract continuous scale as against a discretized
19  * P-state scale which is tied to CPU frequency only. In brief, the basic
20  * operation involves:
21  *
22  * - OS makes a CPU performance request. (Can provide min and max bounds)
23  *
24  * - Platform (such as BMC) is free to optimize request within requested bounds
25  *   depending on power/thermal budgets etc.
26  *
27  * - Platform conveys its decision back to OS
28  *
29  * The communication between OS and platform occurs through another medium
30  * called (PCC) Platform Communication Channel. This is a generic mailbox like
31  * mechanism which includes doorbell semantics to indicate register updates.
32  * See drivers/mailbox/pcc.c for details on PCC.
33  *
34  * Finer details about the PCC and CPPC spec are available in the ACPI v5.1 and
35  * above specifications.
36  */
37 
38 #define pr_fmt(fmt)	"ACPI CPPC: " fmt
39 
40 #include <linux/cpufreq.h>
41 #include <linux/delay.h>
42 #include <linux/ktime.h>
43 #include <linux/rwsem.h>
44 #include <linux/wait.h>
45 
46 #include <acpi/cppc_acpi.h>
47 
48 struct cppc_pcc_data {
49 	struct mbox_chan *pcc_channel;
50 	void __iomem *pcc_comm_addr;
51 	bool pcc_channel_acquired;
52 	ktime_t deadline;
53 	unsigned int pcc_mpar, pcc_mrtt, pcc_nominal;
54 
55 	bool pending_pcc_write_cmd;	/* Any pending/batched PCC write cmds? */
56 	bool platform_owns_pcc;		/* Ownership of PCC subspace */
57 	unsigned int pcc_write_cnt;	/* Running count of PCC write commands */
58 
59 	/*
60 	 * Lock to provide controlled access to the PCC channel.
61 	 *
62 	 * For performance critical usecases(currently cppc_set_perf)
63 	 *	We need to take read_lock and check if channel belongs to OSPM
64 	 * before reading or writing to PCC subspace
65 	 *	We need to take write_lock before transferring the channel
66 	 * ownership to the platform via a Doorbell
67 	 *	This allows us to batch a number of CPPC requests if they happen
68 	 * to originate in about the same time
69 	 *
70 	 * For non-performance critical usecases(init)
71 	 *	Take write_lock for all purposes which gives exclusive access
72 	 */
73 	struct rw_semaphore pcc_lock;
74 
75 	/* Wait queue for CPUs whose requests were batched */
76 	wait_queue_head_t pcc_write_wait_q;
77 	ktime_t last_cmd_cmpl_time;
78 	ktime_t last_mpar_reset;
79 	int mpar_count;
80 	int refcount;
81 };
82 
83 /* Array  to represent the PCC channel per subspace id */
84 static struct cppc_pcc_data *pcc_data[MAX_PCC_SUBSPACES];
85 /* The cpu_pcc_subspace_idx containsper CPU subspace id */
86 static DEFINE_PER_CPU(int, cpu_pcc_subspace_idx);
87 
88 /*
89  * The cpc_desc structure contains the ACPI register details
90  * as described in the per CPU _CPC tables. The details
91  * include the type of register (e.g. PCC, System IO, FFH etc.)
92  * and destination addresses which lets us READ/WRITE CPU performance
93  * information using the appropriate I/O methods.
94  */
95 static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
96 
97 /* pcc mapped address + header size + offset within PCC subspace */
98 #define GET_PCC_VADDR(offs, pcc_ss_id) (pcc_data[pcc_ss_id]->pcc_comm_addr + \
99 						0x8 + (offs))
100 
101 /* Check if a CPC register is in PCC */
102 #define CPC_IN_PCC(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&		\
103 				(cpc)->cpc_entry.reg.space_id ==	\
104 				ACPI_ADR_SPACE_PLATFORM_COMM)
105 
106 /* Evalutes to True if reg is a NULL register descriptor */
107 #define IS_NULL_REG(reg) ((reg)->space_id ==  ACPI_ADR_SPACE_SYSTEM_MEMORY && \
108 				(reg)->address == 0 &&			\
109 				(reg)->bit_width == 0 &&		\
110 				(reg)->bit_offset == 0 &&		\
111 				(reg)->access_width == 0)
112 
113 /* Evalutes to True if an optional cpc field is supported */
114 #define CPC_SUPPORTED(cpc) ((cpc)->type == ACPI_TYPE_INTEGER ?		\
115 				!!(cpc)->cpc_entry.int_value :		\
116 				!IS_NULL_REG(&(cpc)->cpc_entry.reg))
117 /*
118  * Arbitrary Retries in case the remote processor is slow to respond
119  * to PCC commands. Keeping it high enough to cover emulators where
120  * the processors run painfully slow.
121  */
122 #define NUM_RETRIES 500ULL
123 
124 struct cppc_attr {
125 	struct attribute attr;
126 	ssize_t (*show)(struct kobject *kobj,
127 			struct attribute *attr, char *buf);
128 	ssize_t (*store)(struct kobject *kobj,
129 			struct attribute *attr, const char *c, ssize_t count);
130 };
131 
132 #define define_one_cppc_ro(_name)		\
133 static struct cppc_attr _name =			\
134 __ATTR(_name, 0444, show_##_name, NULL)
135 
136 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
137 
138 #define show_cppc_data(access_fn, struct_name, member_name)		\
139 	static ssize_t show_##member_name(struct kobject *kobj,		\
140 					struct attribute *attr,	char *buf) \
141 	{								\
142 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
143 		struct struct_name st_name = {0};			\
144 		int ret;						\
145 									\
146 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
147 		if (ret)						\
148 			return ret;					\
149 									\
150 		return scnprintf(buf, PAGE_SIZE, "%llu\n",		\
151 				(u64)st_name.member_name);		\
152 	}								\
153 	define_one_cppc_ro(member_name)
154 
155 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
156 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
157 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
158 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
159 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, reference_perf);
160 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
161 
162 static ssize_t show_feedback_ctrs(struct kobject *kobj,
163 		struct attribute *attr, char *buf)
164 {
165 	struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);
166 	struct cppc_perf_fb_ctrs fb_ctrs = {0};
167 	int ret;
168 
169 	ret = cppc_get_perf_ctrs(cpc_ptr->cpu_id, &fb_ctrs);
170 	if (ret)
171 		return ret;
172 
173 	return scnprintf(buf, PAGE_SIZE, "ref:%llu del:%llu\n",
174 			fb_ctrs.reference, fb_ctrs.delivered);
175 }
176 define_one_cppc_ro(feedback_ctrs);
177 
178 static struct attribute *cppc_attrs[] = {
179 	&feedback_ctrs.attr,
180 	&reference_perf.attr,
181 	&wraparound_time.attr,
182 	&highest_perf.attr,
183 	&lowest_perf.attr,
184 	&lowest_nonlinear_perf.attr,
185 	&nominal_perf.attr,
186 	NULL
187 };
188 
189 static struct kobj_type cppc_ktype = {
190 	.sysfs_ops = &kobj_sysfs_ops,
191 	.default_attrs = cppc_attrs,
192 };
193 
194 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
195 {
196 	int ret = -EIO, status = 0;
197 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
198 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
199 		pcc_ss_data->pcc_comm_addr;
200 	ktime_t next_deadline = ktime_add(ktime_get(),
201 					  pcc_ss_data->deadline);
202 
203 	if (!pcc_ss_data->platform_owns_pcc)
204 		return 0;
205 
206 	/* Retry in case the remote processor was too slow to catch up. */
207 	while (!ktime_after(ktime_get(), next_deadline)) {
208 		/*
209 		 * Per spec, prior to boot the PCC space wil be initialized by
210 		 * platform and should have set the command completion bit when
211 		 * PCC can be used by OSPM
212 		 */
213 		status = readw_relaxed(&generic_comm_base->status);
214 		if (status & PCC_CMD_COMPLETE_MASK) {
215 			ret = 0;
216 			if (chk_err_bit && (status & PCC_ERROR_MASK))
217 				ret = -EIO;
218 			break;
219 		}
220 		/*
221 		 * Reducing the bus traffic in case this loop takes longer than
222 		 * a few retries.
223 		 */
224 		udelay(3);
225 	}
226 
227 	if (likely(!ret))
228 		pcc_ss_data->platform_owns_pcc = false;
229 	else
230 		pr_err("PCC check channel failed. Status=%x\n", status);
231 
232 	return ret;
233 }
234 
235 /*
236  * This function transfers the ownership of the PCC to the platform
237  * So it must be called while holding write_lock(pcc_lock)
238  */
239 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
240 {
241 	int ret = -EIO, i;
242 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
243 	struct acpi_pcct_shared_memory *generic_comm_base =
244 		(struct acpi_pcct_shared_memory *)pcc_ss_data->pcc_comm_addr;
245 	unsigned int time_delta;
246 
247 	/*
248 	 * For CMD_WRITE we know for a fact the caller should have checked
249 	 * the channel before writing to PCC space
250 	 */
251 	if (cmd == CMD_READ) {
252 		/*
253 		 * If there are pending cpc_writes, then we stole the channel
254 		 * before write completion, so first send a WRITE command to
255 		 * platform
256 		 */
257 		if (pcc_ss_data->pending_pcc_write_cmd)
258 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
259 
260 		ret = check_pcc_chan(pcc_ss_id, false);
261 		if (ret)
262 			goto end;
263 	} else /* CMD_WRITE */
264 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
265 
266 	/*
267 	 * Handle the Minimum Request Turnaround Time(MRTT)
268 	 * "The minimum amount of time that OSPM must wait after the completion
269 	 * of a command before issuing the next command, in microseconds"
270 	 */
271 	if (pcc_ss_data->pcc_mrtt) {
272 		time_delta = ktime_us_delta(ktime_get(),
273 					    pcc_ss_data->last_cmd_cmpl_time);
274 		if (pcc_ss_data->pcc_mrtt > time_delta)
275 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
276 	}
277 
278 	/*
279 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
280 	 * "The maximum number of periodic requests that the subspace channel can
281 	 * support, reported in commands per minute. 0 indicates no limitation."
282 	 *
283 	 * This parameter should be ideally zero or large enough so that it can
284 	 * handle maximum number of requests that all the cores in the system can
285 	 * collectively generate. If it is not, we will follow the spec and just
286 	 * not send the request to the platform after hitting the MPAR limit in
287 	 * any 60s window
288 	 */
289 	if (pcc_ss_data->pcc_mpar) {
290 		if (pcc_ss_data->mpar_count == 0) {
291 			time_delta = ktime_ms_delta(ktime_get(),
292 						    pcc_ss_data->last_mpar_reset);
293 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
294 				pr_debug("PCC cmd not sent due to MPAR limit");
295 				ret = -EIO;
296 				goto end;
297 			}
298 			pcc_ss_data->last_mpar_reset = ktime_get();
299 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
300 		}
301 		pcc_ss_data->mpar_count--;
302 	}
303 
304 	/* Write to the shared comm region. */
305 	writew_relaxed(cmd, &generic_comm_base->command);
306 
307 	/* Flip CMD COMPLETE bit */
308 	writew_relaxed(0, &generic_comm_base->status);
309 
310 	pcc_ss_data->platform_owns_pcc = true;
311 
312 	/* Ring doorbell */
313 	ret = mbox_send_message(pcc_ss_data->pcc_channel, &cmd);
314 	if (ret < 0) {
315 		pr_err("Err sending PCC mbox message. cmd:%d, ret:%d\n",
316 				cmd, ret);
317 		goto end;
318 	}
319 
320 	/* wait for completion and check for PCC errro bit */
321 	ret = check_pcc_chan(pcc_ss_id, true);
322 
323 	if (pcc_ss_data->pcc_mrtt)
324 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
325 
326 	if (pcc_ss_data->pcc_channel->mbox->txdone_irq)
327 		mbox_chan_txdone(pcc_ss_data->pcc_channel, ret);
328 	else
329 		mbox_client_txdone(pcc_ss_data->pcc_channel, ret);
330 
331 end:
332 	if (cmd == CMD_WRITE) {
333 		if (unlikely(ret)) {
334 			for_each_possible_cpu(i) {
335 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
336 				if (!desc)
337 					continue;
338 
339 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
340 					desc->write_cmd_status = ret;
341 			}
342 		}
343 		pcc_ss_data->pcc_write_cnt++;
344 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
345 	}
346 
347 	return ret;
348 }
349 
350 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
351 {
352 	if (ret < 0)
353 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
354 				*(u16 *)msg, ret);
355 	else
356 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
357 				*(u16 *)msg, ret);
358 }
359 
360 struct mbox_client cppc_mbox_cl = {
361 	.tx_done = cppc_chan_tx_done,
362 	.knows_txdone = true,
363 };
364 
365 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
366 {
367 	int result = -EFAULT;
368 	acpi_status status = AE_OK;
369 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
370 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
371 	struct acpi_buffer state = {0, NULL};
372 	union acpi_object  *psd = NULL;
373 	struct acpi_psd_package *pdomain;
374 
375 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL, &buffer,
376 			ACPI_TYPE_PACKAGE);
377 	if (ACPI_FAILURE(status))
378 		return -ENODEV;
379 
380 	psd = buffer.pointer;
381 	if (!psd || psd->package.count != 1) {
382 		pr_debug("Invalid _PSD data\n");
383 		goto end;
384 	}
385 
386 	pdomain = &(cpc_ptr->domain_info);
387 
388 	state.length = sizeof(struct acpi_psd_package);
389 	state.pointer = pdomain;
390 
391 	status = acpi_extract_package(&(psd->package.elements[0]),
392 		&format, &state);
393 	if (ACPI_FAILURE(status)) {
394 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
395 		goto end;
396 	}
397 
398 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
399 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
400 		goto end;
401 	}
402 
403 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
404 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
405 		goto end;
406 	}
407 
408 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
409 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
410 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
411 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
412 		goto end;
413 	}
414 
415 	result = 0;
416 end:
417 	kfree(buffer.pointer);
418 	return result;
419 }
420 
421 /**
422  * acpi_get_psd_map - Map the CPUs in a common freq domain.
423  * @all_cpu_data: Ptrs to CPU specific CPPC data including PSD info.
424  *
425  *	Return: 0 for success or negative value for err.
426  */
427 int acpi_get_psd_map(struct cppc_cpudata **all_cpu_data)
428 {
429 	int count_target;
430 	int retval = 0;
431 	unsigned int i, j;
432 	cpumask_var_t covered_cpus;
433 	struct cppc_cpudata *pr, *match_pr;
434 	struct acpi_psd_package *pdomain;
435 	struct acpi_psd_package *match_pdomain;
436 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
437 
438 	if (!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))
439 		return -ENOMEM;
440 
441 	/*
442 	 * Now that we have _PSD data from all CPUs, lets setup P-state
443 	 * domain info.
444 	 */
445 	for_each_possible_cpu(i) {
446 		pr = all_cpu_data[i];
447 		if (!pr)
448 			continue;
449 
450 		if (cpumask_test_cpu(i, covered_cpus))
451 			continue;
452 
453 		cpc_ptr = per_cpu(cpc_desc_ptr, i);
454 		if (!cpc_ptr) {
455 			retval = -EFAULT;
456 			goto err_ret;
457 		}
458 
459 		pdomain = &(cpc_ptr->domain_info);
460 		cpumask_set_cpu(i, pr->shared_cpu_map);
461 		cpumask_set_cpu(i, covered_cpus);
462 		if (pdomain->num_processors <= 1)
463 			continue;
464 
465 		/* Validate the Domain info */
466 		count_target = pdomain->num_processors;
467 		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
468 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
469 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
470 			pr->shared_type = CPUFREQ_SHARED_TYPE_HW;
471 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
472 			pr->shared_type = CPUFREQ_SHARED_TYPE_ANY;
473 
474 		for_each_possible_cpu(j) {
475 			if (i == j)
476 				continue;
477 
478 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
479 			if (!match_cpc_ptr) {
480 				retval = -EFAULT;
481 				goto err_ret;
482 			}
483 
484 			match_pdomain = &(match_cpc_ptr->domain_info);
485 			if (match_pdomain->domain != pdomain->domain)
486 				continue;
487 
488 			/* Here i and j are in the same domain */
489 			if (match_pdomain->num_processors != count_target) {
490 				retval = -EFAULT;
491 				goto err_ret;
492 			}
493 
494 			if (pdomain->coord_type != match_pdomain->coord_type) {
495 				retval = -EFAULT;
496 				goto err_ret;
497 			}
498 
499 			cpumask_set_cpu(j, covered_cpus);
500 			cpumask_set_cpu(j, pr->shared_cpu_map);
501 		}
502 
503 		for_each_possible_cpu(j) {
504 			if (i == j)
505 				continue;
506 
507 			match_pr = all_cpu_data[j];
508 			if (!match_pr)
509 				continue;
510 
511 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
512 			if (!match_cpc_ptr) {
513 				retval = -EFAULT;
514 				goto err_ret;
515 			}
516 
517 			match_pdomain = &(match_cpc_ptr->domain_info);
518 			if (match_pdomain->domain != pdomain->domain)
519 				continue;
520 
521 			match_pr->shared_type = pr->shared_type;
522 			cpumask_copy(match_pr->shared_cpu_map,
523 				     pr->shared_cpu_map);
524 		}
525 	}
526 
527 err_ret:
528 	for_each_possible_cpu(i) {
529 		pr = all_cpu_data[i];
530 		if (!pr)
531 			continue;
532 
533 		/* Assume no coordination on any error parsing domain info */
534 		if (retval) {
535 			cpumask_clear(pr->shared_cpu_map);
536 			cpumask_set_cpu(i, pr->shared_cpu_map);
537 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
538 		}
539 	}
540 
541 	free_cpumask_var(covered_cpus);
542 	return retval;
543 }
544 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
545 
546 static int register_pcc_channel(int pcc_ss_idx)
547 {
548 	struct acpi_pcct_hw_reduced *cppc_ss;
549 	u64 usecs_lat;
550 
551 	if (pcc_ss_idx >= 0) {
552 		pcc_data[pcc_ss_idx]->pcc_channel =
553 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
554 
555 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
556 			pr_err("Failed to find PCC communication channel\n");
557 			return -ENODEV;
558 		}
559 
560 		/*
561 		 * The PCC mailbox controller driver should
562 		 * have parsed the PCCT (global table of all
563 		 * PCC channels) and stored pointers to the
564 		 * subspace communication region in con_priv.
565 		 */
566 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
567 
568 		if (!cppc_ss) {
569 			pr_err("No PCC subspace found for CPPC\n");
570 			return -ENODEV;
571 		}
572 
573 		/*
574 		 * cppc_ss->latency is just a Nominal value. In reality
575 		 * the remote processor could be much slower to reply.
576 		 * So add an arbitrary amount of wait on top of Nominal.
577 		 */
578 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
579 		pcc_data[pcc_ss_idx]->deadline = ns_to_ktime(usecs_lat * NSEC_PER_USEC);
580 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
581 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
582 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
583 
584 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
585 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
586 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
587 			pr_err("Failed to ioremap PCC comm region mem\n");
588 			return -ENOMEM;
589 		}
590 
591 		/* Set flag so that we dont come here for each CPU. */
592 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
593 	}
594 
595 	return 0;
596 }
597 
598 /**
599  * cpc_ffh_supported() - check if FFH reading supported
600  *
601  * Check if the architecture has support for functional fixed hardware
602  * read/write capability.
603  *
604  * Return: true for supported, false for not supported
605  */
606 bool __weak cpc_ffh_supported(void)
607 {
608 	return false;
609 }
610 
611 
612 /**
613  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
614  *
615  * Check and allocate the cppc_pcc_data memory.
616  * In some processor configurations it is possible that same subspace
617  * is shared between multiple CPU's. This is seen especially in CPU's
618  * with hardware multi-threading support.
619  *
620  * Return: 0 for success, errno for failure
621  */
622 int pcc_data_alloc(int pcc_ss_id)
623 {
624 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
625 		return -EINVAL;
626 
627 	if (pcc_data[pcc_ss_id]) {
628 		pcc_data[pcc_ss_id]->refcount++;
629 	} else {
630 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
631 					      GFP_KERNEL);
632 		if (!pcc_data[pcc_ss_id])
633 			return -ENOMEM;
634 		pcc_data[pcc_ss_id]->refcount++;
635 	}
636 
637 	return 0;
638 }
639 /*
640  * An example CPC table looks like the following.
641  *
642  *	Name(_CPC, Package()
643  *			{
644  *			17,
645  *			NumEntries
646  *			1,
647  *			// Revision
648  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
649  *			// Highest Performance
650  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
651  *			// Nominal Performance
652  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
653  *			// Lowest Nonlinear Performance
654  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
655  *			// Lowest Performance
656  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
657  *			// Guaranteed Performance Register
658  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
659  *			// Desired Performance Register
660  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
661  *			..
662  *			..
663  *			..
664  *
665  *		}
666  * Each Register() encodes how to access that specific register.
667  * e.g. a sample PCC entry has the following encoding:
668  *
669  *	Register (
670  *		PCC,
671  *		AddressSpaceKeyword
672  *		8,
673  *		//RegisterBitWidth
674  *		8,
675  *		//RegisterBitOffset
676  *		0x30,
677  *		//RegisterAddress
678  *		9
679  *		//AccessSize (subspace ID)
680  *		0
681  *		)
682  *	}
683  */
684 
685 /**
686  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
687  * @pr: Ptr to acpi_processor containing this CPUs logical Id.
688  *
689  *	Return: 0 for success or negative value for err.
690  */
691 int acpi_cppc_processor_probe(struct acpi_processor *pr)
692 {
693 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
694 	union acpi_object *out_obj, *cpc_obj;
695 	struct cpc_desc *cpc_ptr;
696 	struct cpc_reg *gas_t;
697 	struct device *cpu_dev;
698 	acpi_handle handle = pr->handle;
699 	unsigned int num_ent, i, cpc_rev;
700 	int pcc_subspace_id = -1;
701 	acpi_status status;
702 	int ret = -EFAULT;
703 
704 	/* Parse the ACPI _CPC table for this cpu. */
705 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
706 			ACPI_TYPE_PACKAGE);
707 	if (ACPI_FAILURE(status)) {
708 		ret = -ENODEV;
709 		goto out_buf_free;
710 	}
711 
712 	out_obj = (union acpi_object *) output.pointer;
713 
714 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
715 	if (!cpc_ptr) {
716 		ret = -ENOMEM;
717 		goto out_buf_free;
718 	}
719 
720 	/* First entry is NumEntries. */
721 	cpc_obj = &out_obj->package.elements[0];
722 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
723 		num_ent = cpc_obj->integer.value;
724 	} else {
725 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
726 				cpc_obj->type);
727 		goto out_free;
728 	}
729 
730 	/* Only support CPPCv2. Bail otherwise. */
731 	if (num_ent != CPPC_NUM_ENT) {
732 		pr_debug("Firmware exports %d entries. Expected: %d\n",
733 				num_ent, CPPC_NUM_ENT);
734 		goto out_free;
735 	}
736 
737 	cpc_ptr->num_entries = num_ent;
738 
739 	/* Second entry should be revision. */
740 	cpc_obj = &out_obj->package.elements[1];
741 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
742 		cpc_rev = cpc_obj->integer.value;
743 	} else {
744 		pr_debug("Unexpected entry type(%d) for Revision\n",
745 				cpc_obj->type);
746 		goto out_free;
747 	}
748 
749 	if (cpc_rev != CPPC_REV) {
750 		pr_debug("Firmware exports revision:%d. Expected:%d\n",
751 				cpc_rev, CPPC_REV);
752 		goto out_free;
753 	}
754 
755 	/* Iterate through remaining entries in _CPC */
756 	for (i = 2; i < num_ent; i++) {
757 		cpc_obj = &out_obj->package.elements[i];
758 
759 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
760 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
761 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
762 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
763 			gas_t = (struct cpc_reg *)
764 				cpc_obj->buffer.pointer;
765 
766 			/*
767 			 * The PCC Subspace index is encoded inside
768 			 * the CPC table entries. The same PCC index
769 			 * will be used for all the PCC entries,
770 			 * so extract it only once.
771 			 */
772 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
773 				if (pcc_subspace_id < 0) {
774 					pcc_subspace_id = gas_t->access_width;
775 					if (pcc_data_alloc(pcc_subspace_id))
776 						goto out_free;
777 				} else if (pcc_subspace_id != gas_t->access_width) {
778 					pr_debug("Mismatched PCC ids.\n");
779 					goto out_free;
780 				}
781 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
782 				if (gas_t->address) {
783 					void __iomem *addr;
784 
785 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
786 					if (!addr)
787 						goto out_free;
788 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
789 				}
790 			} else {
791 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
792 					/* Support only PCC ,SYS MEM and FFH type regs */
793 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
794 					goto out_free;
795 				}
796 			}
797 
798 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
799 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
800 		} else {
801 			pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
802 			goto out_free;
803 		}
804 	}
805 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
806 	/* Store CPU Logical ID */
807 	cpc_ptr->cpu_id = pr->id;
808 
809 	/* Parse PSD data for this CPU */
810 	ret = acpi_get_psd(cpc_ptr, handle);
811 	if (ret)
812 		goto out_free;
813 
814 	/* Register PCC channel once for all PCC subspace id. */
815 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
816 		ret = register_pcc_channel(pcc_subspace_id);
817 		if (ret)
818 			goto out_free;
819 
820 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
821 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
822 	}
823 
824 	/* Everything looks okay */
825 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
826 
827 	/* Add per logical CPU nodes for reading its feedback counters. */
828 	cpu_dev = get_cpu_device(pr->id);
829 	if (!cpu_dev) {
830 		ret = -EINVAL;
831 		goto out_free;
832 	}
833 
834 	/* Plug PSD data into this CPUs CPC descriptor. */
835 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
836 
837 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
838 			"acpi_cppc");
839 	if (ret) {
840 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
841 		goto out_free;
842 	}
843 
844 	kfree(output.pointer);
845 	return 0;
846 
847 out_free:
848 	/* Free all the mapped sys mem areas for this CPU */
849 	for (i = 2; i < cpc_ptr->num_entries; i++) {
850 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
851 
852 		if (addr)
853 			iounmap(addr);
854 	}
855 	kfree(cpc_ptr);
856 
857 out_buf_free:
858 	kfree(output.pointer);
859 	return ret;
860 }
861 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
862 
863 /**
864  * acpi_cppc_processor_exit - Cleanup CPC structs.
865  * @pr: Ptr to acpi_processor containing this CPUs logical Id.
866  *
867  * Return: Void
868  */
869 void acpi_cppc_processor_exit(struct acpi_processor *pr)
870 {
871 	struct cpc_desc *cpc_ptr;
872 	unsigned int i;
873 	void __iomem *addr;
874 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
875 
876 	if (pcc_ss_id >=0 && pcc_data[pcc_ss_id]) {
877 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
878 			pcc_data[pcc_ss_id]->refcount--;
879 			if (!pcc_data[pcc_ss_id]->refcount) {
880 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
881 				pcc_data[pcc_ss_id]->pcc_channel_acquired = 0;
882 				kfree(pcc_data[pcc_ss_id]);
883 			}
884 		}
885 	}
886 
887 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
888 	if (!cpc_ptr)
889 		return;
890 
891 	/* Free all the mapped sys mem areas for this CPU */
892 	for (i = 2; i < cpc_ptr->num_entries; i++) {
893 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
894 		if (addr)
895 			iounmap(addr);
896 	}
897 
898 	kobject_put(&cpc_ptr->kobj);
899 	kfree(cpc_ptr);
900 }
901 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
902 
903 /**
904  * cpc_read_ffh() - Read FFH register
905  * @cpunum:	cpu number to read
906  * @reg:	cppc register information
907  * @val:	place holder for return value
908  *
909  * Read bit_width bits from a specified address and bit_offset
910  *
911  * Return: 0 for success and error code
912  */
913 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
914 {
915 	return -ENOTSUPP;
916 }
917 
918 /**
919  * cpc_write_ffh() - Write FFH register
920  * @cpunum:	cpu number to write
921  * @reg:	cppc register information
922  * @val:	value to write
923  *
924  * Write value of bit_width bits to a specified address and bit_offset
925  *
926  * Return: 0 for success and error code
927  */
928 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
929 {
930 	return -ENOTSUPP;
931 }
932 
933 /*
934  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
935  * as fast as possible. We have already mapped the PCC subspace during init, so
936  * we can directly write to it.
937  */
938 
939 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
940 {
941 	int ret_val = 0;
942 	void __iomem *vaddr = 0;
943 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
944 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
945 
946 	if (reg_res->type == ACPI_TYPE_INTEGER) {
947 		*val = reg_res->cpc_entry.int_value;
948 		return ret_val;
949 	}
950 
951 	*val = 0;
952 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
953 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
954 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
955 		vaddr = reg_res->sys_mem_vaddr;
956 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
957 		return cpc_read_ffh(cpu, reg, val);
958 	else
959 		return acpi_os_read_memory((acpi_physical_address)reg->address,
960 				val, reg->bit_width);
961 
962 	switch (reg->bit_width) {
963 		case 8:
964 			*val = readb_relaxed(vaddr);
965 			break;
966 		case 16:
967 			*val = readw_relaxed(vaddr);
968 			break;
969 		case 32:
970 			*val = readl_relaxed(vaddr);
971 			break;
972 		case 64:
973 			*val = readq_relaxed(vaddr);
974 			break;
975 		default:
976 			pr_debug("Error: Cannot read %u bit width from PCC\n",
977 					reg->bit_width);
978 			ret_val = -EFAULT;
979 	}
980 
981 	return ret_val;
982 }
983 
984 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
985 {
986 	int ret_val = 0;
987 	void __iomem *vaddr = 0;
988 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
989 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
990 
991 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
992 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
993 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
994 		vaddr = reg_res->sys_mem_vaddr;
995 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
996 		return cpc_write_ffh(cpu, reg, val);
997 	else
998 		return acpi_os_write_memory((acpi_physical_address)reg->address,
999 				val, reg->bit_width);
1000 
1001 	switch (reg->bit_width) {
1002 		case 8:
1003 			writeb_relaxed(val, vaddr);
1004 			break;
1005 		case 16:
1006 			writew_relaxed(val, vaddr);
1007 			break;
1008 		case 32:
1009 			writel_relaxed(val, vaddr);
1010 			break;
1011 		case 64:
1012 			writeq_relaxed(val, vaddr);
1013 			break;
1014 		default:
1015 			pr_debug("Error: Cannot write %u bit width to PCC\n",
1016 					reg->bit_width);
1017 			ret_val = -EFAULT;
1018 			break;
1019 	}
1020 
1021 	return ret_val;
1022 }
1023 
1024 /**
1025  * cppc_get_perf_caps - Get a CPUs performance capabilities.
1026  * @cpunum: CPU from which to get capabilities info.
1027  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1028  *
1029  * Return: 0 for success with perf_caps populated else -ERRNO.
1030  */
1031 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1032 {
1033 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1034 	struct cpc_register_resource *highest_reg, *lowest_reg,
1035 		*lowest_non_linear_reg, *nominal_reg;
1036 	u64 high, low, nom, min_nonlinear;
1037 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1038 	struct cppc_pcc_data *pcc_ss_data;
1039 	int ret = 0, regs_in_pcc = 0;
1040 
1041 	if (!cpc_desc || pcc_ss_id < 0) {
1042 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1043 		return -ENODEV;
1044 	}
1045 
1046 	pcc_ss_data = pcc_data[pcc_ss_id];
1047 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1048 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1049 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1050 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1051 
1052 	/* Are any of the regs PCC ?*/
1053 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1054 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg)) {
1055 		regs_in_pcc = 1;
1056 		down_write(&pcc_ss_data->pcc_lock);
1057 		/* Ring doorbell once to update PCC subspace */
1058 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1059 			ret = -EIO;
1060 			goto out_err;
1061 		}
1062 	}
1063 
1064 	cpc_read(cpunum, highest_reg, &high);
1065 	perf_caps->highest_perf = high;
1066 
1067 	cpc_read(cpunum, lowest_reg, &low);
1068 	perf_caps->lowest_perf = low;
1069 
1070 	cpc_read(cpunum, nominal_reg, &nom);
1071 	perf_caps->nominal_perf = nom;
1072 
1073 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1074 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1075 
1076 	if (!high || !low || !nom || !min_nonlinear)
1077 		ret = -EFAULT;
1078 
1079 out_err:
1080 	if (regs_in_pcc)
1081 		up_write(&pcc_ss_data->pcc_lock);
1082 	return ret;
1083 }
1084 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1085 
1086 /**
1087  * cppc_get_perf_ctrs - Read a CPUs performance feedback counters.
1088  * @cpunum: CPU from which to read counters.
1089  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1090  *
1091  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1092  */
1093 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1094 {
1095 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1096 	struct cpc_register_resource *delivered_reg, *reference_reg,
1097 		*ref_perf_reg, *ctr_wrap_reg;
1098 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1099 	struct cppc_pcc_data *pcc_ss_data;
1100 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1101 	int ret = 0, regs_in_pcc = 0;
1102 
1103 	if (!cpc_desc || pcc_ss_id < 0) {
1104 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1105 		return -ENODEV;
1106 	}
1107 
1108 	pcc_ss_data = pcc_data[pcc_ss_id];
1109 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1110 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1111 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1112 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1113 
1114 	/*
1115 	 * If refernce perf register is not supported then we should
1116 	 * use the nominal perf value
1117 	 */
1118 	if (!CPC_SUPPORTED(ref_perf_reg))
1119 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1120 
1121 	/* Are any of the regs PCC ?*/
1122 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1123 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1124 		down_write(&pcc_ss_data->pcc_lock);
1125 		regs_in_pcc = 1;
1126 		/* Ring doorbell once to update PCC subspace */
1127 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1128 			ret = -EIO;
1129 			goto out_err;
1130 		}
1131 	}
1132 
1133 	cpc_read(cpunum, delivered_reg, &delivered);
1134 	cpc_read(cpunum, reference_reg, &reference);
1135 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1136 
1137 	/*
1138 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1139 	 * performance counters are assumed to never wrap during the lifetime of
1140 	 * platform
1141 	 */
1142 	ctr_wrap_time = (u64)(~((u64)0));
1143 	if (CPC_SUPPORTED(ctr_wrap_reg))
1144 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1145 
1146 	if (!delivered || !reference ||	!ref_perf) {
1147 		ret = -EFAULT;
1148 		goto out_err;
1149 	}
1150 
1151 	perf_fb_ctrs->delivered = delivered;
1152 	perf_fb_ctrs->reference = reference;
1153 	perf_fb_ctrs->reference_perf = ref_perf;
1154 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1155 out_err:
1156 	if (regs_in_pcc)
1157 		up_write(&pcc_ss_data->pcc_lock);
1158 	return ret;
1159 }
1160 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1161 
1162 /**
1163  * cppc_set_perf - Set a CPUs performance controls.
1164  * @cpu: CPU for which to set performance controls.
1165  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1166  *
1167  * Return: 0 for success, -ERRNO otherwise.
1168  */
1169 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1170 {
1171 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1172 	struct cpc_register_resource *desired_reg;
1173 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1174 	struct cppc_pcc_data *pcc_ss_data;
1175 	int ret = 0;
1176 
1177 	if (!cpc_desc || pcc_ss_id < 0) {
1178 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1179 		return -ENODEV;
1180 	}
1181 
1182 	pcc_ss_data = pcc_data[pcc_ss_id];
1183 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1184 
1185 	/*
1186 	 * This is Phase-I where we want to write to CPC registers
1187 	 * -> We want all CPUs to be able to execute this phase in parallel
1188 	 *
1189 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1190 	 * achieve that goal here
1191 	 */
1192 	if (CPC_IN_PCC(desired_reg)) {
1193 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1194 		if (pcc_ss_data->platform_owns_pcc) {
1195 			ret = check_pcc_chan(pcc_ss_id, false);
1196 			if (ret) {
1197 				up_read(&pcc_ss_data->pcc_lock);
1198 				return ret;
1199 			}
1200 		}
1201 		/*
1202 		 * Update the pending_write to make sure a PCC CMD_READ will not
1203 		 * arrive and steal the channel during the switch to write lock
1204 		 */
1205 		pcc_ss_data->pending_pcc_write_cmd = true;
1206 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1207 		cpc_desc->write_cmd_status = 0;
1208 	}
1209 
1210 	/*
1211 	 * Skip writing MIN/MAX until Linux knows how to come up with
1212 	 * useful values.
1213 	 */
1214 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1215 
1216 	if (CPC_IN_PCC(desired_reg))
1217 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1218 	/*
1219 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1220 	 *
1221 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1222 	 * requests that happened in short overlapping interval. The last CPU to
1223 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1224 	 *
1225 	 * We have the following requirements for Phase-II:
1226 	 *     1. We want to execute Phase-II only when there are no CPUs
1227 	 * currently executing in Phase-I
1228 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1229 	 * entering Phase-I.
1230 	 *     3. We want only one CPU among all those who went through Phase-I
1231 	 * to run phase-II
1232 	 *
1233 	 * If write_trylock fails to get the lock and doesn't transfer the
1234 	 * PCC ownership to the platform, then one of the following will be TRUE
1235 	 *     1. There is at-least one CPU in Phase-I which will later execute
1236 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1237 	 * executing the Phase-II.
1238 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1239 	 * write_trylock and has already acquired the write_lock. We know for a
1240 	 * fact it(other CPU acquiring the write_lock) couldn't have happened
1241 	 * before this CPU's Phase-I as we held the read_lock.
1242 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1243 	 * down_write, in which case, send_pcc_cmd will check for pending
1244 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1245 	 * So this CPU can be certain that its request will be delivered
1246 	 *    So in all cases, this CPU knows that its request will be delivered
1247 	 * by another CPU and can return
1248 	 *
1249 	 * After getting the down_write we still need to check for
1250 	 * pending_pcc_write_cmd to take care of the following scenario
1251 	 *    The thread running this code could be scheduled out between
1252 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1253 	 * could have delivered the request to Platform by triggering the
1254 	 * doorbell and transferred the ownership of PCC to platform. So this
1255 	 * avoids triggering an unnecessary doorbell and more importantly before
1256 	 * triggering the doorbell it makes sure that the PCC channel ownership
1257 	 * is still with OSPM.
1258 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1259 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1260 	 * before the pcc CMD_WRITE is completed. pcc_send_cmd checks for this
1261 	 * case during a CMD_READ and if there are pending writes it delivers
1262 	 * the write command before servicing the read command
1263 	 */
1264 	if (CPC_IN_PCC(desired_reg)) {
1265 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1266 			/* Update only if there are pending write commands */
1267 			if (pcc_ss_data->pending_pcc_write_cmd)
1268 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1269 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1270 		} else
1271 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1272 			wait_event(pcc_ss_data->pcc_write_wait_q,
1273 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1274 
1275 		/* send_pcc_cmd updates the status in case of failure */
1276 		ret = cpc_desc->write_cmd_status;
1277 	}
1278 	return ret;
1279 }
1280 EXPORT_SYMBOL_GPL(cppc_set_perf);
1281 
1282 /**
1283  * cppc_get_transition_latency - returns frequency transition latency in ns
1284  *
1285  * ACPI CPPC does not explicitly specifiy how a platform can specify the
1286  * transition latency for perfromance change requests. The closest we have
1287  * is the timing information from the PCCT tables which provides the info
1288  * on the number and frequency of PCC commands the platform can handle.
1289  */
1290 unsigned int cppc_get_transition_latency(int cpu_num)
1291 {
1292 	/*
1293 	 * Expected transition latency is based on the PCCT timing values
1294 	 * Below are definition from ACPI spec:
1295 	 * pcc_nominal- Expected latency to process a command, in microseconds
1296 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1297 	 *              channel can support, reported in commands per minute. 0
1298 	 *              indicates no limitation.
1299 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1300 	 *              completion of a command before issuing the next command,
1301 	 *              in microseconds.
1302 	 */
1303 	unsigned int latency_ns = 0;
1304 	struct cpc_desc *cpc_desc;
1305 	struct cpc_register_resource *desired_reg;
1306 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1307 	struct cppc_pcc_data *pcc_ss_data;
1308 
1309 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1310 	if (!cpc_desc)
1311 		return CPUFREQ_ETERNAL;
1312 
1313 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1314 	if (!CPC_IN_PCC(desired_reg))
1315 		return CPUFREQ_ETERNAL;
1316 
1317 	if (pcc_ss_id < 0)
1318 		return CPUFREQ_ETERNAL;
1319 
1320 	pcc_ss_data = pcc_data[pcc_ss_id];
1321 	if (pcc_ss_data->pcc_mpar)
1322 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1323 
1324 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1325 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1326 
1327 	return latency_ns;
1328 }
1329 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1330