xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 78bb17f7)
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 
43 #include <acpi/cppc_acpi.h>
44 
45 struct cppc_pcc_data {
46 	struct mbox_chan *pcc_channel;
47 	void __iomem *pcc_comm_addr;
48 	bool pcc_channel_acquired;
49 	unsigned int deadline_us;
50 	unsigned int pcc_mpar, pcc_mrtt, pcc_nominal;
51 
52 	bool pending_pcc_write_cmd;	/* Any pending/batched PCC write cmds? */
53 	bool platform_owns_pcc;		/* Ownership of PCC subspace */
54 	unsigned int pcc_write_cnt;	/* Running count of PCC write commands */
55 
56 	/*
57 	 * Lock to provide controlled access to the PCC channel.
58 	 *
59 	 * For performance critical usecases(currently cppc_set_perf)
60 	 *	We need to take read_lock and check if channel belongs to OSPM
61 	 * before reading or writing to PCC subspace
62 	 *	We need to take write_lock before transferring the channel
63 	 * ownership to the platform via a Doorbell
64 	 *	This allows us to batch a number of CPPC requests if they happen
65 	 * to originate in about the same time
66 	 *
67 	 * For non-performance critical usecases(init)
68 	 *	Take write_lock for all purposes which gives exclusive access
69 	 */
70 	struct rw_semaphore pcc_lock;
71 
72 	/* Wait queue for CPUs whose requests were batched */
73 	wait_queue_head_t pcc_write_wait_q;
74 	ktime_t last_cmd_cmpl_time;
75 	ktime_t last_mpar_reset;
76 	int mpar_count;
77 	int refcount;
78 };
79 
80 /* Array to represent the PCC channel per subspace ID */
81 static struct cppc_pcc_data *pcc_data[MAX_PCC_SUBSPACES];
82 /* The cpu_pcc_subspace_idx contains per CPU subspace ID */
83 static DEFINE_PER_CPU(int, cpu_pcc_subspace_idx);
84 
85 /*
86  * The cpc_desc structure contains the ACPI register details
87  * as described in the per CPU _CPC tables. The details
88  * include the type of register (e.g. PCC, System IO, FFH etc.)
89  * and destination addresses which lets us READ/WRITE CPU performance
90  * information using the appropriate I/O methods.
91  */
92 static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
93 
94 /* pcc mapped address + header size + offset within PCC subspace */
95 #define GET_PCC_VADDR(offs, pcc_ss_id) (pcc_data[pcc_ss_id]->pcc_comm_addr + \
96 						0x8 + (offs))
97 
98 /* Check if a CPC register is in PCC */
99 #define CPC_IN_PCC(cpc) ((cpc)->type == ACPI_TYPE_BUFFER &&		\
100 				(cpc)->cpc_entry.reg.space_id ==	\
101 				ACPI_ADR_SPACE_PLATFORM_COMM)
102 
103 /* Evalutes to True if reg is a NULL register descriptor */
104 #define IS_NULL_REG(reg) ((reg)->space_id ==  ACPI_ADR_SPACE_SYSTEM_MEMORY && \
105 				(reg)->address == 0 &&			\
106 				(reg)->bit_width == 0 &&		\
107 				(reg)->bit_offset == 0 &&		\
108 				(reg)->access_width == 0)
109 
110 /* Evalutes to True if an optional cpc field is supported */
111 #define CPC_SUPPORTED(cpc) ((cpc)->type == ACPI_TYPE_INTEGER ?		\
112 				!!(cpc)->cpc_entry.int_value :		\
113 				!IS_NULL_REG(&(cpc)->cpc_entry.reg))
114 /*
115  * Arbitrary Retries in case the remote processor is slow to respond
116  * to PCC commands. Keeping it high enough to cover emulators where
117  * the processors run painfully slow.
118  */
119 #define NUM_RETRIES 500ULL
120 
121 struct cppc_attr {
122 	struct attribute attr;
123 	ssize_t (*show)(struct kobject *kobj,
124 			struct attribute *attr, char *buf);
125 	ssize_t (*store)(struct kobject *kobj,
126 			struct attribute *attr, const char *c, ssize_t count);
127 };
128 
129 #define define_one_cppc_ro(_name)		\
130 static struct cppc_attr _name =			\
131 __ATTR(_name, 0444, show_##_name, NULL)
132 
133 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
134 
135 #define show_cppc_data(access_fn, struct_name, member_name)		\
136 	static ssize_t show_##member_name(struct kobject *kobj,		\
137 					struct attribute *attr,	char *buf) \
138 	{								\
139 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
140 		struct struct_name st_name = {0};			\
141 		int ret;						\
142 									\
143 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
144 		if (ret)						\
145 			return ret;					\
146 									\
147 		return scnprintf(buf, PAGE_SIZE, "%llu\n",		\
148 				(u64)st_name.member_name);		\
149 	}								\
150 	define_one_cppc_ro(member_name)
151 
152 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
153 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
154 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
155 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
156 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_freq);
157 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_freq);
158 
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 	&nominal_freq.attr,
187 	&lowest_freq.attr,
188 	NULL
189 };
190 
191 static struct kobj_type cppc_ktype = {
192 	.sysfs_ops = &kobj_sysfs_ops,
193 	.default_attrs = cppc_attrs,
194 };
195 
196 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
197 {
198 	int ret, status;
199 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
200 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
201 		pcc_ss_data->pcc_comm_addr;
202 
203 	if (!pcc_ss_data->platform_owns_pcc)
204 		return 0;
205 
206 	/*
207 	 * Poll PCC status register every 3us(delay_us) for maximum of
208 	 * deadline_us(timeout_us) until PCC command complete bit is set(cond)
209 	 */
210 	ret = readw_relaxed_poll_timeout(&generic_comm_base->status, status,
211 					status & PCC_CMD_COMPLETE_MASK, 3,
212 					pcc_ss_data->deadline_us);
213 
214 	if (likely(!ret)) {
215 		pcc_ss_data->platform_owns_pcc = false;
216 		if (chk_err_bit && (status & PCC_ERROR_MASK))
217 			ret = -EIO;
218 	}
219 
220 	if (unlikely(ret))
221 		pr_err("PCC check channel failed for ss: %d. ret=%d\n",
222 		       pcc_ss_id, ret);
223 
224 	return ret;
225 }
226 
227 /*
228  * This function transfers the ownership of the PCC to the platform
229  * So it must be called while holding write_lock(pcc_lock)
230  */
231 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
232 {
233 	int ret = -EIO, i;
234 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
235 	struct acpi_pcct_shared_memory *generic_comm_base =
236 		(struct acpi_pcct_shared_memory *)pcc_ss_data->pcc_comm_addr;
237 	unsigned int time_delta;
238 
239 	/*
240 	 * For CMD_WRITE we know for a fact the caller should have checked
241 	 * the channel before writing to PCC space
242 	 */
243 	if (cmd == CMD_READ) {
244 		/*
245 		 * If there are pending cpc_writes, then we stole the channel
246 		 * before write completion, so first send a WRITE command to
247 		 * platform
248 		 */
249 		if (pcc_ss_data->pending_pcc_write_cmd)
250 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
251 
252 		ret = check_pcc_chan(pcc_ss_id, false);
253 		if (ret)
254 			goto end;
255 	} else /* CMD_WRITE */
256 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
257 
258 	/*
259 	 * Handle the Minimum Request Turnaround Time(MRTT)
260 	 * "The minimum amount of time that OSPM must wait after the completion
261 	 * of a command before issuing the next command, in microseconds"
262 	 */
263 	if (pcc_ss_data->pcc_mrtt) {
264 		time_delta = ktime_us_delta(ktime_get(),
265 					    pcc_ss_data->last_cmd_cmpl_time);
266 		if (pcc_ss_data->pcc_mrtt > time_delta)
267 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
268 	}
269 
270 	/*
271 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
272 	 * "The maximum number of periodic requests that the subspace channel can
273 	 * support, reported in commands per minute. 0 indicates no limitation."
274 	 *
275 	 * This parameter should be ideally zero or large enough so that it can
276 	 * handle maximum number of requests that all the cores in the system can
277 	 * collectively generate. If it is not, we will follow the spec and just
278 	 * not send the request to the platform after hitting the MPAR limit in
279 	 * any 60s window
280 	 */
281 	if (pcc_ss_data->pcc_mpar) {
282 		if (pcc_ss_data->mpar_count == 0) {
283 			time_delta = ktime_ms_delta(ktime_get(),
284 						    pcc_ss_data->last_mpar_reset);
285 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
286 				pr_debug("PCC cmd for subspace %d not sent due to MPAR limit",
287 					 pcc_ss_id);
288 				ret = -EIO;
289 				goto end;
290 			}
291 			pcc_ss_data->last_mpar_reset = ktime_get();
292 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
293 		}
294 		pcc_ss_data->mpar_count--;
295 	}
296 
297 	/* Write to the shared comm region. */
298 	writew_relaxed(cmd, &generic_comm_base->command);
299 
300 	/* Flip CMD COMPLETE bit */
301 	writew_relaxed(0, &generic_comm_base->status);
302 
303 	pcc_ss_data->platform_owns_pcc = true;
304 
305 	/* Ring doorbell */
306 	ret = mbox_send_message(pcc_ss_data->pcc_channel, &cmd);
307 	if (ret < 0) {
308 		pr_err("Err sending PCC mbox message. ss: %d cmd:%d, ret:%d\n",
309 		       pcc_ss_id, cmd, ret);
310 		goto end;
311 	}
312 
313 	/* wait for completion and check for PCC errro bit */
314 	ret = check_pcc_chan(pcc_ss_id, true);
315 
316 	if (pcc_ss_data->pcc_mrtt)
317 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
318 
319 	if (pcc_ss_data->pcc_channel->mbox->txdone_irq)
320 		mbox_chan_txdone(pcc_ss_data->pcc_channel, ret);
321 	else
322 		mbox_client_txdone(pcc_ss_data->pcc_channel, ret);
323 
324 end:
325 	if (cmd == CMD_WRITE) {
326 		if (unlikely(ret)) {
327 			for_each_possible_cpu(i) {
328 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
329 				if (!desc)
330 					continue;
331 
332 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
333 					desc->write_cmd_status = ret;
334 			}
335 		}
336 		pcc_ss_data->pcc_write_cnt++;
337 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
338 	}
339 
340 	return ret;
341 }
342 
343 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
344 {
345 	if (ret < 0)
346 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
347 				*(u16 *)msg, ret);
348 	else
349 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
350 				*(u16 *)msg, ret);
351 }
352 
353 struct mbox_client cppc_mbox_cl = {
354 	.tx_done = cppc_chan_tx_done,
355 	.knows_txdone = true,
356 };
357 
358 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
359 {
360 	int result = -EFAULT;
361 	acpi_status status = AE_OK;
362 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
363 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
364 	struct acpi_buffer state = {0, NULL};
365 	union acpi_object  *psd = NULL;
366 	struct acpi_psd_package *pdomain;
367 
368 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL,
369 					    &buffer, ACPI_TYPE_PACKAGE);
370 	if (status == AE_NOT_FOUND)	/* _PSD is optional */
371 		return 0;
372 	if (ACPI_FAILURE(status))
373 		return -ENODEV;
374 
375 	psd = buffer.pointer;
376 	if (!psd || psd->package.count != 1) {
377 		pr_debug("Invalid _PSD data\n");
378 		goto end;
379 	}
380 
381 	pdomain = &(cpc_ptr->domain_info);
382 
383 	state.length = sizeof(struct acpi_psd_package);
384 	state.pointer = pdomain;
385 
386 	status = acpi_extract_package(&(psd->package.elements[0]),
387 		&format, &state);
388 	if (ACPI_FAILURE(status)) {
389 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
390 		goto end;
391 	}
392 
393 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
394 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
395 		goto end;
396 	}
397 
398 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
399 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
400 		goto end;
401 	}
402 
403 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
404 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
405 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
406 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
407 		goto end;
408 	}
409 
410 	result = 0;
411 end:
412 	kfree(buffer.pointer);
413 	return result;
414 }
415 
416 /**
417  * acpi_get_psd_map - Map the CPUs in a common freq domain.
418  * @all_cpu_data: Ptrs to CPU specific CPPC data including PSD info.
419  *
420  *	Return: 0 for success or negative value for err.
421  */
422 int acpi_get_psd_map(struct cppc_cpudata **all_cpu_data)
423 {
424 	int count_target;
425 	int retval = 0;
426 	unsigned int i, j;
427 	cpumask_var_t covered_cpus;
428 	struct cppc_cpudata *pr, *match_pr;
429 	struct acpi_psd_package *pdomain;
430 	struct acpi_psd_package *match_pdomain;
431 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
432 
433 	if (!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))
434 		return -ENOMEM;
435 
436 	/*
437 	 * Now that we have _PSD data from all CPUs, let's setup P-state
438 	 * domain info.
439 	 */
440 	for_each_possible_cpu(i) {
441 		if (cpumask_test_cpu(i, covered_cpus))
442 			continue;
443 
444 		pr = all_cpu_data[i];
445 		cpc_ptr = per_cpu(cpc_desc_ptr, i);
446 		if (!cpc_ptr) {
447 			retval = -EFAULT;
448 			goto err_ret;
449 		}
450 
451 		pdomain = &(cpc_ptr->domain_info);
452 		cpumask_set_cpu(i, pr->shared_cpu_map);
453 		cpumask_set_cpu(i, covered_cpus);
454 		if (pdomain->num_processors <= 1)
455 			continue;
456 
457 		/* Validate the Domain info */
458 		count_target = pdomain->num_processors;
459 		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
460 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
461 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
462 			pr->shared_type = CPUFREQ_SHARED_TYPE_HW;
463 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
464 			pr->shared_type = CPUFREQ_SHARED_TYPE_ANY;
465 
466 		for_each_possible_cpu(j) {
467 			if (i == j)
468 				continue;
469 
470 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
471 			if (!match_cpc_ptr) {
472 				retval = -EFAULT;
473 				goto err_ret;
474 			}
475 
476 			match_pdomain = &(match_cpc_ptr->domain_info);
477 			if (match_pdomain->domain != pdomain->domain)
478 				continue;
479 
480 			/* Here i and j are in the same domain */
481 			if (match_pdomain->num_processors != count_target) {
482 				retval = -EFAULT;
483 				goto err_ret;
484 			}
485 
486 			if (pdomain->coord_type != match_pdomain->coord_type) {
487 				retval = -EFAULT;
488 				goto err_ret;
489 			}
490 
491 			cpumask_set_cpu(j, covered_cpus);
492 			cpumask_set_cpu(j, pr->shared_cpu_map);
493 		}
494 
495 		for_each_cpu(j, pr->shared_cpu_map) {
496 			if (i == j)
497 				continue;
498 
499 			match_pr = all_cpu_data[j];
500 			match_pr->shared_type = pr->shared_type;
501 			cpumask_copy(match_pr->shared_cpu_map,
502 				     pr->shared_cpu_map);
503 		}
504 	}
505 	goto out;
506 
507 err_ret:
508 	for_each_possible_cpu(i) {
509 		pr = all_cpu_data[i];
510 
511 		/* Assume no coordination on any error parsing domain info */
512 		cpumask_clear(pr->shared_cpu_map);
513 		cpumask_set_cpu(i, pr->shared_cpu_map);
514 		pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
515 	}
516 out:
517 	free_cpumask_var(covered_cpus);
518 	return retval;
519 }
520 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
521 
522 static int register_pcc_channel(int pcc_ss_idx)
523 {
524 	struct acpi_pcct_hw_reduced *cppc_ss;
525 	u64 usecs_lat;
526 
527 	if (pcc_ss_idx >= 0) {
528 		pcc_data[pcc_ss_idx]->pcc_channel =
529 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
530 
531 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
532 			pr_err("Failed to find PCC channel for subspace %d\n",
533 			       pcc_ss_idx);
534 			return -ENODEV;
535 		}
536 
537 		/*
538 		 * The PCC mailbox controller driver should
539 		 * have parsed the PCCT (global table of all
540 		 * PCC channels) and stored pointers to the
541 		 * subspace communication region in con_priv.
542 		 */
543 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
544 
545 		if (!cppc_ss) {
546 			pr_err("No PCC subspace found for %d CPPC\n",
547 			       pcc_ss_idx);
548 			return -ENODEV;
549 		}
550 
551 		/*
552 		 * cppc_ss->latency is just a Nominal value. In reality
553 		 * the remote processor could be much slower to reply.
554 		 * So add an arbitrary amount of wait on top of Nominal.
555 		 */
556 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
557 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
558 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
559 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
560 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
561 
562 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
563 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
564 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
565 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
566 			       pcc_ss_idx);
567 			return -ENOMEM;
568 		}
569 
570 		/* Set flag so that we don't come here for each CPU. */
571 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
572 	}
573 
574 	return 0;
575 }
576 
577 /**
578  * cpc_ffh_supported() - check if FFH reading supported
579  *
580  * Check if the architecture has support for functional fixed hardware
581  * read/write capability.
582  *
583  * Return: true for supported, false for not supported
584  */
585 bool __weak cpc_ffh_supported(void)
586 {
587 	return false;
588 }
589 
590 /**
591  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
592  *
593  * Check and allocate the cppc_pcc_data memory.
594  * In some processor configurations it is possible that same subspace
595  * is shared between multiple CPUs. This is seen especially in CPUs
596  * with hardware multi-threading support.
597  *
598  * Return: 0 for success, errno for failure
599  */
600 int pcc_data_alloc(int pcc_ss_id)
601 {
602 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
603 		return -EINVAL;
604 
605 	if (pcc_data[pcc_ss_id]) {
606 		pcc_data[pcc_ss_id]->refcount++;
607 	} else {
608 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
609 					      GFP_KERNEL);
610 		if (!pcc_data[pcc_ss_id])
611 			return -ENOMEM;
612 		pcc_data[pcc_ss_id]->refcount++;
613 	}
614 
615 	return 0;
616 }
617 
618 /* Check if CPPC revision + num_ent combination is supported */
619 static bool is_cppc_supported(int revision, int num_ent)
620 {
621 	int expected_num_ent;
622 
623 	switch (revision) {
624 	case CPPC_V2_REV:
625 		expected_num_ent = CPPC_V2_NUM_ENT;
626 		break;
627 	case CPPC_V3_REV:
628 		expected_num_ent = CPPC_V3_NUM_ENT;
629 		break;
630 	default:
631 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
632 			revision);
633 		return false;
634 	}
635 
636 	if (expected_num_ent != num_ent) {
637 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
638 			num_ent, expected_num_ent, revision);
639 		return false;
640 	}
641 
642 	return true;
643 }
644 
645 /*
646  * An example CPC table looks like the following.
647  *
648  *	Name(_CPC, Package()
649  *			{
650  *			17,
651  *			NumEntries
652  *			1,
653  *			// Revision
654  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
655  *			// Highest Performance
656  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
657  *			// Nominal Performance
658  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
659  *			// Lowest Nonlinear Performance
660  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
661  *			// Lowest Performance
662  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
663  *			// Guaranteed Performance Register
664  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
665  *			// Desired Performance Register
666  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
667  *			..
668  *			..
669  *			..
670  *
671  *		}
672  * Each Register() encodes how to access that specific register.
673  * e.g. a sample PCC entry has the following encoding:
674  *
675  *	Register (
676  *		PCC,
677  *		AddressSpaceKeyword
678  *		8,
679  *		//RegisterBitWidth
680  *		8,
681  *		//RegisterBitOffset
682  *		0x30,
683  *		//RegisterAddress
684  *		9
685  *		//AccessSize (subspace ID)
686  *		0
687  *		)
688  *	}
689  */
690 
691 /**
692  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
693  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
694  *
695  *	Return: 0 for success or negative value for err.
696  */
697 int acpi_cppc_processor_probe(struct acpi_processor *pr)
698 {
699 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
700 	union acpi_object *out_obj, *cpc_obj;
701 	struct cpc_desc *cpc_ptr;
702 	struct cpc_reg *gas_t;
703 	struct device *cpu_dev;
704 	acpi_handle handle = pr->handle;
705 	unsigned int num_ent, i, cpc_rev;
706 	int pcc_subspace_id = -1;
707 	acpi_status status;
708 	int ret = -EFAULT;
709 
710 	/* Parse the ACPI _CPC table for this CPU. */
711 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
712 			ACPI_TYPE_PACKAGE);
713 	if (ACPI_FAILURE(status)) {
714 		ret = -ENODEV;
715 		goto out_buf_free;
716 	}
717 
718 	out_obj = (union acpi_object *) output.pointer;
719 
720 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
721 	if (!cpc_ptr) {
722 		ret = -ENOMEM;
723 		goto out_buf_free;
724 	}
725 
726 	/* First entry is NumEntries. */
727 	cpc_obj = &out_obj->package.elements[0];
728 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
729 		num_ent = cpc_obj->integer.value;
730 	} else {
731 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
732 				cpc_obj->type);
733 		goto out_free;
734 	}
735 	cpc_ptr->num_entries = num_ent;
736 
737 	/* Second entry should be revision. */
738 	cpc_obj = &out_obj->package.elements[1];
739 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
740 		cpc_rev = cpc_obj->integer.value;
741 	} else {
742 		pr_debug("Unexpected entry type(%d) for Revision\n",
743 				cpc_obj->type);
744 		goto out_free;
745 	}
746 	cpc_ptr->version = cpc_rev;
747 
748 	if (!is_cppc_supported(cpc_rev, num_ent))
749 		goto out_free;
750 
751 	/* Iterate through remaining entries in _CPC */
752 	for (i = 2; i < num_ent; i++) {
753 		cpc_obj = &out_obj->package.elements[i];
754 
755 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
756 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
757 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
758 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
759 			gas_t = (struct cpc_reg *)
760 				cpc_obj->buffer.pointer;
761 
762 			/*
763 			 * The PCC Subspace index is encoded inside
764 			 * the CPC table entries. The same PCC index
765 			 * will be used for all the PCC entries,
766 			 * so extract it only once.
767 			 */
768 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
769 				if (pcc_subspace_id < 0) {
770 					pcc_subspace_id = gas_t->access_width;
771 					if (pcc_data_alloc(pcc_subspace_id))
772 						goto out_free;
773 				} else if (pcc_subspace_id != gas_t->access_width) {
774 					pr_debug("Mismatched PCC ids.\n");
775 					goto out_free;
776 				}
777 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
778 				if (gas_t->address) {
779 					void __iomem *addr;
780 
781 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
782 					if (!addr)
783 						goto out_free;
784 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
785 				}
786 			} else {
787 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
788 					/* Support only PCC ,SYS MEM and FFH type regs */
789 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
790 					goto out_free;
791 				}
792 			}
793 
794 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
795 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
796 		} else {
797 			pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
798 			goto out_free;
799 		}
800 	}
801 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
802 
803 	/*
804 	 * Initialize the remaining cpc_regs as unsupported.
805 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
806 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
807 	 */
808 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
809 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
810 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
811 	}
812 
813 
814 	/* Store CPU Logical ID */
815 	cpc_ptr->cpu_id = pr->id;
816 
817 	/* Parse PSD data for this CPU */
818 	ret = acpi_get_psd(cpc_ptr, handle);
819 	if (ret)
820 		goto out_free;
821 
822 	/* Register PCC channel once for all PCC subspace ID. */
823 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
824 		ret = register_pcc_channel(pcc_subspace_id);
825 		if (ret)
826 			goto out_free;
827 
828 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
829 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
830 	}
831 
832 	/* Everything looks okay */
833 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
834 
835 	/* Add per logical CPU nodes for reading its feedback counters. */
836 	cpu_dev = get_cpu_device(pr->id);
837 	if (!cpu_dev) {
838 		ret = -EINVAL;
839 		goto out_free;
840 	}
841 
842 	/* Plug PSD data into this CPU's CPC descriptor. */
843 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
844 
845 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
846 			"acpi_cppc");
847 	if (ret) {
848 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
849 		goto out_free;
850 	}
851 
852 	kfree(output.pointer);
853 	return 0;
854 
855 out_free:
856 	/* Free all the mapped sys mem areas for this CPU */
857 	for (i = 2; i < cpc_ptr->num_entries; i++) {
858 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
859 
860 		if (addr)
861 			iounmap(addr);
862 	}
863 	kfree(cpc_ptr);
864 
865 out_buf_free:
866 	kfree(output.pointer);
867 	return ret;
868 }
869 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
870 
871 /**
872  * acpi_cppc_processor_exit - Cleanup CPC structs.
873  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
874  *
875  * Return: Void
876  */
877 void acpi_cppc_processor_exit(struct acpi_processor *pr)
878 {
879 	struct cpc_desc *cpc_ptr;
880 	unsigned int i;
881 	void __iomem *addr;
882 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
883 
884 	if (pcc_ss_id >=0 && pcc_data[pcc_ss_id]) {
885 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
886 			pcc_data[pcc_ss_id]->refcount--;
887 			if (!pcc_data[pcc_ss_id]->refcount) {
888 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
889 				kfree(pcc_data[pcc_ss_id]);
890 				pcc_data[pcc_ss_id] = NULL;
891 			}
892 		}
893 	}
894 
895 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
896 	if (!cpc_ptr)
897 		return;
898 
899 	/* Free all the mapped sys mem areas for this CPU */
900 	for (i = 2; i < cpc_ptr->num_entries; i++) {
901 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
902 		if (addr)
903 			iounmap(addr);
904 	}
905 
906 	kobject_put(&cpc_ptr->kobj);
907 	kfree(cpc_ptr);
908 }
909 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
910 
911 /**
912  * cpc_read_ffh() - Read FFH register
913  * @cpunum:	CPU number to read
914  * @reg:	cppc register information
915  * @val:	place holder for return value
916  *
917  * Read bit_width bits from a specified address and bit_offset
918  *
919  * Return: 0 for success and error code
920  */
921 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
922 {
923 	return -ENOTSUPP;
924 }
925 
926 /**
927  * cpc_write_ffh() - Write FFH register
928  * @cpunum:	CPU number to write
929  * @reg:	cppc register information
930  * @val:	value to write
931  *
932  * Write value of bit_width bits to a specified address and bit_offset
933  *
934  * Return: 0 for success and error code
935  */
936 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
937 {
938 	return -ENOTSUPP;
939 }
940 
941 /*
942  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
943  * as fast as possible. We have already mapped the PCC subspace during init, so
944  * we can directly write to it.
945  */
946 
947 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
948 {
949 	int ret_val = 0;
950 	void __iomem *vaddr = 0;
951 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
952 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
953 
954 	if (reg_res->type == ACPI_TYPE_INTEGER) {
955 		*val = reg_res->cpc_entry.int_value;
956 		return ret_val;
957 	}
958 
959 	*val = 0;
960 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
961 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
962 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
963 		vaddr = reg_res->sys_mem_vaddr;
964 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
965 		return cpc_read_ffh(cpu, reg, val);
966 	else
967 		return acpi_os_read_memory((acpi_physical_address)reg->address,
968 				val, reg->bit_width);
969 
970 	switch (reg->bit_width) {
971 		case 8:
972 			*val = readb_relaxed(vaddr);
973 			break;
974 		case 16:
975 			*val = readw_relaxed(vaddr);
976 			break;
977 		case 32:
978 			*val = readl_relaxed(vaddr);
979 			break;
980 		case 64:
981 			*val = readq_relaxed(vaddr);
982 			break;
983 		default:
984 			pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
985 				 reg->bit_width, pcc_ss_id);
986 			ret_val = -EFAULT;
987 	}
988 
989 	return ret_val;
990 }
991 
992 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
993 {
994 	int ret_val = 0;
995 	void __iomem *vaddr = 0;
996 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
997 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
998 
999 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
1000 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
1001 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1002 		vaddr = reg_res->sys_mem_vaddr;
1003 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1004 		return cpc_write_ffh(cpu, reg, val);
1005 	else
1006 		return acpi_os_write_memory((acpi_physical_address)reg->address,
1007 				val, reg->bit_width);
1008 
1009 	switch (reg->bit_width) {
1010 		case 8:
1011 			writeb_relaxed(val, vaddr);
1012 			break;
1013 		case 16:
1014 			writew_relaxed(val, vaddr);
1015 			break;
1016 		case 32:
1017 			writel_relaxed(val, vaddr);
1018 			break;
1019 		case 64:
1020 			writeq_relaxed(val, vaddr);
1021 			break;
1022 		default:
1023 			pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1024 				 reg->bit_width, pcc_ss_id);
1025 			ret_val = -EFAULT;
1026 			break;
1027 	}
1028 
1029 	return ret_val;
1030 }
1031 
1032 /**
1033  * cppc_get_desired_perf - Get the value of desired performance register.
1034  * @cpunum: CPU from which to get desired performance.
1035  * @desired_perf: address of a variable to store the returned desired performance
1036  *
1037  * Return: 0 for success, -EIO otherwise.
1038  */
1039 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1040 {
1041 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1042 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1043 	struct cpc_register_resource *desired_reg;
1044 	struct cppc_pcc_data *pcc_ss_data = NULL;
1045 
1046 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1047 
1048 	if (CPC_IN_PCC(desired_reg)) {
1049 		int ret = 0;
1050 
1051 		if (pcc_ss_id < 0)
1052 			return -EIO;
1053 
1054 		pcc_ss_data = pcc_data[pcc_ss_id];
1055 
1056 		down_write(&pcc_ss_data->pcc_lock);
1057 
1058 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1059 			cpc_read(cpunum, desired_reg, desired_perf);
1060 		else
1061 			ret = -EIO;
1062 
1063 		up_write(&pcc_ss_data->pcc_lock);
1064 
1065 		return ret;
1066 	}
1067 
1068 	cpc_read(cpunum, desired_reg, desired_perf);
1069 
1070 	return 0;
1071 }
1072 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1073 
1074 /**
1075  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1076  * @cpunum: CPU from which to get capabilities info.
1077  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1078  *
1079  * Return: 0 for success with perf_caps populated else -ERRNO.
1080  */
1081 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1082 {
1083 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1084 	struct cpc_register_resource *highest_reg, *lowest_reg,
1085 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1086 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1087 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1088 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1089 	struct cppc_pcc_data *pcc_ss_data = NULL;
1090 	int ret = 0, regs_in_pcc = 0;
1091 
1092 	if (!cpc_desc) {
1093 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1094 		return -ENODEV;
1095 	}
1096 
1097 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1098 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1099 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1100 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1101 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1102 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1103 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1104 
1105 	/* Are any of the regs PCC ?*/
1106 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1107 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1108 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1109 		if (pcc_ss_id < 0) {
1110 			pr_debug("Invalid pcc_ss_id\n");
1111 			return -ENODEV;
1112 		}
1113 		pcc_ss_data = pcc_data[pcc_ss_id];
1114 		regs_in_pcc = 1;
1115 		down_write(&pcc_ss_data->pcc_lock);
1116 		/* Ring doorbell once to update PCC subspace */
1117 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1118 			ret = -EIO;
1119 			goto out_err;
1120 		}
1121 	}
1122 
1123 	cpc_read(cpunum, highest_reg, &high);
1124 	perf_caps->highest_perf = high;
1125 
1126 	cpc_read(cpunum, lowest_reg, &low);
1127 	perf_caps->lowest_perf = low;
1128 
1129 	cpc_read(cpunum, nominal_reg, &nom);
1130 	perf_caps->nominal_perf = nom;
1131 
1132 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1133 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1134 		perf_caps->guaranteed_perf = 0;
1135 	} else {
1136 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1137 		perf_caps->guaranteed_perf = guaranteed;
1138 	}
1139 
1140 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1141 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1142 
1143 	if (!high || !low || !nom || !min_nonlinear)
1144 		ret = -EFAULT;
1145 
1146 	/* Read optional lowest and nominal frequencies if present */
1147 	if (CPC_SUPPORTED(low_freq_reg))
1148 		cpc_read(cpunum, low_freq_reg, &low_f);
1149 
1150 	if (CPC_SUPPORTED(nom_freq_reg))
1151 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1152 
1153 	perf_caps->lowest_freq = low_f;
1154 	perf_caps->nominal_freq = nom_f;
1155 
1156 
1157 out_err:
1158 	if (regs_in_pcc)
1159 		up_write(&pcc_ss_data->pcc_lock);
1160 	return ret;
1161 }
1162 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1163 
1164 /**
1165  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1166  * @cpunum: CPU from which to read counters.
1167  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1168  *
1169  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1170  */
1171 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1172 {
1173 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1174 	struct cpc_register_resource *delivered_reg, *reference_reg,
1175 		*ref_perf_reg, *ctr_wrap_reg;
1176 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1177 	struct cppc_pcc_data *pcc_ss_data = NULL;
1178 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1179 	int ret = 0, regs_in_pcc = 0;
1180 
1181 	if (!cpc_desc) {
1182 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1183 		return -ENODEV;
1184 	}
1185 
1186 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1187 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1188 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1189 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1190 
1191 	/*
1192 	 * If reference perf register is not supported then we should
1193 	 * use the nominal perf value
1194 	 */
1195 	if (!CPC_SUPPORTED(ref_perf_reg))
1196 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1197 
1198 	/* Are any of the regs PCC ?*/
1199 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1200 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1201 		if (pcc_ss_id < 0) {
1202 			pr_debug("Invalid pcc_ss_id\n");
1203 			return -ENODEV;
1204 		}
1205 		pcc_ss_data = pcc_data[pcc_ss_id];
1206 		down_write(&pcc_ss_data->pcc_lock);
1207 		regs_in_pcc = 1;
1208 		/* Ring doorbell once to update PCC subspace */
1209 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1210 			ret = -EIO;
1211 			goto out_err;
1212 		}
1213 	}
1214 
1215 	cpc_read(cpunum, delivered_reg, &delivered);
1216 	cpc_read(cpunum, reference_reg, &reference);
1217 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1218 
1219 	/*
1220 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1221 	 * performance counters are assumed to never wrap during the lifetime of
1222 	 * platform
1223 	 */
1224 	ctr_wrap_time = (u64)(~((u64)0));
1225 	if (CPC_SUPPORTED(ctr_wrap_reg))
1226 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1227 
1228 	if (!delivered || !reference ||	!ref_perf) {
1229 		ret = -EFAULT;
1230 		goto out_err;
1231 	}
1232 
1233 	perf_fb_ctrs->delivered = delivered;
1234 	perf_fb_ctrs->reference = reference;
1235 	perf_fb_ctrs->reference_perf = ref_perf;
1236 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1237 out_err:
1238 	if (regs_in_pcc)
1239 		up_write(&pcc_ss_data->pcc_lock);
1240 	return ret;
1241 }
1242 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1243 
1244 /**
1245  * cppc_set_perf - Set a CPU's performance controls.
1246  * @cpu: CPU for which to set performance controls.
1247  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1248  *
1249  * Return: 0 for success, -ERRNO otherwise.
1250  */
1251 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1252 {
1253 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1254 	struct cpc_register_resource *desired_reg;
1255 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1256 	struct cppc_pcc_data *pcc_ss_data = NULL;
1257 	int ret = 0;
1258 
1259 	if (!cpc_desc) {
1260 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1261 		return -ENODEV;
1262 	}
1263 
1264 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1265 
1266 	/*
1267 	 * This is Phase-I where we want to write to CPC registers
1268 	 * -> We want all CPUs to be able to execute this phase in parallel
1269 	 *
1270 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1271 	 * achieve that goal here
1272 	 */
1273 	if (CPC_IN_PCC(desired_reg)) {
1274 		if (pcc_ss_id < 0) {
1275 			pr_debug("Invalid pcc_ss_id\n");
1276 			return -ENODEV;
1277 		}
1278 		pcc_ss_data = pcc_data[pcc_ss_id];
1279 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1280 		if (pcc_ss_data->platform_owns_pcc) {
1281 			ret = check_pcc_chan(pcc_ss_id, false);
1282 			if (ret) {
1283 				up_read(&pcc_ss_data->pcc_lock);
1284 				return ret;
1285 			}
1286 		}
1287 		/*
1288 		 * Update the pending_write to make sure a PCC CMD_READ will not
1289 		 * arrive and steal the channel during the switch to write lock
1290 		 */
1291 		pcc_ss_data->pending_pcc_write_cmd = true;
1292 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1293 		cpc_desc->write_cmd_status = 0;
1294 	}
1295 
1296 	/*
1297 	 * Skip writing MIN/MAX until Linux knows how to come up with
1298 	 * useful values.
1299 	 */
1300 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1301 
1302 	if (CPC_IN_PCC(desired_reg))
1303 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1304 	/*
1305 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1306 	 *
1307 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1308 	 * requests that happened in short overlapping interval. The last CPU to
1309 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1310 	 *
1311 	 * We have the following requirements for Phase-II:
1312 	 *     1. We want to execute Phase-II only when there are no CPUs
1313 	 * currently executing in Phase-I
1314 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1315 	 * entering Phase-I.
1316 	 *     3. We want only one CPU among all those who went through Phase-I
1317 	 * to run phase-II
1318 	 *
1319 	 * If write_trylock fails to get the lock and doesn't transfer the
1320 	 * PCC ownership to the platform, then one of the following will be TRUE
1321 	 *     1. There is at-least one CPU in Phase-I which will later execute
1322 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1323 	 * executing the Phase-II.
1324 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1325 	 * write_trylock and has already acquired the write_lock. We know for a
1326 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1327 	 * before this CPU's Phase-I as we held the read_lock.
1328 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1329 	 * down_write, in which case, send_pcc_cmd will check for pending
1330 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1331 	 * So this CPU can be certain that its request will be delivered
1332 	 *    So in all cases, this CPU knows that its request will be delivered
1333 	 * by another CPU and can return
1334 	 *
1335 	 * After getting the down_write we still need to check for
1336 	 * pending_pcc_write_cmd to take care of the following scenario
1337 	 *    The thread running this code could be scheduled out between
1338 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1339 	 * could have delivered the request to Platform by triggering the
1340 	 * doorbell and transferred the ownership of PCC to platform. So this
1341 	 * avoids triggering an unnecessary doorbell and more importantly before
1342 	 * triggering the doorbell it makes sure that the PCC channel ownership
1343 	 * is still with OSPM.
1344 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1345 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1346 	 * before the pcc CMD_WRITE is completed. pcc_send_cmd checks for this
1347 	 * case during a CMD_READ and if there are pending writes it delivers
1348 	 * the write command before servicing the read command
1349 	 */
1350 	if (CPC_IN_PCC(desired_reg)) {
1351 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1352 			/* Update only if there are pending write commands */
1353 			if (pcc_ss_data->pending_pcc_write_cmd)
1354 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1355 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1356 		} else
1357 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1358 			wait_event(pcc_ss_data->pcc_write_wait_q,
1359 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1360 
1361 		/* send_pcc_cmd updates the status in case of failure */
1362 		ret = cpc_desc->write_cmd_status;
1363 	}
1364 	return ret;
1365 }
1366 EXPORT_SYMBOL_GPL(cppc_set_perf);
1367 
1368 /**
1369  * cppc_get_transition_latency - returns frequency transition latency in ns
1370  *
1371  * ACPI CPPC does not explicitly specifiy how a platform can specify the
1372  * transition latency for perfromance change requests. The closest we have
1373  * is the timing information from the PCCT tables which provides the info
1374  * on the number and frequency of PCC commands the platform can handle.
1375  */
1376 unsigned int cppc_get_transition_latency(int cpu_num)
1377 {
1378 	/*
1379 	 * Expected transition latency is based on the PCCT timing values
1380 	 * Below are definition from ACPI spec:
1381 	 * pcc_nominal- Expected latency to process a command, in microseconds
1382 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1383 	 *              channel can support, reported in commands per minute. 0
1384 	 *              indicates no limitation.
1385 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1386 	 *              completion of a command before issuing the next command,
1387 	 *              in microseconds.
1388 	 */
1389 	unsigned int latency_ns = 0;
1390 	struct cpc_desc *cpc_desc;
1391 	struct cpc_register_resource *desired_reg;
1392 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1393 	struct cppc_pcc_data *pcc_ss_data;
1394 
1395 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1396 	if (!cpc_desc)
1397 		return CPUFREQ_ETERNAL;
1398 
1399 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1400 	if (!CPC_IN_PCC(desired_reg))
1401 		return CPUFREQ_ETERNAL;
1402 
1403 	if (pcc_ss_id < 0)
1404 		return CPUFREQ_ETERNAL;
1405 
1406 	pcc_ss_data = pcc_data[pcc_ss_id];
1407 	if (pcc_ss_data->pcc_mpar)
1408 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1409 
1410 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1411 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1412 
1413 	return latency_ns;
1414 }
1415 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1416