xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision b830f94f)
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, &buffer,
369 			ACPI_TYPE_PACKAGE);
370 	if (ACPI_FAILURE(status))
371 		return -ENODEV;
372 
373 	psd = buffer.pointer;
374 	if (!psd || psd->package.count != 1) {
375 		pr_debug("Invalid _PSD data\n");
376 		goto end;
377 	}
378 
379 	pdomain = &(cpc_ptr->domain_info);
380 
381 	state.length = sizeof(struct acpi_psd_package);
382 	state.pointer = pdomain;
383 
384 	status = acpi_extract_package(&(psd->package.elements[0]),
385 		&format, &state);
386 	if (ACPI_FAILURE(status)) {
387 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
388 		goto end;
389 	}
390 
391 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
392 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
393 		goto end;
394 	}
395 
396 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
397 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
398 		goto end;
399 	}
400 
401 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
402 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
403 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
404 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
405 		goto end;
406 	}
407 
408 	result = 0;
409 end:
410 	kfree(buffer.pointer);
411 	return result;
412 }
413 
414 /**
415  * acpi_get_psd_map - Map the CPUs in a common freq domain.
416  * @all_cpu_data: Ptrs to CPU specific CPPC data including PSD info.
417  *
418  *	Return: 0 for success or negative value for err.
419  */
420 int acpi_get_psd_map(struct cppc_cpudata **all_cpu_data)
421 {
422 	int count_target;
423 	int retval = 0;
424 	unsigned int i, j;
425 	cpumask_var_t covered_cpus;
426 	struct cppc_cpudata *pr, *match_pr;
427 	struct acpi_psd_package *pdomain;
428 	struct acpi_psd_package *match_pdomain;
429 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
430 
431 	if (!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))
432 		return -ENOMEM;
433 
434 	/*
435 	 * Now that we have _PSD data from all CPUs, let's setup P-state
436 	 * domain info.
437 	 */
438 	for_each_possible_cpu(i) {
439 		pr = all_cpu_data[i];
440 		if (!pr)
441 			continue;
442 
443 		if (cpumask_test_cpu(i, covered_cpus))
444 			continue;
445 
446 		cpc_ptr = per_cpu(cpc_desc_ptr, i);
447 		if (!cpc_ptr) {
448 			retval = -EFAULT;
449 			goto err_ret;
450 		}
451 
452 		pdomain = &(cpc_ptr->domain_info);
453 		cpumask_set_cpu(i, pr->shared_cpu_map);
454 		cpumask_set_cpu(i, covered_cpus);
455 		if (pdomain->num_processors <= 1)
456 			continue;
457 
458 		/* Validate the Domain info */
459 		count_target = pdomain->num_processors;
460 		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
461 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
462 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
463 			pr->shared_type = CPUFREQ_SHARED_TYPE_HW;
464 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
465 			pr->shared_type = CPUFREQ_SHARED_TYPE_ANY;
466 
467 		for_each_possible_cpu(j) {
468 			if (i == j)
469 				continue;
470 
471 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
472 			if (!match_cpc_ptr) {
473 				retval = -EFAULT;
474 				goto err_ret;
475 			}
476 
477 			match_pdomain = &(match_cpc_ptr->domain_info);
478 			if (match_pdomain->domain != pdomain->domain)
479 				continue;
480 
481 			/* Here i and j are in the same domain */
482 			if (match_pdomain->num_processors != count_target) {
483 				retval = -EFAULT;
484 				goto err_ret;
485 			}
486 
487 			if (pdomain->coord_type != match_pdomain->coord_type) {
488 				retval = -EFAULT;
489 				goto err_ret;
490 			}
491 
492 			cpumask_set_cpu(j, covered_cpus);
493 			cpumask_set_cpu(j, pr->shared_cpu_map);
494 		}
495 
496 		for_each_possible_cpu(j) {
497 			if (i == j)
498 				continue;
499 
500 			match_pr = all_cpu_data[j];
501 			if (!match_pr)
502 				continue;
503 
504 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
505 			if (!match_cpc_ptr) {
506 				retval = -EFAULT;
507 				goto err_ret;
508 			}
509 
510 			match_pdomain = &(match_cpc_ptr->domain_info);
511 			if (match_pdomain->domain != pdomain->domain)
512 				continue;
513 
514 			match_pr->shared_type = pr->shared_type;
515 			cpumask_copy(match_pr->shared_cpu_map,
516 				     pr->shared_cpu_map);
517 		}
518 	}
519 
520 err_ret:
521 	for_each_possible_cpu(i) {
522 		pr = all_cpu_data[i];
523 		if (!pr)
524 			continue;
525 
526 		/* Assume no coordination on any error parsing domain info */
527 		if (retval) {
528 			cpumask_clear(pr->shared_cpu_map);
529 			cpumask_set_cpu(i, pr->shared_cpu_map);
530 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
531 		}
532 	}
533 
534 	free_cpumask_var(covered_cpus);
535 	return retval;
536 }
537 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
538 
539 static int register_pcc_channel(int pcc_ss_idx)
540 {
541 	struct acpi_pcct_hw_reduced *cppc_ss;
542 	u64 usecs_lat;
543 
544 	if (pcc_ss_idx >= 0) {
545 		pcc_data[pcc_ss_idx]->pcc_channel =
546 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
547 
548 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
549 			pr_err("Failed to find PCC channel for subspace %d\n",
550 			       pcc_ss_idx);
551 			return -ENODEV;
552 		}
553 
554 		/*
555 		 * The PCC mailbox controller driver should
556 		 * have parsed the PCCT (global table of all
557 		 * PCC channels) and stored pointers to the
558 		 * subspace communication region in con_priv.
559 		 */
560 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
561 
562 		if (!cppc_ss) {
563 			pr_err("No PCC subspace found for %d CPPC\n",
564 			       pcc_ss_idx);
565 			return -ENODEV;
566 		}
567 
568 		/*
569 		 * cppc_ss->latency is just a Nominal value. In reality
570 		 * the remote processor could be much slower to reply.
571 		 * So add an arbitrary amount of wait on top of Nominal.
572 		 */
573 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
574 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
575 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
576 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
577 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
578 
579 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
580 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
581 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
582 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
583 			       pcc_ss_idx);
584 			return -ENOMEM;
585 		}
586 
587 		/* Set flag so that we don't come here for each CPU. */
588 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
589 	}
590 
591 	return 0;
592 }
593 
594 /**
595  * cpc_ffh_supported() - check if FFH reading supported
596  *
597  * Check if the architecture has support for functional fixed hardware
598  * read/write capability.
599  *
600  * Return: true for supported, false for not supported
601  */
602 bool __weak cpc_ffh_supported(void)
603 {
604 	return false;
605 }
606 
607 /**
608  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
609  *
610  * Check and allocate the cppc_pcc_data memory.
611  * In some processor configurations it is possible that same subspace
612  * is shared between multiple CPUs. This is seen especially in CPUs
613  * with hardware multi-threading support.
614  *
615  * Return: 0 for success, errno for failure
616  */
617 int pcc_data_alloc(int pcc_ss_id)
618 {
619 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
620 		return -EINVAL;
621 
622 	if (pcc_data[pcc_ss_id]) {
623 		pcc_data[pcc_ss_id]->refcount++;
624 	} else {
625 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
626 					      GFP_KERNEL);
627 		if (!pcc_data[pcc_ss_id])
628 			return -ENOMEM;
629 		pcc_data[pcc_ss_id]->refcount++;
630 	}
631 
632 	return 0;
633 }
634 
635 /* Check if CPPC revision + num_ent combination is supported */
636 static bool is_cppc_supported(int revision, int num_ent)
637 {
638 	int expected_num_ent;
639 
640 	switch (revision) {
641 	case CPPC_V2_REV:
642 		expected_num_ent = CPPC_V2_NUM_ENT;
643 		break;
644 	case CPPC_V3_REV:
645 		expected_num_ent = CPPC_V3_NUM_ENT;
646 		break;
647 	default:
648 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
649 			revision);
650 		return false;
651 	}
652 
653 	if (expected_num_ent != num_ent) {
654 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
655 			num_ent, expected_num_ent, revision);
656 		return false;
657 	}
658 
659 	return true;
660 }
661 
662 /*
663  * An example CPC table looks like the following.
664  *
665  *	Name(_CPC, Package()
666  *			{
667  *			17,
668  *			NumEntries
669  *			1,
670  *			// Revision
671  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
672  *			// Highest Performance
673  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
674  *			// Nominal Performance
675  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
676  *			// Lowest Nonlinear Performance
677  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
678  *			// Lowest Performance
679  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
680  *			// Guaranteed Performance Register
681  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
682  *			// Desired Performance Register
683  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
684  *			..
685  *			..
686  *			..
687  *
688  *		}
689  * Each Register() encodes how to access that specific register.
690  * e.g. a sample PCC entry has the following encoding:
691  *
692  *	Register (
693  *		PCC,
694  *		AddressSpaceKeyword
695  *		8,
696  *		//RegisterBitWidth
697  *		8,
698  *		//RegisterBitOffset
699  *		0x30,
700  *		//RegisterAddress
701  *		9
702  *		//AccessSize (subspace ID)
703  *		0
704  *		)
705  *	}
706  */
707 
708 /**
709  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
710  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
711  *
712  *	Return: 0 for success or negative value for err.
713  */
714 int acpi_cppc_processor_probe(struct acpi_processor *pr)
715 {
716 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
717 	union acpi_object *out_obj, *cpc_obj;
718 	struct cpc_desc *cpc_ptr;
719 	struct cpc_reg *gas_t;
720 	struct device *cpu_dev;
721 	acpi_handle handle = pr->handle;
722 	unsigned int num_ent, i, cpc_rev;
723 	int pcc_subspace_id = -1;
724 	acpi_status status;
725 	int ret = -EFAULT;
726 
727 	/* Parse the ACPI _CPC table for this CPU. */
728 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
729 			ACPI_TYPE_PACKAGE);
730 	if (ACPI_FAILURE(status)) {
731 		ret = -ENODEV;
732 		goto out_buf_free;
733 	}
734 
735 	out_obj = (union acpi_object *) output.pointer;
736 
737 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
738 	if (!cpc_ptr) {
739 		ret = -ENOMEM;
740 		goto out_buf_free;
741 	}
742 
743 	/* First entry is NumEntries. */
744 	cpc_obj = &out_obj->package.elements[0];
745 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
746 		num_ent = cpc_obj->integer.value;
747 	} else {
748 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
749 				cpc_obj->type);
750 		goto out_free;
751 	}
752 	cpc_ptr->num_entries = num_ent;
753 
754 	/* Second entry should be revision. */
755 	cpc_obj = &out_obj->package.elements[1];
756 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
757 		cpc_rev = cpc_obj->integer.value;
758 	} else {
759 		pr_debug("Unexpected entry type(%d) for Revision\n",
760 				cpc_obj->type);
761 		goto out_free;
762 	}
763 	cpc_ptr->version = cpc_rev;
764 
765 	if (!is_cppc_supported(cpc_rev, num_ent))
766 		goto out_free;
767 
768 	/* Iterate through remaining entries in _CPC */
769 	for (i = 2; i < num_ent; i++) {
770 		cpc_obj = &out_obj->package.elements[i];
771 
772 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
773 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
774 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
775 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
776 			gas_t = (struct cpc_reg *)
777 				cpc_obj->buffer.pointer;
778 
779 			/*
780 			 * The PCC Subspace index is encoded inside
781 			 * the CPC table entries. The same PCC index
782 			 * will be used for all the PCC entries,
783 			 * so extract it only once.
784 			 */
785 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
786 				if (pcc_subspace_id < 0) {
787 					pcc_subspace_id = gas_t->access_width;
788 					if (pcc_data_alloc(pcc_subspace_id))
789 						goto out_free;
790 				} else if (pcc_subspace_id != gas_t->access_width) {
791 					pr_debug("Mismatched PCC ids.\n");
792 					goto out_free;
793 				}
794 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
795 				if (gas_t->address) {
796 					void __iomem *addr;
797 
798 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
799 					if (!addr)
800 						goto out_free;
801 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
802 				}
803 			} else {
804 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
805 					/* Support only PCC ,SYS MEM and FFH type regs */
806 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
807 					goto out_free;
808 				}
809 			}
810 
811 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
812 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
813 		} else {
814 			pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
815 			goto out_free;
816 		}
817 	}
818 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
819 
820 	/*
821 	 * Initialize the remaining cpc_regs as unsupported.
822 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
823 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
824 	 */
825 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
826 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
827 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
828 	}
829 
830 
831 	/* Store CPU Logical ID */
832 	cpc_ptr->cpu_id = pr->id;
833 
834 	/* Parse PSD data for this CPU */
835 	ret = acpi_get_psd(cpc_ptr, handle);
836 	if (ret)
837 		goto out_free;
838 
839 	/* Register PCC channel once for all PCC subspace ID. */
840 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
841 		ret = register_pcc_channel(pcc_subspace_id);
842 		if (ret)
843 			goto out_free;
844 
845 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
846 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
847 	}
848 
849 	/* Everything looks okay */
850 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
851 
852 	/* Add per logical CPU nodes for reading its feedback counters. */
853 	cpu_dev = get_cpu_device(pr->id);
854 	if (!cpu_dev) {
855 		ret = -EINVAL;
856 		goto out_free;
857 	}
858 
859 	/* Plug PSD data into this CPU's CPC descriptor. */
860 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
861 
862 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
863 			"acpi_cppc");
864 	if (ret) {
865 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
866 		goto out_free;
867 	}
868 
869 	kfree(output.pointer);
870 	return 0;
871 
872 out_free:
873 	/* Free all the mapped sys mem areas for this CPU */
874 	for (i = 2; i < cpc_ptr->num_entries; i++) {
875 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
876 
877 		if (addr)
878 			iounmap(addr);
879 	}
880 	kfree(cpc_ptr);
881 
882 out_buf_free:
883 	kfree(output.pointer);
884 	return ret;
885 }
886 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
887 
888 /**
889  * acpi_cppc_processor_exit - Cleanup CPC structs.
890  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
891  *
892  * Return: Void
893  */
894 void acpi_cppc_processor_exit(struct acpi_processor *pr)
895 {
896 	struct cpc_desc *cpc_ptr;
897 	unsigned int i;
898 	void __iomem *addr;
899 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
900 
901 	if (pcc_ss_id >=0 && pcc_data[pcc_ss_id]) {
902 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
903 			pcc_data[pcc_ss_id]->refcount--;
904 			if (!pcc_data[pcc_ss_id]->refcount) {
905 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
906 				pcc_data[pcc_ss_id]->pcc_channel_acquired = 0;
907 				kfree(pcc_data[pcc_ss_id]);
908 			}
909 		}
910 	}
911 
912 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
913 	if (!cpc_ptr)
914 		return;
915 
916 	/* Free all the mapped sys mem areas for this CPU */
917 	for (i = 2; i < cpc_ptr->num_entries; i++) {
918 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
919 		if (addr)
920 			iounmap(addr);
921 	}
922 
923 	kobject_put(&cpc_ptr->kobj);
924 	kfree(cpc_ptr);
925 }
926 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
927 
928 /**
929  * cpc_read_ffh() - Read FFH register
930  * @cpunum:	CPU number to read
931  * @reg:	cppc register information
932  * @val:	place holder for return value
933  *
934  * Read bit_width bits from a specified address and bit_offset
935  *
936  * Return: 0 for success and error code
937  */
938 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
939 {
940 	return -ENOTSUPP;
941 }
942 
943 /**
944  * cpc_write_ffh() - Write FFH register
945  * @cpunum:	CPU number to write
946  * @reg:	cppc register information
947  * @val:	value to write
948  *
949  * Write value of bit_width bits to a specified address and bit_offset
950  *
951  * Return: 0 for success and error code
952  */
953 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
954 {
955 	return -ENOTSUPP;
956 }
957 
958 /*
959  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
960  * as fast as possible. We have already mapped the PCC subspace during init, so
961  * we can directly write to it.
962  */
963 
964 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
965 {
966 	int ret_val = 0;
967 	void __iomem *vaddr = 0;
968 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
969 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
970 
971 	if (reg_res->type == ACPI_TYPE_INTEGER) {
972 		*val = reg_res->cpc_entry.int_value;
973 		return ret_val;
974 	}
975 
976 	*val = 0;
977 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
978 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
979 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
980 		vaddr = reg_res->sys_mem_vaddr;
981 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
982 		return cpc_read_ffh(cpu, reg, val);
983 	else
984 		return acpi_os_read_memory((acpi_physical_address)reg->address,
985 				val, reg->bit_width);
986 
987 	switch (reg->bit_width) {
988 		case 8:
989 			*val = readb_relaxed(vaddr);
990 			break;
991 		case 16:
992 			*val = readw_relaxed(vaddr);
993 			break;
994 		case 32:
995 			*val = readl_relaxed(vaddr);
996 			break;
997 		case 64:
998 			*val = readq_relaxed(vaddr);
999 			break;
1000 		default:
1001 			pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
1002 				 reg->bit_width, pcc_ss_id);
1003 			ret_val = -EFAULT;
1004 	}
1005 
1006 	return ret_val;
1007 }
1008 
1009 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
1010 {
1011 	int ret_val = 0;
1012 	void __iomem *vaddr = 0;
1013 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1014 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
1015 
1016 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
1017 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
1018 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1019 		vaddr = reg_res->sys_mem_vaddr;
1020 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1021 		return cpc_write_ffh(cpu, reg, val);
1022 	else
1023 		return acpi_os_write_memory((acpi_physical_address)reg->address,
1024 				val, reg->bit_width);
1025 
1026 	switch (reg->bit_width) {
1027 		case 8:
1028 			writeb_relaxed(val, vaddr);
1029 			break;
1030 		case 16:
1031 			writew_relaxed(val, vaddr);
1032 			break;
1033 		case 32:
1034 			writel_relaxed(val, vaddr);
1035 			break;
1036 		case 64:
1037 			writeq_relaxed(val, vaddr);
1038 			break;
1039 		default:
1040 			pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1041 				 reg->bit_width, pcc_ss_id);
1042 			ret_val = -EFAULT;
1043 			break;
1044 	}
1045 
1046 	return ret_val;
1047 }
1048 
1049 /**
1050  * cppc_get_desired_perf - Get the value of desired performance register.
1051  * @cpunum: CPU from which to get desired performance.
1052  * @desired_perf: address of a variable to store the returned desired performance
1053  *
1054  * Return: 0 for success, -EIO otherwise.
1055  */
1056 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1057 {
1058 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1059 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1060 	struct cpc_register_resource *desired_reg;
1061 	struct cppc_pcc_data *pcc_ss_data = NULL;
1062 
1063 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1064 
1065 	if (CPC_IN_PCC(desired_reg)) {
1066 		int ret = 0;
1067 
1068 		if (pcc_ss_id < 0)
1069 			return -EIO;
1070 
1071 		pcc_ss_data = pcc_data[pcc_ss_id];
1072 
1073 		down_write(&pcc_ss_data->pcc_lock);
1074 
1075 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1076 			cpc_read(cpunum, desired_reg, desired_perf);
1077 		else
1078 			ret = -EIO;
1079 
1080 		up_write(&pcc_ss_data->pcc_lock);
1081 
1082 		return ret;
1083 	}
1084 
1085 	cpc_read(cpunum, desired_reg, desired_perf);
1086 
1087 	return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1090 
1091 /**
1092  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1093  * @cpunum: CPU from which to get capabilities info.
1094  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1095  *
1096  * Return: 0 for success with perf_caps populated else -ERRNO.
1097  */
1098 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1099 {
1100 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1101 	struct cpc_register_resource *highest_reg, *lowest_reg,
1102 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1103 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1104 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1105 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1106 	struct cppc_pcc_data *pcc_ss_data = NULL;
1107 	int ret = 0, regs_in_pcc = 0;
1108 
1109 	if (!cpc_desc) {
1110 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1111 		return -ENODEV;
1112 	}
1113 
1114 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1115 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1116 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1117 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1118 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1119 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1120 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1121 
1122 	/* Are any of the regs PCC ?*/
1123 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1124 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1125 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1126 		if (pcc_ss_id < 0) {
1127 			pr_debug("Invalid pcc_ss_id\n");
1128 			return -ENODEV;
1129 		}
1130 		pcc_ss_data = pcc_data[pcc_ss_id];
1131 		regs_in_pcc = 1;
1132 		down_write(&pcc_ss_data->pcc_lock);
1133 		/* Ring doorbell once to update PCC subspace */
1134 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1135 			ret = -EIO;
1136 			goto out_err;
1137 		}
1138 	}
1139 
1140 	cpc_read(cpunum, highest_reg, &high);
1141 	perf_caps->highest_perf = high;
1142 
1143 	cpc_read(cpunum, lowest_reg, &low);
1144 	perf_caps->lowest_perf = low;
1145 
1146 	cpc_read(cpunum, nominal_reg, &nom);
1147 	perf_caps->nominal_perf = nom;
1148 
1149 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1150 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1151 		perf_caps->guaranteed_perf = 0;
1152 	} else {
1153 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1154 		perf_caps->guaranteed_perf = guaranteed;
1155 	}
1156 
1157 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1158 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1159 
1160 	if (!high || !low || !nom || !min_nonlinear)
1161 		ret = -EFAULT;
1162 
1163 	/* Read optional lowest and nominal frequencies if present */
1164 	if (CPC_SUPPORTED(low_freq_reg))
1165 		cpc_read(cpunum, low_freq_reg, &low_f);
1166 
1167 	if (CPC_SUPPORTED(nom_freq_reg))
1168 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1169 
1170 	perf_caps->lowest_freq = low_f;
1171 	perf_caps->nominal_freq = nom_f;
1172 
1173 
1174 out_err:
1175 	if (regs_in_pcc)
1176 		up_write(&pcc_ss_data->pcc_lock);
1177 	return ret;
1178 }
1179 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1180 
1181 /**
1182  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1183  * @cpunum: CPU from which to read counters.
1184  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1185  *
1186  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1187  */
1188 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1189 {
1190 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1191 	struct cpc_register_resource *delivered_reg, *reference_reg,
1192 		*ref_perf_reg, *ctr_wrap_reg;
1193 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1194 	struct cppc_pcc_data *pcc_ss_data = NULL;
1195 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1196 	int ret = 0, regs_in_pcc = 0;
1197 
1198 	if (!cpc_desc) {
1199 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1200 		return -ENODEV;
1201 	}
1202 
1203 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1204 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1205 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1206 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1207 
1208 	/*
1209 	 * If reference perf register is not supported then we should
1210 	 * use the nominal perf value
1211 	 */
1212 	if (!CPC_SUPPORTED(ref_perf_reg))
1213 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1214 
1215 	/* Are any of the regs PCC ?*/
1216 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1217 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1218 		if (pcc_ss_id < 0) {
1219 			pr_debug("Invalid pcc_ss_id\n");
1220 			return -ENODEV;
1221 		}
1222 		pcc_ss_data = pcc_data[pcc_ss_id];
1223 		down_write(&pcc_ss_data->pcc_lock);
1224 		regs_in_pcc = 1;
1225 		/* Ring doorbell once to update PCC subspace */
1226 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1227 			ret = -EIO;
1228 			goto out_err;
1229 		}
1230 	}
1231 
1232 	cpc_read(cpunum, delivered_reg, &delivered);
1233 	cpc_read(cpunum, reference_reg, &reference);
1234 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1235 
1236 	/*
1237 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1238 	 * performance counters are assumed to never wrap during the lifetime of
1239 	 * platform
1240 	 */
1241 	ctr_wrap_time = (u64)(~((u64)0));
1242 	if (CPC_SUPPORTED(ctr_wrap_reg))
1243 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1244 
1245 	if (!delivered || !reference ||	!ref_perf) {
1246 		ret = -EFAULT;
1247 		goto out_err;
1248 	}
1249 
1250 	perf_fb_ctrs->delivered = delivered;
1251 	perf_fb_ctrs->reference = reference;
1252 	perf_fb_ctrs->reference_perf = ref_perf;
1253 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1254 out_err:
1255 	if (regs_in_pcc)
1256 		up_write(&pcc_ss_data->pcc_lock);
1257 	return ret;
1258 }
1259 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1260 
1261 /**
1262  * cppc_set_perf - Set a CPU's performance controls.
1263  * @cpu: CPU for which to set performance controls.
1264  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1265  *
1266  * Return: 0 for success, -ERRNO otherwise.
1267  */
1268 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1269 {
1270 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1271 	struct cpc_register_resource *desired_reg;
1272 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1273 	struct cppc_pcc_data *pcc_ss_data = NULL;
1274 	int ret = 0;
1275 
1276 	if (!cpc_desc) {
1277 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1278 		return -ENODEV;
1279 	}
1280 
1281 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1282 
1283 	/*
1284 	 * This is Phase-I where we want to write to CPC registers
1285 	 * -> We want all CPUs to be able to execute this phase in parallel
1286 	 *
1287 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1288 	 * achieve that goal here
1289 	 */
1290 	if (CPC_IN_PCC(desired_reg)) {
1291 		if (pcc_ss_id < 0) {
1292 			pr_debug("Invalid pcc_ss_id\n");
1293 			return -ENODEV;
1294 		}
1295 		pcc_ss_data = pcc_data[pcc_ss_id];
1296 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1297 		if (pcc_ss_data->platform_owns_pcc) {
1298 			ret = check_pcc_chan(pcc_ss_id, false);
1299 			if (ret) {
1300 				up_read(&pcc_ss_data->pcc_lock);
1301 				return ret;
1302 			}
1303 		}
1304 		/*
1305 		 * Update the pending_write to make sure a PCC CMD_READ will not
1306 		 * arrive and steal the channel during the switch to write lock
1307 		 */
1308 		pcc_ss_data->pending_pcc_write_cmd = true;
1309 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1310 		cpc_desc->write_cmd_status = 0;
1311 	}
1312 
1313 	/*
1314 	 * Skip writing MIN/MAX until Linux knows how to come up with
1315 	 * useful values.
1316 	 */
1317 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1318 
1319 	if (CPC_IN_PCC(desired_reg))
1320 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1321 	/*
1322 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1323 	 *
1324 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1325 	 * requests that happened in short overlapping interval. The last CPU to
1326 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1327 	 *
1328 	 * We have the following requirements for Phase-II:
1329 	 *     1. We want to execute Phase-II only when there are no CPUs
1330 	 * currently executing in Phase-I
1331 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1332 	 * entering Phase-I.
1333 	 *     3. We want only one CPU among all those who went through Phase-I
1334 	 * to run phase-II
1335 	 *
1336 	 * If write_trylock fails to get the lock and doesn't transfer the
1337 	 * PCC ownership to the platform, then one of the following will be TRUE
1338 	 *     1. There is at-least one CPU in Phase-I which will later execute
1339 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1340 	 * executing the Phase-II.
1341 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1342 	 * write_trylock and has already acquired the write_lock. We know for a
1343 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1344 	 * before this CPU's Phase-I as we held the read_lock.
1345 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1346 	 * down_write, in which case, send_pcc_cmd will check for pending
1347 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1348 	 * So this CPU can be certain that its request will be delivered
1349 	 *    So in all cases, this CPU knows that its request will be delivered
1350 	 * by another CPU and can return
1351 	 *
1352 	 * After getting the down_write we still need to check for
1353 	 * pending_pcc_write_cmd to take care of the following scenario
1354 	 *    The thread running this code could be scheduled out between
1355 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1356 	 * could have delivered the request to Platform by triggering the
1357 	 * doorbell and transferred the ownership of PCC to platform. So this
1358 	 * avoids triggering an unnecessary doorbell and more importantly before
1359 	 * triggering the doorbell it makes sure that the PCC channel ownership
1360 	 * is still with OSPM.
1361 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1362 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1363 	 * before the pcc CMD_WRITE is completed. pcc_send_cmd checks for this
1364 	 * case during a CMD_READ and if there are pending writes it delivers
1365 	 * the write command before servicing the read command
1366 	 */
1367 	if (CPC_IN_PCC(desired_reg)) {
1368 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1369 			/* Update only if there are pending write commands */
1370 			if (pcc_ss_data->pending_pcc_write_cmd)
1371 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1372 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1373 		} else
1374 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1375 			wait_event(pcc_ss_data->pcc_write_wait_q,
1376 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1377 
1378 		/* send_pcc_cmd updates the status in case of failure */
1379 		ret = cpc_desc->write_cmd_status;
1380 	}
1381 	return ret;
1382 }
1383 EXPORT_SYMBOL_GPL(cppc_set_perf);
1384 
1385 /**
1386  * cppc_get_transition_latency - returns frequency transition latency in ns
1387  *
1388  * ACPI CPPC does not explicitly specifiy how a platform can specify the
1389  * transition latency for perfromance change requests. The closest we have
1390  * is the timing information from the PCCT tables which provides the info
1391  * on the number and frequency of PCC commands the platform can handle.
1392  */
1393 unsigned int cppc_get_transition_latency(int cpu_num)
1394 {
1395 	/*
1396 	 * Expected transition latency is based on the PCCT timing values
1397 	 * Below are definition from ACPI spec:
1398 	 * pcc_nominal- Expected latency to process a command, in microseconds
1399 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1400 	 *              channel can support, reported in commands per minute. 0
1401 	 *              indicates no limitation.
1402 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1403 	 *              completion of a command before issuing the next command,
1404 	 *              in microseconds.
1405 	 */
1406 	unsigned int latency_ns = 0;
1407 	struct cpc_desc *cpc_desc;
1408 	struct cpc_register_resource *desired_reg;
1409 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1410 	struct cppc_pcc_data *pcc_ss_data;
1411 
1412 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1413 	if (!cpc_desc)
1414 		return CPUFREQ_ETERNAL;
1415 
1416 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1417 	if (!CPC_IN_PCC(desired_reg))
1418 		return CPUFREQ_ETERNAL;
1419 
1420 	if (pcc_ss_id < 0)
1421 		return CPUFREQ_ETERNAL;
1422 
1423 	pcc_ss_data = pcc_data[pcc_ss_id];
1424 	if (pcc_ss_data->pcc_mpar)
1425 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1426 
1427 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1428 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1429 
1430 	return latency_ns;
1431 }
1432 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1433