xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 7b73a9c8)
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 		pr = all_cpu_data[i];
442 		if (!pr)
443 			continue;
444 
445 		if (cpumask_test_cpu(i, covered_cpus))
446 			continue;
447 
448 		cpc_ptr = per_cpu(cpc_desc_ptr, i);
449 		if (!cpc_ptr) {
450 			retval = -EFAULT;
451 			goto err_ret;
452 		}
453 
454 		pdomain = &(cpc_ptr->domain_info);
455 		cpumask_set_cpu(i, pr->shared_cpu_map);
456 		cpumask_set_cpu(i, covered_cpus);
457 		if (pdomain->num_processors <= 1)
458 			continue;
459 
460 		/* Validate the Domain info */
461 		count_target = pdomain->num_processors;
462 		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
463 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
464 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
465 			pr->shared_type = CPUFREQ_SHARED_TYPE_HW;
466 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
467 			pr->shared_type = CPUFREQ_SHARED_TYPE_ANY;
468 
469 		for_each_possible_cpu(j) {
470 			if (i == j)
471 				continue;
472 
473 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
474 			if (!match_cpc_ptr) {
475 				retval = -EFAULT;
476 				goto err_ret;
477 			}
478 
479 			match_pdomain = &(match_cpc_ptr->domain_info);
480 			if (match_pdomain->domain != pdomain->domain)
481 				continue;
482 
483 			/* Here i and j are in the same domain */
484 			if (match_pdomain->num_processors != count_target) {
485 				retval = -EFAULT;
486 				goto err_ret;
487 			}
488 
489 			if (pdomain->coord_type != match_pdomain->coord_type) {
490 				retval = -EFAULT;
491 				goto err_ret;
492 			}
493 
494 			cpumask_set_cpu(j, covered_cpus);
495 			cpumask_set_cpu(j, pr->shared_cpu_map);
496 		}
497 
498 		for_each_possible_cpu(j) {
499 			if (i == j)
500 				continue;
501 
502 			match_pr = all_cpu_data[j];
503 			if (!match_pr)
504 				continue;
505 
506 			match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
507 			if (!match_cpc_ptr) {
508 				retval = -EFAULT;
509 				goto err_ret;
510 			}
511 
512 			match_pdomain = &(match_cpc_ptr->domain_info);
513 			if (match_pdomain->domain != pdomain->domain)
514 				continue;
515 
516 			match_pr->shared_type = pr->shared_type;
517 			cpumask_copy(match_pr->shared_cpu_map,
518 				     pr->shared_cpu_map);
519 		}
520 	}
521 
522 err_ret:
523 	for_each_possible_cpu(i) {
524 		pr = all_cpu_data[i];
525 		if (!pr)
526 			continue;
527 
528 		/* Assume no coordination on any error parsing domain info */
529 		if (retval) {
530 			cpumask_clear(pr->shared_cpu_map);
531 			cpumask_set_cpu(i, pr->shared_cpu_map);
532 			pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
533 		}
534 	}
535 
536 	free_cpumask_var(covered_cpus);
537 	return retval;
538 }
539 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
540 
541 static int register_pcc_channel(int pcc_ss_idx)
542 {
543 	struct acpi_pcct_hw_reduced *cppc_ss;
544 	u64 usecs_lat;
545 
546 	if (pcc_ss_idx >= 0) {
547 		pcc_data[pcc_ss_idx]->pcc_channel =
548 			pcc_mbox_request_channel(&cppc_mbox_cl,	pcc_ss_idx);
549 
550 		if (IS_ERR(pcc_data[pcc_ss_idx]->pcc_channel)) {
551 			pr_err("Failed to find PCC channel for subspace %d\n",
552 			       pcc_ss_idx);
553 			return -ENODEV;
554 		}
555 
556 		/*
557 		 * The PCC mailbox controller driver should
558 		 * have parsed the PCCT (global table of all
559 		 * PCC channels) and stored pointers to the
560 		 * subspace communication region in con_priv.
561 		 */
562 		cppc_ss = (pcc_data[pcc_ss_idx]->pcc_channel)->con_priv;
563 
564 		if (!cppc_ss) {
565 			pr_err("No PCC subspace found for %d CPPC\n",
566 			       pcc_ss_idx);
567 			return -ENODEV;
568 		}
569 
570 		/*
571 		 * cppc_ss->latency is just a Nominal value. In reality
572 		 * the remote processor could be much slower to reply.
573 		 * So add an arbitrary amount of wait on top of Nominal.
574 		 */
575 		usecs_lat = NUM_RETRIES * cppc_ss->latency;
576 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
577 		pcc_data[pcc_ss_idx]->pcc_mrtt = cppc_ss->min_turnaround_time;
578 		pcc_data[pcc_ss_idx]->pcc_mpar = cppc_ss->max_access_rate;
579 		pcc_data[pcc_ss_idx]->pcc_nominal = cppc_ss->latency;
580 
581 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
582 			acpi_os_ioremap(cppc_ss->base_address, cppc_ss->length);
583 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
584 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
585 			       pcc_ss_idx);
586 			return -ENOMEM;
587 		}
588 
589 		/* Set flag so that we don't come here for each CPU. */
590 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
591 	}
592 
593 	return 0;
594 }
595 
596 /**
597  * cpc_ffh_supported() - check if FFH reading supported
598  *
599  * Check if the architecture has support for functional fixed hardware
600  * read/write capability.
601  *
602  * Return: true for supported, false for not supported
603  */
604 bool __weak cpc_ffh_supported(void)
605 {
606 	return false;
607 }
608 
609 /**
610  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
611  *
612  * Check and allocate the cppc_pcc_data memory.
613  * In some processor configurations it is possible that same subspace
614  * is shared between multiple CPUs. This is seen especially in CPUs
615  * with hardware multi-threading support.
616  *
617  * Return: 0 for success, errno for failure
618  */
619 int pcc_data_alloc(int pcc_ss_id)
620 {
621 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
622 		return -EINVAL;
623 
624 	if (pcc_data[pcc_ss_id]) {
625 		pcc_data[pcc_ss_id]->refcount++;
626 	} else {
627 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
628 					      GFP_KERNEL);
629 		if (!pcc_data[pcc_ss_id])
630 			return -ENOMEM;
631 		pcc_data[pcc_ss_id]->refcount++;
632 	}
633 
634 	return 0;
635 }
636 
637 /* Check if CPPC revision + num_ent combination is supported */
638 static bool is_cppc_supported(int revision, int num_ent)
639 {
640 	int expected_num_ent;
641 
642 	switch (revision) {
643 	case CPPC_V2_REV:
644 		expected_num_ent = CPPC_V2_NUM_ENT;
645 		break;
646 	case CPPC_V3_REV:
647 		expected_num_ent = CPPC_V3_NUM_ENT;
648 		break;
649 	default:
650 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
651 			revision);
652 		return false;
653 	}
654 
655 	if (expected_num_ent != num_ent) {
656 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
657 			num_ent, expected_num_ent, revision);
658 		return false;
659 	}
660 
661 	return true;
662 }
663 
664 /*
665  * An example CPC table looks like the following.
666  *
667  *	Name(_CPC, Package()
668  *			{
669  *			17,
670  *			NumEntries
671  *			1,
672  *			// Revision
673  *			ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
674  *			// Highest Performance
675  *			ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
676  *			// Nominal Performance
677  *			ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
678  *			// Lowest Nonlinear Performance
679  *			ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
680  *			// Lowest Performance
681  *			ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
682  *			// Guaranteed Performance Register
683  *			ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
684  *			// Desired Performance Register
685  *			ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
686  *			..
687  *			..
688  *			..
689  *
690  *		}
691  * Each Register() encodes how to access that specific register.
692  * e.g. a sample PCC entry has the following encoding:
693  *
694  *	Register (
695  *		PCC,
696  *		AddressSpaceKeyword
697  *		8,
698  *		//RegisterBitWidth
699  *		8,
700  *		//RegisterBitOffset
701  *		0x30,
702  *		//RegisterAddress
703  *		9
704  *		//AccessSize (subspace ID)
705  *		0
706  *		)
707  *	}
708  */
709 
710 /**
711  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
712  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
713  *
714  *	Return: 0 for success or negative value for err.
715  */
716 int acpi_cppc_processor_probe(struct acpi_processor *pr)
717 {
718 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
719 	union acpi_object *out_obj, *cpc_obj;
720 	struct cpc_desc *cpc_ptr;
721 	struct cpc_reg *gas_t;
722 	struct device *cpu_dev;
723 	acpi_handle handle = pr->handle;
724 	unsigned int num_ent, i, cpc_rev;
725 	int pcc_subspace_id = -1;
726 	acpi_status status;
727 	int ret = -EFAULT;
728 
729 	/* Parse the ACPI _CPC table for this CPU. */
730 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
731 			ACPI_TYPE_PACKAGE);
732 	if (ACPI_FAILURE(status)) {
733 		ret = -ENODEV;
734 		goto out_buf_free;
735 	}
736 
737 	out_obj = (union acpi_object *) output.pointer;
738 
739 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
740 	if (!cpc_ptr) {
741 		ret = -ENOMEM;
742 		goto out_buf_free;
743 	}
744 
745 	/* First entry is NumEntries. */
746 	cpc_obj = &out_obj->package.elements[0];
747 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
748 		num_ent = cpc_obj->integer.value;
749 	} else {
750 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
751 				cpc_obj->type);
752 		goto out_free;
753 	}
754 	cpc_ptr->num_entries = num_ent;
755 
756 	/* Second entry should be revision. */
757 	cpc_obj = &out_obj->package.elements[1];
758 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
759 		cpc_rev = cpc_obj->integer.value;
760 	} else {
761 		pr_debug("Unexpected entry type(%d) for Revision\n",
762 				cpc_obj->type);
763 		goto out_free;
764 	}
765 	cpc_ptr->version = cpc_rev;
766 
767 	if (!is_cppc_supported(cpc_rev, num_ent))
768 		goto out_free;
769 
770 	/* Iterate through remaining entries in _CPC */
771 	for (i = 2; i < num_ent; i++) {
772 		cpc_obj = &out_obj->package.elements[i];
773 
774 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
775 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
776 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
777 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
778 			gas_t = (struct cpc_reg *)
779 				cpc_obj->buffer.pointer;
780 
781 			/*
782 			 * The PCC Subspace index is encoded inside
783 			 * the CPC table entries. The same PCC index
784 			 * will be used for all the PCC entries,
785 			 * so extract it only once.
786 			 */
787 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
788 				if (pcc_subspace_id < 0) {
789 					pcc_subspace_id = gas_t->access_width;
790 					if (pcc_data_alloc(pcc_subspace_id))
791 						goto out_free;
792 				} else if (pcc_subspace_id != gas_t->access_width) {
793 					pr_debug("Mismatched PCC ids.\n");
794 					goto out_free;
795 				}
796 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
797 				if (gas_t->address) {
798 					void __iomem *addr;
799 
800 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
801 					if (!addr)
802 						goto out_free;
803 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
804 				}
805 			} else {
806 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
807 					/* Support only PCC ,SYS MEM and FFH type regs */
808 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
809 					goto out_free;
810 				}
811 			}
812 
813 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
814 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
815 		} else {
816 			pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
817 			goto out_free;
818 		}
819 	}
820 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
821 
822 	/*
823 	 * Initialize the remaining cpc_regs as unsupported.
824 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
825 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
826 	 */
827 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
828 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
829 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
830 	}
831 
832 
833 	/* Store CPU Logical ID */
834 	cpc_ptr->cpu_id = pr->id;
835 
836 	/* Parse PSD data for this CPU */
837 	ret = acpi_get_psd(cpc_ptr, handle);
838 	if (ret)
839 		goto out_free;
840 
841 	/* Register PCC channel once for all PCC subspace ID. */
842 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
843 		ret = register_pcc_channel(pcc_subspace_id);
844 		if (ret)
845 			goto out_free;
846 
847 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
848 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
849 	}
850 
851 	/* Everything looks okay */
852 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
853 
854 	/* Add per logical CPU nodes for reading its feedback counters. */
855 	cpu_dev = get_cpu_device(pr->id);
856 	if (!cpu_dev) {
857 		ret = -EINVAL;
858 		goto out_free;
859 	}
860 
861 	/* Plug PSD data into this CPU's CPC descriptor. */
862 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
863 
864 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
865 			"acpi_cppc");
866 	if (ret) {
867 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
868 		goto out_free;
869 	}
870 
871 	kfree(output.pointer);
872 	return 0;
873 
874 out_free:
875 	/* Free all the mapped sys mem areas for this CPU */
876 	for (i = 2; i < cpc_ptr->num_entries; i++) {
877 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
878 
879 		if (addr)
880 			iounmap(addr);
881 	}
882 	kfree(cpc_ptr);
883 
884 out_buf_free:
885 	kfree(output.pointer);
886 	return ret;
887 }
888 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
889 
890 /**
891  * acpi_cppc_processor_exit - Cleanup CPC structs.
892  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
893  *
894  * Return: Void
895  */
896 void acpi_cppc_processor_exit(struct acpi_processor *pr)
897 {
898 	struct cpc_desc *cpc_ptr;
899 	unsigned int i;
900 	void __iomem *addr;
901 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
902 
903 	if (pcc_ss_id >=0 && pcc_data[pcc_ss_id]) {
904 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
905 			pcc_data[pcc_ss_id]->refcount--;
906 			if (!pcc_data[pcc_ss_id]->refcount) {
907 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
908 				kfree(pcc_data[pcc_ss_id]);
909 				pcc_data[pcc_ss_id] = NULL;
910 			}
911 		}
912 	}
913 
914 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
915 	if (!cpc_ptr)
916 		return;
917 
918 	/* Free all the mapped sys mem areas for this CPU */
919 	for (i = 2; i < cpc_ptr->num_entries; i++) {
920 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
921 		if (addr)
922 			iounmap(addr);
923 	}
924 
925 	kobject_put(&cpc_ptr->kobj);
926 	kfree(cpc_ptr);
927 }
928 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
929 
930 /**
931  * cpc_read_ffh() - Read FFH register
932  * @cpunum:	CPU number to read
933  * @reg:	cppc register information
934  * @val:	place holder for return value
935  *
936  * Read bit_width bits from a specified address and bit_offset
937  *
938  * Return: 0 for success and error code
939  */
940 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
941 {
942 	return -ENOTSUPP;
943 }
944 
945 /**
946  * cpc_write_ffh() - Write FFH register
947  * @cpunum:	CPU number to write
948  * @reg:	cppc register information
949  * @val:	value to write
950  *
951  * Write value of bit_width bits to a specified address and bit_offset
952  *
953  * Return: 0 for success and error code
954  */
955 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
956 {
957 	return -ENOTSUPP;
958 }
959 
960 /*
961  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
962  * as fast as possible. We have already mapped the PCC subspace during init, so
963  * we can directly write to it.
964  */
965 
966 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
967 {
968 	int ret_val = 0;
969 	void __iomem *vaddr = 0;
970 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
971 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
972 
973 	if (reg_res->type == ACPI_TYPE_INTEGER) {
974 		*val = reg_res->cpc_entry.int_value;
975 		return ret_val;
976 	}
977 
978 	*val = 0;
979 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
980 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
981 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
982 		vaddr = reg_res->sys_mem_vaddr;
983 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
984 		return cpc_read_ffh(cpu, reg, val);
985 	else
986 		return acpi_os_read_memory((acpi_physical_address)reg->address,
987 				val, reg->bit_width);
988 
989 	switch (reg->bit_width) {
990 		case 8:
991 			*val = readb_relaxed(vaddr);
992 			break;
993 		case 16:
994 			*val = readw_relaxed(vaddr);
995 			break;
996 		case 32:
997 			*val = readl_relaxed(vaddr);
998 			break;
999 		case 64:
1000 			*val = readq_relaxed(vaddr);
1001 			break;
1002 		default:
1003 			pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
1004 				 reg->bit_width, pcc_ss_id);
1005 			ret_val = -EFAULT;
1006 	}
1007 
1008 	return ret_val;
1009 }
1010 
1011 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
1012 {
1013 	int ret_val = 0;
1014 	void __iomem *vaddr = 0;
1015 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1016 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
1017 
1018 	if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
1019 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
1020 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
1021 		vaddr = reg_res->sys_mem_vaddr;
1022 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1023 		return cpc_write_ffh(cpu, reg, val);
1024 	else
1025 		return acpi_os_write_memory((acpi_physical_address)reg->address,
1026 				val, reg->bit_width);
1027 
1028 	switch (reg->bit_width) {
1029 		case 8:
1030 			writeb_relaxed(val, vaddr);
1031 			break;
1032 		case 16:
1033 			writew_relaxed(val, vaddr);
1034 			break;
1035 		case 32:
1036 			writel_relaxed(val, vaddr);
1037 			break;
1038 		case 64:
1039 			writeq_relaxed(val, vaddr);
1040 			break;
1041 		default:
1042 			pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1043 				 reg->bit_width, pcc_ss_id);
1044 			ret_val = -EFAULT;
1045 			break;
1046 	}
1047 
1048 	return ret_val;
1049 }
1050 
1051 /**
1052  * cppc_get_desired_perf - Get the value of desired performance register.
1053  * @cpunum: CPU from which to get desired performance.
1054  * @desired_perf: address of a variable to store the returned desired performance
1055  *
1056  * Return: 0 for success, -EIO otherwise.
1057  */
1058 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1059 {
1060 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1061 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1062 	struct cpc_register_resource *desired_reg;
1063 	struct cppc_pcc_data *pcc_ss_data = NULL;
1064 
1065 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1066 
1067 	if (CPC_IN_PCC(desired_reg)) {
1068 		int ret = 0;
1069 
1070 		if (pcc_ss_id < 0)
1071 			return -EIO;
1072 
1073 		pcc_ss_data = pcc_data[pcc_ss_id];
1074 
1075 		down_write(&pcc_ss_data->pcc_lock);
1076 
1077 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1078 			cpc_read(cpunum, desired_reg, desired_perf);
1079 		else
1080 			ret = -EIO;
1081 
1082 		up_write(&pcc_ss_data->pcc_lock);
1083 
1084 		return ret;
1085 	}
1086 
1087 	cpc_read(cpunum, desired_reg, desired_perf);
1088 
1089 	return 0;
1090 }
1091 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1092 
1093 /**
1094  * cppc_get_perf_caps - Get a CPU's performance capabilities.
1095  * @cpunum: CPU from which to get capabilities info.
1096  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
1097  *
1098  * Return: 0 for success with perf_caps populated else -ERRNO.
1099  */
1100 int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
1101 {
1102 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1103 	struct cpc_register_resource *highest_reg, *lowest_reg,
1104 		*lowest_non_linear_reg, *nominal_reg, *guaranteed_reg,
1105 		*low_freq_reg = NULL, *nom_freq_reg = NULL;
1106 	u64 high, low, guaranteed, nom, min_nonlinear, low_f = 0, nom_f = 0;
1107 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1108 	struct cppc_pcc_data *pcc_ss_data = NULL;
1109 	int ret = 0, regs_in_pcc = 0;
1110 
1111 	if (!cpc_desc) {
1112 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1113 		return -ENODEV;
1114 	}
1115 
1116 	highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
1117 	lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
1118 	lowest_non_linear_reg = &cpc_desc->cpc_regs[LOW_NON_LINEAR_PERF];
1119 	nominal_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1120 	low_freq_reg = &cpc_desc->cpc_regs[LOWEST_FREQ];
1121 	nom_freq_reg = &cpc_desc->cpc_regs[NOMINAL_FREQ];
1122 	guaranteed_reg = &cpc_desc->cpc_regs[GUARANTEED_PERF];
1123 
1124 	/* Are any of the regs PCC ?*/
1125 	if (CPC_IN_PCC(highest_reg) || CPC_IN_PCC(lowest_reg) ||
1126 		CPC_IN_PCC(lowest_non_linear_reg) || CPC_IN_PCC(nominal_reg) ||
1127 		CPC_IN_PCC(low_freq_reg) || CPC_IN_PCC(nom_freq_reg)) {
1128 		if (pcc_ss_id < 0) {
1129 			pr_debug("Invalid pcc_ss_id\n");
1130 			return -ENODEV;
1131 		}
1132 		pcc_ss_data = pcc_data[pcc_ss_id];
1133 		regs_in_pcc = 1;
1134 		down_write(&pcc_ss_data->pcc_lock);
1135 		/* Ring doorbell once to update PCC subspace */
1136 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1137 			ret = -EIO;
1138 			goto out_err;
1139 		}
1140 	}
1141 
1142 	cpc_read(cpunum, highest_reg, &high);
1143 	perf_caps->highest_perf = high;
1144 
1145 	cpc_read(cpunum, lowest_reg, &low);
1146 	perf_caps->lowest_perf = low;
1147 
1148 	cpc_read(cpunum, nominal_reg, &nom);
1149 	perf_caps->nominal_perf = nom;
1150 
1151 	if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
1152 	    IS_NULL_REG(&guaranteed_reg->cpc_entry.reg)) {
1153 		perf_caps->guaranteed_perf = 0;
1154 	} else {
1155 		cpc_read(cpunum, guaranteed_reg, &guaranteed);
1156 		perf_caps->guaranteed_perf = guaranteed;
1157 	}
1158 
1159 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
1160 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
1161 
1162 	if (!high || !low || !nom || !min_nonlinear)
1163 		ret = -EFAULT;
1164 
1165 	/* Read optional lowest and nominal frequencies if present */
1166 	if (CPC_SUPPORTED(low_freq_reg))
1167 		cpc_read(cpunum, low_freq_reg, &low_f);
1168 
1169 	if (CPC_SUPPORTED(nom_freq_reg))
1170 		cpc_read(cpunum, nom_freq_reg, &nom_f);
1171 
1172 	perf_caps->lowest_freq = low_f;
1173 	perf_caps->nominal_freq = nom_f;
1174 
1175 
1176 out_err:
1177 	if (regs_in_pcc)
1178 		up_write(&pcc_ss_data->pcc_lock);
1179 	return ret;
1180 }
1181 EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
1182 
1183 /**
1184  * cppc_get_perf_ctrs - Read a CPU's performance feedback counters.
1185  * @cpunum: CPU from which to read counters.
1186  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
1187  *
1188  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
1189  */
1190 int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
1191 {
1192 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1193 	struct cpc_register_resource *delivered_reg, *reference_reg,
1194 		*ref_perf_reg, *ctr_wrap_reg;
1195 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1196 	struct cppc_pcc_data *pcc_ss_data = NULL;
1197 	u64 delivered, reference, ref_perf, ctr_wrap_time;
1198 	int ret = 0, regs_in_pcc = 0;
1199 
1200 	if (!cpc_desc) {
1201 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1202 		return -ENODEV;
1203 	}
1204 
1205 	delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
1206 	reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
1207 	ref_perf_reg = &cpc_desc->cpc_regs[REFERENCE_PERF];
1208 	ctr_wrap_reg = &cpc_desc->cpc_regs[CTR_WRAP_TIME];
1209 
1210 	/*
1211 	 * If reference perf register is not supported then we should
1212 	 * use the nominal perf value
1213 	 */
1214 	if (!CPC_SUPPORTED(ref_perf_reg))
1215 		ref_perf_reg = &cpc_desc->cpc_regs[NOMINAL_PERF];
1216 
1217 	/* Are any of the regs PCC ?*/
1218 	if (CPC_IN_PCC(delivered_reg) || CPC_IN_PCC(reference_reg) ||
1219 		CPC_IN_PCC(ctr_wrap_reg) || CPC_IN_PCC(ref_perf_reg)) {
1220 		if (pcc_ss_id < 0) {
1221 			pr_debug("Invalid pcc_ss_id\n");
1222 			return -ENODEV;
1223 		}
1224 		pcc_ss_data = pcc_data[pcc_ss_id];
1225 		down_write(&pcc_ss_data->pcc_lock);
1226 		regs_in_pcc = 1;
1227 		/* Ring doorbell once to update PCC subspace */
1228 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) < 0) {
1229 			ret = -EIO;
1230 			goto out_err;
1231 		}
1232 	}
1233 
1234 	cpc_read(cpunum, delivered_reg, &delivered);
1235 	cpc_read(cpunum, reference_reg, &reference);
1236 	cpc_read(cpunum, ref_perf_reg, &ref_perf);
1237 
1238 	/*
1239 	 * Per spec, if ctr_wrap_time optional register is unsupported, then the
1240 	 * performance counters are assumed to never wrap during the lifetime of
1241 	 * platform
1242 	 */
1243 	ctr_wrap_time = (u64)(~((u64)0));
1244 	if (CPC_SUPPORTED(ctr_wrap_reg))
1245 		cpc_read(cpunum, ctr_wrap_reg, &ctr_wrap_time);
1246 
1247 	if (!delivered || !reference ||	!ref_perf) {
1248 		ret = -EFAULT;
1249 		goto out_err;
1250 	}
1251 
1252 	perf_fb_ctrs->delivered = delivered;
1253 	perf_fb_ctrs->reference = reference;
1254 	perf_fb_ctrs->reference_perf = ref_perf;
1255 	perf_fb_ctrs->wraparound_time = ctr_wrap_time;
1256 out_err:
1257 	if (regs_in_pcc)
1258 		up_write(&pcc_ss_data->pcc_lock);
1259 	return ret;
1260 }
1261 EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
1262 
1263 /**
1264  * cppc_set_perf - Set a CPU's performance controls.
1265  * @cpu: CPU for which to set performance controls.
1266  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1267  *
1268  * Return: 0 for success, -ERRNO otherwise.
1269  */
1270 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1271 {
1272 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1273 	struct cpc_register_resource *desired_reg;
1274 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1275 	struct cppc_pcc_data *pcc_ss_data = NULL;
1276 	int ret = 0;
1277 
1278 	if (!cpc_desc) {
1279 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1280 		return -ENODEV;
1281 	}
1282 
1283 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1284 
1285 	/*
1286 	 * This is Phase-I where we want to write to CPC registers
1287 	 * -> We want all CPUs to be able to execute this phase in parallel
1288 	 *
1289 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1290 	 * achieve that goal here
1291 	 */
1292 	if (CPC_IN_PCC(desired_reg)) {
1293 		if (pcc_ss_id < 0) {
1294 			pr_debug("Invalid pcc_ss_id\n");
1295 			return -ENODEV;
1296 		}
1297 		pcc_ss_data = pcc_data[pcc_ss_id];
1298 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1299 		if (pcc_ss_data->platform_owns_pcc) {
1300 			ret = check_pcc_chan(pcc_ss_id, false);
1301 			if (ret) {
1302 				up_read(&pcc_ss_data->pcc_lock);
1303 				return ret;
1304 			}
1305 		}
1306 		/*
1307 		 * Update the pending_write to make sure a PCC CMD_READ will not
1308 		 * arrive and steal the channel during the switch to write lock
1309 		 */
1310 		pcc_ss_data->pending_pcc_write_cmd = true;
1311 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1312 		cpc_desc->write_cmd_status = 0;
1313 	}
1314 
1315 	/*
1316 	 * Skip writing MIN/MAX until Linux knows how to come up with
1317 	 * useful values.
1318 	 */
1319 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1320 
1321 	if (CPC_IN_PCC(desired_reg))
1322 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1323 	/*
1324 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1325 	 *
1326 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1327 	 * requests that happened in short overlapping interval. The last CPU to
1328 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1329 	 *
1330 	 * We have the following requirements for Phase-II:
1331 	 *     1. We want to execute Phase-II only when there are no CPUs
1332 	 * currently executing in Phase-I
1333 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1334 	 * entering Phase-I.
1335 	 *     3. We want only one CPU among all those who went through Phase-I
1336 	 * to run phase-II
1337 	 *
1338 	 * If write_trylock fails to get the lock and doesn't transfer the
1339 	 * PCC ownership to the platform, then one of the following will be TRUE
1340 	 *     1. There is at-least one CPU in Phase-I which will later execute
1341 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1342 	 * executing the Phase-II.
1343 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1344 	 * write_trylock and has already acquired the write_lock. We know for a
1345 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1346 	 * before this CPU's Phase-I as we held the read_lock.
1347 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1348 	 * down_write, in which case, send_pcc_cmd will check for pending
1349 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1350 	 * So this CPU can be certain that its request will be delivered
1351 	 *    So in all cases, this CPU knows that its request will be delivered
1352 	 * by another CPU and can return
1353 	 *
1354 	 * After getting the down_write we still need to check for
1355 	 * pending_pcc_write_cmd to take care of the following scenario
1356 	 *    The thread running this code could be scheduled out between
1357 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1358 	 * could have delivered the request to Platform by triggering the
1359 	 * doorbell and transferred the ownership of PCC to platform. So this
1360 	 * avoids triggering an unnecessary doorbell and more importantly before
1361 	 * triggering the doorbell it makes sure that the PCC channel ownership
1362 	 * is still with OSPM.
1363 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1364 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1365 	 * before the pcc CMD_WRITE is completed. pcc_send_cmd checks for this
1366 	 * case during a CMD_READ and if there are pending writes it delivers
1367 	 * the write command before servicing the read command
1368 	 */
1369 	if (CPC_IN_PCC(desired_reg)) {
1370 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1371 			/* Update only if there are pending write commands */
1372 			if (pcc_ss_data->pending_pcc_write_cmd)
1373 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1374 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1375 		} else
1376 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1377 			wait_event(pcc_ss_data->pcc_write_wait_q,
1378 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1379 
1380 		/* send_pcc_cmd updates the status in case of failure */
1381 		ret = cpc_desc->write_cmd_status;
1382 	}
1383 	return ret;
1384 }
1385 EXPORT_SYMBOL_GPL(cppc_set_perf);
1386 
1387 /**
1388  * cppc_get_transition_latency - returns frequency transition latency in ns
1389  *
1390  * ACPI CPPC does not explicitly specifiy how a platform can specify the
1391  * transition latency for perfromance change requests. The closest we have
1392  * is the timing information from the PCCT tables which provides the info
1393  * on the number and frequency of PCC commands the platform can handle.
1394  */
1395 unsigned int cppc_get_transition_latency(int cpu_num)
1396 {
1397 	/*
1398 	 * Expected transition latency is based on the PCCT timing values
1399 	 * Below are definition from ACPI spec:
1400 	 * pcc_nominal- Expected latency to process a command, in microseconds
1401 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1402 	 *              channel can support, reported in commands per minute. 0
1403 	 *              indicates no limitation.
1404 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1405 	 *              completion of a command before issuing the next command,
1406 	 *              in microseconds.
1407 	 */
1408 	unsigned int latency_ns = 0;
1409 	struct cpc_desc *cpc_desc;
1410 	struct cpc_register_resource *desired_reg;
1411 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1412 	struct cppc_pcc_data *pcc_ss_data;
1413 
1414 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1415 	if (!cpc_desc)
1416 		return CPUFREQ_ETERNAL;
1417 
1418 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1419 	if (!CPC_IN_PCC(desired_reg))
1420 		return CPUFREQ_ETERNAL;
1421 
1422 	if (pcc_ss_id < 0)
1423 		return CPUFREQ_ETERNAL;
1424 
1425 	pcc_ss_data = pcc_data[pcc_ss_id];
1426 	if (pcc_ss_data->pcc_mpar)
1427 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1428 
1429 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1430 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1431 
1432 	return latency_ns;
1433 }
1434 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1435