xref: /openbmc/linux/drivers/acpi/cppc_acpi.c (revision 2cf1c348)
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/delay.h>
37 #include <linux/iopoll.h>
38 #include <linux/ktime.h>
39 #include <linux/rwsem.h>
40 #include <linux/wait.h>
41 #include <linux/topology.h>
42 
43 #include <acpi/cppc_acpi.h>
44 
45 struct cppc_pcc_data {
46 	struct pcc_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 /* Evaluates 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 /* Evaluates 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 #define OVER_16BTS_MASK ~0xFFFFULL
122 
123 #define define_one_cppc_ro(_name)		\
124 static struct kobj_attribute _name =		\
125 __ATTR(_name, 0444, show_##_name, NULL)
126 
127 #define to_cpc_desc(a) container_of(a, struct cpc_desc, kobj)
128 
129 #define show_cppc_data(access_fn, struct_name, member_name)		\
130 	static ssize_t show_##member_name(struct kobject *kobj,		\
131 				struct kobj_attribute *attr, char *buf)	\
132 	{								\
133 		struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);		\
134 		struct struct_name st_name = {0};			\
135 		int ret;						\
136 									\
137 		ret = access_fn(cpc_ptr->cpu_id, &st_name);		\
138 		if (ret)						\
139 			return ret;					\
140 									\
141 		return scnprintf(buf, PAGE_SIZE, "%llu\n",		\
142 				(u64)st_name.member_name);		\
143 	}								\
144 	define_one_cppc_ro(member_name)
145 
146 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, highest_perf);
147 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_perf);
148 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_perf);
149 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_nonlinear_perf);
150 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, lowest_freq);
151 show_cppc_data(cppc_get_perf_caps, cppc_perf_caps, nominal_freq);
152 
153 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, reference_perf);
154 show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
155 
156 static ssize_t show_feedback_ctrs(struct kobject *kobj,
157 		struct kobj_attribute *attr, char *buf)
158 {
159 	struct cpc_desc *cpc_ptr = to_cpc_desc(kobj);
160 	struct cppc_perf_fb_ctrs fb_ctrs = {0};
161 	int ret;
162 
163 	ret = cppc_get_perf_ctrs(cpc_ptr->cpu_id, &fb_ctrs);
164 	if (ret)
165 		return ret;
166 
167 	return scnprintf(buf, PAGE_SIZE, "ref:%llu del:%llu\n",
168 			fb_ctrs.reference, fb_ctrs.delivered);
169 }
170 define_one_cppc_ro(feedback_ctrs);
171 
172 static struct attribute *cppc_attrs[] = {
173 	&feedback_ctrs.attr,
174 	&reference_perf.attr,
175 	&wraparound_time.attr,
176 	&highest_perf.attr,
177 	&lowest_perf.attr,
178 	&lowest_nonlinear_perf.attr,
179 	&nominal_perf.attr,
180 	&nominal_freq.attr,
181 	&lowest_freq.attr,
182 	NULL
183 };
184 ATTRIBUTE_GROUPS(cppc);
185 
186 static struct kobj_type cppc_ktype = {
187 	.sysfs_ops = &kobj_sysfs_ops,
188 	.default_groups = cppc_groups,
189 };
190 
191 static int check_pcc_chan(int pcc_ss_id, bool chk_err_bit)
192 {
193 	int ret, status;
194 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
195 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
196 		pcc_ss_data->pcc_comm_addr;
197 
198 	if (!pcc_ss_data->platform_owns_pcc)
199 		return 0;
200 
201 	/*
202 	 * Poll PCC status register every 3us(delay_us) for maximum of
203 	 * deadline_us(timeout_us) until PCC command complete bit is set(cond)
204 	 */
205 	ret = readw_relaxed_poll_timeout(&generic_comm_base->status, status,
206 					status & PCC_CMD_COMPLETE_MASK, 3,
207 					pcc_ss_data->deadline_us);
208 
209 	if (likely(!ret)) {
210 		pcc_ss_data->platform_owns_pcc = false;
211 		if (chk_err_bit && (status & PCC_ERROR_MASK))
212 			ret = -EIO;
213 	}
214 
215 	if (unlikely(ret))
216 		pr_err("PCC check channel failed for ss: %d. ret=%d\n",
217 		       pcc_ss_id, ret);
218 
219 	return ret;
220 }
221 
222 /*
223  * This function transfers the ownership of the PCC to the platform
224  * So it must be called while holding write_lock(pcc_lock)
225  */
226 static int send_pcc_cmd(int pcc_ss_id, u16 cmd)
227 {
228 	int ret = -EIO, i;
229 	struct cppc_pcc_data *pcc_ss_data = pcc_data[pcc_ss_id];
230 	struct acpi_pcct_shared_memory __iomem *generic_comm_base =
231 		pcc_ss_data->pcc_comm_addr;
232 	unsigned int time_delta;
233 
234 	/*
235 	 * For CMD_WRITE we know for a fact the caller should have checked
236 	 * the channel before writing to PCC space
237 	 */
238 	if (cmd == CMD_READ) {
239 		/*
240 		 * If there are pending cpc_writes, then we stole the channel
241 		 * before write completion, so first send a WRITE command to
242 		 * platform
243 		 */
244 		if (pcc_ss_data->pending_pcc_write_cmd)
245 			send_pcc_cmd(pcc_ss_id, CMD_WRITE);
246 
247 		ret = check_pcc_chan(pcc_ss_id, false);
248 		if (ret)
249 			goto end;
250 	} else /* CMD_WRITE */
251 		pcc_ss_data->pending_pcc_write_cmd = FALSE;
252 
253 	/*
254 	 * Handle the Minimum Request Turnaround Time(MRTT)
255 	 * "The minimum amount of time that OSPM must wait after the completion
256 	 * of a command before issuing the next command, in microseconds"
257 	 */
258 	if (pcc_ss_data->pcc_mrtt) {
259 		time_delta = ktime_us_delta(ktime_get(),
260 					    pcc_ss_data->last_cmd_cmpl_time);
261 		if (pcc_ss_data->pcc_mrtt > time_delta)
262 			udelay(pcc_ss_data->pcc_mrtt - time_delta);
263 	}
264 
265 	/*
266 	 * Handle the non-zero Maximum Periodic Access Rate(MPAR)
267 	 * "The maximum number of periodic requests that the subspace channel can
268 	 * support, reported in commands per minute. 0 indicates no limitation."
269 	 *
270 	 * This parameter should be ideally zero or large enough so that it can
271 	 * handle maximum number of requests that all the cores in the system can
272 	 * collectively generate. If it is not, we will follow the spec and just
273 	 * not send the request to the platform after hitting the MPAR limit in
274 	 * any 60s window
275 	 */
276 	if (pcc_ss_data->pcc_mpar) {
277 		if (pcc_ss_data->mpar_count == 0) {
278 			time_delta = ktime_ms_delta(ktime_get(),
279 						    pcc_ss_data->last_mpar_reset);
280 			if ((time_delta < 60 * MSEC_PER_SEC) && pcc_ss_data->last_mpar_reset) {
281 				pr_debug("PCC cmd for subspace %d not sent due to MPAR limit",
282 					 pcc_ss_id);
283 				ret = -EIO;
284 				goto end;
285 			}
286 			pcc_ss_data->last_mpar_reset = ktime_get();
287 			pcc_ss_data->mpar_count = pcc_ss_data->pcc_mpar;
288 		}
289 		pcc_ss_data->mpar_count--;
290 	}
291 
292 	/* Write to the shared comm region. */
293 	writew_relaxed(cmd, &generic_comm_base->command);
294 
295 	/* Flip CMD COMPLETE bit */
296 	writew_relaxed(0, &generic_comm_base->status);
297 
298 	pcc_ss_data->platform_owns_pcc = true;
299 
300 	/* Ring doorbell */
301 	ret = mbox_send_message(pcc_ss_data->pcc_channel->mchan, &cmd);
302 	if (ret < 0) {
303 		pr_err("Err sending PCC mbox message. ss: %d cmd:%d, ret:%d\n",
304 		       pcc_ss_id, cmd, ret);
305 		goto end;
306 	}
307 
308 	/* wait for completion and check for PCC errro bit */
309 	ret = check_pcc_chan(pcc_ss_id, true);
310 
311 	if (pcc_ss_data->pcc_mrtt)
312 		pcc_ss_data->last_cmd_cmpl_time = ktime_get();
313 
314 	if (pcc_ss_data->pcc_channel->mchan->mbox->txdone_irq)
315 		mbox_chan_txdone(pcc_ss_data->pcc_channel->mchan, ret);
316 	else
317 		mbox_client_txdone(pcc_ss_data->pcc_channel->mchan, ret);
318 
319 end:
320 	if (cmd == CMD_WRITE) {
321 		if (unlikely(ret)) {
322 			for_each_possible_cpu(i) {
323 				struct cpc_desc *desc = per_cpu(cpc_desc_ptr, i);
324 
325 				if (!desc)
326 					continue;
327 
328 				if (desc->write_cmd_id == pcc_ss_data->pcc_write_cnt)
329 					desc->write_cmd_status = ret;
330 			}
331 		}
332 		pcc_ss_data->pcc_write_cnt++;
333 		wake_up_all(&pcc_ss_data->pcc_write_wait_q);
334 	}
335 
336 	return ret;
337 }
338 
339 static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
340 {
341 	if (ret < 0)
342 		pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
343 				*(u16 *)msg, ret);
344 	else
345 		pr_debug("TX completed. CMD sent:%x, ret:%d\n",
346 				*(u16 *)msg, ret);
347 }
348 
349 static struct mbox_client cppc_mbox_cl = {
350 	.tx_done = cppc_chan_tx_done,
351 	.knows_txdone = true,
352 };
353 
354 static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
355 {
356 	int result = -EFAULT;
357 	acpi_status status = AE_OK;
358 	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
359 	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
360 	struct acpi_buffer state = {0, NULL};
361 	union acpi_object  *psd = NULL;
362 	struct acpi_psd_package *pdomain;
363 
364 	status = acpi_evaluate_object_typed(handle, "_PSD", NULL,
365 					    &buffer, ACPI_TYPE_PACKAGE);
366 	if (status == AE_NOT_FOUND)	/* _PSD is optional */
367 		return 0;
368 	if (ACPI_FAILURE(status))
369 		return -ENODEV;
370 
371 	psd = buffer.pointer;
372 	if (!psd || psd->package.count != 1) {
373 		pr_debug("Invalid _PSD data\n");
374 		goto end;
375 	}
376 
377 	pdomain = &(cpc_ptr->domain_info);
378 
379 	state.length = sizeof(struct acpi_psd_package);
380 	state.pointer = pdomain;
381 
382 	status = acpi_extract_package(&(psd->package.elements[0]),
383 		&format, &state);
384 	if (ACPI_FAILURE(status)) {
385 		pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
386 		goto end;
387 	}
388 
389 	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
390 		pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
391 		goto end;
392 	}
393 
394 	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
395 		pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
396 		goto end;
397 	}
398 
399 	if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
400 	    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
401 	    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
402 		pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
403 		goto end;
404 	}
405 
406 	result = 0;
407 end:
408 	kfree(buffer.pointer);
409 	return result;
410 }
411 
412 bool acpi_cpc_valid(void)
413 {
414 	struct cpc_desc *cpc_ptr;
415 	int cpu;
416 
417 	for_each_present_cpu(cpu) {
418 		cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
419 		if (!cpc_ptr)
420 			return false;
421 	}
422 
423 	return true;
424 }
425 EXPORT_SYMBOL_GPL(acpi_cpc_valid);
426 
427 /**
428  * acpi_get_psd_map - Map the CPUs in the freq domain of a given cpu
429  * @cpu: Find all CPUs that share a domain with cpu.
430  * @cpu_data: Pointer to CPU specific CPPC data including PSD info.
431  *
432  *	Return: 0 for success or negative value for err.
433  */
434 int acpi_get_psd_map(unsigned int cpu, struct cppc_cpudata *cpu_data)
435 {
436 	struct cpc_desc *cpc_ptr, *match_cpc_ptr;
437 	struct acpi_psd_package *match_pdomain;
438 	struct acpi_psd_package *pdomain;
439 	int count_target, i;
440 
441 	/*
442 	 * Now that we have _PSD data from all CPUs, let's setup P-state
443 	 * domain info.
444 	 */
445 	cpc_ptr = per_cpu(cpc_desc_ptr, cpu);
446 	if (!cpc_ptr)
447 		return -EFAULT;
448 
449 	pdomain = &(cpc_ptr->domain_info);
450 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
451 	if (pdomain->num_processors <= 1)
452 		return 0;
453 
454 	/* Validate the Domain info */
455 	count_target = pdomain->num_processors;
456 	if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
457 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ALL;
458 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
459 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_HW;
460 	else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
461 		cpu_data->shared_type = CPUFREQ_SHARED_TYPE_ANY;
462 
463 	for_each_possible_cpu(i) {
464 		if (i == cpu)
465 			continue;
466 
467 		match_cpc_ptr = per_cpu(cpc_desc_ptr, i);
468 		if (!match_cpc_ptr)
469 			goto err_fault;
470 
471 		match_pdomain = &(match_cpc_ptr->domain_info);
472 		if (match_pdomain->domain != pdomain->domain)
473 			continue;
474 
475 		/* Here i and cpu are in the same domain */
476 		if (match_pdomain->num_processors != count_target)
477 			goto err_fault;
478 
479 		if (pdomain->coord_type != match_pdomain->coord_type)
480 			goto err_fault;
481 
482 		cpumask_set_cpu(i, cpu_data->shared_cpu_map);
483 	}
484 
485 	return 0;
486 
487 err_fault:
488 	/* Assume no coordination on any error parsing domain info */
489 	cpumask_clear(cpu_data->shared_cpu_map);
490 	cpumask_set_cpu(cpu, cpu_data->shared_cpu_map);
491 	cpu_data->shared_type = CPUFREQ_SHARED_TYPE_NONE;
492 
493 	return -EFAULT;
494 }
495 EXPORT_SYMBOL_GPL(acpi_get_psd_map);
496 
497 static int register_pcc_channel(int pcc_ss_idx)
498 {
499 	struct pcc_mbox_chan *pcc_chan;
500 	u64 usecs_lat;
501 
502 	if (pcc_ss_idx >= 0) {
503 		pcc_chan = pcc_mbox_request_channel(&cppc_mbox_cl, pcc_ss_idx);
504 
505 		if (IS_ERR(pcc_chan)) {
506 			pr_err("Failed to find PCC channel for subspace %d\n",
507 			       pcc_ss_idx);
508 			return -ENODEV;
509 		}
510 
511 		pcc_data[pcc_ss_idx]->pcc_channel = pcc_chan;
512 		/*
513 		 * cppc_ss->latency is just a Nominal value. In reality
514 		 * the remote processor could be much slower to reply.
515 		 * So add an arbitrary amount of wait on top of Nominal.
516 		 */
517 		usecs_lat = NUM_RETRIES * pcc_chan->latency;
518 		pcc_data[pcc_ss_idx]->deadline_us = usecs_lat;
519 		pcc_data[pcc_ss_idx]->pcc_mrtt = pcc_chan->min_turnaround_time;
520 		pcc_data[pcc_ss_idx]->pcc_mpar = pcc_chan->max_access_rate;
521 		pcc_data[pcc_ss_idx]->pcc_nominal = pcc_chan->latency;
522 
523 		pcc_data[pcc_ss_idx]->pcc_comm_addr =
524 			acpi_os_ioremap(pcc_chan->shmem_base_addr,
525 					pcc_chan->shmem_size);
526 		if (!pcc_data[pcc_ss_idx]->pcc_comm_addr) {
527 			pr_err("Failed to ioremap PCC comm region mem for %d\n",
528 			       pcc_ss_idx);
529 			return -ENOMEM;
530 		}
531 
532 		/* Set flag so that we don't come here for each CPU. */
533 		pcc_data[pcc_ss_idx]->pcc_channel_acquired = true;
534 	}
535 
536 	return 0;
537 }
538 
539 /**
540  * cpc_ffh_supported() - check if FFH reading supported
541  *
542  * Check if the architecture has support for functional fixed hardware
543  * read/write capability.
544  *
545  * Return: true for supported, false for not supported
546  */
547 bool __weak cpc_ffh_supported(void)
548 {
549 	return false;
550 }
551 
552 /**
553  * pcc_data_alloc() - Allocate the pcc_data memory for pcc subspace
554  *
555  * Check and allocate the cppc_pcc_data memory.
556  * In some processor configurations it is possible that same subspace
557  * is shared between multiple CPUs. This is seen especially in CPUs
558  * with hardware multi-threading support.
559  *
560  * Return: 0 for success, errno for failure
561  */
562 static int pcc_data_alloc(int pcc_ss_id)
563 {
564 	if (pcc_ss_id < 0 || pcc_ss_id >= MAX_PCC_SUBSPACES)
565 		return -EINVAL;
566 
567 	if (pcc_data[pcc_ss_id]) {
568 		pcc_data[pcc_ss_id]->refcount++;
569 	} else {
570 		pcc_data[pcc_ss_id] = kzalloc(sizeof(struct cppc_pcc_data),
571 					      GFP_KERNEL);
572 		if (!pcc_data[pcc_ss_id])
573 			return -ENOMEM;
574 		pcc_data[pcc_ss_id]->refcount++;
575 	}
576 
577 	return 0;
578 }
579 
580 /* Check if CPPC revision + num_ent combination is supported */
581 static bool is_cppc_supported(int revision, int num_ent)
582 {
583 	int expected_num_ent;
584 
585 	switch (revision) {
586 	case CPPC_V2_REV:
587 		expected_num_ent = CPPC_V2_NUM_ENT;
588 		break;
589 	case CPPC_V3_REV:
590 		expected_num_ent = CPPC_V3_NUM_ENT;
591 		break;
592 	default:
593 		pr_debug("Firmware exports unsupported CPPC revision: %d\n",
594 			revision);
595 		return false;
596 	}
597 
598 	if (expected_num_ent != num_ent) {
599 		pr_debug("Firmware exports %d entries. Expected: %d for CPPC rev:%d\n",
600 			num_ent, expected_num_ent, revision);
601 		return false;
602 	}
603 
604 	return true;
605 }
606 
607 /*
608  * An example CPC table looks like the following.
609  *
610  *  Name (_CPC, Package() {
611  *      17,							// NumEntries
612  *      1,							// Revision
613  *      ResourceTemplate() {Register(PCC, 32, 0, 0x120, 2)},	// Highest Performance
614  *      ResourceTemplate() {Register(PCC, 32, 0, 0x124, 2)},	// Nominal Performance
615  *      ResourceTemplate() {Register(PCC, 32, 0, 0x128, 2)},	// Lowest Nonlinear Performance
616  *      ResourceTemplate() {Register(PCC, 32, 0, 0x12C, 2)},	// Lowest Performance
617  *      ResourceTemplate() {Register(PCC, 32, 0, 0x130, 2)},	// Guaranteed Performance Register
618  *      ResourceTemplate() {Register(PCC, 32, 0, 0x110, 2)},	// Desired Performance Register
619  *      ResourceTemplate() {Register(SystemMemory, 0, 0, 0, 0)},
620  *      ...
621  *      ...
622  *      ...
623  *  }
624  * Each Register() encodes how to access that specific register.
625  * e.g. a sample PCC entry has the following encoding:
626  *
627  *  Register (
628  *      PCC,	// AddressSpaceKeyword
629  *      8,	// RegisterBitWidth
630  *      8,	// RegisterBitOffset
631  *      0x30,	// RegisterAddress
632  *      9,	// AccessSize (subspace ID)
633  *  )
634  */
635 
636 #ifndef init_freq_invariance_cppc
637 static inline void init_freq_invariance_cppc(void) { }
638 #endif
639 
640 /**
641  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
642  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
643  *
644  *	Return: 0 for success or negative value for err.
645  */
646 int acpi_cppc_processor_probe(struct acpi_processor *pr)
647 {
648 	struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
649 	union acpi_object *out_obj, *cpc_obj;
650 	struct cpc_desc *cpc_ptr;
651 	struct cpc_reg *gas_t;
652 	struct device *cpu_dev;
653 	acpi_handle handle = pr->handle;
654 	unsigned int num_ent, i, cpc_rev;
655 	int pcc_subspace_id = -1;
656 	acpi_status status;
657 	int ret = -EFAULT;
658 
659 	/* Parse the ACPI _CPC table for this CPU. */
660 	status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
661 			ACPI_TYPE_PACKAGE);
662 	if (ACPI_FAILURE(status)) {
663 		ret = -ENODEV;
664 		goto out_buf_free;
665 	}
666 
667 	out_obj = (union acpi_object *) output.pointer;
668 
669 	cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
670 	if (!cpc_ptr) {
671 		ret = -ENOMEM;
672 		goto out_buf_free;
673 	}
674 
675 	/* First entry is NumEntries. */
676 	cpc_obj = &out_obj->package.elements[0];
677 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
678 		num_ent = cpc_obj->integer.value;
679 	} else {
680 		pr_debug("Unexpected entry type(%d) for NumEntries\n",
681 				cpc_obj->type);
682 		goto out_free;
683 	}
684 	cpc_ptr->num_entries = num_ent;
685 
686 	/* Second entry should be revision. */
687 	cpc_obj = &out_obj->package.elements[1];
688 	if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
689 		cpc_rev = cpc_obj->integer.value;
690 	} else {
691 		pr_debug("Unexpected entry type(%d) for Revision\n",
692 				cpc_obj->type);
693 		goto out_free;
694 	}
695 	cpc_ptr->version = cpc_rev;
696 
697 	if (!is_cppc_supported(cpc_rev, num_ent))
698 		goto out_free;
699 
700 	/* Iterate through remaining entries in _CPC */
701 	for (i = 2; i < num_ent; i++) {
702 		cpc_obj = &out_obj->package.elements[i];
703 
704 		if (cpc_obj->type == ACPI_TYPE_INTEGER)	{
705 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
706 			cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
707 		} else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
708 			gas_t = (struct cpc_reg *)
709 				cpc_obj->buffer.pointer;
710 
711 			/*
712 			 * The PCC Subspace index is encoded inside
713 			 * the CPC table entries. The same PCC index
714 			 * will be used for all the PCC entries,
715 			 * so extract it only once.
716 			 */
717 			if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
718 				if (pcc_subspace_id < 0) {
719 					pcc_subspace_id = gas_t->access_width;
720 					if (pcc_data_alloc(pcc_subspace_id))
721 						goto out_free;
722 				} else if (pcc_subspace_id != gas_t->access_width) {
723 					pr_debug("Mismatched PCC ids.\n");
724 					goto out_free;
725 				}
726 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
727 				if (gas_t->address) {
728 					void __iomem *addr;
729 
730 					addr = ioremap(gas_t->address, gas_t->bit_width/8);
731 					if (!addr)
732 						goto out_free;
733 					cpc_ptr->cpc_regs[i-2].sys_mem_vaddr = addr;
734 				}
735 			} else if (gas_t->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
736 				if (gas_t->access_width < 1 || gas_t->access_width > 3) {
737 					/*
738 					 * 1 = 8-bit, 2 = 16-bit, and 3 = 32-bit.
739 					 * SystemIO doesn't implement 64-bit
740 					 * registers.
741 					 */
742 					pr_debug("Invalid access width %d for SystemIO register\n",
743 						gas_t->access_width);
744 					goto out_free;
745 				}
746 				if (gas_t->address & OVER_16BTS_MASK) {
747 					/* SystemIO registers use 16-bit integer addresses */
748 					pr_debug("Invalid IO port %llu for SystemIO register\n",
749 						gas_t->address);
750 					goto out_free;
751 				}
752 			} else {
753 				if (gas_t->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE || !cpc_ffh_supported()) {
754 					/* Support only PCC, SystemMemory, SystemIO, and FFH type regs. */
755 					pr_debug("Unsupported register type: %d\n", gas_t->space_id);
756 					goto out_free;
757 				}
758 			}
759 
760 			cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
761 			memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
762 		} else {
763 			pr_debug("Err in entry:%d in CPC table of CPU:%d\n", i, pr->id);
764 			goto out_free;
765 		}
766 	}
767 	per_cpu(cpu_pcc_subspace_idx, pr->id) = pcc_subspace_id;
768 
769 	/*
770 	 * Initialize the remaining cpc_regs as unsupported.
771 	 * Example: In case FW exposes CPPC v2, the below loop will initialize
772 	 * LOWEST_FREQ and NOMINAL_FREQ regs as unsupported
773 	 */
774 	for (i = num_ent - 2; i < MAX_CPC_REG_ENT; i++) {
775 		cpc_ptr->cpc_regs[i].type = ACPI_TYPE_INTEGER;
776 		cpc_ptr->cpc_regs[i].cpc_entry.int_value = 0;
777 	}
778 
779 
780 	/* Store CPU Logical ID */
781 	cpc_ptr->cpu_id = pr->id;
782 
783 	/* Parse PSD data for this CPU */
784 	ret = acpi_get_psd(cpc_ptr, handle);
785 	if (ret)
786 		goto out_free;
787 
788 	/* Register PCC channel once for all PCC subspace ID. */
789 	if (pcc_subspace_id >= 0 && !pcc_data[pcc_subspace_id]->pcc_channel_acquired) {
790 		ret = register_pcc_channel(pcc_subspace_id);
791 		if (ret)
792 			goto out_free;
793 
794 		init_rwsem(&pcc_data[pcc_subspace_id]->pcc_lock);
795 		init_waitqueue_head(&pcc_data[pcc_subspace_id]->pcc_write_wait_q);
796 	}
797 
798 	/* Everything looks okay */
799 	pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
800 
801 	/* Add per logical CPU nodes for reading its feedback counters. */
802 	cpu_dev = get_cpu_device(pr->id);
803 	if (!cpu_dev) {
804 		ret = -EINVAL;
805 		goto out_free;
806 	}
807 
808 	/* Plug PSD data into this CPU's CPC descriptor. */
809 	per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
810 
811 	ret = kobject_init_and_add(&cpc_ptr->kobj, &cppc_ktype, &cpu_dev->kobj,
812 			"acpi_cppc");
813 	if (ret) {
814 		per_cpu(cpc_desc_ptr, pr->id) = NULL;
815 		kobject_put(&cpc_ptr->kobj);
816 		goto out_free;
817 	}
818 
819 	init_freq_invariance_cppc();
820 
821 	kfree(output.pointer);
822 	return 0;
823 
824 out_free:
825 	/* Free all the mapped sys mem areas for this CPU */
826 	for (i = 2; i < cpc_ptr->num_entries; i++) {
827 		void __iomem *addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
828 
829 		if (addr)
830 			iounmap(addr);
831 	}
832 	kfree(cpc_ptr);
833 
834 out_buf_free:
835 	kfree(output.pointer);
836 	return ret;
837 }
838 EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
839 
840 /**
841  * acpi_cppc_processor_exit - Cleanup CPC structs.
842  * @pr: Ptr to acpi_processor containing this CPU's logical ID.
843  *
844  * Return: Void
845  */
846 void acpi_cppc_processor_exit(struct acpi_processor *pr)
847 {
848 	struct cpc_desc *cpc_ptr;
849 	unsigned int i;
850 	void __iomem *addr;
851 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, pr->id);
852 
853 	if (pcc_ss_id >= 0 && pcc_data[pcc_ss_id]) {
854 		if (pcc_data[pcc_ss_id]->pcc_channel_acquired) {
855 			pcc_data[pcc_ss_id]->refcount--;
856 			if (!pcc_data[pcc_ss_id]->refcount) {
857 				pcc_mbox_free_channel(pcc_data[pcc_ss_id]->pcc_channel);
858 				kfree(pcc_data[pcc_ss_id]);
859 				pcc_data[pcc_ss_id] = NULL;
860 			}
861 		}
862 	}
863 
864 	cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
865 	if (!cpc_ptr)
866 		return;
867 
868 	/* Free all the mapped sys mem areas for this CPU */
869 	for (i = 2; i < cpc_ptr->num_entries; i++) {
870 		addr = cpc_ptr->cpc_regs[i-2].sys_mem_vaddr;
871 		if (addr)
872 			iounmap(addr);
873 	}
874 
875 	kobject_put(&cpc_ptr->kobj);
876 	kfree(cpc_ptr);
877 }
878 EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
879 
880 /**
881  * cpc_read_ffh() - Read FFH register
882  * @cpunum:	CPU number to read
883  * @reg:	cppc register information
884  * @val:	place holder for return value
885  *
886  * Read bit_width bits from a specified address and bit_offset
887  *
888  * Return: 0 for success and error code
889  */
890 int __weak cpc_read_ffh(int cpunum, struct cpc_reg *reg, u64 *val)
891 {
892 	return -ENOTSUPP;
893 }
894 
895 /**
896  * cpc_write_ffh() - Write FFH register
897  * @cpunum:	CPU number to write
898  * @reg:	cppc register information
899  * @val:	value to write
900  *
901  * Write value of bit_width bits to a specified address and bit_offset
902  *
903  * Return: 0 for success and error code
904  */
905 int __weak cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
906 {
907 	return -ENOTSUPP;
908 }
909 
910 /*
911  * Since cpc_read and cpc_write are called while holding pcc_lock, it should be
912  * as fast as possible. We have already mapped the PCC subspace during init, so
913  * we can directly write to it.
914  */
915 
916 static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
917 {
918 	void __iomem *vaddr = NULL;
919 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
920 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
921 
922 	if (reg_res->type == ACPI_TYPE_INTEGER) {
923 		*val = reg_res->cpc_entry.int_value;
924 		return 0;
925 	}
926 
927 	*val = 0;
928 
929 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
930 		u32 width = 8 << (reg->access_width - 1);
931 		u32 val_u32;
932 		acpi_status status;
933 
934 		status = acpi_os_read_port((acpi_io_address)reg->address,
935 					   &val_u32, width);
936 		if (ACPI_FAILURE(status)) {
937 			pr_debug("Error: Failed to read SystemIO port %llx\n",
938 				 reg->address);
939 			return -EFAULT;
940 		}
941 
942 		*val = val_u32;
943 		return 0;
944 	} else if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
945 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
946 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
947 		vaddr = reg_res->sys_mem_vaddr;
948 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
949 		return cpc_read_ffh(cpu, reg, val);
950 	else
951 		return acpi_os_read_memory((acpi_physical_address)reg->address,
952 				val, reg->bit_width);
953 
954 	switch (reg->bit_width) {
955 	case 8:
956 		*val = readb_relaxed(vaddr);
957 		break;
958 	case 16:
959 		*val = readw_relaxed(vaddr);
960 		break;
961 	case 32:
962 		*val = readl_relaxed(vaddr);
963 		break;
964 	case 64:
965 		*val = readq_relaxed(vaddr);
966 		break;
967 	default:
968 		pr_debug("Error: Cannot read %u bit width from PCC for ss: %d\n",
969 			 reg->bit_width, pcc_ss_id);
970 		return -EFAULT;
971 	}
972 
973 	return 0;
974 }
975 
976 static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
977 {
978 	int ret_val = 0;
979 	void __iomem *vaddr = NULL;
980 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
981 	struct cpc_reg *reg = &reg_res->cpc_entry.reg;
982 
983 	if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
984 		u32 width = 8 << (reg->access_width - 1);
985 		acpi_status status;
986 
987 		status = acpi_os_write_port((acpi_io_address)reg->address,
988 					    (u32)val, width);
989 		if (ACPI_FAILURE(status)) {
990 			pr_debug("Error: Failed to write SystemIO port %llx\n",
991 				 reg->address);
992 			return -EFAULT;
993 		}
994 
995 		return 0;
996 	} else if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM && pcc_ss_id >= 0)
997 		vaddr = GET_PCC_VADDR(reg->address, pcc_ss_id);
998 	else if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
999 		vaddr = reg_res->sys_mem_vaddr;
1000 	else if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE)
1001 		return cpc_write_ffh(cpu, reg, val);
1002 	else
1003 		return acpi_os_write_memory((acpi_physical_address)reg->address,
1004 				val, reg->bit_width);
1005 
1006 	switch (reg->bit_width) {
1007 	case 8:
1008 		writeb_relaxed(val, vaddr);
1009 		break;
1010 	case 16:
1011 		writew_relaxed(val, vaddr);
1012 		break;
1013 	case 32:
1014 		writel_relaxed(val, vaddr);
1015 		break;
1016 	case 64:
1017 		writeq_relaxed(val, vaddr);
1018 		break;
1019 	default:
1020 		pr_debug("Error: Cannot write %u bit width to PCC for ss: %d\n",
1021 			 reg->bit_width, pcc_ss_id);
1022 		ret_val = -EFAULT;
1023 		break;
1024 	}
1025 
1026 	return ret_val;
1027 }
1028 
1029 static int cppc_get_perf(int cpunum, enum cppc_regs reg_idx, u64 *perf)
1030 {
1031 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
1032 	struct cpc_register_resource *reg;
1033 
1034 	if (!cpc_desc) {
1035 		pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
1036 		return -ENODEV;
1037 	}
1038 
1039 	reg = &cpc_desc->cpc_regs[reg_idx];
1040 
1041 	if (CPC_IN_PCC(reg)) {
1042 		int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
1043 		struct cppc_pcc_data *pcc_ss_data = NULL;
1044 		int ret = 0;
1045 
1046 		if (pcc_ss_id < 0)
1047 			return -EIO;
1048 
1049 		pcc_ss_data = pcc_data[pcc_ss_id];
1050 
1051 		down_write(&pcc_ss_data->pcc_lock);
1052 
1053 		if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
1054 			cpc_read(cpunum, reg, perf);
1055 		else
1056 			ret = -EIO;
1057 
1058 		up_write(&pcc_ss_data->pcc_lock);
1059 
1060 		return ret;
1061 	}
1062 
1063 	cpc_read(cpunum, reg, perf);
1064 
1065 	return 0;
1066 }
1067 
1068 /**
1069  * cppc_get_desired_perf - Get the desired performance register value.
1070  * @cpunum: CPU from which to get desired performance.
1071  * @desired_perf: Return address.
1072  *
1073  * Return: 0 for success, -EIO otherwise.
1074  */
1075 int cppc_get_desired_perf(int cpunum, u64 *desired_perf)
1076 {
1077 	return cppc_get_perf(cpunum, DESIRED_PERF, desired_perf);
1078 }
1079 EXPORT_SYMBOL_GPL(cppc_get_desired_perf);
1080 
1081 /**
1082  * cppc_get_nominal_perf - Get the nominal performance register value.
1083  * @cpunum: CPU from which to get nominal performance.
1084  * @nominal_perf: Return address.
1085  *
1086  * Return: 0 for success, -EIO otherwise.
1087  */
1088 int cppc_get_nominal_perf(int cpunum, u64 *nominal_perf)
1089 {
1090 	return cppc_get_perf(cpunum, NOMINAL_PERF, nominal_perf);
1091 }
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_enable - Set to enable CPPC on the processor by writing the
1265  * Continuous Performance Control package EnableRegister field.
1266  * @cpu: CPU for which to enable CPPC register.
1267  * @enable: 0 - disable, 1 - enable CPPC feature on the processor.
1268  *
1269  * Return: 0 for success, -ERRNO or -EIO otherwise.
1270  */
1271 int cppc_set_enable(int cpu, bool enable)
1272 {
1273 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1274 	struct cpc_register_resource *enable_reg;
1275 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1276 	struct cppc_pcc_data *pcc_ss_data = NULL;
1277 	int ret = -EINVAL;
1278 
1279 	if (!cpc_desc) {
1280 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1281 		return -EINVAL;
1282 	}
1283 
1284 	enable_reg = &cpc_desc->cpc_regs[ENABLE];
1285 
1286 	if (CPC_IN_PCC(enable_reg)) {
1287 
1288 		if (pcc_ss_id < 0)
1289 			return -EIO;
1290 
1291 		ret = cpc_write(cpu, enable_reg, enable);
1292 		if (ret)
1293 			return ret;
1294 
1295 		pcc_ss_data = pcc_data[pcc_ss_id];
1296 
1297 		down_write(&pcc_ss_data->pcc_lock);
1298 		/* after writing CPC, transfer the ownership of PCC to platfrom */
1299 		ret = send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1300 		up_write(&pcc_ss_data->pcc_lock);
1301 		return ret;
1302 	}
1303 
1304 	return cpc_write(cpu, enable_reg, enable);
1305 }
1306 EXPORT_SYMBOL_GPL(cppc_set_enable);
1307 
1308 /**
1309  * cppc_set_perf - Set a CPU's performance controls.
1310  * @cpu: CPU for which to set performance controls.
1311  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
1312  *
1313  * Return: 0 for success, -ERRNO otherwise.
1314  */
1315 int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
1316 {
1317 	struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
1318 	struct cpc_register_resource *desired_reg;
1319 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
1320 	struct cppc_pcc_data *pcc_ss_data = NULL;
1321 	int ret = 0;
1322 
1323 	if (!cpc_desc) {
1324 		pr_debug("No CPC descriptor for CPU:%d\n", cpu);
1325 		return -ENODEV;
1326 	}
1327 
1328 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1329 
1330 	/*
1331 	 * This is Phase-I where we want to write to CPC registers
1332 	 * -> We want all CPUs to be able to execute this phase in parallel
1333 	 *
1334 	 * Since read_lock can be acquired by multiple CPUs simultaneously we
1335 	 * achieve that goal here
1336 	 */
1337 	if (CPC_IN_PCC(desired_reg)) {
1338 		if (pcc_ss_id < 0) {
1339 			pr_debug("Invalid pcc_ss_id\n");
1340 			return -ENODEV;
1341 		}
1342 		pcc_ss_data = pcc_data[pcc_ss_id];
1343 		down_read(&pcc_ss_data->pcc_lock); /* BEGIN Phase-I */
1344 		if (pcc_ss_data->platform_owns_pcc) {
1345 			ret = check_pcc_chan(pcc_ss_id, false);
1346 			if (ret) {
1347 				up_read(&pcc_ss_data->pcc_lock);
1348 				return ret;
1349 			}
1350 		}
1351 		/*
1352 		 * Update the pending_write to make sure a PCC CMD_READ will not
1353 		 * arrive and steal the channel during the switch to write lock
1354 		 */
1355 		pcc_ss_data->pending_pcc_write_cmd = true;
1356 		cpc_desc->write_cmd_id = pcc_ss_data->pcc_write_cnt;
1357 		cpc_desc->write_cmd_status = 0;
1358 	}
1359 
1360 	/*
1361 	 * Skip writing MIN/MAX until Linux knows how to come up with
1362 	 * useful values.
1363 	 */
1364 	cpc_write(cpu, desired_reg, perf_ctrls->desired_perf);
1365 
1366 	if (CPC_IN_PCC(desired_reg))
1367 		up_read(&pcc_ss_data->pcc_lock);	/* END Phase-I */
1368 	/*
1369 	 * This is Phase-II where we transfer the ownership of PCC to Platform
1370 	 *
1371 	 * Short Summary: Basically if we think of a group of cppc_set_perf
1372 	 * requests that happened in short overlapping interval. The last CPU to
1373 	 * come out of Phase-I will enter Phase-II and ring the doorbell.
1374 	 *
1375 	 * We have the following requirements for Phase-II:
1376 	 *     1. We want to execute Phase-II only when there are no CPUs
1377 	 * currently executing in Phase-I
1378 	 *     2. Once we start Phase-II we want to avoid all other CPUs from
1379 	 * entering Phase-I.
1380 	 *     3. We want only one CPU among all those who went through Phase-I
1381 	 * to run phase-II
1382 	 *
1383 	 * If write_trylock fails to get the lock and doesn't transfer the
1384 	 * PCC ownership to the platform, then one of the following will be TRUE
1385 	 *     1. There is at-least one CPU in Phase-I which will later execute
1386 	 * write_trylock, so the CPUs in Phase-I will be responsible for
1387 	 * executing the Phase-II.
1388 	 *     2. Some other CPU has beaten this CPU to successfully execute the
1389 	 * write_trylock and has already acquired the write_lock. We know for a
1390 	 * fact it (other CPU acquiring the write_lock) couldn't have happened
1391 	 * before this CPU's Phase-I as we held the read_lock.
1392 	 *     3. Some other CPU executing pcc CMD_READ has stolen the
1393 	 * down_write, in which case, send_pcc_cmd will check for pending
1394 	 * CMD_WRITE commands by checking the pending_pcc_write_cmd.
1395 	 * So this CPU can be certain that its request will be delivered
1396 	 *    So in all cases, this CPU knows that its request will be delivered
1397 	 * by another CPU and can return
1398 	 *
1399 	 * After getting the down_write we still need to check for
1400 	 * pending_pcc_write_cmd to take care of the following scenario
1401 	 *    The thread running this code could be scheduled out between
1402 	 * Phase-I and Phase-II. Before it is scheduled back on, another CPU
1403 	 * could have delivered the request to Platform by triggering the
1404 	 * doorbell and transferred the ownership of PCC to platform. So this
1405 	 * avoids triggering an unnecessary doorbell and more importantly before
1406 	 * triggering the doorbell it makes sure that the PCC channel ownership
1407 	 * is still with OSPM.
1408 	 *   pending_pcc_write_cmd can also be cleared by a different CPU, if
1409 	 * there was a pcc CMD_READ waiting on down_write and it steals the lock
1410 	 * before the pcc CMD_WRITE is completed. send_pcc_cmd checks for this
1411 	 * case during a CMD_READ and if there are pending writes it delivers
1412 	 * the write command before servicing the read command
1413 	 */
1414 	if (CPC_IN_PCC(desired_reg)) {
1415 		if (down_write_trylock(&pcc_ss_data->pcc_lock)) {/* BEGIN Phase-II */
1416 			/* Update only if there are pending write commands */
1417 			if (pcc_ss_data->pending_pcc_write_cmd)
1418 				send_pcc_cmd(pcc_ss_id, CMD_WRITE);
1419 			up_write(&pcc_ss_data->pcc_lock);	/* END Phase-II */
1420 		} else
1421 			/* Wait until pcc_write_cnt is updated by send_pcc_cmd */
1422 			wait_event(pcc_ss_data->pcc_write_wait_q,
1423 				   cpc_desc->write_cmd_id != pcc_ss_data->pcc_write_cnt);
1424 
1425 		/* send_pcc_cmd updates the status in case of failure */
1426 		ret = cpc_desc->write_cmd_status;
1427 	}
1428 	return ret;
1429 }
1430 EXPORT_SYMBOL_GPL(cppc_set_perf);
1431 
1432 /**
1433  * cppc_get_transition_latency - returns frequency transition latency in ns
1434  *
1435  * ACPI CPPC does not explicitly specify how a platform can specify the
1436  * transition latency for performance change requests. The closest we have
1437  * is the timing information from the PCCT tables which provides the info
1438  * on the number and frequency of PCC commands the platform can handle.
1439  */
1440 unsigned int cppc_get_transition_latency(int cpu_num)
1441 {
1442 	/*
1443 	 * Expected transition latency is based on the PCCT timing values
1444 	 * Below are definition from ACPI spec:
1445 	 * pcc_nominal- Expected latency to process a command, in microseconds
1446 	 * pcc_mpar   - The maximum number of periodic requests that the subspace
1447 	 *              channel can support, reported in commands per minute. 0
1448 	 *              indicates no limitation.
1449 	 * pcc_mrtt   - The minimum amount of time that OSPM must wait after the
1450 	 *              completion of a command before issuing the next command,
1451 	 *              in microseconds.
1452 	 */
1453 	unsigned int latency_ns = 0;
1454 	struct cpc_desc *cpc_desc;
1455 	struct cpc_register_resource *desired_reg;
1456 	int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu_num);
1457 	struct cppc_pcc_data *pcc_ss_data;
1458 
1459 	cpc_desc = per_cpu(cpc_desc_ptr, cpu_num);
1460 	if (!cpc_desc)
1461 		return CPUFREQ_ETERNAL;
1462 
1463 	desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
1464 	if (!CPC_IN_PCC(desired_reg))
1465 		return CPUFREQ_ETERNAL;
1466 
1467 	if (pcc_ss_id < 0)
1468 		return CPUFREQ_ETERNAL;
1469 
1470 	pcc_ss_data = pcc_data[pcc_ss_id];
1471 	if (pcc_ss_data->pcc_mpar)
1472 		latency_ns = 60 * (1000 * 1000 * 1000 / pcc_ss_data->pcc_mpar);
1473 
1474 	latency_ns = max(latency_ns, pcc_ss_data->pcc_nominal * 1000);
1475 	latency_ns = max(latency_ns, pcc_ss_data->pcc_mrtt * 1000);
1476 
1477 	return latency_ns;
1478 }
1479 EXPORT_SYMBOL_GPL(cppc_get_transition_latency);
1480