1 /*
2  * Copyright 2012 by Oracle Inc
3  * Author: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
4  *
5  * This code borrows ideas from https://lkml.org/lkml/2011/11/30/249
6  * so many thanks go to Kevin Tian <kevin.tian@intel.com>
7  * and Yu Ke <ke.yu@intel.com>.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms and conditions of the GNU General Public License,
11  * version 2, as published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  */
19 
20 #include <linux/cpumask.h>
21 #include <linux/cpufreq.h>
22 #include <linux/freezer.h>
23 #include <linux/kernel.h>
24 #include <linux/kthread.h>
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <acpi/acpi_bus.h>
29 #include <acpi/acpi_drivers.h>
30 #include <acpi/processor.h>
31 
32 #include <xen/xen.h>
33 #include <xen/interface/platform.h>
34 #include <asm/xen/hypercall.h>
35 
36 #define DRV_NAME "xen-acpi-processor: "
37 
38 static int no_hypercall;
39 MODULE_PARM_DESC(off, "Inhibit the hypercall.");
40 module_param_named(off, no_hypercall, int, 0400);
41 
42 /*
43  * Note: Do not convert the acpi_id* below to cpumask_var_t or use cpumask_bit
44  * - as those shrink to nr_cpu_bits (which is dependent on possible_cpu), which
45  * can be less than what we want to put in. Instead use the 'nr_acpi_bits'
46  * which is dynamically computed based on the MADT or x2APIC table.
47  */
48 static unsigned int nr_acpi_bits;
49 /* Mutex to protect the acpi_ids_done - for CPU hotplug use. */
50 static DEFINE_MUTEX(acpi_ids_mutex);
51 /* Which ACPI ID we have processed from 'struct acpi_processor'. */
52 static unsigned long *acpi_ids_done;
53 /* Which ACPI ID exist in the SSDT/DSDT processor definitions. */
54 static unsigned long __initdata *acpi_id_present;
55 /* And if there is an _CST definition (or a PBLK) for the ACPI IDs */
56 static unsigned long __initdata *acpi_id_cst_present;
57 
58 static int push_cxx_to_hypervisor(struct acpi_processor *_pr)
59 {
60 	struct xen_platform_op op = {
61 		.cmd			= XENPF_set_processor_pminfo,
62 		.interface_version	= XENPF_INTERFACE_VERSION,
63 		.u.set_pminfo.id	= _pr->acpi_id,
64 		.u.set_pminfo.type	= XEN_PM_CX,
65 	};
66 	struct xen_processor_cx *dst_cx, *dst_cx_states = NULL;
67 	struct acpi_processor_cx *cx;
68 	unsigned int i, ok;
69 	int ret = 0;
70 
71 	dst_cx_states = kcalloc(_pr->power.count,
72 				sizeof(struct xen_processor_cx), GFP_KERNEL);
73 	if (!dst_cx_states)
74 		return -ENOMEM;
75 
76 	for (ok = 0, i = 1; i <= _pr->power.count; i++) {
77 		cx = &_pr->power.states[i];
78 		if (!cx->valid)
79 			continue;
80 
81 		dst_cx = &(dst_cx_states[ok++]);
82 
83 		dst_cx->reg.space_id = ACPI_ADR_SPACE_SYSTEM_IO;
84 		if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
85 			dst_cx->reg.bit_width = 8;
86 			dst_cx->reg.bit_offset = 0;
87 			dst_cx->reg.access_size = 1;
88 		} else {
89 			dst_cx->reg.space_id = ACPI_ADR_SPACE_FIXED_HARDWARE;
90 			if (cx->entry_method == ACPI_CSTATE_FFH) {
91 				/* NATIVE_CSTATE_BEYOND_HALT */
92 				dst_cx->reg.bit_offset = 2;
93 				dst_cx->reg.bit_width = 1; /* VENDOR_INTEL */
94 			}
95 			dst_cx->reg.access_size = 0;
96 		}
97 		dst_cx->reg.address = cx->address;
98 
99 		dst_cx->type = cx->type;
100 		dst_cx->latency = cx->latency;
101 		dst_cx->power = cx->power;
102 
103 		dst_cx->dpcnt = 0;
104 		set_xen_guest_handle(dst_cx->dp, NULL);
105 	}
106 	if (!ok) {
107 		pr_debug(DRV_NAME "No _Cx for ACPI CPU %u\n", _pr->acpi_id);
108 		kfree(dst_cx_states);
109 		return -EINVAL;
110 	}
111 	op.u.set_pminfo.power.count = ok;
112 	op.u.set_pminfo.power.flags.bm_control = _pr->flags.bm_control;
113 	op.u.set_pminfo.power.flags.bm_check = _pr->flags.bm_check;
114 	op.u.set_pminfo.power.flags.has_cst = _pr->flags.has_cst;
115 	op.u.set_pminfo.power.flags.power_setup_done =
116 		_pr->flags.power_setup_done;
117 
118 	set_xen_guest_handle(op.u.set_pminfo.power.states, dst_cx_states);
119 
120 	if (!no_hypercall)
121 		ret = HYPERVISOR_dom0_op(&op);
122 
123 	if (!ret) {
124 		pr_debug("ACPI CPU%u - C-states uploaded.\n", _pr->acpi_id);
125 		for (i = 1; i <= _pr->power.count; i++) {
126 			cx = &_pr->power.states[i];
127 			if (!cx->valid)
128 				continue;
129 			pr_debug("     C%d: %s %d uS\n",
130 				 cx->type, cx->desc, (u32)cx->latency);
131 		}
132 	} else if (ret != -EINVAL)
133 		/* EINVAL means the ACPI ID is incorrect - meaning the ACPI
134 		 * table is referencing a non-existing CPU - which can happen
135 		 * with broken ACPI tables. */
136 		pr_err(DRV_NAME "(CX): Hypervisor error (%d) for ACPI CPU%u\n",
137 		       ret, _pr->acpi_id);
138 
139 	kfree(dst_cx_states);
140 
141 	return ret;
142 }
143 static struct xen_processor_px *
144 xen_copy_pss_data(struct acpi_processor *_pr,
145 		  struct xen_processor_performance *dst_perf)
146 {
147 	struct xen_processor_px *dst_states = NULL;
148 	unsigned int i;
149 
150 	BUILD_BUG_ON(sizeof(struct xen_processor_px) !=
151 		     sizeof(struct acpi_processor_px));
152 
153 	dst_states = kcalloc(_pr->performance->state_count,
154 			     sizeof(struct xen_processor_px), GFP_KERNEL);
155 	if (!dst_states)
156 		return ERR_PTR(-ENOMEM);
157 
158 	dst_perf->state_count = _pr->performance->state_count;
159 	for (i = 0; i < _pr->performance->state_count; i++) {
160 		/* Fortunatly for us, they are both the same size */
161 		memcpy(&(dst_states[i]), &(_pr->performance->states[i]),
162 		       sizeof(struct acpi_processor_px));
163 	}
164 	return dst_states;
165 }
166 static int xen_copy_psd_data(struct acpi_processor *_pr,
167 			     struct xen_processor_performance *dst)
168 {
169 	struct acpi_psd_package *pdomain;
170 
171 	BUILD_BUG_ON(sizeof(struct xen_psd_package) !=
172 		     sizeof(struct acpi_psd_package));
173 
174 	/* This information is enumerated only if acpi_processor_preregister_performance
175 	 * has been called.
176 	 */
177 	dst->shared_type = _pr->performance->shared_type;
178 
179 	pdomain = &(_pr->performance->domain_info);
180 
181 	/* 'acpi_processor_preregister_performance' does not parse if the
182 	 * num_processors <= 1, but Xen still requires it. Do it manually here.
183 	 */
184 	if (pdomain->num_processors <= 1) {
185 		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
186 			dst->shared_type = CPUFREQ_SHARED_TYPE_ALL;
187 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
188 			dst->shared_type = CPUFREQ_SHARED_TYPE_HW;
189 		else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
190 			dst->shared_type = CPUFREQ_SHARED_TYPE_ANY;
191 
192 	}
193 	memcpy(&(dst->domain_info), pdomain, sizeof(struct acpi_psd_package));
194 	return 0;
195 }
196 static int xen_copy_pct_data(struct acpi_pct_register *pct,
197 			     struct xen_pct_register *dst_pct)
198 {
199 	/* It would be nice if you could just do 'memcpy(pct, dst_pct') but
200 	 * sadly the Xen structure did not have the proper padding so the
201 	 * descriptor field takes two (dst_pct) bytes instead of one (pct).
202 	 */
203 	dst_pct->descriptor = pct->descriptor;
204 	dst_pct->length = pct->length;
205 	dst_pct->space_id = pct->space_id;
206 	dst_pct->bit_width = pct->bit_width;
207 	dst_pct->bit_offset = pct->bit_offset;
208 	dst_pct->reserved = pct->reserved;
209 	dst_pct->address = pct->address;
210 	return 0;
211 }
212 static int push_pxx_to_hypervisor(struct acpi_processor *_pr)
213 {
214 	int ret = 0;
215 	struct xen_platform_op op = {
216 		.cmd			= XENPF_set_processor_pminfo,
217 		.interface_version	= XENPF_INTERFACE_VERSION,
218 		.u.set_pminfo.id	= _pr->acpi_id,
219 		.u.set_pminfo.type	= XEN_PM_PX,
220 	};
221 	struct xen_processor_performance *dst_perf;
222 	struct xen_processor_px *dst_states = NULL;
223 
224 	dst_perf = &op.u.set_pminfo.perf;
225 
226 	dst_perf->platform_limit = _pr->performance_platform_limit;
227 	dst_perf->flags |= XEN_PX_PPC;
228 	xen_copy_pct_data(&(_pr->performance->control_register),
229 			  &dst_perf->control_register);
230 	xen_copy_pct_data(&(_pr->performance->status_register),
231 			  &dst_perf->status_register);
232 	dst_perf->flags |= XEN_PX_PCT;
233 	dst_states = xen_copy_pss_data(_pr, dst_perf);
234 	if (!IS_ERR_OR_NULL(dst_states)) {
235 		set_xen_guest_handle(dst_perf->states, dst_states);
236 		dst_perf->flags |= XEN_PX_PSS;
237 	}
238 	if (!xen_copy_psd_data(_pr, dst_perf))
239 		dst_perf->flags |= XEN_PX_PSD;
240 
241 	if (dst_perf->flags != (XEN_PX_PSD | XEN_PX_PSS | XEN_PX_PCT | XEN_PX_PPC)) {
242 		pr_warn(DRV_NAME "ACPI CPU%u missing some P-state data (%x), skipping.\n",
243 			_pr->acpi_id, dst_perf->flags);
244 		ret = -ENODEV;
245 		goto err_free;
246 	}
247 
248 	if (!no_hypercall)
249 		ret = HYPERVISOR_dom0_op(&op);
250 
251 	if (!ret) {
252 		struct acpi_processor_performance *perf;
253 		unsigned int i;
254 
255 		perf = _pr->performance;
256 		pr_debug("ACPI CPU%u - P-states uploaded.\n", _pr->acpi_id);
257 		for (i = 0; i < perf->state_count; i++) {
258 			pr_debug("     %cP%d: %d MHz, %d mW, %d uS\n",
259 			(i == perf->state ? '*' : ' '), i,
260 			(u32) perf->states[i].core_frequency,
261 			(u32) perf->states[i].power,
262 			(u32) perf->states[i].transition_latency);
263 		}
264 	} else if (ret != -EINVAL)
265 		/* EINVAL means the ACPI ID is incorrect - meaning the ACPI
266 		 * table is referencing a non-existing CPU - which can happen
267 		 * with broken ACPI tables. */
268 		pr_warn(DRV_NAME "(_PXX): Hypervisor error (%d) for ACPI CPU%u\n",
269 		       ret, _pr->acpi_id);
270 err_free:
271 	if (!IS_ERR_OR_NULL(dst_states))
272 		kfree(dst_states);
273 
274 	return ret;
275 }
276 static int upload_pm_data(struct acpi_processor *_pr)
277 {
278 	int err = 0;
279 
280 	mutex_lock(&acpi_ids_mutex);
281 	if (__test_and_set_bit(_pr->acpi_id, acpi_ids_done)) {
282 		mutex_unlock(&acpi_ids_mutex);
283 		return -EBUSY;
284 	}
285 	if (_pr->flags.power)
286 		err = push_cxx_to_hypervisor(_pr);
287 
288 	if (_pr->performance && _pr->performance->states)
289 		err |= push_pxx_to_hypervisor(_pr);
290 
291 	mutex_unlock(&acpi_ids_mutex);
292 	return err;
293 }
294 static unsigned int __init get_max_acpi_id(void)
295 {
296 	struct xenpf_pcpuinfo *info;
297 	struct xen_platform_op op = {
298 		.cmd = XENPF_get_cpuinfo,
299 		.interface_version = XENPF_INTERFACE_VERSION,
300 	};
301 	int ret = 0;
302 	unsigned int i, last_cpu, max_acpi_id = 0;
303 
304 	info = &op.u.pcpu_info;
305 	info->xen_cpuid = 0;
306 
307 	ret = HYPERVISOR_dom0_op(&op);
308 	if (ret)
309 		return NR_CPUS;
310 
311 	/* The max_present is the same irregardless of the xen_cpuid */
312 	last_cpu = op.u.pcpu_info.max_present;
313 	for (i = 0; i <= last_cpu; i++) {
314 		info->xen_cpuid = i;
315 		ret = HYPERVISOR_dom0_op(&op);
316 		if (ret)
317 			continue;
318 		max_acpi_id = max(info->acpi_id, max_acpi_id);
319 	}
320 	max_acpi_id *= 2; /* Slack for CPU hotplug support. */
321 	pr_debug(DRV_NAME "Max ACPI ID: %u\n", max_acpi_id);
322 	return max_acpi_id;
323 }
324 /*
325  * The read_acpi_id and check_acpi_ids are there to support the Xen
326  * oddity of virtual CPUs != physical CPUs in the initial domain.
327  * The user can supply 'xen_max_vcpus=X' on the Xen hypervisor line
328  * which will band the amount of CPUs the initial domain can see.
329  * In general that is OK, except it plays havoc with any of the
330  * for_each_[present|online]_cpu macros which are banded to the virtual
331  * CPU amount.
332  */
333 static acpi_status __init
334 read_acpi_id(acpi_handle handle, u32 lvl, void *context, void **rv)
335 {
336 	u32 acpi_id;
337 	acpi_status status;
338 	acpi_object_type acpi_type;
339 	unsigned long long tmp;
340 	union acpi_object object = { 0 };
341 	struct acpi_buffer buffer = { sizeof(union acpi_object), &object };
342 	acpi_io_address pblk = 0;
343 
344 	status = acpi_get_type(handle, &acpi_type);
345 	if (ACPI_FAILURE(status))
346 		return AE_OK;
347 
348 	switch (acpi_type) {
349 	case ACPI_TYPE_PROCESSOR:
350 		status = acpi_evaluate_object(handle, NULL, NULL, &buffer);
351 		if (ACPI_FAILURE(status))
352 			return AE_OK;
353 		acpi_id = object.processor.proc_id;
354 		pblk = object.processor.pblk_address;
355 		break;
356 	case ACPI_TYPE_DEVICE:
357 		status = acpi_evaluate_integer(handle, "_UID", NULL, &tmp);
358 		if (ACPI_FAILURE(status))
359 			return AE_OK;
360 		acpi_id = tmp;
361 		break;
362 	default:
363 		return AE_OK;
364 	}
365 	/* There are more ACPI Processor objects than in x2APIC or MADT.
366 	 * This can happen with incorrect ACPI SSDT declerations. */
367 	if (acpi_id > nr_acpi_bits) {
368 		pr_debug(DRV_NAME "We only have %u, trying to set %u\n",
369 			 nr_acpi_bits, acpi_id);
370 		return AE_OK;
371 	}
372 	/* OK, There is a ACPI Processor object */
373 	__set_bit(acpi_id, acpi_id_present);
374 
375 	pr_debug(DRV_NAME "ACPI CPU%u w/ PBLK:0x%lx\n", acpi_id,
376 		 (unsigned long)pblk);
377 
378 	status = acpi_evaluate_object(handle, "_CST", NULL, &buffer);
379 	if (ACPI_FAILURE(status)) {
380 		if (!pblk)
381 			return AE_OK;
382 	}
383 	/* .. and it has a C-state */
384 	__set_bit(acpi_id, acpi_id_cst_present);
385 
386 	return AE_OK;
387 }
388 static int __init check_acpi_ids(struct acpi_processor *pr_backup)
389 {
390 
391 	if (!pr_backup)
392 		return -ENODEV;
393 
394 	/* All online CPUs have been processed at this stage. Now verify
395 	 * whether in fact "online CPUs" == physical CPUs.
396 	 */
397 	acpi_id_present = kcalloc(BITS_TO_LONGS(nr_acpi_bits), sizeof(unsigned long), GFP_KERNEL);
398 	if (!acpi_id_present)
399 		return -ENOMEM;
400 
401 	acpi_id_cst_present = kcalloc(BITS_TO_LONGS(nr_acpi_bits), sizeof(unsigned long), GFP_KERNEL);
402 	if (!acpi_id_cst_present) {
403 		kfree(acpi_id_present);
404 		return -ENOMEM;
405 	}
406 
407 	acpi_walk_namespace(ACPI_TYPE_PROCESSOR, ACPI_ROOT_OBJECT,
408 			    ACPI_UINT32_MAX,
409 			    read_acpi_id, NULL, NULL, NULL);
410 	acpi_get_devices("ACPI0007", read_acpi_id, NULL, NULL);
411 
412 	if (!bitmap_equal(acpi_id_present, acpi_ids_done, nr_acpi_bits)) {
413 		unsigned int i;
414 		for_each_set_bit(i, acpi_id_present, nr_acpi_bits) {
415 			pr_backup->acpi_id = i;
416 			/* Mask out C-states if there are no _CST or PBLK */
417 			pr_backup->flags.power = test_bit(i, acpi_id_cst_present);
418 			(void)upload_pm_data(pr_backup);
419 		}
420 	}
421 	kfree(acpi_id_present);
422 	acpi_id_present = NULL;
423 	kfree(acpi_id_cst_present);
424 	acpi_id_cst_present = NULL;
425 	return 0;
426 }
427 static int __init check_prereq(void)
428 {
429 	struct cpuinfo_x86 *c = &cpu_data(0);
430 
431 	if (!xen_initial_domain())
432 		return -ENODEV;
433 
434 	if (!acpi_gbl_FADT.smi_command)
435 		return -ENODEV;
436 
437 	if (c->x86_vendor == X86_VENDOR_INTEL) {
438 		if (!cpu_has(c, X86_FEATURE_EST))
439 			return -ENODEV;
440 
441 		return 0;
442 	}
443 	if (c->x86_vendor == X86_VENDOR_AMD) {
444 		/* Copied from powernow-k8.h, can't include ../cpufreq/powernow
445 		 * as we get compile warnings for the static functions.
446 		 */
447 #define CPUID_FREQ_VOLT_CAPABILITIES    0x80000007
448 #define USE_HW_PSTATE                   0x00000080
449 		u32 eax, ebx, ecx, edx;
450 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
451 		if ((edx & USE_HW_PSTATE) != USE_HW_PSTATE)
452 			return -ENODEV;
453 		return 0;
454 	}
455 	return -ENODEV;
456 }
457 /* acpi_perf_data is a pointer to percpu data. */
458 static struct acpi_processor_performance __percpu *acpi_perf_data;
459 
460 static void free_acpi_perf_data(void)
461 {
462 	unsigned int i;
463 
464 	/* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
465 	for_each_possible_cpu(i)
466 		free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
467 				 ->shared_cpu_map);
468 	free_percpu(acpi_perf_data);
469 }
470 
471 static int __init xen_acpi_processor_init(void)
472 {
473 	struct acpi_processor *pr_backup = NULL;
474 	unsigned int i;
475 	int rc = check_prereq();
476 
477 	if (rc)
478 		return rc;
479 
480 	nr_acpi_bits = get_max_acpi_id() + 1;
481 	acpi_ids_done = kcalloc(BITS_TO_LONGS(nr_acpi_bits), sizeof(unsigned long), GFP_KERNEL);
482 	if (!acpi_ids_done)
483 		return -ENOMEM;
484 
485 	acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
486 	if (!acpi_perf_data) {
487 		pr_debug(DRV_NAME "Memory allocation error for acpi_perf_data.\n");
488 		kfree(acpi_ids_done);
489 		return -ENOMEM;
490 	}
491 	for_each_possible_cpu(i) {
492 		if (!zalloc_cpumask_var_node(
493 			&per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
494 			GFP_KERNEL, cpu_to_node(i))) {
495 			rc = -ENOMEM;
496 			goto err_out;
497 		}
498 	}
499 
500 	/* Do initialization in ACPI core. It is OK to fail here. */
501 	(void)acpi_processor_preregister_performance(acpi_perf_data);
502 
503 	for_each_possible_cpu(i) {
504 		struct acpi_processor_performance *perf;
505 
506 		perf = per_cpu_ptr(acpi_perf_data, i);
507 		rc = acpi_processor_register_performance(perf, i);
508 		if (rc)
509 			goto err_out;
510 	}
511 	rc = acpi_processor_notify_smm(THIS_MODULE);
512 	if (rc)
513 		goto err_unregister;
514 
515 	for_each_possible_cpu(i) {
516 		struct acpi_processor *_pr;
517 		_pr = per_cpu(processors, i /* APIC ID */);
518 		if (!_pr)
519 			continue;
520 
521 		if (!pr_backup) {
522 			pr_backup = kzalloc(sizeof(struct acpi_processor), GFP_KERNEL);
523 			if (pr_backup)
524 				memcpy(pr_backup, _pr, sizeof(struct acpi_processor));
525 		}
526 		(void)upload_pm_data(_pr);
527 	}
528 	rc = check_acpi_ids(pr_backup);
529 
530 	kfree(pr_backup);
531 	pr_backup = NULL;
532 
533 	if (rc)
534 		goto err_unregister;
535 
536 	return 0;
537 err_unregister:
538 	for_each_possible_cpu(i) {
539 		struct acpi_processor_performance *perf;
540 		perf = per_cpu_ptr(acpi_perf_data, i);
541 		acpi_processor_unregister_performance(perf, i);
542 	}
543 err_out:
544 	/* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
545 	free_acpi_perf_data();
546 	kfree(acpi_ids_done);
547 	return rc;
548 }
549 static void __exit xen_acpi_processor_exit(void)
550 {
551 	int i;
552 
553 	kfree(acpi_ids_done);
554 	for_each_possible_cpu(i) {
555 		struct acpi_processor_performance *perf;
556 		perf = per_cpu_ptr(acpi_perf_data, i);
557 		acpi_processor_unregister_performance(perf, i);
558 	}
559 	free_acpi_perf_data();
560 }
561 
562 MODULE_AUTHOR("Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>");
563 MODULE_DESCRIPTION("Xen ACPI Processor P-states (and Cx) driver which uploads PM data to Xen hypervisor");
564 MODULE_LICENSE("GPL");
565 
566 /* We want to be loaded before the CPU freq scaling drivers are loaded.
567  * They are loaded in late_initcall. */
568 device_initcall(xen_acpi_processor_init);
569 module_exit(xen_acpi_processor_exit);
570