xref: /openbmc/linux/arch/x86/kernel/cpu/resctrl/core.c (revision 6491d698)
1 /*
2  * Resource Director Technology(RDT)
3  * - Cache Allocation code.
4  *
5  * Copyright (C) 2016 Intel Corporation
6  *
7  * Authors:
8  *    Fenghua Yu <fenghua.yu@intel.com>
9  *    Tony Luck <tony.luck@intel.com>
10  *    Vikas Shivappa <vikas.shivappa@intel.com>
11  *
12  * This program is free software; you can redistribute it and/or modify it
13  * under the terms and conditions of the GNU General Public License,
14  * version 2, as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope it will be useful, but WITHOUT
17  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
18  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
19  * more details.
20  *
21  * More information about RDT be found in the Intel (R) x86 Architecture
22  * Software Developer Manual June 2016, volume 3, section 17.17.
23  */
24 
25 #define pr_fmt(fmt)	"resctrl: " fmt
26 
27 #include <linux/slab.h>
28 #include <linux/err.h>
29 #include <linux/cacheinfo.h>
30 #include <linux/cpuhotplug.h>
31 
32 #include <asm/intel-family.h>
33 #include <asm/resctrl_sched.h>
34 #include "internal.h"
35 
36 /* Mutex to protect rdtgroup access. */
37 DEFINE_MUTEX(rdtgroup_mutex);
38 
39 /*
40  * The cached resctrl_pqr_state is strictly per CPU and can never be
41  * updated from a remote CPU. Functions which modify the state
42  * are called with interrupts disabled and no preemption, which
43  * is sufficient for the protection.
44  */
45 DEFINE_PER_CPU(struct resctrl_pqr_state, pqr_state);
46 
47 /*
48  * Used to store the max resource name width and max resource data width
49  * to display the schemata in a tabular format
50  */
51 int max_name_width, max_data_width;
52 
53 /*
54  * Global boolean for rdt_alloc which is true if any
55  * resource allocation is enabled.
56  */
57 bool rdt_alloc_capable;
58 
59 static void
60 mba_wrmsr_intel(struct rdt_domain *d, struct msr_param *m,
61 		struct rdt_resource *r);
62 static void
63 cat_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r);
64 static void
65 mba_wrmsr_amd(struct rdt_domain *d, struct msr_param *m,
66 	      struct rdt_resource *r);
67 
68 #define domain_init(id) LIST_HEAD_INIT(rdt_resources_all[id].domains)
69 
70 struct rdt_resource rdt_resources_all[] = {
71 	[RDT_RESOURCE_L3] =
72 	{
73 		.rid			= RDT_RESOURCE_L3,
74 		.name			= "L3",
75 		.domains		= domain_init(RDT_RESOURCE_L3),
76 		.msr_base		= MSR_IA32_L3_CBM_BASE,
77 		.msr_update		= cat_wrmsr,
78 		.cache_level		= 3,
79 		.cache = {
80 			.min_cbm_bits	= 1,
81 			.cbm_idx_mult	= 1,
82 			.cbm_idx_offset	= 0,
83 		},
84 		.parse_ctrlval		= parse_cbm,
85 		.format_str		= "%d=%0*x",
86 		.fflags			= RFTYPE_RES_CACHE,
87 	},
88 	[RDT_RESOURCE_L3DATA] =
89 	{
90 		.rid			= RDT_RESOURCE_L3DATA,
91 		.name			= "L3DATA",
92 		.domains		= domain_init(RDT_RESOURCE_L3DATA),
93 		.msr_base		= MSR_IA32_L3_CBM_BASE,
94 		.msr_update		= cat_wrmsr,
95 		.cache_level		= 3,
96 		.cache = {
97 			.min_cbm_bits	= 1,
98 			.cbm_idx_mult	= 2,
99 			.cbm_idx_offset	= 0,
100 		},
101 		.parse_ctrlval		= parse_cbm,
102 		.format_str		= "%d=%0*x",
103 		.fflags			= RFTYPE_RES_CACHE,
104 	},
105 	[RDT_RESOURCE_L3CODE] =
106 	{
107 		.rid			= RDT_RESOURCE_L3CODE,
108 		.name			= "L3CODE",
109 		.domains		= domain_init(RDT_RESOURCE_L3CODE),
110 		.msr_base		= MSR_IA32_L3_CBM_BASE,
111 		.msr_update		= cat_wrmsr,
112 		.cache_level		= 3,
113 		.cache = {
114 			.min_cbm_bits	= 1,
115 			.cbm_idx_mult	= 2,
116 			.cbm_idx_offset	= 1,
117 		},
118 		.parse_ctrlval		= parse_cbm,
119 		.format_str		= "%d=%0*x",
120 		.fflags			= RFTYPE_RES_CACHE,
121 	},
122 	[RDT_RESOURCE_L2] =
123 	{
124 		.rid			= RDT_RESOURCE_L2,
125 		.name			= "L2",
126 		.domains		= domain_init(RDT_RESOURCE_L2),
127 		.msr_base		= MSR_IA32_L2_CBM_BASE,
128 		.msr_update		= cat_wrmsr,
129 		.cache_level		= 2,
130 		.cache = {
131 			.min_cbm_bits	= 1,
132 			.cbm_idx_mult	= 1,
133 			.cbm_idx_offset	= 0,
134 		},
135 		.parse_ctrlval		= parse_cbm,
136 		.format_str		= "%d=%0*x",
137 		.fflags			= RFTYPE_RES_CACHE,
138 	},
139 	[RDT_RESOURCE_L2DATA] =
140 	{
141 		.rid			= RDT_RESOURCE_L2DATA,
142 		.name			= "L2DATA",
143 		.domains		= domain_init(RDT_RESOURCE_L2DATA),
144 		.msr_base		= MSR_IA32_L2_CBM_BASE,
145 		.msr_update		= cat_wrmsr,
146 		.cache_level		= 2,
147 		.cache = {
148 			.min_cbm_bits	= 1,
149 			.cbm_idx_mult	= 2,
150 			.cbm_idx_offset	= 0,
151 		},
152 		.parse_ctrlval		= parse_cbm,
153 		.format_str		= "%d=%0*x",
154 		.fflags			= RFTYPE_RES_CACHE,
155 	},
156 	[RDT_RESOURCE_L2CODE] =
157 	{
158 		.rid			= RDT_RESOURCE_L2CODE,
159 		.name			= "L2CODE",
160 		.domains		= domain_init(RDT_RESOURCE_L2CODE),
161 		.msr_base		= MSR_IA32_L2_CBM_BASE,
162 		.msr_update		= cat_wrmsr,
163 		.cache_level		= 2,
164 		.cache = {
165 			.min_cbm_bits	= 1,
166 			.cbm_idx_mult	= 2,
167 			.cbm_idx_offset	= 1,
168 		},
169 		.parse_ctrlval		= parse_cbm,
170 		.format_str		= "%d=%0*x",
171 		.fflags			= RFTYPE_RES_CACHE,
172 	},
173 	[RDT_RESOURCE_MBA] =
174 	{
175 		.rid			= RDT_RESOURCE_MBA,
176 		.name			= "MB",
177 		.domains		= domain_init(RDT_RESOURCE_MBA),
178 		.cache_level		= 3,
179 		.format_str		= "%d=%*u",
180 		.fflags			= RFTYPE_RES_MB,
181 	},
182 };
183 
184 static unsigned int cbm_idx(struct rdt_resource *r, unsigned int closid)
185 {
186 	return closid * r->cache.cbm_idx_mult + r->cache.cbm_idx_offset;
187 }
188 
189 /*
190  * cache_alloc_hsw_probe() - Have to probe for Intel haswell server CPUs
191  * as they do not have CPUID enumeration support for Cache allocation.
192  * The check for Vendor/Family/Model is not enough to guarantee that
193  * the MSRs won't #GP fault because only the following SKUs support
194  * CAT:
195  *	Intel(R) Xeon(R)  CPU E5-2658  v3  @  2.20GHz
196  *	Intel(R) Xeon(R)  CPU E5-2648L v3  @  1.80GHz
197  *	Intel(R) Xeon(R)  CPU E5-2628L v3  @  2.00GHz
198  *	Intel(R) Xeon(R)  CPU E5-2618L v3  @  2.30GHz
199  *	Intel(R) Xeon(R)  CPU E5-2608L v3  @  2.00GHz
200  *	Intel(R) Xeon(R)  CPU E5-2658A v3  @  2.20GHz
201  *
202  * Probe by trying to write the first of the L3 cach mask registers
203  * and checking that the bits stick. Max CLOSids is always 4 and max cbm length
204  * is always 20 on hsw server parts. The minimum cache bitmask length
205  * allowed for HSW server is always 2 bits. Hardcode all of them.
206  */
207 static inline void cache_alloc_hsw_probe(void)
208 {
209 	struct rdt_resource *r  = &rdt_resources_all[RDT_RESOURCE_L3];
210 	u32 l, h, max_cbm = BIT_MASK(20) - 1;
211 
212 	if (wrmsr_safe(MSR_IA32_L3_CBM_BASE, max_cbm, 0))
213 		return;
214 
215 	rdmsr(MSR_IA32_L3_CBM_BASE, l, h);
216 
217 	/* If all the bits were set in MSR, return success */
218 	if (l != max_cbm)
219 		return;
220 
221 	r->num_closid = 4;
222 	r->default_ctrl = max_cbm;
223 	r->cache.cbm_len = 20;
224 	r->cache.shareable_bits = 0xc0000;
225 	r->cache.min_cbm_bits = 2;
226 	r->alloc_capable = true;
227 	r->alloc_enabled = true;
228 
229 	rdt_alloc_capable = true;
230 }
231 
232 bool is_mba_sc(struct rdt_resource *r)
233 {
234 	if (!r)
235 		return rdt_resources_all[RDT_RESOURCE_MBA].membw.mba_sc;
236 
237 	return r->membw.mba_sc;
238 }
239 
240 /*
241  * rdt_get_mb_table() - get a mapping of bandwidth(b/w) percentage values
242  * exposed to user interface and the h/w understandable delay values.
243  *
244  * The non-linear delay values have the granularity of power of two
245  * and also the h/w does not guarantee a curve for configured delay
246  * values vs. actual b/w enforced.
247  * Hence we need a mapping that is pre calibrated so the user can
248  * express the memory b/w as a percentage value.
249  */
250 static inline bool rdt_get_mb_table(struct rdt_resource *r)
251 {
252 	/*
253 	 * There are no Intel SKUs as of now to support non-linear delay.
254 	 */
255 	pr_info("MBA b/w map not implemented for cpu:%d, model:%d",
256 		boot_cpu_data.x86, boot_cpu_data.x86_model);
257 
258 	return false;
259 }
260 
261 static bool __get_mem_config_intel(struct rdt_resource *r)
262 {
263 	union cpuid_0x10_3_eax eax;
264 	union cpuid_0x10_x_edx edx;
265 	u32 ebx, ecx;
266 
267 	cpuid_count(0x00000010, 3, &eax.full, &ebx, &ecx, &edx.full);
268 	r->num_closid = edx.split.cos_max + 1;
269 	r->membw.max_delay = eax.split.max_delay + 1;
270 	r->default_ctrl = MAX_MBA_BW;
271 	if (ecx & MBA_IS_LINEAR) {
272 		r->membw.delay_linear = true;
273 		r->membw.min_bw = MAX_MBA_BW - r->membw.max_delay;
274 		r->membw.bw_gran = MAX_MBA_BW - r->membw.max_delay;
275 	} else {
276 		if (!rdt_get_mb_table(r))
277 			return false;
278 	}
279 	r->data_width = 3;
280 
281 	r->alloc_capable = true;
282 	r->alloc_enabled = true;
283 
284 	return true;
285 }
286 
287 static bool __rdt_get_mem_config_amd(struct rdt_resource *r)
288 {
289 	union cpuid_0x10_3_eax eax;
290 	union cpuid_0x10_x_edx edx;
291 	u32 ebx, ecx;
292 
293 	cpuid_count(0x80000020, 1, &eax.full, &ebx, &ecx, &edx.full);
294 	r->num_closid = edx.split.cos_max + 1;
295 	r->default_ctrl = MAX_MBA_BW_AMD;
296 
297 	/* AMD does not use delay */
298 	r->membw.delay_linear = false;
299 
300 	r->membw.min_bw = 0;
301 	r->membw.bw_gran = 1;
302 	/* Max value is 2048, Data width should be 4 in decimal */
303 	r->data_width = 4;
304 
305 	r->alloc_capable = true;
306 	r->alloc_enabled = true;
307 
308 	return true;
309 }
310 
311 static void rdt_get_cache_alloc_cfg(int idx, struct rdt_resource *r)
312 {
313 	union cpuid_0x10_1_eax eax;
314 	union cpuid_0x10_x_edx edx;
315 	u32 ebx, ecx;
316 
317 	cpuid_count(0x00000010, idx, &eax.full, &ebx, &ecx, &edx.full);
318 	r->num_closid = edx.split.cos_max + 1;
319 	r->cache.cbm_len = eax.split.cbm_len + 1;
320 	r->default_ctrl = BIT_MASK(eax.split.cbm_len + 1) - 1;
321 	r->cache.shareable_bits = ebx & r->default_ctrl;
322 	r->data_width = (r->cache.cbm_len + 3) / 4;
323 	r->alloc_capable = true;
324 	r->alloc_enabled = true;
325 }
326 
327 static void rdt_get_cdp_config(int level, int type)
328 {
329 	struct rdt_resource *r_l = &rdt_resources_all[level];
330 	struct rdt_resource *r = &rdt_resources_all[type];
331 
332 	r->num_closid = r_l->num_closid / 2;
333 	r->cache.cbm_len = r_l->cache.cbm_len;
334 	r->default_ctrl = r_l->default_ctrl;
335 	r->cache.shareable_bits = r_l->cache.shareable_bits;
336 	r->data_width = (r->cache.cbm_len + 3) / 4;
337 	r->alloc_capable = true;
338 	/*
339 	 * By default, CDP is disabled. CDP can be enabled by mount parameter
340 	 * "cdp" during resctrl file system mount time.
341 	 */
342 	r->alloc_enabled = false;
343 }
344 
345 static void rdt_get_cdp_l3_config(void)
346 {
347 	rdt_get_cdp_config(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA);
348 	rdt_get_cdp_config(RDT_RESOURCE_L3, RDT_RESOURCE_L3CODE);
349 }
350 
351 static void rdt_get_cdp_l2_config(void)
352 {
353 	rdt_get_cdp_config(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA);
354 	rdt_get_cdp_config(RDT_RESOURCE_L2, RDT_RESOURCE_L2CODE);
355 }
356 
357 static int get_cache_id(int cpu, int level)
358 {
359 	struct cpu_cacheinfo *ci = get_cpu_cacheinfo(cpu);
360 	int i;
361 
362 	for (i = 0; i < ci->num_leaves; i++) {
363 		if (ci->info_list[i].level == level)
364 			return ci->info_list[i].id;
365 	}
366 
367 	return -1;
368 }
369 
370 static void
371 mba_wrmsr_amd(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r)
372 {
373 	unsigned int i;
374 
375 	for (i = m->low; i < m->high; i++)
376 		wrmsrl(r->msr_base + i, d->ctrl_val[i]);
377 }
378 
379 /*
380  * Map the memory b/w percentage value to delay values
381  * that can be written to QOS_MSRs.
382  * There are currently no SKUs which support non linear delay values.
383  */
384 u32 delay_bw_map(unsigned long bw, struct rdt_resource *r)
385 {
386 	if (r->membw.delay_linear)
387 		return MAX_MBA_BW - bw;
388 
389 	pr_warn_once("Non Linear delay-bw map not supported but queried\n");
390 	return r->default_ctrl;
391 }
392 
393 static void
394 mba_wrmsr_intel(struct rdt_domain *d, struct msr_param *m,
395 		struct rdt_resource *r)
396 {
397 	unsigned int i;
398 
399 	/*  Write the delay values for mba. */
400 	for (i = m->low; i < m->high; i++)
401 		wrmsrl(r->msr_base + i, delay_bw_map(d->ctrl_val[i], r));
402 }
403 
404 static void
405 cat_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r)
406 {
407 	unsigned int i;
408 
409 	for (i = m->low; i < m->high; i++)
410 		wrmsrl(r->msr_base + cbm_idx(r, i), d->ctrl_val[i]);
411 }
412 
413 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r)
414 {
415 	struct rdt_domain *d;
416 
417 	list_for_each_entry(d, &r->domains, list) {
418 		/* Find the domain that contains this CPU */
419 		if (cpumask_test_cpu(cpu, &d->cpu_mask))
420 			return d;
421 	}
422 
423 	return NULL;
424 }
425 
426 void rdt_ctrl_update(void *arg)
427 {
428 	struct msr_param *m = arg;
429 	struct rdt_resource *r = m->res;
430 	int cpu = smp_processor_id();
431 	struct rdt_domain *d;
432 
433 	d = get_domain_from_cpu(cpu, r);
434 	if (d) {
435 		r->msr_update(d, m, r);
436 		return;
437 	}
438 	pr_warn_once("cpu %d not found in any domain for resource %s\n",
439 		     cpu, r->name);
440 }
441 
442 /*
443  * rdt_find_domain - Find a domain in a resource that matches input resource id
444  *
445  * Search resource r's domain list to find the resource id. If the resource
446  * id is found in a domain, return the domain. Otherwise, if requested by
447  * caller, return the first domain whose id is bigger than the input id.
448  * The domain list is sorted by id in ascending order.
449  */
450 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
451 				   struct list_head **pos)
452 {
453 	struct rdt_domain *d;
454 	struct list_head *l;
455 
456 	if (id < 0)
457 		return ERR_PTR(-ENODEV);
458 
459 	list_for_each(l, &r->domains) {
460 		d = list_entry(l, struct rdt_domain, list);
461 		/* When id is found, return its domain. */
462 		if (id == d->id)
463 			return d;
464 		/* Stop searching when finding id's position in sorted list. */
465 		if (id < d->id)
466 			break;
467 	}
468 
469 	if (pos)
470 		*pos = l;
471 
472 	return NULL;
473 }
474 
475 void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm)
476 {
477 	int i;
478 
479 	/*
480 	 * Initialize the Control MSRs to having no control.
481 	 * For Cache Allocation: Set all bits in cbm
482 	 * For Memory Allocation: Set b/w requested to 100%
483 	 * and the bandwidth in MBps to U32_MAX
484 	 */
485 	for (i = 0; i < r->num_closid; i++, dc++, dm++) {
486 		*dc = r->default_ctrl;
487 		*dm = MBA_MAX_MBPS;
488 	}
489 }
490 
491 static int domain_setup_ctrlval(struct rdt_resource *r, struct rdt_domain *d)
492 {
493 	struct msr_param m;
494 	u32 *dc, *dm;
495 
496 	dc = kmalloc_array(r->num_closid, sizeof(*d->ctrl_val), GFP_KERNEL);
497 	if (!dc)
498 		return -ENOMEM;
499 
500 	dm = kmalloc_array(r->num_closid, sizeof(*d->mbps_val), GFP_KERNEL);
501 	if (!dm) {
502 		kfree(dc);
503 		return -ENOMEM;
504 	}
505 
506 	d->ctrl_val = dc;
507 	d->mbps_val = dm;
508 	setup_default_ctrlval(r, dc, dm);
509 
510 	m.low = 0;
511 	m.high = r->num_closid;
512 	r->msr_update(d, &m, r);
513 	return 0;
514 }
515 
516 static int domain_setup_mon_state(struct rdt_resource *r, struct rdt_domain *d)
517 {
518 	size_t tsize;
519 
520 	if (is_llc_occupancy_enabled()) {
521 		d->rmid_busy_llc = bitmap_zalloc(r->num_rmid, GFP_KERNEL);
522 		if (!d->rmid_busy_llc)
523 			return -ENOMEM;
524 		INIT_DELAYED_WORK(&d->cqm_limbo, cqm_handle_limbo);
525 	}
526 	if (is_mbm_total_enabled()) {
527 		tsize = sizeof(*d->mbm_total);
528 		d->mbm_total = kcalloc(r->num_rmid, tsize, GFP_KERNEL);
529 		if (!d->mbm_total) {
530 			bitmap_free(d->rmid_busy_llc);
531 			return -ENOMEM;
532 		}
533 	}
534 	if (is_mbm_local_enabled()) {
535 		tsize = sizeof(*d->mbm_local);
536 		d->mbm_local = kcalloc(r->num_rmid, tsize, GFP_KERNEL);
537 		if (!d->mbm_local) {
538 			bitmap_free(d->rmid_busy_llc);
539 			kfree(d->mbm_total);
540 			return -ENOMEM;
541 		}
542 	}
543 
544 	if (is_mbm_enabled()) {
545 		INIT_DELAYED_WORK(&d->mbm_over, mbm_handle_overflow);
546 		mbm_setup_overflow_handler(d, MBM_OVERFLOW_INTERVAL);
547 	}
548 
549 	return 0;
550 }
551 
552 /*
553  * domain_add_cpu - Add a cpu to a resource's domain list.
554  *
555  * If an existing domain in the resource r's domain list matches the cpu's
556  * resource id, add the cpu in the domain.
557  *
558  * Otherwise, a new domain is allocated and inserted into the right position
559  * in the domain list sorted by id in ascending order.
560  *
561  * The order in the domain list is visible to users when we print entries
562  * in the schemata file and schemata input is validated to have the same order
563  * as this list.
564  */
565 static void domain_add_cpu(int cpu, struct rdt_resource *r)
566 {
567 	int id = get_cache_id(cpu, r->cache_level);
568 	struct list_head *add_pos = NULL;
569 	struct rdt_domain *d;
570 
571 	d = rdt_find_domain(r, id, &add_pos);
572 	if (IS_ERR(d)) {
573 		pr_warn("Could't find cache id for cpu %d\n", cpu);
574 		return;
575 	}
576 
577 	if (d) {
578 		cpumask_set_cpu(cpu, &d->cpu_mask);
579 		return;
580 	}
581 
582 	d = kzalloc_node(sizeof(*d), GFP_KERNEL, cpu_to_node(cpu));
583 	if (!d)
584 		return;
585 
586 	d->id = id;
587 	cpumask_set_cpu(cpu, &d->cpu_mask);
588 
589 	if (r->alloc_capable && domain_setup_ctrlval(r, d)) {
590 		kfree(d);
591 		return;
592 	}
593 
594 	if (r->mon_capable && domain_setup_mon_state(r, d)) {
595 		kfree(d);
596 		return;
597 	}
598 
599 	list_add_tail(&d->list, add_pos);
600 
601 	/*
602 	 * If resctrl is mounted, add
603 	 * per domain monitor data directories.
604 	 */
605 	if (static_branch_unlikely(&rdt_mon_enable_key))
606 		mkdir_mondata_subdir_allrdtgrp(r, d);
607 }
608 
609 static void domain_remove_cpu(int cpu, struct rdt_resource *r)
610 {
611 	int id = get_cache_id(cpu, r->cache_level);
612 	struct rdt_domain *d;
613 
614 	d = rdt_find_domain(r, id, NULL);
615 	if (IS_ERR_OR_NULL(d)) {
616 		pr_warn("Could't find cache id for cpu %d\n", cpu);
617 		return;
618 	}
619 
620 	cpumask_clear_cpu(cpu, &d->cpu_mask);
621 	if (cpumask_empty(&d->cpu_mask)) {
622 		/*
623 		 * If resctrl is mounted, remove all the
624 		 * per domain monitor data directories.
625 		 */
626 		if (static_branch_unlikely(&rdt_mon_enable_key))
627 			rmdir_mondata_subdir_allrdtgrp(r, d->id);
628 		list_del(&d->list);
629 		if (is_mbm_enabled())
630 			cancel_delayed_work(&d->mbm_over);
631 		if (is_llc_occupancy_enabled() &&  has_busy_rmid(r, d)) {
632 			/*
633 			 * When a package is going down, forcefully
634 			 * decrement rmid->ebusy. There is no way to know
635 			 * that the L3 was flushed and hence may lead to
636 			 * incorrect counts in rare scenarios, but leaving
637 			 * the RMID as busy creates RMID leaks if the
638 			 * package never comes back.
639 			 */
640 			__check_limbo(d, true);
641 			cancel_delayed_work(&d->cqm_limbo);
642 		}
643 
644 		/*
645 		 * rdt_domain "d" is going to be freed below, so clear
646 		 * its pointer from pseudo_lock_region struct.
647 		 */
648 		if (d->plr)
649 			d->plr->d = NULL;
650 
651 		kfree(d->ctrl_val);
652 		kfree(d->mbps_val);
653 		bitmap_free(d->rmid_busy_llc);
654 		kfree(d->mbm_total);
655 		kfree(d->mbm_local);
656 		kfree(d);
657 		return;
658 	}
659 
660 	if (r == &rdt_resources_all[RDT_RESOURCE_L3]) {
661 		if (is_mbm_enabled() && cpu == d->mbm_work_cpu) {
662 			cancel_delayed_work(&d->mbm_over);
663 			mbm_setup_overflow_handler(d, 0);
664 		}
665 		if (is_llc_occupancy_enabled() && cpu == d->cqm_work_cpu &&
666 		    has_busy_rmid(r, d)) {
667 			cancel_delayed_work(&d->cqm_limbo);
668 			cqm_setup_limbo_handler(d, 0);
669 		}
670 	}
671 }
672 
673 static void clear_closid_rmid(int cpu)
674 {
675 	struct resctrl_pqr_state *state = this_cpu_ptr(&pqr_state);
676 
677 	state->default_closid = 0;
678 	state->default_rmid = 0;
679 	state->cur_closid = 0;
680 	state->cur_rmid = 0;
681 	wrmsr(IA32_PQR_ASSOC, 0, 0);
682 }
683 
684 static int resctrl_online_cpu(unsigned int cpu)
685 {
686 	struct rdt_resource *r;
687 
688 	mutex_lock(&rdtgroup_mutex);
689 	for_each_capable_rdt_resource(r)
690 		domain_add_cpu(cpu, r);
691 	/* The cpu is set in default rdtgroup after online. */
692 	cpumask_set_cpu(cpu, &rdtgroup_default.cpu_mask);
693 	clear_closid_rmid(cpu);
694 	mutex_unlock(&rdtgroup_mutex);
695 
696 	return 0;
697 }
698 
699 static void clear_childcpus(struct rdtgroup *r, unsigned int cpu)
700 {
701 	struct rdtgroup *cr;
702 
703 	list_for_each_entry(cr, &r->mon.crdtgrp_list, mon.crdtgrp_list) {
704 		if (cpumask_test_and_clear_cpu(cpu, &cr->cpu_mask)) {
705 			break;
706 		}
707 	}
708 }
709 
710 static int resctrl_offline_cpu(unsigned int cpu)
711 {
712 	struct rdtgroup *rdtgrp;
713 	struct rdt_resource *r;
714 
715 	mutex_lock(&rdtgroup_mutex);
716 	for_each_capable_rdt_resource(r)
717 		domain_remove_cpu(cpu, r);
718 	list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
719 		if (cpumask_test_and_clear_cpu(cpu, &rdtgrp->cpu_mask)) {
720 			clear_childcpus(rdtgrp, cpu);
721 			break;
722 		}
723 	}
724 	clear_closid_rmid(cpu);
725 	mutex_unlock(&rdtgroup_mutex);
726 
727 	return 0;
728 }
729 
730 /*
731  * Choose a width for the resource name and resource data based on the
732  * resource that has widest name and cbm.
733  */
734 static __init void rdt_init_padding(void)
735 {
736 	struct rdt_resource *r;
737 	int cl;
738 
739 	for_each_alloc_capable_rdt_resource(r) {
740 		cl = strlen(r->name);
741 		if (cl > max_name_width)
742 			max_name_width = cl;
743 
744 		if (r->data_width > max_data_width)
745 			max_data_width = r->data_width;
746 	}
747 }
748 
749 enum {
750 	RDT_FLAG_CMT,
751 	RDT_FLAG_MBM_TOTAL,
752 	RDT_FLAG_MBM_LOCAL,
753 	RDT_FLAG_L3_CAT,
754 	RDT_FLAG_L3_CDP,
755 	RDT_FLAG_L2_CAT,
756 	RDT_FLAG_L2_CDP,
757 	RDT_FLAG_MBA,
758 };
759 
760 #define RDT_OPT(idx, n, f)	\
761 [idx] = {			\
762 	.name = n,		\
763 	.flag = f		\
764 }
765 
766 struct rdt_options {
767 	char	*name;
768 	int	flag;
769 	bool	force_off, force_on;
770 };
771 
772 static struct rdt_options rdt_options[]  __initdata = {
773 	RDT_OPT(RDT_FLAG_CMT,	    "cmt",	X86_FEATURE_CQM_OCCUP_LLC),
774 	RDT_OPT(RDT_FLAG_MBM_TOTAL, "mbmtotal", X86_FEATURE_CQM_MBM_TOTAL),
775 	RDT_OPT(RDT_FLAG_MBM_LOCAL, "mbmlocal", X86_FEATURE_CQM_MBM_LOCAL),
776 	RDT_OPT(RDT_FLAG_L3_CAT,    "l3cat",	X86_FEATURE_CAT_L3),
777 	RDT_OPT(RDT_FLAG_L3_CDP,    "l3cdp",	X86_FEATURE_CDP_L3),
778 	RDT_OPT(RDT_FLAG_L2_CAT,    "l2cat",	X86_FEATURE_CAT_L2),
779 	RDT_OPT(RDT_FLAG_L2_CDP,    "l2cdp",	X86_FEATURE_CDP_L2),
780 	RDT_OPT(RDT_FLAG_MBA,	    "mba",	X86_FEATURE_MBA),
781 };
782 #define NUM_RDT_OPTIONS ARRAY_SIZE(rdt_options)
783 
784 static int __init set_rdt_options(char *str)
785 {
786 	struct rdt_options *o;
787 	bool force_off;
788 	char *tok;
789 
790 	if (*str == '=')
791 		str++;
792 	while ((tok = strsep(&str, ",")) != NULL) {
793 		force_off = *tok == '!';
794 		if (force_off)
795 			tok++;
796 		for (o = rdt_options; o < &rdt_options[NUM_RDT_OPTIONS]; o++) {
797 			if (strcmp(tok, o->name) == 0) {
798 				if (force_off)
799 					o->force_off = true;
800 				else
801 					o->force_on = true;
802 				break;
803 			}
804 		}
805 	}
806 	return 1;
807 }
808 __setup("rdt", set_rdt_options);
809 
810 static bool __init rdt_cpu_has(int flag)
811 {
812 	bool ret = boot_cpu_has(flag);
813 	struct rdt_options *o;
814 
815 	if (!ret)
816 		return ret;
817 
818 	for (o = rdt_options; o < &rdt_options[NUM_RDT_OPTIONS]; o++) {
819 		if (flag == o->flag) {
820 			if (o->force_off)
821 				ret = false;
822 			if (o->force_on)
823 				ret = true;
824 			break;
825 		}
826 	}
827 	return ret;
828 }
829 
830 static __init bool get_mem_config(void)
831 {
832 	if (!rdt_cpu_has(X86_FEATURE_MBA))
833 		return false;
834 
835 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
836 		return __get_mem_config_intel(&rdt_resources_all[RDT_RESOURCE_MBA]);
837 	else if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
838 		return __rdt_get_mem_config_amd(&rdt_resources_all[RDT_RESOURCE_MBA]);
839 
840 	return false;
841 }
842 
843 static __init bool get_rdt_alloc_resources(void)
844 {
845 	bool ret = false;
846 
847 	if (rdt_alloc_capable)
848 		return true;
849 
850 	if (!boot_cpu_has(X86_FEATURE_RDT_A))
851 		return false;
852 
853 	if (rdt_cpu_has(X86_FEATURE_CAT_L3)) {
854 		rdt_get_cache_alloc_cfg(1, &rdt_resources_all[RDT_RESOURCE_L3]);
855 		if (rdt_cpu_has(X86_FEATURE_CDP_L3))
856 			rdt_get_cdp_l3_config();
857 		ret = true;
858 	}
859 	if (rdt_cpu_has(X86_FEATURE_CAT_L2)) {
860 		/* CPUID 0x10.2 fields are same format at 0x10.1 */
861 		rdt_get_cache_alloc_cfg(2, &rdt_resources_all[RDT_RESOURCE_L2]);
862 		if (rdt_cpu_has(X86_FEATURE_CDP_L2))
863 			rdt_get_cdp_l2_config();
864 		ret = true;
865 	}
866 
867 	if (get_mem_config())
868 		ret = true;
869 
870 	return ret;
871 }
872 
873 static __init bool get_rdt_mon_resources(void)
874 {
875 	if (rdt_cpu_has(X86_FEATURE_CQM_OCCUP_LLC))
876 		rdt_mon_features |= (1 << QOS_L3_OCCUP_EVENT_ID);
877 	if (rdt_cpu_has(X86_FEATURE_CQM_MBM_TOTAL))
878 		rdt_mon_features |= (1 << QOS_L3_MBM_TOTAL_EVENT_ID);
879 	if (rdt_cpu_has(X86_FEATURE_CQM_MBM_LOCAL))
880 		rdt_mon_features |= (1 << QOS_L3_MBM_LOCAL_EVENT_ID);
881 
882 	if (!rdt_mon_features)
883 		return false;
884 
885 	return !rdt_get_mon_l3_config(&rdt_resources_all[RDT_RESOURCE_L3]);
886 }
887 
888 static __init void __check_quirks_intel(void)
889 {
890 	switch (boot_cpu_data.x86_model) {
891 	case INTEL_FAM6_HASWELL_X:
892 		if (!rdt_options[RDT_FLAG_L3_CAT].force_off)
893 			cache_alloc_hsw_probe();
894 		break;
895 	case INTEL_FAM6_SKYLAKE_X:
896 		if (boot_cpu_data.x86_stepping <= 4)
897 			set_rdt_options("!cmt,!mbmtotal,!mbmlocal,!l3cat");
898 		else
899 			set_rdt_options("!l3cat");
900 	}
901 }
902 
903 static __init void check_quirks(void)
904 {
905 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
906 		__check_quirks_intel();
907 }
908 
909 static __init bool get_rdt_resources(void)
910 {
911 	rdt_alloc_capable = get_rdt_alloc_resources();
912 	rdt_mon_capable = get_rdt_mon_resources();
913 
914 	return (rdt_mon_capable || rdt_alloc_capable);
915 }
916 
917 static __init void rdt_init_res_defs_intel(void)
918 {
919 	struct rdt_resource *r;
920 
921 	for_each_rdt_resource(r) {
922 		if (r->rid == RDT_RESOURCE_L3 ||
923 		    r->rid == RDT_RESOURCE_L3DATA ||
924 		    r->rid == RDT_RESOURCE_L3CODE ||
925 		    r->rid == RDT_RESOURCE_L2 ||
926 		    r->rid == RDT_RESOURCE_L2DATA ||
927 		    r->rid == RDT_RESOURCE_L2CODE)
928 			r->cbm_validate = cbm_validate_intel;
929 		else if (r->rid == RDT_RESOURCE_MBA) {
930 			r->msr_base = MSR_IA32_MBA_THRTL_BASE;
931 			r->msr_update = mba_wrmsr_intel;
932 			r->parse_ctrlval = parse_bw_intel;
933 		}
934 	}
935 }
936 
937 static __init void rdt_init_res_defs_amd(void)
938 {
939 	struct rdt_resource *r;
940 
941 	for_each_rdt_resource(r) {
942 		if (r->rid == RDT_RESOURCE_L3 ||
943 		    r->rid == RDT_RESOURCE_L3DATA ||
944 		    r->rid == RDT_RESOURCE_L3CODE ||
945 		    r->rid == RDT_RESOURCE_L2 ||
946 		    r->rid == RDT_RESOURCE_L2DATA ||
947 		    r->rid == RDT_RESOURCE_L2CODE)
948 			r->cbm_validate = cbm_validate_amd;
949 		else if (r->rid == RDT_RESOURCE_MBA) {
950 			r->msr_base = MSR_IA32_MBA_BW_BASE;
951 			r->msr_update = mba_wrmsr_amd;
952 			r->parse_ctrlval = parse_bw_amd;
953 		}
954 	}
955 }
956 
957 static __init void rdt_init_res_defs(void)
958 {
959 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
960 		rdt_init_res_defs_intel();
961 	else if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
962 		rdt_init_res_defs_amd();
963 }
964 
965 static enum cpuhp_state rdt_online;
966 
967 static int __init resctrl_late_init(void)
968 {
969 	struct rdt_resource *r;
970 	int state, ret;
971 
972 	/*
973 	 * Initialize functions(or definitions) that are different
974 	 * between vendors here.
975 	 */
976 	rdt_init_res_defs();
977 
978 	check_quirks();
979 
980 	if (!get_rdt_resources())
981 		return -ENODEV;
982 
983 	rdt_init_padding();
984 
985 	state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN,
986 				  "x86/resctrl/cat:online:",
987 				  resctrl_online_cpu, resctrl_offline_cpu);
988 	if (state < 0)
989 		return state;
990 
991 	ret = rdtgroup_init();
992 	if (ret) {
993 		cpuhp_remove_state(state);
994 		return ret;
995 	}
996 	rdt_online = state;
997 
998 	for_each_alloc_capable_rdt_resource(r)
999 		pr_info("%s allocation detected\n", r->name);
1000 
1001 	for_each_mon_capable_rdt_resource(r)
1002 		pr_info("%s monitoring detected\n", r->name);
1003 
1004 	return 0;
1005 }
1006 
1007 late_initcall(resctrl_late_init);
1008 
1009 static void __exit resctrl_exit(void)
1010 {
1011 	cpuhp_remove_state(rdt_online);
1012 	rdtgroup_exit();
1013 }
1014 
1015 __exitcall(resctrl_exit);
1016