]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/kernel/cpu/intel_rdt.c
Merge branch 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kernel / cpu / intel_rdt.c
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) KBUILD_MODNAME ": " 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/intel_rdt_sched.h>
34 #include "intel_rdt.h"
35
36 #define MAX_MBA_BW 100u
37 #define MBA_IS_LINEAR 0x4
38
39 /* Mutex to protect rdtgroup access. */
40 DEFINE_MUTEX(rdtgroup_mutex);
41
42 /*
43 * The cached intel_pqr_state is strictly per CPU and can never be
44 * updated from a remote CPU. Functions which modify the state
45 * are called with interrupts disabled and no preemption, which
46 * is sufficient for the protection.
47 */
48 DEFINE_PER_CPU(struct intel_pqr_state, pqr_state);
49
50 /*
51 * Used to store the max resource name width and max resource data width
52 * to display the schemata in a tabular format
53 */
54 int max_name_width, max_data_width;
55
56 /*
57 * Global boolean for rdt_alloc which is true if any
58 * resource allocation is enabled.
59 */
60 bool rdt_alloc_capable;
61
62 static void
63 mba_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r);
64 static void
65 cat_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r);
66
67 #define domain_init(id) LIST_HEAD_INIT(rdt_resources_all[id].domains)
68
69 struct rdt_resource rdt_resources_all[] = {
70 [RDT_RESOURCE_L3] =
71 {
72 .rid = RDT_RESOURCE_L3,
73 .name = "L3",
74 .domains = domain_init(RDT_RESOURCE_L3),
75 .msr_base = IA32_L3_CBM_BASE,
76 .msr_update = cat_wrmsr,
77 .cache_level = 3,
78 .cache = {
79 .min_cbm_bits = 1,
80 .cbm_idx_mult = 1,
81 .cbm_idx_offset = 0,
82 },
83 .parse_ctrlval = parse_cbm,
84 .format_str = "%d=%0*x",
85 .fflags = RFTYPE_RES_CACHE,
86 },
87 [RDT_RESOURCE_L3DATA] =
88 {
89 .rid = RDT_RESOURCE_L3DATA,
90 .name = "L3DATA",
91 .domains = domain_init(RDT_RESOURCE_L3DATA),
92 .msr_base = IA32_L3_CBM_BASE,
93 .msr_update = cat_wrmsr,
94 .cache_level = 3,
95 .cache = {
96 .min_cbm_bits = 1,
97 .cbm_idx_mult = 2,
98 .cbm_idx_offset = 0,
99 },
100 .parse_ctrlval = parse_cbm,
101 .format_str = "%d=%0*x",
102 .fflags = RFTYPE_RES_CACHE,
103 },
104 [RDT_RESOURCE_L3CODE] =
105 {
106 .rid = RDT_RESOURCE_L3CODE,
107 .name = "L3CODE",
108 .domains = domain_init(RDT_RESOURCE_L3CODE),
109 .msr_base = IA32_L3_CBM_BASE,
110 .msr_update = cat_wrmsr,
111 .cache_level = 3,
112 .cache = {
113 .min_cbm_bits = 1,
114 .cbm_idx_mult = 2,
115 .cbm_idx_offset = 1,
116 },
117 .parse_ctrlval = parse_cbm,
118 .format_str = "%d=%0*x",
119 .fflags = RFTYPE_RES_CACHE,
120 },
121 [RDT_RESOURCE_L2] =
122 {
123 .rid = RDT_RESOURCE_L2,
124 .name = "L2",
125 .domains = domain_init(RDT_RESOURCE_L2),
126 .msr_base = IA32_L2_CBM_BASE,
127 .msr_update = cat_wrmsr,
128 .cache_level = 2,
129 .cache = {
130 .min_cbm_bits = 1,
131 .cbm_idx_mult = 1,
132 .cbm_idx_offset = 0,
133 },
134 .parse_ctrlval = parse_cbm,
135 .format_str = "%d=%0*x",
136 .fflags = RFTYPE_RES_CACHE,
137 },
138 [RDT_RESOURCE_MBA] =
139 {
140 .rid = RDT_RESOURCE_MBA,
141 .name = "MB",
142 .domains = domain_init(RDT_RESOURCE_MBA),
143 .msr_base = IA32_MBA_THRTL_BASE,
144 .msr_update = mba_wrmsr,
145 .cache_level = 3,
146 .parse_ctrlval = parse_bw,
147 .format_str = "%d=%*d",
148 .fflags = RFTYPE_RES_MB,
149 },
150 };
151
152 static unsigned int cbm_idx(struct rdt_resource *r, unsigned int closid)
153 {
154 return closid * r->cache.cbm_idx_mult + r->cache.cbm_idx_offset;
155 }
156
157 /*
158 * cache_alloc_hsw_probe() - Have to probe for Intel haswell server CPUs
159 * as they do not have CPUID enumeration support for Cache allocation.
160 * The check for Vendor/Family/Model is not enough to guarantee that
161 * the MSRs won't #GP fault because only the following SKUs support
162 * CAT:
163 * Intel(R) Xeon(R) CPU E5-2658 v3 @ 2.20GHz
164 * Intel(R) Xeon(R) CPU E5-2648L v3 @ 1.80GHz
165 * Intel(R) Xeon(R) CPU E5-2628L v3 @ 2.00GHz
166 * Intel(R) Xeon(R) CPU E5-2618L v3 @ 2.30GHz
167 * Intel(R) Xeon(R) CPU E5-2608L v3 @ 2.00GHz
168 * Intel(R) Xeon(R) CPU E5-2658A v3 @ 2.20GHz
169 *
170 * Probe by trying to write the first of the L3 cach mask registers
171 * and checking that the bits stick. Max CLOSids is always 4 and max cbm length
172 * is always 20 on hsw server parts. The minimum cache bitmask length
173 * allowed for HSW server is always 2 bits. Hardcode all of them.
174 */
175 static inline void cache_alloc_hsw_probe(void)
176 {
177 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_L3];
178 u32 l, h, max_cbm = BIT_MASK(20) - 1;
179
180 if (wrmsr_safe(IA32_L3_CBM_BASE, max_cbm, 0))
181 return;
182 rdmsr(IA32_L3_CBM_BASE, l, h);
183
184 /* If all the bits were set in MSR, return success */
185 if (l != max_cbm)
186 return;
187
188 r->num_closid = 4;
189 r->default_ctrl = max_cbm;
190 r->cache.cbm_len = 20;
191 r->cache.shareable_bits = 0xc0000;
192 r->cache.min_cbm_bits = 2;
193 r->alloc_capable = true;
194 r->alloc_enabled = true;
195
196 rdt_alloc_capable = true;
197 }
198
199 /*
200 * rdt_get_mb_table() - get a mapping of bandwidth(b/w) percentage values
201 * exposed to user interface and the h/w understandable delay values.
202 *
203 * The non-linear delay values have the granularity of power of two
204 * and also the h/w does not guarantee a curve for configured delay
205 * values vs. actual b/w enforced.
206 * Hence we need a mapping that is pre calibrated so the user can
207 * express the memory b/w as a percentage value.
208 */
209 static inline bool rdt_get_mb_table(struct rdt_resource *r)
210 {
211 /*
212 * There are no Intel SKUs as of now to support non-linear delay.
213 */
214 pr_info("MBA b/w map not implemented for cpu:%d, model:%d",
215 boot_cpu_data.x86, boot_cpu_data.x86_model);
216
217 return false;
218 }
219
220 static bool rdt_get_mem_config(struct rdt_resource *r)
221 {
222 union cpuid_0x10_3_eax eax;
223 union cpuid_0x10_x_edx edx;
224 u32 ebx, ecx;
225
226 cpuid_count(0x00000010, 3, &eax.full, &ebx, &ecx, &edx.full);
227 r->num_closid = edx.split.cos_max + 1;
228 r->membw.max_delay = eax.split.max_delay + 1;
229 r->default_ctrl = MAX_MBA_BW;
230 if (ecx & MBA_IS_LINEAR) {
231 r->membw.delay_linear = true;
232 r->membw.min_bw = MAX_MBA_BW - r->membw.max_delay;
233 r->membw.bw_gran = MAX_MBA_BW - r->membw.max_delay;
234 } else {
235 if (!rdt_get_mb_table(r))
236 return false;
237 }
238 r->data_width = 3;
239
240 r->alloc_capable = true;
241 r->alloc_enabled = true;
242
243 return true;
244 }
245
246 static void rdt_get_cache_alloc_cfg(int idx, struct rdt_resource *r)
247 {
248 union cpuid_0x10_1_eax eax;
249 union cpuid_0x10_x_edx edx;
250 u32 ebx, ecx;
251
252 cpuid_count(0x00000010, idx, &eax.full, &ebx, &ecx, &edx.full);
253 r->num_closid = edx.split.cos_max + 1;
254 r->cache.cbm_len = eax.split.cbm_len + 1;
255 r->default_ctrl = BIT_MASK(eax.split.cbm_len + 1) - 1;
256 r->cache.shareable_bits = ebx & r->default_ctrl;
257 r->data_width = (r->cache.cbm_len + 3) / 4;
258 r->alloc_capable = true;
259 r->alloc_enabled = true;
260 }
261
262 static void rdt_get_cdp_l3_config(int type)
263 {
264 struct rdt_resource *r_l3 = &rdt_resources_all[RDT_RESOURCE_L3];
265 struct rdt_resource *r = &rdt_resources_all[type];
266
267 r->num_closid = r_l3->num_closid / 2;
268 r->cache.cbm_len = r_l3->cache.cbm_len;
269 r->default_ctrl = r_l3->default_ctrl;
270 r->cache.shareable_bits = r_l3->cache.shareable_bits;
271 r->data_width = (r->cache.cbm_len + 3) / 4;
272 r->alloc_capable = true;
273 /*
274 * By default, CDP is disabled. CDP can be enabled by mount parameter
275 * "cdp" during resctrl file system mount time.
276 */
277 r->alloc_enabled = false;
278 }
279
280 static int get_cache_id(int cpu, int level)
281 {
282 struct cpu_cacheinfo *ci = get_cpu_cacheinfo(cpu);
283 int i;
284
285 for (i = 0; i < ci->num_leaves; i++) {
286 if (ci->info_list[i].level == level)
287 return ci->info_list[i].id;
288 }
289
290 return -1;
291 }
292
293 /*
294 * Map the memory b/w percentage value to delay values
295 * that can be written to QOS_MSRs.
296 * There are currently no SKUs which support non linear delay values.
297 */
298 static u32 delay_bw_map(unsigned long bw, struct rdt_resource *r)
299 {
300 if (r->membw.delay_linear)
301 return MAX_MBA_BW - bw;
302
303 pr_warn_once("Non Linear delay-bw map not supported but queried\n");
304 return r->default_ctrl;
305 }
306
307 static void
308 mba_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r)
309 {
310 unsigned int i;
311
312 /* Write the delay values for mba. */
313 for (i = m->low; i < m->high; i++)
314 wrmsrl(r->msr_base + i, delay_bw_map(d->ctrl_val[i], r));
315 }
316
317 static void
318 cat_wrmsr(struct rdt_domain *d, struct msr_param *m, struct rdt_resource *r)
319 {
320 unsigned int i;
321
322 for (i = m->low; i < m->high; i++)
323 wrmsrl(r->msr_base + cbm_idx(r, i), d->ctrl_val[i]);
324 }
325
326 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r)
327 {
328 struct rdt_domain *d;
329
330 list_for_each_entry(d, &r->domains, list) {
331 /* Find the domain that contains this CPU */
332 if (cpumask_test_cpu(cpu, &d->cpu_mask))
333 return d;
334 }
335
336 return NULL;
337 }
338
339 void rdt_ctrl_update(void *arg)
340 {
341 struct msr_param *m = arg;
342 struct rdt_resource *r = m->res;
343 int cpu = smp_processor_id();
344 struct rdt_domain *d;
345
346 d = get_domain_from_cpu(cpu, r);
347 if (d) {
348 r->msr_update(d, m, r);
349 return;
350 }
351 pr_warn_once("cpu %d not found in any domain for resource %s\n",
352 cpu, r->name);
353 }
354
355 /*
356 * rdt_find_domain - Find a domain in a resource that matches input resource id
357 *
358 * Search resource r's domain list to find the resource id. If the resource
359 * id is found in a domain, return the domain. Otherwise, if requested by
360 * caller, return the first domain whose id is bigger than the input id.
361 * The domain list is sorted by id in ascending order.
362 */
363 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
364 struct list_head **pos)
365 {
366 struct rdt_domain *d;
367 struct list_head *l;
368
369 if (id < 0)
370 return ERR_PTR(id);
371
372 list_for_each(l, &r->domains) {
373 d = list_entry(l, struct rdt_domain, list);
374 /* When id is found, return its domain. */
375 if (id == d->id)
376 return d;
377 /* Stop searching when finding id's position in sorted list. */
378 if (id < d->id)
379 break;
380 }
381
382 if (pos)
383 *pos = l;
384
385 return NULL;
386 }
387
388 static int domain_setup_ctrlval(struct rdt_resource *r, struct rdt_domain *d)
389 {
390 struct msr_param m;
391 u32 *dc;
392 int i;
393
394 dc = kmalloc_array(r->num_closid, sizeof(*d->ctrl_val), GFP_KERNEL);
395 if (!dc)
396 return -ENOMEM;
397
398 d->ctrl_val = dc;
399
400 /*
401 * Initialize the Control MSRs to having no control.
402 * For Cache Allocation: Set all bits in cbm
403 * For Memory Allocation: Set b/w requested to 100
404 */
405 for (i = 0; i < r->num_closid; i++, dc++)
406 *dc = r->default_ctrl;
407
408 m.low = 0;
409 m.high = r->num_closid;
410 r->msr_update(d, &m, r);
411 return 0;
412 }
413
414 static int domain_setup_mon_state(struct rdt_resource *r, struct rdt_domain *d)
415 {
416 size_t tsize;
417
418 if (is_llc_occupancy_enabled()) {
419 d->rmid_busy_llc = kcalloc(BITS_TO_LONGS(r->num_rmid),
420 sizeof(unsigned long),
421 GFP_KERNEL);
422 if (!d->rmid_busy_llc)
423 return -ENOMEM;
424 INIT_DELAYED_WORK(&d->cqm_limbo, cqm_handle_limbo);
425 }
426 if (is_mbm_total_enabled()) {
427 tsize = sizeof(*d->mbm_total);
428 d->mbm_total = kcalloc(r->num_rmid, tsize, GFP_KERNEL);
429 if (!d->mbm_total) {
430 kfree(d->rmid_busy_llc);
431 return -ENOMEM;
432 }
433 }
434 if (is_mbm_local_enabled()) {
435 tsize = sizeof(*d->mbm_local);
436 d->mbm_local = kcalloc(r->num_rmid, tsize, GFP_KERNEL);
437 if (!d->mbm_local) {
438 kfree(d->rmid_busy_llc);
439 kfree(d->mbm_total);
440 return -ENOMEM;
441 }
442 }
443
444 if (is_mbm_enabled()) {
445 INIT_DELAYED_WORK(&d->mbm_over, mbm_handle_overflow);
446 mbm_setup_overflow_handler(d, MBM_OVERFLOW_INTERVAL);
447 }
448
449 return 0;
450 }
451
452 /*
453 * domain_add_cpu - Add a cpu to a resource's domain list.
454 *
455 * If an existing domain in the resource r's domain list matches the cpu's
456 * resource id, add the cpu in the domain.
457 *
458 * Otherwise, a new domain is allocated and inserted into the right position
459 * in the domain list sorted by id in ascending order.
460 *
461 * The order in the domain list is visible to users when we print entries
462 * in the schemata file and schemata input is validated to have the same order
463 * as this list.
464 */
465 static void domain_add_cpu(int cpu, struct rdt_resource *r)
466 {
467 int id = get_cache_id(cpu, r->cache_level);
468 struct list_head *add_pos = NULL;
469 struct rdt_domain *d;
470
471 d = rdt_find_domain(r, id, &add_pos);
472 if (IS_ERR(d)) {
473 pr_warn("Could't find cache id for cpu %d\n", cpu);
474 return;
475 }
476
477 if (d) {
478 cpumask_set_cpu(cpu, &d->cpu_mask);
479 return;
480 }
481
482 d = kzalloc_node(sizeof(*d), GFP_KERNEL, cpu_to_node(cpu));
483 if (!d)
484 return;
485
486 d->id = id;
487 cpumask_set_cpu(cpu, &d->cpu_mask);
488
489 if (r->alloc_capable && domain_setup_ctrlval(r, d)) {
490 kfree(d);
491 return;
492 }
493
494 if (r->mon_capable && domain_setup_mon_state(r, d)) {
495 kfree(d);
496 return;
497 }
498
499 list_add_tail(&d->list, add_pos);
500
501 /*
502 * If resctrl is mounted, add
503 * per domain monitor data directories.
504 */
505 if (static_branch_unlikely(&rdt_mon_enable_key))
506 mkdir_mondata_subdir_allrdtgrp(r, d);
507 }
508
509 static void domain_remove_cpu(int cpu, struct rdt_resource *r)
510 {
511 int id = get_cache_id(cpu, r->cache_level);
512 struct rdt_domain *d;
513
514 d = rdt_find_domain(r, id, NULL);
515 if (IS_ERR_OR_NULL(d)) {
516 pr_warn("Could't find cache id for cpu %d\n", cpu);
517 return;
518 }
519
520 cpumask_clear_cpu(cpu, &d->cpu_mask);
521 if (cpumask_empty(&d->cpu_mask)) {
522 /*
523 * If resctrl is mounted, remove all the
524 * per domain monitor data directories.
525 */
526 if (static_branch_unlikely(&rdt_mon_enable_key))
527 rmdir_mondata_subdir_allrdtgrp(r, d->id);
528 kfree(d->ctrl_val);
529 kfree(d->rmid_busy_llc);
530 kfree(d->mbm_total);
531 kfree(d->mbm_local);
532 list_del(&d->list);
533 if (is_mbm_enabled())
534 cancel_delayed_work(&d->mbm_over);
535 if (is_llc_occupancy_enabled() && has_busy_rmid(r, d)) {
536 /*
537 * When a package is going down, forcefully
538 * decrement rmid->ebusy. There is no way to know
539 * that the L3 was flushed and hence may lead to
540 * incorrect counts in rare scenarios, but leaving
541 * the RMID as busy creates RMID leaks if the
542 * package never comes back.
543 */
544 __check_limbo(d, true);
545 cancel_delayed_work(&d->cqm_limbo);
546 }
547
548 kfree(d);
549 return;
550 }
551
552 if (r == &rdt_resources_all[RDT_RESOURCE_L3]) {
553 if (is_mbm_enabled() && cpu == d->mbm_work_cpu) {
554 cancel_delayed_work(&d->mbm_over);
555 mbm_setup_overflow_handler(d, 0);
556 }
557 if (is_llc_occupancy_enabled() && cpu == d->cqm_work_cpu &&
558 has_busy_rmid(r, d)) {
559 cancel_delayed_work(&d->cqm_limbo);
560 cqm_setup_limbo_handler(d, 0);
561 }
562 }
563 }
564
565 static void clear_closid_rmid(int cpu)
566 {
567 struct intel_pqr_state *state = this_cpu_ptr(&pqr_state);
568
569 state->default_closid = 0;
570 state->default_rmid = 0;
571 state->cur_closid = 0;
572 state->cur_rmid = 0;
573 wrmsr(IA32_PQR_ASSOC, 0, 0);
574 }
575
576 static int intel_rdt_online_cpu(unsigned int cpu)
577 {
578 struct rdt_resource *r;
579
580 mutex_lock(&rdtgroup_mutex);
581 for_each_capable_rdt_resource(r)
582 domain_add_cpu(cpu, r);
583 /* The cpu is set in default rdtgroup after online. */
584 cpumask_set_cpu(cpu, &rdtgroup_default.cpu_mask);
585 clear_closid_rmid(cpu);
586 mutex_unlock(&rdtgroup_mutex);
587
588 return 0;
589 }
590
591 static void clear_childcpus(struct rdtgroup *r, unsigned int cpu)
592 {
593 struct rdtgroup *cr;
594
595 list_for_each_entry(cr, &r->mon.crdtgrp_list, mon.crdtgrp_list) {
596 if (cpumask_test_and_clear_cpu(cpu, &cr->cpu_mask)) {
597 break;
598 }
599 }
600 }
601
602 static int intel_rdt_offline_cpu(unsigned int cpu)
603 {
604 struct rdtgroup *rdtgrp;
605 struct rdt_resource *r;
606
607 mutex_lock(&rdtgroup_mutex);
608 for_each_capable_rdt_resource(r)
609 domain_remove_cpu(cpu, r);
610 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
611 if (cpumask_test_and_clear_cpu(cpu, &rdtgrp->cpu_mask)) {
612 clear_childcpus(rdtgrp, cpu);
613 break;
614 }
615 }
616 clear_closid_rmid(cpu);
617 mutex_unlock(&rdtgroup_mutex);
618
619 return 0;
620 }
621
622 /*
623 * Choose a width for the resource name and resource data based on the
624 * resource that has widest name and cbm.
625 */
626 static __init void rdt_init_padding(void)
627 {
628 struct rdt_resource *r;
629 int cl;
630
631 for_each_alloc_capable_rdt_resource(r) {
632 cl = strlen(r->name);
633 if (cl > max_name_width)
634 max_name_width = cl;
635
636 if (r->data_width > max_data_width)
637 max_data_width = r->data_width;
638 }
639 }
640
641 enum {
642 RDT_FLAG_CMT,
643 RDT_FLAG_MBM_TOTAL,
644 RDT_FLAG_MBM_LOCAL,
645 RDT_FLAG_L3_CAT,
646 RDT_FLAG_L3_CDP,
647 RDT_FLAG_L2_CAT,
648 RDT_FLAG_MBA,
649 };
650
651 #define RDT_OPT(idx, n, f) \
652 [idx] = { \
653 .name = n, \
654 .flag = f \
655 }
656
657 struct rdt_options {
658 char *name;
659 int flag;
660 bool force_off, force_on;
661 };
662
663 static struct rdt_options rdt_options[] __initdata = {
664 RDT_OPT(RDT_FLAG_CMT, "cmt", X86_FEATURE_CQM_OCCUP_LLC),
665 RDT_OPT(RDT_FLAG_MBM_TOTAL, "mbmtotal", X86_FEATURE_CQM_MBM_TOTAL),
666 RDT_OPT(RDT_FLAG_MBM_LOCAL, "mbmlocal", X86_FEATURE_CQM_MBM_LOCAL),
667 RDT_OPT(RDT_FLAG_L3_CAT, "l3cat", X86_FEATURE_CAT_L3),
668 RDT_OPT(RDT_FLAG_L3_CDP, "l3cdp", X86_FEATURE_CDP_L3),
669 RDT_OPT(RDT_FLAG_L2_CAT, "l2cat", X86_FEATURE_CAT_L2),
670 RDT_OPT(RDT_FLAG_MBA, "mba", X86_FEATURE_MBA),
671 };
672 #define NUM_RDT_OPTIONS ARRAY_SIZE(rdt_options)
673
674 static int __init set_rdt_options(char *str)
675 {
676 struct rdt_options *o;
677 bool force_off;
678 char *tok;
679
680 if (*str == '=')
681 str++;
682 while ((tok = strsep(&str, ",")) != NULL) {
683 force_off = *tok == '!';
684 if (force_off)
685 tok++;
686 for (o = rdt_options; o < &rdt_options[NUM_RDT_OPTIONS]; o++) {
687 if (strcmp(tok, o->name) == 0) {
688 if (force_off)
689 o->force_off = true;
690 else
691 o->force_on = true;
692 break;
693 }
694 }
695 }
696 return 1;
697 }
698 __setup("rdt", set_rdt_options);
699
700 static bool __init rdt_cpu_has(int flag)
701 {
702 bool ret = boot_cpu_has(flag);
703 struct rdt_options *o;
704
705 if (!ret)
706 return ret;
707
708 for (o = rdt_options; o < &rdt_options[NUM_RDT_OPTIONS]; o++) {
709 if (flag == o->flag) {
710 if (o->force_off)
711 ret = false;
712 if (o->force_on)
713 ret = true;
714 break;
715 }
716 }
717 return ret;
718 }
719
720 static __init bool get_rdt_alloc_resources(void)
721 {
722 bool ret = false;
723
724 if (rdt_alloc_capable)
725 return true;
726
727 if (!boot_cpu_has(X86_FEATURE_RDT_A))
728 return false;
729
730 if (rdt_cpu_has(X86_FEATURE_CAT_L3)) {
731 rdt_get_cache_alloc_cfg(1, &rdt_resources_all[RDT_RESOURCE_L3]);
732 if (rdt_cpu_has(X86_FEATURE_CDP_L3)) {
733 rdt_get_cdp_l3_config(RDT_RESOURCE_L3DATA);
734 rdt_get_cdp_l3_config(RDT_RESOURCE_L3CODE);
735 }
736 ret = true;
737 }
738 if (rdt_cpu_has(X86_FEATURE_CAT_L2)) {
739 /* CPUID 0x10.2 fields are same format at 0x10.1 */
740 rdt_get_cache_alloc_cfg(2, &rdt_resources_all[RDT_RESOURCE_L2]);
741 ret = true;
742 }
743
744 if (rdt_cpu_has(X86_FEATURE_MBA)) {
745 if (rdt_get_mem_config(&rdt_resources_all[RDT_RESOURCE_MBA]))
746 ret = true;
747 }
748 return ret;
749 }
750
751 static __init bool get_rdt_mon_resources(void)
752 {
753 if (rdt_cpu_has(X86_FEATURE_CQM_OCCUP_LLC))
754 rdt_mon_features |= (1 << QOS_L3_OCCUP_EVENT_ID);
755 if (rdt_cpu_has(X86_FEATURE_CQM_MBM_TOTAL))
756 rdt_mon_features |= (1 << QOS_L3_MBM_TOTAL_EVENT_ID);
757 if (rdt_cpu_has(X86_FEATURE_CQM_MBM_LOCAL))
758 rdt_mon_features |= (1 << QOS_L3_MBM_LOCAL_EVENT_ID);
759
760 if (!rdt_mon_features)
761 return false;
762
763 return !rdt_get_mon_l3_config(&rdt_resources_all[RDT_RESOURCE_L3]);
764 }
765
766 static __init void rdt_quirks(void)
767 {
768 switch (boot_cpu_data.x86_model) {
769 case INTEL_FAM6_HASWELL_X:
770 if (!rdt_options[RDT_FLAG_L3_CAT].force_off)
771 cache_alloc_hsw_probe();
772 break;
773 case INTEL_FAM6_SKYLAKE_X:
774 if (boot_cpu_data.x86_mask <= 4)
775 set_rdt_options("!cmt,!mbmtotal,!mbmlocal,!l3cat");
776 }
777 }
778
779 static __init bool get_rdt_resources(void)
780 {
781 rdt_quirks();
782 rdt_alloc_capable = get_rdt_alloc_resources();
783 rdt_mon_capable = get_rdt_mon_resources();
784
785 return (rdt_mon_capable || rdt_alloc_capable);
786 }
787
788 static int __init intel_rdt_late_init(void)
789 {
790 struct rdt_resource *r;
791 int state, ret;
792
793 if (!get_rdt_resources())
794 return -ENODEV;
795
796 rdt_init_padding();
797
798 state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN,
799 "x86/rdt/cat:online:",
800 intel_rdt_online_cpu, intel_rdt_offline_cpu);
801 if (state < 0)
802 return state;
803
804 ret = rdtgroup_init();
805 if (ret) {
806 cpuhp_remove_state(state);
807 return ret;
808 }
809
810 for_each_alloc_capable_rdt_resource(r)
811 pr_info("Intel RDT %s allocation detected\n", r->name);
812
813 for_each_mon_capable_rdt_resource(r)
814 pr_info("Intel RDT %s monitoring detected\n", r->name);
815
816 return 0;
817 }
818
819 late_initcall(intel_rdt_late_init);