arch/x86/kernel/cpu/intel_rdt_ctrlmondata.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  *
  11  * This program is free software; you can redistribute it and/or modify it
  12  * under the terms and conditions of the GNU General Public License,
  13  * version 2, as published by the Free Software Foundation.
  14  *
  15  * This program is distributed in the hope it will be useful, but WITHOUT
  16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  17  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  18  * more details.
  19  *
  20  * More information about RDT be found in the Intel (R) x86 Architecture
  21  * Software Developer Manual June 2016, volume 3, section 17.17.
  22  */
  23
  24 #define pr_fmt(fmt)     KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/kernfs.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/slab.h>
  29 #include "intel_rdt.h"
  30
  31 /*
  32  * Check whether MBA bandwidth percentage value is correct. The value is
  33  * checked against the minimum and max bandwidth values specified by the
  34  * hardware. The allocated bandwidth percentage is rounded to the next
  35  * control step available on the hardware.
  36  */
  37 static bool bw_validate(char *buf, unsigned long *data, struct rdt_resource *r)
  38 {
  39         unsigned long bw;
  40         int ret;
  41
  42         /*
  43          * Only linear delay values is supported for current Intel SKUs.
  44          */
  45         if (!r->membw.delay_linear) {
  46                 rdt_last_cmd_puts("No support for non-linear MB domains\n");
  47                 return false;
  48         }
  49
  50         ret = kstrtoul(buf, 10, &bw);
  51         if (ret) {
  52                 rdt_last_cmd_printf("Non-decimal digit in MB value %s\n", buf);
  53                 return false;
  54         }
  55
  56         if ((bw < r->membw.min_bw || bw > r->default_ctrl) &&
  57             !is_mba_sc(r)) {
  58                 rdt_last_cmd_printf("MB value %ld out of range [%d,%d]\n", bw,
  59                                     r->membw.min_bw, r->default_ctrl);
  60                 return false;
  61         }
  62
  63         *data = roundup(bw, (unsigned long)r->membw.bw_gran);
  64         return true;
  65 }
  66
  67 int parse_bw(struct rdt_parse_data *data, struct rdt_resource *r,
  68              struct rdt_domain *d)
  69 {
  70         unsigned long bw_val;
  71
  72         if (d->have_new_ctrl) {
  73                 rdt_last_cmd_printf("duplicate domain %d\n", d->id);
  74                 return -EINVAL;
  75         }
  76
  77         if (!bw_validate(data->buf, &bw_val, r))
  78                 return -EINVAL;
  79         d->new_ctrl = bw_val;
  80         d->have_new_ctrl = true;
  81
  82         return 0;
  83 }
  84
  85 /*
  86  * Check whether a cache bit mask is valid. The SDM says:
  87  *      Please note that all (and only) contiguous '1' combinations
  88  *      are allowed (e.g. FFFFH, 0FF0H, 003CH, etc.).
  89  * Additionally Haswell requires at least two bits set.
  90  */
  91 static bool cbm_validate(char *buf, u32 *data, struct rdt_resource *r)
  92 {
  93         unsigned long first_bit, zero_bit, val;
  94         unsigned int cbm_len = r->cache.cbm_len;
  95         int ret;
  96
  97         ret = kstrtoul(buf, 16, &val);
  98         if (ret) {
  99                 rdt_last_cmd_printf("non-hex character in mask %s\n", buf);
 100                 return false;
 101         }
 102
 103         if (val == 0 || val > r->default_ctrl) {
 104                 rdt_last_cmd_puts("mask out of range\n");
 105                 return false;
 106         }
 107
 108         first_bit = find_first_bit(&val, cbm_len);
 109         zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
 110
 111         if (find_next_bit(&val, cbm_len, zero_bit) < cbm_len) {
 112                 rdt_last_cmd_printf("mask %lx has non-consecutive 1-bits\n", val);
 113                 return false;
 114         }
 115
 116         if ((zero_bit - first_bit) < r->cache.min_cbm_bits) {
 117                 rdt_last_cmd_printf("Need at least %d bits in mask\n",
 118                                     r->cache.min_cbm_bits);
 119                 return false;
 120         }
 121
 122         *data = val;
 123         return true;
 124 }
 125
 126 /*
 127  * Read one cache bit mask (hex). Check that it is valid for the current
 128  * resource type.
 129  */
 130 int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
 131               struct rdt_domain *d)
 132 {
 133         struct rdtgroup *rdtgrp = data->rdtgrp;
 134         u32 cbm_val;
 135
 136         if (d->have_new_ctrl) {
 137                 rdt_last_cmd_printf("duplicate domain %d\n", d->id);
 138                 return -EINVAL;
 139         }
 140
 141         /*
 142          * Cannot set up more than one pseudo-locked region in a cache
 143          * hierarchy.
 144          */
 145         if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP &&
 146             rdtgroup_pseudo_locked_in_hierarchy(d)) {
 147                 rdt_last_cmd_printf("pseudo-locked region in hierarchy\n");
 148                 return -EINVAL;
 149         }
 150
 151         if (!cbm_validate(data->buf, &cbm_val, r))
 152                 return -EINVAL;
 153
 154         if ((rdtgrp->mode == RDT_MODE_EXCLUSIVE ||
 155              rdtgrp->mode == RDT_MODE_SHAREABLE) &&
 156             rdtgroup_cbm_overlaps_pseudo_locked(d, cbm_val)) {
 157                 rdt_last_cmd_printf("CBM overlaps with pseudo-locked region\n");
 158                 return -EINVAL;
 159         }
 160
 161         /*
 162          * The CBM may not overlap with the CBM of another closid if
 163          * either is exclusive.
 164          */
 165         if (rdtgroup_cbm_overlaps(r, d, cbm_val, rdtgrp->closid, true)) {
 166                 rdt_last_cmd_printf("overlaps with exclusive group\n");
 167                 return -EINVAL;
 168         }
 169
 170         if (rdtgroup_cbm_overlaps(r, d, cbm_val, rdtgrp->closid, false)) {
 171                 if (rdtgrp->mode == RDT_MODE_EXCLUSIVE ||
 172                     rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
 173                         rdt_last_cmd_printf("overlaps with other group\n");
 174                         return -EINVAL;
 175                 }
 176         }
 177
 178         d->new_ctrl = cbm_val;
 179         d->have_new_ctrl = true;
 180
 181         return 0;
 182 }
 183
 184 /*
 185  * For each domain in this resource we expect to find a series of:
 186  *      id=mask
 187  * separated by ";". The "id" is in decimal, and must match one of
 188  * the "id"s for this resource.
 189  */
 190 static int parse_line(char *line, struct rdt_resource *r,
 191                       struct rdtgroup *rdtgrp)
 192 {
 193         struct rdt_parse_data data;
 194         char *dom = NULL, *id;
 195         struct rdt_domain *d;
 196         unsigned long dom_id;
 197
 198         if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP &&
 199             r->rid == RDT_RESOURCE_MBA) {
 200                 rdt_last_cmd_puts("Cannot pseudo-lock MBA resource\n");
 201                 return -EINVAL;
 202         }
 203
 204 next:
 205         if (!line || line[0] == '\0')
 206                 return 0;
 207         dom = strsep(&line, ";");
 208         id = strsep(&dom, "=");
 209         if (!dom || kstrtoul(id, 10, &dom_id)) {
 210                 rdt_last_cmd_puts("Missing '=' or non-numeric domain\n");
 211                 return -EINVAL;
 212         }
 213         dom = strim(dom);
 214         list_for_each_entry(d, &r->domains, list) {
 215                 if (d->id == dom_id) {
 216                         data.buf = dom;
 217                         data.rdtgrp = rdtgrp;
 218                         if (r->parse_ctrlval(&data, r, d))
 219                                 return -EINVAL;
 220                         if (rdtgrp->mode ==  RDT_MODE_PSEUDO_LOCKSETUP) {
 221                                 /*
 222                                  * In pseudo-locking setup mode and just
 223                                  * parsed a valid CBM that should be
 224                                  * pseudo-locked. Only one locked region per
 225                                  * resource group and domain so just do
 226                                  * the required initialization for single
 227                                  * region and return.
 228                                  */
 229                                 rdtgrp->plr->r = r;
 230                                 rdtgrp->plr->d = d;
 231                                 rdtgrp->plr->cbm = d->new_ctrl;
 232                                 d->plr = rdtgrp->plr;
 233                                 return 0;
 234                         }
 235                         goto next;
 236                 }
 237         }
 238         return -EINVAL;
 239 }
 240
 241 int update_domains(struct rdt_resource *r, int closid)
 242 {
 243         struct msr_param msr_param;
 244         cpumask_var_t cpu_mask;
 245         struct rdt_domain *d;
 246         bool mba_sc;
 247         u32 *dc;
 248         int cpu;
 249
 250         if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
 251                 return -ENOMEM;
 252
 253         msr_param.low = closid;
 254         msr_param.high = msr_param.low + 1;
 255         msr_param.res = r;
 256
 257         mba_sc = is_mba_sc(r);
 258         list_for_each_entry(d, &r->domains, list) {
 259                 dc = !mba_sc ? d->ctrl_val : d->mbps_val;
 260                 if (d->have_new_ctrl && d->new_ctrl != dc[closid]) {
 261                         cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
 262                         dc[closid] = d->new_ctrl;
 263                 }
 264         }
 265
 266         /*
 267          * Avoid writing the control msr with control values when
 268          * MBA software controller is enabled
 269          */
 270         if (cpumask_empty(cpu_mask) || mba_sc)
 271                 goto done;
 272         cpu = get_cpu();
 273         /* Update CBM on this cpu if it's in cpu_mask. */
 274         if (cpumask_test_cpu(cpu, cpu_mask))
 275                 rdt_ctrl_update(&msr_param);
 276         /* Update CBM on other cpus. */
 277         smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
 278         put_cpu();
 279
 280 done:
 281         free_cpumask_var(cpu_mask);
 282
 283         return 0;
 284 }
 285
 286 static int rdtgroup_parse_resource(char *resname, char *tok,
 287                                    struct rdtgroup *rdtgrp)
 288 {
 289         struct rdt_resource *r;
 290
 291         for_each_alloc_enabled_rdt_resource(r) {
 292                 if (!strcmp(resname, r->name) && rdtgrp->closid < r->num_closid)
 293                         return parse_line(tok, r, rdtgrp);
 294         }
 295         rdt_last_cmd_printf("unknown/unsupported resource name '%s'\n", resname);
 296         return -EINVAL;
 297 }
 298
 299 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
 300                                 char *buf, size_t nbytes, loff_t off)
 301 {
 302         struct rdtgroup *rdtgrp;
 303         struct rdt_domain *dom;
 304         struct rdt_resource *r;
 305         char *tok, *resname;
 306         int ret = 0;
 307
 308         /* Valid input requires a trailing newline */
 309         if (nbytes == 0 || buf[nbytes - 1] != '\n')
 310                 return -EINVAL;
 311         buf[nbytes - 1] = '\0';
 312
 313         rdtgrp = rdtgroup_kn_lock_live(of->kn);
 314         if (!rdtgrp) {
 315                 rdtgroup_kn_unlock(of->kn);
 316                 return -ENOENT;
 317         }
 318         rdt_last_cmd_clear();
 319
 320         /*
 321          * No changes to pseudo-locked region allowed. It has to be removed
 322          * and re-created instead.
 323          */
 324         if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
 325                 ret = -EINVAL;
 326                 rdt_last_cmd_puts("resource group is pseudo-locked\n");
 327                 goto out;
 328         }
 329
 330         for_each_alloc_enabled_rdt_resource(r) {
 331                 list_for_each_entry(dom, &r->domains, list)
 332                         dom->have_new_ctrl = false;
 333         }
 334
 335         while ((tok = strsep(&buf, "\n")) != NULL) {
 336                 resname = strim(strsep(&tok, ":"));
 337                 if (!tok) {
 338                         rdt_last_cmd_puts("Missing ':'\n");
 339                         ret = -EINVAL;
 340                         goto out;
 341                 }
 342                 if (tok[0] == '\0') {
 343                         rdt_last_cmd_printf("Missing '%s' value\n", resname);
 344                         ret = -EINVAL;
 345                         goto out;
 346                 }
 347                 ret = rdtgroup_parse_resource(resname, tok, rdtgrp);
 348                 if (ret)
 349                         goto out;
 350         }
 351
 352         for_each_alloc_enabled_rdt_resource(r) {
 353                 ret = update_domains(r, rdtgrp->closid);
 354                 if (ret)
 355                         goto out;
 356         }
 357
 358         if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
 359                 /*
 360                  * If pseudo-locking fails we keep the resource group in
 361                  * mode RDT_MODE_PSEUDO_LOCKSETUP with its class of service
 362                  * active and updated for just the domain the pseudo-locked
 363                  * region was requested for.
 364                  */
 365                 ret = rdtgroup_pseudo_lock_create(rdtgrp);
 366         }
 367
 368 out:
 369         rdtgroup_kn_unlock(of->kn);
 370         return ret ?: nbytes;
 371 }
 372
 373 static void show_doms(struct seq_file *s, struct rdt_resource *r, int closid)
 374 {
 375         struct rdt_domain *dom;
 376         bool sep = false;
 377         u32 ctrl_val;
 378
 379         seq_printf(s, "%*s:", max_name_width, r->name);
 380         list_for_each_entry(dom, &r->domains, list) {
 381                 if (sep)
 382                         seq_puts(s, ";");
 383
 384                 ctrl_val = (!is_mba_sc(r) ? dom->ctrl_val[closid] :
 385                             dom->mbps_val[closid]);
 386                 seq_printf(s, r->format_str, dom->id, max_data_width,
 387                            ctrl_val);
 388                 sep = true;
 389         }
 390         seq_puts(s, "\n");
 391 }
 392
 393 int rdtgroup_schemata_show(struct kernfs_open_file *of,
 394                            struct seq_file *s, void *v)
 395 {
 396         struct rdtgroup *rdtgrp;
 397         struct rdt_resource *r;
 398         int ret = 0;
 399         u32 closid;
 400
 401         rdtgrp = rdtgroup_kn_lock_live(of->kn);
 402         if (rdtgrp) {
 403                 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
 404                         for_each_alloc_enabled_rdt_resource(r)
 405                                 seq_printf(s, "%s:uninitialized\n", r->name);
 406                 } else if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
 407                         if (!rdtgrp->plr->d) {
 408                                 rdt_last_cmd_clear();
 409                                 rdt_last_cmd_puts("Cache domain offline\n");
 410                                 ret = -ENODEV;
 411                         } else {
 412                                 seq_printf(s, "%s:%d=%x\n",
 413                                            rdtgrp->plr->r->name,
 414                                            rdtgrp->plr->d->id,
 415                                            rdtgrp->plr->cbm);
 416                         }
 417                 } else {
 418                         closid = rdtgrp->closid;
 419                         for_each_alloc_enabled_rdt_resource(r) {
 420                                 if (closid < r->num_closid)
 421                                         show_doms(s, r, closid);
 422                         }
 423                 }
 424         } else {
 425                 ret = -ENOENT;
 426         }
 427         rdtgroup_kn_unlock(of->kn);
 428         return ret;
 429 }
 430
 431 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
 432                     struct rdtgroup *rdtgrp, int evtid, int first)
 433 {
 434         /*
 435          * setup the parameters to send to the IPI to read the data.
 436          */
 437         rr->rgrp = rdtgrp;
 438         rr->evtid = evtid;
 439         rr->d = d;
 440         rr->val = 0;
 441         rr->first = first;
 442
 443         smp_call_function_any(&d->cpu_mask, mon_event_count, rr, 1);
 444 }
 445
 446 int rdtgroup_mondata_show(struct seq_file *m, void *arg)
 447 {
 448         struct kernfs_open_file *of = m->private;
 449         u32 resid, evtid, domid;
 450         struct rdtgroup *rdtgrp;
 451         struct rdt_resource *r;
 452         union mon_data_bits md;
 453         struct rdt_domain *d;
 454         struct rmid_read rr;
 455         int ret = 0;
 456
 457         rdtgrp = rdtgroup_kn_lock_live(of->kn);
 458
 459         md.priv = of->kn->priv;
 460         resid = md.u.rid;
 461         domid = md.u.domid;
 462         evtid = md.u.evtid;
 463
 464         r = &rdt_resources_all[resid];
 465         d = rdt_find_domain(r, domid, NULL);
 466         if (!d) {
 467                 ret = -ENOENT;
 468                 goto out;
 469         }
 470
 471         mon_event_read(&rr, d, rdtgrp, evtid, false);
 472
 473         if (rr.val & RMID_VAL_ERROR)
 474                 seq_puts(m, "Error\n");
 475         else if (rr.val & RMID_VAL_UNAVAIL)
 476                 seq_puts(m, "Unavailable\n");
 477         else
 478                 seq_printf(m, "%llu\n", rr.val * r->mon_scale);
 479
 480 out:
 481         rdtgroup_kn_unlock(of->kn);
 482         return ret;
 483 }