2 * Resource Director Technology(RDT)
3 * - Cache Allocation code.
5 * Copyright (C) 2016 Intel Corporation
8 * Fenghua Yu <fenghua.yu@intel.com>
9 * Tony Luck <tony.luck@intel.com>
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.
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
20 * More information about RDT be found in the Intel (R) x86 Architecture
21 * Software Developer Manual June 2016, volume 3, section 17.17.
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 #include <linux/cpu.h>
27 #include <linux/kernfs.h>
28 #include <linux/seq_file.h>
29 #include <linux/slab.h>
30 #include "intel_rdt.h"
33 * Check whether MBA bandwidth percentage value is correct. The value is
34 * checked against the minimum and max bandwidth values specified by the
35 * hardware. The allocated bandwidth percentage is rounded to the next
36 * control step available on the hardware.
38 static bool bw_validate(char *buf, unsigned long *data, struct rdt_resource *r)
44 * Only linear delay values is supported for current Intel SKUs.
46 if (!r->membw.delay_linear) {
47 rdt_last_cmd_puts("No support for non-linear MB domains\n");
51 ret = kstrtoul(buf, 10, &bw);
53 rdt_last_cmd_printf("Non-decimal digit in MB value %s\n", buf);
57 if ((bw < r->membw.min_bw || bw > r->default_ctrl) &&
59 rdt_last_cmd_printf("MB value %ld out of range [%d,%d]\n", bw,
60 r->membw.min_bw, r->default_ctrl);
64 *data = roundup(bw, (unsigned long)r->membw.bw_gran);
68 int parse_bw(struct rdt_parse_data *data, struct rdt_resource *r,
73 if (d->have_new_ctrl) {
74 rdt_last_cmd_printf("duplicate domain %d\n", d->id);
78 if (!bw_validate(data->buf, &bw_val, r))
81 d->have_new_ctrl = true;
87 * Check whether a cache bit mask is valid. The SDM says:
88 * Please note that all (and only) contiguous '1' combinations
89 * are allowed (e.g. FFFFH, 0FF0H, 003CH, etc.).
90 * Additionally Haswell requires at least two bits set.
92 static bool cbm_validate(char *buf, u32 *data, struct rdt_resource *r)
94 unsigned long first_bit, zero_bit, val;
95 unsigned int cbm_len = r->cache.cbm_len;
98 ret = kstrtoul(buf, 16, &val);
100 rdt_last_cmd_printf("non-hex character in mask %s\n", buf);
104 if (val == 0 || val > r->default_ctrl) {
105 rdt_last_cmd_puts("mask out of range\n");
109 first_bit = find_first_bit(&val, cbm_len);
110 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
112 if (find_next_bit(&val, cbm_len, zero_bit) < cbm_len) {
113 rdt_last_cmd_printf("mask %lx has non-consecutive 1-bits\n", val);
117 if ((zero_bit - first_bit) < r->cache.min_cbm_bits) {
118 rdt_last_cmd_printf("Need at least %d bits in mask\n",
119 r->cache.min_cbm_bits);
128 * Read one cache bit mask (hex). Check that it is valid for the current
131 int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
132 struct rdt_domain *d)
134 struct rdtgroup *rdtgrp = data->rdtgrp;
137 if (d->have_new_ctrl) {
138 rdt_last_cmd_printf("duplicate domain %d\n", d->id);
143 * Cannot set up more than one pseudo-locked region in a cache
146 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP &&
147 rdtgroup_pseudo_locked_in_hierarchy(d)) {
148 rdt_last_cmd_printf("pseudo-locked region in hierarchy\n");
152 if (!cbm_validate(data->buf, &cbm_val, r))
155 if ((rdtgrp->mode == RDT_MODE_EXCLUSIVE ||
156 rdtgrp->mode == RDT_MODE_SHAREABLE) &&
157 rdtgroup_cbm_overlaps_pseudo_locked(d, cbm_val)) {
158 rdt_last_cmd_printf("CBM overlaps with pseudo-locked region\n");
163 * The CBM may not overlap with the CBM of another closid if
164 * either is exclusive.
166 if (rdtgroup_cbm_overlaps(r, d, cbm_val, rdtgrp->closid, true)) {
167 rdt_last_cmd_printf("overlaps with exclusive group\n");
171 if (rdtgroup_cbm_overlaps(r, d, cbm_val, rdtgrp->closid, false)) {
172 if (rdtgrp->mode == RDT_MODE_EXCLUSIVE ||
173 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
174 rdt_last_cmd_printf("overlaps with other group\n");
179 d->new_ctrl = cbm_val;
180 d->have_new_ctrl = true;
186 * For each domain in this resource we expect to find a series of:
188 * separated by ";". The "id" is in decimal, and must match one of
189 * the "id"s for this resource.
191 static int parse_line(char *line, struct rdt_resource *r,
192 struct rdtgroup *rdtgrp)
194 struct rdt_parse_data data;
195 char *dom = NULL, *id;
196 struct rdt_domain *d;
197 unsigned long dom_id;
199 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP &&
200 r->rid == RDT_RESOURCE_MBA) {
201 rdt_last_cmd_puts("Cannot pseudo-lock MBA resource\n");
206 if (!line || line[0] == '\0')
208 dom = strsep(&line, ";");
209 id = strsep(&dom, "=");
210 if (!dom || kstrtoul(id, 10, &dom_id)) {
211 rdt_last_cmd_puts("Missing '=' or non-numeric domain\n");
215 list_for_each_entry(d, &r->domains, list) {
216 if (d->id == dom_id) {
218 data.rdtgrp = rdtgrp;
219 if (r->parse_ctrlval(&data, r, d))
221 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
223 * In pseudo-locking setup mode and just
224 * parsed a valid CBM that should be
225 * pseudo-locked. Only one locked region per
226 * resource group and domain so just do
227 * the required initialization for single
232 rdtgrp->plr->cbm = d->new_ctrl;
233 d->plr = rdtgrp->plr;
242 int update_domains(struct rdt_resource *r, int closid)
244 struct msr_param msr_param;
245 cpumask_var_t cpu_mask;
246 struct rdt_domain *d;
251 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
254 msr_param.low = closid;
255 msr_param.high = msr_param.low + 1;
258 mba_sc = is_mba_sc(r);
259 list_for_each_entry(d, &r->domains, list) {
260 dc = !mba_sc ? d->ctrl_val : d->mbps_val;
261 if (d->have_new_ctrl && d->new_ctrl != dc[closid]) {
262 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
263 dc[closid] = d->new_ctrl;
268 * Avoid writing the control msr with control values when
269 * MBA software controller is enabled
271 if (cpumask_empty(cpu_mask) || mba_sc)
274 /* Update CBM on this cpu if it's in cpu_mask. */
275 if (cpumask_test_cpu(cpu, cpu_mask))
276 rdt_ctrl_update(&msr_param);
277 /* Update CBM on other cpus. */
278 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
282 free_cpumask_var(cpu_mask);
287 static int rdtgroup_parse_resource(char *resname, char *tok,
288 struct rdtgroup *rdtgrp)
290 struct rdt_resource *r;
292 for_each_alloc_enabled_rdt_resource(r) {
293 if (!strcmp(resname, r->name) && rdtgrp->closid < r->num_closid)
294 return parse_line(tok, r, rdtgrp);
296 rdt_last_cmd_printf("unknown/unsupported resource name '%s'\n", resname);
300 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
301 char *buf, size_t nbytes, loff_t off)
303 struct rdtgroup *rdtgrp;
304 struct rdt_domain *dom;
305 struct rdt_resource *r;
309 /* Valid input requires a trailing newline */
310 if (nbytes == 0 || buf[nbytes - 1] != '\n')
312 buf[nbytes - 1] = '\0';
315 rdtgrp = rdtgroup_kn_lock_live(of->kn);
317 rdtgroup_kn_unlock(of->kn);
321 rdt_last_cmd_clear();
324 * No changes to pseudo-locked region allowed. It has to be removed
325 * and re-created instead.
327 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
329 rdt_last_cmd_puts("resource group is pseudo-locked\n");
333 for_each_alloc_enabled_rdt_resource(r) {
334 list_for_each_entry(dom, &r->domains, list)
335 dom->have_new_ctrl = false;
338 while ((tok = strsep(&buf, "\n")) != NULL) {
339 resname = strim(strsep(&tok, ":"));
341 rdt_last_cmd_puts("Missing ':'\n");
345 if (tok[0] == '\0') {
346 rdt_last_cmd_printf("Missing '%s' value\n", resname);
350 ret = rdtgroup_parse_resource(resname, tok, rdtgrp);
355 for_each_alloc_enabled_rdt_resource(r) {
356 ret = update_domains(r, rdtgrp->closid);
361 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
363 * If pseudo-locking fails we keep the resource group in
364 * mode RDT_MODE_PSEUDO_LOCKSETUP with its class of service
365 * active and updated for just the domain the pseudo-locked
366 * region was requested for.
368 ret = rdtgroup_pseudo_lock_create(rdtgrp);
372 rdtgroup_kn_unlock(of->kn);
374 return ret ?: nbytes;
377 static void show_doms(struct seq_file *s, struct rdt_resource *r, int closid)
379 struct rdt_domain *dom;
383 seq_printf(s, "%*s:", max_name_width, r->name);
384 list_for_each_entry(dom, &r->domains, list) {
388 ctrl_val = (!is_mba_sc(r) ? dom->ctrl_val[closid] :
389 dom->mbps_val[closid]);
390 seq_printf(s, r->format_str, dom->id, max_data_width,
397 int rdtgroup_schemata_show(struct kernfs_open_file *of,
398 struct seq_file *s, void *v)
400 struct rdtgroup *rdtgrp;
401 struct rdt_resource *r;
405 rdtgrp = rdtgroup_kn_lock_live(of->kn);
407 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
408 for_each_alloc_enabled_rdt_resource(r)
409 seq_printf(s, "%s:uninitialized\n", r->name);
410 } else if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
411 if (!rdtgrp->plr->d) {
412 rdt_last_cmd_clear();
413 rdt_last_cmd_puts("Cache domain offline\n");
416 seq_printf(s, "%s:%d=%x\n",
417 rdtgrp->plr->r->name,
422 closid = rdtgrp->closid;
423 for_each_alloc_enabled_rdt_resource(r) {
424 if (closid < r->num_closid)
425 show_doms(s, r, closid);
431 rdtgroup_kn_unlock(of->kn);
435 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
436 struct rdtgroup *rdtgrp, int evtid, int first)
439 * setup the parameters to send to the IPI to read the data.
447 smp_call_function_any(&d->cpu_mask, mon_event_count, rr, 1);
450 int rdtgroup_mondata_show(struct seq_file *m, void *arg)
452 struct kernfs_open_file *of = m->private;
453 u32 resid, evtid, domid;
454 struct rdtgroup *rdtgrp;
455 struct rdt_resource *r;
456 union mon_data_bits md;
457 struct rdt_domain *d;
461 rdtgrp = rdtgroup_kn_lock_live(of->kn);
467 md.priv = of->kn->priv;
472 r = &rdt_resources_all[resid];
473 d = rdt_find_domain(r, domid, NULL);
474 if (IS_ERR_OR_NULL(d)) {
479 mon_event_read(&rr, d, rdtgrp, evtid, false);
481 if (rr.val & RMID_VAL_ERROR)
482 seq_puts(m, "Error\n");
483 else if (rr.val & RMID_VAL_UNAVAIL)
484 seq_puts(m, "Unavailable\n");
486 seq_printf(m, "%llu\n", rr.val * r->mon_scale);
489 rdtgroup_kn_unlock(of->kn);