x86/intel_rdt: Add schemata file
authorTony Luck <tony.luck@intel.com>
Fri, 28 Oct 2016 22:04:47 +0000 (15:04 -0700)
committerThomas Gleixner <tglx@linutronix.de>
Mon, 31 Oct 2016 01:10:16 +0000 (19:10 -0600)
Last of the per resource group files. Also mode 0644. This one shows
the resources available to the group. Syntax depends on whether the
"cdp" mount option was given. With code/data prioritization disabled
it is simply a list of masks for each cache domain. Initial value
allows access to all of the L3 cache on all domains. E.g. on a 2 socket
Broadwell:
        L3:0=fffff;1=fffff
With CDP enabled, separate masks for data and instructions are provided:
        L3DATA:0=fffff;1=fffff
        L3CODE:0=fffff;1=fffff

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Cc: "Ravi V Shankar" <ravi.v.shankar@intel.com>
Cc: "Shaohua Li" <shli@fb.com>
Cc: "Sai Prakhya" <sai.praneeth.prakhya@intel.com>
Cc: "Peter Zijlstra" <peterz@infradead.org>
Cc: "Stephane Eranian" <eranian@google.com>
Cc: "Dave Hansen" <dave.hansen@intel.com>
Cc: "David Carrillo-Cisneros" <davidcc@google.com>
Cc: "Nilay Vaish" <nilayvaish@gmail.com>
Cc: "Vikas Shivappa" <vikas.shivappa@linux.intel.com>
Cc: "Ingo Molnar" <mingo@elte.hu>
Cc: "Borislav Petkov" <bp@suse.de>
Cc: "H. Peter Anvin" <h.peter.anvin@intel.com>
Link: http://lkml.kernel.org/r/1477692289-37412-9-git-send-email-fenghua.yu@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/include/asm/intel_rdt.h
arch/x86/kernel/cpu/Makefile
arch/x86/kernel/cpu/intel_rdt_rdtgroup.c
arch/x86/kernel/cpu/intel_rdt_schemata.c [new file with mode: 0644]

index a0dd3e9..2e5eab0 100644 (file)
@@ -73,6 +73,7 @@ struct rftype {
  * @num_domains:               Number of domains active
  * @msr_base:                  Base MSR address for CBMs
  * @tmp_cbms:                  Scratch space when updating schemata
+ * @num_tmp_cbms:              Number of CBMs in tmp_cbms
  * @cache_level:               Which cache level defines scope of this domain
  * @cbm_idx_multi:             Multiplier of CBM index
  * @cbm_idx_offset:            Offset of CBM index. CBM index is computed by:
@@ -90,6 +91,7 @@ struct rdt_resource {
        int                     num_domains;
        int                     msr_base;
        u32                     *tmp_cbms;
+       int                     num_tmp_cbms;
        int                     cache_level;
        int                     cbm_idx_multi;
        int                     cbm_idx_offset;
@@ -170,4 +172,8 @@ DECLARE_PER_CPU_READ_MOSTLY(int, cpu_closid);
 void rdt_cbm_update(void *arg);
 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
 void rdtgroup_kn_unlock(struct kernfs_node *kn);
+ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
+                               char *buf, size_t nbytes, loff_t off);
+int rdtgroup_schemata_show(struct kernfs_open_file *of,
+                          struct seq_file *s, void *v);
 #endif /* _ASM_X86_INTEL_RDT_H */
index b4334e8..c9f8c81 100644 (file)
@@ -34,7 +34,7 @@ obj-$(CONFIG_CPU_SUP_CENTAUR)         += centaur.o
 obj-$(CONFIG_CPU_SUP_TRANSMETA_32)     += transmeta.o
 obj-$(CONFIG_CPU_SUP_UMC_32)           += umc.o
 
-obj-$(CONFIG_INTEL_RDT_A)      += intel_rdt.o intel_rdt_rdtgroup.o
+obj-$(CONFIG_INTEL_RDT_A)      += intel_rdt.o intel_rdt_rdtgroup.o intel_rdt_schemata.o
 
 obj-$(CONFIG_X86_MCE)                  += mcheck/
 obj-$(CONFIG_MTRR)                     += mtrr/
index 5cc0865..5c4bab9 100644 (file)
@@ -440,6 +440,13 @@ static struct rftype rdtgroup_base_files[] = {
                .write          = rdtgroup_tasks_write,
                .seq_show       = rdtgroup_tasks_show,
        },
+       {
+               .name           = "schemata",
+               .mode           = 0644,
+               .kf_ops         = &rdtgroup_kf_single_ops,
+               .write          = rdtgroup_schemata_write,
+               .seq_show       = rdtgroup_schemata_show,
+       },
 };
 
 static int rdt_num_closids_show(struct kernfs_open_file *of,
diff --git a/arch/x86/kernel/cpu/intel_rdt_schemata.c b/arch/x86/kernel/cpu/intel_rdt_schemata.c
new file mode 100644 (file)
index 0000000..f369cb8
--- /dev/null
@@ -0,0 +1,245 @@
+/*
+ * Resource Director Technology(RDT)
+ * - Cache Allocation code.
+ *
+ * Copyright (C) 2016 Intel Corporation
+ *
+ * Authors:
+ *    Fenghua Yu <fenghua.yu@intel.com>
+ *    Tony Luck <tony.luck@intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * More information about RDT be found in the Intel (R) x86 Architecture
+ * Software Developer Manual June 2016, volume 3, section 17.17.
+ */
+
+#define pr_fmt(fmt)    KBUILD_MODNAME ": " fmt
+
+#include <linux/kernfs.h>
+#include <linux/seq_file.h>
+#include <linux/slab.h>
+#include <asm/intel_rdt.h>
+
+/*
+ * Check whether a cache bit mask is valid. The SDM says:
+ *     Please note that all (and only) contiguous '1' combinations
+ *     are allowed (e.g. FFFFH, 0FF0H, 003CH, etc.).
+ * Additionally Haswell requires at least two bits set.
+ */
+static bool cbm_validate(unsigned long var, struct rdt_resource *r)
+{
+       unsigned long first_bit, zero_bit;
+
+       if (var == 0 || var > r->max_cbm)
+               return false;
+
+       first_bit = find_first_bit(&var, r->cbm_len);
+       zero_bit = find_next_zero_bit(&var, r->cbm_len, first_bit);
+
+       if (find_next_bit(&var, r->cbm_len, zero_bit) < r->cbm_len)
+               return false;
+
+       if ((zero_bit - first_bit) < r->min_cbm_bits)
+               return false;
+       return true;
+}
+
+/*
+ * Read one cache bit mask (hex). Check that it is valid for the current
+ * resource type.
+ */
+static int parse_cbm(char *buf, struct rdt_resource *r)
+{
+       unsigned long data;
+       int ret;
+
+       ret = kstrtoul(buf, 16, &data);
+       if (ret)
+               return ret;
+       if (!cbm_validate(data, r))
+               return -EINVAL;
+       r->tmp_cbms[r->num_tmp_cbms++] = data;
+
+       return 0;
+}
+
+/*
+ * For each domain in this resource we expect to find a series of:
+ *     id=mask
+ * separated by ";". The "id" is in decimal, and must appear in the
+ * right order.
+ */
+static int parse_line(char *line, struct rdt_resource *r)
+{
+       char *dom = NULL, *id;
+       struct rdt_domain *d;
+       unsigned long dom_id;
+
+       list_for_each_entry(d, &r->domains, list) {
+               dom = strsep(&line, ";");
+               if (!dom)
+                       return -EINVAL;
+               id = strsep(&dom, "=");
+               if (kstrtoul(id, 10, &dom_id) || dom_id != d->id)
+                       return -EINVAL;
+               if (parse_cbm(dom, r))
+                       return -EINVAL;
+       }
+
+       /* Any garbage at the end of the line? */
+       if (line && line[0])
+               return -EINVAL;
+       return 0;
+}
+
+static int update_domains(struct rdt_resource *r, int closid)
+{
+       struct msr_param msr_param;
+       cpumask_var_t cpu_mask;
+       struct rdt_domain *d;
+       int cpu, idx = 0;
+
+       if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
+               return -ENOMEM;
+
+       msr_param.low = closid;
+       msr_param.high = msr_param.low + 1;
+       msr_param.res = r;
+
+       list_for_each_entry(d, &r->domains, list) {
+               cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
+               d->cbm[msr_param.low] = r->tmp_cbms[idx++];
+       }
+       cpu = get_cpu();
+       /* Update CBM on this cpu if it's in cpu_mask. */
+       if (cpumask_test_cpu(cpu, cpu_mask))
+               rdt_cbm_update(&msr_param);
+       /* Update CBM on other cpus. */
+       smp_call_function_many(cpu_mask, rdt_cbm_update, &msr_param, 1);
+       put_cpu();
+
+       free_cpumask_var(cpu_mask);
+
+       return 0;
+}
+
+ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
+                               char *buf, size_t nbytes, loff_t off)
+{
+       struct rdtgroup *rdtgrp;
+       struct rdt_resource *r;
+       char *tok, *resname;
+       int closid, ret = 0;
+       u32 *l3_cbms = NULL;
+
+       /* Valid input requires a trailing newline */
+       if (nbytes == 0 || buf[nbytes - 1] != '\n')
+               return -EINVAL;
+       buf[nbytes - 1] = '\0';
+
+       rdtgrp = rdtgroup_kn_lock_live(of->kn);
+       if (!rdtgrp) {
+               rdtgroup_kn_unlock(of->kn);
+               return -ENOENT;
+       }
+
+       closid = rdtgrp->closid;
+
+       /* get scratch space to save all the masks while we validate input */
+       for_each_enabled_rdt_resource(r) {
+               r->tmp_cbms = kcalloc(r->num_domains, sizeof(*l3_cbms),
+                                     GFP_KERNEL);
+               if (!r->tmp_cbms) {
+                       ret = -ENOMEM;
+                       goto out;
+               }
+               r->num_tmp_cbms = 0;
+       }
+
+       while ((tok = strsep(&buf, "\n")) != NULL) {
+               resname = strsep(&tok, ":");
+               if (!tok) {
+                       ret = -EINVAL;
+                       goto out;
+               }
+               for_each_enabled_rdt_resource(r) {
+                       if (!strcmp(resname, r->name) &&
+                           closid < r->num_closid) {
+                               ret = parse_line(tok, r);
+                               if (ret)
+                                       goto out;
+                               break;
+                       }
+               }
+               if (!r->name) {
+                       ret = -EINVAL;
+                       goto out;
+               }
+       }
+
+       /* Did the parser find all the masks we need? */
+       for_each_enabled_rdt_resource(r) {
+               if (r->num_tmp_cbms != r->num_domains) {
+                       ret = -EINVAL;
+                       goto out;
+               }
+       }
+
+       for_each_enabled_rdt_resource(r) {
+               ret = update_domains(r, closid);
+               if (ret)
+                       goto out;
+       }
+
+out:
+       rdtgroup_kn_unlock(of->kn);
+       for_each_enabled_rdt_resource(r) {
+               kfree(r->tmp_cbms);
+               r->tmp_cbms = NULL;
+       }
+       return ret ?: nbytes;
+}
+
+static void show_doms(struct seq_file *s, struct rdt_resource *r, int closid)
+{
+       struct rdt_domain *dom;
+       bool sep = false;
+
+       seq_printf(s, "%s:", r->name);
+       list_for_each_entry(dom, &r->domains, list) {
+               if (sep)
+                       seq_puts(s, ";");
+               seq_printf(s, "%d=%x", dom->id, dom->cbm[closid]);
+               sep = true;
+       }
+       seq_puts(s, "\n");
+}
+
+int rdtgroup_schemata_show(struct kernfs_open_file *of,
+                          struct seq_file *s, void *v)
+{
+       struct rdtgroup *rdtgrp;
+       struct rdt_resource *r;
+       int closid, ret = 0;
+
+       rdtgrp = rdtgroup_kn_lock_live(of->kn);
+       if (rdtgrp) {
+               closid = rdtgrp->closid;
+               for_each_enabled_rdt_resource(r) {
+                       if (closid < r->num_closid)
+                               show_doms(s, r, closid);
+               }
+       } else {
+               ret = -ENOENT;
+       }
+       rdtgroup_kn_unlock(of->kn);
+       return ret;
+}