Add resctrl, a tool for setting up reservations

resctrl is a new tool intended to configure reservations in
reservation-aware plugins. For now, it only supports the partitioned
reservations (P-RES) plugin.

(While at it, also update .gitignore)

3 files changed, 246 insertions, 1 deletions

diff --git a/.gitignore b/.gitignore
index a52500b..f79d4fa 100644
--- a/.gitignore
+++ b/.gitignore
@@ -3,10 +3,13 @@
 *.a
 
 
+# editor backup files
+*~
 
 # generated files
 tests/test_catalog.inc
 *.d
+*.d.*
 docs/
 
 # executables
@@ -23,6 +26,8 @@ showst
 rtspin
 cycles
 measure_syscall
+uncache
+resctrl
 
 # build system files
 .config
@@ -32,3 +37,5 @@ test_catalog.inc
 # files copied in from the kernel tree
 include/litmus
 arch/*/include/asm
+arch/x86/include/generated
+arch/x86/include/uapi
diff --git a/Makefile b/Makefile
index 4742fd2..34c428f 100644
--- a/Makefile
+++ b/Makefile
@@ -73,7 +73,7 @@ AR  := ${CROSS_COMPILE}${AR}
 
 all     = lib ${rt-apps}
 rt-apps = cycles base_task rt_launch rtspin release_ts measure_syscall \
-          base_mt_task uncache runtests
+          base_mt_task uncache runtests resctrl
 
 .PHONY: all lib clean dump-config TAGS tags cscope help doc
 
@@ -234,6 +234,9 @@ obj-release_ts = release_ts.o
 obj-measure_syscall = null_call.o
 lib-measure_syscall = -lm
 
+obj-resctrl = resctrl.o
+
+
 # ##############################################################################
 # Build everything that depends on liblitmus.
 
diff --git a/bin/resctrl.c b/bin/resctrl.c
new file mode 100644
index 0000000..917f49a
--- /dev/null
+++ b/bin/resctrl.c
@@ -0,0 +1,235 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <limits.h>
+#include <signal.h>
+
+#include <sched.h>
+
+#include "litmus.h"
+#include "common.h"
+
+const char *usage_msg =
+        "Usage: resctrl OPTIONS [INTERVAL-START,INTERVAL-END]*\n"
+        "    -n ID             create new reservation with id ID\n"
+        "    -a PID            attach already-running process PID to reservation\n"
+        "    -r ID             specify which reservation to attach to (not needed with -n)\n"
+        "    -t TYPE           type of reservation (polling-periodic, polling-sporadic, table-driven)\n"
+        "    -c CPU            physical partition or cluster to assign to\n"
+        "    -b BUDGET         polling reservation budget (in ms, default: 10ms)\n"
+        "    -p PERIOD         polling reservation period (in ms, default: 100ms)\n"
+        "    -d DEADLINE       relative deadline, implicit by default (in ms)\n"
+        "    -o OFFSET         offset (also known as phase), zero by default (in ms)\n"
+        "    -q PRIORITY       priority to use (EDF by default, max=0)\n"
+        "    -m MAJOR-CYCLE    major cycle length (in ms, for table-driven reservations) \n"
+        "\n";
+
+void usage(char *error) {
+        fprintf(stderr, "%s\n%s", error, usage_msg);
+        exit(1);
+}
+
+
+static void attach_task(int attach_pid, struct reservation_config *config)
+{
+        int ret;
+        struct rt_task param;
+        struct sched_param linux_param;
+
+        ret = be_migrate_thread_to_cpu(attach_pid, config->cpu);
+        if (ret < 0) {
+                fprintf(stderr, "failed to migrate task %d to CPU %d\n",
+                        attach_pid, config->cpu);
+                exit(4);
+        }
+
+        init_rt_task_param(&param);
+        /* dummy values */
+        param.exec_cost = ms2ns(100);
+        param.period    = ms2ns(100);
+        /* specify reservation as "virtual" CPU */
+        param.cpu       = config->id;
+
+        ret = set_rt_task_param(attach_pid, &param);
+        if (ret < 0) {
+                fprintf(stderr, "failed to set RT task parameters for task %d (%m)\n",
+                        attach_pid);
+                exit(2);
+        }
+
+        linux_param.sched_priority = 0;
+        ret = sched_setscheduler(attach_pid, SCHED_LITMUS, &linux_param);
+        if (ret < 0) {
+                fprintf(stderr, "failed to transition task %d to LITMUS^RT class (%m)\n",
+                        attach_pid);
+                exit(3);
+        }
+}
+
+static struct lt_interval* parse_td_intervals(int argc, char** argv,
+        unsigned int *num_intervals, lt_t major_cycle)
+{
+        int i, matched;
+        struct lt_interval *slots = malloc(sizeof(slots[0]) * argc);
+        double start, end;
+
+        *num_intervals = 0;
+        for (i = 0; i < argc; i++) {
+                matched = sscanf(argv[i], "[%lf,%lf]", &start, &end);
+                if (matched != 2) {
+                        fprintf(stderr, "could not parse '%s' as interval\n", argv[i]);
+                        exit(5);
+                }
+                if (start < 0) {
+                        fprintf(stderr, "interval %s: must not start before zero\n", argv[i]);
+                        exit(5);
+                }
+                if (end <= start) {
+                        fprintf(stderr, "interval %s: end before start\n", argv[i]);
+                        exit(5);
+                }
+
+                slots[i].start = ms2ns(start);
+                slots[i].end   = ms2ns(end);
+
+                if (i > 0 && slots[i - 1].end >= slots[i].start) {
+                        fprintf(stderr, "interval %s: overlaps with previous interval\n", argv[i]);
+                        exit(5);
+                }
+
+                if (slots[i].end >= major_cycle) {
+                        fprintf(stderr, "interval %s: exceeds major cycle length\n", argv[i]);
+                        exit(5);
+                }
+
+                (*num_intervals)++;
+        }
+
+        return slots;
+}
+
+#define OPTSTR "n:a:r:t:c:b:p:d:o:q:m:h"
+
+int main(int argc, char** argv)
+{
+        int ret, opt;
+        double budget_ms, period_ms, offset_ms, deadline_ms, major_cycle_ms;
+        int create_new = 0;
+        int attach_pid = 0;
+        int res_type = SPORADIC_POLLING;
+
+        struct reservation_config config;
+
+        /* Reasonable defaults */
+        offset_ms      =    0;
+        deadline_ms    =    0;
+        budget_ms      =   10;
+        period_ms      =  100;
+        major_cycle_ms = 1000;
+
+        config.id = 0;
+        config.priority = LITMUS_NO_PRIORITY; /* use EDF by default */
+        config.cpu = 0;
+
+        while ((opt = getopt(argc, argv, OPTSTR)) != -1) {
+                switch (opt) {
+                case 'b':
+                        budget_ms = atof(optarg);
+                        break;
+                case 'p':
+                        period_ms = atof(optarg);
+                        break;
+                case 'd':
+                        deadline_ms = atof(optarg);
+                        if (deadline_ms <= 0) {
+                                usage("The relative deadline must be a positive"
+                                        " number.");
+                        }
+                        break;
+                case 'o':
+                        offset_ms = atof(optarg);
+                        break;
+                case 'm':
+                        major_cycle_ms = atof(optarg);
+                        break;
+
+                case 'q':
+                        config.priority = atoi(optarg);
+                        break;
+                case 'c':
+                        config.cpu = atoi(optarg);
+                        break;
+
+                case 'n':
+                        create_new = 1;
+                        config.id  = atoi(optarg);
+                        break;
+                case 'a':
+                        attach_pid = atoi(optarg);
+                        if (!attach_pid)
+                                usage("-a: invalid PID");
+                        break;
+
+                case 'r':
+                        config.id  = atoi(optarg);
+                        break;
+
+                case 't':
+                        if (strcmp(optarg, "polling-periodic") == 0) {
+                                res_type = PERIODIC_POLLING;
+                        } else if (strcmp(optarg, "polling-sporadic") == 0) {
+                                res_type = SPORADIC_POLLING;
+                        }  else if (strcmp(optarg, "table-driven") == 0) {
+                                res_type = TABLE_DRIVEN;
+                        } else {
+                                usage("Unknown reservation type.");
+                        }
+                        break;
+
+                case 'h':
+                        usage("");
+                        break;
+                case ':':
+                        usage("Argument missing.");
+                        break;
+                case '?':
+                default:
+                        usage("Bad argument.");
+                        break;
+                }
+        }
+
+        if (res_type != TABLE_DRIVEN) {
+                config.polling_params.budget = ms2ns(budget_ms);
+                config.polling_params.period = ms2ns(period_ms);
+                config.polling_params.offset = ms2ns(offset_ms);
+                config.polling_params.relative_deadline = ms2ns(deadline_ms);
+                if (config.polling_params.budget > config.polling_params.period) {
+                        usage("The budget must not exceed the period.");
+                }
+        } else {
+                config.table_driven_params.major_cycle_length = ms2ns(major_cycle_ms);
+                argc -= optind;
+                argv += optind;
+                config.table_driven_params.intervals = parse_td_intervals(
+                        argc, argv, &config.table_driven_params.num_intervals,
+                        config.table_driven_params.major_cycle_length);
+                if (!config.table_driven_params.num_intervals)
+                        usage("Table-driven reservations require at least one interval to be specified.");
+        }
+
+        if (create_new) {
+                ret = reservation_create(res_type, &config);
+                if (ret < 0) {
+                        fprintf(stderr, "failed to create reservation %u (%m)\n",
+                                config.id);
+                        exit(1);
+                }
+        }
+
+        if (attach_pid)
+                attach_task(attach_pid, &config);
+
+        return 0;
+}