Turbostat: check return value of vsnprintf
[collectd.git] / src / turbostat.c
index 482ec5e..6fa172d 100644 (file)
@@ -196,21 +196,30 @@ static _Bool allocated = 0;
 static _Bool initialized = 0;
 
 #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \
-       (thread_base + (pkg_no) * topo.num_cores_per_pkg * \
-               topo.num_threads_per_core + \
-               (core_no) * topo.num_threads_per_core + (thread_no))
+       (thread_base + \
+               (pkg_no) * topology.num_cores * topology.num_threads + \
+               (core_no) * topology.num_threads + \
+               (thread_no))
 #define GET_CORE(core_base, core_no, pkg_no) \
-       (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no))
+       (core_base + \
+               (pkg_no) * topology.num_cores + \
+               (core_no))
 #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no)
 
-struct topo_params {
+struct cpu_topology {
+       int package_id;
+       int core_id;
+       _Bool first_core_in_package;
+       _Bool first_thread_in_core;
+};
+
+struct topology {
+       int max_cpu_id;
        int num_packages;
-       int num_cpus;
        int num_cores;
-       int max_cpu_num;
-       int num_cores_per_pkg;
-       int num_threads_per_core;
-} topo;
+       int num_threads;
+       struct cpu_topology *cpus;
+} topology;
 
 struct timeval tv_even, tv_odd, tv_delta;
 
@@ -253,6 +262,7 @@ enum return_values {
        ERR_CPU_ALLOC,
        ERR_NOT_ROOT,
        UNSUPPORTED_CPU,
+       ERR_PATH_TOO_LONG,
 };
 
 
@@ -281,7 +291,7 @@ open_msr(int cpu, _Bool multiple_read)
                CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
                CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set);
                if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) {
-                       ERROR("Could not migrate to CPU %d", cpu);
+                       ERROR("Turbostat plugin: Could not migrate to CPU %d", cpu);
                        return -ERR_CPU_MIGRATE;
                }
        }
@@ -304,7 +314,8 @@ read_msr(int fd, off_t offset, unsigned long long *msr)
        retval = pread(fd, msr, sizeof *msr, offset);
 
        if (retval != sizeof *msr) {
-               ERROR("MSR offset 0x%llx read failed", (unsigned long long)offset);
+               ERROR("Turbostat plugin: MSR offset 0x%llx read failed",
+                     (unsigned long long)offset);
                return -1;
        }
        return 0;
@@ -393,20 +404,20 @@ do {                                              \
                goto out;
        }
 
-        if (do_pkg_cstate & (1 << 2))
-                READ_MSR(MSR_PKG_C2_RESIDENCY, &p->pc2);
-        if (do_pkg_cstate & (1 << 3))
-                READ_MSR(MSR_PKG_C3_RESIDENCY, &p->pc3);
-        if (do_pkg_cstate & (1 << 6))
-                READ_MSR(MSR_PKG_C6_RESIDENCY, &p->pc6);
-        if (do_pkg_cstate & (1 << 7))
-                READ_MSR(MSR_PKG_C7_RESIDENCY, &p->pc7);
-        if (do_pkg_cstate & (1 << 8))
-                READ_MSR(MSR_PKG_C8_RESIDENCY, &p->pc8);
-        if (do_pkg_cstate & (1 << 9))
-                READ_MSR(MSR_PKG_C9_RESIDENCY, &p->pc9);
-        if (do_pkg_cstate & (1 << 10))
-                READ_MSR(MSR_PKG_C10_RESIDENCY, &p->pc10);
+       if (do_pkg_cstate & (1 << 2))
+               READ_MSR(MSR_PKG_C2_RESIDENCY, &p->pc2);
+       if (do_pkg_cstate & (1 << 3))
+               READ_MSR(MSR_PKG_C3_RESIDENCY, &p->pc3);
+       if (do_pkg_cstate & (1 << 6))
+               READ_MSR(MSR_PKG_C6_RESIDENCY, &p->pc6);
+       if (do_pkg_cstate & (1 << 7))
+               READ_MSR(MSR_PKG_C7_RESIDENCY, &p->pc7);
+       if (do_pkg_cstate & (1 << 8))
+               READ_MSR(MSR_PKG_C8_RESIDENCY, &p->pc8);
+       if (do_pkg_cstate & (1 << 9))
+               READ_MSR(MSR_PKG_C9_RESIDENCY, &p->pc9);
+       if (do_pkg_cstate & (1 << 10))
+               READ_MSR(MSR_PKG_C10_RESIDENCY, &p->pc10);
 
        if (do_rapl & RAPL_PKG) {
                READ_MSR(MSR_PKG_ENERGY_STATUS, &msr);
@@ -506,9 +517,10 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str
 
        /* check for TSC < 1 Mcycles over interval */
        if (delta->tsc < (1000 * 1000)) {
-               WARNING("Insanely slow TSC rate, TSC stops in idle? ");
-               WARNING("You can disable all c-states by booting with \"idle=poll\" ");
-               WARNING("or just the deep ones with \"processor.max_cstate=1\"");
+               WARNING("Turbostat plugin: Insanely slow TSC rate, TSC stops "
+                       "in idle? You can disable all c-states by booting with"
+                       " 'idle=poll' or just the deep ones with"
+                       " 'processor.max_cstate=1'");
                return -1;
        }
 
@@ -519,9 +531,10 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str
                delta->mperf = new->mperf - old->mperf;
        } else {
                if (!aperf_mperf_unstable) {
-                       WARNING(" APERF or MPERF went backwards * ");
-                       WARNING("* Frequency results do not cover entire interval *");
-                       WARNING("* fix this by running Linux-2.6.30 or later *");
+                       WARNING("Turbostat plugin: APERF or MPERF went "
+                               "backwards. Frequency results do not cover "
+                               "the entire interval. Fix this by running "
+                               "Linux-2.6.30 or later.");
 
                        aperf_mperf_unstable = 1;
                }
@@ -541,7 +554,7 @@ delta_thread(struct thread_data *delta, const struct thread_data *new, const str
        }
 
        if (delta->mperf == 0) {
-               WARNING("cpu%d MPERF 0!", old->cpu_id);
+               WARNING("Turbostat plugin: cpu%d MPERF 0!", old->cpu_id);
                delta->mperf = 1;       /* divide by 0 protection */
        }
 
@@ -661,14 +674,13 @@ done:
        return 0;
 }
 
+
 /**********************************
  * Looping function over all CPUs *
  **********************************/
 
 /*
  * Check if a given cpu id is in our compiled list of existing CPUs
- *
- * CHECKME: Why do we need this?
  */
 static int
 cpu_is_not_present(int cpu)
@@ -679,7 +691,7 @@ cpu_is_not_present(int cpu)
 /*
  * Loop on all CPUs in topological order
  *
- * Skip 'non-present' cpus (CHECKME: Why do we need this?)
+ * Skip non-present cpus
  * Return the error code at the first error or 0
  */
 static int __attribute__((warn_unused_result))
@@ -688,10 +700,9 @@ for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_dat
 {
        int retval, pkg_no, core_no, thread_no;
 
-       for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
-               for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
-                       for (thread_no = 0; thread_no <
-                               topo.num_threads_per_core; ++thread_no) {
+       for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
+               for (core_no = 0; core_no < topology.num_cores; ++core_no) {
+                       for (thread_no = 0; thread_no < topology.num_threads; ++thread_no) {
                                struct thread_data *t;
                                struct core_data *c;
                                struct pkg_data *p;
@@ -716,6 +727,9 @@ for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_dat
 /*
  * Dedicated loop: Extract every data evolution for all CPU
  *
+ * Skip non-present cpus
+ * Return the error code at the first error or 0
+ *
  * Core data is shared for all threads in one core: extracted only for the first thread
  * Package data is shared for all core in one package: extracted only for the first thread of the first core
  */
@@ -725,10 +739,9 @@ for_all_cpus_delta(const struct thread_data *thread_new_base, const struct core_
 {
        int retval, pkg_no, core_no, thread_no;
 
-       for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) {
-               for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) {
-                       for (thread_no = 0; thread_no <
-                               topo.num_threads_per_core; ++thread_no) {
+       for (pkg_no = 0; pkg_no < topology.num_packages; ++pkg_no) {
+               for (core_no = 0; core_no < topology.num_cores; ++core_no) {
+                       for (thread_no = 0; thread_no < topology.num_threads; ++thread_no) {
                                struct thread_data *t_delta;
                                const struct thread_data *t_old, *t_new;
                                struct core_data *c_delta;
@@ -778,284 +791,9 @@ for_all_cpus_delta(const struct thread_data *thread_new_base, const struct core_
 }
 
 
-static void
-free_all_buffers(void)
-{
-       allocated = 0;
-       initialized = 0;
-
-       CPU_FREE(cpu_present_set);
-       cpu_present_set = NULL;
-       cpu_present_set = 0;
-
-       CPU_FREE(cpu_affinity_set);
-       cpu_affinity_set = NULL;
-       cpu_affinity_setsize = 0;
-
-       CPU_FREE(cpu_saved_affinity_set);
-       cpu_saved_affinity_set = NULL;
-       cpu_saved_affinity_setsize = 0;
-
-       free(thread_even);
-       free(core_even);
-       free(package_even);
-
-       thread_even = NULL;
-       core_even = NULL;
-       package_even = NULL;
-
-       free(thread_odd);
-       free(core_odd);
-       free(package_odd);
-
-       thread_odd = NULL;
-       core_odd = NULL;
-       package_odd = NULL;
-
-       free(thread_delta);
-       free(core_delta);
-       free(package_delta);
-
-       thread_delta = NULL;
-       core_delta = NULL;
-       package_delta = NULL;
-}
-
-/*
- * Parse a file containing a single int.
- */
-static int __attribute__ ((format(printf,1,2)))
-parse_int_file(const char *fmt, ...)
-{
-       va_list args;
-       char path[PATH_MAX];
-       FILE *filep;
-       int value;
-
-       va_start(args, fmt);
-       vsnprintf(path, sizeof(path), fmt, args);
-       va_end(args);
-       filep = fopen(path, "r");
-       if (!filep) {
-               ERROR("%s: open failed", path);
-               return -ERR_CANT_OPEN_FILE;
-       }
-       if (fscanf(filep, "%d", &value) != 1) {
-               ERROR("%s: failed to parse number from file", path);
-               return -ERR_CANT_READ_NUMBER;
-       }
-       fclose(filep);
-       return value;
-}
-
-/*
- * cpu_is_first_sibling_in_core(cpu)
- * return 1 if given CPU is 1st HT sibling in the core
- */
-static int
-cpu_is_first_sibling_in_core(int cpu)
-{
-       return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
-}
-
-/*
- * cpu_is_first_core_in_package(cpu)
- * return 1 if given CPU is 1st core in package
- */
-static int
-cpu_is_first_core_in_package(int cpu)
-{
-       return cpu == parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu);
-}
-
-static int
-get_physical_package_id(int cpu)
-{
-       return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
-}
-
-static int
-get_core_id(int cpu)
-{
-       return parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
-}
-
-static int
-get_num_ht_siblings(int cpu)
-{
-       char path[80];
-       FILE *filep;
-       int sib1, sib2;
-       int matches;
-       char character;
-
-       ssnprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
-       filep = fopen(path, "r");
-        if (!filep) {
-                ERROR("%s: open failed", path);
-                return -ERR_CANT_OPEN_FILE;
-        }
-       /*
-        * file format:
-        * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
-        * otherwinse 1 sibling (self).
-        */
-       matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
-
-       fclose(filep);
-
-       if (matches == 3)
-               return 2;
-       else
-               return 1;
-}
-
-/*
- * run func(cpu) on every cpu in /proc/stat
- * return max_cpu number
- */
-static int __attribute__((warn_unused_result))
-for_all_proc_cpus(int (func)(int))
-{
-       FILE *fp;
-       int cpu_num;
-       int retval;
-
-       fp = fopen("/proc/stat", "r");
-        if (!fp) {
-                ERROR("Failed to open /proc/stat");
-                return -ERR_CANT_OPEN_FILE;
-        }
-
-       retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
-       if (retval != 0) {
-               ERROR("Failed to parse /proc/stat");
-               fclose(fp);
-               return -ERR_CANT_READ_PROC_STAT;
-       }
-
-       while (1) {
-               retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
-               if (retval != 1)
-                       break;
-
-               retval = func(cpu_num);
-               if (retval) {
-                       fclose(fp);
-                       return(retval);
-               }
-       }
-       fclose(fp);
-       return 0;
-}
-
-/*
- * count_cpus()
- * remember the last one seen, it will be the max
- */
-static int
-count_cpus(int cpu)
-{
-       if (topo.max_cpu_num < cpu)
-               topo.max_cpu_num = cpu;
-
-       topo.num_cpus += 1;
-       return 0;
-}
-static int
-mark_cpu_present(int cpu)
-{
-       CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
-       return 0;
-}
-
-
-static int setup_all_buffers(void);
-
-static int
-turbostat_read(user_data_t * not_used)
-{
-       int ret;
-
-       if (!allocated) {
-               if ((ret = setup_all_buffers()) < 0)
-                       return ret;
-       }
-
-       if (for_all_proc_cpus(cpu_is_not_present)) {
-               free_all_buffers();
-               if ((ret = setup_all_buffers()) < 0)
-                       return ret;
-               if (for_all_proc_cpus(cpu_is_not_present))
-                       return -ERR_CPU_NOT_PRESENT;
-       }
-
-       /* Saving the scheduling affinity, as it will be modified by get_counters */
-       if (sched_getaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set) != 0)
-               return -ERR_CPU_SAVE_SCHED_AFFINITY;
-
-       if (!initialized) {
-               if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
-                       goto out;
-               gettimeofday(&tv_even, (struct timezone *)NULL);
-               is_even = 1;
-               initialized = 1;
-               ret = 0;
-               goto out;
-       }
-
-       if (is_even) {
-               if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
-                       goto out;
-               gettimeofday(&tv_odd, (struct timezone *)NULL);
-               is_even = 0;
-               timersub(&tv_odd, &tv_even, &tv_delta);
-               if ((ret = for_all_cpus_delta(ODD_COUNTERS, EVEN_COUNTERS)) < 0)
-                       goto out;
-               if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
-                       goto out;
-       } else {
-               if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
-                       goto out;
-               gettimeofday(&tv_even, (struct timezone *)NULL);
-               is_even = 1;
-               timersub(&tv_even, &tv_odd, &tv_delta);
-               if ((ret = for_all_cpus_delta(EVEN_COUNTERS, ODD_COUNTERS)) < 0)
-                       goto out;
-               if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
-                       goto out;
-       }
-       ret = 0;
-out:
-       /*
-        * Let's restore the affinity
-        * This might fail if the number of CPU changed, but we can't do anything in that case..
-        */
-       (void)sched_setaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set);
-       return ret;
-}
-
-static int __attribute__((warn_unused_result))
-check_dev_msr()
-{
-       struct stat sb;
-
-       if (stat("/dev/cpu/0/msr", &sb)) {
-               ERROR("no /dev/cpu/0/msr, try \"# modprobe msr\"");
-               return -ERR_NO_MSR;
-       }
-       return 0;
-}
-
-static int __attribute__((warn_unused_result))
-check_super_user()
-{
-       if (getuid() != 0) {
-               ERROR("must be root");
-               return -ERR_NOT_ROOT;
-       }
-       return 0;
-}
+/***************
+ * CPU Probing *
+ ***************/
 
 /*
  * MSR_IA32_TEMPERATURE_TARGET indicates the temperature where
@@ -1102,7 +840,8 @@ set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_da
 
 guess:
        p->tcc_activation_temp = TJMAX_DEFAULT;
-       WARNING("cpu%d: Guessing tjMax %d C, Please use TCCActivationTemp to specify",
+       WARNING("Turbostat plugin: cpu%d: Guessing tjMax %d C,"
+               " Please use TCCActivationTemp to specify it.",
                t->cpu_id, p->tcc_activation_temp);
 
        return 0;
@@ -1121,25 +860,25 @@ probe_cpu()
         * - EAX: Maximum Input Value for Basic CPUID Information
         * - EBX: "Genu" (0x756e6547)
         * - EDX: "ineI" (0x49656e69)
-         * - ECX: "ntel" (0x6c65746e)
-         */
+        * - ECX: "ntel" (0x6c65746e)
+        */
        max_level = ebx = ecx = edx = 0;
        __get_cpuid(0, &max_level, &ebx, &ecx, &edx);
        if (ebx != 0x756e6547 && edx != 0x49656e69 && ecx != 0x6c65746e) {
-               ERROR("Unsupported CPU");
+               ERROR("Turbostat plugin: Unsupported CPU (not Intel)");
                return -UNSUPPORTED_CPU;
        }
 
        /* CPUID(1):
         * - EAX: Version Information: Type, Family, Model, and Stepping ID
-         *  + 4-7:   Model ID
-         *  + 8-11:  Family ID
+        *  + 4-7:   Model ID
+        *  + 8-11:  Family ID
         *  + 12-13: Processor type
         *  + 16-19: Extended Model ID
         *  + 20-27: Extended Family ID
         * - EDX: Feature Information:
         *  + 5: Support for MSR read/write operations
-         */
+        */
        fms = ebx = ecx = edx = 0;
        __get_cpuid(1, &fms, &ebx, &ecx, &edx);
        family = (fms >> 8) & 0xf;
@@ -1149,7 +888,7 @@ probe_cpu()
        if (family == 6 || family == 0xf)
                model += ((fms >> 16) & 0xf) << 4;
        if (!(edx & (1 << 5))) {
-               ERROR("CPUID: no MSR");
+               ERROR("Turbostat plugin: Unsupported CPU (no MSR support)");
                return -ERR_NO_MSR;
        }
 
@@ -1164,7 +903,8 @@ probe_cpu()
        max_level = ebx = ecx = edx = 0;
        __get_cpuid(0x80000000, &max_level, &ebx, &ecx, &edx);
        if (max_level < 0x80000007) {
-               ERROR("CPUID: no invariant TSC (max_level 0x%x)", max_level);
+               ERROR("Turbostat plugin: Unsupported CPU (no invariant TSC, "
+                     " Maximum Extended Function: 0x%x)", max_level);
                return -ERR_NO_INVARIANT_TSC;
        }
 
@@ -1178,7 +918,7 @@ probe_cpu()
        eax = ebx = ecx = edx = 0;
        __get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
        if (!(edx & (1 << 8))) {
-               ERROR("No invariant TSC");
+               ERROR("Turbostat plugin: Unsupported CPU (No invariant TSC)");
                return -ERR_NO_INVARIANT_TSC;
        }
 
@@ -1199,7 +939,7 @@ probe_cpu()
        do_dts = eax & (1 << 0);
        do_ptm = eax & (1 << 6);
        if (!(ecx & (1 << 0))) {
-               ERROR("No APERF");
+               ERROR("Turbostat plugin: Unsupported CPU (No APERF)");
                return -ERR_NO_APERF;
        }
 
@@ -1271,7 +1011,8 @@ probe_cpu()
                        do_pkg_cstate = (1 << 2) | (1 << 3) | (1 << 6) | (1 << 7) | (1 << 8) | (1 << 9) | (1 << 10);
                        break;
                default:
-                       ERROR("Unsupported CPU");
+                       ERROR("Turbostat plugin: Unsupported CPU (family: %#x,"
+                             " model: %#x)", family, model);
                }
                switch (model) {
                case 0x2A:
@@ -1296,7 +1037,8 @@ probe_cpu()
                        do_rapl = 0;
                }
        } else {
-               ERROR("Unsupported CPU");
+               ERROR("Turbostat plugin: Unsupported CPU (family: %#x, "
+                     "model: %#x)", family, model);
                return -UNSUPPORTED_CPU;
        }
 
@@ -1315,236 +1057,390 @@ probe_cpu()
 }
 
 
+/********************
+ * Topology Probing *
+ ********************/
 
-static int __attribute__((warn_unused_result))
-topology_probe()
+/*
+ * Read a single int from a file.
+ */
+static int __attribute__ ((format(printf,1,2)))
+parse_int_file(const char *fmt, ...)
 {
-       int i;
-       int ret;
-       int max_core_id = 0;
-       int max_package_id = 0;
-       int max_siblings = 0;
-       struct cpu_topology {
-               int core_id;
-               int physical_package_id;
-       } *cpus;
-
-       /* Initialize num_cpus, max_cpu_num */
-       topo.num_cpus = 0;
-       topo.max_cpu_num = 0;
-       ret = for_all_proc_cpus(count_cpus);
-       if (ret < 0)
-               return ret;
-
-       DEBUG("num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num);
-
-       cpus = calloc(1, (topo.max_cpu_num  + 1) * sizeof(struct cpu_topology));
-       if (cpus == NULL) {
-               ERROR("calloc cpus");
-               return -ERR_CALLOC;
+       va_list args;
+       char path[PATH_MAX];
+       FILE *filep;
+       int len, value;
+
+       va_start(args, fmt);
+       len = vsnprintf(path, sizeof(path), fmt, args);
+       va_end(args);
+       if (len < 0 || len >= PATH_MAX) {
+               ERROR("Turbostat plugin: path truncated: '%s'", path);
+               return -ERR_PATH_TOO_LONG;
        }
 
-       /*
-        * Allocate and initialize cpu_present_set
-        */
-       cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1));
-       if (cpu_present_set == NULL) {
-               free(cpus);
-               ERROR("CPU_ALLOC");
-               return -ERR_CPU_ALLOC;
+       filep = fopen(path, "r");
+       if (!filep) {
+               ERROR("Turbostat plugin: Failed to open '%s'", path);
+               return -ERR_CANT_OPEN_FILE;
        }
-       cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
-       CPU_ZERO_S(cpu_present_setsize, cpu_present_set);
-       ret = for_all_proc_cpus(mark_cpu_present);
-       if (ret < 0) {
-               free(cpus);
-               return ret;
+       if (fscanf(filep, "%d", &value) != 1) {
+               ERROR("Turbostat plugin: Failed to parse number from '%s'", path);
+               return -ERR_CANT_READ_NUMBER;
        }
+       fclose(filep);
+       return value;
+}
 
+static int
+get_threads_on_core(int cpu)
+{
+       char path[80];
+       FILE *filep;
+       int sib1, sib2;
+       int matches;
+       char character;
+
+       ssnprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu);
+       filep = fopen(path, "r");
+       if (!filep) {
+               ERROR("Turbostat plugin: Failed to open '%s'", path);
+               return -ERR_CANT_OPEN_FILE;
+       }
        /*
-        * Allocate and initialize cpu_affinity_set
+        * file format:
+        * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4)
+        * otherwinse 1 sibling (self).
         */
-       cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
-       if (cpu_affinity_set == NULL) {
-               free(cpus);
-               ERROR("CPU_ALLOC");
-               return -ERR_CPU_ALLOC;
+       matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2);
+
+       fclose(filep);
+
+       if (matches == 3)
+               return 2;
+       else
+               return 1;
+}
+
+/*
+ * run func(cpu) on every cpu in /proc/stat
+ * return max_cpu number
+ */
+static int __attribute__((warn_unused_result))
+for_all_proc_cpus(int (func)(int))
+{
+       FILE *fp;
+       int cpu_num;
+       int retval;
+
+       fp = fopen("/proc/stat", "r");
+       if (!fp) {
+               ERROR("Turbostat plugin: Failed to open /proc/stat");
+               return -ERR_CANT_OPEN_FILE;
+       }
+
+       retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n");
+       if (retval != 0) {
+               ERROR("Turbostat plugin: Failed to parse /proc/stat");
+               fclose(fp);
+               return -ERR_CANT_READ_PROC_STAT;
        }
-       cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
-       CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set);
 
+       while (1) {
+               retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num);
+               if (retval != 1)
+                       break;
 
-       /*
-        * Allocate and initialize cpu_saved_affinity_set
-        */
-       cpu_saved_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1));
-       if (cpu_saved_affinity_set == NULL) {
-               free(cpus);
-               ERROR("CPU_ALLOC");
+               retval = func(cpu_num);
+               if (retval) {
+                       fclose(fp);
+                       return(retval);
+               }
+       }
+       fclose(fp);
+       return 0;
+}
+
+/*
+ * Update the stored topology.max_cpu_id
+ */
+static int
+update_max_cpu_id(int cpu)
+{
+       if (topology.max_cpu_id < cpu)
+               topology.max_cpu_id = cpu;
+       return 0;
+}
+
+static int
+mark_cpu_present(int cpu)
+{
+       CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set);
+       return 0;
+}
+
+static int __attribute__((warn_unused_result))
+allocate_cpu_set(cpu_set_t ** set, size_t * size) {
+       *set = CPU_ALLOC(topology.max_cpu_id  + 1);
+       if (*set == NULL) {
+               ERROR("Turbostat plugin: Unable to allocate CPU state");
                return -ERR_CPU_ALLOC;
        }
-       cpu_saved_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1));
-       CPU_ZERO_S(cpu_saved_affinity_setsize, cpu_saved_affinity_set);
+       *size = CPU_ALLOC_SIZE(topology.max_cpu_id  + 1);
+       CPU_ZERO_S(*size, *set);
+       return 0;
+}
 
+/*
+ * Build a local representation of the cpu distribution
+ */
+static int __attribute__((warn_unused_result))
+topology_probe()
+{
+       int i;
+       int ret;
+       int max_package_id, max_core_id, max_thread_id;
+       max_package_id = max_core_id = max_thread_id = 0;
+
+       /* Clean topology */
+       free(topology.cpus);
+       memset(&topology, 0, sizeof(topology));
+
+       ret = for_all_proc_cpus(update_max_cpu_id);
+       if (ret != 0)
+               goto err;
+
+       topology.cpus = calloc(1, (topology.max_cpu_id  + 1) * sizeof(struct cpu_topology));
+       if (topology.cpus == NULL) {
+               ERROR("Turbostat plugin: Unable to allocate memory for CPU topology");
+               return -ERR_CALLOC;
+       }
+
+       ret = allocate_cpu_set(&cpu_present_set, &cpu_present_setsize);
+       if (ret != 0)
+               goto err;
+       ret = allocate_cpu_set(&cpu_affinity_set, &cpu_affinity_setsize);
+       if (ret != 0)
+               goto err;
+       ret = allocate_cpu_set(&cpu_saved_affinity_set, &cpu_saved_affinity_setsize);
+       if (ret != 0)
+               goto err;
+
+       ret = for_all_proc_cpus(mark_cpu_present);
+       if (ret != 0)
+               goto err;
 
        /*
         * For online cpus
         * find max_core_id, max_package_id
         */
-       for (i = 0; i <= topo.max_cpu_num; ++i) {
-               int siblings;
+       for (i = 0; i <= topology.max_cpu_id; ++i) {
+               int num_threads;
+               struct cpu_topology *cpu = &topology.cpus[i];
 
                if (cpu_is_not_present(i)) {
-                       WARNING("cpu%d NOT PRESENT", i);
+                       WARNING("Turbostat plugin: cpu%d NOT PRESENT", i);
                        continue;
                }
-               cpus[i].core_id = get_core_id(i);
-               if (cpus[i].core_id < 0)
-                       return cpus[i].core_id;
-               if (cpus[i].core_id > max_core_id)
-                       max_core_id = cpus[i].core_id;
-
-               cpus[i].physical_package_id = get_physical_package_id(i);
-               if (cpus[i].physical_package_id < 0)
-                       return cpus[i].physical_package_id;
-               if (cpus[i].physical_package_id > max_package_id)
-                       max_package_id = cpus[i].physical_package_id;
-
-               siblings = get_num_ht_siblings(i);
-               if (siblings < 0)
-                       return siblings;
-               if (siblings > max_siblings)
-                       max_siblings = siblings;
-               DEBUG("cpu %d pkg %d core %d\n",
-                       i, cpus[i].physical_package_id, cpus[i].core_id);
-       }
-       topo.num_cores_per_pkg = max_core_id + 1;
-       DEBUG("max_core_id %d, sizing for %d cores per package\n",
-               max_core_id, topo.num_cores_per_pkg);
 
-       topo.num_packages = max_package_id + 1;
-       DEBUG("max_package_id %d, sizing for %d packages\n",
-               max_package_id, topo.num_packages);
+               ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/physical_package_id", i);
+               if (ret < 0)
+                       goto err;
+               else
+                       cpu->package_id = ret;
+               if (cpu->package_id > max_package_id)
+                       max_package_id = cpu->package_id;
 
-       topo.num_threads_per_core = max_siblings;
-       DEBUG("max_siblings %d\n", max_siblings);
+               ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_id", i);
+               if (ret < 0)
+                       goto err;
+               else
+                       cpu->core_id = ret;
+               if (cpu->core_id > max_core_id)
+                       max_core_id = cpu->core_id;
+               ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", i);
+               if (ret < 0)
+                       goto err;
+               else if (ret == i)
+                       cpu->first_core_in_package = 1;
+
+               ret = get_threads_on_core(i);
+               if (ret < 0)
+                       goto err;
+               else
+                       num_threads = ret;
+               if (num_threads > max_thread_id)
+                       max_thread_id = num_threads;
+               ret = parse_int_file("/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", i);
+               if (ret < 0)
+                       goto err;
+               else if (ret == i)
+                       cpu->first_thread_in_core = 1;
+
+               DEBUG("Turbostat plugin: cpu %d pkg %d core %d\n",
+                       i, cpu->package_id, cpu->core_id);
+       }
+       /* Num is max + 1 (need to count 0) */
+       topology.num_packages = max_package_id + 1;
+       topology.num_cores = max_core_id + 1;
+       topology.num_threads = max_thread_id + 1;
 
-       free(cpus);
        return 0;
+err:
+       free(topology.cpus);
+       return ret;
 }
 
+
+/************************
+ * Main alloc/init/free *
+ ************************/
+
 static int
-allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p)
+allocate_counters(struct thread_data **threads, struct core_data **cores, struct pkg_data **packages)
 {
        int i;
+       int total_threads, total_cores;
 
-       *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg *
-               topo.num_packages, sizeof(struct thread_data));
-       if (*t == NULL)
-               goto error;
+       total_threads = topology.num_threads * topology.num_cores * topology.num_packages;
+       *threads = calloc(total_threads, sizeof(struct thread_data));
+       if (*threads == NULL)
+               goto err;
 
-       for (i = 0; i < topo.num_threads_per_core *
-               topo.num_cores_per_pkg * topo.num_packages; i++)
-               (*t)[i].cpu_id = -1;
+       for (i = 0; i < total_threads; ++i)
+               (*threads)[i].cpu_id = -1;
 
-       *c = calloc(topo.num_cores_per_pkg * topo.num_packages,
-               sizeof(struct core_data));
-       if (*c == NULL)
-               goto error;
+       total_cores = topology.num_cores * topology.num_packages;
+       *cores = calloc(total_cores, sizeof(struct core_data));
+       if (*cores == NULL)
+               goto err_clean_threads;
 
-       for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++)
-               (*c)[i].core_id = -1;
+       for (i = 0; i < total_cores; ++i)
+               (*cores)[i].core_id = -1;
 
-       *p = calloc(topo.num_packages, sizeof(struct pkg_data));
-       if (*p == NULL)
-               goto error;
+       *packages = calloc(topology.num_packages, sizeof(struct pkg_data));
+       if (*packages == NULL)
+               goto err_clean_cores;
 
-       for (i = 0; i < topo.num_packages; i++)
-               (*p)[i].package_id = i;
+       for (i = 0; i < topology.num_packages; i++)
+               (*packages)[i].package_id = i;
 
        return 0;
-error:
-       ERROR("calloc counters");
+
+err_clean_cores:
+       free(*cores);
+err_clean_threads:
+       free(*threads);
+err:
+       ERROR("Turbostat plugin: Failled to allocate memory for counters");
        return -ERR_CALLOC;
 }
-/*
- * init_counter()
- *
- * set cpu_id, core_num, pkg_num
- * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE
- *
- * increment topo.num_cores when 1st core in pkg seen
- */
+
 static int
 init_counter(struct thread_data *thread_base, struct core_data *core_base,
-       struct pkg_data *pkg_base, int thread_num, int core_num,
-       int pkg_num, int cpu_id)
+       struct pkg_data *pkg_base, int cpu_id)
 {
-       int ret;
        struct thread_data *t;
        struct core_data *c;
        struct pkg_data *p;
+       struct cpu_topology *cpu = &topology.cpus[cpu_id];
 
-       t = GET_THREAD(thread_base, thread_num, core_num, pkg_num);
-       c = GET_CORE(core_base, core_num, pkg_num);
-       p = GET_PKG(pkg_base, pkg_num);
+       t = GET_THREAD(thread_base, !(cpu->first_thread_in_core), cpu->core_id, cpu->package_id);
+       c = GET_CORE(core_base, cpu->core_id, cpu->package_id);
+       p = GET_PKG(pkg_base, cpu->package_id);
 
        t->cpu_id = cpu_id;
-       if (thread_num == 0) {
+       if (cpu->first_thread_in_core)
                t->flags |= CPU_IS_FIRST_THREAD_IN_CORE;
-               if ((ret = cpu_is_first_core_in_package(cpu_id)) < 0) {
-                       return ret;
-               } else if (ret != 0) {
-                       t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
-               }
-       }
+       if (cpu->first_core_in_package)
+               t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE;
 
-       c->core_id = core_num;
-       p->package_id = pkg_num;
+       c->core_id = cpu->core_id;
+       p->package_id = cpu->package_id;
 
        return 0;
 }
 
-
 static int
-initialize_counters(int cpu_id)
+initialize_counters(void)
 {
-       int my_thread_id, my_core_id, my_package_id;
        int ret;
+       int cpu_id;
 
-       my_package_id = get_physical_package_id(cpu_id);
-       if (my_package_id < 0)
-               return my_package_id;
-       my_core_id = get_core_id(cpu_id);
-       if (my_core_id < 0)
-               return my_core_id;
-
-       if ((ret = cpu_is_first_sibling_in_core(cpu_id)) < 0) {
-               return ret;
-       } else if (ret != 0) {
-               my_thread_id = 0;
-               topo.num_cores++;
-       } else {
-               my_thread_id = 1;
-       }
+       for (cpu_id = 0; cpu_id <= topology.max_cpu_id; ++cpu_id) {
+               if (cpu_is_not_present(cpu_id)) {
+                       continue;
+               }
 
-       ret = init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
-       if (ret < 0)
-               return ret;
-       ret = init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
-       if (ret < 0)
-               return ret;
-       ret = init_counter(DELTA_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id);
-       if (ret < 0)
-               return ret;
+               ret = init_counter(EVEN_COUNTERS, cpu_id);
+               if (ret < 0)
+                       return ret;
+               ret = init_counter(ODD_COUNTERS, cpu_id);
+               if (ret < 0)
+                       return ret;
+               ret = init_counter(DELTA_COUNTERS, cpu_id);
+               if (ret < 0)
+                       return ret;
+       }
        return 0;
 }
 
+
+
+static void
+free_all_buffers(void)
+{
+       allocated = 0;
+       initialized = 0;
+
+       CPU_FREE(cpu_present_set);
+       cpu_present_set = NULL;
+       cpu_present_set = 0;
+
+       CPU_FREE(cpu_affinity_set);
+       cpu_affinity_set = NULL;
+       cpu_affinity_setsize = 0;
+
+       CPU_FREE(cpu_saved_affinity_set);
+       cpu_saved_affinity_set = NULL;
+       cpu_saved_affinity_setsize = 0;
+
+       free(thread_even);
+       free(core_even);
+       free(package_even);
+
+       thread_even = NULL;
+       core_even = NULL;
+       package_even = NULL;
+
+       free(thread_odd);
+       free(core_odd);
+       free(package_odd);
+
+       thread_odd = NULL;
+       core_odd = NULL;
+       package_odd = NULL;
+
+       free(thread_delta);
+       free(core_delta);
+       free(package_delta);
+
+       thread_delta = NULL;
+       core_delta = NULL;
+       package_delta = NULL;
+}
+
+
+/**********************
+ * Collectd functions *
+ **********************/
+
 #define DO_OR_GOTO_ERR(something) \
-do {                         \
-       ret = (something);     \
-       if (ret < 0)         \
-               goto err;    \
+do {                             \
+       ret = (something);        \
+       if (ret < 0)              \
+               goto err;         \
 } while (0)
 
 static int setup_all_buffers(void)
@@ -1555,7 +1451,7 @@ static int setup_all_buffers(void)
        DO_OR_GOTO_ERR(allocate_counters(&thread_even, &core_even, &package_even));
        DO_OR_GOTO_ERR(allocate_counters(&thread_odd, &core_odd, &package_odd));
        DO_OR_GOTO_ERR(allocate_counters(&thread_delta, &core_delta, &package_delta));
-       DO_OR_GOTO_ERR(for_all_proc_cpus(initialize_counters));
+       DO_OR_GOTO_ERR(initialize_counters());
        DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, EVEN_COUNTERS));
        DO_OR_GOTO_ERR(for_all_cpus(set_temperature_target, ODD_COUNTERS));
 
@@ -1567,16 +1463,94 @@ err:
 }
 
 static int
+turbostat_read(void)
+{
+       int ret;
+
+       if (!allocated) {
+               if ((ret = setup_all_buffers()) < 0)
+                       return ret;
+       }
+
+       if (for_all_proc_cpus(cpu_is_not_present)) {
+               free_all_buffers();
+               if ((ret = setup_all_buffers()) < 0)
+                       return ret;
+               if (for_all_proc_cpus(cpu_is_not_present))
+                       return -ERR_CPU_NOT_PRESENT;
+       }
+
+       /* Saving the scheduling affinity, as it will be modified by get_counters */
+       if (sched_getaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set) != 0)
+               return -ERR_CPU_SAVE_SCHED_AFFINITY;
+
+       if (!initialized) {
+               if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
+                       goto out;
+               gettimeofday(&tv_even, (struct timezone *)NULL);
+               is_even = 1;
+               initialized = 1;
+               ret = 0;
+               goto out;
+       }
+
+       if (is_even) {
+               if ((ret = for_all_cpus(get_counters, ODD_COUNTERS)) < 0)
+                       goto out;
+               gettimeofday(&tv_odd, (struct timezone *)NULL);
+               is_even = 0;
+               timersub(&tv_odd, &tv_even, &tv_delta);
+               if ((ret = for_all_cpus_delta(ODD_COUNTERS, EVEN_COUNTERS)) < 0)
+                       goto out;
+               if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
+                       goto out;
+       } else {
+               if ((ret = for_all_cpus(get_counters, EVEN_COUNTERS)) < 0)
+                       goto out;
+               gettimeofday(&tv_even, (struct timezone *)NULL);
+               is_even = 1;
+               timersub(&tv_even, &tv_odd, &tv_delta);
+               if ((ret = for_all_cpus_delta(EVEN_COUNTERS, ODD_COUNTERS)) < 0)
+                       goto out;
+               if ((ret = for_all_cpus(submit_counters, DELTA_COUNTERS)) < 0)
+                       goto out;
+       }
+       ret = 0;
+out:
+       /*
+        * Let's restore the affinity
+        * This might fail if the number of CPU changed, but we can't do anything in that case..
+        */
+       (void)sched_setaffinity(0, cpu_saved_affinity_setsize, cpu_saved_affinity_set);
+       return ret;
+}
+
+static int
 turbostat_init(void)
 {
+       struct stat sb;
        int ret;
 
-       DO_OR_GOTO_ERR(check_super_user());
+       if (getuid() != 0) {
+               ERROR("Turbostat plugin: Initialization failed: this plugin "
+                     "requires collectd to run as root in order to read "
+                     "special CPU registers");
+               return -ERR_NOT_ROOT;
+       }
+
        DO_OR_GOTO_ERR(probe_cpu());
-       DO_OR_GOTO_ERR(check_dev_msr());
+
+       if (stat("/dev/cpu/0/msr", &sb)) {
+               ERROR("Turbostat plugin: Initialization failed: /dev/cpu/0/msr"
+                     " does not exist while the CPU supports MSR. You may be "
+                     "missing the corresponding kernel module, please try '# "
+                     "modprobe msr'");
+               return -ERR_NO_MSR;
+       }
+
        DO_OR_GOTO_ERR(setup_all_buffers());
 
-       plugin_register_complex_read(NULL, PLUGIN_NAME, turbostat_read, NULL, NULL);
+       plugin_register_read(PLUGIN_NAME, turbostat_read);
 
        return 0;
 err:
@@ -1607,7 +1581,6 @@ turbostat_config(const char *key, const char *value)
        return 0;
 }
 
-void module_register(void);
 void module_register(void)
 {
        plugin_register_init(PLUGIN_NAME, turbostat_init);