int linuxNodeInfoCPUPopulate(FILE *cpuinfo,
const char *sysfs_dir,
virNodeInfoPtr nodeinfo)
{
char line[1024];
DIR *nodedir = NULL;
struct dirent *nodedirent = NULL;
int cpus, cores, socks, threads, offline = 0;
unsigned int node;
int ret = -1;
char *sysfs_nodedir = NULL;
char *sysfs_cpudir = NULL;
/* Start with parsing CPU clock speed from /proc/cpuinfo */
while (fgets(line, sizeof(line), cpuinfo) != NULL) {
# if defined(__x86_64__) || \
defined(__amd64__) || \
defined(__i386__)
char *buf = line;
if (STRPREFIX(buf, "cpu MHz")) {
char *p;
unsigned int ui;
buf += 7;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
_("parsing cpu MHz from cpuinfo"));
goto cleanup;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0 &&
/* Accept trailing fractional part. */
(*p == '\0' || *p == '.' || c_isspace(*p)))
nodeinfo->mhz = ui;
}
# elif defined(__powerpc__) || \
defined(__powerpc64__)
char *buf = line;
if (STRPREFIX(buf, "clock")) {
char *p;
unsigned int ui;
buf += 5;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
_("parsing cpu MHz from cpuinfo"));
goto cleanup;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0 &&
/* Accept trailing fractional part. */
(*p == '\0' || *p == '.' || c_isspace(*p)))
nodeinfo->mhz = ui;
/* No other interesting infos are available in /proc/cpuinfo.
* However, there is a line identifying processor's version,
* identification and machine, but we don't want it to be caught
* and parsed in next iteration, because it is not in expected
* format and thus lead to error. */
}
# elif defined(__arm__)
char *buf = line;
if (STRPREFIX(buf, "BogoMIPS")) {
char *p;
unsigned int ui;
buf += 8;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
virReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("parsing cpu MHz from cpuinfo"));
goto cleanup;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0
/* Accept trailing fractional part. */
&& (*p == '\0' || *p == '.' || c_isspace(*p)))
nodeinfo->mhz = ui;
}
# elif defined(__s390__) || \
defined(__s390x__)
/* s390x has no realistic value for CPU speed,
* assign a value of zero to signify this */
nodeinfo->mhz = 0;
# else
# warning Parser for /proc/cpuinfo needs to be adapted for your architecture
# endif
}
/* OK, we've parsed clock speed out of /proc/cpuinfo. Get the
* core, node, socket, thread and topology information from /sys
*/
if (virAsprintf(&sysfs_nodedir, "%s/node", sysfs_dir) < 0)
goto cleanup;
if (!(nodedir = opendir(sysfs_nodedir))) {
/* the host isn't probably running a NUMA architecture */
goto fallback;
}
errno = 0;
while ((nodedirent = readdir(nodedir))) {
if (sscanf(nodedirent->d_name, "node%u", &node) != 1)
continue;
nodeinfo->nodes++;
if (virAsprintf(&sysfs_cpudir, "%s/node/%s",
sysfs_dir, nodedirent->d_name) < 0)
goto cleanup;
if ((cpus = virNodeParseNode(sysfs_cpudir, &socks, &cores,
&threads, &offline)) < 0)
goto cleanup;
VIR_FREE(sysfs_cpudir);
nodeinfo->cpus += cpus;
if (socks > nodeinfo->sockets)
nodeinfo->sockets = socks;
if (cores > nodeinfo->cores)
nodeinfo->cores = cores;
if (threads > nodeinfo->threads)
nodeinfo->threads = threads;
errno = 0;
}
if (errno) {
virReportSystemError(errno, _("problem reading %s"), sysfs_nodedir);
goto cleanup;
}
if (nodeinfo->cpus && nodeinfo->nodes)
goto done;
fallback:
VIR_FREE(sysfs_cpudir);
if (virAsprintf(&sysfs_cpudir, "%s/cpu", sysfs_dir) < 0)
goto cleanup;
if ((cpus = virNodeParseNode(sysfs_cpudir, &socks, &cores,
&threads, &offline)) < 0)
goto cleanup;
nodeinfo->nodes = 1;
nodeinfo->cpus = cpus;
nodeinfo->sockets = socks;
nodeinfo->cores = cores;
nodeinfo->threads = threads;
done:
/* There should always be at least one cpu, socket, node, and thread. */
if (nodeinfo->cpus == 0) {
virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("no CPUs found"));
goto cleanup;
}
if (nodeinfo->sockets == 0) {
virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("no sockets found"));
goto cleanup;
}
if (nodeinfo->threads == 0) {
virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("no threads found"));
goto cleanup;
}
/* Now check if the topology makes sense. There are machines that don't
* expose their real number of nodes or for example the AMD Bulldozer
* architecture that exposes their Clustered integer core modules as both
* threads and cores. This approach throws off our detection. Unfortunately
* the nodeinfo structure isn't designed to carry the full topology so
* we're going to lie about the detected topology to notify the user
* to check the host capabilities for the actual topology. */
if ((nodeinfo->nodes *
nodeinfo->sockets *
nodeinfo->cores *
nodeinfo->threads) != (nodeinfo->cpus + offline)) {
nodeinfo->nodes = 1;
nodeinfo->sockets = 1;
nodeinfo->cores = nodeinfo->cpus + offline;
nodeinfo->threads = 1;
}
ret = 0;
cleanup:
/* don't shadow a more serious error */
if (nodedir && closedir(nodedir) < 0 && ret >= 0) {
virReportSystemError(errno, _("problem closing %s"), sysfs_nodedir);
ret = -1;
}
VIR_FREE(sysfs_nodedir);
VIR_FREE(sysfs_cpudir);
return ret;
}
int linuxNodeInfoCPUPopulate(FILE *cpuinfo,
const char *sysfs_dir,
virNodeInfoPtr nodeinfo)
{
char line[1024];
DIR *cpudir = NULL;
struct dirent *cpudirent = NULL;
unsigned int cpu;
unsigned long core, sock, cur_threads;
cpu_set_t core_mask;
cpu_set_t socket_mask;
int online;
int ret = -1;
char *sysfs_cpudir = NULL;
unsigned int cpu_cores = 0;
nodeinfo->cpus = 0;
nodeinfo->mhz = 0;
nodeinfo->cores = 0;
/* Start with parsing /proc/cpuinfo; although it tends to have
* fewer details. Hyperthreads are ignored at this stage. */
while (fgets(line, sizeof(line), cpuinfo) != NULL) {
# if defined(__x86_64__) || \
defined(__amd64__) || \
defined(__i386__)
char *buf = line;
if (STRPREFIX(buf, "cpu MHz")) {
char *p;
unsigned int ui;
buf += 9;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("parsing cpu MHz from cpuinfo"));
goto cleanup;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0
/* Accept trailing fractional part. */
&& (*p == '\0' || *p == '.' || c_isspace(*p)))
nodeinfo->mhz = ui;
}
if (STRPREFIX(buf, "cpu cores")) {
char *p;
unsigned int ui;
buf += 9;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("parsing cpu cores from cpuinfo"));
return -1;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0
&& (*p == '\0' || c_isspace(*p)))
cpu_cores = ui;
}
# elif defined(__powerpc__) || \
defined(__powerpc64__)
char *buf = line;
if (STRPREFIX(buf, "clock")) {
char *p;
unsigned int ui;
buf += 5;
while (*buf && c_isspace(*buf))
buf++;
if (*buf != ':' || !buf[1]) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("parsing cpu MHz from cpuinfo"));
goto cleanup;
}
if (virStrToLong_ui(buf+1, &p, 10, &ui) == 0
/* Accept trailing fractional part. */
&& (*p == '\0' || *p == '.' || c_isspace(*p)))
nodeinfo->mhz = ui;
/* No other interesting infos are available in /proc/cpuinfo.
* However, there is a line identifying processor's version,
* identification and machine, but we don't want it to be caught
* and parsed in next iteration, because it is not in expected
* format and thus lead to error. */
}
# else
# warning Parser for /proc/cpuinfo needs to be adapted for your architecture
# endif
}
/* OK, we've parsed clock speed out of /proc/cpuinfo. Get the
* core, node, socket, thread and topology information from /sys
*/
if (virAsprintf(&sysfs_cpudir, "%s/cpu", sysfs_dir) < 0) {
virReportOOMError();
goto cleanup;
}
cpudir = opendir(sysfs_cpudir);
if (cpudir == NULL) {
virReportSystemError(errno, _("cannot opendir %s"), sysfs_cpudir);
goto cleanup;
}
CPU_ZERO(&core_mask);
CPU_ZERO(&socket_mask);
while ((cpudirent = readdir(cpudir))) {
if (sscanf(cpudirent->d_name, "cpu%u", &cpu) != 1)
continue;
online = virNodeGetCpuValue(sysfs_cpudir, cpu, "online", true);
if (online < 0) {
closedir(cpudir);
goto cleanup;
}
if (!online)
continue;
nodeinfo->cpus++;
/* Parse core */
core = virNodeGetCpuValue(sysfs_cpudir, cpu, "topology/core_id", false);
if (!CPU_ISSET(core, &core_mask)) {
CPU_SET(core, &core_mask);
nodeinfo->cores++;
}
/* Parse socket */
sock = virNodeParseSocket(sysfs_cpudir, cpu);
if (!CPU_ISSET(sock, &socket_mask)) {
CPU_SET(sock, &socket_mask);
nodeinfo->sockets++;
}
cur_threads = virNodeCountThreadSiblings(sysfs_cpudir, cpu);
if (cur_threads == 0) {
closedir(cpudir);
goto cleanup;
}
if (cur_threads > nodeinfo->threads)
nodeinfo->threads = cur_threads;
}
if (errno) {
virReportSystemError(errno,
_("problem reading %s"), sysfs_cpudir);
closedir(cpudir);
goto cleanup;
}
if (closedir(cpudir) < 0) {
virReportSystemError(errno,
_("problem closing %s"), sysfs_cpudir);
goto cleanup;
}
if ((nodeinfo->nodes = virNodeParseNode(sysfs_dir)) <= 0)
goto cleanup;
/* There should always be at least one cpu, socket, node, and thread. */
if (nodeinfo->cpus == 0) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("no CPUs found"));
goto cleanup;
}
if (nodeinfo->sockets == 0) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("no sockets found"));
goto cleanup;
}
if (nodeinfo->threads == 0) {
nodeReportError(VIR_ERR_INTERNAL_ERROR,
"%s", _("no threads found"));
goto cleanup;
}
/* Platform like AMD Magny Cours has two NUMA nodes each package, and
* the two nodes share the same core ID set, it results in the cores
* number calculated from sysfs is not the actual cores number. Use
* "cpu cores" in /proc/cpuinfo as the cores number instead in this case.
* More details about the problem:
* https://www.redhat.com/archives/libvir-list/2012-May/msg00607.html
*/
if (cpu_cores && (cpu_cores > nodeinfo->cores))
nodeinfo->cores = cpu_cores;
/* nodeinfo->sockets is supposed to be a number of sockets per NUMA node,
* however if NUMA nodes are not composed of whole sockets, we just lie
* about the number of NUMA nodes and force apps to check capabilities XML
* for the actual NUMA topology.
*/
if (nodeinfo->sockets % nodeinfo->nodes == 0)
nodeinfo->sockets /= nodeinfo->nodes;
else
nodeinfo->nodes = 1;
ret = 0;
cleanup:
VIR_FREE(sysfs_cpudir);
return ret;
}
