fix: num_possible_cpus() with hot-unplugged CPUs

[lttng-ust.git] / libringbuffer / smp.c

diff --git a/libringbuffer/smp.c b/libringbuffer/smp.c
index 656a75da2787a40d6ac1188263863069f42745cd..8b66b9e83fd08b88c6b70798000781898af1a562 100644 (file)
--- a/libringbuffer/smp.c
+++ b/libringbuffer/smp.c
@@ -21,17 +21,23 @@
 
 #define _GNU_SOURCE
 #define _LGPL_SOURCE
-#include <unistd.h>
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <limits.h>
 #include <pthread.h>
+#include <stdlib.h>
+#include <unistd.h>
+
 #include "smp.h"
+#include "usterr-signal-safe.h"
 
 int __num_possible_cpus;
 
 #if (defined(__GLIBC__) || defined( __UCLIBC__))
-void _get_num_possible_cpus(void)
+int get_num_possible_cpus_fallback(void)
 {
-       int result;
-
        /* On Linux, when some processors are offline
         * _SC_NPROCESSORS_CONF counts the offline
         * processors, whereas _SC_NPROCESSORS_ONLN
@@ -40,10 +46,7 @@ void _get_num_possible_cpus(void)
         * this sysconf, in which case the arrays
         * indexed by processor would overflow.
         */
-       result = sysconf(_SC_NPROCESSORS_CONF);
-       if (result == -1)
-               return;
-       __num_possible_cpus = result;
+       return sysconf(_SC_NPROCESSORS_CONF);
 }
 
 #else
@@ -66,9 +69,9 @@ void _get_num_possible_cpus(void)
 
 #define __max(a,b) ((a)>(b)?(a):(b))
 
-void _get_num_possible_cpus(void)
+int get_num_possible_cpus_fallback(void)
 {
-       int result, count = 0;
+       int count = 0;
        DIR *cpudir;
        struct dirent *entry;
 
@@ -99,13 +102,131 @@ end:
        /*
         * Get the sysconf value as a fallback. Keep the highest number.
         */
-       result = __max(sysconf(_SC_NPROCESSORS_CONF), count);
+       return __max(sysconf(_SC_NPROCESSORS_CONF), count);
+}
+#endif
+
+/*
+ * Get the CPU possible mask string from sysfs.
+ *
+ * buf: the buffer where the mask will be read.
+ * max_bytes: the maximum number of bytes to write in the buffer.
+ *
+ * Returns the number of bytes read or -1 on error.
+ */
+int get_possible_cpu_mask_from_sysfs(char *buf, size_t max_bytes)
+{
+       ssize_t bytes_read = 0;
+       size_t total_bytes_read = 0;
+       int fd = -1, ret = -1;
+
+       if (buf == NULL)
+               goto end;
+
+       fd = open("/sys/devices/system/cpu/possible", O_RDONLY);
+       if (fd < 0)
+               goto end;
+
+       do {
+               bytes_read = read(fd, buf + total_bytes_read,
+                               max_bytes - total_bytes_read);
+
+               if (bytes_read < 0) {
+                       if (errno == EINTR) {
+                               continue;       /* retry operation */
+                       } else {
+                               goto end;
+                       }
+               }
+
+               total_bytes_read += bytes_read;
+               assert(total_bytes_read <= max_bytes);
+       } while (max_bytes > total_bytes_read && bytes_read > 0);
+
+       /*
+        * Make sure the mask read is a null terminated string.
+        */
+       if (total_bytes_read < max_bytes)
+               buf[total_bytes_read] = '\0';
+       else
+               buf[max_bytes - 1] = '\0';
+
+       if (total_bytes_read > INT_MAX)
+               goto end;
+
+       ret = (int) total_bytes_read;
+
+end:
+       if (fd >= 0 && close(fd) < 0)
+               PERROR("close");
+
+       return ret;
+}
+
+/*
+ * Get the number of CPUs from the possible cpu mask.
+ *
+ * pmask: the mask to parse.
+ * len: the len of the mask excluding '\0'.
+ *
+ * Returns the number of possible CPUs from the mask or 0 on error.
+ */
+int get_num_possible_cpus_from_mask(const char *pmask, size_t len)
+{
+       ssize_t i;
+       unsigned long cpu_index;
+       char *endptr;
+
+       /* We need at least one char to read */
+       if (len < 1)
+               goto error;
+
+       /* Start from the end to read the last CPU index. */
+       for (i = len - 1; i > 0; i--) {
+               /* Break when we hit the first separator. */
+               if ((pmask[i] == ',') || (pmask[i] == '-')) {
+                       i++;
+                       break;
+               }
+       }
+
+       cpu_index = strtoul(&pmask[i], &endptr, 10);
 
        /*
-        * If both methods failed, don't store the value.
+        * If we read a CPU index, increment it by one to return a number of
+        * CPUs.
         */
-       if (result < 1)
+       if ((&pmask[i] != endptr) && (cpu_index < INT_MAX))
+               return (int) cpu_index + 1;
+
+error:
+       return 0;
+}
+
+void _get_num_possible_cpus(void)
+{
+       int ret;
+       int buf_len = sysconf(_SC_PAGE_SIZE);
+       char buf[buf_len];
+
+       /* Get the possible cpu mask from sysfs, fallback to sysconf. */
+       ret = get_possible_cpu_mask_from_sysfs((char *) &buf, buf_len);
+       if (ret <= 0)
+               goto fallback;
+
+       /* Parse the possible cpu mask, on failure fallback to sysconf. */
+       ret = get_num_possible_cpus_from_mask((char *) &buf, ret);
+       if (ret > 0)
+               goto end;
+
+fallback:
+       /* Fallback to sysconf. */
+       ret = get_num_possible_cpus_fallback();
+
+end:
+       /* If all methods failed, don't store the value. */
+       if (ret < 1)
                return;
-       __num_possible_cpus = result;
+
+       __num_possible_cpus = ret;
 }
-#endif