perflib: add ability to set thread affinity

Co-Authored-by: Alexandr Nedvedicky <[email protected]>
Signed-off-by: Eugene Syromiatnikov <[email protected]>

Author: esyr

source/perflib/perfhelper.c

@@ -9,6 +9,7 @@
 
 #include <string.h>
 #include <openssl/crypto.h>
+#include <openssl/macros.h>
 #include <openssl/bio.h>
 #include <openssl/err.h>
 #include <openssl/ssl.h>

source/perflib/perflib.h

@@ -21,6 +21,7 @@
 #  include <windows.h>
 
 typedef HANDLE thread_t;
+typedef DWORD_PTR affinity_t;
 
 #  define strcasecmp(_a, _b) _stricmp((_a), (_b))
 
@@ -29,6 +30,7 @@ typedef HANDLE thread_t;
 #  include <pthread.h>
 
 typedef pthread_t thread_t;
+typedef unsigned long affinity_t;
 
 # endif
 
@@ -37,10 +39,60 @@ struct thread_arg_st {
     size_t num;
 };
 
-int perflib_run_thread(thread_t *t, struct thread_arg_st *arg);
+/**
+ * A callback that allows setting CPU affinity for the threads being run.
+ * Gets the set of available CPUs in the cpu_set argument and expected
+ * to update it in accordance with the information provided in num and arg
+ * arguments.
+ *
+ * Currently supported only on glibc on Linux (because of a non-privileged
+ * pthread_attr_setaffinity_np, with a maximum of 1024 CPUs) and Windows
+ * (with a limitation of using only the initial process group).
+ *
+ * @param[in,out] cpu_set       On entering, contains the set of CPUs available
+ *                              to the test.  The callback is supposed
+ *                              to update the set in accordance
+ *                              with the information provided in num, cnt,
+ *                              and arg parameters.
+ * @param[in]     cpu_set_bits  Size of cpu_set_bits, in bits.
+ * @param[in]     num           Index of a thread index being run,
+ *                              counted from 0.
+ * @param[in]     cnt           Total count of available CPUs
+ *                              (popcnt(cpu_set_bits)).
+ * @param         arg           An opaque pointer that a caller has provided
+ *                              along the callback.
+ * @return                      1 on success, 0 on error.
+ */
+typedef int (*perflib_affinity_fn)(affinity_t *cpu_set, size_t cpu_set_bits,
+                                   size_t num, size_t cnt, void *arg);
+
+/**
+ * A simple affinity callback that assigns each thread a single CPU
+ * in a round robin fashion.  arg must be NULL.
+ */
+int perflib_roundrobin_affinity(affinity_t *cpu_set_bits, size_t cpu_set_size,
+                                size_t num, size_t cnt, void *arg);
+
+int perflib_run_thread_ex(thread_t *t, struct thread_arg_st *arg,
+                          perflib_affinity_fn affinity_cb,
+                          void *affinity_cb_arg);
+static ossl_unused ossl_inline int
+perflib_run_thread(thread_t *t, struct thread_arg_st *arg)
+{
+    return perflib_run_thread_ex(t, arg, NULL, NULL);
+}
 int perflib_wait_for_thread(thread_t thread);
-int perflib_run_multi_thread_test(void (*f)(size_t), size_t threadcount,
-                                  OSSL_TIME *duration);
+int perflib_run_multi_thread_test_ex(void (*f)(size_t), size_t threadcount,
+                                     OSSL_TIME *duration,
+                                     perflib_affinity_fn affinity_cb,
+                                     void *affinity_cb_arg);
+static ossl_unused ossl_inline int
+perflib_run_multi_thread_test(void (*f)(size_t), size_t threadcount,
+                              OSSL_TIME *duration)
+{
+    return perflib_run_multi_thread_test_ex(f, threadcount, duration,
+                                            NULL, NULL);
+}
 char *perflib_mk_file_path(const char *dir, const char *file);
 char *perflib_glue_strings(const char *list[], size_t *out_len);