2 * Copyright (C) 2016 Red Hat, Inc.
3 * Author: Michael S. Tsirkin <mst@redhat.com>
4 * This work is licensed under the terms of the GNU GPL, version 2.
6 * Common macros and functions for ring benchmarking.
17 #if defined(__x86_64__) || defined(__i386__)
18 #include "x86intrin.h"
20 static inline void wait_cycles(unsigned long long cycles
)
25 while (__rdtsc() - t
< cycles
) {}
28 #define VMEXIT_CYCLES 500
29 #define VMENTRY_CYCLES 500
31 #elif defined(__s390x__)
32 static inline void wait_cycles(unsigned long long cycles
)
34 asm volatile("0: brctg %0,0b" : : "d" (cycles
));
38 #define VMEXIT_CYCLES 200
39 #define VMENTRY_CYCLES 200
42 static inline void wait_cycles(unsigned long long cycles
)
46 #define VMEXIT_CYCLES 0
47 #define VMENTRY_CYCLES 0
50 static inline void vmexit(void)
55 wait_cycles(VMEXIT_CYCLES
);
57 static inline void vmentry(void)
62 wait_cycles(VMENTRY_CYCLES
);
65 /* implemented by ring */
66 void alloc_ring(void);
68 int add_inbuf(unsigned, void *, void *);
69 void *get_buf(unsigned *, void **);
73 void kick_available();
78 bool use_buf(unsigned *, void **);
81 /* implemented by main */
84 void wait_for_kick(void);
86 void wait_for_call(void);
88 extern unsigned ring_size
;
90 /* Compiler barrier - similar to what Linux uses */
91 #define barrier() asm volatile("" ::: "memory")
93 /* Is there a portable way to do this? */
94 #if defined(__x86_64__) || defined(__i386__)
95 #define cpu_relax() asm ("rep; nop" ::: "memory")
96 #elif defined(__s390x__)
97 #define cpu_relax() barrier()
99 #define cpu_relax() assert(0)
102 extern bool do_relax
;
104 static inline void busy_wait(void)
109 /* prevent compiler from removing busy loops */
114 * Not using __ATOMIC_SEQ_CST since gcc docs say they are only synchronized
115 * with other __ATOMIC_SEQ_CST calls.
117 #define smp_mb() __sync_synchronize()
120 * This abuses the atomic builtins for thread fences, and
121 * adds a compiler barrier.
123 #define smp_release() do { \
125 __atomic_thread_fence(__ATOMIC_RELEASE); \
128 #define smp_acquire() do { \
129 __atomic_thread_fence(__ATOMIC_ACQUIRE); \