When we implemented per-vCPU TCG contexts, we forgot to also
distribute the tcg_time counter, which has remained as a global
accessed without any serialization, leading to potentially missed
counts.
Fix it by distributing the field over the TCG contexts, embedding
it into TCGProfile with a field called "cpu_exec_time", which is more
descriptive than "tcg_time". Add a function to query this value
directly, and for completeness, fill in the field in
tcg_profile_snapshot, even though its callers do not use it.
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <
20181010144853.13005-5-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
ret = cpu_exec(cpu);
cpu_exec_end(cpu);
#ifdef CONFIG_PROFILER
- tcg_time += profile_getclock() - ti;
+ atomic_set(&tcg_ctx->prof.cpu_exec_time,
+ tcg_ctx->prof.cpu_exec_time + profile_getclock() - ti);
#endif
return ret;
}
return get_clock();
}
-extern int64_t tcg_time;
extern int64_t dev_time;
#endif
#include "sysemu/cpus.h"
#include "sysemu/iothread.h"
#include "qemu/cutils.h"
+#include "tcg/tcg.h"
#if defined(TARGET_S390X)
#include "hw/s390x/storage-keys.h"
#ifdef CONFIG_PROFILER
-int64_t tcg_time;
int64_t dev_time;
static void hmp_info_profile(Monitor *mon, const QDict *qdict)
{
+ static int64_t last_cpu_exec_time;
+ int64_t cpu_exec_time;
+ int64_t delta;
+
+ cpu_exec_time = tcg_cpu_exec_time();
+ delta = cpu_exec_time - last_cpu_exec_time;
+
monitor_printf(mon, "async time %" PRId64 " (%0.3f)\n",
dev_time, dev_time / (double)NANOSECONDS_PER_SECOND);
monitor_printf(mon, "qemu time %" PRId64 " (%0.3f)\n",
- tcg_time, tcg_time / (double)NANOSECONDS_PER_SECOND);
- tcg_time = 0;
+ delta, delta / (double)NANOSECONDS_PER_SECOND);
+ last_cpu_exec_time = cpu_exec_time;
dev_time = 0;
}
#else
/* Define to jump the ELF file used to communicate with GDB. */
#undef DEBUG_JIT
+#include "qemu/error-report.h"
#include "qemu/cutils.h"
#include "qemu/host-utils.h"
#include "qemu/timer.h"
const TCGProfile *orig = &s->prof;
if (counters) {
+ PROF_ADD(prof, orig, cpu_exec_time);
PROF_ADD(prof, orig, tb_count1);
PROF_ADD(prof, orig, tb_count);
PROF_ADD(prof, orig, op_count);
prof.table_op_count[i]);
}
}
+
+int64_t tcg_cpu_exec_time(void)
+{
+ unsigned int n_ctxs = atomic_read(&n_tcg_ctxs);
+ unsigned int i;
+ int64_t ret = 0;
+
+ for (i = 0; i < n_ctxs; i++) {
+ const TCGContext *s = atomic_read(&tcg_ctxs[i]);
+ const TCGProfile *prof = &s->prof;
+
+ ret += atomic_read(&prof->cpu_exec_time);
+ }
+ return ret;
+}
#else
void tcg_dump_op_count(FILE *f, fprintf_function cpu_fprintf)
{
cpu_fprintf(f, "[TCG profiler not compiled]\n");
}
+
+int64_t tcg_cpu_exec_time(void)
+{
+ error_report("%s: TCG profiler not compiled", __func__);
+ exit(EXIT_FAILURE);
+}
#endif
QEMU_BUILD_BUG_ON(NB_OPS > (1 << 8));
typedef struct TCGProfile {
+ int64_t cpu_exec_time;
int64_t tb_count1;
int64_t tb_count;
int64_t op_count; /* total insn count */
#define tcg_check_temp_count() 0
#endif
+int64_t tcg_cpu_exec_time(void);
void tcg_dump_info(FILE *f, fprintf_function cpu_fprintf);
void tcg_dump_op_count(FILE *f, fprintf_function cpu_fprintf);