static int SetThreadAffinity( pthread_t *pt, affinity_type_t at, ...)
{
int i, res, numcpus;
cpu_set_t cpuset;
res = pthread_getaffinity_np( *pt, sizeof(cpu_set_t), &cpuset);
if (res != 0) {
return 1;
}
numcpus = CPU_COUNT(&cpuset);
switch(at) {
case MASKMOD2:
CPU_ZERO(&cpuset);
for( i=0; i<numcpus; i+=2) {
CPU_SET( i, &cpuset);
}
break;
case STRICTLYFIRST:
if( numcpus > 1) {
CPU_ZERO(&cpuset);
CPU_SET(0, &cpuset);
}
break;
case ALLBUTFIRST:
if( numcpus > 1) {
CPU_CLR(0, &cpuset);
}
break;
case MAXCORES: {
int j, num_cores;
va_list args;
va_start( args, at);
num_cores = va_arg( args, int);
va_end( args);
for( j=num_cores; j<numcpus; j++) {
CPU_CLR( j, &cpuset);
}
}
break;
default:
break;
}
/* set the affinity mask */
res = pthread_setaffinity_np( *pt, sizeof(cpu_set_t), &cpuset);
if( res != 0) {
return 1;
}
return 0;
}
static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t *maskp)
{
int i, cpu = -1, nrcpus = 1024;
realloc:
CPU_ZERO(maskp);
if (sched_getaffinity(pid, sizeof(*maskp), maskp) == -1) {
if (errno == EINVAL && nrcpus < (1024 << 8)) {
nrcpus = nrcpus << 2;
goto realloc;
}
perror("sched_getaffinity");
return -1;
}
for (i = 0; i < nrcpus; i++) {
if (CPU_ISSET(i, maskp)) {
if (cpu == -1)
cpu = i;
else
CPU_CLR(i, maskp);
}
}
return cpu;
}
void main()
{
cpu_set_t set;
int ret,i;
//Zero out the struct
CPU_ZERO(&set);
CPU_SET(0, &set); //allow CPU 0
CPU_CLR(1, &set); //disallow CPU 1. This is just for completion. All CPUS were zeroed out by CPU_ZERO
//Set the processor affinity for this process
ret = sched_setaffinity(0, sizeof(cpu_set_t), &set);
if(ret == -1)
perror("sched_setaffinity");
/*
* CPU_SETSIZE is the number of processors that can be represented by cpu_set_t
*/
for(i=0; i < CPU_SETSIZE; i++)
{
int cpu;
cpu = CPU_ISSET(i, &set);
if(cpu)
printf("cpu=%i is %s\n", i, cpu ? "set" : "unset");
}
}
int
acpi_sleep_machdep(struct acpi_softc *sc, int state)
{
ACPI_STATUS status;
if (sc->acpi_wakeaddr == 0ul)
return (-1); /* couldn't alloc wake memory */
#ifdef SMP
suspcpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &suspcpus);
#endif
if (acpi_resume_beep != 0)
timer_spkr_acquire();
AcpiSetFirmwareWakingVector(WAKECODE_PADDR(sc));
intr_suspend();
if (savectx(susppcbs[0])) {
fpususpend(suspfpusave[0]);
#ifdef SMP
if (!CPU_EMPTY(&suspcpus) &&
suspend_cpus(suspcpus) == 0) {
device_printf(sc->acpi_dev, "Failed to suspend APs\n");
return (0); /* couldn't sleep */
}
#endif
WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[0]);
WAKECODE_FIXUP(wakeup_fpusave, void *, suspfpusave[0]);
WAKECODE_FIXUP(wakeup_gdt, uint16_t,
susppcbs[0]->pcb_gdt.rd_limit);
WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
susppcbs[0]->pcb_gdt.rd_base);
WAKECODE_FIXUP(wakeup_cpu, int, 0);
/* Call ACPICA to enter the desired sleep state */
if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
status = AcpiEnterSleepStateS4bios();
else
status = AcpiEnterSleepState(state);
if (status != AE_OK) {
device_printf(sc->acpi_dev,
"AcpiEnterSleepState failed - %s\n",
AcpiFormatException(status));
return (0); /* couldn't sleep */
}
for (;;)
ia32_pause();
}
return (1); /* wakeup successfully */
}
RTDECL(int) RTMpOnOthers(PFNRTMPWORKER pfnWorker, void *pvUser1, void *pvUser2)
{
/* Will panic if no rendezvousing cpus, so check up front. */
if (RTMpGetOnlineCount() > 1)
{
#if __FreeBSD_version >= 900000
cpuset_t Mask;
#elif __FreeBSD_version >= 700000
cpumask_t Mask;
#endif
RTMPARGS Args;
Args.pfnWorker = pfnWorker;
Args.pvUser1 = pvUser1;
Args.pvUser2 = pvUser2;
Args.idCpu = RTMpCpuId();
Args.cHits = 0;
#if __FreeBSD_version >= 700000
# if __FreeBSD_version >= 900000
Mask = all_cpus;
CPU_CLR(curcpu, &Mask);
# else
Mask = ~(cpumask_t)curcpu;
# endif
smp_rendezvous_cpus(Mask, NULL, rtmpOnOthersFreeBSDWrapper, smp_no_rendezvous_barrier, &Args);
#else
smp_rendezvous(NULL, rtmpOnOthersFreeBSDWrapper, NULL, &Args);
#endif
}
return VINF_SUCCESS;
}
/* The job has specialized cores, synchronize user mask with available cores */
static void _validate_mask(uint32_t task_id, hwloc_obj_t obj, cpu_set_t *ts)
{
int i, j, overlaps = 0;
bool superset = true;
for (i = 0; i < CPU_SETSIZE; i++) {
if (!CPU_ISSET(i, ts))
continue;
j = hwloc_bitmap_isset(obj->allowed_cpuset, i);
if (j > 0) {
overlaps++;
} else if (j == 0) {
CPU_CLR(i, ts);
superset = false;
}
}
if (overlaps == 0) {
/* The task's cpu map is completely invalid.
* Give it all allowed CPUs */
for (i = 0; i < CPU_SETSIZE; i++) {
if (hwloc_bitmap_isset(obj->allowed_cpuset, i) > 0)
CPU_SET(i, ts);
}
}
if (!superset) {
info("task/cgroup: Ignoring user CPU binding outside of job "
"step allocation for task[%u]", task_id);
fprintf(stderr, "Requested cpu_bind option outside of job "
"step allocation for task[%u]\n", task_id);
}
}
int
main(int argc, char *argv[])
{
int s, j, nprocs;
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
nprocs = sysconf(_SC_NPROCESSORS_ONLN);
/* Set affinity mask to include CPUs 0 to 7 */
CPU_ZERO(&cpuset);
for (j = 0; j < nprocs; j++)
CPU_SET(j, &cpuset);
CPU_CLR(1, &cpuset);
CPU_CLR(2, &cpuset);
CPU_CLR(3, &cpuset);
CPU_CLR(4, &cpuset);
CPU_CLR(5, &cpuset);
/* check if the cpu's have actually been set */
for (j = 0; j < nprocs; j++)
fprintf(stdout, "CPU: %d, status: %d\n", j, CPU_ISSET(j, &cpuset));
s = pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
if (s != 0)
handle_error_en(s, "pthread_setaffinity_np");
/* Check the actual affinity mask assigned to the thread */
s = pthread_getaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
if (s != 0)
handle_error_en(s, "pthread_getaffinity_np");
printf("Set returned by pthread_getaffinity_np() contained:\n");
for (j = 0; j < CPU_SETSIZE; j++)
if (CPU_ISSET(j, &cpuset))
printf(" CPU %d\n", j);
exit(EXIT_SUCCESS);
}
static void
smp_targeted_tlb_shootdown(cpuset_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2)
{
int cpu, ncpu, othercpus;
struct _call_data data;
othercpus = mp_ncpus - 1;
if (CPU_ISFULLSET(&mask)) {
if (othercpus < 1)
return;
} else {
CPU_CLR(PCPU_GET(cpuid), &mask);
if (CPU_EMPTY(&mask))
return;
}
if (!(read_eflags() & PSL_I))
panic("%s: interrupts disabled", __func__);
mtx_lock_spin(&smp_ipi_mtx);
KASSERT(call_data == NULL, ("call_data isn't null?!"));
call_data = &data;
call_data->func_id = vector;
call_data->arg1 = addr1;
call_data->arg2 = addr2;
atomic_store_rel_int(&smp_tlb_wait, 0);
if (CPU_ISFULLSET(&mask)) {
ncpu = othercpus;
ipi_all_but_self(vector);
} else {
ncpu = 0;
while ((cpu = cpusetobj_ffs(&mask)) != 0) {
cpu--;
CPU_CLR(cpu, &mask);
CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu,
vector);
ipi_send_cpu(cpu, vector);
ncpu++;
}
}
while (smp_tlb_wait < ncpu)
ia32_pause();
call_data = NULL;
mtx_unlock_spin(&smp_ipi_mtx);
}
static int cpu_enable(cpu_set_t *cpu_set, size_t setsize, int enable)
{
unsigned int cpu;
int online, rc;
int configured = -1;
for (cpu = 0; cpu < setsize; cpu++) {
if (!CPU_ISSET(cpu, cpu_set))
continue;
if (!path_exist(_PATH_SYS_CPU "/cpu%d", cpu)) {
printf(_("CPU %d does not exist\n"), cpu);
continue;
}
if (!path_exist(_PATH_SYS_CPU "/cpu%d/online", cpu)) {
printf(_("CPU %d is not hot pluggable\n"), cpu);
continue;
}
online = path_getnum(_PATH_SYS_CPU "/cpu%d/online", cpu);
if ((online == 1) && (enable == 1)) {
printf(_("CPU %d is already enabled\n"), cpu);
continue;
}
if ((online == 0) && (enable == 0)) {
printf(_("CPU %d is already disabled\n"), cpu);
continue;
}
if (path_exist(_PATH_SYS_CPU "/cpu%d/configure", cpu))
configured = path_getnum(_PATH_SYS_CPU "/cpu%d/configure", cpu);
if (enable) {
rc = path_writestr("1", _PATH_SYS_CPU "/cpu%d/online", cpu);
if ((rc == -1) && (configured == 0))
printf(_("CPU %d enable failed "
"(CPU is deconfigured)\n"), cpu);
else if (rc == -1)
printf(_("CPU %d enable failed (%m)\n"), cpu);
else
printf(_("CPU %d enabled\n"), cpu);
} else {
if (onlinecpus && num_online_cpus() == 1) {
printf(_("CPU %d disable failed "
"(last enabled CPU)\n"), cpu);
continue;
}
rc = path_writestr("0", _PATH_SYS_CPU "/cpu%d/online", cpu);
if (rc == -1)
printf(_("CPU %d disable failed (%m)\n"), cpu);
else {
printf(_("CPU %d disabled\n"), cpu);
if (onlinecpus)
CPU_CLR(cpu, onlinecpus);
}
}
}
return EXIT_SUCCESS;
}
// 프로세스의 affinity를 get/set함.
void affinity_control()
{
// CPU affinity 설정 구조체
cpu_set_t set;
// 구조체의 각 CPU값을 모두 reset하는 매크
CPU_ZERO(&set);
// 현재 프로세스에 대해 조회함.
if(sched_getaffinity(0, sizeof(cpu_set_t), &set) == -1)
errexit("sched_getaffinity");
// affinity 설정값 확인
int i;
// CPU_SETSIZE = 1024
for (i = 0; i < CPU_SETSIZE; ++i) {
int cpu;
// i번 CPU가 set상태인지 확인하는 매크로
cpu = CPU_ISSET(i, &set);
printf("CPU #%d = %d\n", i, cpu);
}
// affinity 설정
CPU_ZERO(&set);
// CPU 0번, 1번을 set한다.
CPU_SET(0, &set);
CPU_SET(1, &set);
// CPU 2번, 3번을 clear한다.
CPU_CLR(2, &set);
CPU_CLR(3, &set);
// 현재 프로세스에 대해 affinity를 set에 설정한 값대로 설정한다.
if(sched_setaffinity(0, sizeof(cpu_set_t), &set) == -1)
errexit("sched_getaffinity");
for (i = 0; i < CPU_SETSIZE; ++i) {
int cpu;
// i번 CPU가 set상태인지 확인하는 매크로
cpu = CPU_ISSET(i, &set);
printf("CPU #%d = %d\n", i, cpu);
}
}
mt::LinuxCPUAffinityThreadInitializer::
LinuxCPUAffinityThreadInitializer(const cpu_set_t& cpu)
{
for (int i = 0; i < CPU_SETSIZE; ++i)
{
CPU_CLR(i, &mCPU);
if (CPU_ISSET(i, &cpu))
CPU_SET(i, &mCPU);
}
}
/**
* Combines the two given bitsets, destination and source, storing their logical AND result into destination
* @param destination[in/out] One source and a destination of the logical AND
* @param source[in] The other source of the logical AND
*/
static void
cpuset_logical_and(cpu_set_t *destination, const cpu_set_t *source)
{
uintptr_t i = 0;
for (i = 0; i < CPU_SETSIZE; i++) {
if (!CPU_ISSET(i, source) && CPU_ISSET(i, destination)) {
CPU_CLR(i, destination);
}
}
}
void
kdb_panic(const char *msg)
{
#ifdef SMP
cpuset_t other_cpus;
other_cpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &other_cpus);
stop_cpus_hard(other_cpus);
#endif
printf("KDB: panic\n");
panic("%s", msg);
}
void *
start_server_thread_func (void *ptr)
{
client_thread = false;
client_t *client = (client_t *)ptr;
server_t *server = server_new (&client->buffer);
server->server_signal = &client->server_signal;
server->client_signal = &client->client_signal;
mutex_unlock (client->server_started_mutex);
prctl (PR_SET_TIMERSLACK, 1);
pthread_t id = pthread_self ();
int online_cpus = sysconf (_SC_NPROCESSORS_ONLN);
int available_cpus = sysconf (_SC_NPROCESSORS_CONF);
if (online_cpus > 1) {
cpu_set_t cpu_set;
CPU_ZERO (&cpu_set);
if (pthread_getaffinity_np (id, sizeof (cpu_set_t), &cpu_set) == 0) {
/* find first cpu to run on */
int cpu = 0;
int i;
for (i = 1; i < available_cpus; i++) {
if (CPU_ISSET (i, &cpu_set)) {
cpu = i;
break;
}
}
/* force server to run on cpu1 */
if (cpu == 0)
cpu = 1;
if (cpu != 0) {
for (i = 0; i < available_cpus; i++) {
if (i != cpu)
CPU_CLR (i, &cpu_set);
}
CPU_SET (cpu, &cpu_set);
pthread_setaffinity_np (id, sizeof (cpu_set_t), &cpu_set);
}
}
}
server_start_work_loop (server);
server_destroy(server);
return NULL;
}
void
cpu_reset()
{
#ifdef SMP
cpuset_t map;
u_int cnt;
if (smp_started) {
map = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &map);
CPU_NAND(&map, &stopped_cpus);
if (!CPU_EMPTY(&map)) {
printf("cpu_reset: Stopping other CPUs\n");
stop_cpus(map);
}
if (PCPU_GET(cpuid) != 0) {
cpu_reset_proxyid = PCPU_GET(cpuid);
cpustop_restartfunc = cpu_reset_proxy;
cpu_reset_proxy_active = 0;
printf("cpu_reset: Restarting BSP\n");
/* Restart CPU #0. */
CPU_SETOF(0, &started_cpus);
wmb();
cnt = 0;
while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
ia32_pause();
cnt++; /* Wait for BSP to announce restart */
}
if (cpu_reset_proxy_active == 0)
printf("cpu_reset: Failed to restart BSP\n");
enable_intr();
cpu_reset_proxy_active = 2;
while (1)
ia32_pause();
/* NOTREACHED */
}
DELAY(1000000);
}
#endif
cpu_reset_real();
/* NOTREACHED */
}
/* The job has specialized cores, synchronize user mask with available cores */
static void _validate_mask(launch_tasks_request_msg_t *req, char *avail_mask)
{
char *new_mask = NULL, *save_ptr = NULL, *tok;
cpu_set_t avail_cpus, task_cpus;
bool superset = true;
CPU_ZERO(&avail_cpus);
(void) str_to_cpuset(&avail_cpus, avail_mask);
tok = strtok_r(req->cpu_bind, ",", &save_ptr);
while (tok) {
int i, overlaps = 0;
char mask_str[1 + CPU_SETSIZE / 4];
CPU_ZERO(&task_cpus);
(void) str_to_cpuset(&task_cpus, tok);
for (i = 0; i < CPU_SETSIZE; i++) {
if (!CPU_ISSET(i, &task_cpus))
continue;
if (CPU_ISSET(i, &avail_cpus)) {
overlaps++;
} else {
CPU_CLR(i, &task_cpus);
superset = false;
}
}
if (overlaps == 0) {
/* The task's CPU mask is completely invalid.
* Give it all allowed CPUs. */
for (i = 0; i < CPU_SETSIZE; i++) {
if (CPU_ISSET(i, &avail_cpus))
CPU_SET(i, &task_cpus);
}
}
cpuset_to_str(&task_cpus, mask_str);
if (new_mask)
xstrcat(new_mask, ",");
xstrcat(new_mask, mask_str);
tok = strtok_r(NULL, ",", &save_ptr);
}
if (!superset) {
info("task/affinity: Ignoring user CPU binding outside of job "
"step allocation");
}
xfree(req->cpu_bind);
req->cpu_bind = new_mask;
}
int main(int argc, char *argv[])
{
QCoreApplication app(argc, argv);
qRegisterMetaType< struct timespec >();
if( Tuning::DoMemLock ) {
// Prevent this process from being swapped out (technically this is
// moot since we don't use a swapfile on the Pi)
mlockall(MCL_CURRENT | MCL_FUTURE);
}
if( Tuning::DoRTLimit ){
struct rlimit rlimits;
getrlimit(RLIMIT_RTTIME, &rlimits );
qDebug() << "Current RLIMIT_RTTIME = " << rlimits.rlim_cur;
rlimits.rlim_cur = Tuning::RTLimit;
if( setrlimit(RLIMIT_RTTIME, &rlimits ) ) {
qDebug() << "Error setting RLIMIT_RTTIME (" << errno << "): " << strerror(errno);
}
}
// Mask out CPU affinity for all processes
if( Tuning::DoCpuAffinity ) {
cpu_set_t cpu_set;
if( sched_getaffinity(0, sizeof(cpu_set_t), &cpu_set ) != 0 ) {
qDebug() << "Unable to get CPU affinity main thread (" << errno << "): " << strerror(errno);
}
CPU_CLR( Tuning::CriticalCpu, &cpu_set );
if( sched_setaffinity(0, sizeof(cpu_set_t), &cpu_set ) != 0 ) {
qDebug() << "Unable to set CPU affinity for main thread (" << errno << "): " << strerror(errno);
}
sched_getaffinity(0, sizeof(cpu_set_t), &cpu_set );
qDebug() << "Main thread runs on " << (CPU_COUNT( &cpu_set )) << " cpus";
}
setScheduler( Tuning::DoFIFOScheduler, false );
MainThread *mt = new MainThread( &app );
return app.exec();
}
int set_cpu_affinity(const char *str, int inverted)
{
int ret, i, cpus;
const char *p, *q;
cpu_set_t cpu_bitmask;
q = str;
cpus = get_number_cpus();
CPU_ZERO(&cpu_bitmask);
for (i = 0; inverted && i < cpus; ++i)
CPU_SET(i, &cpu_bitmask);
while (p = q, q = next_token(q, ','), p) {
unsigned int a; /* Beginning of range */
unsigned int b; /* End of range */
unsigned int s; /* Stride */
const char *c1, *c2;
if (sscanf(p, "%u", &a) < 1)
return -EINVAL;
b = a;
s = 1;
c1 = next_token(p, '-');
c2 = next_token(p, ',');
if (c1 != NULL && (c2 == NULL || c1 < c2)) {
if (sscanf(c1, "%u", &b) < 1)
return -EINVAL;
c1 = next_token(c1, ':');
if (c1 != NULL && (c2 == NULL || c1 < c2))
if (sscanf(c1, "%u", &s) < 1)
return -EINVAL;
}
if (!(a <= b))
return -EINVAL;
while (a <= b) {
if (inverted)
CPU_CLR(a, &cpu_bitmask);
else
CPU_SET(a, &cpu_bitmask);
a += s;
}
}
ret = sched_setaffinity(getpid(), sizeof(cpu_bitmask),
&cpu_bitmask);
if (ret)
panic("Can't set this cpu affinity!\n");
return 0;
}
static int
do_test (void)
{
cpu_set_t cs;
if (sched_getaffinity (getpid (), sizeof (cs), &cs) != 0)
{
printf ("getaffinity failed: %m\n");
return 1;
}
int result = 0;
int cpu = 0;
while (CPU_COUNT (&cs) != 0)
{
if (CPU_ISSET (cpu, &cs))
{
cpu_set_t cs2;
CPU_ZERO (&cs2);
CPU_SET (cpu, &cs2);
if (sched_setaffinity (getpid (), sizeof (cs2), &cs2) != 0)
{
printf ("setaffinity(%d) failed: %m\n", cpu);
result = 1;
}
else
{
int cpu2 = sched_getcpu ();
if (cpu2 == -1 && errno == ENOSYS)
{
puts ("getcpu syscall not implemented");
return 0;
}
if (cpu2 != cpu)
{
printf ("getcpu results %d not possible\n", cpu2);
result = 1;
}
}
CPU_CLR (cpu, &cs);
}
++cpu;
}
return result;
}
/*
* send an IPI to all CPUs EXCEPT myself
*/
void
ipi_all_but_self(u_int ipi)
{
cpuset_t other_cpus;
/*
* IPI_STOP_HARD maps to a NMI and the trap handler needs a bit
* of help in order to understand what is the source.
* Set the mask of receiving CPUs for this purpose.
*/
other_cpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &other_cpus);
if (ipi == IPI_STOP_HARD)
CPU_OR_ATOMIC(&ipi_nmi_pending, &other_cpus);
CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi);
ipi_selected(other_cpus, ipi);
}
static void test_cpu_clr_case_1(size_t cpu)
{
size_t i;
/*
* Set to all zeros and verify
*/
printf( "Exercise CPU_FILL, CPU_CLR(%u), and CPU_ISET\n", cpu );
CPU_FILL(&set1);
CPU_CLR(cpu, &set1);
/* test if all bits except 5 are set */
for (i=0 ; i<CPU_SETSIZE ; i++) {
if (i==cpu)
rtems_test_assert( CPU_ISSET(i, &set1) == 0 );
else
rtems_test_assert( CPU_ISSET(i, &set1) == 1 );
}
}
/*
* send an IPI to a set of cpus.
*/
void
ipi_selected(cpuset_t cpus, u_int ipi)
{
int cpu;
/*
* IPI_STOP_HARD maps to a NMI and the trap handler needs a bit
* of help in order to understand what is the source.
* Set the mask of receiving CPUs for this purpose.
*/
if (ipi == IPI_STOP_HARD)
CPU_OR_ATOMIC(&ipi_nmi_pending, &cpus);
while ((cpu = cpusetobj_ffs(&cpus)) != 0) {
cpu--;
CPU_CLR(cpu, &cpus);
CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi);
ipi_send_cpu(cpu, ipi);
}
}
int main(int argc, char *argv[])
{
pid_t pid;
if (argc < 3) {
printf("%s <PID> <CPU_AFFINITY>\ne.g. %s 9527 0110\n", argv[0], argv[0]);
return 0;
}
sscanf(argv[1], "%d", &pid);
int i, cpu_num = get_cpu_num();
if (cpu_num < 0) {
printf("can't get cpu number\n");
return 1;
}
cpu_set_t mask;
unsigned int len = sizeof(mask);
CPU_ZERO(&mask);
for (i = 0; i < cpu_num && argv[2][i] != '\0'; i++) {
if (argv[2][i] == '1') {
CPU_SET(i, &mask);
}
else if (argv[2][i] == '0') {
CPU_CLR(i, &mask);
}
else {
printf("bad cpu_affinity, only 0/1 is allowed\n");
return 2;
}
}
if (sched_setaffinity(pid, len, &mask) < 0) {
perror("sched_setaffinity");
return -1;
}
printf("PID[%d] set to [%s] ok\n", pid, argv[2]);
return 0;
}
int main( int argc, char *argv[])
{
cpu_set_t set;
int ret, i;
CPU_ZERO(&set);
ret = sched_getaffinity(0, sizeof(cpu_set_t), &set);
if( ret == -1)
perror("sched_getaffinity");
print_status(&set);
printf("Now, let's trying to unset CPU #1\n");
CPU_CLR(1, &set);
ret = sched_setaffinity(0, sizeof(cpu_set_t), &set);
if( ret == -1)
perror("sched_setaffinity");
print_status(&set);
return 0;
}
static void PtyReader_28979140(PtyReader_28979140_Arg* arg) {
arg->finished = false;
cpu_set_t cpus;
ASSERT_EQ(0, sched_getaffinity(0, sizeof(cpu_set_t), &cpus));
CPU_CLR(arg->main_cpu_id, &cpus);
ASSERT_EQ(0, sched_setaffinity(0, sizeof(cpu_set_t), &cpus));
uint32_t counter = 0;
while (counter <= arg->data_count) {
char buf[4096]; // Use big buffer to read to hit the bug more easily.
size_t to_read = std::min(sizeof(buf), (arg->data_count + 1 - counter) * sizeof(uint32_t));
ASSERT_TRUE(android::base::ReadFully(arg->slave_fd, buf, to_read));
size_t num_of_value = to_read / sizeof(uint32_t);
uint32_t* p = reinterpret_cast<uint32_t*>(buf);
while (num_of_value-- > 0) {
if (*p++ != counter++) {
arg->matched = false;
}
}
}
close(arg->slave_fd);
arg->finished = true;
}
int test__openat_syscall_event_on_all_cpus(int subtest __maybe_unused)
{
int err = -1, fd, cpu;
struct cpu_map *cpus;
struct perf_evsel *evsel;
unsigned int nr_openat_calls = 111, i;
cpu_set_t cpu_set;
struct thread_map *threads = thread_map__new(-1, getpid(), UINT_MAX);
char sbuf[STRERR_BUFSIZE];
char errbuf[BUFSIZ];
if (threads == NULL) {
pr_debug("thread_map__new\n");
return -1;
}
cpus = cpu_map__new(NULL);
if (cpus == NULL) {
pr_debug("cpu_map__new\n");
goto out_thread_map_delete;
}
CPU_ZERO(&cpu_set);
evsel = perf_evsel__newtp("syscalls", "sys_enter_openat");
if (IS_ERR(evsel)) {
tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "syscalls", "sys_enter_openat");
pr_debug("%s\n", errbuf);
goto out_thread_map_delete;
}
if (perf_evsel__open(evsel, cpus, threads) < 0) {
pr_debug("failed to open counter: %s, "
"tweak /proc/sys/kernel/perf_event_paranoid?\n",
str_error_r(errno, sbuf, sizeof(sbuf)));
goto out_evsel_delete;
}
for (cpu = 0; cpu < cpus->nr; ++cpu) {
unsigned int ncalls = nr_openat_calls + cpu;
/*
* XXX eventually lift this restriction in a way that
* keeps perf building on older glibc installations
* without CPU_ALLOC. 1024 cpus in 2010 still seems
* a reasonable upper limit tho :-)
*/
if (cpus->map[cpu] >= CPU_SETSIZE) {
pr_debug("Ignoring CPU %d\n", cpus->map[cpu]);
continue;
}
CPU_SET(cpus->map[cpu], &cpu_set);
if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) {
pr_debug("sched_setaffinity() failed on CPU %d: %s ",
cpus->map[cpu],
str_error_r(errno, sbuf, sizeof(sbuf)));
goto out_close_fd;
}
for (i = 0; i < ncalls; ++i) {
fd = openat(0, "/etc/passwd", O_RDONLY);
close(fd);
}
CPU_CLR(cpus->map[cpu], &cpu_set);
}
/*
* Here we need to explicitly preallocate the counts, as if
* we use the auto allocation it will allocate just for 1 cpu,
* as we start by cpu 0.
*/
if (perf_evsel__alloc_counts(evsel, cpus->nr, 1) < 0) {
pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus->nr);
goto out_close_fd;
}
err = 0;
for (cpu = 0; cpu < cpus->nr; ++cpu) {
unsigned int expected;
if (cpus->map[cpu] >= CPU_SETSIZE)
continue;
if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) {
pr_debug("perf_evsel__read_on_cpu\n");
err = -1;
break;
}
expected = nr_openat_calls + cpu;
if (perf_counts(evsel->counts, cpu, 0)->val != expected) {
pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %" PRIu64 "\n",
expected, cpus->map[cpu], perf_counts(evsel->counts, cpu, 0)->val);
err = -1;
}
}
perf_evsel__free_counts(evsel);
out_close_fd:
perf_evsel__close_fd(evsel, 1, threads->nr);
out_evsel_delete:
perf_evsel__delete(evsel);
out_thread_map_delete:
thread_map__put(threads);
return err;
}
/*
* stress_tlb_shootdown()
* stress out TLB shootdowns
*/
static int stress_tlb_shootdown(const args_t *args)
{
const size_t page_size = args->page_size;
const size_t mmap_size = page_size * MMAP_PAGES;
pid_t pids[MAX_TLB_PROCS];
cpu_set_t proc_mask_initial;
if (sched_getaffinity(0, sizeof(proc_mask_initial), &proc_mask_initial) < 0) {
pr_fail_err("could not get CPU affinity");
return EXIT_FAILURE;
}
do {
uint8_t *mem, *ptr;
int retry = 128;
cpu_set_t proc_mask;
int32_t tlb_procs, i;
const int32_t max_cpus = stress_get_processors_configured();
CPU_ZERO(&proc_mask);
CPU_OR(&proc_mask, &proc_mask_initial, &proc_mask);
tlb_procs = max_cpus;
if (tlb_procs > MAX_TLB_PROCS)
tlb_procs = MAX_TLB_PROCS;
if (tlb_procs < MIN_TLB_PROCS)
tlb_procs = MIN_TLB_PROCS;
for (;;) {
mem = mmap(NULL, mmap_size, PROT_WRITE | PROT_READ,
MAP_SHARED | MAP_ANONYMOUS, -1, 0);
if ((void *)mem == MAP_FAILED) {
if ((errno == EAGAIN) ||
(errno == ENOMEM) ||
(errno == ENFILE)) {
if (--retry < 0)
return EXIT_NO_RESOURCE;
} else {
pr_fail_err("mmap");
}
} else {
break;
}
}
(void)memset(mem, 0, mmap_size);
for (i = 0; i < tlb_procs; i++)
pids[i] = -1;
for (i = 0; i < tlb_procs; i++) {
int32_t j, cpu = -1;
for (j = 0; j < max_cpus; j++) {
if (CPU_ISSET(j, &proc_mask)) {
cpu = j;
CPU_CLR(j, &proc_mask);
break;
}
}
if (cpu == -1)
break;
pids[i] = fork();
if (pids[i] < 0)
break;
if (pids[i] == 0) {
cpu_set_t mask;
char buffer[page_size];
(void)setpgid(0, g_pgrp);
stress_parent_died_alarm();
/* Make sure this is killable by OOM killer */
set_oom_adjustment(args->name, true);
CPU_ZERO(&mask);
CPU_SET(cpu % max_cpus, &mask);
(void)sched_setaffinity(args->pid, sizeof(mask), &mask);
for (ptr = mem; ptr < mem + mmap_size; ptr += page_size) {
/* Force tlb shoot down on page */
(void)mprotect(ptr, page_size, PROT_READ);
(void)memcpy(buffer, ptr, page_size);
(void)munmap(ptr, page_size);
}
_exit(0);
}
}
for (i = 0; i < tlb_procs; i++) {
if (pids[i] != -1) {
int status, ret;
ret = shim_waitpid(pids[i], &status, 0);
if ((ret < 0) && (errno == EINTR)) {
int j;
/*
* We got interrupted, so assume
* it was the alarm (timedout) or
* SIGINT so force terminate
*/
for (j = i; j < tlb_procs; j++) {
if (pids[j] != -1)
(void)kill(pids[j], SIGKILL);
}
/* re-wait on the failed wait */
(void)shim_waitpid(pids[i], &status, 0);
/* and continue waitpid on the pids */
}
}
}
(void)munmap(mem, mmap_size);
(void)sched_setaffinity(0, sizeof(proc_mask_initial), &proc_mask_initial);
inc_counter(args);
} while (keep_stressing());
return EXIT_SUCCESS;
}
/* The main CPU accumulator thread */
void guppi_accum_thread(void *_args) {
float **accumulator; //indexed accumulator[accum_id][chan][subband][stokes]
char accum_dirty[NUM_SW_STATES];
struct sdfits_data_columns data_cols[NUM_SW_STATES];
int payload_type;
int i, j, k, rv;
/* Get arguments */
struct guppi_thread_args *args = (struct guppi_thread_args *)_args;
/* Set cpu affinity */
cpu_set_t cpuset, cpuset_orig;
sched_getaffinity(0, sizeof(cpu_set_t), &cpuset_orig);
//CPU_ZERO(&cpuset);
CPU_CLR(13, &cpuset);
CPU_SET(9, &cpuset);
rv = sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
if (rv<0) {
guppi_error("guppi_accum_thread", "Error setting cpu affinity.");
perror("sched_setaffinity");
}
/* Set priority */
rv = setpriority(PRIO_PROCESS, 0, args->priority);
if (rv<0) {
guppi_error("guppi_accum_thread", "Error setting priority level.");
perror("set_priority");
}
/* Attach to status shared mem area */
struct guppi_status st;
rv = guppi_status_attach(&st);
if (rv!=GUPPI_OK) {
guppi_error("guppi_accum_thread",
"Error attaching to status shared memory.");
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_status_detach, &st);
pthread_cleanup_push((void *)set_exit_status, &st);
pthread_cleanup_push((void *)guppi_thread_set_finished, args);
/* Init status */
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "init");
guppi_status_unlock_safe(&st);
/* Read in general parameters */
struct guppi_params gp;
struct sdfits sf;
pthread_cleanup_push((void *)guppi_free_sdfits, &sf);
/* Attach to databuf shared mem */
struct guppi_databuf *db_in, *db_out;
db_in = guppi_databuf_attach(args->input_buffer);
char errmsg[256];
if (db_in==NULL) {
sprintf(errmsg,
"Error attaching to input databuf(%d) shared memory.",
args->input_buffer);
guppi_error("guppi_accum_thread", errmsg);
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_databuf_detach, db_in);
db_out = guppi_databuf_attach(args->output_buffer);
if (db_out==NULL) {
sprintf(errmsg,
"Error attaching to output databuf(%d) shared memory.",
args->output_buffer);
guppi_error("guppi_accum_thread", errmsg);
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_databuf_detach, db_out);
/* Determine high/low bandwidth mode */
char bw_mode[16];
if (hgets(st.buf, "BW_MODE", 16, bw_mode))
{
if(strncmp(bw_mode, "high", 4) == 0)
payload_type = INT_PAYLOAD;
else if(strncmp(bw_mode, "low", 3) == 0)
payload_type = FLOAT_PAYLOAD;
else
guppi_error("guppi_net_thread", "Unsupported bandwidth mode");
}
else
guppi_error("guppi_net_thread", "BW_MODE not set");
/* Read nchan and nsubband from status shared memory */
guppi_read_obs_params(st.buf, &gp, &sf);
/* Allocate memory for vector accumulators */
create_accumulators(&accumulator, sf.hdr.nchan, sf.hdr.nsubband);
pthread_cleanup_push((void *)destroy_accumulators, accumulator);
/* Clear the vector accumulators */
for(i = 0; i < NUM_SW_STATES; i++) accum_dirty[i] = 1;
reset_accumulators(accumulator, data_cols, accum_dirty, sf.hdr.nsubband, sf.hdr.nchan);
/* Loop */
int curblock_in=0, curblock_out=0;
int first=1;
float reqd_exposure=0;
double accum_time=0;
int integ_num;
float pfb_rate;
int heap, accumid, struct_offset, array_offset;
char *hdr_in=NULL, *hdr_out=NULL;
struct databuf_index *index_in, *index_out;
int nblock_int=0, npacket=0, n_pkt_drop=0, n_heap_drop=0;
signal(SIGINT,cc);
while (run) {
/* Note waiting status */
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "waiting");
guppi_status_unlock_safe(&st);
/* Wait for buf to have data */
rv = guppi_databuf_wait_filled(db_in, curblock_in);
if (rv!=0) continue;
/* Note waiting status and current block*/
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "accumulating");
hputi4(st.buf, "ACCBLKIN", curblock_in);
guppi_status_unlock_safe(&st);
/* Read param struct for this block */
hdr_in = guppi_databuf_header(db_in, curblock_in);
if (first)
guppi_read_obs_params(hdr_in, &gp, &sf);
else
guppi_read_subint_params(hdr_in, &gp, &sf);
/* Do any first time stuff: first time code runs, not first time process this block */
if (first) {
/* Set up first output header. This header is copied from block to block
each time a new block is created */
hdr_out = guppi_databuf_header(db_out, curblock_out);
memcpy(hdr_out, guppi_databuf_header(db_in, curblock_in),
GUPPI_STATUS_SIZE);
/* Read required exposure and PFB rate from status shared memory */
reqd_exposure = sf.data_columns.exposure;
pfb_rate = sf.hdr.efsampfr / (2 * sf.hdr.nchan);
/* Initialise the index in the output block */
index_out = (struct databuf_index*)guppi_databuf_index(db_out, curblock_out);
index_out->num_datasets = 0;
index_out->array_size = sf.hdr.nsubband * sf.hdr.nchan * NUM_STOKES * 4;
first=0;
}
/* Loop through each spectrum (heap) in input buffer */
index_in = (struct databuf_index*)guppi_databuf_index(db_in, curblock_in);
for(heap = 0; heap < index_in->num_heaps; heap++)
{
/* If invalid, record it and move to next heap */
if(!index_in->cpu_gpu_buf[heap].heap_valid)
{
n_heap_drop++;
continue;
}
/* Read in heap from buffer */
char* heap_addr = (char*)(guppi_databuf_data(db_in, curblock_in) +
sizeof(struct freq_spead_heap) * heap);
struct freq_spead_heap* freq_heap = (struct freq_spead_heap*)(heap_addr);
char* payload_addr = (char*)(guppi_databuf_data(db_in, curblock_in) +
sizeof(struct freq_spead_heap) * MAX_HEAPS_PER_BLK +
(index_in->heap_size - sizeof(struct freq_spead_heap)) * heap );
int *i_payload = (int*)(payload_addr);
float *f_payload = (float*)(payload_addr);
accumid = freq_heap->status_bits & 0x7;
/*Debug: print heap */
/* printf("%d, %d, %d, %d, %d, %d\n", freq_heap->time_cntr, freq_heap->spectrum_cntr,
freq_heap->integ_size, freq_heap->mode, freq_heap->status_bits,
freq_heap->payload_data_off);
*/
/* If we have accumulated for long enough, write vectors to output block */
if(accum_time >= reqd_exposure)
{
for(i = 0; i < NUM_SW_STATES; i++)
{
/*If a particular accumulator is dirty, write it to output buffer */
if(accum_dirty[i])
{
/*If insufficient space, first mark block as filled and request new block*/
index_out = (struct databuf_index*)(guppi_databuf_index(db_out, curblock_out));
if( (index_out->num_datasets+1) *
(index_out->array_size + sizeof(struct sdfits_data_columns)) >
db_out->block_size)
{
printf("Accumulator finished with output block %d\n", curblock_out);
/* Write block number to status buffer */
guppi_status_lock_safe(&st);
hputi4(st.buf, "ACCBLKOU", curblock_out);
guppi_status_unlock_safe(&st);
/* Update packet count and loss fields in output header */
hputi4(hdr_out, "NBLOCK", nblock_int);
hputi4(hdr_out, "NPKT", npacket);
hputi4(hdr_out, "NPKTDROP", n_pkt_drop);
hputi4(hdr_out, "NHPDROP", n_heap_drop);
/* Close out current integration */
guppi_databuf_set_filled(db_out, curblock_out);
/* Wait for next output buf */
curblock_out = (curblock_out + 1) % db_out->n_block;
guppi_databuf_wait_free(db_out, curblock_out);
while ((rv=guppi_databuf_wait_free(db_out, curblock_out)) != GUPPI_OK)
{
if (rv==GUPPI_TIMEOUT) {
guppi_warn("guppi_accum_thread", "timeout while waiting for output block");
continue;
} else {
guppi_error("guppi_accum_thread", "error waiting for free databuf");
run=0;
pthread_exit(NULL);
break;
}
}
hdr_out = guppi_databuf_header(db_out, curblock_out);
memcpy(hdr_out, guppi_databuf_header(db_in, curblock_in),
GUPPI_STATUS_SIZE);
/* Initialise the index in new output block */
index_out = (struct databuf_index*)guppi_databuf_index(db_out, curblock_out);
index_out->num_datasets = 0;
index_out->array_size = sf.hdr.nsubband * sf.hdr.nchan * NUM_STOKES * 4;
nblock_int=0;
npacket=0;
n_pkt_drop=0;
n_heap_drop=0;
}
/*Update index for output buffer*/
index_out = (struct databuf_index*)(guppi_databuf_index(db_out, curblock_out));
if(index_out->num_datasets == 0)
struct_offset = 0;
else
struct_offset = index_out->disk_buf[index_out->num_datasets-1].array_offset +
index_out->array_size;
array_offset = struct_offset + sizeof(struct sdfits_data_columns);
index_out->disk_buf[index_out->num_datasets].struct_offset = struct_offset;
index_out->disk_buf[index_out->num_datasets].array_offset = array_offset;
/*Copy sdfits_data_columns struct to disk buffer */
memcpy(guppi_databuf_data(db_out, curblock_out) + struct_offset,
&data_cols[i], sizeof(struct sdfits_data_columns));
/*Copy data array to disk buffer */
memcpy(guppi_databuf_data(db_out, curblock_out) + array_offset,
accumulator[i], index_out->array_size);
/*Update SDFITS data_columns pointer to data array */
((struct sdfits_data_columns*)
(guppi_databuf_data(db_out, curblock_out) + struct_offset))->data =
(unsigned char*)(guppi_databuf_data(db_out, curblock_out) + array_offset);
index_out->num_datasets = index_out->num_datasets + 1;
}
}
accum_time = 0;
integ_num += 1;
reset_accumulators(accumulator, data_cols, accum_dirty,
sf.hdr.nsubband, sf.hdr.nchan);
}
/* Only add spectrum to accumulator if blanking bit is low */
if((freq_heap->status_bits & 0x08) == 0)
{
/* Fill in data columns header fields */
if(!accum_dirty[accumid])
{
/*Record SPEAD header fields*/
data_cols[accumid].time = index_in->cpu_gpu_buf[heap].heap_rcvd_mjd;
data_cols[accumid].time_counter = freq_heap->time_cntr;
data_cols[accumid].integ_num = integ_num;
data_cols[accumid].sttspec = freq_heap->spectrum_cntr;
data_cols[accumid].accumid = accumid;
/* Fill in rest of fields from status buffer */
strcpy(data_cols[accumid].object, sf.data_columns.object);
data_cols[accumid].azimuth = sf.data_columns.azimuth;
data_cols[accumid].elevation = sf.data_columns.elevation;
data_cols[accumid].bmaj = sf.data_columns.bmaj;
data_cols[accumid].bmin = sf.data_columns.bmin;
data_cols[accumid].bpa = sf.data_columns.bpa;
data_cols[accumid].centre_freq_idx = sf.data_columns.centre_freq_idx;
data_cols[accumid].ra = sf.data_columns.ra;
data_cols[accumid].dec = sf.data_columns.dec;
data_cols[accumid].exposure = 0.0;
for(i = 0; i < NUM_SW_STATES; i++)
data_cols[accumid].centre_freq[i] = sf.data_columns.centre_freq[i];
accum_dirty[accumid] = 1;
}
data_cols[accumid].exposure += (float)(freq_heap->integ_size)/pfb_rate;
data_cols[accumid].stpspec = freq_heap->spectrum_cntr;
/* Add spectrum to appropriate vector accumulator (high-bw mode) */
if(payload_type == INT_PAYLOAD)
{
for(i = 0; i < sf.hdr.nchan; i++)
{
for(j = 0; j < sf.hdr.nsubband; j++)
{
for(k = 0; k < NUM_STOKES; k++)
{
accumulator[accumid]
[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k] +=
(float)i_payload[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k];
}
}
}
}
/* Add spectrum to appropriate vector accumulator (low-bw mode) */
else
{
for(i = 0; i < sf.hdr.nchan; i++)
{
for(j = 0; j < sf.hdr.nsubband; j++)
{
for(k = 0; k < NUM_STOKES; k++)
{
accumulator[accumid]
[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k] +=
f_payload[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k];
}
}
}
}
}
accum_time += (double)freq_heap->integ_size / pfb_rate;
}
/* Update packet count and loss fields from input header */
nblock_int++;
npacket += gp.num_pkts_rcvd;
n_pkt_drop += gp.num_pkts_dropped;
/* Done with current input block */
guppi_databuf_set_free(db_in, curblock_in);
curblock_in = (curblock_in + 1) % db_in->n_block;
/* Check for cancel */
pthread_testcancel();
}
pthread_exit(NULL);
pthread_cleanup_pop(0); /* Closes set_exit_status */
pthread_cleanup_pop(0); /* Closes set_finished */
pthread_cleanup_pop(0); /* Closes guppi_free_sdfits */
pthread_cleanup_pop(0); /* Closes ? */
pthread_cleanup_pop(0); /* Closes destroy_accumulators */
pthread_cleanup_pop(0); /* Closes guppi_status_detach */
pthread_cleanup_pop(0); /* Closes guppi_databuf_detach */
}
void
smp_rendezvous_cpus(cpuset_t map,
void (* setup_func)(void *),
void (* action_func)(void *),
void (* teardown_func)(void *),
void *arg)
{
int curcpumap, i, ncpus = 0;
/* Look comments in the !SMP case. */
if (!smp_started) {
spinlock_enter();
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
teardown_func(arg);
spinlock_exit();
return;
}
CPU_FOREACH(i) {
if (CPU_ISSET(i, &map))
ncpus++;
}
if (ncpus == 0)
panic("ncpus is 0 with non-zero map");
mtx_lock_spin(&smp_ipi_mtx);
/* Pass rendezvous parameters via global variables. */
smp_rv_ncpus = ncpus;
smp_rv_setup_func = setup_func;
smp_rv_action_func = action_func;
smp_rv_teardown_func = teardown_func;
smp_rv_func_arg = arg;
smp_rv_waiters[1] = 0;
smp_rv_waiters[2] = 0;
smp_rv_waiters[3] = 0;
atomic_store_rel_int(&smp_rv_waiters[0], 0);
/*
* Signal other processors, which will enter the IPI with
* interrupts off.
*/
curcpumap = CPU_ISSET(curcpu, &map);
CPU_CLR(curcpu, &map);
ipi_selected(map, IPI_RENDEZVOUS);
/* Check if the current CPU is in the map */
if (curcpumap != 0)
smp_rendezvous_action();
/*
* Ensure that the master CPU waits for all the other
* CPUs to finish the rendezvous, so that smp_rv_*
* pseudo-structure and the arg are guaranteed to not
* be in use.
*/
while (atomic_load_acq_int(&smp_rv_waiters[3]) < ncpus)
cpu_spinwait();
mtx_unlock_spin(&smp_ipi_mtx);
}
static void
xctrl_suspend()
{
#ifdef SMP
cpuset_t cpu_suspend_map;
#endif
int suspend_cancelled;
EVENTHANDLER_INVOKE(power_suspend);
if (smp_started) {
thread_lock(curthread);
sched_bind(curthread, 0);
thread_unlock(curthread);
}
KASSERT((PCPU_GET(cpuid) == 0), ("Not running on CPU#0"));
/*
* Clear our XenStore node so the toolstack knows we are
* responding to the suspend request.
*/
xs_write(XST_NIL, "control", "shutdown", "");
/*
* Be sure to hold Giant across DEVICE_SUSPEND/RESUME since non-MPSAFE
* drivers need this.
*/
mtx_lock(&Giant);
if (DEVICE_SUSPEND(root_bus) != 0) {
mtx_unlock(&Giant);
printf("%s: device_suspend failed\n", __func__);
return;
}
mtx_unlock(&Giant);
#ifdef SMP
CPU_ZERO(&cpu_suspend_map); /* silence gcc */
if (smp_started) {
/*
* Suspend other CPUs. This prevents IPIs while we
* are resuming, and will allow us to reset per-cpu
* vcpu_info on resume.
*/
cpu_suspend_map = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &cpu_suspend_map);
if (!CPU_EMPTY(&cpu_suspend_map))
suspend_cpus(cpu_suspend_map);
}
#endif
/*
* Prevent any races with evtchn_interrupt() handler.
*/
disable_intr();
intr_suspend();
xen_hvm_suspend();
suspend_cancelled = HYPERVISOR_suspend(0);
xen_hvm_resume(suspend_cancelled != 0);
intr_resume(suspend_cancelled != 0);
enable_intr();
/*
* Reset grant table info.
*/
gnttab_resume(NULL);
#ifdef SMP
/* Send an IPI_BITMAP in case there are pending bitmap IPIs. */
lapic_ipi_vectored(IPI_BITMAP_VECTOR, APIC_IPI_DEST_ALL);
if (smp_started && !CPU_EMPTY(&cpu_suspend_map)) {
/*
* Now that event channels have been initialized,
* resume CPUs.
*/
resume_cpus(cpu_suspend_map);
}
#endif
/*
* FreeBSD really needs to add DEVICE_SUSPEND_CANCEL or
* similar.
*/
mtx_lock(&Giant);
DEVICE_RESUME(root_bus);
mtx_unlock(&Giant);
if (smp_started) {
thread_lock(curthread);
sched_unbind(curthread);
thread_unlock(curthread);
}
EVENTHANDLER_INVOKE(power_resume);
if (bootverbose)
printf("System resumed after suspension\n");
}
