darling-xnu/tests/kevent_qos.c
2023-05-16 21:41:14 -07:00

1958 lines
64 KiB
C

/*
* kevent_qos: Tests Synchronous IPC QOS override.
*/
#ifdef T_NAMESPACE
#undef T_NAMESPACE
#endif
#include <darwintest.h>
#include <darwintest_multiprocess.h>
#include <dispatch/dispatch.h>
#include <pthread.h>
#include <launch.h>
#include <mach/mach.h>
#include <mach/message.h>
#include <mach/mach_voucher.h>
#include <pthread/workqueue_private.h>
#include <voucher/ipc_pthread_priority_types.h>
#include <servers/bootstrap.h>
#include <stdlib.h>
#include <sys/event.h>
#include <unistd.h>
#include <crt_externs.h>
#include <mach/mach_port.h>
#include <mach/mach_sync_ipc.h>
T_GLOBAL_META(T_META_NAMESPACE("xnu.kevent_qos"));
#define ARRAYLEN(arr) (sizeof(arr) / sizeof(arr[0]))
#define INTERMITTENT_TIMEOUT_SEC (3)
#define RECV_TIMEOUT_SECS (4)
#define SEND_TIMEOUT_SECS (6)
#define HELPER_TIMEOUT_SECS (15)
#define ENV_VAR_QOS (3)
static const char *qos_env[ENV_VAR_QOS] = {"XNU_TEST_QOS_BO", "XNU_TEST_QOS_QO", "XNU_TEST_QOS_AO"};
static const char *qos_name_env[ENV_VAR_QOS] = {"XNU_TEST_QOS_NAME_BO", "XNU_TEST_QOS_NAME_QO", "XNU_TEST_QOS_NAME_AO"};
#define ENV_VAR_FUNCTION (1)
static const char *wl_function_name = "XNU_TEST_WL_FUNCTION";
static qos_class_t g_expected_qos[ENV_VAR_QOS];
static const char *g_expected_qos_name[ENV_VAR_QOS];
#define ENV_QOS_BEFORE_OVERRIDE (0)
#define ENV_QOS_QUEUE_OVERRIDE (1)
#define ENV_QOS_AFTER_OVERRIDE (2)
struct test_msg {
mach_msg_header_t header;
mach_msg_body_t body;
mach_msg_port_descriptor_t port_descriptor;
mach_msg_option_t opts;
mach_msg_priority_t qos;
};
#pragma mark pthread callbacks
static pthread_t
thread_create_at_qos(qos_class_t qos, void * (*function)(void *));
static void
send(mach_port_t send_port, mach_port_t reply_port, mach_port_t msg_port, mach_msg_priority_t qos, mach_msg_option_t options);
static void
enable_kevent(uint64_t *workloop_id, unsigned long long port);
static void
populate_kevent(struct kevent_qos_s *kev, unsigned long long port);
static void
worker_cb(pthread_priority_t __unused priority)
{
T_FAIL("a worker thread was created");
}
static void
event_cb(void ** __unused events, int * __unused nevents)
{
T_FAIL("a kevent routine was called instead of workloop");
}
static uint32_t
get_user_promotion_basepri(void)
{
mach_msg_type_number_t count = THREAD_POLICY_STATE_COUNT;
struct thread_policy_state thread_policy;
boolean_t get_default = FALSE;
mach_port_t thread_port = pthread_mach_thread_np(pthread_self());
kern_return_t kr = thread_policy_get(thread_port, THREAD_POLICY_STATE,
(thread_policy_t)&thread_policy, &count, &get_default);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "thread_policy_get");
return thread_policy.thps_user_promotion_basepri;
}
#define EXPECT_QOS_EQ(qos, ...) do { \
if ((qos) == QOS_CLASS_USER_INTERACTIVE) { \
T_EXPECT_EFFECTIVE_QOS_EQ(QOS_CLASS_USER_INITIATED, __VA_ARGS__); \
T_EXPECT_EQ(47u, get_user_promotion_basepri(), __VA_ARGS__); \
} else { \
T_EXPECT_EFFECTIVE_QOS_EQ(qos, __VA_ARGS__); \
} \
} while (0)
#define EXPECT_TEST_MSG(_ke) do { \
struct kevent_qos_s *ke = _ke; \
mach_msg_header_t *hdr = (mach_msg_header_t *)ke->ext[0]; \
T_ASSERT_NOTNULL(hdr, "has a message"); \
T_ASSERT_EQ(hdr->msgh_size, (uint32_t)sizeof(struct test_msg), "of the right size"); \
struct test_msg *tmsg = (struct test_msg *)hdr; \
if (tmsg->opts & MACH_SEND_PROPAGATE_QOS) { \
T_EXPECT_EQ(tmsg->qos, ((uint32_t)(ke->ext[2] >> 32)), \
"propagation works"); \
} \
} while (0)
/*
* Basic WL handler callback, it sleeps for n seconds and then checks the
* effective Qos of the servicer thread.
*/
static void
workloop_cb_test_intransit(uint64_t *workloop_id __unused, void **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_intransit called. "
"Will wait for %d seconds to make sure client enqueues the sync msg \n",
2 * RECV_TIMEOUT_SECS);
EXPECT_TEST_MSG(*eventslist);
/* Wait for the client to send the high priority message to override the qos */
sleep(2 * RECV_TIMEOUT_SECS);
/* Skip the test if we can't check Qos */
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
*events = 0;
T_END;
}
/*
* WL handler which checks if the servicer thread has correct Qos.
*/
static void
workloop_cb_test_sync_send(uint64_t *workloop_id __unused, void **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send called");
EXPECT_TEST_MSG(*eventslist);
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
*events = 0;
T_END;
}
/*
* WL handler which checks the overridden Qos and then enables the knote and checks
* for the Qos again if that dropped the sync ipc override.
*/
static void
workloop_cb_test_sync_send_and_enable(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
unsigned override_priority;
unsigned reenable_priority;
T_LOG("Workloop handler workloop_cb_test_sync_send_and_enable called");
EXPECT_TEST_MSG(*eventslist);
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
/* Snapshot the current override priority */
override_priority = get_user_promotion_basepri();
/* Enable the knote */
struct kevent_qos_s *kev = *eventslist;
enable_kevent(workloop_id, kev->ident);
/*
* Check if the override has been dropped, check for priority instead of qos since
* there will be async qos push.
*/
reenable_priority = get_user_promotion_basepri();
T_EXPECT_LT(reenable_priority, override_priority,
"thread's current override priority %d should be less than override priority prior to enabling knote %d",
reenable_priority, override_priority);
*events = 0;
T_END;
}
/*
* WL handler which checks the overridden Qos and then handoffs the IPC,
* enables the knote and checks for the Qos again that it hasn't dropped the sync ipc override.
*/
static void
workloop_cb_test_sync_send_and_enable_handoff(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
unsigned override_priority;
int error;
T_LOG("Workloop handler workloop_cb_test_sync_send_and_enable_handoff called");
EXPECT_TEST_MSG(*eventslist);
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s",
g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
/* Snapshot the current override priority */
override_priority = get_user_promotion_basepri();
struct kevent_qos_s *kev = *eventslist;
mach_msg_header_t *hdr = (mach_msg_header_t *)kev->ext[0];
/* handoff the IPC */
struct kevent_qos_s handoff_kev = {
.filter = EVFILT_WORKLOOP,
.ident = hdr->msgh_remote_port,
.flags = EV_ADD | EV_DISABLE,
.fflags = 0x80000000,
};
error = kevent_id(*workloop_id, &handoff_kev, 1, &handoff_kev, 1, NULL,
NULL, KEVENT_FLAG_WORKLOOP | KEVENT_FLAG_ERROR_EVENTS | KEVENT_FLAG_DYNAMIC_KQ_MUST_EXIST);
T_QUIET; T_ASSERT_POSIX_SUCCESS(error, "kevent_id");
T_ASSERT_EQ(0, error, "Handed off the sync IPC");
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
/*
* Check if the override has not been dropped.
*/
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should still be %s",
g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
*events = 0;
T_END;
}
/*
* WL handler receives the first message and checks sync ipc override, then enables the knote
* and receives 2nd message and checks it sync ipc override.
*/
static int send_two_sync_handler_called = 0;
static void
workloop_cb_test_send_two_sync(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_send_two_sync called for %d time", send_two_sync_handler_called + 1);
EXPECT_TEST_MSG(*eventslist);
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_LOG("Number of events received is %d\n", *events);
if (send_two_sync_handler_called == 0) {
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
/* Enable the knote to get 2nd message */
struct kevent_qos_s *kev = *eventslist;
uint64_t port = kev->ident;
populate_kevent(kev, port);
*events = 1;
} else {
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_BEFORE_OVERRIDE]);
*events = 0;
T_END;
}
send_two_sync_handler_called++;
}
/*
* Checks the sync ipc override and then waits for client to destroy the
* special reply port and checks if that removes the sync ipc override.
*/
static boolean_t two_send_and_destroy_test_passed = FALSE;
static int two_send_and_destroy_handler = 0;
static void
workloop_cb_test_two_send_and_destroy(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist __unused, int *events)
{
T_LOG("Workloop handler workloop_cb_test_two_send_and_destroy called %d times", two_send_and_destroy_handler + 1);
EXPECT_TEST_MSG(*eventslist);
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
if (two_send_and_destroy_handler == 0) {
/* Sleep to make sure the mqueue gets full */
sleep(RECV_TIMEOUT_SECS);
/* The effective Qos should be the one expected after override */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
sleep(SEND_TIMEOUT_SECS);
/* Special reply port should have been destroyed, check Qos again */
EXPECT_QOS_EQ(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE],
"dispatch_source event handler QoS should be %s", g_expected_qos_name[ENV_QOS_BEFORE_OVERRIDE]);
two_send_and_destroy_test_passed = TRUE;
} else {
if (two_send_and_destroy_test_passed) {
T_END;
}
}
/* Enable the knote to get next message */
struct kevent_qos_s *kev = *eventslist;
uint64_t port = kev->ident;
populate_kevent(kev, port);
*events = 1;
two_send_and_destroy_handler++;
T_LOG("Handler returning \n");
}
static mach_port_type_t
get_reply_port(struct kevent_qos_s *kev)
{
mach_msg_header_t *hdr;
mach_port_t reply_port;
mach_port_type_t type;
kern_return_t kr;
hdr = (void*)kev->ext[0];
T_QUIET; T_ASSERT_NOTNULL(hdr, "msg hdr");
reply_port = hdr->msgh_remote_port;
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(reply_port), "reply port valid");
kr = mach_port_type(mach_task_self(), reply_port, &type);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_type");
T_QUIET; T_ASSERT_TRUE(type & MACH_PORT_TYPE_SEND_ONCE, "send once received");
return reply_port;
}
static void
send_reply(mach_port_t reply_port)
{
kern_return_t kr;
struct {
mach_msg_header_t header;
} send_msg = {
.header = {
.msgh_remote_port = reply_port,
.msgh_local_port = MACH_PORT_NULL,
.msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_MOVE_SEND_ONCE, 0),
.msgh_id = 0x100,
.msgh_size = sizeof(send_msg),
},
};
kr = mach_msg(&(send_msg.header),
MACH_SEND_MSG,
send_msg.header.msgh_size,
0,
MACH_PORT_NULL,
0,
0);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "server mach_msg");
}
static void
populate_kevent(struct kevent_qos_s *kev, unsigned long long port)
{
memset(kev, 0, sizeof(struct kevent_qos_s));
kev->ident = port;
kev->filter = EVFILT_MACHPORT;
kev->flags = EV_ADD | EV_ENABLE | EV_UDATA_SPECIFIC | EV_DISPATCH | EV_VANISHED;
kev->fflags = (MACH_RCV_MSG | MACH_RCV_VOUCHER | MACH_RCV_LARGE | MACH_RCV_LARGE_IDENTITY |
MACH_RCV_TRAILER_ELEMENTS(MACH_RCV_TRAILER_AV) |
MACH_RCV_TRAILER_TYPE(MACH_MSG_TRAILER_FORMAT_0));
kev->data = 1;
}
static void
enable_kevent(uint64_t *workloop_id, unsigned long long port)
{
struct kevent_qos_s kev;
int error;
populate_kevent(&kev, port);
struct kevent_qos_s kev_err[] = {{ 0 }};
error = kevent_id(*workloop_id, &kev, 1, kev_err, 1, NULL,
NULL, KEVENT_FLAG_WORKLOOP | KEVENT_FLAG_ERROR_EVENTS | KEVENT_FLAG_DYNAMIC_KQ_MUST_EXIST);
T_QUIET; T_ASSERT_POSIX_SUCCESS(error, "kevent_id");
}
/*
* WL handler which sends a msg to the client from handler.
*/
static void
workloop_cb_test_sync_send_reply(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_reply called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
/* send reply */
send_reply(get_reply_port(*eventslist));
*events = 0;
}
/*
* WL handler which deallocates reply port.
*/
static void
workloop_cb_test_sync_send_deallocate(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
mach_port_t reply_port;
kern_return_t kr;
T_LOG("Workloop handler workloop_cb_test_sync_send_deallocate called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
reply_port = get_reply_port(*eventslist);
/* deallocate port */
kr = mach_port_deallocate(mach_task_self(), reply_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_deallocate");
*events = 0;
T_LOG("Handler returning \n");
}
/*
* WL handler which sends a msg to the client before enabling the event from handler.
*/
static void
workloop_cb_test_sync_send_reply_kevent(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_reply_kevent called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT(((*eventslist)->filter), EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
/* send reply */
send_reply(get_reply_port(kev));
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
*events = 0;
T_LOG("Handler returning \n");
}
/*
* WL handler which sends a msg to the client before enabling the event from pthread.
*/
static void
workloop_cb_test_sync_send_reply_kevent_pthread(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_reply_kevent_pthread called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
/* send reply */
send_reply(get_reply_port(kev));
populate_kevent(kev, kev->ident);
*events = 1;
T_LOG("Handler returning \n");
}
/*
* WL handler which sends a msg to the client after reenabling the event.
*/
static void
workloop_cb_test_sync_send_kevent_reply(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("workloop handler workloop_cb_test_sync_send_kevent_reply called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
mach_port_t reply_port = get_reply_port(*eventslist);
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
/* send reply */
send_reply(reply_port);
*events = 0;
T_LOG("Handler returning \n");
}
/*
* WL handler that does nothing.
*/
static void
workloop_cb_test_sync_send_do_nothing(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_do_nothing called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
/* do nothing */
*events = 0;
T_LOG("Handler returning \n");
}
/*
* WL handler that returns the event to reenable.
*/
static void
workloop_cb_test_sync_send_do_nothing_kevent_pthread(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_do_nothing_kevent_pthread called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
populate_kevent(kev, kev->ident);
*events = 1;
T_LOG("handler returning \n");
}
/*
* WL handler that exits.
*/
static void
workloop_cb_test_sync_send_do_nothing_exit(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, __unused int *events)
{
T_LOG("workloop handler workloop_cb_test_sync_send_do_nothing_exit called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
/* call exit */
exit(0);
}
/*
* WL handler which:
* first sync sends a msg to the client and reenables kevent after
* second sync sends a msg and reenables kevent after.
*/
static void
workloop_cb_test_sync_send_reply_kevent_reply_kevent(uint64_t *workloop_id __unused, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("Workloop handler workloop_cb_test_sync_send_reply_kevent_reply_kevent called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
/* send reply */
send_reply(get_reply_port(kev));
populate_kevent(kev, kev->ident);
*events = 1;
T_LOG("Handler returning \n");
}
/*
* WL handler which:
* first sync reenables kevent and after sends a msg
* second sync sends a msg and reenables kevent after.
*/
static int workloop_cb_test_sync_send_kevent_reply_reply_kevent_handler_called = 0;
static void
workloop_cb_test_sync_send_kevent_reply_reply_kevent(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("workloop handler workloop_cb_test_sync_send_kevent_reply_reply_kevent called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
mach_port_t reply_port = get_reply_port(kev);
if (workloop_cb_test_sync_send_kevent_reply_reply_kevent_handler_called == 0) {
workloop_cb_test_sync_send_kevent_reply_reply_kevent_handler_called = 1;
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
/* send reply */
send_reply(reply_port);
*events = 0;
} else {
/* send reply */
send_reply(reply_port);
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
*events = 0;
}
T_LOG("Handler returning \n");
}
/*
* WL handler which:
* first sync reenables kevent and after sends a msg
* second sync reenables kevent and after sends a msg
*/
static void
workloop_cb_test_sync_send_kevent_reply_kevent_reply(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("workloop handler workloop_cb_test_sync_send_kevent_reply_kevent_reply called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
mach_port_t reply_port = get_reply_port(kev);
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
/* send reply */
send_reply(reply_port);
*events = 0;
T_LOG("Handler returning \n");
}
/*
* WL handler which:
* first sync ends a msg and reenables kevent after
* second sync reenables kevent and sends a msg after
*/
static int workloop_cb_test_sync_send_reply_kevent_kevent_reply_handler_called = 0;
static void
workloop_cb_test_sync_send_reply_kevent_kevent_reply(uint64_t *workloop_id, struct kevent_qos_s **eventslist, int *events)
{
T_LOG("workloop handler workloop_cb_test_sync_send_reply_kevent_kevent_reply called");
if (geteuid() != 0) {
T_SKIP("kevent_qos test requires root privileges to run.");
}
T_QUIET; T_ASSERT_EQ_INT(*events, 1, "events received");
T_QUIET; T_ASSERT_EQ_INT((*eventslist)->filter, EVFILT_MACHPORT, "received EVFILT_MACHPORT");
struct kevent_qos_s *kev = *eventslist;
mach_port_t reply_port = get_reply_port(kev);
if (workloop_cb_test_sync_send_reply_kevent_kevent_reply_handler_called == 0) {
workloop_cb_test_sync_send_reply_kevent_kevent_reply_handler_called = 1;
/* send reply */
send_reply(reply_port);
populate_kevent(kev, kev->ident);
*events = 1;
} else {
/* Enable the knote */
enable_kevent(workloop_id, kev->ident);
/* send reply */
send_reply(reply_port);
*events = 0;
}
T_LOG("Handler returning \n");
}
#pragma mark Mach receive
#define KEVENT_QOS_SERVICE_NAME "com.apple.xnu.test.kevent_qos"
static mach_port_t
get_server_port(void)
{
mach_port_t port;
kern_return_t kr = bootstrap_check_in(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "server bootstrap_check_in");
return port;
}
static void
env_set_qos(char **env, qos_class_t qos[], const char *qos_name[], const char *wl_function)
{
int i;
char *qos_str, *qos_name_str;
for (i = 0; i < ENV_VAR_QOS; i++) {
T_QUIET; T_ASSERT_POSIX_SUCCESS(asprintf(&qos_str, "%s=%d", qos_env[i], qos[i]),
NULL);
T_QUIET; T_ASSERT_POSIX_SUCCESS(
asprintf(&qos_name_str, "%s=%s", qos_name_env[i], qos_name[i]), NULL);
env[2 * i] = qos_str;
env[2 * i + 1] = qos_name_str;
}
T_QUIET; T_ASSERT_POSIX_SUCCESS(asprintf(&env[2 * i], "%s=%s", wl_function_name, wl_function),
NULL);
env[2 * i + 1] = NULL;
}
static void
environ_get_qos(qos_class_t qos[], const char *qos_name[], const char **wl_function)
{
char *qos_str;
char *qos_end;
int i;
for (i = 0; i < ENV_VAR_QOS; i++) {
qos_str = getenv(qos_env[i]);
T_QUIET; T_ASSERT_NOTNULL(qos_str, "getenv(%s)", qos_env[i]);
unsigned long qos_l = strtoul(qos_str, &qos_end, 10);
T_QUIET; T_ASSERT_EQ(*qos_end, '\0', "getenv(%s) = '%s' should be an "
"integer", qos_env[i], qos_str);
T_QUIET; T_ASSERT_LT(qos_l, (unsigned long)100, "getenv(%s) = '%s' should "
"be less than 100", qos_env[i], qos_str);
qos[i] = (qos_class_t)qos_l;
qos_name[i] = getenv(qos_name_env[i]);
T_QUIET; T_ASSERT_NOTNULL(qos_name[i], "getenv(%s)", qos_name_env[i]);
}
*wl_function = getenv(wl_function_name);
T_QUIET; T_ASSERT_NOTNULL(*wl_function, "getenv(%s)", wl_function_name);
}
static mach_voucher_t
create_pthpriority_voucher(mach_msg_priority_t qos)
{
char voucher_buf[sizeof(mach_voucher_attr_recipe_data_t) + sizeof(ipc_pthread_priority_value_t)];
mach_voucher_t voucher = MACH_PORT_NULL;
kern_return_t ret;
ipc_pthread_priority_value_t ipc_pthread_priority_value =
(ipc_pthread_priority_value_t)qos;
mach_voucher_attr_raw_recipe_array_t recipes;
mach_voucher_attr_raw_recipe_size_t recipe_size = 0;
mach_voucher_attr_recipe_t recipe =
(mach_voucher_attr_recipe_t)&voucher_buf[recipe_size];
recipe->key = MACH_VOUCHER_ATTR_KEY_PTHPRIORITY;
recipe->command = MACH_VOUCHER_ATTR_PTHPRIORITY_CREATE;
recipe->previous_voucher = MACH_VOUCHER_NULL;
memcpy((char *)&recipe->content[0], &ipc_pthread_priority_value, sizeof(ipc_pthread_priority_value));
recipe->content_size = sizeof(ipc_pthread_priority_value_t);
recipe_size += sizeof(mach_voucher_attr_recipe_data_t) + recipe->content_size;
recipes = (mach_voucher_attr_raw_recipe_array_t)&voucher_buf[0];
ret = host_create_mach_voucher(mach_host_self(),
recipes,
recipe_size,
&voucher);
T_QUIET; T_ASSERT_MACH_SUCCESS(ret, "client host_create_mach_voucher");
return voucher;
}
static void
send(
mach_port_t send_port,
mach_port_t reply_port,
mach_port_t msg_port,
mach_msg_priority_t qos,
mach_msg_option_t options)
{
kern_return_t ret = 0;
struct test_msg send_msg = {
.header = {
.msgh_remote_port = send_port,
.msgh_local_port = reply_port,
.msgh_bits = MACH_MSGH_BITS_SET(MACH_MSG_TYPE_COPY_SEND,
reply_port ? MACH_MSG_TYPE_MAKE_SEND_ONCE : 0,
MACH_MSG_TYPE_MOVE_SEND,
MACH_MSGH_BITS_COMPLEX),
.msgh_id = 0x100,
.msgh_size = sizeof(send_msg),
},
.body = {
.msgh_descriptor_count = 1,
},
.port_descriptor = {
.name = msg_port,
.disposition = MACH_MSG_TYPE_MOVE_RECEIVE,
.type = MACH_MSG_PORT_DESCRIPTOR,
},
.opts = options,
};
if (msg_port == MACH_PORT_NULL) {
send_msg.body.msgh_descriptor_count = 0;
}
if ((options & MACH_SEND_PROPAGATE_QOS) == 0) {
send_msg.header.msgh_voucher_port = create_pthpriority_voucher(qos);
send_msg.qos = qos;
} else {
qos_class_t qc;
int relpri;
pthread_get_qos_class_np(pthread_self(), &qc, &relpri);
send_msg.qos = (uint32_t)_pthread_qos_class_encode(qc, relpri, 0);
}
mach_msg_option_t send_opts = options;
if (reply_port) {
send_opts |= MACH_SEND_SYNC_OVERRIDE;
}
send_opts |= MACH_SEND_MSG | MACH_SEND_TIMEOUT | MACH_SEND_OVERRIDE;
ret = mach_msg(&send_msg.header, send_opts, send_msg.header.msgh_size,
0, MACH_PORT_NULL, 10000, qos);
T_QUIET; T_ASSERT_MACH_SUCCESS(ret, "client mach_msg");
}
static kern_return_t
receive(
mach_port_t rcv_port,
mach_port_t notify_port)
{
kern_return_t ret = 0;
struct test_msg rcv_msg = {
.header = {
.msgh_remote_port = MACH_PORT_NULL,
.msgh_local_port = rcv_port,
.msgh_size = sizeof(rcv_msg),
},
};
T_LOG("Client: Starting sync receive\n");
ret = mach_msg(&(rcv_msg.header),
MACH_RCV_MSG |
MACH_RCV_TIMEOUT |
MACH_RCV_SYNC_WAIT,
0,
rcv_msg.header.msgh_size,
rcv_port,
SEND_TIMEOUT_SECS * 1000,
notify_port);
return ret;
}
T_HELPER_DECL(qos_get_special_reply_port,
"Test get_special_reply_port and it's corner cases.")
{
mach_port_t special_reply_port;
mach_port_t new_special_reply_port;
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
new_special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(new_special_reply_port), "get_thread_special_reply_port");
mach_port_destroy(mach_task_self(), special_reply_port);
mach_port_destroy(mach_task_self(), new_special_reply_port);
new_special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(new_special_reply_port), "get_thread_special_reply_port");
T_END;
}
static void *
qos_client_send_to_intransit(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t msg_port;
mach_port_t special_reply_port;
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* Create a rcv right to send in a msg */
kr = mach_port_allocate(mach_task_self(),
MACH_PORT_RIGHT_RECEIVE,
&msg_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client mach_port_allocate");
kr = mach_port_insert_right(mach_task_self(),
msg_port,
msg_port,
MACH_MSG_TYPE_MAKE_SEND);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client mach_port_insert_right");
/* Send an empty msg on the port to fire the WL thread */
send(qos_send_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
/* Sleep 3 seconds for the server to start */
sleep(3);
/* Send the message with msg port as in-transit port, this msg will not be dequeued */
send(qos_send_port, MACH_PORT_NULL, msg_port,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
/* Send 5 messages to msg port to make sure the port is full */
for (int i = 0; i < 5; i++) {
send(msg_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
}
T_LOG("Sent 5 msgs, now trying to send sync ipc message, which will block with a timeout\n");
/* Send the message to the in-transit port, it should block and override the rcv's workloop */
send(msg_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
T_LOG("Client done sending messages, now waiting for server to end the test");
T_ASSERT_FAIL("client timed out");
return NULL;
}
T_HELPER_DECL(qos_client_send_to_intransit_with_thr_pri,
"Send synchronous messages from a pri thread to an intransit port")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_client_send_to_intransit);
sleep(HELPER_TIMEOUT_SECS);
}
static pthread_t
thread_create_at_qos(qos_class_t qos, void * (*function)(void *))
{
qos_class_t qos_thread;
pthread_t thread;
pthread_attr_t attr;
int ret;
ret = setpriority(PRIO_DARWIN_ROLE, 0, PRIO_DARWIN_ROLE_UI_FOCAL);
if (ret != 0) {
T_LOG("set priority failed\n");
}
pthread_attr_init(&attr);
pthread_attr_set_qos_class_np(&attr, qos, 0);
pthread_create(&thread, &attr, function, NULL);
T_LOG("pthread created\n");
pthread_get_qos_class_np(thread, &qos_thread, NULL);
T_EXPECT_EQ(qos_thread, (qos_class_t)qos, NULL);
return thread;
}
static void *
qos_send_and_sync_rcv(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
T_LOG("Client: from created thread\n");
T_EXPECT_EFFECTIVE_QOS_EQ(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
"pthread QoS should be %s", g_expected_qos_name[ENV_QOS_AFTER_OVERRIDE]);
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* enqueue two messages to make sure that mqueue is not empty */
send(qos_send_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_QUEUE_OVERRIDE], 0, 0), 0);
send(qos_send_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_QUEUE_OVERRIDE], 0, 0), 0);
sleep(SEND_TIMEOUT_SECS);
/* sync wait on msg port */
receive(special_reply_port, qos_send_port);
T_LOG("Client done doing sync rcv, now waiting for server to end the test");
sleep(SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
static void *
qos_sync_rcv(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
T_LOG("Client: from created thread\n");
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* enqueue two messages to make sure that mqueue is not empty */
send(qos_send_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_QUEUE_OVERRIDE], 0, 0), 0);
sleep(RECV_TIMEOUT_SECS);
/* sync wait on msg port */
receive(special_reply_port, qos_send_port);
T_LOG("Client done doing sync rcv, now waiting for server to end the test");
sleep(SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
static void
thread_wait_to_block(mach_port_t thread_port)
{
thread_extended_info_data_t extended_info;
kern_return_t kr;
while (1) {
mach_msg_type_number_t count = THREAD_EXTENDED_INFO_COUNT;
kr = thread_info(thread_port, THREAD_EXTENDED_INFO,
(thread_info_t)&extended_info, &count);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "thread_info");
if (extended_info.pth_run_state == TH_STATE_WAITING) {
T_LOG("Target thread blocked\n");
break;
}
thread_switch(thread_port, SWITCH_OPTION_DEPRESS, 0);
}
}
T_HELPER_DECL(qos_client_send_sync_and_sync_rcv,
"Send messages and syncronously wait for rcv")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_send_and_sync_rcv);
sleep(HELPER_TIMEOUT_SECS);
}
T_HELPER_DECL(qos_client_sync_rcv_qos_change,
"Send messages and syncronously wait for rcv and change qos of waiting thread")
{
pthread_t rcv_thread;
rcv_thread = thread_create_at_qos(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], qos_sync_rcv);
T_LOG("Waiting for %d seconds before changing qos of rcv thread", SEND_TIMEOUT_SECS);
sleep(SEND_TIMEOUT_SECS);
/* Wait for the thread to block */
thread_wait_to_block(pthread_mach_thread_np(rcv_thread));
/* Update the rcv thread's qos */
pthread_override_qos_class_start_np(rcv_thread, g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0);
sleep(HELPER_TIMEOUT_SECS);
}
static void *
qos_client_send_sync_msg_and_test_link(void *arg)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
boolean_t in_effect = FALSE;
kern_return_t kr;
unsigned long expected_result = (unsigned long) arg;
kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
/* start monitoring sync ipc link */
kr = mach_sync_ipc_link_monitoring_start(&special_reply_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_sync_ipc_link_monitoring_start");
/* Send the message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
/*
* wait for the reply
* some tests do not send a msg back so the receive
* might fail
*/
receive(special_reply_port, qos_send_port);
/* stop monitoring link */
kr = mach_sync_ipc_link_monitoring_stop(special_reply_port, &in_effect);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_sync_ipc_link_monitoring_stop");
if (!in_effect) {
T_LOG("Link was broken");
} else {
T_LOG("Link correct");
}
if (expected_result == 1) {
T_ASSERT_TRUE(in_effect, "special reply port link after rcv");
} else {
T_ASSERT_FALSE(in_effect, "special reply port link after rcv");
}
T_END;
}
static void *
qos_client_send_2sync_msg_and_test_link(void *arg)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
boolean_t in_effect = FALSE;
kern_return_t kr;
unsigned long expected_result = (unsigned long) arg;
kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
/* start monitoring sync ipc link */
kr = mach_sync_ipc_link_monitoring_start(&special_reply_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_sync_ipc_link_monitoring_start");
/* Send the first message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
/* wait for the reply */
kr = receive(special_reply_port, qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "receive");
/* Send the second message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
/* wait for the reply */
kr = receive(special_reply_port, qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "receive");
/* stop monitoring link */
kr = mach_sync_ipc_link_monitoring_stop(special_reply_port, &in_effect);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_sync_ipc_link_monitoring_stop");
if (!in_effect) {
T_LOG("Link was broken");
} else {
T_LOG("Link correct");
}
if (expected_result == 1) {
T_ASSERT_TRUE(in_effect, "special reply port link after rcv");
} else {
T_ASSERT_FALSE(in_effect, "special reply port link after rcv");
}
T_END;
}
T_HELPER_DECL(qos_client_send_sync_msg_with_link_check_correct_server,
"Send sync message, wait for reply and check sync ipc link")
{
pthread_t thread;
pthread_attr_t attr;
unsigned long expected_result = 1;
pthread_attr_init(&attr);
pthread_create(&thread, &attr, qos_client_send_sync_msg_and_test_link, (void *)expected_result);
sleep(HELPER_TIMEOUT_SECS);
}
T_HELPER_DECL(qos_client_send_sync_msg_with_link_check_incorrect_server,
"Send sync message, wait for reply and check sync ipc link")
{
pthread_t thread;
pthread_attr_t attr;
unsigned long expected_result = 0;
pthread_attr_init(&attr);
pthread_create(&thread, &attr, qos_client_send_sync_msg_and_test_link, (void *)expected_result);
sleep(HELPER_TIMEOUT_SECS);
}
T_HELPER_DECL(qos_client_send_2sync_msg_with_link_check_correct_server,
"Send sync message, wait for reply and check sync ipc link")
{
pthread_t thread;
pthread_attr_t attr;
unsigned long expected_result = 1;
pthread_attr_init(&attr);
pthread_create(&thread, &attr, qos_client_send_2sync_msg_and_test_link, (void *)expected_result);
sleep(HELPER_TIMEOUT_SECS);
}
T_HELPER_DECL(qos_client_send_2sync_msg_with_link_check_incorrect_server,
"Send sync message, wait for reply and check sync ipc link")
{
pthread_t thread;
pthread_attr_t attr;
unsigned long expected_result = 0;
pthread_attr_init(&attr);
pthread_create(&thread, &attr, qos_client_send_2sync_msg_and_test_link, (void *)expected_result);
sleep(HELPER_TIMEOUT_SECS);
}
static void *
qos_client_send_sync_msg(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* Send the message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
/* wait for the reply */
receive(special_reply_port, qos_send_port);
T_LOG("Client done sending messages, now waiting for server to end the test");
sleep(2 * SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
T_HELPER_DECL(qos_client_send_sync_msg_with_pri,
"Send sync message and wait for reply")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_client_send_sync_msg);
sleep(HELPER_TIMEOUT_SECS);
}
static void *
qos_client_send_two_sync_msg_high_qos(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* Send the message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], 0, 0), 0);
/* wait for the reply */
receive(special_reply_port, qos_send_port);
T_LOG("Client done sending messages, now waiting for server to end the test");
sleep(SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
static void *
qos_client_send_two_sync_msg_low_qos(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
/* Send the message to msg port */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
/* wait for the reply */
receive(special_reply_port, qos_send_port);
T_LOG("Client done sending messages, now waiting for server to end the test");
sleep(SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
T_HELPER_DECL(qos_client_send_two_sync_msg_with_thr_pri,
"Send messages sync msgs from 2 threads at given thread pri")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_client_send_two_sync_msg_high_qos);
sleep(INTERMITTENT_TIMEOUT_SEC);
thread_create_at_qos(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], qos_client_send_two_sync_msg_low_qos);
sleep(HELPER_TIMEOUT_SECS);
}
static mach_port_t other_thread_reply_port = MACH_PORT_NULL;
static void *
qos_client_destroy_other_threads_port(void *arg __unused)
{
T_LOG("Waiting 6 seconds before destroying other thread's reply port");
sleep(SEND_TIMEOUT_SECS);
T_LOG("Destroying other thread's special reply port ");
mach_port_destroy(mach_task_self(), other_thread_reply_port);
T_LOG("Other thread done destroying ");
sleep(3 * SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
static void *
qos_client_create_sepcial_reply_and_spawn_thread(void *arg __unused)
{
mach_port_t qos_send_port;
mach_port_t special_reply_port;
kern_return_t kr = bootstrap_look_up(bootstrap_port,
KEVENT_QOS_SERVICE_NAME, &qos_send_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
special_reply_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port");
other_thread_reply_port = special_reply_port;
/* Send an async message */
send(qos_send_port, MACH_PORT_NULL, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
/* Send the sync ipc message */
send(qos_send_port, special_reply_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0), 0);
/* Create a new thread to send the sync message on our special reply port */
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_client_destroy_other_threads_port);
/* Client starting to receive message */
receive(special_reply_port, qos_send_port);
sleep(3 * SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
T_HELPER_DECL(qos_client_send_two_msg_and_destroy,
"Send a message with another threads special reply port while that thread destroys the port")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE], qos_client_create_sepcial_reply_and_spawn_thread);
sleep(HELPER_TIMEOUT_SECS);
}
static mach_port_t send_complex_connection_port = MACH_PORT_NULL;
static void *
qos_client_send_complex_msg_to_service_port(void *arg __unused)
{
mach_port_t svc_port, tsr_port, conn_port;
kern_return_t kr;
kr = bootstrap_look_up(bootstrap_port, KEVENT_QOS_SERVICE_NAME, &svc_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up");
tsr_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(tsr_port), "get_thread_special_reply_port");
conn_port = send_complex_connection_port;
T_LOG("Sending to the service port with a sync IPC");
send(svc_port, tsr_port, conn_port,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0),
MACH_SEND_PROPAGATE_QOS);
receive(tsr_port, svc_port);
sleep(3 * SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
static void *
qos_client_send_to_connection_then_service_port(void *arg __unused)
{
mach_port_t tsr_port, conn_port;
mach_port_options_t opts = {
.flags = MPO_INSERT_SEND_RIGHT,
};
kern_return_t kr;
kr = mach_port_construct(mach_task_self(), &opts, 0ull, &conn_port);
T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct");
send_complex_connection_port = conn_port;
tsr_port = thread_get_special_reply_port();
T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(tsr_port), "get_thread_special_reply_port");
T_LOG("Sending to the connection port with a sync IPC");
send(conn_port, tsr_port, MACH_PORT_NULL,
(uint32_t)_pthread_qos_class_encode(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE], 0, 0),
MACH_SEND_PROPAGATE_QOS);
thread_create_at_qos(g_expected_qos[ENV_QOS_AFTER_OVERRIDE],
qos_client_send_complex_msg_to_service_port);
receive(tsr_port, conn_port);
sleep(3 * SEND_TIMEOUT_SECS);
T_ASSERT_FAIL("client timed out");
return NULL;
}
T_HELPER_DECL(qos_client_send_complex_msg_with_pri,
"Send a message with several ports causing links")
{
thread_create_at_qos(g_expected_qos[ENV_QOS_BEFORE_OVERRIDE],
qos_client_send_to_connection_then_service_port);
sleep(HELPER_TIMEOUT_SECS);
}
static void
run_client_server(const char *server_name, const char *client_name, qos_class_t qos[],
const char *qos_name[], const char *wl_function)
{
char *env[2 * ENV_VAR_QOS + ENV_VAR_FUNCTION + 1];
env_set_qos(env, qos, qos_name, wl_function);
for (int i = 0; i < ENV_VAR_QOS; i++) {
g_expected_qos[i] = qos[i];
g_expected_qos_name[i] = qos_name[i];
}
dt_helper_t helpers[] = {
dt_launchd_helper_env("com.apple.xnu.test.kevent_qos.plist",
server_name, env),
dt_fork_helper(client_name)
};
dt_run_helpers(helpers, 2, HELPER_TIMEOUT_SECS);
}
#pragma mark Mach receive - kevent_qos
static void
expect_kevent_id_recv(mach_port_t port, qos_class_t qos[], const char *qos_name[], const char *wl_function)
{
int r;
/* Qos expected by workloop thread */
for (int i = 0; i < ENV_VAR_QOS; i++) {
g_expected_qos[i] = qos[i];
g_expected_qos_name[i] = qos_name[i];
}
if (strcmp(wl_function, "workloop_cb_test_intransit") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_intransit, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_and_enable") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_and_enable, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_and_enable_handoff") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_and_enable_handoff, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_send_two_sync") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_send_two_sync, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_two_send_and_destroy") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_two_send_and_destroy, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_reply") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_reply, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_deallocate") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_deallocate, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_reply_kevent") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_reply_kevent, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_reply_kevent_pthread") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_reply_kevent_pthread, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_kevent_reply") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_kevent_reply, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_do_nothing") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_do_nothing, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_do_nothing_kevent_pthread") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_do_nothing_kevent_pthread, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_do_nothing_exit") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_do_nothing_exit, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_reply_kevent_reply_kevent") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_reply_kevent_reply_kevent, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_kevent_reply_reply_kevent") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_kevent_reply_reply_kevent, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_kevent_reply_kevent_reply") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_kevent_reply_kevent_reply, 0, 0), NULL);
} else if (strcmp(wl_function, "workloop_cb_test_sync_send_reply_kevent_kevent_reply") == 0) {
T_QUIET; T_ASSERT_POSIX_ZERO(_pthread_workqueue_init_with_workloop(
worker_cb, event_cb,
(pthread_workqueue_function_workloop_t)workloop_cb_test_sync_send_reply_kevent_kevent_reply, 0, 0), NULL);
} else {
T_ASSERT_FAIL("no workloop function specified \n");
}
struct kevent_qos_s kev[] = {{
.ident = port,
.filter = EVFILT_MACHPORT,
.flags = EV_ADD | EV_UDATA_SPECIFIC | EV_DISPATCH | EV_VANISHED,
.fflags = (MACH_RCV_MSG | MACH_RCV_VOUCHER | MACH_RCV_LARGE | MACH_RCV_LARGE_IDENTITY |
MACH_RCV_TRAILER_ELEMENTS(MACH_RCV_TRAILER_AV) |
MACH_RCV_TRAILER_TYPE(MACH_MSG_TRAILER_FORMAT_0)),
.data = 1,
.qos = (int32_t)_pthread_qos_class_encode(qos[ENV_QOS_QUEUE_OVERRIDE], 0, 0)
}};
struct kevent_qos_s kev_err[] = {{ 0 }};
/* Setup workloop for mach msg rcv */
r = kevent_id(25, kev, 1, kev_err, 1, NULL,
NULL, KEVENT_FLAG_WORKLOOP | KEVENT_FLAG_ERROR_EVENTS);
T_QUIET; T_ASSERT_POSIX_SUCCESS(r, "kevent_id");
T_QUIET; T_ASSERT_EQ(r, 0, "no errors returned from kevent_id");
sleep(HELPER_TIMEOUT_SECS);
}
T_HELPER_DECL(server_kevent_id,
"Reply with the QoS that a dispatch source event handler ran with")
{
qos_class_t qos[ENV_VAR_QOS];
const char *qos_name[ENV_VAR_QOS];
const char *wl_function;
environ_get_qos(qos, qos_name, &wl_function);
expect_kevent_id_recv(get_server_port(), qos, qos_name, wl_function);
sleep(HELPER_TIMEOUT_SECS);
T_ASSERT_FAIL("should receive a message within %d seconds",
RECV_TIMEOUT_SECS);
}
static void *
special_reply_port_thread(void *ctxt)
{
kern_return_t ret;
mach_port_t rcv_port = *(mach_port_t *)ctxt;
struct test_msg rcv_msg = {
.header = {
.msgh_remote_port = MACH_PORT_NULL,
.msgh_local_port = rcv_port,
.msgh_size = sizeof(rcv_msg),
},
};
ret = mach_msg(&rcv_msg.header, MACH_RCV_MSG | MACH_RCV_TIMEOUT, 0,
rcv_msg.header.msgh_size, rcv_port, 1000, MACH_PORT_NULL);
T_EXPECT_EQ(ret, MACH_RCV_TIMED_OUT, "receive should not panic");
*(mach_port_t *)ctxt = MACH_PORT_NULL;
sleep(1); // give some time to pthread_exit
ret = mach_msg(&rcv_msg.header, MACH_RCV_MSG | MACH_RCV_TIMEOUT, 0,
rcv_msg.header.msgh_size, rcv_port, 1000, MACH_PORT_NULL);
T_EXPECT_EQ(ret, MACH_RCV_TIMED_OUT, "receive should not panic");
T_END;
}
T_DECL(special_reply_port, "basic special reply port robustness checks",
T_META_RUN_CONCURRENTLY(true))
{
pthread_t thread;
mach_port_t srp = thread_get_special_reply_port();
pthread_create(&thread, NULL, special_reply_port_thread, &srp);
while (srp) {
usleep(1000);
}
pthread_exit(NULL);
}
#define TEST_QOS(server_name, client_name, name, wl_function_name, qos_bo, qos_bo_name, qos_qo, qos_qo_name, qos_ao, qos_ao_name) \
T_DECL(server_kevent_id_##name, \
"Event delivery at " qos_ao_name " QoS using a kevent_id", \
T_META_ASROOT(YES)) \
{ \
qos_class_t qos_array[ENV_VAR_QOS] = {qos_bo, qos_qo, qos_ao}; \
const char *qos_name_array[ENV_VAR_QOS] = {qos_bo_name, qos_qo_name, qos_ao_name}; \
run_client_server(server_name, client_name, qos_array, qos_name_array, wl_function_name); \
}
/*
* Test 1: Test special reply port SPI
*
* Create thread special reply port and check any subsequent calls to
* the same should return MACH_PORT_NULL, unless the reply port is destroyed.
*/
TEST_QOS("server_kevent_id", "qos_get_special_reply_port", special_reply_port, "workloop_cb_test_intransit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
/*
* Test 2: Test sync ipc send to an in-transit port
*
* Send a sync ipc message (at IN qos) to an in-transit port enqueued in a port
* attached to a workloop. Test that the servicer of the workloop gets
* sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_to_intransit_with_thr_pri", transit_IN, "workloop_cb_test_intransit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INITIATED, "user initiated")
/*
* Test 3: Test sync ipc send to an in-transit port
*
* Send a sync ipc message (at UI qos) to an in-transit port enqueued in a port
* attached to a workloop. Test that the servicer of the workloop gets
* sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_to_intransit_with_thr_pri", transit_UI, "workloop_cb_test_intransit",
QOS_CLASS_USER_INITIATED, "user initiated",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 4: Test starting a sync rcv overrides the servicer
*
* Send an async message to a port and then start waiting on
* the port in mach msg rcv (at IN qos) with sync wait and test if the
* servicer of the workloop gets sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_and_sync_rcv", rcv_IN, "workloop_cb_test_intransit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INITIATED, "user initiated")
/*
* Test 5: Test starting a sync rcv overrides the servicer
*
* Send an async message to a port and then start waiting on
* the port in mach msg rcv (at UI qos) with sync wait and test if the
* servicer of the workloop gets sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_and_sync_rcv", rcv_UI, "workloop_cb_test_intransit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INTERACTIVE, "user interactive with 47 basepri promotion")
/*
* Test 6: test sending sync ipc message (at IN qos) to port will override the servicer
*
* Send a message with sync ipc override to a port and check if the servicer
* of the workloop on other side gets sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_pri", send_sync_IN, "workloop_cb_test_sync_send",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INITIATED, "user initiated")
/*
* Test 7: test sending sync ipc message (at UI qos) to port will override the servicer
*
* Send a message with sync ipc override to a port and check if the servicer
* of the workloop on other side gets sync ipc override.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_pri", send_sync_UI, "workloop_cb_test_sync_send",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 8: test enabling a knote in workloop handler will drop the sync ipc override of delivered message
*
* Send a sync ipc message to port and check the servicer of the workloop
* on other side gets sync ipc override and once the handler enables the knote,
* that sync ipc override is dropped.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_pri", send_sync_UI_and_enable, "workloop_cb_test_sync_send_and_enable",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 9: test returning to begin processing drops sync ipc override of delivered message
*
* Send a sync ipc message and check if enabling the knote clears the override of
* the delivered message, but should still have the override of an enqueued message.
*/
TEST_QOS("server_kevent_id", "qos_client_send_two_sync_msg_with_thr_pri", send_two_sync_UI, "workloop_cb_test_send_two_sync",
QOS_CLASS_BACKGROUND, "background",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 10: test destroying the special reply port drops the override
*
* Send an async messages and a sync ipc message, the workloop handler
* should get a sync ipc override, now test if destroying the special
* reply port drops the sync ipc override on the servicer.
*/
TEST_QOS("server_kevent_id", "qos_client_send_two_msg_and_destroy", send_two_UI_and_destroy, "workloop_cb_test_two_send_and_destroy",
QOS_CLASS_BACKGROUND, "background",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 11: test sending two ports with chaining
*
* Send a sync IPC to a connection port, which itself is embedded in a message
* sent as a sync IPC to a service port.
*/
TEST_QOS("server_kevent_id", "qos_client_send_complex_msg_with_pri", send_complex_sync_UI_and_enable, "workloop_cb_test_sync_send_and_enable",
QOS_CLASS_USER_INITIATED, "user initiated",
QOS_CLASS_USER_INITIATED, "user initiated",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 12: test sending two ports with chaining
*
* Send a sync IPC to a connection port, which itself is embedded in a message
* sent as a sync IPC to a service port.
*/
TEST_QOS("server_kevent_id", "qos_client_send_complex_msg_with_pri", send_complex_sync_UI_and_enable_and_handoff, "workloop_cb_test_sync_send_and_enable_handoff",
QOS_CLASS_USER_INITIATED, "user initiated",
QOS_CLASS_USER_INITIATED, "user initiated",
QOS_CLASS_USER_INTERACTIVE, "user initiated with 47 basepri promotion")
/*
* Test 13: test changing qos of a thread to trigger turnstile push
*
* Send a sync IPC to a service port and change the qos of the blocked thread
* to verify that changing qos triggers a turnstile push.
*/
TEST_QOS("server_kevent_id", "qos_client_sync_rcv_qos_change", qos_change_to_IN, "workloop_cb_test_intransit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_MAINTENANCE, "maintenance",
QOS_CLASS_USER_INITIATED, "user initiated")
/*
* Test 14 - 21
*
* Test single sync ipc link with server that breaks/preserves the link in different ways.
*/
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_s, "workloop_cb_test_sync_send_reply",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_d, "workloop_cb_test_sync_send_deallocate",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_sk, "workloop_cb_test_sync_send_reply_kevent",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_skp, "workloop_cb_test_sync_send_reply_kevent_pthread",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_incorrect_server", send_sync_link_incorrect_server_ks, "workloop_cb_test_sync_send_kevent_reply",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_n, "workloop_cb_test_sync_send_do_nothing",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_incorrect_server", send_sync_link_incorrect_server_kp, "workloop_cb_test_sync_send_do_nothing_kevent_pthread",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_sync_msg_with_link_check_correct_server", send_sync_link_correct_server_e, "workloop_cb_test_sync_send_do_nothing_exit",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
/*
* Test 22 - 25
*
* Test sequential sync ipc link with server that breaks/preserves the link.
*/
TEST_QOS("server_kevent_id", "qos_client_send_2sync_msg_with_link_check_correct_server", send_2sync_link_correct_server_sksk, "workloop_cb_test_sync_send_reply_kevent_reply_kevent",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_2sync_msg_with_link_check_incorrect_server", send_2sync_link_incorrect_server_kssk, "workloop_cb_test_sync_send_kevent_reply_reply_kevent",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_2sync_msg_with_link_check_incorrect_server", send_2sync_link_incorrect_server_ksks, "workloop_cb_test_sync_send_kevent_reply_kevent_reply",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")
TEST_QOS("server_kevent_id", "qos_client_send_2sync_msg_with_link_check_incorrect_server", send_2sync_link_incorrect_server_skks, "workloop_cb_test_sync_send_reply_kevent_kevent_reply",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default",
QOS_CLASS_DEFAULT, "default")