| Message ID | 1440632493-31734-8-git-send-email-bill.fischofer@linaro.org |
|---|---|
| State | New |
| Headers | show |
How can you prevent that if odp_schedule_group_leave() is supported? A group with no member threads is no different than any other group. It was pointed out in earlier discussions on this subject that it's convenient during initialization to create various scheduler groups of interest because they are needed before queues are created (so the group ID can be passed to queue create). This may well be before any threads are created, so we don't know their thread IDs in advance. Of course you can create scheduler groups with an initial non-zero mask, but I don't see what advantage there is to trying to make this a requirement. On Thu, Aug 27, 2015 at 3:37 AM, Savolainen, Petri (Nokia - FI/Espoo) < petri.savolainen@nokia.com> wrote: > > > > -----Original Message----- > > From: lng-odp [mailto:lng-odp-bounces@lists.linaro.org] On Behalf Of > > ext Bill Fischofer > > Sent: Thursday, August 27, 2015 2:42 AM > > To: lng-odp@lists.linaro.org > > Subject: [lng-odp] [API-NEXT PATCHv3 07/10] validation: schedule: test > > scheduler groups > > > > Signed-off-by: Bill Fischofer <bill.fischofer@linaro.org> > > --- > > test/validation/scheduler/scheduler.c | 188 > > ++++++++++++++++++++++++++++++++++ > > test/validation/scheduler/scheduler.h | 1 + > > 2 files changed, 189 insertions(+) > > > > diff --git a/test/validation/scheduler/scheduler.c > > b/test/validation/scheduler/scheduler.c > > index afd505f..1e9a669 100644 > > --- a/test/validation/scheduler/scheduler.c > > +++ b/test/validation/scheduler/scheduler.c > > @@ -35,6 +35,8 @@ > > #define ENABLE_EXCL_ATOMIC 1 > > > > #define MAGIC 0xdeadbeef > > +#define MAGIC1 0xdeadbeef > > +#define MAGIC2 0xcafef00d > > > > /* Test global variables */ > > typedef struct { > > @@ -162,6 +164,191 @@ void scheduler_test_queue_destroy(void) > > CU_ASSERT_FATAL(odp_pool_destroy(p) == 0); > > } > > > > +void scheduler_test_groups(void) > > +{ > > + odp_pool_t p; > > + odp_pool_param_t params; > > + odp_queue_param_t qp; > > + odp_queue_t queue_grp1, queue_grp2, from; > > + odp_buffer_t buf; > > + odp_event_t ev; > > + uint32_t *u32; > > + int i, j, rc; > > + odp_schedule_sync_t sync[] = {ODP_SCHED_SYNC_NONE, > > + ODP_SCHED_SYNC_ATOMIC, > > + ODP_SCHED_SYNC_ORDERED}; > > + int thr_id = odp_thread_id(); > > + odp_thrmask_t zeromask, mymask, testmask; > > + odp_schedule_group_t mygrp1, mygrp2, lookup; > > + > > + odp_thrmask_zero(&zeromask); > > + odp_thrmask_zero(&mymask); > > + odp_thrmask_set(&mymask, thr_id); > > + > > + /* Can't find a group before we create it */ > > + lookup = odp_schedule_group_lookup("Test Group 1"); > > + CU_ASSERT(lookup == ODP_SCHED_GROUP_INVALID); > > + > > + /* Now create the group */ > > + mygrp1 = odp_schedule_group_create("Test Group 1", &zeromask); > > + CU_ASSERT_FATAL(mygrp1 != ODP_SCHED_GROUP_INVALID); > > + > > + /* Verify we can now find it */ > > + lookup = odp_schedule_group_lookup("Test Group 1"); > > + CU_ASSERT(lookup == mygrp1); > > + > > + /* Threadmask should be retrievable and be what we expect */ > > + rc = odp_schedule_group_thrmask(mygrp1, &testmask); > > + CU_ASSERT(rc == 0); > > + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); > > + > > + /* Now join the group and verify we're part of it */ > > + rc = odp_schedule_group_join(mygrp1, &mymask); > > + CU_ASSERT(rc == 0); > > + > > + rc = odp_schedule_group_thrmask(mygrp1, &testmask); > > + CU_ASSERT(rc == 0); > > + CU_ASSERT(odp_thrmask_isset(&testmask, thr_id)); > > + > > + /* We can't join or leave an unknown group */ > > + rc = odp_schedule_group_join(ODP_SCHED_GROUP_INVALID, &mymask); > > + CU_ASSERT(rc != 0); > > + > > + rc = odp_schedule_group_leave(ODP_SCHED_GROUP_INVALID, &mymask); > > + CU_ASSERT(rc != 0); > > + > > + /* But we can leave our group */ > > + rc = odp_schedule_group_leave(mygrp1, &mymask); > > + CU_ASSERT(rc == 0); > > + > > + rc = odp_schedule_group_thrmask(mygrp1, &testmask); > > + CU_ASSERT(rc == 0); > > + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); > > + > > + /* We shouldn't be able to find our second group before creating > > it */ > > + lookup = odp_schedule_group_lookup("Test Group 2"); > > + CU_ASSERT(lookup == ODP_SCHED_GROUP_INVALID); > > + > > + /* Now create it and verify we can find it */ > > + mygrp2 = odp_schedule_group_create("Test Group 2", &zeromask); > > + CU_ASSERT_FATAL(mygrp2 != ODP_SCHED_GROUP_INVALID); > > + > > + lookup = odp_schedule_group_lookup("Test Group 2"); > > + CU_ASSERT(lookup == mygrp2); > > + > > + /* Verify we're not part of it */ > > + rc = odp_schedule_group_thrmask(mygrp2, &testmask); > > + CU_ASSERT(rc == 0); > > + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); > > + > > + /* Now join the group and verify we're part of it */ > > + rc = odp_schedule_group_join(mygrp2, &mymask); > > + CU_ASSERT(rc == 0); > > + > > + rc = odp_schedule_group_thrmask(mygrp2, &testmask); > > + CU_ASSERT(rc == 0); > > + CU_ASSERT(odp_thrmask_isset(&testmask, thr_id)); > > + > > + /* Now verify scheduler adherence to groups */ > > + odp_queue_param_init(&qp); > > + odp_pool_param_init(¶ms); > > + params.buf.size = 100; > > + params.buf.align = 0; > > + params.buf.num = 2; > > + params.type = ODP_POOL_BUFFER; > > + > > + p = odp_pool_create("sched_group_pool", ¶ms); > > + > > + CU_ASSERT_FATAL(p != ODP_POOL_INVALID); > > + > > + for (i = 0; i < 3; i++) { > > + qp.sched.prio = ODP_SCHED_PRIO_DEFAULT; > > + qp.sched.sync = sync[i]; > > + qp.sched.group = mygrp1; > > + > > + /* Create and populate a group in group 1 */ > > + queue_grp1 = odp_queue_create("sched_group_test_queue 1", > > + ODP_QUEUE_TYPE_SCHED, &qp); > > + CU_ASSERT_FATAL(queue_grp1 != ODP_QUEUE_INVALID); > > + CU_ASSERT_FATAL(odp_queue_sched_group(queue_grp1) == > > mygrp1); > > + > > + buf = odp_buffer_alloc(p); > > + > > + CU_ASSERT_FATAL(buf != ODP_BUFFER_INVALID); > > + > > + u32 = odp_buffer_addr(buf); > > + u32[0] = MAGIC1; > > + > > + ev = odp_buffer_to_event(buf); > > + if (!(CU_ASSERT(odp_queue_enq(queue_grp1, ev) == 0))) > > + odp_buffer_free(buf); > > + > > + /* Now create and populate a queue in group 2 */ > > + qp.sched.group = mygrp2; > > + queue_grp2 = odp_queue_create("sched_group_test_queue_2", > > + ODP_QUEUE_TYPE_SCHED, &qp); > > + CU_ASSERT_FATAL(queue_grp2 != ODP_QUEUE_INVALID); > > + CU_ASSERT_FATAL(odp_queue_sched_group(queue_grp2) == > > mygrp2); > > + > > + buf = odp_buffer_alloc(p); > > + CU_ASSERT_FATAL(buf != ODP_BUFFER_INVALID); > > + > > + u32 = odp_buffer_addr(buf); > > + u32[0] = MAGIC2; > > + > > + ev = odp_buffer_to_event(buf); > > + if (!(CU_ASSERT(odp_queue_enq(queue_grp2, ev) == 0))) > > + odp_buffer_free(buf); > > + > > + /* Scheduler should give us the event from Group 2 */ > > + ev = odp_schedule(&from, ODP_SCHED_WAIT); > > + CU_ASSERT_FATAL(ev != ODP_EVENT_INVALID); > > + CU_ASSERT_FATAL(from == queue_grp2); > > + > > + buf = odp_buffer_from_event(ev); > > + u32 = odp_buffer_addr(buf); > > + > > + CU_ASSERT_FATAL(u32[0] == MAGIC2); > > + > > + odp_buffer_free(buf); > > + > > + /* Scheduler should not return anything now since we're > > + * not in Group 1 and Queue 2 is empty. Do this several > > + * times to confirm. > > + */ > > + > > + for (j = 0; j < 10; j++) { > > + ev = odp_schedule(&from, ODP_SCHED_NO_WAIT); > > + CU_ASSERT_FATAL(ev == ODP_EVENT_INVALID) > > + } > > + > > + /* Now join group 1 and verify we can get the event */ > > + rc = odp_schedule_group_join(mygrp1, &mymask); > > + CU_ASSERT_FATAL(rc == 0); > > This expects that scheduler can have groups that don't have any threads in > it (mask == 0) and can store events enqueued in such a group. It may not > be generally implementable. So, better to move enqueue into grp1 in here > and add "enqueue to empty group" spec (it can be undefined for now) and > test later on if needed. > > > + > > + ev = odp_schedule(&from, ODP_SCHED_WAIT); > > + CU_ASSERT_FATAL(ev != ODP_EVENT_INVALID); > > + CU_ASSERT_FATAL(from == queue_grp1); > > + > > + buf = odp_buffer_from_event(ev); > > + u32 = odp_buffer_addr(buf); > > + > > + CU_ASSERT_FATAL(u32[0] == MAGIC1); > > + > > + odp_buffer_free(buf); > > + > > Need to call schedule() one more time (so that it returns invalid event) > and thus be sure that the current scheduling context have been released > (before continue the test or exit). > > -Petri > > > + /* Leave group 1 for next pass */ > > + rc = odp_schedule_group_leave(mygrp1, &mymask); > > + CU_ASSERT_FATAL(rc == 0); > > + > > + /* Done with queues for this round */ > > + CU_ASSERT_FATAL(odp_queue_destroy(queue_grp1) == 0); > > + CU_ASSERT_FATAL(odp_queue_destroy(queue_grp2) == 0); > > + } > > + > > + CU_ASSERT_FATAL(odp_pool_destroy(p) == 0); > > +} > > + > > static void *schedule_common_(void *arg) > > { > > thread_args_t *args = (thread_args_t *)arg; > > @@ -780,6 +967,7 @@ CU_TestInfo scheduler_suite[] = { > > {"schedule_wait_time", scheduler_test_wait_time}, > > {"schedule_num_prio", scheduler_test_num_prio}, > > {"schedule_queue_destroy", scheduler_test_queue_destroy}, > > + {"schedule_groups", scheduler_test_groups}, > > {"schedule_1q_1t_n", scheduler_test_1q_1t_n}, > > {"schedule_1q_1t_a", scheduler_test_1q_1t_a}, > > {"schedule_1q_1t_o", scheduler_test_1q_1t_o}, > > diff --git a/test/validation/scheduler/scheduler.h > > b/test/validation/scheduler/scheduler.h > > index 31f02ca..eab8787 100644 > > --- a/test/validation/scheduler/scheduler.h > > +++ b/test/validation/scheduler/scheduler.h > > @@ -13,6 +13,7 @@ > > void scheduler_test_wait_time(void); > > void scheduler_test_num_prio(void); > > void scheduler_test_queue_destroy(void); > > +void scheduler_test_groups(void); > > void scheduler_test_1q_1t_n(void); > > void scheduler_test_1q_1t_a(void); > > void scheduler_test_1q_1t_o(void); > > -- > > 2.1.4 > > > > _______________________________________________ > > lng-odp mailing list > > lng-odp@lists.linaro.org > > https://lists.linaro.org/mailman/listinfo/lng-odp >
diff --git a/test/validation/scheduler/scheduler.c b/test/validation/scheduler/scheduler.c index afd505f..1e9a669 100644 --- a/test/validation/scheduler/scheduler.c +++ b/test/validation/scheduler/scheduler.c @@ -35,6 +35,8 @@ #define ENABLE_EXCL_ATOMIC 1 #define MAGIC 0xdeadbeef +#define MAGIC1 0xdeadbeef +#define MAGIC2 0xcafef00d /* Test global variables */ typedef struct { @@ -162,6 +164,191 @@ void scheduler_test_queue_destroy(void) CU_ASSERT_FATAL(odp_pool_destroy(p) == 0); } +void scheduler_test_groups(void) +{ + odp_pool_t p; + odp_pool_param_t params; + odp_queue_param_t qp; + odp_queue_t queue_grp1, queue_grp2, from; + odp_buffer_t buf; + odp_event_t ev; + uint32_t *u32; + int i, j, rc; + odp_schedule_sync_t sync[] = {ODP_SCHED_SYNC_NONE, + ODP_SCHED_SYNC_ATOMIC, + ODP_SCHED_SYNC_ORDERED}; + int thr_id = odp_thread_id(); + odp_thrmask_t zeromask, mymask, testmask; + odp_schedule_group_t mygrp1, mygrp2, lookup; + + odp_thrmask_zero(&zeromask); + odp_thrmask_zero(&mymask); + odp_thrmask_set(&mymask, thr_id); + + /* Can't find a group before we create it */ + lookup = odp_schedule_group_lookup("Test Group 1"); + CU_ASSERT(lookup == ODP_SCHED_GROUP_INVALID); + + /* Now create the group */ + mygrp1 = odp_schedule_group_create("Test Group 1", &zeromask); + CU_ASSERT_FATAL(mygrp1 != ODP_SCHED_GROUP_INVALID); + + /* Verify we can now find it */ + lookup = odp_schedule_group_lookup("Test Group 1"); + CU_ASSERT(lookup == mygrp1); + + /* Threadmask should be retrievable and be what we expect */ + rc = odp_schedule_group_thrmask(mygrp1, &testmask); + CU_ASSERT(rc == 0); + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); + + /* Now join the group and verify we're part of it */ + rc = odp_schedule_group_join(mygrp1, &mymask); + CU_ASSERT(rc == 0); + + rc = odp_schedule_group_thrmask(mygrp1, &testmask); + CU_ASSERT(rc == 0); + CU_ASSERT(odp_thrmask_isset(&testmask, thr_id)); + + /* We can't join or leave an unknown group */ + rc = odp_schedule_group_join(ODP_SCHED_GROUP_INVALID, &mymask); + CU_ASSERT(rc != 0); + + rc = odp_schedule_group_leave(ODP_SCHED_GROUP_INVALID, &mymask); + CU_ASSERT(rc != 0); + + /* But we can leave our group */ + rc = odp_schedule_group_leave(mygrp1, &mymask); + CU_ASSERT(rc == 0); + + rc = odp_schedule_group_thrmask(mygrp1, &testmask); + CU_ASSERT(rc == 0); + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); + + /* We shouldn't be able to find our second group before creating it */ + lookup = odp_schedule_group_lookup("Test Group 2"); + CU_ASSERT(lookup == ODP_SCHED_GROUP_INVALID); + + /* Now create it and verify we can find it */ + mygrp2 = odp_schedule_group_create("Test Group 2", &zeromask); + CU_ASSERT_FATAL(mygrp2 != ODP_SCHED_GROUP_INVALID); + + lookup = odp_schedule_group_lookup("Test Group 2"); + CU_ASSERT(lookup == mygrp2); + + /* Verify we're not part of it */ + rc = odp_schedule_group_thrmask(mygrp2, &testmask); + CU_ASSERT(rc == 0); + CU_ASSERT(!odp_thrmask_isset(&testmask, thr_id)); + + /* Now join the group and verify we're part of it */ + rc = odp_schedule_group_join(mygrp2, &mymask); + CU_ASSERT(rc == 0); + + rc = odp_schedule_group_thrmask(mygrp2, &testmask); + CU_ASSERT(rc == 0); + CU_ASSERT(odp_thrmask_isset(&testmask, thr_id)); + + /* Now verify scheduler adherence to groups */ + odp_queue_param_init(&qp); + odp_pool_param_init(¶ms); + params.buf.size = 100; + params.buf.align = 0; + params.buf.num = 2; + params.type = ODP_POOL_BUFFER; + + p = odp_pool_create("sched_group_pool", ¶ms); + + CU_ASSERT_FATAL(p != ODP_POOL_INVALID); + + for (i = 0; i < 3; i++) { + qp.sched.prio = ODP_SCHED_PRIO_DEFAULT; + qp.sched.sync = sync[i]; + qp.sched.group = mygrp1; + + /* Create and populate a group in group 1 */ + queue_grp1 = odp_queue_create("sched_group_test_queue 1", + ODP_QUEUE_TYPE_SCHED, &qp); + CU_ASSERT_FATAL(queue_grp1 != ODP_QUEUE_INVALID); + CU_ASSERT_FATAL(odp_queue_sched_group(queue_grp1) == mygrp1); + + buf = odp_buffer_alloc(p); + + CU_ASSERT_FATAL(buf != ODP_BUFFER_INVALID); + + u32 = odp_buffer_addr(buf); + u32[0] = MAGIC1; + + ev = odp_buffer_to_event(buf); + if (!(CU_ASSERT(odp_queue_enq(queue_grp1, ev) == 0))) + odp_buffer_free(buf); + + /* Now create and populate a queue in group 2 */ + qp.sched.group = mygrp2; + queue_grp2 = odp_queue_create("sched_group_test_queue_2", + ODP_QUEUE_TYPE_SCHED, &qp); + CU_ASSERT_FATAL(queue_grp2 != ODP_QUEUE_INVALID); + CU_ASSERT_FATAL(odp_queue_sched_group(queue_grp2) == mygrp2); + + buf = odp_buffer_alloc(p); + CU_ASSERT_FATAL(buf != ODP_BUFFER_INVALID); + + u32 = odp_buffer_addr(buf); + u32[0] = MAGIC2; + + ev = odp_buffer_to_event(buf); + if (!(CU_ASSERT(odp_queue_enq(queue_grp2, ev) == 0))) + odp_buffer_free(buf); + + /* Scheduler should give us the event from Group 2 */ + ev = odp_schedule(&from, ODP_SCHED_WAIT); + CU_ASSERT_FATAL(ev != ODP_EVENT_INVALID); + CU_ASSERT_FATAL(from == queue_grp2); + + buf = odp_buffer_from_event(ev); + u32 = odp_buffer_addr(buf); + + CU_ASSERT_FATAL(u32[0] == MAGIC2); + + odp_buffer_free(buf); + + /* Scheduler should not return anything now since we're + * not in Group 1 and Queue 2 is empty. Do this several + * times to confirm. + */ + + for (j = 0; j < 10; j++) { + ev = odp_schedule(&from, ODP_SCHED_NO_WAIT); + CU_ASSERT_FATAL(ev == ODP_EVENT_INVALID) + } + + /* Now join group 1 and verify we can get the event */ + rc = odp_schedule_group_join(mygrp1, &mymask); + CU_ASSERT_FATAL(rc == 0); + + ev = odp_schedule(&from, ODP_SCHED_WAIT); + CU_ASSERT_FATAL(ev != ODP_EVENT_INVALID); + CU_ASSERT_FATAL(from == queue_grp1); + + buf = odp_buffer_from_event(ev); + u32 = odp_buffer_addr(buf); + + CU_ASSERT_FATAL(u32[0] == MAGIC1); + + odp_buffer_free(buf); + + /* Leave group 1 for next pass */ + rc = odp_schedule_group_leave(mygrp1, &mymask); + CU_ASSERT_FATAL(rc == 0); + + /* Done with queues for this round */ + CU_ASSERT_FATAL(odp_queue_destroy(queue_grp1) == 0); + CU_ASSERT_FATAL(odp_queue_destroy(queue_grp2) == 0); + } + + CU_ASSERT_FATAL(odp_pool_destroy(p) == 0); +} + static void *schedule_common_(void *arg) { thread_args_t *args = (thread_args_t *)arg; @@ -780,6 +967,7 @@ CU_TestInfo scheduler_suite[] = { {"schedule_wait_time", scheduler_test_wait_time}, {"schedule_num_prio", scheduler_test_num_prio}, {"schedule_queue_destroy", scheduler_test_queue_destroy}, + {"schedule_groups", scheduler_test_groups}, {"schedule_1q_1t_n", scheduler_test_1q_1t_n}, {"schedule_1q_1t_a", scheduler_test_1q_1t_a}, {"schedule_1q_1t_o", scheduler_test_1q_1t_o}, diff --git a/test/validation/scheduler/scheduler.h b/test/validation/scheduler/scheduler.h index 31f02ca..eab8787 100644 --- a/test/validation/scheduler/scheduler.h +++ b/test/validation/scheduler/scheduler.h @@ -13,6 +13,7 @@ void scheduler_test_wait_time(void); void scheduler_test_num_prio(void); void scheduler_test_queue_destroy(void); +void scheduler_test_groups(void); void scheduler_test_1q_1t_n(void); void scheduler_test_1q_1t_a(void); void scheduler_test_1q_1t_o(void);
Signed-off-by: Bill Fischofer <bill.fischofer@linaro.org> --- test/validation/scheduler/scheduler.c | 188 ++++++++++++++++++++++++++++++++++ test/validation/scheduler/scheduler.h | 1 + 2 files changed, 189 insertions(+)