michaelh/zephyr
1
0
Fork 0
Code Releases Activity

tests: latency_measure: Misc updates

Browse source 
Adds userspace configuration to testcase.yaml.

Updates the README.rst with output from runs on a frdm_k64f board
as not only have the descriptions changed, but there is now
additional output for userspace configurations.

Signed-off-by: Peter Mitsis <peter.mitsis@intel.com>
This commit is contained in:
Peter Mitsis 2023-10-06 10:36:23 -04:00 • committed by Carles Cufí
commit 6899c0d73a
4 changed files with 140 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

127
tests/benchmarks/latency_measure/README.rst
View file

@ -1,44 +1,109 @@
Latency Measurements Latency Measurements
#################### ####################
This benchmark measures the latency of selected kernel capabilities, including: This benchmark measures the average latency of selected kernel capabilities,
including:
* Context switch time between preemptive threads using k_yield
* Measure time to switch from ISR back to interrupted thread * Context switch time between cooperative threads using k_yield
* Measure time from ISR to executing a different thread (rescheduled) * Time to switch from ISR back to interrupted thread
* Measure average time to signal a semaphore then test that semaphore * Time from ISR to executing a different thread (rescheduled)
* Measure average time to signal a semaphore then test that semaphore with a context switch * Times to signal a semaphore then test that semaphore
* Measure average time to lock a mutex then unlock that mutex * Times to signal a semaphore then test that semaphore with a context switch
* Measure average context switch time between threads using (k_yield) * Times to lock a mutex then unlock that mutex
* Measure average context switch time between threads (coop)
* Time it takes to suspend a thread
* Time it takes to resume a suspended thread
* Time it takes to create a new thread (without starting it) * Time it takes to create a new thread (without starting it)
* Time it takes to start a newly created thread * Time it takes to start a newly created thread
* Time it takes to suspend a thread
* Time it takes to resume a suspended thread
* Time it takes to abort a thread
* Measure average time to alloc memory from heap then free that memory * Measure average time to alloc memory from heap then free that memory
When userspace is enabled using the prj_user.conf configuration file, this benchmark will
where possible, also test the above capabilities using various configurations involving user
threads:
Sample output of the benchmark:: * Kernel thread to kernel thread
* Kernel thread to user thread
* User thread to kernel thread
* User thread to user thread
*** Booting Zephyr OS build zephyr-v2.6.0-1119-g378a1e082ac5 *** Sample output of the benchmark (without userspace enabled)::
*** Booting Zephyr OS build v3.5.0-rc1-139-gdab69aeed11d ***
START - Time Measurement START - Time Measurement
Timing results: Clock frequency: 1000 MHz Timing results: Clock frequency: 120 MHz
Average thread context switch using yield : 9060 cycles , 9060 ns Preemptive threads ctx switch via k_yield (K -> K) : 519 cycles , 4325 ns :
Average context switch time between threads (coop) : 9503 cycles , 9503 ns Cooperative threads ctx switch via k_yield (K -> K) : 519 cycles , 4325 ns :
Switch from ISR back to interrupted thread : 14208 cycles , 14208 ns Switch from ISR back to interrupted thread : 508 cycles , 4241 ns :
Time from ISR to executing a different thread : 9664 cycles , 9664 ns Switch from ISR to another thread (kernel) : 554 cycles , 4616 ns :
Time to create a thread (without start) : 3968 cycles , 3968 ns Create kernel thread from kernel thread : 396 cycles , 3308 ns :
Time to start a thread : 12064 cycles , 12064 ns Start kernel thread from kernel thread : 603 cycles , 5033 ns :
Time to suspend a thread : 12640 cycles , 12640 ns Suspend kernel thread from kernel thread : 599 cycles , 4992 ns :
Time to resume a thread : 12096 cycles , 12096 ns Resume kernel thread from kernel thread : 547 cycles , 4558 ns :
Time to abort a thread (not running) : 2208 cycles , 2208 ns Abort kernel thread from kernel thread : 339 cycles , 2825 ns :
Average semaphore signal time : 8928 cycles , 8928 ns Give a semaphore (no waiters) from kernel thread : 134 cycles , 1116 ns :
Average semaphore test time : 2048 cycles , 2048 ns Take a semaphore (no blocking) from kernel thread : 53 cycles , 441 ns :
Semaphore take time (context switch) : 13472 cycles , 13472 ns Take a semaphore (context switch K -> K) : 689 cycles , 5742 ns :
Semaphore give time (context switch) : 18400 cycles , 18400 ns Give a semaphore (context switch K -> K) : 789 cycles , 6575 ns :
Average time to lock a mutex : 3072 cycles , 3072 ns Lock a mutex from kernel thread : 94 cycles , 783 ns :
Average time to unlock a mutex : 9251 cycles , 9251 ns Unlock a mutex from kernel thread : 24 cycles , 200 ns :
Average time for heap malloc : 13056 cycles , 13056 ns Average time for heap malloc : 620 cycles , 5166 ns :
Average time for heap free : 7776 cycles , 7776 ns Average time for heap free : 431 cycles , 3591 ns :
===================================================================
PROJECT EXECUTION SUCCESSFUL
Sample output of the benchmark (with userspace enabled)::
*** Booting Zephyr OS build v3.5.0-rc1-139-gdab69aeed11d ***
START - Time Measurement
Timing results: Clock frequency: 120 MHz
Preemptive threads ctx switch via k_yield (K -> K) : 1195 cycles , 9958 ns :
Preemptive threads ctx switch via k_yield (U -> U) : 1485 cycles , 12379 ns :
Preemptive threads ctx switch via k_yield (K -> U) : 1390 cycles , 11587 ns :
Preemptive threads ctx switch via k_yield (U -> K) : 1289 cycles , 10749 ns :
Cooperative threads ctx switch via k_yield (K -> K) : 1185 cycles , 9875 ns :
Cooperative threads ctx switch via k_yield (U -> U) : 1475 cycles , 12295 ns :
Cooperative threads ctx switch via k_yield (K -> U) : 1380 cycles , 11504 ns :
Cooperative threads ctx switch via k_yield (U -> K) : 1280 cycles , 10666 ns :
Switch from ISR back to interrupted thread : 1130 cycles , 9416 ns :
Switch from ISR to another thread (kernel) : 1184 cycles , 9874 ns :
Switch from ISR to another thread (user) : 1390 cycles , 11583 ns :
Create kernel thread from kernel thread : 985 cycles , 8208 ns :
Start kernel thread from kernel thread : 1275 cycles , 10625 ns :
Suspend kernel thread from kernel thread : 1220 cycles , 10167 ns :
Resume kernel thread from kernel thread : 1193 cycles , 9942 ns :
Abort kernel thread from kernel thread : 2555 cycles , 21292 ns :
Create user thread from kernel thread : 849 cycles , 7083 ns :
Start user thread from kernel thread : 6715 cycles , 55960 ns :
Suspend user thread from kernel thread : 1585 cycles , 13208 ns :
Resume user thread from kernel thread : 1383 cycles , 11525 ns :
Abort user thread from kernel thread : 2420 cycles , 20167 ns :
Create user thread from user thread : 2110 cycles , 17584 ns :
Start user thread from user thread : 7070 cycles , 58919 ns :
Suspend user thread from user thread : 1784 cycles , 14874 ns :
Resume user thread from user thread : 1740 cycles , 14502 ns :
Abort user thread from user thread : 3000 cycles , 25000 ns :
Start kernel thread from user thread : 1630 cycles , 13583 ns :
Suspend kernel thread from user thread : 1420 cycles , 11833 ns :
Resume kernel thread from user thread : 1550 cycles , 12917 ns :
Abort kernel thread from user thread : 3135 cycles , 26125 ns :
Give a semaphore (no waiters) from kernel thread : 160 cycles , 1333 ns :
Take a semaphore (no blocking) from kernel thread : 95 cycles , 791 ns :
Give a semaphore (no waiters) from user thread : 380 cycles , 3166 ns :
Take a semaphore (no blocking) from user thread : 315 cycles , 2625 ns :
Take a semaphore (context switch K -> K) : 1340 cycles , 11167 ns :
Give a semaphore (context switch K -> K) : 1460 cycles , 12167 ns :
Take a semaphore (context switch K -> U) : 1540 cycles , 12838 ns :
Give a semaphore (context switch U -> K) : 1800 cycles , 15000 ns :
Take a semaphore (context switch U -> K) : 1690 cycles , 14084 ns :
Give a semaphore (context switch K -> U) : 1650 cycles , 13750 ns :
Take a semaphore (context switch U -> U) : 1890 cycles , 15756 ns :
Give a semaphore (context switch U -> U) : 1990 cycles , 16583 ns :
Lock a mutex from kernel thread : 105 cycles , 875 ns :
Unlock a mutex from kernel thread : 17 cycles , 141 ns :
Lock a mutex from user thread : 330 cycles , 2750 ns :
Unlock a mutex from user thread : 255 cycles , 2125 ns :
Average time for heap malloc : 606 cycles , 5058 ns :
Average time for heap free : 422 cycles , 3516 ns :
=================================================================== ===================================================================
PROJECT EXECUTION SUCCESSFUL PROJECT EXECUTION SUCCESSFUL

1
tests/benchmarks/latency_measure/prj.conf
View file

@ -2,7 +2,6 @@ CONFIG_TEST=y
# eliminate timer interrupts during the benchmark # eliminate timer interrupts during the benchmark
CONFIG_SYS_CLOCK_TICKS_PER_SEC=1 CONFIG_SYS_CLOCK_TICKS_PER_SEC=1
CONFIG_TICKLESS_KERNEL=n
# We use irq_offload(), enable it # We use irq_offload(), enable it
CONFIG_IRQ_OFFLOAD=y CONFIG_IRQ_OFFLOAD=y

27
tests/benchmarks/latency_measure/prj_user.conf Normal file
View file

@ -0,0 +1,27 @@
CONFIG_TEST=y
# eliminate timer interrupts during the benchmark
CONFIG_SYS_CLOCK_TICKS_PER_SEC=1
# We use irq_offload(), enable it
CONFIG_IRQ_OFFLOAD=y
# Reduce memory/code footprint
CONFIG_BT=n
CONFIG_FORCE_NO_ASSERT=y
CONFIG_TEST_HW_STACK_PROTECTION=n
# Disable HW Stack Protection (see #28664)
CONFIG_HW_STACK_PROTECTION=n
CONFIG_COVERAGE=n
# Disable system power management
CONFIG_PM=n
# Can only run under 1 CPU
CONFIG_MP_MAX_NUM_CPUS=1
CONFIG_TIMING_FUNCTIONS=y
CONFIG_HEAP_MEM_POOL_SIZE=2048
CONFIG_APPLICATION_DEFINED_SYSCALL=y
CONFIG_USERSPACE=y

17
tests/benchmarks/latency_measure/testcase.yaml
View file

@ -36,3 +36,20 @@ tests:
regex: "(?P<metric>.*):(?P<cycles>.*) cycles ,(?P<nanoseconds>.*) ns" regex: "(?P<metric>.*):(?P<cycles>.*) cycles ,(?P<nanoseconds>.*) ns"
regex: regex:
- "PROJECT EXECUTION SUCCESSFUL" - "PROJECT EXECUTION SUCCESSFUL"
# Obtain the benchmark results for various user thread / kernel thread
# configurations on platforms that support user space.
benchmark.user.latency:
filter: CONFIG_ARCH_HAS_USERSPACE
timeout: 300
extra_args: CONF_FILE=prj_user.conf
harness: console
integration_platforms:
- qemu_x86
- qemu_cortex_a53
harness_config:
type: one_line
record:
regex: "(?P<metric>.*):(?P<cycles>.*) cycles ,(?P<nanoseconds>.*) ns"
regex:
- "PROJECT EXECUTION SUCCESSFUL"