[RFC PATCH 0/5] selftests/damon: add Python modules for DAMON control and functionality tests
DAMON provides almost all control to the user via its sysfs interface. For that, the interface provides plenty of files and hierarchies. The interface is simple enough to be controlled by shell commands including 'cat', 'echo', and redirection. However, due to the number of files and the hierarchies, doing that repeatedly is quite tedious. As a result, DAMON selftests are containing only simple test cases rather than real functionality tests. Having a wrapper script that can be reused to implement more functionality tests could be helpful. Writing such wrapper with shell script might be challenging and not easy to further maintain and extend for future DAMON interface extensions, though.
To this end, implement a Python-written DAMON sysfs interface wrapper that could be easily managed and extended for future DAMON interface extensions. Further implement one simple functionality test and a corner case regression test for a previously found bug, using the wrapper module. In fact, the bug was found by the test this patchset is introducing.
Note that the Python wrapper is not supporting full features of DAMON interface, but only some of those that essential for the tests that this patchset is introducing. The wrapper would extended to support more features, but only with essential ones for such future tests. The wrapper will hence keep being simple, small, and constrained. For convenient and general use cases of DAMON, users should use DAMON user-space tools for such purpose, like damo[1].
[1] https://github.com/damonitor/damo
Patches Sequence ----------------
This patchset is constructed with five patches. The first three patches implement the Python-written DAMON sysfs interface wrapper in small steps. The basic data structure (first patch), kdamond startup command (second patch), and finally DAMOS tried bytes command (third patch).
Then two patches for adding selftests using the wrapper follows. The fourth patch implements a basic functionality test of DAMON for working set estimation accuracy. Finally, the fifth patch implements a corner case test for a previously found bug.
SeongJae Park (5): selftests/damon: add a DAMON interface wrapper python module selftests/damon/_damon: implement sysfs-based kdamonds start function selftests/damon/_damon: implement sysfs updat_schemes_tried_bytes command selftests/damon: add a test for update_schemes_tried_regions sysfs command selftests/damon: add a test for update_schemes_tried_regions hang bug
tools/testing/selftests/damon/Makefile | 3 + tools/testing/selftests/damon/_damon.py | 322 ++++++++++++++++++ tools/testing/selftests/damon/access_memory.c | 41 +++ ...sysfs_update_schemes_tried_regions_hang.py | 33 ++ ...te_schemes_tried_regions_wss_estimation.py | 48 +++ 5 files changed, 447 insertions(+) create mode 100644 tools/testing/selftests/damon/_damon.py create mode 100644 tools/testing/selftests/damon/access_memory.c create mode 100755 tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_hang.py create mode 100755 tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_wss_estimation.py
base-commit: 1be383c41197b82cfd51b2edc7ee515c0b786496
DAMON sysfs interface supports almost every DAMON functionality. For that, it provides a number of files and hierarchies that not simple to be manually manipulated, or writing shell script every time. For the reason, DAMON is not having good functionality selftests. Adding an existing DAMON user space tool and using it could also be a way, but it would add unnecessarily huge change that not essential for the test purpose. Write a minimum python module that can further extended to be used as an easy-to-use DAMON functionality test purpose DAMON user interface wrapper.
Note that this commit is implementing only basic data structures for representing the sysfs files. More operations will be implemented by following commits, as needed for specific real test cases.
Signed-off-by: SeongJae Park sj@kernel.org --- tools/testing/selftests/damon/_damon.py | 102 ++++++++++++++++++++++++ 1 file changed, 102 insertions(+) create mode 100644 tools/testing/selftests/damon/_damon.py
diff --git a/tools/testing/selftests/damon/_damon.py b/tools/testing/selftests/damon/_damon.py new file mode 100644 index 000000000000..78101846ab66 --- /dev/null +++ b/tools/testing/selftests/damon/_damon.py @@ -0,0 +1,102 @@ +# SPDX-License-Identifier: GPL-2.0 + +class DamosAccessPattern: + size = None + nr_accesses = None + age = None + scheme = None + + def __init__(self, size=None, nr_accesses=None, age=None): + self.size = size + self.nr_accesses = nr_accesses + self.age = age + + if self.size == None: + self.size = [0, 2**64 - 1] + if self.nr_accesses == None: + self.nr_accesses = [0, 2**64 - 1] + if self.age == None: + self.age = [0, 2**64 - 1] + +class Damos: + action = None + access_pattern = None + # todo: Support quotas, watermarks, stats, tried_regions + idx = None + context = None + + def __init__(self, action='stat', access_pattern=DamosAccessPattern()): + self.action = action + self.access_pattern = access_pattern + self.access_pattern.scheme = self + +class DamonTarget: + pid = None + # todo: Support target regions if test is made + idx = None + context = None + + def __init__(self, pid): + self.pid = pid + +class DamonAttrs: + sample_us = None + aggr_us = None + update_us = None + min_nr_regions = None + max_nr_regions = None + context = None + + def __init__(self, sample_us=5000, aggr_us=100000, update_us=1000000, + min_nr_regions=10, max_nr_regions=1000): + self.sample_us = sample_us + self.aggr_us = aggr_us + self.update_us = update_us + self.min_nr_regions = min_nr_regions + self.max_nr_regions = max_nr_regions + +class DamonCtx: + ops = None + monitoring_attrs = None + targets = None + schemes = None + kdamond = None + idx = None + + def __init__(self, ops='paddr', monitoring_attrs=DamonAttrs(), targets=[], + schemes=[]): + self.ops = ops + self.monitoring_attrs = monitoring_attrs + self.monitoring_attrs.context = self + + self.targets = targets + for idx, target in enumerate(self.targets): + target.idx = idx + target.context = self + + self.schemes = schemes + for idx, scheme in enumerate(self.schemes): + scheme.idx = idx + scheme.context = self + +class Kdamond: + state = None + pid = None + contexts = None + idx = None # index of this kdamond between siblings + kdamonds = None # parent + + def __init__(self, contexts=[]): + self.contexts = contexts + for idx, context in enumerate(self.contexts): + context.idx = idx + context.kdamond = self + +class Kdamonds: + kdamonds = [] + + def __init__(self, kdamonds=[]): + self.kdamonds = kdamonds + for idx, kdamond in enumerate(self.kdamonds): + kdamond.idx = idx + kdamond.kdamonds = self
Extend the test-writing-purpose DAMON sysfs wrapper Python module to support turning DAMON on with basic parameter inputs.
Signed-off-by: SeongJae Park sj@kernel.org --- tools/testing/selftests/damon/_damon.py | 206 ++++++++++++++++++++++++ 1 file changed, 206 insertions(+)
diff --git a/tools/testing/selftests/damon/_damon.py b/tools/testing/selftests/damon/_damon.py index 78101846ab66..6b99f87a5f1e 100644 --- a/tools/testing/selftests/damon/_damon.py +++ b/tools/testing/selftests/damon/_damon.py @@ -1,5 +1,28 @@ # SPDX-License-Identifier: GPL-2.0
+import os + +sysfs_root = '/sys/kernel/mm/damon/admin' + +def write_file(path, string): + "Returns error string if failed, or None otherwise" + string = '%s' % string + try: + with open(path, 'w') as f: + f.write(string) + except Exception as e: + return '%s' % e + return None + +def read_file(path): + '''Returns the read content and error string. The read content is None if + the reading failed''' + try: + with open(path, 'r') as f: + return f.read(), None + except Exception as e: + return None, '%s' % e + class DamosAccessPattern: size = None nr_accesses = None @@ -18,6 +41,35 @@ class DamosAccessPattern: if self.age == None: self.age = [0, 2**64 - 1]
+ def sysfs_dir(self): + return os.path.join(self.scheme.sysfs_dir(), 'access_pattern') + + def stage(self): + err = write_file( + os.path.join(self.sysfs_dir(), 'sz', 'min'), self.size[0]) + if err != None: + return err + err = write_file( + os.path.join(self.sysfs_dir(), 'sz', 'max'), self.size[1]) + if err != None: + return err + err = write_file(os.path.join(self.sysfs_dir(), 'nr_accesses', 'min'), + self.nr_accesses[0]) + if err != None: + return err + err = write_file(os.path.join(self.sysfs_dir(), 'nr_accesses', 'max'), + self.nr_accesses[1]) + if err != None: + return err + err = write_file( + os.path.join(self.sysfs_dir(), 'age', 'min'), self.age[0]) + if err != None: + return err + err = write_file( + os.path.join(self.sysfs_dir(), 'age', 'max'), self.age[1]) + if err != None: + return err + class Damos: action = None access_pattern = None @@ -30,6 +82,39 @@ class Damos: self.access_pattern = access_pattern self.access_pattern.scheme = self
+ def sysfs_dir(self): + return os.path.join( + self.context.sysfs_dir(), 'schemes', '%d' % self.idx) + + def stage(self): + err = write_file(os.path.join(self.sysfs_dir(), 'action'), self.action) + if err != None: + return err + err = self.access_pattern.stage() + if err != None: + return err + + # disable quotas + err = write_file(os.path.join(self.sysfs_dir(), 'quotas', 'ms'), '0') + if err != None: + return err + err = write_file( + os.path.join(self.sysfs_dir(), 'quotas', 'bytes'), '0') + if err != None: + return err + + # disable watermarks + err = write_file( + os.path.join(self.sysfs_dir(), 'watermarks', 'metric'), 'none') + if err != None: + return err + + # disable filters + err = write_file( + os.path.join(self.sysfs_dir(), 'filters', 'nr_filters'), '0') + if err != None: + return err + class DamonTarget: pid = None # todo: Support target regions if test is made @@ -39,6 +124,18 @@ class DamonTarget: def __init__(self, pid): self.pid = pid
+ def sysfs_dir(self): + return os.path.join( + self.context.sysfs_dir(), 'targets', '%d' % self.idx) + + def stage(self): + err = write_file( + os.path.join(self.sysfs_dir(), 'regions', 'nr_regions'), '0') + if err != None: + return err + return write_file( + os.path.join(self.sysfs_dir(), 'pid_target'), self.pid) + class DamonAttrs: sample_us = None aggr_us = None @@ -55,6 +152,40 @@ class DamonAttrs: self.min_nr_regions = min_nr_regions self.max_nr_regions = max_nr_regions
+ def interval_sysfs_dir(self): + return os.path.join(self.context.sysfs_dir(), 'monitoring_attrs', + 'intervals') + + def nr_regions_range_sysfs_dir(self): + return os.path.join(self.context.sysfs_dir(), 'monitoring_attrs', + 'nr_regions') + + def stage(self): + err = write_file(os.path.join(self.interval_sysfs_dir(), 'sample_us'), + self.sample_us) + if err != None: + return err + err = write_file(os.path.join(self.interval_sysfs_dir(), 'aggr_us'), + self.aggr_us) + if err != None: + return err + err = write_file(os.path.join(self.interval_sysfs_dir(), 'update_us'), + self.update_us) + if err != None: + return err + + err = write_file( + os.path.join(self.nr_regions_range_sysfs_dir(), 'min'), + self.min_nr_regions) + if err != None: + return err + + err = write_file( + os.path.join(self.nr_regions_range_sysfs_dir(), 'max'), + self.max_nr_regions) + if err != None: + return err + class DamonCtx: ops = None monitoring_attrs = None @@ -79,6 +210,46 @@ class DamonCtx: scheme.idx = idx scheme.context = self
+ def sysfs_dir(self): + return os.path.join(self.kdamond.sysfs_dir(), 'contexts', + '%d' % self.idx) + + def stage(self): + err = write_file( + os.path.join(self.sysfs_dir(), 'operations'), self.ops) + if err != None: + return err + err = self.monitoring_attrs.stage() + if err != None: + return err + + nr_targets_file = os.path.join( + self.sysfs_dir(), 'targets', 'nr_targets') + content, err = read_file(nr_targets_file) + if err != None: + return err + if int(content) != len(self.targets): + err = write_file(nr_targets_file, '%d' % len(self.targets)) + if err != None: + return err + for target in self.targets: + err = target.stage() + if err != None: + return err + + nr_schemes_file = os.path.join( + self.sysfs_dir(), 'schemes', 'nr_schemes') + content, err = read_file(nr_schemes_file) + if int(content) != len(self.schemes): + err = write_file(nr_schemes_file, '%d' % len(self.schemes)) + if err != None: + return err + for scheme in self.schemes: + err = scheme.stage() + if err != None: + return err + return None + class Kdamond: state = None pid = None @@ -92,6 +263,27 @@ class Kdamond: context.idx = idx context.kdamond = self
+ def sysfs_dir(self): + return os.path.join(self.kdamonds.sysfs_dir(), '%d' % self.idx) + + def start(self): + nr_contexts_file = os.path.join(self.sysfs_dir(), + 'contexts', 'nr_contexts') + content, err = read_file(nr_contexts_file) + if err != None: + return err + if int(content) != len(self.contexts): + err = write_file(nr_contexts_file, '%d' % len(self.contexts)) + if err != None: + return err + + for context in self.contexts: + err = context.stage() + if err != None: + return err + err = write_file(os.path.join(self.sysfs_dir(), 'state'), 'on') + return err + class Kdamonds: kdamonds = []
@@ -100,3 +292,17 @@ class Kdamonds: for idx, kdamond in enumerate(self.kdamonds): kdamond.idx = idx kdamond.kdamonds = self + + def sysfs_dir(self): + return os.path.join(sysfs_root, 'kdamonds') + + def start(self): + err = write_file(os.path.join(self.sysfs_dir(), 'nr_kdamonds'), + '%s' % len(self.kdamonds)) + if err != None: + return err + for kdamond in self.kdamonds: + err = kdamond.start() + if err != None: + return err + return None
Implement update_schemes_tried_bytes command in the test-writing-purpose DAMON sysfs interface wrapper Python module. It is not only making the update, but also read the updated value from the sysfs interface and store on the Kdamond python objects so that user of the module can get it by reading the field of the object.
Signed-off-by: SeongJae Park sj@kernel.org --- tools/testing/selftests/damon/_damon.py | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
diff --git a/tools/testing/selftests/damon/_damon.py b/tools/testing/selftests/damon/_damon.py index 6b99f87a5f1e..e98cf4b6a4b7 100644 --- a/tools/testing/selftests/damon/_damon.py +++ b/tools/testing/selftests/damon/_damon.py @@ -76,6 +76,7 @@ class Damos: # todo: Support quotas, watermarks, stats, tried_regions idx = None context = None + tried_bytes = None
def __init__(self, action='stat', access_pattern=DamosAccessPattern()): self.action = action @@ -284,6 +285,19 @@ class Kdamond: err = write_file(os.path.join(self.sysfs_dir(), 'state'), 'on') return err
+ def update_schemes_tried_bytes(self): + err = write_file(os.path.join(self.sysfs_dir(), 'state'), + 'update_schemes_tried_bytes') + if err != None: + return err + for context in self.contexts: + for scheme in context.schemes: + content, err = read_file(os.path.join(scheme.sysfs_dir(), + 'tried_regions', 'total_bytes')) + if err != None: + return err + scheme.tried_bytes = int(content) + class Kdamonds: kdamonds = []
Add a basic test for verifying accuracy of DAMON's access monitoring functionality. The test starts a program of artificial access pattern, monitor the accesses using DAMON, and check if DAMON gets working set size with reasonable amount of error rate. Note that the acceptable error rate is set with only a naive assumption. Failure of this test may not always mean DAMON is broken, and we could adjust it in future.
Signed-off-by: SeongJae Park sj@kernel.org --- tools/testing/selftests/damon/Makefile | 2 + tools/testing/selftests/damon/access_memory.c | 41 ++++++++++++++++ ...te_schemes_tried_regions_wss_estimation.py | 48 +++++++++++++++++++ 3 files changed, 91 insertions(+) create mode 100644 tools/testing/selftests/damon/access_memory.c create mode 100755 tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_wss_estimation.py
diff --git a/tools/testing/selftests/damon/Makefile b/tools/testing/selftests/damon/Makefile index d2105d41ea25..1363987709c6 100644 --- a/tools/testing/selftests/damon/Makefile +++ b/tools/testing/selftests/damon/Makefile @@ -4,6 +4,7 @@ TEST_GEN_FILES += huge_count_read_write TEST_GEN_FILES += dbgfs_target_ids_read_before_terminate_race TEST_GEN_FILES += dbgfs_target_ids_pid_leak +TEST_GEN_FILES += access_memory
TEST_FILES = _chk_dependency.sh _debugfs_common.sh TEST_PROGS = debugfs_attrs.sh debugfs_schemes.sh debugfs_target_ids.sh @@ -11,6 +12,7 @@ TEST_PROGS += debugfs_empty_targets.sh debugfs_huge_count_read_write.sh TEST_PROGS += debugfs_duplicate_context_creation.sh TEST_PROGS += debugfs_rm_non_contexts.sh TEST_PROGS += sysfs.sh sysfs_update_removed_scheme_dir.sh +TEST_PROGS += sysfs_update_schemes_tried_regions_wss_estimation.py TEST_PROGS += reclaim.sh lru_sort.sh TEST_PROGS += dbgfs_target_ids_read_before_terminate_race.sh TEST_PROGS += dbgfs_target_ids_pid_leak.sh diff --git a/tools/testing/selftests/damon/access_memory.c b/tools/testing/selftests/damon/access_memory.c new file mode 100644 index 000000000000..585a2fa54329 --- /dev/null +++ b/tools/testing/selftests/damon/access_memory.c @@ -0,0 +1,41 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Artificial memory access program for testing DAMON. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <time.h> + +int main(int argc, char *argv[]) +{ + char **regions; + clock_t start_clock; + int nr_regions; + int sz_region; + int access_time_ms; + int i; + + if (argc != 4) { + printf("Usage: %s <number> <size (bytes)> <time (ms)>\n", + argv[0]); + return -1; + } + + nr_regions = atoi(argv[1]); + sz_region = atoi(argv[2]); + access_time_ms = atoi(argv[3]); + + regions = malloc(sizeof(*regions) * nr_regions); + for (i = 0; i < nr_regions; i++) + regions[i] = malloc(sz_region); + + for (i = 0; i < nr_regions; i++) { + start_clock = clock(); + while ((clock() - start_clock) * 1000 / CLOCKS_PER_SEC < + access_time_ms) + memset(regions[i], i, 1024 * 1024 * 10); + } + return 0; +} diff --git a/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_wss_estimation.py b/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_wss_estimation.py new file mode 100755 index 000000000000..2aa66f4ad2dc --- /dev/null +++ b/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_wss_estimation.py @@ -0,0 +1,48 @@ +#!/usr/bin/env python3 +# SPDX-License-Identifier: GPL-2.0 + +import subprocess +import time + +import _damon + +def main(): + # access three 10 MiB memory regions, 1 second per each + sz_region = 10 * 1024 * 1024 + proc = subprocess.Popen(['./access_memory', '3', '%d' % sz_region, '1000']) + kdamonds = _damon.Kdamonds([_damon.Kdamond( + contexts=[_damon.DamonCtx( + ops='vaddr', + targets=[_damon.DamonTarget(pid=proc.pid)], + schemes=[_damon.Damos( + access_pattern=_damon.DamosAccessPattern( + nr_accesses=[1, 20]))] # schemes + )] # contexts + )]) # kdamonds + + err = kdamonds.start() + if err != None: + print('kdmaond start failed: %s' % err) + exit(1) + + wss_collected = [] + while proc.poll() == None: + time.sleep(0.1) + err = kdamonds.kdamonds[0].update_schemes_tried_bytes() + if err != None: + print('tried bytes update failed: %s' % err) + exit(1) + + wss_collected.append( + kdamonds.kdamonds[0].contexts[0].schemes[0].tried_bytes) + + wss_collected.sort() + for percentile in [25, 75]: + sample = wss_collected[int(len(wss_collected) * percentile / 100)] + error_rate = abs(sample - sz_region) / sz_region + print('%d-th percentile error %f' % (percentile, error_rate)) + if error_rate > 0.5: + exit(1) + +if __name__ == '__main__': + main()
The update_schemees_tried_{regions,bytes} command was able to be indefinitely hang in some corner cases. It has fixed by introducing a timeout for the command[1]. Add a test for the corner case to not introduce the problem again.
[1] https://lore.kernel.org/damon/20231124213840.39157-1-sj@kernel.org/
Signed-off-by: SeongJae Park sj@kernel.org --- tools/testing/selftests/damon/Makefile | 1 + ...sysfs_update_schemes_tried_regions_hang.py | 33 +++++++++++++++++++ 2 files changed, 34 insertions(+) create mode 100755 tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_hang.py
diff --git a/tools/testing/selftests/damon/Makefile b/tools/testing/selftests/damon/Makefile index 1363987709c6..d90bdba28ff4 100644 --- a/tools/testing/selftests/damon/Makefile +++ b/tools/testing/selftests/damon/Makefile @@ -12,6 +12,7 @@ TEST_PROGS += debugfs_empty_targets.sh debugfs_huge_count_read_write.sh TEST_PROGS += debugfs_duplicate_context_creation.sh TEST_PROGS += debugfs_rm_non_contexts.sh TEST_PROGS += sysfs.sh sysfs_update_removed_scheme_dir.sh +TEST_PROGS += sysfs_update_schemes_tried_regions_hang.py TEST_PROGS += sysfs_update_schemes_tried_regions_wss_estimation.py TEST_PROGS += reclaim.sh lru_sort.sh TEST_PROGS += dbgfs_target_ids_read_before_terminate_race.sh diff --git a/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_hang.py b/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_hang.py new file mode 100755 index 000000000000..14187be3b004 --- /dev/null +++ b/tools/testing/selftests/damon/sysfs_update_schemes_tried_regions_hang.py @@ -0,0 +1,33 @@ +#!/usr/bin/env python3 +# SPDX-License-Identifier: GPL-2.0 + +import subprocess +import time + +import _damon + +def main(): + proc = subprocess.Popen(['sleep', '2']) + kdamonds = _damon.Kdamonds([_damon.Kdamond( + contexts=[_damon.DamonCtx( + ops='vaddr', + targets=[_damon.DamonTarget(pid=proc.pid)], + schemes=[_damon.Damos( + access_pattern=_damon.DamosAccessPattern( + nr_accesses=[200, 200]))] # schemes + )] # contexts + )]) # kdamonds + + err = kdamonds.start() + if err != None: + print('kdmaond start failed: %s' % err) + exit(1) + + while proc.poll() == None: + err = kdamonds.kdamonds[0].update_schemes_tried_bytes() + if err != None: + print('tried bytes update failed: %s' % err) + exit(1) + +if __name__ == '__main__': + main()
linux-kselftest-mirror@lists.linaro.org
-
SeongJae Park