|
| From: | Maksim Davydov |
| Subject: | Re: [PATCH v2 4/4] scripts: add script to compare compatible properties |
| Date: | Mon, 19 Sep 2022 13:04:22 +0300 |
On 9/14/22 13:55, Maksim Davydov wrote:
This script run QEMU to obtain compat_props of machines and default
values of different types and produce appropriate table. This table
can be used to compare machine types to choose the most suitable
machine. Also table in json or csv format should be used to check that
new machine doesn't affect previous ones via comparisin tables with and
without new machine.
Default values of properties are needed to fill "holes" in the table (one
machine has these properties and another not).
Notes:
* some init values from the devices can't be available like properties
from virtio-9p when configure has --disable-virtfs. This situations will
be seen in the table as "unavailable driver".
* Default values can be get can be obtained in an unobvious way, like
x86 features. If the script doesn't know how to get property default value
to compare one machine with another it fills "holes" with "unavailable
method". This is done because script uses whitelist model to get default
values of different types. It means that the method that can't be applied
to a new type that can crash this script. It is better to get an
"unavailable driver" when creating a new machine with new compatible
properties than to break this script. So it turns out a more stable and
generic script.
* If the default value can't be obtained because this property doesn't
exist or because this property can't have default value, appropriate
"hole" will be filled by "unknown property" or "no default value"
* If the property is applied to the abstract class, the script collects
default values from all child classes (set of default values)
Example:
./scripts/compare_mt.py --mt pc-q35-3.1 pc-q35-2.12
╒════════════════════════════════════╤═══════════════╤═══════════════╕
│ │ pc-q35-2.12 │ pc-q35-3.1 │
╞════════════════════════════════════╪═══════════════╪═══════════════╡
│ EPYC-IBPB-x86_64-cpu-xlevel │ 0x8000000a │ 2147483678 │
├────────────────────────────────────┼───────────────┼───────────────┤
│ EPYC-x86_64-cpu-xlevel │ 0x8000000a │ 2147483678 │
├────────────────────────────────────┼───────────────┼───────────────┤
│ Skylake-Server-IBRS-x86_64-cpu-pku │ False │ True │
├────────────────────────────────────┼───────────────┼───────────────┤
│ Skylake-Server-x86_64-cpu-pku │ False │ True │
├────────────────────────────────────┼───────────────┼───────────────┤
│ VGA-global-vmstate │ True │ False │
├────────────────────────────────────┼───────────────┼───────────────┤
│ cirrus-vga-global-vmstate │ True │ False │
├────────────────────────────────────┼───────────────┼───────────────┤
│ hda-audio-use-timer │ False │ True │
├────────────────────────────────────┼───────────────┼───────────────┤
│ migration-decompress-error-check │ False │ True │
├────────────────────────────────────┼───────────────┼───────────────┤
│ qxl-vga-global-vmstate │ True │ False │
├────────────────────────────────────┼───────────────┼───────────────┤
│ vmware-svga-global-vmstate │ True │ False │
├────────────────────────────────────┼───────────────┼───────────────┤
│ x86_64-cpu-legacy-cache │ True │ [True, False] │
├────────────────────────────────────┼───────────────┼───────────────┤
│ x86_64-cpu-topoext │ False │ [True, False] │
├────────────────────────────────────┼───────────────┼───────────────┤
│ x86_64-cpu-x-hv-synic-kvm-only │ True │ False │
╘════════════════════════════════════╧═══════════════╧═══════════════╛
Looks great!
Some nitpicking:
1. I'd use left-align, at least for the first column as more readable.
2. Seems '-' is bad separator for driver-property, as both drivers and properties sometimes has own '-' inside. Let's use something else. Maybe colon, like driver:property. Or, maybe, just add one more column, to keep driver and property names in different columns.
Hmm, and what means [True, False] ?
Signed-off-by: Maksim Davydov <davydov-max@yandex-team.ru>
---
scripts/compare_mt.py | 374 ++++++++++++++++++++++++++++++++++++++++++
1 file changed, 374 insertions(+)
create mode 100755 scripts/compare_mt.py
diff --git a/scripts/compare_mt.py b/scripts/compare_mt.py
new file mode 100755
index 0000000000..1df5e6c4f4
--- /dev/null
+++ b/scripts/compare_mt.py
@@ -0,0 +1,374 @@
+#!/usr/bin/env python3
+#
+# Script to compare machine type compatible properties (include/hw/boards.h).
+# compat_props are applied to the driver during initialization to change
+# default values, for instance, to maintain compatibility.
+# This script constructs table with machines and values of their compat_props
+# to compare and to find places for improvements or places with bugs. If
+# during the comparision, some machine type doesn't have a property (it is in
+# the comparision table because another machine type has it), then the
+# appropriate method will be used to obtain the default value of this driver
+# property via qmp command (e.g. query-cpu-model-expansion for x86_64-cpu).
+# These methods are defined below in propery_methods.
+#
+# Copyright (c) Yandex Technologies LLC, 2022
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, see <http://www.gnu.org/licenses/>.
+
+from tabulate import tabulate
+import sys
+from os import path
+from argparse import ArgumentParser, RawTextHelpFormatter
+import pandas as pd
+
+try:
+ qemu_dir = path.abspath(path.dirname(path.dirname(__file__)))
+ sys.path.append(path.join(qemu_dir, 'python'))
+ from qemu.machine import QEMUMachine
+except ModuleNotFoundError as exc:
+ print(f"Module '{exc.name}' not found.")
+ print("Try export PYTHONPATH=top-qemu-dir/python or run from top-qemu-dir")
+ sys.exit(1)
+
+
+default_cmd_line = 'build/qemu-system-x86_64 -enable-kvm -machine none'
+
+
+# Methods to get right values of drivers props
+#
+# Use these methods as a 'whitelist' and add entries only if necessary. It's
+# important to be stable and predictable in analysis and tests.
+# * Names should be in qom-list-types format (486-x86_64-cpu, not 486)
+# * The latest definition wins:
+# 1) device
+# 2) cpu
+# 3) 486-x86_64-cpu
+# For 486 will be used - rule #3, for virtio device - #1, for Haswell - #2
+def get_device_prop(vm, device, prop_name):
+ args = {'typename': device}
+ device_props = vm.command('device-list-properties', **args)
+ for prop in device_props:
+ if prop['name'] == prop_name:
+ return str(prop.get('default-value', 'No default value'))
+
+ return 'Unknown property'
+
+
+def get_x86_64_cpu_prop(vm, device, prop_name):
+ # crop last 11 chars '-x86_64-cpu'
+ args = {'type': 'full', 'model': {'name': device[:-11]}}
+ props = vm.command('query-cpu-model-expansion', **args)['model']['props']
+ return str(props.get(prop_name, 'Unknown property'))
+
+
+# Now it's stub, because all memory_backend types don't have default values
+# but this behaviour can be changed
+def get_memory_backend_prop(vm, driver, prop_name):
+ args = {'typename': driver}
+ memory_backend_props = vm.command('qom-list-properties', **args)
+ for prop in memory_backend_props:
+ if prop['name'] == prop_name:
+ return str(prop.get('default-value', 'No default value'))
+
+ return 'Unknown property'
+
+
+property_methods = [
+ {'name': 'device', 'method': get_device_prop},
+ {'name': 'x86_64-cpu', 'method': get_x86_64_cpu_prop},
+ {'name': 'memory-backend', 'method': get_memory_backend_prop}
+ ]
+# End of methods definition
+
+
+script_desc="""Script to compare machine types (their compat_props).
PEP8: whitespaces around = operator
+
+If a property applies to an abstract class this script collects default \
+values of all child classes and prints them as a set.
+
+"Unavailable method" - means that this script doesn\'t know how to get \
you don't need to escape quote (') inside """string""" in triple-double quotes.
+default values of the driver. To add method use the construction described \
+at the top of the script.
+"Unavailable driver" - means that this script doesn\'t know this driver. \
+For instance, this can happen if you configure QEMU without this device or \
+if machine type definition has error.
+"No default value" - means that the appropriate method can\'t get the default \
+value and most likely that this property doesn\'t have it.
+"Unknown property" - means that the appropriate method can\'t find property \
+with this name."""
+
+
+def parse_args():
+ parser = ArgumentParser(formatter_class=RawTextHelpFormatter,
+ description=script_desc)
+ parser.add_argument('--json', action='',
+ help='returns table in json format')
+ parser.add_argument('--csv', action='',
+ help='returns table in csv format')
More usual would be to add one argument instead, like --format, that will take "json" or "csv" argument (or "ascii" by default). See here: https://docs.python.org/3/library/argparse.html#choices
+ parser.add_argument('--all', action='',
+ help='prints all available machine types (list of '
+ 'machine types will be ignored). Execution with '
+ '--all may take several minutes!')
+ parser.add_argument('--mt', nargs="*", type=str,
+ help='list of Machine Types that will be compared')
To restrict using both --all and --mt, you may use mutual exclusion: https://docs.python.org/3/library/argparse.html#mutual-exclusion
+ parser.add_argument('--raw', action='',
+ help='prints ALL defined properties without value '
+ 'transformation. By default, only properties '
+ 'with different values will be printed and with '
+ 'value transformation(e.g. "on" -> True)')
+ parser.add_argument('--cmd-line', default=default_cmd_line,
+ help='command line to start qemu. '
+ 'Default: "{}"'.format(default_cmd_line))
+ return parser.parse_args()
+
+
+# return tuple (name, major, minor, revision)
+def mt_comp(mt):
+ splited_name = mt['name'].rsplit('-', 1)
+ if len(splited_name) == 2:
+ version = splited_name[1].split('.')
+ if len(version) == 2:
+ return (splited_name[0], int(version[0]), int(version[1]), 0)
+ if len(version) == 3:
+ return (splited_name[0],
+ int(version[0]), int(version[1]), int(version[2]))
what if len(version) == 1 or len(version) > 3 ? Should we add an assertion?
also, you don't need parentheses to return tuple
+
+ return (splited_name[0], 0, 0, 0)
what about something like this:
name = mt['name']
if '-' not in name:
return name, 0, 0, 0
name, ver = name.rsplit('-', 1)
ver_list = map(int, ver.split('.', 2))
ver_list += [0] * (3 - len(ver_list))
return name, *ver_list
+
+
+def get_mt_definitions(vm):
+ args = {'compat-props': True}
+ raw_mt_defs = vm.command('query-machines', **args)
+ mt_defs = [] # MT: {driver_1: set_of_props, ...}
PEP8: at least two spaces before '#'.
+ for raw_mt in raw_mt_defs:
+ compat_props = {}
+ for prop in raw_mt['compat-props']:
+ if prop['driver'] not in compat_props:
+ compat_props[prop['driver']] = {}
+ compat_props[prop['driver']][prop['property']] = prop['value']
+ mt_defs.append({'name': raw_mt['name'], 'compat-props': compat_props})
+
+ mt_defs.sort(key=mt_comp)
+ return mt_defs
+
+
+def get_req_props(mt_defs):
+ driver_props = {}
+ for mt in mt_defs:
+ compat_props = mt['compat-props']
+ for driver, prop in compat_props.items():
+ if driver not in driver_props:
+ driver_props[driver] = set()
+ driver_props[driver].update(prop.keys())
+
+ return driver_props
+
+
+def get_driver_definitions(vm):
+ args = {'abstract': True}
+ qom_all_types = vm.command('qom-list-types', **args)
+
+ driver_to_def = {}
+ for obj_type in qom_all_types:
+ # parent of Object is None
+ parent = obj_type.get('parent', None)
+ abstr = obj_type.get('abstract', False)
+ driver_to_def[obj_type['name']] = {
+ 'parent': parent,
+ 'abstract': abstr,
+ 'method': None}
+ if abstr:
+ type_args = {'implements': obj_type['name'], 'abstract': True}
+ list_child_objs = vm.command('qom-list-types', **type_args)
I don't insist, but we already have two popular styles of calling qmp commands:
list_child_objs = vm.command('qom-list-types', **{
'implemnts': obj_type['name'],
'abstract': True
}
and
list_child_objs = vm.command('qom-list-types', implements=obj_type['name'],
abstract=True)
don't you like one of them?
+ child_list = [child['name'] for child in list_child_objs]
+ driver_to_def[obj_type['name']]['child_list'] = child_list
+
+ for driver in property_methods:
+ if not driver_to_def[driver['name']]['abstract']:
+ driver_to_def[driver['name']]['method'] = driver['method']
+ continue
+
+ for child in driver_to_def[driver['name']]['child_list']:
+ driver_to_def[child]['method'] = driver['method']
+
+ return driver_to_def
+
+
+def fill_prop_table(vm, mt_list, driver_props, driver_defs):
+ table = {}
+ for driver, props in sorted(driver_props.items()):
+ for prop in sorted(props):
+ name = '{}-{}'.format(driver, prop)
+ table[name] = []
+ for mt in mt_list:
+ compat_props = mt['compat-props']
+ if driver in compat_props:
+ if prop in compat_props[driver]:
+ table[name].append(compat_props[driver][prop])
+ continue
+
+ # properties from QEMU (not from machine type compat_props)
+ # properties from another architecture or config
+ if driver not in driver_defs:
+ table[name].append('Unavailable driver')
+ continue
+
+ if not driver_defs[driver]['abstract']:
+ if driver_defs[driver]['method'] is None:
+ table[name].append('Unavailable method')
+ else:
+ table[name].append(
+ driver_defs[driver]['method'](vm, driver, prop))
+ else:
+ children = driver_defs[driver]['child_list']
+ values = set()
+ for child in children:
+ if driver_defs[child]['abstract']:
+ continue
+
+ if driver_defs[child]['method'] is None:
+ values.add('Unavailable method')
+ else:
+ values.add(
+ driver_defs[child]['method'](vm, child, prop))
+
+ table[name].append(list(values))
+
+ headers = [mt['name'] for mt in mt_list]
+ return pd.DataFrame.from_dict(table, orient='index', columns=headers)
+
+
+def transform_value(value):
+ true_list = ['true', 'on']
+ false_list = ['false', 'off']
+
+ out = value.lower()
+
+ if out in true_list:
+ return True
+
+ if out in false_list:
+ return False
+
+ return out
+
+
+# Only hex, dec and oct formats
+def transform_number(value):
+ try:
+ # C doesn't work with underscore ('2_5' != 25)
+ if '_' in value:
+ raise ValueError
+
+ if 'x' in value or 'X' in value:
+ return int(value, 16)
+
+ if 'o' in value or 'O' in value:
+ return int(value, 8)
+
+ return int(value)
For this, you can simply use int(value, 0). It converts automatically 0x.. as hex and 0o.. as oct.
+
+ except ValueError:
+ return None
+
+
+def transformed_table(table):
+ new_table = {}
+ for index, row in table.iterrows():
+ new_row = []
+ all_values = set()
+ # We want to save original hex/decimal format if not all values
+ # are the same in the row. So, transformed and not transformed will be
+ # stored
+ numeric_values = set()
+ for mt_prop_val in row:
+ if type(mt_prop_val) is list:
+ transformed = [transform_value(val) for val in mt_prop_val]
the same:
transformed = map(transform_value, mt_prop_val)
+ if len(transformed) == 1:
+ new_row.append(transformed[0])
+ else:
+ new_row.append(transformed)
+
+ numeric_values.update(set([transform_number(val)
+ for val in mt_prop_val]))
+ all_values.update(set(transformed))
+ else:
+ transformed = transform_value(mt_prop_val)
+ new_row.append(transformed)
+ numeric_values.add(transform_number(mt_prop_val))
+ all_values.add(transformed)
+
+ if len(table.columns) > 1:
+ if len(all_values) == 1:
+ continue
+
+ if None not in numeric_values and len(numeric_values) == 1:
+ continue
+
+ new_table[index] = new_row
+
+ return pd.DataFrame.from_dict(new_table, orient='index',
+ columns=table.columns.values)
+
+
+if __name__ == '__main__':
+ args = parse_args()
+ qemu_arg_list = args.cmd_line.split(' ')
Not sure about good practices, but that's how I do for scripts:
add a separate main function, that will have arguments according to your ArgumentParser, like
def main(all, mt, raw, json, csv ...):
...
and in "if __name__ == '__main__'" block you'll just do
main(**vars(parse_args()))
But anyway, having large code block in "if __name__ == '__main__'" is not a good practices. Better add separate function and call it.
+ with QEMUMachine(binary=qemu_arg_list[0],
+ qmp_timer=15, args=qemu_arg_list[1:]) as vm:
+ vm.launch()
+ mt_defs = get_mt_definitions(vm)
+ list_mt = [mt['name'] for mt in mt_defs]
+
+ if not args.all:
+ if args.mt is None:
+ print('Enter machine types for comparision or use --help')
+ print('List of available machine types:')
+ print(*list_mt, sep='\n')
+ sys.exit(1)
+
+ for mt in args.mt:
+ if mt not in list_mt:
+ print('Wrong machine type name')
+ print('List of available machine types:')
+ print(*list_mt, sep='\n')
+ sys.exit(1)
+
+ req_mt = []
+ if args.all:
+ req_mt = mt_defs
+ else:
+ for mt in mt_defs:
+ if mt['name'] in args.mt:
+ req_mt.append(mt)
+
+ if len(req_mt) == 1:
+ args.full = True
+
+ req_driver_props = get_req_props(req_mt)
+ driver_defs = get_driver_definitions(vm)
+ comp_table = fill_prop_table(vm, req_mt, req_driver_props, driver_defs)
+ if not args.raw:
+ comp_table = transformed_table(comp_table)
+
+ if args.json:
+ print(comp_table.to_json())
+ elif args.csv:
+ print(comp_table.to_csv())
+ else:
+ print(tabulate(comp_table, showindex=True, stralign='center',
+ tablefmt='fancy_grid', headers='keys',
+ disable_numparse=True))
+
+ vm.shutdown()--
I'm afraid to be too strict :( But I have some advises to make the script more readable and more maintainable :)
1. Use classes instead of nested-nested dicts. For example you may have Driver object, that would have name, parent, abstract ... properties.
2. Avoid high level of nesting in functions (like in fill_prop_table, you have double if inside triple for). Try add some helper functions to reduce complexity and nesting of the original function.
3. Add a short doc-string for each function, what it does.
4. Use python typing. In Qemu we tend to use it for all new python code.
5. Use f-strings instead of .format() where appropriate (yes, that's a nitpicking, but still)
Best regards,
Vladimir
| [Prev in Thread] | Current Thread | [Next in Thread] |