[Linux] Run glib event loop on dedicated glib Matter context #25960

arkq · 2023-04-04T11:58:09Z

Problem

Currently Linux uses GLib global main context when running main even loop. This causes issues with glib-based applications which use default main context in their global main event loop. Such an example is a Linux application based on Qt (Qt on Linux runs GLib global main context in order to link seamlessly with glib-based libraries). The problem is that on Linux platform chip::DeviceLayer::PlatformManagerImpl::_InitChipStack() waits until main event loop is ready to process events on given context (currently, global context). E.g.:

import chip.native
from PySide6.QtWidgets import QApplication

app = QApplication()   # this grabs glib global main context
chip.native.Init(0)    # DEADLOCK: (we need event loop to be running, but our `g_main_loop_run()` waits until context will be released)
app.exec()             # grabbed context will be processed here

See this for reference: #23320 (comment)

This PR is a follow-up for #25916

Changes

do not use glib global main context in our glib main event loop
rewrite all places where global context was used, to use glib Matter main context (functions like g_idle_add() or g_io_add_watch_full() can not be used any more)
add asserts to check whether g_main_context_get_thread_default() was set in all places where g_signal_connect() or GIO call are made

Testing

Tested Python bindings:

>>> import chip.native
>>> from PySide6.QtWidgets import QApplication
>>> app = QApplication()
>>> chip.native.Init(0)
[1680609380.171509][1551207:1551207] CHIP:CTL: Setting attestation nonce to random value
[1680609380.172885][1551207:1551207] CHIP:CTL: Setting CSR nonce to random value
[1680609380.173417][1551207:1551207] CHIP:DL: ChipLinuxStorage::Init: Using KVS config file: /tmp/chip_kvs
[1680609380.173493][1551207:1551207] CHIP:DL: ChipLinuxStorage::Init: Using KVS config file: /tmp/chip_factory.ini
[1680609380.173560][1551207:1551207] CHIP:DL: ChipLinuxStorage::Init: Using KVS config file: /tmp/chip_config.ini
[1680609380.173605][1551207:1551207] CHIP:DL: ChipLinuxStorage::Init: Using KVS config file: /tmp/chip_counters.ini
[1680609380.173770][1551207:1551207] CHIP:DL: writing settings to file (/tmp/chip_counters.ini-yTuqpO)
[1680609380.173865][1551207:1551207] CHIP:DL: renamed tmp file to file (/tmp/chip_counters.ini)
[1680609380.173879][1551207:1551207] CHIP:DL: NVS set: chip-counters/reboot-count = 3 (0x3)
[1680609380.174271][1551207:1551207] CHIP:DL: Got Ethernet interface: eno1
[1680609380.174567][1551207:1551207] CHIP:DL: Found the primary Ethernet interface:eno1
[1680609380.175111][1551207:1551207] CHIP:DL: Got WiFi interface: wlxf4f26d1a6d27
[1680609380.175629][1551207:1551207] CHIP:DL: Found the primary WiFi interface:wlxf4f26d1a6d27
>>> app.exec()

Tested BLE commissioning with

./chip-lighting-app --wifi

./out/linux-x64-chip-tool/chip-tool pairing ble-wifi 1 <SSID> <password> 20202021 3840

Postscriptum

For anyone reading this description because of some break caused by this PR:

In order to quickly verify that the problem is with glib global main context, one can add this snippet at the beginning of PlatformManagerImpl::_InitChipStack():

    g_thread_unref(g_thread_new(
        "gmain-global",
        [](void *) -> void * {
            g_main_loop_run(g_main_loop_new(nullptr, FALSE));
            return nullptr;
        },
        nullptr));

If this will help, it means that somewhere there is a source attached to the glib global main context. The next step is to find this place and use GLibMatterContextAttachSource() if applicable or run particular block of code with GLibMatterContextInvokeSync(). For finding GIO calls attached to glib global main context, one can use gdb -x <script> --args <cmd> with script:

start
# using global context in g_main_context_ref_thread_default
# https://github.com/GNOME/glib/blob/180713772f4e7bcdddf2c793f2f34a498184ed15/glib/gmain.c#L999
break ../../../glib/gmain.c:999
commands
backtrace
end
# run app
continue

arkq · 2023-04-04T12:01:07Z

@yzm157 can you verify on your side whether this PR fixes the issue with Qt?

github-actions · 2023-04-04T12:10:05Z

PR #25960: Size comparison from d81d4e3 to 2f96585

Decreases (1 build for cc32xx)

platform	target	config	section	`d81d4e3`	`2f96585`	change	% change
cc32xx	lock	CC3235SF_LAUNCHXL	.debug_info	20302987	20302986	-1	-0.0

Full report (1 build for cc32xx)

platform	target	config	section	`d81d4e3`	`2f96585`	change	% change
cc32xx	lock	CC3235SF_LAUNCHXL		0	0	0	0.0
			(read only)	644409	644409	0	0.0
			(read/write)	203848	203848	0	0.0
			.ARM.attributes	44	44	0	0.0
			.ARM.exidx	8	8	0	0.0
			.bss	197248	197248	0	0.0
			.comment	194	194	0	0.0
			.data	1480	1480	0	0.0
			.debug_abbrev	933099	933099	0	0.0
			.debug_aranges	87608	87608	0	0.0
			.debug_frame	301328	301328	0	0.0
			.debug_info	20302987	20302986	-1	-0.0
			.debug_line	`2679748`	`2679748`	0	0.0
			.debug_loc	2824679	2824679	0	0.0
			.debug_ranges	286200	286200	0	0.0
			.debug_str	`3039406`	`3039406`	0	0.0
			.ramVecs	780	780	0	0.0
			.resetVecs	64	64	0	0.0
			.rodata	105953	105953	0	0.0
			.shstrtab	232	232	0	0.0
			.stab	204	204	0	0.0
			.stabstr	441	441	0	0.0
			.stack	2048	2048	0	0.0
			.strtab	377422	377422	0	0.0
			.symtab	256768	256768	0	0.0
			.text	536336	536336	0	0.0

github-actions · 2023-04-04T14:08:51Z

PR #25960: Size comparison from d81d4e3 to 58ba991

Full report (1 build for cc32xx)

platform	target	config	section	`d81d4e3`	`58ba991`
cc32xx	lock	CC3235SF_LAUNCHXL		0	0
			(read only)	644409	644409
			(read/write)	203848	203848
			.ARM.attributes	44	44
			.ARM.exidx	8	8
			.bss	197248	197248
			.comment	194	194
			.data	1480	1480
			.debug_abbrev	933099	933099
			.debug_aranges	87608	87608
			.debug_frame	301328	301328
			.debug_info	20302987	20302987
			.debug_line	`2679748`	`2679748`
			.debug_loc	2824679	2824679
			.debug_ranges	286200	286200
			.debug_str	`3039406`	`3039406`
			.ramVecs	780	780
			.resetVecs	64	64
			.rodata	105953	105953
			.shstrtab	232	232
			.stab	204	204
			.stabstr	441	441
			.stack	2048	2048
			.strtab	377422	377422
			.symtab	256768	256768
			.text	536336	536336

yzm157 · 2023-04-07T02:39:27Z

@yzm157 can you verify on your side whether this PR fixes the issue with Qt?

Looks like it works, works well.

All D-Bus proxy objects created with GLib *_new_for_bus and *_new_for_bus_sync functions have to be created with thread default main context set. Otherwise, proxy object internal signals will be scheduled on GLib global main context. We do not want that, because we do not know whether Matter SDK will be used in GLib-based application or not, so we can not run global main even loop on our own - all Matter-related GLib signals shell be run on a dedicated Matter GLib context.

github-actions · 2023-04-14T08:12:05Z

PR #25960: Size comparison from 62974a0 to ec685cd

Increases (1 build for cc32xx)

platform	target	config	section	`62974a0`	`ec685cd`	change	% change
cc32xx	lock	CC3235SF_LAUNCHXL	.debug_info	20326060	20326061	1	0.0

Full report (1 build for cc32xx)

platform	target	config	section	`62974a0`	`ec685cd`	change
cc32xx	lock	CC3235SF_LAUNCHXL		0	0	0
			(read only)	643081	643081	0
			(read/write)	203848	203848	0
			.ARM.attributes	44	44	0
			.ARM.exidx	8	8	0
			.bss	197248	197248	0
			.comment	194	194	0
			.data	1480	1480	0
			.debug_abbrev	933224	933224	0
			.debug_aranges	87760	87760	0
			.debug_frame	302028	302028	0
			.debug_info	20326060	20326061	1
			.debug_line	`2687403`	`2687403`	0
			.debug_loc	2838361	2838361	0
			.debug_ranges	288040	288040	0
			.debug_str	3042379	3042379	0
			.ramVecs	780	780	0
			.resetVecs	64	64	0
			.rodata	104393	104393	0
			.shstrtab	232	232	0
			.stab	204	204	0
			.stabstr	441	441	0
			.stack	2048	2048	0
			.strtab	377626	377626	0
			.symtab	256832	256832	0
			.text	536568	536568	0

src/platform/Linux/bluez/ChipDeviceScanner.cpp

github-actions · 2023-04-24T09:25:50Z

PR #25960: Size comparison from de0dfcf to a84d073

Decreases (1 build for cc32xx)

platform	target	config	section	`de0dfcf`	`a84d073`	change	% change
cc32xx	lock	CC3235SF_LAUNCHXL	.debug_info	20330829	20330828	-1	-0.0

Full report (1 build for cc32xx)

platform	target	config	section	`de0dfcf`	`a84d073`	change	% change
cc32xx	lock	CC3235SF_LAUNCHXL		0	0	0	0.0
			(read only)	643249	643249	0	0.0
			(read/write)	203848	203848	0	0.0
			.ARM.attributes	44	44	0	0.0
			.ARM.exidx	8	8	0	0.0
			.bss	197248	197248	0	0.0
			.comment	194	194	0	0.0
			.data	1480	1480	0	0.0
			.debug_abbrev	933224	933224	0	0.0
			.debug_aranges	87792	87792	0	0.0
			.debug_frame	302140	302140	0	0.0
			.debug_info	20330829	20330828	-1	-0.0
			.debug_line	2687904	2687904	0	0.0
			.debug_loc	`2838960`	`2838960`	0	0.0
			.debug_ranges	288072	288072	0	0.0
			.debug_str	3042335	3042335	0	0.0
			.ramVecs	780	780	0	0.0
			.resetVecs	64	64	0	0.0
			.rodata	104401	104401	0	0.0
			.shstrtab	232	232	0	0.0
			.stab	204	204	0	0.0
			.stabstr	441	441	0	0.0
			.stack	2048	2048	0	0.0
			.strtab	377963	377963	0	0.0
			.symtab	256976	256976	0	0.0
			.text	536728	536728	0	0.0

…is started (#26338) * chip-repl hits the Code is unsafe/racy assert when BLE commissioning is started * Add lock/unlock to CancelTimer * Fix a few other issue * Restyle * Small fix * Last few fixes

github-actions bot added linux platform labels Apr 4, 2023

arkq changed the title ~~Glib matter context~~ [Linux] Run glib event loop on dedicated glib Matter context Apr 4, 2023

arkq requested a review from msandstedt April 4, 2023 12:02

arkq mentioned this pull request Apr 11, 2023

[Tizen] Run single thread for processing all GLib sources #25762

Merged

arkq marked this pull request as ready for review April 12, 2023 11:15

pullapprove bot requested review from vivien-apple, woody-apple, xylophone21, younghak-hwang, yufengwangca and yunhanw-google April 12, 2023 11:16

pullapprove bot added the review - pending label Apr 12, 2023

arkq added 11 commits April 14, 2023 10:00

Run Glib thread on a dedicated context

de3bafe

Run WiFiIPChangeListener on dedicated context

77e5733

Do not use g_io_add_watch when adding IO watch

e78dd2f

Initialize BlueZ D-Bus object proxy on Matter GLib event loop

a661fde

Initialize WPA D-Bus object proxy on Matter GLib event loop

2ab13cb

Fix typo: devcie -> device

06cc96f

Initialize OTBR D-Bus object proxy on Matter GLib event loop

e19cad1

Fix glib source watch removal in BLE module

029c124

Enable glib event loop in Thread-only configuration

cc44d01

Run OTBR async calls on Matter glib context

ec685cd

arkq force-pushed the glib-matter-context branch from 2f8e778 to ec685cd Compare April 14, 2023 08:01

andy31415 reviewed Apr 14, 2023

View reviewed changes

src/platform/Linux/bluez/ChipDeviceScanner.cpp Outdated Show resolved Hide resolved

arkq added 2 commits April 24, 2023 11:13

Document GDBusCreateObjectManagerContext usage

2b3664c

Merge branch 'master' into glib-matter-context

a84d073

andy31415 approved these changes May 9, 2023

View reviewed changes

andy31415 added the fast track label May 9, 2023

andy31415 merged commit 0ada46a into project-chip:master May 9, 2023

arkq deleted the glib-matter-context branch May 9, 2023 19:03

arkq mentioned this pull request May 12, 2023

Initialize BLE iterator on Matter glib context #26540

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Linux] Run glib event loop on dedicated glib Matter context #25960

[Linux] Run glib event loop on dedicated glib Matter context #25960

arkq commented Apr 4, 2023 •

edited

Loading

arkq commented Apr 4, 2023

github-actions bot commented Apr 4, 2023

github-actions bot commented Apr 4, 2023

yzm157 commented Apr 7, 2023

github-actions bot commented Apr 14, 2023

github-actions bot commented Apr 24, 2023

[Linux] Run glib event loop on dedicated glib Matter context #25960

[Linux] Run glib event loop on dedicated glib Matter context #25960

Conversation

arkq commented Apr 4, 2023 • edited Loading

Problem

Changes

Testing

Postscriptum

arkq commented Apr 4, 2023

github-actions bot commented Apr 4, 2023

github-actions bot commented Apr 4, 2023

yzm157 commented Apr 7, 2023

github-actions bot commented Apr 14, 2023

github-actions bot commented Apr 24, 2023

arkq commented Apr 4, 2023 •

edited

Loading