[SoundWire] Kernel Oops on machine/codec driver remove #3531

pablos-here · 2022-03-21T20:33:41Z

Describe the bug
I was attempting to configure the sound card on my Dell XPS 9700 as the mic was not working (speakers work). This resulted in a kernel call trace which I submitted a bug against[1] against my distro.

I distilled the developer's test case so blame me for non-intuitive cruft. There are two conditions which cause the oops - see below.

Just in case, I have captured separately the call trace for each condition.

I'm also uploading alsa-info.txt.

Refernces
1 - https://bugzilla.suse.com/show_bug.cgi?id=1197315

Condition 1

/sbin/modprobe -r snd_soc_sof_sdw
/sbin/modprobe -r snd_soc_rt711

Condition 2

# 
# Run me twice
# 
/sbin/modprobe -r snd_soc_sof_sdw
/sbin/modprobe -r snd_soc_rt1308_sdw

Call trace per condition
condition-1.txt
condition-2.txt

The text was updated successfully, but these errors were encountered:

plbossart · 2022-03-21T23:34:24Z

@looney-loon Can you try and remove the parent PCI device first to see if this helps? this is what we do in our remove/insert tests. I am not sure if we've ever tested removing the machine driver this way.

/sbin/modprobe -r snd_sof_pci_intel_cnl
/sbin/modprobe -r snd_soc_sof_sdw
/sbin/modprobe -r snd_soc_rt711
/sbin/modprobe -r snd_soc_rt1308_sdw
```

plbossart · 2022-03-21T23:35:23Z

For reference, these are the scripts we use:

https://github.com/thesofproject/sof-test/blob/main/tools/kmod/sof_remove.sh
https://github.com/thesofproject/sof-test/blob/main/tools/kmod/sof_insert.sh

Some information is not cleared on driver remove, which can lead to kernel oopses BugLink: thesofproject#3531 Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

Some information is not cleared on driver remove, which can lead to kernel oopses when removing codec drivers, e.g. with the following sequence. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shuts down the bus. This patch fixes the kernel oopses - however the users will still see a set of 'probed timed out' errors since a driver can no longer be found for a physical device. FIXME: do we need any sort of bus locking? BugLink: thesofproject#3531 Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

plbossart · 2022-03-22T00:49:41Z

I can reproduce the problem, see tentative fix at 8274036 PR #3534

pablos-here · 2022-03-22T02:09:42Z

Can you try and remove the parent PCI device first to see if this helps? this is what we do in our remove/insert tests. I am not sure if we've ever tested removing the machine driver this way.

/sbin/modprobe -r snd_sof_pci_intel_cnl
/sbin/modprobe -r snd_soc_sof_sdw
/sbin/modprobe -r snd_soc_rt711
/sbin/modprobe -r snd_soc_rt1308_sdw

Hi Pierre-Louis,

I just tested the commands above and it works. I'll update the SUSE bug.

Glad the issue netted something productive on your end.

Let me know if you'd like me to close the issue or whether your workflow handles it.

Merci beaucoup!

Some information is not cleared on driver remove, which can lead to kernel oopses when removing codec drivers, e.g. with the following sequence. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shuts down the bus. This patch fixes the kernel oopses - however the users will still see a set of 'probed timed out' errors since a driver can no longer be found for a physical device. The issue with the ops pointer has been there forever, but the Fixes tag below should be more than enough for actual SoundWire uses. BugLink: thesofproject#3531 Fixes: 2140b66 ("soundwire: bus: fix race condition with probe_complete signaling") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shuts down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. This patch removes the use the 'slave' ops and uses proper locking to make sure there are no live callbacks based on a dangling pointer executed during or after the driver unbinding sequence. In some cases, indicated with comments in the code, the device lock is already taken at a higher level when starting the resume operations. The issue with the ops pointer has been there since December 2017, but there were so many changes in the bus handling code that this patch will not apply cleanly all the way to this initial commit. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. This patch removes the use the 'slave' ops and uses proper locking to make sure there are no live callbacks based on a dangling pointer executed during or after the driver unbinding sequence. In one specific case, indicated with comments in the code, the device lock is already taken at a higher level when starting the resume operations. The issue with the ops pointer has been there since December 2017, but there were so many changes in the bus handling code that this patch will not apply cleanly all the way to this initial commit. The changes can be easily backported though. Thanks to Dan Williams for his suggestions on an earlier version of this patch. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. This patch removes the use the 'slave' ops and uses proper locking to make sure there are no live callbacks based on a dangling pointer executed during or after the driver unbinding sequence. The routine sdw_update_slave_status() can be called from two different places, and the device lock may be taken already by the device core. In this case, taking the lock unconditionally results in a seld-inflicted deadlock. Using device_trylock() helps prevent this issue - and in case the device is locked already by the device core, it's safe to proceed since the call to sdw_update_slave_status() happens during the parent device PM routines. The parent-child relationship guarantees that the lock will not be released while the dev->driver pointer is used. The issue with the ops pointer has been there since December 2017, but there were so many changes in the bus handling code that this patch will not apply cleanly all the way to this initial commit. The changes can be easily backported though. Thanks to Dan Williams for his suggestions on an earlier version of this patch. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. This patch removes the use the 'slave' ops and uses proper locking to make sure there are no live callbacks based on a dangling pointer executed during or after the driver unbinding sequence. Adding device_lock()/unlock() seems fitting since it's what is already used by the device and power management core routines, however this needs to be done carefully to avoid self-inflicted dead-locks. There are two cases where the device core code will take a lock, and taking the same lock while doing through suspend/resume routines would be problematic. The first case is on attachment. The attachment and device resume can happen asynchronously, and in the case where the resume happens first adding device_lock() does not work at all. Fortunately, in this case the codec driver resume routine needs to wait on the 'initialization_complete', so using device_trylock() is fine. In all other cases unrelated to attachment, a regular device_lock() needs to be taken. The second case is the routine sdw_update_slave_status(), which can be called from two different places, and the device lock may be taken already by the device core. Using device_trylock() also helps prevent this issue - and in case the device is locked already by the device core, it's safe to proceed since the call to sdw_update_slave_status() happens during the parent device PM routines. The parent-child relationship guarantees that the lock will not be released while the dev->driver pointer is used. The issue with the ops pointer has been there since December 2017, but there were so many changes in the bus handling code that this patch will not apply cleanly all the way to this initial commit. The changes can be easily backported though. Thanks to Dan Williams for his suggestions on an earlier version of this patch. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. This patch removes the use the 'slave' ops, and checks with a state variable if a driver is bound before invoking the callbacks to make sure there are no live callbacks based on a dangling pointer executed during or after the driver unbinding sequence. That state variable is set after a driver is bound and cleared before a driver is unbound. An alternate solution based on device_lock() and device_unlock() was tested earlier. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, the bus notifier seemed like a less invasive solution to prevent kernels oopses in corner cases. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject/linux#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject/linux#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject/linux#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

[ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject/linux#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>

BugLink: https://bugs.launchpad.net/bugs/1990162 [ Upstream commit bd29c00 ] In the SoundWire probe, we store a pointer from the driver ops into the 'slave' structure. This can lead to kernel oopses when unbinding codec drivers, e.g. with the following sequence to remove machine driver and codec driver. /sbin/modprobe -r snd_soc_sof_sdw /sbin/modprobe -r snd_soc_rt711 The full details can be found in the BugLink below, for reference the two following examples show different cases of driver ops/callbacks being invoked after the driver .remove(). kernel: BUG: kernel NULL pointer dereference, address: 0000000000000150 kernel: Workqueue: events cdns_update_slave_status_work [soundwire_cadence] kernel: RIP: 0010:mutex_lock+0x19/0x30 kernel: Call Trace: kernel: ? sdw_handle_slave_status+0x426/0xe00 [soundwire_bus 94ff184bf398570c3f8ff7efe9e32529f532e4ae] kernel: ? newidle_balance+0x26a/0x400 kernel: ? cdns_update_slave_status_work+0x1e9/0x200 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: BUG: unable to handle page fault for address: ffffffffc07654c8 kernel: Workqueue: pm pm_runtime_work kernel: RIP: 0010:sdw_bus_prep_clk_stop+0x6f/0x160 [soundwire_bus] kernel: Call Trace: kernel: <TASK> kernel: sdw_cdns_clock_stop+0xb5/0x1b0 [soundwire_cadence 1bcf98eebe5ba9833cd433323769ac923c9c6f82] kernel: intel_suspend_runtime+0x5f/0x120 [soundwire_intel aca858f7c87048d3152a4a41bb68abb9b663a1dd] kernel: ? dpm_sysfs_remove+0x60/0x60 This was not detected earlier in Intel tests since the tests first remove the parent PCI device and shut down the bus. The sequence above is a corner case which keeps the bus operational but without a driver bound. While trying to solve this kernel oopses, it became clear that the existing SoundWire bus does not deal well with the unbind case. Commit 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") added a 'probed' status variable and a 'probe_complete' struct completion. This status is however not reset on remove and likewise the 'probe complete' is not re-initialized, so the bind/unbind/bind test cases would fail. The timeout used before the 'update_status' callback was also a bad idea in hindsight, there should really be no timing assumption as to if and when a driver is bound to a device. An initial draft was based on device_lock() and device_unlock() was tested. This proved too complicated, with deadlocks created during the suspend-resume sequences, which also use the same device_lock/unlock() as the bind/unbind sequences. On a CometLake device, a bad DSDT/BIOS caused spurious resumes and the use of device_lock() caused hangs during suspend. After multiple weeks or testing and painful reverse-engineering of deadlocks on different devices, we looked for alternatives that did not interfere with the device core. A bus notifier was used successfully to keep track of DRIVER_BOUND and DRIVER_UNBIND events. This solved the bind-unbind-bind case in tests, but it can still be defeated with a theoretical corner case where the memory is freed by a .remove while the callback is in use. The notifier only helps make sure the driver callbacks are valid, but not that the memory allocated in probe remains valid while the callbacks are invoked. This patch suggests the introduction of a new 'sdw_dev_lock' mutex protecting probe/remove and all driver callbacks. Since this mutex is 'local' to SoundWire only, it does not interfere with existing locks and does not create deadlocks. In addition, this patch removes the 'probe_complete' completion, instead we directly invoke the 'update_status' from the probe routine. That removes any sort of timing dependency and a much better support for the device/driver model, the driver could be bound before the bus started, or eons after the bus started and the hardware would be properly initialized in all cases. BugLink: thesofproject/linux#3531 Fixes: 56d4fe3 ("soundwire: Add MIPI DisCo property helpers") Fixes: 528be50 ("soundwire: sdw_slave: add probe_complete structure and new fields") Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> Reviewed-by: Rander Wang <rander.wang@intel.com> Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com> Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://lore.kernel.org/r/20220621225641.221170-2-pierre-louis.bossart@linux.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org> Signed-off-by: Kamal Mostafa <kamal@canonical.com> Signed-off-by: Stefan Bader <stefan.bader@canonical.com>

pablos-here mentioned this issue Mar 21, 2022

Mic not working, although speakers are ... speaking #3532

Closed

plbossart mentioned this issue Mar 22, 2022

ASoC/SoundWire: fix two issues for rmmod/modprobe machine driver tests #3534

Closed

plbossart changed the title ~~Oops > trying to configure audio~~ [SoundWire] Kernel Oops on machine/codec driver remove Mar 22, 2022

plbossart added SDW Applies to SoundWire bus for codec connection bug Something isn't working P1 Blocker bugs or important features labels Mar 22, 2022

plbossart mentioned this issue Apr 5, 2022

SoundWire: remove the use of slave->ops #3568

Closed

plbossart mentioned this issue May 18, 2022

SoundWire: rt711-sdca: transfers timed out on component remove/probe #3650

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SoundWire] Kernel Oops on machine/codec driver remove #3531

[SoundWire] Kernel Oops on machine/codec driver remove #3531

pablos-here commented Mar 21, 2022

plbossart commented Mar 21, 2022

plbossart commented Mar 21, 2022

plbossart commented Mar 22, 2022 •

edited

pablos-here commented Mar 22, 2022

[SoundWire] Kernel Oops on machine/codec driver remove #3531

[SoundWire] Kernel Oops on machine/codec driver remove #3531

Comments

pablos-here commented Mar 21, 2022

plbossart commented Mar 21, 2022

plbossart commented Mar 21, 2022

plbossart commented Mar 22, 2022 • edited

pablos-here commented Mar 22, 2022

plbossart commented Mar 22, 2022 •

edited