I suspect this just caused a regression:

https://sof-ci.01.org/sofpr/PR7684/build8467/devicetest/index.html?model=TGLU_RVP_SDW_IPC4ZPH&testcase=check-kmod-load-unload-after-playback

RMMOD	regmap_sdw
rmmod: ERROR: Module regmap_sdw is in use by: snd_soc_rt712_sdca_dmic snd_soc_rt712_sdca

Same error across multiple devices.

EDIT: or maybe this is an expected kernel change that this PR was not enough to support?

Failed the same in totally unrelated https://sof-ci.01.org/linuxpr/PR4376/build5091/devicetest/index.html

Also observed all across daily test run 26460 (stable-v2.2) and 26441 (IPC4) on 2023-05-25. Not affecting daily MTL because it's not running kmod-unload-reload (forgot why).

@marc-hb @plbossart Below change can fix the issue.

diff --git a/tools/kmod/sof_remove.sh b/tools/kmod/sof_remove.sh
index 45cb3e4..d87096e 100755
--- a/tools/kmod/sof_remove.sh
+++ b/tools/kmod/sof_remove.sh
@@ -210,8 +210,8 @@ remove_module snd_soc_rt298
 remove_module snd_soc_rt700
 remove_module snd_soc_rt711
 remove_module snd_soc_rt711_sdca
-remove_module snd-soc-rt712-sdca
-remove_module snd-soc-rt712-sdca-dmic
+remove_module snd_soc_rt712_sdca
+remove_module snd_soc_rt712_sdca_dmic
 remove_module snd_soc_rt715
 remove_module snd_soc_rt715_sdca
 remove_module snd_soc_rt1308

sudo modprobe -r snd-soc-rt712-sdca works, but somehow snd-soc-rt712-sdca doesn't work in sof_remove.sh

Thanks @bardliao , this looks great! Can you please submit the PR?

but somehow snd-soc-rt712-sdca doesn't work in sof_remove.sh

That's because sof_remove.sh searches for the "real" name in /proc/modules.

Fixed in #1040 .

Any idea why the failure did not seem systematic?

Fixed in #1040 .

Any idea why the failure did not seem systematic?

it's probably a miss where the RT712 was not included in the build when this change was added. I copied the name of the module from the Makefile, and didn't think of the difference between _ and -. Not sure what this matters, usually modprobe does not care.

it's probably a miss where the RT712 was not included in the build when this change was added.

Thanks, that would explain.

Not sure what this matters, usually modprobe does not care.

As mentioned above the script searches /proc/modules so it can show a nice message that the module was already unloaded as opposed to trying to unload it anyway, having to ignore errors and then fail to catch actual failures to unload.