Load proper sound drivers for QEMU x86
cover Load proper sound drivers for QEMU x86
From: Mark Jonas <[email protected]>
The patch series fixes a regression where images built for QEMU x86 and x86-64 did not contain sound drivers matching the QEMU device emulation configuration. Additionally, the right drivers are now also auto-loaded when starting the image.
root@qemux86:~# uname -a Linux qemux86 5.19.17-yocto-standard #1 SMP PREEMPT_DYNAMIC Fri Nov 18 21:13:18 UTC 2022 i686 GNU/Linux
root@qemux86:~# lsmod Module Size Used by bnep 20480 2 snd_intel8x0 32768 0 snd_ac97_codec 114688 1 snd_intel8x0 ac97_bus 16384 1 snd_ac97_codec snd_pcm 86016 2 snd_ac97_codec,snd_intel8x0 snd_timer 28672 1 snd_pcm snd 57344 4 snd_ac97_codec,snd_timer,snd_intel8x0,snd_pcm soundcore 16384 1 snd
root@qemux86:~# cat /proc/asound/cards
0 [I82801AAICH ]: ICH - Intel 82801AA-ICH
Intel 82801AA-ICH with STAC9700,83,84 at irq 10Mark Jonas (2):
conf/machine: Match QEMU x86 sound kernel module
linux-yocto: Autoload sound driver on QEMU x86
meta/conf/machine/qemux86-64.conf | 3 ++- meta/conf/machine/qemux86.conf | 3 ++- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 3 files changed, 9 insertions(+), 2 deletions(-)
--
2.34.1
PATCH 1/2 conf/machine: Match QEMU x86 sound kernel module
From: Mark Jonas <[email protected]>
In "conf/machine: fix QEMU x86 sound options" b802a5dd the sound devices emulated by QEMU were changed from AC97 (intel8x0) and ES1370 (Soundblaster) to just AC97. Yet, the sound kernel modules added to the file system still matched ES1370.
This patch adds the matching snd-intel8x0 kernel module for the emulated AC97 sound card. It removes the unused ES1370 kernel modules.
--- meta/conf/machine/qemux86-64.conf | 3 ++- meta/conf/machine/qemux86.conf | 3 ++- 2 files changed, 4 insertions(+), 2 deletions(-)
meta/conf/machine/qemux86-64.conf +2 -1
@@ -36,7 +36,8 @@ MACHINE_FEATURES += "x86 pci" MACHINE_ESSENTIAL_EXTRA_RDEPENDS += "v86d" -MACHINE_EXTRA_RRECOMMENDS = "kernel-module-snd-ens1370 kernel-module-snd-rawmidi" +# matches QEMU -device AC97 +MACHINE_EXTRA_RRECOMMENDS = "kernel-module-snd-intel8x0" WKS_FILE ?= "qemux86-directdisk.wks" do_image_wic[depends] += "syslinux:do_populate_sysroot syslinux-native:do_populate_sysroot mtools-native:do_populate_sysroot dosfstools-native:do_populate_sysroot"
meta/conf/machine/qemux86.conf +2 -1
@@ -33,7 +33,8 @@ MACHINE_FEATURES += "x86 pci" MACHINE_ESSENTIAL_EXTRA_RDEPENDS += "v86d" -MACHINE_EXTRA_RRECOMMENDS = "kernel-module-snd-ens1370 kernel-module-snd-rawmidi" +# matches QEMU -device AC97 +MACHINE_EXTRA_RRECOMMENDS = "kernel-module-snd-intel8x0" WKS_FILE ?= "qemux86-directdisk.wks" do_image_wic[depends] += "syslinux:do_populate_sysroot syslinux-native:do_populate_sysroot mtools-native:do_populate_sysroot dosfstools-native:do_populate_sysroot"
Signature · 1 line
2.34.1
PATCH 2/2 linux-yocto: Autoload sound driver on QEMU x86
In reply to [email protected]From: Mark Jonas <[email protected]>
If DISTRO_FEATURES includes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurations conf/machine/qemux86.conf and qemux86-64.conf.
--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)
meta/recipes-kernel/linux/linux-yocto.inc +5 -0
@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}" + # A KMACHINE is the mapping of a yocto $MACHINE to what is built # by the kernel. This is typically the branch that should be built, # and it can be specific to the machine or shared
Signature · 1 line
2.34.1
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0
meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"
This gets us most of the way, but if we are going to do this we should complete the job.
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.
It would then be something we'd add to the KERNEL_FEATRES triggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.
Bruce
Quoted 16 lines
meta/recipes-kernel/linux/linux-yocto.inc +1
-1+ # A KMACHINE is the mapping of a yocto $MACHINE to what is built # by the kernel. This is typically the branch that should be built, # and it can be specific to the machine or shared --2.34.1
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#175471): https://lists.openembedded.org/g/openembedded-core/message/175471
Mute This Topic: https://lists.openembedded.org/mt/96048827/1050810
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end - "Use the force Harry" - Gandalf, Star Trek II
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.
In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Does that maybe mean I should better add the KERNEL_MODULE_AUTOLOAD into the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.
Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
Cheers,
Mark
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a KERNEL_FEATURE that is set based on a machine or distro feature.
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
Bruce
Quoted 3 lines
Cheers,
Mark
--
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by MACHINE_EXTRA_RRECOMMENDS in the machine file.
Quoted 5 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a MACHINE_FEATURE. And that is set in machine/include/qemu.inc.
Quoted 4 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
I was testing on Poky with the default init system i.e, sysvinit. Without the KERNEL_MODULE_AUTOLOAD line the snd-intel8x0 module is not loaded.
And it seems you are right: The KERNEL_MODULE_AUTOLOAD is superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.
Cheers,
Mark
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Quoted 77 lines · Mark Jonas
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.
Quoted 7 lines
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by
MACHINE_EXTRA_RRECOMMENDSin the machine file.
I might also be misunderstanding but I think what Bruce is asking is that the qemux86 sound pieces be broken out of cfg/sound.scc. In case you didn't find them, the files are here:
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.scc
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.cfg
and the question is which of the options in sound.cfg are the drivers which qemu needs on x86.
I think Bruce would like a patch which splits those items out and triggers them on qemux86 specifically.
This means if things change in future, we know exactly which module options we need for sound on qemux86.
Bruce can correct me if I misunderstand! :)
Quoted 32 lines
Quoted 5 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a
MACHINE_FEATURE. And that is set in machine/include/qemu.inc.Quoted 4 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
Quoted 7 lines
I was testing on Poky with the default init system i.e, sysvinit. Without the
KERNEL_MODULE_AUTOLOADline the snd-intel8x0 module is not loaded.And it seems you are right: The
KERNEL_MODULE_AUTOLOADis superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.
I'm not sure on that. The kernel will try and autoprobe for drivers for hardware it has present where it can though.
Cheers,
Richard
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Hi Richard and Bruce,
Quoted 97 lines · Richard Purdie
Quoted 77 lines · Mark Jonas
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.Quoted 7 lines
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by
MACHINE_EXTRA_RRECOMMENDSin the machine file.I might also be misunderstanding but I think what Bruce is asking is that the qemux86 sound pieces be broken out of cfg/sound.scc. In case you didn't find them, the files are here:
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.scc
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.cfgand the question is which of the options in sound.cfg are the drivers which qemu needs on x86.
Quoted 7 lines
I think Bruce would like a patch which splits those items out and triggers them on qemux86 specifically.
This means if things change in future, we know exactly which module options we need for sound on qemux86.
Bruce can correct me if I misunderstand! :)
Yes, that makes sense to me. But I have to admit that I neither have the time nor the sound driver knowledge to do this properly. I am afraid that it would not be sufficient to simply break out the driver needed for QEMU ie., intel8x0 and its dependencies. sound.cfg contains support for a lot of different sound cards which share kernel config options / modules. It can be done and the scc files support that but that is too much time needed for me right now. Sorry.
What I actually wanted to achieve is to build Doom with YP and run it in QEMU. And even though the correct sound drivers are now loaded I do not get sound using e.g., aplay /usr/share/sounds/alsa/Front_Center.wav.
Quoted 36 lines
Quoted 36 lines
Quoted 5 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a
MACHINE_FEATURE. And that is set in machine/include/qemu.inc.Quoted 4 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
I was testing on Poky with the default init system i.e, sysvinit. Without the
KERNEL_MODULE_AUTOLOADline the snd-intel8x0 module is not loaded.
Quoted 6 lines
Quoted 3 lines
And it seems you are right: The
KERNEL_MODULE_AUTOLOADis superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.I'm not sure on that. The kernel will try and autoprobe for drivers for hardware it has present where it can though.
I'll wait for Bruce's feedback and then send an updated patch which fixes the QEMU machine file.
Cheers,
Mark
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
Quoted 110 lines · Mark Jonas
Hi Richard and Bruce,
Quoted 97 lines · Richard Purdie
Quoted 77 lines · Mark Jonas
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.Quoted 7 lines
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by
MACHINE_EXTRA_RRECOMMENDSin the machine file.I might also be misunderstanding but I think what Bruce is asking is that the qemux86 sound pieces be broken out of cfg/sound.scc. In case you didn't find them, the files are here:
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.scc
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.cfgand the question is which of the options in sound.cfg are the drivers which qemu needs on x86.
Quoted 7 lines
I think Bruce would like a patch which splits those items out and triggers them on qemux86 specifically.
This means if things change in future, we know exactly which module options we need for sound on qemux86.
Bruce can correct me if I misunderstand! :)
Quoted 7 lines
Yes, that makes sense to me. But I have to admit that I neither have the time nor the sound driver knowledge to do this properly. I am afraid that it would not be sufficient to simply break out the driver needed for QEMU ie., intel8x0 and its dependencies. sound.cfg contains support for a lot of different sound cards which share kernel config options / modules. It can be done and the scc files support that but that is too much time needed for me right now. Sorry.
I can help with this part, since I don't want to lose the contribution by asking for too much!
It really is just putting the =m configuration for the modules into a separate config, with any direct dependencies in that same file.
If you send a patch that does the tweaking of the enablement, I can pull it into my queue and split out the fragments to something specific to what we need, and submit them all as a series.
Bruce
Quoted 54 lines
What I actually wanted to achieve is to build Doom with YP and run it in QEMU. And even though the correct sound drivers are now loaded I do not get sound using e.g., aplay
/usr/share/sounds/alsa/Front_Center.wav.Quoted 36 lines
Quoted 36 lines
Quoted 5 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a
MACHINE_FEATURE. And that is set in machine/include/qemu.inc.Quoted 4 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
I was testing on Poky with the default init system i.e, sysvinit. Without the
KERNEL_MODULE_AUTOLOADline the snd-intel8x0 module is not loaded.Quoted 6 lines
Quoted 3 lines
And it seems you are right: The
KERNEL_MODULE_AUTOLOADis superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.I'm not sure on that. The kernel will try and autoprobe for drivers for hardware it has present where it can though.
I'll wait for Bruce's feedback and then send an updated patch which fixes the QEMU machine file.
Cheers,
Mark
Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end - "Use the force Harry" - Gandalf, Star Trek II
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
In reply to Richard PurdieQuoted 97 lines · Richard Purdie
Quoted 77 lines · Mark Jonas
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.Quoted 7 lines
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by
MACHINE_EXTRA_RRECOMMENDSin the machine file.I might also be misunderstanding but I think what Bruce is asking is that the qemux86 sound pieces be broken out of cfg/sound.scc. In case you didn't find them, the files are here:
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.scc
https://git.yoctoproject.org/yocto-kernel-cache/tree/cfg/sound.cfgand the question is which of the options in sound.cfg are the drivers which qemu needs on x86.
Quoted 7 lines
I think Bruce would like a patch which splits those items out and triggers them on qemux86 specifically.
This means if things change in future, we know exactly which module options we need for sound on qemux86.
Bruce can correct me if I misunderstand! :)
That is correct!
Something like sonud-intel.scc/cfg or sound-qemu.scc/cfg that just contains what we need for those particular drivers, as they have a well defined use case, and there's no need for ALL the other drivers to be enabled, built and packaged if you really know your target machine.
The KERNEL_FEATURES append would then be something like:
SOUND_FEATURES="${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound-intel.scc', '', d)}" KERNEL_FEATURES:append:qemux86=" ${SOUND_FEATURES} cfg/paravirt_kvm.scc"
Or it could all be done in the single append, etc, but having it in a variable might
give us more flexibility for other qemu or target machines to re-use it.
My example keeps the machine specific enablement in the KERNEL_FEATURES append, but there are any number of ways to do it (since my proposed configuration
fragment has "intel" in it, it could be confusing :))
Bruce
Quoted 49 lines
Quoted 33 lines
Quoted 6 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a
MACHINE_FEATURE. And that is set in machine/include/qemu.inc.Quoted 4 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
Quoted 7 lines
I was testing on Poky with the default init system i.e, sysvinit. Without the
KERNEL_MODULE_AUTOLOADline the snd-intel8x0 module is not loaded.And it seems you are right: The
KERNEL_MODULE_AUTOLOADis superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.I'm not sure on that. The kernel will try and autoprobe for drivers for hardware it has present where it can though.
Cheers,
Richard
Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end - "Use the force Harry" - Gandalf, Star Trek II
Re: [OE-core] [PATCH 2/2] linux-yocto: Autoload sound driver on QEMU x86
In reply to Mark JonasQuoted 78 lines · Mark Jonas
Hi Bruce,
Quoted 66 lines · Bruce Ashfield
Quoted 49 lines · Mark Jonas
Hi Bruce,
Quoted 30 lines · Bruce Ashfield
Quoted 17 lines · Mark Jonas
From: Mark Jonas <[email protected]>
If
DISTRO_FEATURESincludes ALSA then automatically load the snd-intel8x0 kernel module on qemux86 and qemux86-64. This matches the machine configurationsconf/machine/qemux86.confand qemux86-64.conf.Signed-off-by Mark Jonas <[email protected]>--- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 1 file changed, 5 insertions(+)meta/recipes-kernel/linux/linux-yocto.inc +0
-0meta/recipes-kernel/linux/linux-yocto.inc KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ · line 43
Quoted 7 lines
meta/recipes-kernel/linux/linux-yocto.inc +4
-0@@ -37,6 +37,11 @@ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'efi', 'cfg/ KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'numa', 'features/numa/numa.scc', '', d)}" KERNEL_FEATURES:append = " ${@bb.utils.contains('MACHINE_FEATURES', 'vfat', 'cfg/fs/vfat.scc', '', d)}" +# sound driver recommended by conf/machine/qemux86*.conf +ALSA_MODULES = "${@bb.utils.contains("DISTRO_FEATURES", "alsa", "snd-intel8x0", "", d)}" +KERNEL_MODULE_AUTOLOAD:qemux86 += "${ALSA_MODULES}" +KERNEL_MODULE_AUTOLOAD:qemux86-64 += "${ALSA_MODULES}"This gets us most of the way, but if we are going to do this we should complete the job.
Quoted 8 lines
We really need to make sure there's a configuration fragment that explicitly enables
the modules we need (and not count on defaults, or other selects). That would go into the kernel-cache.It would then be something we'd add to the
KERNEL_FEATREStriggered off the distro feature. Just like we are doing with numa and vfat that is visible in the context of the patch.In my understanding this is already the case. See for example linux-yocto_5.19.bb which contains the following lines.
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
Quoted 7 lines
Does that maybe mean I should better add the
KERNEL_MODULE_AUTOLOADinto the same file where the corresponding configuration fragment is added? But that would mean to duplicate the lines over five recipes.Alternatively I could move the cfg/sound.scc out of the recipes into the linux-yocto.inc. But then linux-yocto-tiny_*.bb recipes would also get it.
I definitely wasn't clear. Sorry about that, I'm still partially on holidays and didn't have enough coffee when I wrote that.
That is close to what I meant. It is one thing to define the options in our generic sound "buckets" and another to put a specific module into the autoload.
Quoted 7 lines
If we are going to autoload a module, the option really should be pulled out of the generic bucket, and put into something smaller / specific to what we are trying to do, and trigger the inclusion of that fragment only when the appropriate distro feature is set. The machine conf files specifying specific modules isn't ideal, since it is largely "aspirational" unless it is coupled with a
KERNEL_FEATUREthat is set based on a machine or distro feature.
Quoted 2 lines
I have to admit that I feel a bit lost. I am not really sure I understood what you meant.
I'll reply here, and then to the follow up email. I've read them all and will clarify.
Quoted 5 lines
Is the "generic bucket" cfg/sound.scc? But why should we break that up? A module only ends up in an image if the module is installed. And the installation is controlled e.g. by
MACHINE_EXTRA_RRECOMMENDSin the machine file.
Having really broad configuration buckets that build a lot of "just in case" drivers, aren't a great idea. We've been making things more feature based and refined over the past decade or so.
The "kernel-modules" package will install them all. Just because in this case we are using the individual module packages, doesn't mean that is always the case.
Spreading very specific module package names into machine configuration files is also problematic, as the modules come and go, change names, etc. I'm always juggling a few each kernel upgrade. Keeping that knowledge centralized, makes things easier .. as witnessed by the need to tweak the sound modules in this thread.
Quoted 19 lines
Quoted 6 lines
If you search the mailing list archives, you'll see me musing about how I need to change those unconditional appends of option to qemu, as it makes defining a new qemu machine problematic (if the options are causing issues). The tweak I'm describing is a small step in that direction.
What I understood is that you propose to change the existing
KERNEL_FEATURES:append:qemux86=" cfg/sound.scc cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/sound.scc cfg/paravirt_kvm.scc"
in e.g. linux-yocto_5.19.bb to (remove cfg/sound.scc)
KERNEL_FEATURES:append:qemux86=" cfg/paravirt_kvm.scc" KERNEL_FEATURES:append:qemux86-64=" cfg/paravirt_kvm.scc"
and add (move) the following into linux-yocto.inc
Quoted 8 lines
KERNEL_FEATURES:append:qemuall=" ${@bb.utils.contains('MACHINE_FEATURES', 'alsa', 'cfg/sound.scc', '', d)}"
That is, instead of coupling the compilation of QEMU sound driver modules based on machine name qemux86 and qemux86-64 we activate it for all QEMU machines which also have alsa as a
MACHINE_FEATURE. And that is set in machine/include/qemu.inc.
I'll follow up to this in a reply to Richard's clarification.
Quoted 6 lines
Quoted 5 lines
One new question comes to mind, what init system were you testing with (default poky and sysvinit ?) ? If we now have the proper init system, are no events being triggered to load the modules now that the proper ones are in the machine files ?
Quoted 7 lines
I was testing on Poky with the default init system i.e, sysvinit. Without the
KERNEL_MODULE_AUTOLOADline the snd-intel8x0 module is not loaded.And it seems you are right: The
KERNEL_MODULE_AUTOLOADis superfluous. Even when I remove the second patch the modules are still loaded. I am a little puzzled why it originally seemed to me that it was necessary.
The autoloading is the behaviour that I expected, but I can't explain why it wasn't working in your earlier tests!
Bruce
Quoted 3 lines
Cheers,
Mark
Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end - "Use the force Harry" - Gandalf, Star Trek II
Quoted 25 lines · On Wed, 2023-01-04 at 13:18 +0100, [email protected] wrote:
From: Mark Jonas <[email protected]>
The patch series fixes a regression where images built for QEMU x86 and x86-64 did not contain sound drivers matching the QEMU device emulation configuration. Additionally, the right drivers are now also auto-loaded when starting the image.
root@qemux86:~# uname -a Linux qemux86 5.19.17-yocto-standard #1 SMP PREEMPT_DYNAMIC Fri Nov 18 21:13:18 UTC 2022 i686 GNU/Linuxroot@qemux86:~# lsmod Module Size Used by bnep 20480 2 snd_intel8x0 32768 0 snd_ac97_codec 114688 1 snd_intel8x0 ac97_bus 16384 1 snd_ac97_codec snd_pcm 86016 2 snd_ac97_codec,snd_intel8x0 snd_timer 28672 1 snd_pcm snd 57344 4 snd_ac97_codec,snd_timer,snd_intel8x0,snd_pcm soundcore 16384 1 sndroot@qemux86:~# cat /proc/asound/cards 0 [I82801AAICH ]: ICH - Intel 82801AA-ICH Intel 82801AA-ICH with STAC9700,83,84 at irq 10
Quoted 8 lines
Mark Jonas (2):
conf/machine: Match QEMU x86 sound kernel module
linux-yocto: Autoload sound driver on QEMU x86meta/conf/machine/qemux86-64.conf | 3 ++- meta/conf/machine/qemux86.conf | 3 ++- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 3 files changed, 9 insertions(+), 2 deletions(-)
These look good, I'm happy to see the patches, thanks!
As Bruce mentions, a kernel config fragment would be a nice added bonus to ensure the right config options stay around or error if they're changed upstream.
One thing I've learnt is that the best way to ensure these kinds of things stay "fixed" once we fix them is to tweak our automated testing to check them.
I'm wondering if there is something we could/should add to:
"bitbake core-image-sato -c testimage"
to check for the presence of a /proc/asound/cards entry for example?
The idea here is that now it is working, we'd try and ensure it stays working!
Cheers,
Richard
Hi Richard,
Quoted 36 lines · Richard Purdie
Quoted 25 lines · On Wed, 2023-01-04 at 13:18 +0100, [email protected] wrote:
From: Mark Jonas <[email protected]>
The patch series fixes a regression where images built for QEMU x86 and x86-64 did not contain sound drivers matching the QEMU device emulation configuration. Additionally, the right drivers are now also auto-loaded when starting the image.
root@qemux86:~# uname -a Linux qemux86 5.19.17-yocto-standard #1 SMP PREEMPT_DYNAMIC Fri Nov 18 21:13:18 UTC 2022 i686 GNU/Linuxroot@qemux86:~# lsmod Module Size Used by bnep 20480 2 snd_intel8x0 32768 0 snd_ac97_codec 114688 1 snd_intel8x0 ac97_bus 16384 1 snd_ac97_codec snd_pcm 86016 2 snd_ac97_codec,snd_intel8x0 snd_timer 28672 1 snd_pcm snd 57344 4 snd_ac97_codec,snd_timer,snd_intel8x0,snd_pcm soundcore 16384 1 sndroot@qemux86:~# cat /proc/asound/cards 0 [I82801AAICH ]: ICH - Intel 82801AA-ICH Intel 82801AA-ICH with STAC9700,83,84 at irq 10Quoted 8 lines
Mark Jonas (2):
conf/machine: Match QEMU x86 sound kernel module
linux-yocto: Autoload sound driver on QEMU x86meta/conf/machine/qemux86-64.conf | 3 ++- meta/conf/machine/qemux86.conf | 3 ++- meta/recipes-kernel/linux/linux-yocto.inc | 5 +++++ 3 files changed, 9 insertions(+), 2 deletions(-)
Quoted 5 lines
These look good, I'm happy to see the patches, thanks!
As Bruce mentions, a kernel config fragment would be a nice added bonus to ensure the right config options stay around or error if they're changed upstream.
Quick summary of what I replied to Bruce: The linux-yocto*.bb recipes (except for the tiny) already use a cfg/sound.scc to enable building of sound driver modules.
Quoted 4 lines
One thing I've learnt is that the best way to ensure these kinds of things stay "fixed" once we fix them is to tweak our automated testing to check them.
Quoted 8 lines
I'm wondering if there is something we could/should add to:
"bitbake core-image-sato -c testimage"
to check for the presence of a
/proc/asound/cardsentry for example?The idea here is that now it is working, we'd try and ensure it stays working!
Sounds reasonable. I haven't used automated tests for OE yet so it might take me some time to dig into it.
Cheers,
Mark