[PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

Currently some Qualcomm interconnect drivers are enabled
as modules which isn't overly useful since the interconnects
are required to be loaded during early boot.

Loading the interconnects late (e.g. in initrd or as module)
can cause boot issues, such as slowdown or even not booting
at all (since the interconnect would be required for storage
devices).

Be consistent and enable all of the Qualcomm interconnect
drivers as built-in to the kernel image.

Signed-off-by: Christopher Obbard <christopher.obbard@linaro.org>
---
 arch/arm64/configs/defconfig | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
index 219ef05ee5a757c43a37ec9f8571ce9976354830..6582baee2ab02ecb2ff442c6e73aa6a23fee8d7f 100644
--- a/arch/arm64/configs/defconfig
+++ b/arch/arm64/configs/defconfig
@@ -1656,11 +1656,11 @@ CONFIG_INTERCONNECT_IMX8MN=m
 CONFIG_INTERCONNECT_IMX8MQ=m
 CONFIG_INTERCONNECT_IMX8MP=y
 CONFIG_INTERCONNECT_QCOM=y
-CONFIG_INTERCONNECT_QCOM_MSM8916=m
+CONFIG_INTERCONNECT_QCOM_MSM8916=y
 CONFIG_INTERCONNECT_QCOM_MSM8996=y
-CONFIG_INTERCONNECT_QCOM_OSM_L3=m
+CONFIG_INTERCONNECT_QCOM_OSM_L3=y
 CONFIG_INTERCONNECT_QCOM_QCM2290=y
-CONFIG_INTERCONNECT_QCOM_QCS404=m
+CONFIG_INTERCONNECT_QCOM_QCS404=y
 CONFIG_INTERCONNECT_QCOM_QCS615=y
 CONFIG_INTERCONNECT_QCOM_QCS8300=y
 CONFIG_INTERCONNECT_QCOM_QDU1000=y

---
base-commit: b098bcd8278b89cb3eb73fdb6e06dc49af75ad37
change-id: 20250311-wip-obbardc-qcom-defconfig-interconnects-builtin-258fcc961b11

Best regards,
-- 
Christopher Obbard <christopher.obbard@linaro.org>

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

I sent this patch to start the discussion, some things I found:

1) Some interconnects are missing from arm defconfig. Should they be =y too ?
$ grep CONFIG_INTERCONNECT_QCOM arch/arm/configs/qcom_defconfig
CONFIG_INTERCONNECT_QCOM=y
CONFIG_INTERCONNECT_QCOM_MSM8974=m
CONFIG_INTERCONNECT_QCOM_SDX55=m

2) Some interconnects are missing from arm64 defconfig (which should
probably be in there) (I have included just two examples):
$ grep CONFIG_INTERCONNECT drivers/interconnect/qcom/Makefile
obj-$(CONFIG_INTERCONNECT_QCOM_QCS615) += qnoc-qcs615.o
obj-$(CONFIG_INTERCONNECT_QCOM_SM7150) += qnoc-sm7150.o

I can handle these in follow-up or v2 of the patchset as follow-up
commits, please let me know what you'd prefer.

On Tue, 11 Mar 2025 at 19:03, Christopher Obbard
<christopher.obbard@linaro.org> wrote:
>
> Currently some Qualcomm interconnect drivers are enabled
> as modules which isn't overly useful since the interconnects
> are required to be loaded during early boot.
>
> Loading the interconnects late (e.g. in initrd or as module)
> can cause boot issues, such as slowdown or even not booting
> at all (since the interconnect would be required for storage
> devices).
>
> Be consistent and enable all of the Qualcomm interconnect
> drivers as built-in to the kernel image.
>
> Signed-off-by: Christopher Obbard <christopher.obbard@linaro.org>
> ---
>  arch/arm64/configs/defconfig | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
> index 219ef05ee5a757c43a37ec9f8571ce9976354830..6582baee2ab02ecb2ff442c6e73aa6a23fee8d7f 100644
> --- a/arch/arm64/configs/defconfig
> +++ b/arch/arm64/configs/defconfig
> @@ -1656,11 +1656,11 @@ CONFIG_INTERCONNECT_IMX8MN=m
>  CONFIG_INTERCONNECT_IMX8MQ=m
>  CONFIG_INTERCONNECT_IMX8MP=y
>  CONFIG_INTERCONNECT_QCOM=y
> -CONFIG_INTERCONNECT_QCOM_MSM8916=m
> +CONFIG_INTERCONNECT_QCOM_MSM8916=y
>  CONFIG_INTERCONNECT_QCOM_MSM8996=y
> -CONFIG_INTERCONNECT_QCOM_OSM_L3=m
> +CONFIG_INTERCONNECT_QCOM_OSM_L3=y
>  CONFIG_INTERCONNECT_QCOM_QCM2290=y
> -CONFIG_INTERCONNECT_QCOM_QCS404=m
> +CONFIG_INTERCONNECT_QCOM_QCS404=y
>  CONFIG_INTERCONNECT_QCOM_QCS615=y
>  CONFIG_INTERCONNECT_QCOM_QCS8300=y
>  CONFIG_INTERCONNECT_QCOM_QDU1000=y
>
> ---
> base-commit: b098bcd8278b89cb3eb73fdb6e06dc49af75ad37
> change-id: 20250311-wip-obbardc-qcom-defconfig-interconnects-builtin-258fcc961b11
>
> Best regards,
> --
> Christopher Obbard <christopher.obbard@linaro.org>
>

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Dmitry Baryshkov]

On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
> I sent this patch to start the discussion, some things I found:
> 
> 1) Some interconnects are missing from arm defconfig. Should they be =y too ?

No, unless those are required for the UART console.

> $ grep CONFIG_INTERCONNECT_QCOM arch/arm/configs/qcom_defconfig
> CONFIG_INTERCONNECT_QCOM=y
> CONFIG_INTERCONNECT_QCOM_MSM8974=m
> CONFIG_INTERCONNECT_QCOM_SDX55=m
> 
> 2) Some interconnects are missing from arm64 defconfig (which should
> probably be in there) (I have included just two examples):

I think `git log -S CONFIG_INTERCONNECT_QCOM
arch/arm64/configs/defconfig` will answer this question. The drivers are
enabled on the premises of being required for a particular device, not
because they exist in the Linux kernel.

> $ grep CONFIG_INTERCONNECT drivers/interconnect/qcom/Makefile
> obj-$(CONFIG_INTERCONNECT_QCOM_QCS615) += qnoc-qcs615.o
> obj-$(CONFIG_INTERCONNECT_QCOM_SM7150) += qnoc-sm7150.o
> 
> I can handle these in follow-up or v2 of the patchset as follow-up
> commits, please let me know what you'd prefer.
> 
> On Tue, 11 Mar 2025 at 19:03, Christopher Obbard
> <christopher.obbard@linaro.org> wrote:
> >
> > Currently some Qualcomm interconnect drivers are enabled
> > as modules which isn't overly useful since the interconnects
> > are required to be loaded during early boot.
> >
> > Loading the interconnects late (e.g. in initrd or as module)
> > can cause boot issues, such as slowdown or even not booting
> > at all (since the interconnect would be required for storage
> > devices).
> >
> > Be consistent and enable all of the Qualcomm interconnect
> > drivers as built-in to the kernel image.
> >
> > Signed-off-by: Christopher Obbard <christopher.obbard@linaro.org>
> > ---
> >  arch/arm64/configs/defconfig | 6 +++---
> >  1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
> > index 219ef05ee5a757c43a37ec9f8571ce9976354830..6582baee2ab02ecb2ff442c6e73aa6a23fee8d7f 100644
> > --- a/arch/arm64/configs/defconfig
> > +++ b/arch/arm64/configs/defconfig
> > @@ -1656,11 +1656,11 @@ CONFIG_INTERCONNECT_IMX8MN=m
> >  CONFIG_INTERCONNECT_IMX8MQ=m
> >  CONFIG_INTERCONNECT_IMX8MP=y
> >  CONFIG_INTERCONNECT_QCOM=y
> > -CONFIG_INTERCONNECT_QCOM_MSM8916=m
> > +CONFIG_INTERCONNECT_QCOM_MSM8916=y
> >  CONFIG_INTERCONNECT_QCOM_MSM8996=y
> > -CONFIG_INTERCONNECT_QCOM_OSM_L3=m
> > +CONFIG_INTERCONNECT_QCOM_OSM_L3=y
> >  CONFIG_INTERCONNECT_QCOM_QCM2290=y
> > -CONFIG_INTERCONNECT_QCOM_QCS404=m
> > +CONFIG_INTERCONNECT_QCOM_QCS404=y
> >  CONFIG_INTERCONNECT_QCOM_QCS615=y
> >  CONFIG_INTERCONNECT_QCOM_QCS8300=y
> >  CONFIG_INTERCONNECT_QCOM_QDU1000=y
> >
> > ---
> > base-commit: b098bcd8278b89cb3eb73fdb6e06dc49af75ad37
> > change-id: 20250311-wip-obbardc-qcom-defconfig-interconnects-builtin-258fcc961b11
> >
> > Best regards,
> > --
> > Christopher Obbard <christopher.obbard@linaro.org>
> >

-- 
With best wishes
Dmitry

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

Hi Dmitry,

On Tue, 11 Mar 2025 at 19:58, Dmitry Baryshkov <lumag@kernel.org> wrote:
>
> On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
> > I sent this patch to start the discussion, some things I found:
> >
> > 1) Some interconnects are missing from arm defconfig. Should they be =y too ?
>
> No, unless those are required for the UART console.

OK, that makes sense. FWIW the cryptic (to me, at least) commit log on
https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6eee808134ecf1c1093ff1ddfc056dc5e469d0c3
made me think that the interconnects should be built-in on all devices.

Of course, the real problem here is RB3gen2 not actually finding the
UFS/eMMC device due to no interconnect driver.
Until now, I have been building that into the kernel. I will
investigate instead shoving into the initrd (in both debian and
fedora) which should solve my issue and render this patchset useless.


Thanks,

Chris

>
> > $ grep CONFIG_INTERCONNECT_QCOM arch/arm/configs/qcom_defconfig
> > CONFIG_INTERCONNECT_QCOM=y
> > CONFIG_INTERCONNECT_QCOM_MSM8974=m
> > CONFIG_INTERCONNECT_QCOM_SDX55=m
> >
> > 2) Some interconnects are missing from arm64 defconfig (which should
> > probably be in there) (I have included just two examples):
>
> I think `git log -S CONFIG_INTERCONNECT_QCOM
> arch/arm64/configs/defconfig` will answer this question. The drivers are
> enabled on the premises of being required for a particular device, not
> because they exist in the Linux kernel.
>
> > $ grep CONFIG_INTERCONNECT drivers/interconnect/qcom/Makefile
> > obj-$(CONFIG_INTERCONNECT_QCOM_QCS615) += qnoc-qcs615.o
> > obj-$(CONFIG_INTERCONNECT_QCOM_SM7150) += qnoc-sm7150.o
> >
> > I can handle these in follow-up or v2 of the patchset as follow-up
> > commits, please let me know what you'd prefer.
> >
> > On Tue, 11 Mar 2025 at 19:03, Christopher Obbard
> > <christopher.obbard@linaro.org> wrote:
> > >
> > > Currently some Qualcomm interconnect drivers are enabled
> > > as modules which isn't overly useful since the interconnects
> > > are required to be loaded during early boot.
> > >
> > > Loading the interconnects late (e.g. in initrd or as module)
> > > can cause boot issues, such as slowdown or even not booting
> > > at all (since the interconnect would be required for storage
> > > devices).
> > >
> > > Be consistent and enable all of the Qualcomm interconnect
> > > drivers as built-in to the kernel image.
> > >
> > > Signed-off-by: Christopher Obbard <christopher.obbard@linaro.org>
> > > ---
> > >  arch/arm64/configs/defconfig | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
> > > index 219ef05ee5a757c43a37ec9f8571ce9976354830..6582baee2ab02ecb2ff442c6e73aa6a23fee8d7f 100644
> > > --- a/arch/arm64/configs/defconfig
> > > +++ b/arch/arm64/configs/defconfig
> > > @@ -1656,11 +1656,11 @@ CONFIG_INTERCONNECT_IMX8MN=m
> > >  CONFIG_INTERCONNECT_IMX8MQ=m
> > >  CONFIG_INTERCONNECT_IMX8MP=y
> > >  CONFIG_INTERCONNECT_QCOM=y
> > > -CONFIG_INTERCONNECT_QCOM_MSM8916=m
> > > +CONFIG_INTERCONNECT_QCOM_MSM8916=y
> > >  CONFIG_INTERCONNECT_QCOM_MSM8996=y
> > > -CONFIG_INTERCONNECT_QCOM_OSM_L3=m
> > > +CONFIG_INTERCONNECT_QCOM_OSM_L3=y
> > >  CONFIG_INTERCONNECT_QCOM_QCM2290=y
> > > -CONFIG_INTERCONNECT_QCOM_QCS404=m
> > > +CONFIG_INTERCONNECT_QCOM_QCS404=y
> > >  CONFIG_INTERCONNECT_QCOM_QCS615=y
> > >  CONFIG_INTERCONNECT_QCOM_QCS8300=y
> > >  CONFIG_INTERCONNECT_QCOM_QDU1000=y
> > >
> > > ---
> > > base-commit: b098bcd8278b89cb3eb73fdb6e06dc49af75ad37
> > > change-id: 20250311-wip-obbardc-qcom-defconfig-interconnects-builtin-258fcc961b11
> > >
> > > Best regards,
> > > --
> > > Christopher Obbard <christopher.obbard@linaro.org>
> > >
>
> --
> With best wishes
> Dmitry

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Krzysztof Kozlowski]

On 11/03/2025 20:15, Christopher Obbard wrote:
> Hi Dmitry,
> 
> On Tue, 11 Mar 2025 at 19:58, Dmitry Baryshkov <lumag@kernel.org> wrote:
>>
>> On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
>>> I sent this patch to start the discussion, some things I found:
>>>
>>> 1) Some interconnects are missing from arm defconfig. Should they be =y too ?
>>
>> No, unless those are required for the UART console.
> 
> OK, that makes sense. FWIW the cryptic (to me, at least) commit log on
> https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6eee808134ecf1c1093ff1ddfc056dc5e469d0c3
> made me think that the interconnects should be built-in on all devices.
> 
> Of course, the real problem here is RB3gen2 not actually finding the
> UFS/eMMC device due to no interconnect driver.
> Until now, I have been building that into the kernel. I will
> investigate instead shoving into the initrd (in both debian and
> fedora) which should solve my issue and render this patchset useless.

For Qualcomm platforms you are expected to always have initramfs, thus
you will have the modules for UFS/eMMC mounts. I don't understand the
problem which you were trying to solve.

The interconnects were built in *only* because of need for serial
console. Only.

Best regards,
Krzysztof

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

Hi Krzysztof,

On Wed, 12 Mar 2025 at 09:56, Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On 11/03/2025 20:15, Christopher Obbard wrote:
> > Hi Dmitry,
> >
> > On Tue, 11 Mar 2025 at 19:58, Dmitry Baryshkov <lumag@kernel.org> wrote:
> >>
> >> On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
> >>> I sent this patch to start the discussion, some things I found:
> >>>
> >>> 1) Some interconnects are missing from arm defconfig. Should they be =y too ?
> >>
> >> No, unless those are required for the UART console.
> >
> > OK, that makes sense. FWIW the cryptic (to me, at least) commit log on
> > https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6eee808134ecf1c1093ff1ddfc056dc5e469d0c3
> > made me think that the interconnects should be built-in on all devices.
> >
> > Of course, the real problem here is RB3gen2 not actually finding the
> > UFS/eMMC device due to no interconnect driver.
> > Until now, I have been building that into the kernel. I will
> > investigate instead shoving into the initrd (in both debian and
> > fedora) which should solve my issue and render this patchset useless.
>
> For Qualcomm platforms you are expected to always have initramfs, thus
> you will have the modules for UFS/eMMC mounts. I don't understand the
> problem which you were trying to solve.
>
> The interconnects were built in *only* because of need for serial
> console. Only.

Thanks for confirming. It is all clear now.

Consider this patch dropped from my side.

For reference, I am working on updating initramfs generation tools in
Debian/Fedora to include the required interconnect modules. Currently
the interconnect drivers are built as modules in these distros, but
are not included in the initrd. That is where my confusion initially
stemmed from.

Thanks,

Chris

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Krzysztof Kozlowski]

On 12/03/2025 12:10, Christopher Obbard wrote:
> Hi Krzysztof,
> 
> On Wed, 12 Mar 2025 at 09:56, Krzysztof Kozlowski <krzk@kernel.org> wrote:
>>
>> On 11/03/2025 20:15, Christopher Obbard wrote:
>>> Hi Dmitry,
>>>
>>> On Tue, 11 Mar 2025 at 19:58, Dmitry Baryshkov <lumag@kernel.org> wrote:
>>>>
>>>> On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
>>>>> I sent this patch to start the discussion, some things I found:
>>>>>
>>>>> 1) Some interconnects are missing from arm defconfig. Should they be =y too ?
>>>>
>>>> No, unless those are required for the UART console.
>>>
>>> OK, that makes sense. FWIW the cryptic (to me, at least) commit log on
>>> https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6eee808134ecf1c1093ff1ddfc056dc5e469d0c3
>>> made me think that the interconnects should be built-in on all devices.
>>>
>>> Of course, the real problem here is RB3gen2 not actually finding the
>>> UFS/eMMC device due to no interconnect driver.
>>> Until now, I have been building that into the kernel. I will
>>> investigate instead shoving into the initrd (in both debian and
>>> fedora) which should solve my issue and render this patchset useless.
>>
>> For Qualcomm platforms you are expected to always have initramfs, thus
>> you will have the modules for UFS/eMMC mounts. I don't understand the
>> problem which you were trying to solve.
>>
>> The interconnects were built in *only* because of need for serial
>> console. Only.
> 
> Thanks for confirming. It is all clear now.
> 
> Consider this patch dropped from my side.
> 
> For reference, I am working on updating initramfs generation tools in
> Debian/Fedora to include the required interconnect modules. Currently
> the interconnect drivers are built as modules in these distros, but
> are not included in the initrd. That is where my confusion initially
> stemmed from.

Sure. This defconfig is anyway for us - developers, not for the distros
to use directly. Distros have much bigger configs and use almost
everything as module.

Best regards,
Krzysztof

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

Hi Krzysztof,

On Wed, 12 Mar 2025 at 12:34, Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On 12/03/2025 12:10, Christopher Obbard wrote:
> > Hi Krzysztof,
> >
> > On Wed, 12 Mar 2025 at 09:56, Krzysztof Kozlowski <krzk@kernel.org> wrote:
> >>
> >> On 11/03/2025 20:15, Christopher Obbard wrote:
> >>> Hi Dmitry,
> >>>
> >>> On Tue, 11 Mar 2025 at 19:58, Dmitry Baryshkov <lumag@kernel.org> wrote:
> >>>>
> >>>> On Tue, Mar 11, 2025 at 07:10:06PM +0100, Christopher Obbard wrote:
> >>>>> I sent this patch to start the discussion, some things I found:
> >>>>>
> >>>>> 1) Some interconnects are missing from arm defconfig. Should they be =y too ?
> >>>>
> >>>> No, unless those are required for the UART console.
> >>>
> >>> OK, that makes sense. FWIW the cryptic (to me, at least) commit log on
> >>> https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6eee808134ecf1c1093ff1ddfc056dc5e469d0c3
> >>> made me think that the interconnects should be built-in on all devices.
> >>>
> >>> Of course, the real problem here is RB3gen2 not actually finding the
> >>> UFS/eMMC device due to no interconnect driver.
> >>> Until now, I have been building that into the kernel. I will
> >>> investigate instead shoving into the initrd (in both debian and
> >>> fedora) which should solve my issue and render this patchset useless.
> >>
> >> For Qualcomm platforms you are expected to always have initramfs, thus
> >> you will have the modules for UFS/eMMC mounts. I don't understand the
> >> problem which you were trying to solve.
> >>
> >> The interconnects were built in *only* because of need for serial
> >> console. Only.
> >
> > Thanks for confirming. It is all clear now.
> >
> > Consider this patch dropped from my side.
> >
> > For reference, I am working on updating initramfs generation tools in
> > Debian/Fedora to include the required interconnect modules. Currently
> > the interconnect drivers are built as modules in these distros, but
> > are not included in the initrd. That is where my confusion initially
> > stemmed from.
>
> Sure. This defconfig is anyway for us - developers, not for the distros
> to use directly. Distros have much bigger configs and use almost
> everything as module.

Completely agree - my usual workflow is to build with the kernel's
defconfig (disto config is generally huge) and install the resulting
kernel into my Debian image.

Installing a kernel built with `make bindeb-pkg` then calls
update-initramfs to create the initrd - which then should pack the
RIGHT modules needed for loading storage/network boot/etc into the
initrd. In this case it simply didn't yet copy that module into the
initrd in Debian, so I will sort it out in the distro space ;-).

We can likely stop this discussion now as it's clear to everyone
what's going on I think.

Thanks,

Chris

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Brian Masney]

Hi Christopher,

On Wed, Mar 12, 2025 at 12:10:56PM +0100, Christopher Obbard wrote:
> For reference, I am working on updating initramfs generation tools in
> Debian/Fedora to include the required interconnect modules. Currently
> the interconnect drivers are built as modules in these distros, but
> are not included in the initrd. That is where my confusion initially
> stemmed from.

From a Fedora and centos-stream-9/10 perspective, we have dracut
updated so that the interconnect modules are included in the initramfs
by default.

https://github.com/dracutdevs/dracut/blob/master/modules.d/90kernel-modules/module-setup.sh#L74

Let me know if you are seeing a specific issue with the initramfs on
Fedora and I can help you.

We also configured all of the QC pinctrl drivers to be modules in
Fedora, and this causes a large number of probe deferrals on boot up.
One of the boots I measured was about a half second:

https://lore.kernel.org/linux-embedded/Z6J3WpeJKIKENC81@x13s/

Everything eventually loads and works as expected though. The upstream
defconfig configures most of the pinctrl drivers to be built in for
Qualcomm.

Brian

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Christopher Obbard]

Hi Brian,

On Wed, 12 Mar 2025 at 19:40, Brian Masney <bmasney@redhat.com> wrote:
>
> Hi Christopher,
>
> On Wed, Mar 12, 2025 at 12:10:56PM +0100, Christopher Obbard wrote:
> > For reference, I am working on updating initramfs generation tools in
> > Debian/Fedora to include the required interconnect modules. Currently
> > the interconnect drivers are built as modules in these distros, but
> > are not included in the initrd. That is where my confusion initially
> > stemmed from.
>
> From a Fedora and centos-stream-9/10 perspective, we have dracut
> updated so that the interconnect modules are included in the initramfs
> by default.
>
> https://github.com/dracutdevs/dracut/blob/master/modules.d/90kernel-modules/module-setup.sh#L74
>
> Let me know if you are seeing a specific issue with the initramfs on
> Fedora and I can help you.

Awesome, turns out I am wrong and the interconnect drivers are in fact
present in the latest nightly Fedora 42 image I am testing:

$ lsinitramfs initramfs-6.14.0-0.rc3.29.fc42.aarch64.img  | grep interconnect
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx/imx-interconnect.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx/imx8mm-interconnect.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx/imx8mn-interconnect.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx/imx8mp-interconnect.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/imx/imx8mq-interconnect.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/icc-osm-l3.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/icc-smd-rpm.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-msm8916.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-msm8953.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-msm8996.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-qcm2290.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sa8775p.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sc7280.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sc8180x.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sc8280xp.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sdm845.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sdx75.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sm6115.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sm8150.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sm8250.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-sm8450.ko.xz
usr/lib/modules/6.14.0-0.rc3.29.fc42.aarch64/kernel/drivers/interconnect/qcom/qnoc-x1e80100.ko.xz



Thanks,

Chris

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Dmitry Baryshkov]

On Tue, Mar 11, 2025 at 07:02:56PM +0100, Christopher Obbard wrote:
> Currently some Qualcomm interconnect drivers are enabled
> as modules which isn't overly useful since the interconnects
> are required to be loaded during early boot.
> 
> Loading the interconnects late (e.g. in initrd or as module)
> can cause boot issues, such as slowdown or even not booting
> at all (since the interconnect would be required for storage
> devices).

This is not a good justification. It should be perfectly fine to load
block drivers (including their dependencies) from the initramfs.

Up to now we have been enabling only those interconnect (and pinctrl,
btw) drivers, which are required to be able to open the consoe (thanks,
systemd).

> 
> Be consistent and enable all of the Qualcomm interconnect
> drivers as built-in to the kernel image.
> 
> Signed-off-by: Christopher Obbard <christopher.obbard@linaro.org>
> ---
>  arch/arm64/configs/defconfig | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
> index 219ef05ee5a757c43a37ec9f8571ce9976354830..6582baee2ab02ecb2ff442c6e73aa6a23fee8d7f 100644
> --- a/arch/arm64/configs/defconfig
> +++ b/arch/arm64/configs/defconfig
> @@ -1656,11 +1656,11 @@ CONFIG_INTERCONNECT_IMX8MN=m
>  CONFIG_INTERCONNECT_IMX8MQ=m
>  CONFIG_INTERCONNECT_IMX8MP=y
>  CONFIG_INTERCONNECT_QCOM=y
> -CONFIG_INTERCONNECT_QCOM_MSM8916=m
> +CONFIG_INTERCONNECT_QCOM_MSM8916=y
>  CONFIG_INTERCONNECT_QCOM_MSM8996=y
> -CONFIG_INTERCONNECT_QCOM_OSM_L3=m
> +CONFIG_INTERCONNECT_QCOM_OSM_L3=y

This is the memory / L3 / cpufreq interconnect, it has nothing to do
with the block devices.

>  CONFIG_INTERCONNECT_QCOM_QCM2290=y
> -CONFIG_INTERCONNECT_QCOM_QCS404=m
> +CONFIG_INTERCONNECT_QCOM_QCS404=y
>  CONFIG_INTERCONNECT_QCOM_QCS615=y
>  CONFIG_INTERCONNECT_QCOM_QCS8300=y
>  CONFIG_INTERCONNECT_QCOM_QDU1000=y
> 
> ---
> base-commit: b098bcd8278b89cb3eb73fdb6e06dc49af75ad37
> change-id: 20250311-wip-obbardc-qcom-defconfig-interconnects-builtin-258fcc961b11
> 
> Best regards,
> -- 
> Christopher Obbard <christopher.obbard@linaro.org>
> 

-- 
With best wishes
Dmitry

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Konrad Dybcio]

On 3/11/25 7:52 PM, Dmitry Baryshkov wrote:
> On Tue, Mar 11, 2025 at 07:02:56PM +0100, Christopher Obbard wrote:
>> Currently some Qualcomm interconnect drivers are enabled
>> as modules which isn't overly useful since the interconnects
>> are required to be loaded during early boot.
>>
>> Loading the interconnects late (e.g. in initrd or as module)
>> can cause boot issues, such as slowdown or even not booting
>> at all (since the interconnect would be required for storage
>> devices).
> 
> This is not a good justification. It should be perfectly fine to load
> block drivers (including their dependencies) from the initramfs.
> 
> Up to now we have been enabling only those interconnect (and pinctrl,
> btw) drivers, which are required to be able to open the consoe (thanks,
> systemd).

(non-earlycon) console is a better argument here, though

Konrad

Re: [PATCH] arm64: defconfig: Enable Qualcomm interconnects as built-in

[Krzysztof Kozlowski]

On 11/03/2025 19:02, Christopher Obbard wrote:
> Currently some Qualcomm interconnect drivers are enabled
> as modules which isn't overly useful since the interconnects
> are required to be loaded during early boot.
> 
> Loading the interconnects late (e.g. in initrd or as module)
> can cause boot issues, such as slowdown or even not booting
> at all (since the interconnect would be required for storage
> devices).
> 
> Be consistent and enable all of the Qualcomm interconnect
> drivers as built-in to the kernel image.


I don't agree. If you want consistency, fix booting process and make
everything module. We switch them to builtin on exception, so come with
reasoning for given exceptions instead of making exceptions the new normal.

Best regards,
Krzysztof