[OE-core] [PATCH 07/37] modutils-initscripts: fix wrong order of module loading happening in udev

[Darren Hart]

On 04/25/2011 10:14 AM, Gary Thomas wrote:
> On 04/25/2011 09:49 AM, Saul Wold wrote:
>> On 04/25/2011 05:02 AM, Gary Thomas wrote:
>>> On 04/23/2011 12:28 AM, Saul Wold wrote:
>>>> From: =?UTF-8?q?Petr=20=C5=A0tetiar?=<ynezz at true.cz>
>>>>
>>>> In the current state, udev init script is loaded as 3rd and
>>>> modutils.sh as
>>>> 20th, so udevadm called in udev init script loads all kernel modules
>>>> in some
>>>> unpredictable order, not honoring in any way specified order of
>>>> modules in
>>>> /etc/modules. This causes some troubles mainly in the first boot. So
>>>> to fix
>>>> this we now move loading of the kernel modules just before we exec
>>>> udev init
>>>> script. Example of the current state:
>>>>
>>>> Starting udev
>>>> rtc-m48t86 rtc-m48t86: rtc core: registered m48t86 as rtc0
>>>> rtc-m48t86 rtc-m48t86: battery ok
>>>> ep93xx-rtc ep93xx-rtc: rtc core: registered ep93xx-rtc as rtc1
>>>>
>>>> root at ts72xx:~# cat /etc/modules
>>>> rtc-ep93xx
>>>> ohci-hcd
>>>> ts72xx_sbcinfo
>>>> rtc-m48t86
>>>>
>>>> As you can guess, this is really wrong behaviour, because I've rtc1 as
>>>> the
>>>> main time source for my system.
>>>>
>>>> root at ts72xx:~# cat /etc/default/hwclock
>>>> HWCLOCKACCESS=yes
>>>> HWCLOCKDEVICE=/dev/rtc1
>>>>
>>>> Expected behaviour with this patch applied:
>>>>
>>>> Please wait: booting...
>>>> ep93xx-rtc ep93xx-rtc: rtc core: registered ep93xx-rtc as rtc0
>>>> usbcore: registered new device driver usb
>>>> ohci_hcd: USB 1.1 'Open' Host Controller (OHCI) Driver
>>>> ...
>>>> rtc-m48t86 rtc-m48t86: rtc core: registered m48t86 as rtc1
>>>> rtc-m48t86 rtc-m48t86: battery ok
>>>> Starting udev
>>>>
>>>> Signed-off-by: Petr Štetiar<ynezz at true.cz>
>>>>
>>>> (based on http://patches.openembedded.org/patch/1917/)
>>>> Signed-off-by: Koen Kooi<koen at dominion.thruhere.net>
>>>> Signed-off-by: Saul Wold<sgw at linux.intel.com>
>>>> ---
>>>> .../modutils/modutils-initscripts.bb | 4 ++--
>>>> 1 files changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/meta/recipes-kernel/modutils/modutils-initscripts.bb
>>>> b/meta/recipes-kernel/modutils/modutils-initscripts.bb
>>>> index 1665955..5ae34b4 100644
>>>> --- a/meta/recipes-kernel/modutils/modutils-initscripts.bb
>>>> +++ b/meta/recipes-kernel/modutils/modutils-initscripts.bb
>>>> @@ -4,10 +4,10 @@ LICENSE = "PD"
>>>> LIC_FILES_CHKSUM = "file://LICENSE;md5=7bf87fc37976e93ec66ad84fac58c098"
>>>> SRC_URI = "file://modutils.sh \
>>>> file://PD.patch"
>>>> -PR = "r4"
>>>> +PR = "r5"
>>>>
>>>> INITSCRIPT_NAME = "modutils.sh"
>>>> -INITSCRIPT_PARAMS = "start 20 S ."
>>>> +INITSCRIPT_PARAMS = "start 2 S ."
>>>>
>>>> inherit update-rc.d
>>>>
>>>
>>> Should there not be a corresponding change to modutils_2.4.27.bb? It's
>>> not clear to me
>>> where there are these two separate recipes, but they should at least be
>>> kept in sync.
>>>
>> Gary,
>>
>> Could you please resend this as a proper patch with Signed-off-by and commit info.
> 
> Attached.  However, as Koen pointed out, perhaps this recipe should just be dropped,
> at least for Poky?  I don't know that 2.4 kernels are built or supported?


Right, so for clarification, for 2.4 kernels, images should be built
using just the modutils_2.4.27.bb recipe. For 2.6 kernels, images should
be built using module_init_tools and modutils-initscripts.

In oe-core, task-base requires module-init-tools and task-core-boot
requires modutils-initscripts. That actually seems backwards to me as
modutils-initscripts is useless without module-init-tools.

There is a considerable amount of cruft in the kernel base classes
related to older kernels which I'd like to see purged. Perhaps a
meta-linux-2.4 layer would be a good place to keep things like these as
well as the modutils_2.4.27 recipe.

However, in terms of the immediate issue, I agree with Gary's patch.

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel