All of lore.kernel.org
 help / color / mirror / Atom feed
* Issues creating /dev/ nodes in an initramfs
@ 2014-10-19 18:09 Mitchell Skiba
  2014-10-22  8:05 ` ChenQi
  0 siblings, 1 reply; 2+ messages in thread
From: Mitchell Skiba @ 2014-10-19 18:09 UTC (permalink / raw)
  To: yocto

[-- Attachment #1: Type: text/plain, Size: 1699 bytes --]

I've been building a kernel with an initramfs, by using INITRAMFS_IMAGE to
specify a recipe that generates an image as a .cpio.gz archive. I've been
having issues creating /dev/console (or any other device file) in the
image. (The 3.14 kernel panics if /dev/console is not a character device.)

It seems there is already a way to create dev nodes via the
IMAGE_DEVICE_TABLE and IMAGE_DEVICE_TABLES variables. However this
mechanism, through the makedevs utility, appears to require the building
user to have permission to call mknod. (Makedevs ignores the return code
from the mknod call, so it doesn't outright fail on the permissions error,
but device files will not be created.)

I've hacked around it in my image recipe by appending a step onto
do_rootfs, but I'm not too happy with the way I did it. It seems like the
proper way to fix this is to change the image commands in
image_types.bbclass. However, each image type would need a different way of
adding in the special files. (At this point I only care about cpio archives
for the initramfs.)


Does anyone have the current IMAGE_DEVICE_TABLE implementation working? (As
in, is there something that I have missed?)

I'm considering fixing this by changing the IMAGE_COMMAND_cpio function in
image_typyes.bbclass to generate a cpio archive with only the contents
described in IMAGE_DEIVCE_TABLE(S), then use the cpio utility's append mode
to add in the rootfs contents. (It is simpler to create a new archive with
a script than it is to append to an existing one, so I thought I'd let the
cpio utility handle the archive appending.)
Does this sound like a reasonable approach, and the correct place to fix it?

[-- Attachment #2: Type: text/html, Size: 2454 bytes --]

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Issues creating /dev/ nodes in an initramfs
  2014-10-19 18:09 Issues creating /dev/ nodes in an initramfs Mitchell Skiba
@ 2014-10-22  8:05 ` ChenQi
  0 siblings, 0 replies; 2+ messages in thread
From: ChenQi @ 2014-10-22  8:05 UTC (permalink / raw)
  To: yocto

[-- Attachment #1: Type: text/plain, Size: 1915 bytes --]

On 10/20/2014 02:09 AM, Mitchell Skiba wrote:
> I've been building a kernel with an initramfs, by using 
> INITRAMFS_IMAGE to specify a recipe that generates an image as a 
> .cpio.gz archive. I've been having issues creating /dev/console (or 
> any other device file) in the image. (The 3.14 kernel panics if 
> /dev/console is not a character device.)
>
> It seems there is already a way to create dev nodes via the 
> IMAGE_DEVICE_TABLE and IMAGE_DEVICE_TABLES variables. However this 
> mechanism, through the makedevs utility, appears to require the 
> building user to have permission to call mknod. (Makedevs ignores the 
> return code from the mknod call, so it doesn't outright fail on the 
> permissions error, but device files will not be created.)
>
> I've hacked around it in my image recipe by appending a step onto 
> do_rootfs, but I'm not too happy with the way I did it. It seems like 
> the proper way to fix this is to change the image commands in 
> image_types.bbclass. However, each image type would need a different 
> way of adding in the special files. (At this point I only care about 
> cpio archives for the initramfs.)
>
>
> Does anyone have the current IMAGE_DEVICE_TABLE implementation 
> working? (As in, is there something that I have missed?)
>
> I'm considering fixing this by changing the IMAGE_COMMAND_cpio 
> function in image_typyes.bbclass to generate a cpio archive with only 
> the contents described in IMAGE_DEIVCE_TABLE(S), then use the cpio 
> utility's append mode to add in the rootfs contents. (It is simpler to 
> create a new archive with a script than it is to append to an existing 
> one, so I thought I'd let the cpio utility handle the archive appending.)
> Does this sound like a reasonable approach, and the correct place to 
> fix it?
>
>

How about setting USE_DEVFS to "0" in your image recipe?

Regards,
Chen Qi

[-- Attachment #2: Type: text/html, Size: 3656 bytes --]

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2014-10-22  8:03 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-10-19 18:09 Issues creating /dev/ nodes in an initramfs Mitchell Skiba
2014-10-22  8:05 ` ChenQi

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.