howto? Build something but not deploy in root image
I know, this has come up before. Reading through some of the old messages I see this discussed, but, I do not see any solution(s) mentioned?
I have several parts to this project that must be built during the bitbake process (bootstrap-image), but are not part of the target filesystem image. The total system parts are:
* bootloader - ARM - resides in Flash.
* kernel - ARM - resides in Flash.
* target rootfs image - ARM - resides on MMC card.
* Flash Utility - native (x86) - runs on host computer.
The concept for the system is that the bootloader starts the RAM and other hardware up. It then copies the kernel image into RAM and launches it. The kernel is configured to look for its root filesystem on the MMC card. Essentially, I boot the kernel from Flash, but the kernel boots from the MMC card.
The only thing that I need in the root filesystem (tmp/rootfs/) is the packages I named from the local conf files. The kernel does need to be there, it soaks up space uselessly.
As to the bootloader, it comes in two sections (files), these are programmed by the Flash Utility into the target system. The bootloader is not needed in the root filesystem.
The problem is not getting these things built, but preventing them from being installed in the tmp/rootfs/ filesystem or tarball. They should only go into tmp/deploy/image/, which I currently have them going into... However, the bitbake system assumes that anything that is packaged, must be deployed into the root filesystem!?
How do you build a bootloader that resides in Flash and doesn't get stuck into the filesystem image? The same would hold true for the native (linux PC) app which flashes the target memory.
Regards,
TomW
Signature · 4 lines
Tom Walsh - WN3L - Embedded Systems Consultant http://openhardware.net http://cyberiansoftware.com http://openzipit.org "Windows? No thanks, I have work to do..." ----------------------------------------------------
Quoted 26 lines · Tom Walsh
I know, this has come up before. Reading through some of the old messages I see this discussed, but, I do not see any solution(s) mentioned?
I have several parts to this project that must be built during the bitbake process (bootstrap-image), but are not part of the target filesystem image. The total system parts are:
* bootloader - ARM - resides in Flash.
* kernel - ARM - resides in Flash.
* target rootfs image - ARM - resides on MMC card.
* Flash Utility - native (x86) - runs on host computer.The concept for the system is that the bootloader starts the RAM and other hardware up. It then copies the kernel image into RAM and launches it. The kernel is configured to look for its root filesystem on the MMC card. Essentially, I boot the kernel from Flash, but the kernel boots from the MMC card.
The only thing that I need in the root filesystem (tmp/rootfs/) is the packages I named from the local conf files. The kernel does need to be there, it soaks up space uselessly.
As to the bootloader, it comes in two sections (files), these are programmed by the Flash Utility into the target system. The bootloader is not needed in the root filesystem.
Quoted 5 lines
The problem is not getting these things built, but preventing them from being installed in the tmp/rootfs/ filesystem or tarball. They should only go into tmp/deploy/image/, which I currently have them going into... However, the bitbake system assumes that anything that is packaged, must be deployed into the root filesystem!?
This isn't quite true. Bitbake installs whatever packages are in the RDEPENDS of the image. I think what you would do it make sure that your bootloader, etc are in the DEPENDS but not the RDEPENDS.
Quoted 5 lines
How do you build a bootloader that resides in Flash and doesn't get stuck into the filesystem image? The same would hold true for the native (linux PC) app which flashes the target memory.
Signature · 1 line
Justin Patrin
Hello Tom,
Quoted 2 lines · you
I know, this has come up before. Reading through some of the old messages I see this discussed, but, I do not see any solution(s) mentioned?
Well, until recently, OE had issues with building truly minimal and
well-polished images. This yet needs to be described in howto-like manner, that's on my todo.
Quoted 9 lines
I have several parts to this project that must be built during the bitbake process (bootstrap-image), but are not part of the target filesystem image. The total system parts are:
* bootloader - ARM - resides in Flash.
* kernel - ARM - resides in Flash.
* target rootfs image - ARM - resides on MMC card.
* Flash Utility - native (x86) - runs on host computer.
Quoted 8 lines
The concept for the system is that the bootloader starts the RAM and other hardware up. It then copies the kernel image into RAM and launches it. The kernel is configured to look for its root filesystem on the MMC card. Essentially, I boot the kernel from Flash, but the kernel boots from the MMC card.
The only thing that I need in the root filesystem (tmp/rootfs/) is the packages I named from the local conf files.
This is one feature which is possible now. You name exact and
complete list of packages you want to install into image, and nothing else gets installed, not even dependencies. This is of course highly special-purpose feature, and you don't want to use it on realistic rootfs with dozens of packages, it is indtended for initrds/initramfs, which consist of few packages and where each KB counts.
Quoted 2 lines
The kernel does need to be
there, it soaks up space uselessly.
Kernel is indeed special case, and there was some kind of adhoc
support for doing what you want all the time in OE. Couple of months ago Richard Purdie implemented more scalable support for this though, so search archives for his RFC.
Quoted 3 lines
As to the bootloader, it comes in two sections (files), these are programmed by the Flash Utility into the target system. The bootloader is not needed in the root filesystem.
Besides two special cases above, when you really need to subvert
OE's dependency handling, for all other cases you just need to get it right. The only reason why bootloader is installed into image is that something RDEPENDS on it (or it listed as top-level package to install ;-) ). So, fix your dependencies, and it will be ok.
Quoted 13 lines
The problem is not getting these things built, but preventing them from being installed in the tmp/rootfs/ filesystem or tarball. They should only go into tmp/deploy/image/, which I currently have them going into... However, the bitbake system assumes that anything that is packaged, must be deployed into the root filesystem!?
How do you build a bootloader that resides in Flash and doesn't get stuck into the filesystem image? The same would hold true for the native (linux PC) app which flashes the target memory.
Regards,
TomW
Signature · 2 lines
Best regards, Paul mailto:[email protected]
Quoted 8 lines · Tom Walsh
I have several parts to this project that must be built during the bitbake process (bootstrap-image), but are not part of the target filesystem image. The total system parts are:
* bootloader - ARM - resides in Flash.
* kernel - ARM - resides in Flash.
* target rootfs image - ARM - resides on MMC card.
* Flash Utility - native (x86) - runs on host computer.
In machine config add:
IMAGEDEPENDS = "bootloader flash-utility-native"
Basically it should be MACHINE_EXTRA_DEPENDS but we do not support it yet properly.
Quoted 2 lines
The only thing that I need in the root filesystem (tmp/rootfs/) is the packages I named from the local conf files.
Create own image recipe and add those packages to RDEPENDS and PACKAGE_INSTALL.
Quoted 8 lines
As to the bootloader, it comes in two sections (files), these are programmed by the Flash Utility into the target system. The bootloader is not needed in the root filesystem.
The problem is not getting these things built, but preventing them from being installed in the tmp/rootfs/ filesystem or tarball. They should only go into tmp/deploy/image/, which I currently have them going into...
Look at uboot/u-boot* recipes and their do_deploy() tasks which copy bootloader into tmp/deploy/images/ dir.
Quoted 2 lines
However, the bitbake system assumes that anything that is packaged, must be deployed into the root filesystem!?
No. Image is built from packages, but from selected ones only.
Quoted 3 lines
How do you build a bootloader that resides in Flash and doesn't get stuck into the filesystem image? The same would hold true for the native (linux PC) app which flashes the target memory.
In their recipes add do_deploy() task which copies them to tmp/deploy/images/ dir.
Signature · 5 lines
JID: hrw-jabber.org
OpenEmbedded developer/consultant
Free speech includes the right not to listen, if not interested
[Robert A. Heinlein]Quoted 17 lines · Marcin Juszkiewicz
Quoted 9 lines · Tom Walsh
I have several parts to this project that must be built during the bitbake process (bootstrap-image), but are not part of the target filesystem image. The total system parts are:
* bootloader - ARM - resides in Flash.
* kernel - ARM - resides in Flash.
* target rootfs image - ARM - resides on MMC card.
* Flash Utility - native (x86) - runs on host computer.In machine config add:
IMAGEDEPENDS = "bootloader flash-utility-native"
Quoted 10 lines
Basically it should be
MACHINE_EXTRA_DEPENDSbut we do not support it yet properly.Quoted 3 lines
The only thing that I need in the root filesystem (tmp/rootfs/) is the packages I named from the local conf files.
Create own image recipe and add those packages to RDEPENDS and
PACKAGE_INSTALL.
Yes, that is what I was trying to avoid: building a custom image recipe. It just seemed that it should be possible to thwart the copy of the kernel image into /boot from setting a variable or something. The OE system changes quite frequently and I found myself in trouble with my own image recipe where I was "away" for several months and "returned".
Adding a DEPENDS for the bootloader and utils within the kernel recipe did resolve the issue of having the image recipe attempting to install those two packages into the IMAGE. I still have a 1.3Meg kernel in the /boot directory, though, I guess half a loaf is better than none? At least OE now builds the entire project without stopping along the way to complain about "iconv.h" and other such things.
Regards,
TomW
Signature · 4 lines
Tom Walsh - WN3L - Embedded Systems Consultant http://openhardware.net http://cyberiansoftware.com http://openzipit.org "Windows? No thanks, I have work to do..." ----------------------------------------------------
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Tom Walsh schreef:
Quoted 5 lines
Yes, that is what I was trying to avoid: building a custom image recipe. It just seemed that it should be possible to thwart the copy of the kernel image into /boot from setting a variable or something. The OE system changes quite frequently and I found myself in trouble with my own image recipe where I was "away" for several months and "returned".
You are highlighting the exact reason why not submitting your work for inclusion into OE is bad.
regards,
Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
iD8DBQFGmiITMkyGM64RGpERAok3AJ9oXLOIVJaL1E9S/Dj3RousCPNY7QCdHHL1 wtaaFQRvQ8pwr9XCTLmdgR4=
=eKIP
-----END PGP SIGNATURE-----
Am 15.07.2007 um 15:33 schrieb Koen Kooi:
Quoted 5 lines
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Tom Walsh schreef:
Quoted 8 lines
Quoted 5 lines
Yes, that is what I was trying to avoid: building a custom image recipe. It just seemed that it should be possible to thwart the copy of the kernel image into /boot from setting a variable or something. The OE system changes quite frequently and I found myself in trouble with my own image recipe where I was "away" for several months and "returned".
You are highlighting the exact reason why not submitting your work for inclusion into OE is bad.
And that a meaningful ChangeLog could help as well.
z.