[meta-freescale] Large rootfs corrupted by build system

[George Hilliard]

I have run into a very odd issue when building a custom Yocto "dora" image for the i.MX6 SABRE Auto.  When building a small image (final image is around 80MB), the image is generated just fine.  However, when I build a much larger image (around 2GB), the root filesystem is generated corrupt!  This is not a fluke; every time I regenerate the image, it is corrupt (according to fsck).

I worked around it by actually removing packages (git and kernel-dev) from the recipe, causing the image to fall to about 1.91GB.  After this, the rootfs was generated correctly.  This doesn't look like anything but a bug in the build system to me; maybe the 2GB threshold is significant (?).  I'm currently building fsl-image-test to see if the problem occurs with that image too.

Am I doing everything right?  Attached are my two build scripts.  The -dev image is the bigger one.

Thanks!
George Hilliard
-------------- next part --------------
require custom-image-production.bb

# Add extra image features; this is like Debian's package groups
IMAGE_FEATURES += " \
    dev-pkgs \
    dbg-pkgs \
    tools-sdk \
    tools-debug \
    tools-testapps \
    ssh-server-dropbear \
    package-management \
    debug-tweaks \
"

# Add individual recipes
IMAGE_INSTALL += " \
    python-scons \
    kernel-dev \
    jsoncpp \
    libsocket \
    git \
"

export IMAGE_BASENAME = "custom-image-production-dev"
-------------- next part --------------
include recipes-core/images/core-image-base.bb

# Add extra image features; this is like Debian's package groups
IMAGE_FEATURES += " \
"

# Add individual recipes
IMAGE_INSTALL += " \
    canutils \
    iproute2 \
    hostap-daemon \
    dnsmasq \
"

LICENSE = "CLOSED"

export IMAGE_BASENAME = "custom-image-production"