| Age | Commit message (Collapse) | Author |
|---|
|
stat command from stat package doesn't support '-L' option, so avoid
using it to get rid of boot errors like below.
Populating dev cache
stat: invalid option -- 'L'
Usage: stat [-l] [-f] [-s] [-v] [-h] [-t] [-c format] file1 [file2 ...]
[YOCTO #7210]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
|
|
Most of /etc/init.d/udev-cache is in a conditional block which can be
replaced by a `[ ... ] || exit 0` to reduce nesting.
This also provides an opportunity to add some additional messages
when VERBOSE is set.
Capture and report errors encountered in the cache generation process,
using set -e and trap EXIT. These errors were previously being ignored.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
Don't hold up the boot while the cache is being updated.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Ben Shelton <ben.shelton@ni.com>
Acked-by: Gratian Crisan <gratian.crisan@ni.com>
|
|
Presently, the cache is not regenerated if udev rules are modified,
which may cause the cache to preserve an old configuration. To fix,
include the size, mtime, and filename of all udev rules in the system
configuration.
This change requires `stat`. If busybox supplies stat,
CONFIG_FEATURE_STAT_FORMAT must be enabled.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
The current system configuration needs to be generated both inside
udev (to compare against the cached system configuration) and
udev-cache (to regenerate the cached system configuration). Use a single
function definition for this task, duplicated across both initscripts.
This also allows administrators to modify it for machine-specific
requirements.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
The udev initscript signals udev-cache to run by generating a new
sysconf; but udev-cache now overwrites that with its own copy. To
eliminate the needless sysconf generating in udev, we instead trigger
udev-cache to run by touching a new file $DEVCACHE_REGEN.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
Currently, udev-cache system configurations are compared as shell string
variables, read into memory with the readfiles() function. This is more
complex, and significantly (27-41%) slower, than comparing them using
`cmp`. (Performance was verified on both Cortex-A9 and Intel Nehalem
systems.)
So just use cmp. This requires a few other small changes:
exclude /proc/atags from CMP_FILE_LIST if it doesn't exist to avoid
errors in `cat` and `cmp`.
`cmp -q` doesn't exist in busybox, so instead, redirect output to
/dev/null.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
The validity of the udev cache is determined by the sysconf
file (/etc/udev/cache.data). Currently, there is a substantial delay
between sysconf generation in /etc/init.d/udev and cache generation in
/etc/init.d/udev-cache. If a hotplug event arrives in the middle of
this, then the sysconf will be out of date with respect to the cache.
The solution is two-pronged. First, we minimize the race window by
regenerating the sysconf immediately before the cache, in
/etc/init.d/udev-cache. This allows us to kill the race entirely by
stopping the udev event queue while the sysconf and cache are being
generated.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
|
|
The udev-cache facility uses files that represent system states, to
ensure that the cache tarball is valid to apply. These paths were
hardcoded in several places; collect them into SYSCONF_CACHED and
SYSCONF_TMP.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Archiving sockets causes tar to report an error and return a nonzero
exit status. Archiving a mounted filesystem is harmless, but may greatly
bloat the size of the cache tarball, and wastes time on boot.
To fix these issues, use `find` to only include the files we want, which
are the file types that udev will create (block/char devices and
symlinks) that are on the same filesystem as /dev.
While we're at it, remove a subshell by archiving /dev as an absolute
path. However, `tar` will complain about stripping the leading slash on
stderr. To inhibit this, `cut` out the leading slash.
An alternative solution is to use `tar --exclude`, but that is modestly
more brittle, since we'd need to explicitly list every socket and
filesystem to exclude. Note that `tar --one-file-system` is
GNU-specific, and tar implementations generally have nothing equivalent
to `find -type`.
If using busybox `find`, this change requires CONFIG_FEATURE_FIND_TYPE=y
and CONFIG_FEATURE_FIND_XDEV=y. If using busybox `tar`, this change
requires CONFIG_FEATURE_TAR_FROM=y.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
$DEVCACHE is observed to be 100k uncompressed; compressing it reduces
its size to ~5k. But compress it outside of `tar` so that archival
operation takes as little time as possible, to minimize the risk of
devices being created/removed during execution.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Overwriting the tarball in-place could cause a partial write, if the
system stops at an inopportune time. This is mitigated by first writing
to a temporary file, then moving that file on top of the final location.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Ben Shelton <ben.shelton@ni.com>
Acked-by: Gratian Crisan <gratian.crisan@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In case of a read-only rootfs, we skip the process of generating
udev cache, as the data cannot be persisted between reboots.
However, it's possbile that the $DEVCACHE (default to /etc/dev.tar)
exists in a read-only rootfs, no matter how it's generated or installed.
In such situation, we try to use $DEVCACHE if possible.
Besides the basic changes in the logic of udev cache handling,
this patch also adds code to output more information if the udev
cache is not used and VERBOSE enabled.
This patch also changes the readfile function to readfiles function
so that it could handle more than one file at once.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Previouly, the udev-cache has no real effect even if it's installed
into the system. The key problem here is that at first boot, the
/etc/dev.tar is not present, thus resulting /dev/shm/udev.cache not
created on first boot even if udev-cache is enabled.
This patch fixes this problem. The /dev/shm/udev.cache will be created
if necessary, that is, on first boot or when some part of the system is
changed. In the latter case, the udev cache may not be valid.
[YOCTO #4738]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Ensure we can update the script base don the location of the udevd installation
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This is the final upgrade of udev. Futher upgrades will only
come in conjunction with systemd.
The v4l1 removal patch is deprecated as the bug is fixed inside udev.
There is a new patch fixing the path for default sh interpreter.
New debug binaries are generated, and udev.inc is modified to package
those correctly.
The install locations changed for udevd and udevadm, so the scripts
are updated accordingly.
(From OE-Core rev: 3cbe52b94c4d559a037347ac419fafee5af84fe6)
Signed-off-by: Alexandru DAMIAN <alexandru.damian@intel.com>
Conflicts:
meta/recipes-core/udev/udev_164.bb
sgw - Fixed up DEPENDS += and added some OECONF options that where in the
meta-oe version and make sense to be included.
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* allow udev-cache to be disabled at runtime (using
/etc/default/udev-cache);
* make cache invalidated if kernel, bootparams or device list
changes;
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
|
|
We merged both files and udev-164 directory onto a single directory
and renamed udev-new.inc onto udev.inc as we have a single version of
udev now.
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
|
Chen Qi
Richard Tollerton
Saul Wold
Alexandru DAMIAN
Otavio Salvador