Chapter 2 Known Issues
This chapter describes the known issues for the Unbreakable Enterprise Kernel Release 3.
Updating Oracle Linux Fails if the kernel-uek Package Cannot Be Updated
By default, the installation of Oracle Linux includes the
dtrace-modules
package for UEK R3. This
package requires a specific kernel-uek
version. However, a yum update fails if it
cannot update the kernel-uek
package when the
installonly_limit
of three updates is reached
as the dtrace-modules
package does not allow
the existing kernel-uek
packages to be
removed. yum displays an error message
similar to the following example:
--> Finished Dependency Resolution Error: Package: kernel-uek-debug-3.8.13-55.1.1.el6uek.x86_64 (public_ol6_UEKR3_latest) Requires: kernel-firmware = 3.8.13-55.1.1.el6uek
The workaround is to remove any existing
dtrace-modules
packages before updating the
kernel-uek
package, for example:
# for package in `rpm -qa | grep dtrace-modules`; do yum remove -y $package; done
When you have removed all dtrace-modules
packages, you can update Oracle Linux, including the UEK R3
kernel. If you want to use DTrace with UEK R3, reinstall the
dtrace-modules
package for the current
kernel:
# yum install dtrace-modules-`uname -r`
(Bug ID 21669543)
ACFS
Oracle ASM Cluster File System (ACFS) is currently not supported for use with UEK R3. (Bug ID 16318126)
ACPI
On some systems you might see ACPI-related error messages in
dmesg
similar to the following:ACPI Error: [CDW1] Namespace lookup failure, AE_NOT_FOUND ACPI Error: Method parse/execution failed [_SB_._OSC||\||] ACPI Error: Field [CDW3] at 96 exceeds Buffer [NULL] size 64 (bits)]]>
These messages, which are not fatal, are caused by bugs in the BIOS. Contact your system vendor for a BIOS update. (Bug ID 13100702)
The following messages indicate that the BIOS does not present a suitable interface, such as
_PSS
or_PPC
, that theacpi-cpufreq
module requires:kernel: powernow-k8: this CPU is not supported anymore, using acpi-cpufreq instead. modprobe: FATAL: Error inserting acpi_cpufreq
There is no known workaround for this error. (Bug ID 17034535)
ASM
Calling the oracleasm
init
script, /etc/init.d/oracleasm
, with the
parameter scandisks
can lead to error
messages about missing devices similar to the following:
oracleasm-read-label: Unable to open device "device
": No such file or directory
However, the device actually exists. You can ignore this error
message, which is triggered by a timing issue. Only use the
init
script to start and stop the
oracleasm
service. All other options, such as
scandisks, listdisk
, and
createdisk
, are deprecated. For these and
other administrative tasks, use
/usr/sbin/oracleasm
instead. (Bug ID
13639337)
bnx2x driver
When using the bnx2x
driver in a bridge,
disable Transparent Packet Aggregation (TPA) by including the
statement options bnx2x disable_tpa=1
in
/etc/modprobe.conf
. (Bug ID 14626070)
btrfs
If you use the --alloc-start option with mkfs.btrfs to specify an offset for the start of the file system, the size of the file system should be smaller but this is not the case. It is also possible to specify an offset that is higher than the device size. (Bug ID 16946255)
The usage information for mkfs.btrfs reports
raid5
andraid6
as possible profiles for both data and metadata. However, the kernel does not support these features and cannot mount file systems that use them. (Bug ID 16946303)The btrfs filesystem balance command does not warn that the RAID level can be changed under certain circumstances, and does not provide the choice of cancelling the operation. (Bug ID 16472824)
Converting an existing ext2, ext3, or ext4 root file system to btrfs does not carry over the associated security contexts that are stored as part of a file's extended attributes. With SELinux enabled and set to enforcing mode, you might experience many permission denied errors after reboot, and the system might be unbootable. To avoid this problem, enforce automatic file system relabeling to run at bootup time. To trigger automatic relabeling, create an empty file named
.autorelabel
(for example, by using touch) in the file system'sroot
directory before rebooting the system after the initial conversion. The presence of this file instruct SELinux to recreate the security attributes for all files on the file system. If you forget to do this and rebooting fails, either temporarily disable SELinux completely by addingselinux=0
to the kernel boot parameters, or disable enforcing of the SELinux policy by addingenforcing=0
. (Bug ID 13806043)Commands such as du can show inconsistent results for file sizes in a btrfs file system when the number of bytes that is under delayed allocation is changing. (Bug ID 13096268)
The copy-on-write nature of btrfs means that every operation on the file system initially requires disk space. It is possible that you cannot execute any operation on a disk that has no space left; even removing a file might not be possible. The workaround is to run sync before retrying the operation. If this does not help, remount the file system with the -o nodatacow option and delete some files to free up space. See https://btrfs.wiki.kernel.org/index.php/ENOSPC.
Btrfs has a limit of 237 or fewer hard links to a file from a single directory. The exact limit depends on the number of characters in the file name. The limit is 237 for a file with up to eight characters in its file name; the limit is lower for longer file names. Attempting to create more than this number of links results in the error
Too many links
. You can create more hard links to the same file from another directory. Although the limitation of the number of hard links in a single directory has been increased to 65535, the version of mkfs.btrfs that is provided in thebtrfs-progs
package does not yet support the compatibility flag for this feature. (Bug ID 16278563)The -c option to the btrfs qgroup limit command is redundant as the quota limit is always enforced after compression. (Bug ID 16557528)
If you run the btrfs quota enable command on a non-empty file system, any existing files do not count toward space usage. Removing these files can cause usage reports to display negative numbers and the file system to be inaccessible. The workaround is to enable quotas immediately after creating the file system. If you have already written data to the file system, it is too late to enable quotas. (Bug ID 16569350)
The btrfs quota rescan command is not currently implemented. The command does not perform a rescan and returns without displaying any message. (Bug ID 16569350)
When you overwrite data in a file, starting somewhere in the middle of the file, the overwritten space is counted twice in the space usage numbers that btrfs qgroup show displays. (Bug ID 16609467)
If you run btrfsck --init-csum-tree on a file system and then run a simple btrfsck on the same file system, the command displays a Backref mismatch error that was not previously present. (Bug ID 16972799)
Btrfs tracks the devices on which you create btrfs file systems. If you subsequently reuse these devices in a file system other than btrfs, you might see error messages such as the following when performing a device scan or creating a RAID-1 file system, for example:
ERROR: device scan failed '/dev/cciss/c0d0p1' - Invalid argument
You can safely ignore these errors. (Bug ID 17087097)
If you use the -s option to specify a sector size to mkfs.btrfs that is different from the page size, the created file system cannot be mounted. By default, the sector size is set to be the same as the page size. (Bug ID 17087232)
CPU microcode update failures on PVM or PVHVM guests
When running Oracle Linux 6 with UEK R3, you might see error
messages in dmesg
or
/var/log/messages
similar to this one:
microcode: CPU0 update to revision 0x6b failed.
You can ignore this warning. You do not need to upgrade the microcode for virtual CPUs as presented to the guest. (Bug ID 12576264, 13782843)
DHCP lease is not obtained at boot time
If DHCP lease negotiation takes more than 5 seconds at boot time, the following message is displayed:
ethX
: failed. No link present. Check cable?
If the ethtool
ethX
command confirms that
the interface is present, edit
/etc/sysconfig/network-scripts/ifcfg-eth
and set
X
LINKDELAY=
, where
N
N
is a value greater than 5 seconds
(for example, 30 seconds). Alternatively, use NetworkManager to
configure the interface. (Bug ID 16620177)
dm-nfs obsoleted
In UEK R2, the dm-nfs
module provided the
ability to create a loopback device for a mounted NFS file or
file system. For example, the feature allowed you to create the
shared storage for an Oracle 3 VM cluster on an NFS file system.
The dm-nfs
module provided direct I/O to the
server and bypassed the loop
driver to avoid
an additional level of page caching. The
dm-nfs
module is not provided with UEK R3.
The loop
driver can now provide the same I/O
functionality as dm-nfs
by extending the AIO
interface to perform direct I/O. To create the loopback device,
use the losetup command instead of
dmsetup.
DTrace
Using kill -9 to terminate dtrace can leave breakpoints outstanding in processes being traced, which might sooner or later kill them.
Argument declarations for probe definitions cannot be declared with derived types such as
enum
,struct
, orunion
.The following compiler warning can be ignored for probe definition arguments of type
string
(which is a D type but not a C type):provider_def
.h:line#
: warning: parameter names (without types) in function declaration
ERST message
You can safely ignore the following message that might be
displayed in syslog
or
dmesg
:
ERST: Failed to get Error Log Address Range.
The message indicates that the system BIOS does not support an Error Record Serialization Table (ERST). (Bug ID 17034576)
ext4 inline data
The inline data feature that allows the data of small files to be stored inside their inodes is not yet available. The -O inline_data option to the mkfs.ext4 and tune2fs commands is not supported. (Bug ID 17210654)
Firmware warning message
You can safely ignore the following firmware warning message that might be displayed on some Sun hardware:
[Firmware Warn]: GHES: Poll interval is 0 for generic hardware error source: 1, disabled.
(Bug ID 13696512)
Huge pages
One-gigabyte (1 GB) huge pages are not currently supported for the following configurations:
HVM guests
PV guests
Oracle Database
Two-megabyte (2 MB) huge pages have been tested and work with these configurations.
(Bug ID 17299364, 17299871, 17271305)
I/O scheduler
The Unbreakable Enterprise Kernel uses the
deadline
scheduler as the default I/O
scheduler. For the Red Hat Compatible Kernel, the default I/O
scheduler is the cfq
scheduler.
ioapic failure messages
You can safely ignore messages such as ioapic: probe of
0000:00:05.4 failed with error -22
. Such messages are
the result of the ioapic
driver attempting to
re-register I/O APIC PCI devices that were already registered at
boot time. (Bug ID 17034993)
InfiniBand warning messages when disabling a switch port
You might see the following warning messages if you use the ibportstate disable command to disable a switch port:
ibwarn: [2696] _do_madrpc: recv failed: Connection timed out ibwarn: [2696] mad_rpc: _do_madrpc failed; dport (Lid 38) ibportstate: iberror: failed: smp set portinfo failed
You can safely ignore these warnings. (Bug ID 16248314)
IPoIB mode switching
The Internet Protocol over InfiniBand (IPoIB) driver supports
the use of either connected mode or datagram mode with an
interface, where datagram mode is the default mode. Changing the
mode of an InfiniBand interface by echoing either
connected
or datagram
to
/sys/class/net/ib
is not supported. It is also not possible to change the mode of
an InfiniBand interface while it is enabled.
N
/mode
To change the IPoIB mode of an InfiniBand interface:
Edit the
/etc/sysconfig/network-scripts/ifcfg-ib
configuration file, whereN
N
is the number of the interface:To configure connected mode, specify
CONNECTED_MODE=yes
in the file.To configure datagram mode, either specify
CONNECTED_MODE=no
in the file or do not specify this setting at all (datagram mode is enabled by default).
NoteBefore saving your changes, make sure that you have not specified more than one setting for
CONNECTED_MODE
in the file.To enable the specified mode on the interface, use the following commands to take down the interface and bring it back up:
#
ifdown ib
#N
ifup ib
N
(Bug ID 17479833)
libfprint
The following message might appear in dmesg
or /var/log/messages
:
WARNING! power/level is deprecated; use power/control instead.
The USB subsystem in UEK R3 deprecates the power/level
sysfs
attribute in favor of the
power/control
attribute. The
libfprint
fingerprinting library triggers
this warning via udev
rules that try to use
the old attribute first. You can safely ignore this warning. The
setting of the appropriate power level still succeeds. (Bug ID
13523418)
Large memory system fails to boot
If a large memory system fails to start, boot it using an alternate kernel to UEK R3 and disable the kdump service before booting into the UEK R3 kernel:
# chkconfig kdump off
(Bug ID 16765434)
Linux Containers (LXC)
The correct operation of containers might require that you completely disable SELinux on the host system. For example, SELinux can interfere with container operation under the following conditions:
Running the halt or shutdown command from inside the container hangs the container or results in a
permission denied
error. (An alternate workaround is to use the init 0 command from inside the container to shut it down.)Setting a password inside the container results in a
permission denied
error, even when run asroot
.You want to allow ssh logins to the container.
To disable SELinux on the host:
Edit the configuration file for SELinux,
/etc/selinux/config
and set the value of theSELINUX
directive todisabled
.Shut down and reboot the host system.
The
root
user in a container can affect the configuration of the host system by setting some/proc
entries. (Bug ID 17190287)Using yum to update packages inside the container that use
init
scripts can undo changes made by the Oracle template.Migrating live containers (lxc-checkpoint) is not yet supported.
Oracle Database is not yet supported for use with Linux Containers. The following information is intended for those who want to experiment with such a configuration.
The following
/proc
parameter files may only be set on the host and not for individual containers:/proc/sys/fs/aio-max-nr
/proc/sys/net/core/rmem_default
/proc/sys/net/core/rmem_max
/proc/sys/net/core/wmem_default
/proc/sys/net/core/wmem_max
/proc/sys/net/ipv4/ip_local_port_range
Setting the parameters in the host to the Oracle recommended values sets them for all containers and allows the Oracle database to run in a container. For more information, see Configuring Kernel Parameters. (Bug ID 17217854)
NUMA warning messages on a non-NUMA system
You can safely ignore the following warning messages in
dmesg
and /var/log
messages
if you see them on a non-NUMA system:
kernel: NUMA: Warning: node ids are out of bound, from=-1 to=-1 distance=10 hcid[4293]: Register path:/org/bluez fallback:1 kernel: No NUMA configuration found
(Bug ID 13711370)
pcspkr driver error message
You can safely ignore the following error message:
Error: Driver 'pcspkr' is already registered, aborting...
The message arises from an alias conflict between
snd-pcsp
and pcspkr
. To
prevent the message from being displayed, add the following line
to /etc/modprobe.d/blacklist.conf
:
blacklist snd-pcsp
(Bug ID 10355937)
sched_yield() settings for CFS
For the Unbreakable Enterprise Kernel,
kernel.sched_compat_yield=1
is set by
default. For the Red Hat Compatible Kernel,
kernel.sched_compat_yield=0
is used by
default.
Soft lockup errors when booting
When upgrading or installing the UEK R3 kernel on fast hardware,
usually with SAN storage attached, the kernel can fail to boot
and BUG: soft lockup
messages are displayed
in the console log. The workaround is to increase the baud rate
from the default value of 9600 by amending the kernel boot line
in /boot/grub/grub.conf
to include an
appropriate console setting, for example:
console=ttyS0,115200n8
A value of 115200 is recommended as smaller values such as 19200 are known to be insufficient for some systems (for example, see https://docs.oracle.com/cd/E19045-01/blade.x6220/820-0048-18/sp.html#0_pgfId-1002490). If the host implements an integrated system management infrastructure, such as ILOM on Sun and Oracle systems or iLO on HP systems, configure the integrated console baud rate to match the setting for the host system. Otherwise, the integrated console is likely to display garbage characters. (Bug ID 17064059, 17252160)
Transparent Huge Pages
This release removes the Transparent Huge Pages (THP) feature. Following extensive benchmarking and testing, Oracle found that THP caused a performance degradation of between 5 and 10% for some workloads. This performance degradation was a result of a slower memory allocator code path being used even when the applications were not using THP. When the fact that huge pages are not swappable was taken into account, the positive effect that THP should provide was outweighed by its negative effects.
After installing this UEK release, you cannot enable THP (for
example, by specifying kernel boot parameters). The THP settings
under /sys/kernel/mm/transparent_hugepage
have also been removed. A future update might contain an updated
THP implementation which resolves the performance issue.
This change does not affect support for applications that use explicit huge pages (for example, Oracle Database).
(Bug ID 16823432)
User Namespaces
The kernel functionality (CONFIG_USER_NS
)
that allows unprivileged processes to create namespaces for
users inside which they have root privileges is not currently
implemented because of a clash with the implementation of XFS.
This functionality is primarily intended for use with Linux
Containers. As a result, the lxc-checkconfig
command displays User namespace: missing
.
(Bug ID 16656850)
Virtualization
When booting UEK R3 as a PVHVM guest, you can safely ignore the following kernel message:
register_vcpu_info failed: err=-38
(Bug ID 13713774)
Under Oracle VM Server 3.1.1, migrating a PVHVM guest that is running the UEK R3 kernel causes a disparity between the date and time as displayed by date and hwclock. The workaround post migration is either to run the command hwclock --hctosys on the guest or to reboot the guest. (Bug ID 16861041)
On virtualized systems that are built on Xen version 3, including all releases of Oracle VM 2 including 2.2.2 and 2.2.3, disk synchronization requests for ext3 and ext4 file systems result in journal corruption with kernel messages similar to the following being logged:
blkfront: barrier: empty write xvda op failed blkfront: xvda: barrier or flush: disabled
In addition, journal failures such as the following might be reported:
Aborting journal on device xvda1
The workaround is to add the mount option barrier=0 to all ext3 and ext4 file systems in the guest VM before upgrading to UEK R3. For example, you would change a mount entry such as:
UUID=4e4287b1-87dc-47a8-b69a-075c7579eaf1 / ext3 defaults 1 1
so that it reads:
UUID=4e4287b1-87dc-47a8-b69a-075c7579eaf1 / ext3 defaults,barrier=0 1 1
This issue does not apply to Xen 4 based systems, such as Oracle VM 3. (Bug ID 17310816)
X.509 Certificates for module verification
The system reports a message similar to the following if there is a problem loading an in-kernel X.509 module verification certificate at boot time:
Loading module verification certificates X.509: Cert 0c21da3d73dcdbaffc799e3d26f3c846a3afdc43 is not yet valid MODSIGN: Problem loading in-kernel X.509 certificate (-129)
This error occurs because the hardware clock lags behind the system time as shown by hwclock, for example:
# hwclock
Tue 20 Aug 2013 01:41:40 PM EDT -0.767004 seconds
The solution is to set the hardware clock from the system time by running the following command:
# hwclock --systohc
After correcting the hardware clock, no error should be seen at boot time, for example:
Loading module verification certificates MODSIGN: Loaded cert 'Slarti: Josteldalsbreen signing key: 0c21da3d73dcdbaffc799e3d26f3c846a3afdc43'
(Bug ID 17346862)