When Single Root I/O Virtualization (SR-IOV) is enabled on mlx4 adapters, initialization of the
mlx4_core module can take longer than 30 seconds. However,
udev has a 30-second timer for hardware initialization and terminates the modprobe sequence if the time is exceeded. As a consequence, the
mlx4_ib modules are not loaded after the
mlx4_core module has finished initializing.
Two workarounds are available:
To work around the
udev time limit, run the
rmmod mlx4_core; modprobe mlx4_core command on a system that does not run on mlx4 hardware. Running this command enables you to have the full configured number of virtual devices, but for example guest operating systems that are configured to start on boot may need to be started manually once all the devices are present.
Alternatively, reduce either the total number of virtual functions that are enabled on the card, or functions that are probed by the guest driver, or both, until the card initialization sequence can be reliably completed in less than 30 seconds. However, you can be limited in the number of guests you can run due to insufficient virtual functions.
kernel component, BZ#1126294
The modprobe configuration file does not use the full path name of the
modprobe binary file. As a consequence, modprobe fails to load the
ib_qib module. To work around this problem, run the
modprobe ib_qib command after each boot. Alternatively, to fix the problem permanently, edit the
/etc/modprobe.d/truescale.conf and change the instance of modprobe to be /sbin/modprobe. As a result, the
ib_qib module loads as expected.
Sun Fire X4500 data server enumerates the e1000 card with Peripheral Component Interconnect Extended (PCI-X) and enables 64-bit direct memory access (DMA), however, 64-bit DMA is not fully supported on this hardware. If possible, disable 64-bit DMA in BIOS.
Use of multiboot images makes discerning different image types problematic during kernel updates. As a consequence, using the tboot package and multiple types of kernels at the same time does not work properly. If, for example, tboot is in use and the kernel-debug package is installed, bootloader configuration can sometimes reflect an incorrect image list. To avoid this, do not use the kernel-debug on a system utilizing tboot, or vice versa. If such a situation is unavoidable, manually verify that the bootloader configuration is reasonable after each update before rebooting.
When the debug kernel is installed and also used as the Red Hat Enterprise Linux kdump kernel, the reserved kdump memory must be increased to a minimum of 256 MB. To assure this setting, start the system-config-kdump
tool, modify the kdump memory, and reboot your Linux instance. Alternatively, you can configure a particular kernel that is always used as the kdump kernel, independently of the running kernel. For more information, consult the Red Hat Enterprise Linux 6 Deployment Guide
Red Hat Enterprise Linux 6.4 changed the maximum read/write socket memory default value to be higher, allowing for better performance on some machines. It was observed that if the values of
?mem_max are not symmetrical between two machines, the performance can be negatively affected. To work around this problem, adjust the value of
?mem_max to be equal across all Red Hat Enterprise Linux systems in the network.
The vxfs module might not work properly on Red Hat Enterprise Linux 6.4 and later because of the broken
radix_tree_gang_lookup_slot symbol. Consult Symantec should you require a workaround for this issue.
Enabling TCP Segmentation Offload (TSO) on TAP interface may cause low throughput when the uplink is a high-speed interface. To improve throughput, turn off TSO on the tap interface of the virtual machine.
When using Chelsio's iSCSI HBAs for an iSCSI root partition, the first boot after install fails. This occurs because Chelsio's iSCSI HBA is not properly detected. To work around this issue, users must add the
iscsi_firmware parameter to grub's kernel command line. This will signal to dracut to boot from the iSCSI HBA.
The installation of Red Hat Enterprise Linux 6.3 i386 and later may occasionally fail. To work around this issue, add the following parameter to the kernel command line:
If a device reports an error, while it is opened (via the
open(2) system call), then the device is closed (via the
close(2) system call), and the
/dev/disk/by-id link for the device may be removed. When the problem on the device that caused the error is resolved, the
by-id link is not re-created. To work around this issue, run the following command:
echo 'change' > /sys/class/block/sdX/uevent
When an HBA that uses the
mpt2sas driver is connected to a storage using an SAS switch LSI SAS 6160, the driver may become unresponsive during Controller Fail Drive Fail (CFDF) testing. This is due to faulty firmware that is present on the switch. To fix this issue, use a newer version (14.00.00.00 or later) of firmware for the LSI SAS 6160 switch.
kernel component, BZ#745713
In some cases, Red Hat Enterprise Linux 6 guests running fully-virtualized under Red Hat Enterprise Linux 5 experience a time drift or fail to boot. In other cases, drifting may start after migration of the virtual machine to a host with different speed. This is due to limitations in the Red Hat Enterprise Linux 5 Xen hypervisor. To work around this, add the
nohpet parameter or, alternatively, the
clocksource=jiffies parameter to the kernel command line of the guest. Or, if running under Red Hat Enterprise Linux 5.7 or newer, locate the guest configuration file for the guest and add the
hpet=0 parameter in it.
On some systems, Xen full-virt guests may print the following message when booting:
WARNING: BIOS bug: CPU MTRRs don't cover all of memory, losing <number>MB of RAM
It is possible to avoid the memory trimming by using the
disable_mtrr_trim kernel command line option.
On 64-bit PowerPC, the following command may cause kernel panic:
./perf record -agT -e sched:sched_switch -F 100 -- sleep 3
Applications are increasingly using more than 1024 file descriptors. It is not recommended to increase the default soft limit of file descriptors because it may break applications that use the
select() call. However, it is safe to increase the default hard limit; that way, applications requiring a large amount of file descriptors can increase their soft limit without needing root privileges and without any user intervention.
In network only use of Brocade Converged Network Adapters (CNAs), switches that are not properly configured to work with Brocade FCoE functionality can cause a continuous linkup/linkdown condition. This causes continuous messages on the host console:
bfa xxxx:xx:xx.x: Base port (WWN = xx:xx:xx:xx:xx:xx:xx:xx) lost fabric connectivity
To work around this issue, unload the Brocade
In Red Hat Enterprise Linux 6, a legacy bug in the PowerEdge Expandable RAID Controller 5 (PERC5) which causes the kdump kernel to fail to scan for
scsi devices. It is usually triggered when a large amounts of I/O operations are pending on the controller in the first kernel before performing a kdump.
kernel component, BZ#679262
In Red Hat Enterprise Linux 6.2 and later, due to security concerns, addresses in
/proc/modules show all zeros when accessed by a non-root user.
Superfluous information is displayed on the console due to a correctable machine check error occurring. This information can be safely ignored by the user. Machine check error reporting can be disabled by using the
nomce kernel boot option, which disables machine check error reporting, or the
mce=ignore_ce kernel boot option, which disables correctable machine check error reporting.
The order in which PCI devices are scanned may change from one major Red Hat Enterprise Linux release to another. This may result in device names changing, for example, when upgrading from Red Hat Enterprise Linux 5 to 6. You must confirm that a device you refer to during installation, is the intended device.
One way to assure the correctness of device names is to, in some configurations, determine the mapping from the controller name to the controller's PCI address in the older release, and then compare this to the mapping in the newer release, to ensure that the device name is as expected.
The following is an example from /var/log/messages:
kernel: cciss0: <0x3230> at PCI 0000:1f:00.0 IRQ 71 using DAC
kernel: cciss1: <0x3230> at PCI 0000:02:00.0 IRQ 75 using DAC
If the device name is incorrect, add the
pci=bfsort parameter to the kernel command line, and check again.
The minimum firmware version for NIC adapters managed by
netxen_nic is 4.0.550. This includes the boot firmware which is flashed in option ROM on the adapter itself.
High stress on 64-bit IBM POWER series machines prevents kdump from successfully capturing the
vmcore. As a result, the second kernel is not loaded, and the system becomes unresponsive.
Triggering kdump to capture a
vmcore through the network using the Intel 82575EB ethernet device in a 32 bit environment causes the networking driver to not function properly in the kdump kernel, and prevent the
vmcore from being captured.
Memory Type Range Register (MTRR) setup on some hyperthreaded machines may be incorrect following a suspend/resume cycle. This can cause graphics performance (specifically, scrolling) to slow considerably after a suspend/resume cycle.
To work around this issue, disable and then re-enable the hyperthreaded sibling CPUs around suspend/resume, for example:
# Disable hyper-threading processor cores on suspend and hibernate, re-enable
# on resume.
# This file goes into /etc/pm/sleep.d/
case $1 in
echo 0 > /sys/devices/system/cpu/cpu1/online
echo 0 > /sys/devices/system/cpu/cpu3/online
echo 1 > /sys/devices/system/cpu/cpu1/online
echo 1 > /sys/devices/system/cpu/cpu3/online
In Red Hat Enterprise Linux 6.2,
nmi_watchdog registers with the
perf subsystem. Consequently, during boot, the
perf subsystem grabs control of the performance counter registers, blocking OProfile from working. To resolve this, either boot with the
nmi_watchdog=0 kernel parameter set, or run the following command to disable it at run time:
echo 0 > /proc/sys/kernel/nmi_watchdog
nmi-watchdog, use the following command
echo 1 > /proc/sys/kernel/nmi_watchdog
kernel component, BZ#603911
Due to the way ftrace works when modifying the code during start-up, the NMI watchdog causes too much noise and ftrace can not find a quiet period to instrument the code. Consequently, machines with more than 512 CPUs will encounter issues with the NMI watchdog. Such issues will return error messages similar to
BUG: NMI Watchdog detected LOCKUP and have either
ipi_handler in the backtrace. To work around this issue, disable NMI watchdog by setting the
nmi_watchdog=0 kernel parameter, or using the following command at run time:
echo 0 > /proc/sys/kernel/nmi_watchdog
On 64-bit POWER systems the EHEA NIC driver will fail when attempting to dump a
vmcore via NFS. To work around this issue, utilize other kdump facilities, for example dumping to the local file system, or dumping over SSH.
kernel component, BZ#587909
A BIOS emulated floppy disk might cause the installation or kernel boot process to hang. To avoid this, disable emulated floppy disk support in the BIOS.
The preferred method to enable nmi_watchdog on 32-bit x86 systems is to use either
nmi_watchdog=lapic parameters. The parameter
nmi_watchdog=1 is not supported.
The kernel parameter,
pci=noioapicquirk, is required when installing the 32-bit variant of Red Hat Enterprise Linux 6 on HP xw9300 workstations. Note that the parameter change is not required when installing the 64-bit variant.