Red Hat Training

A Red Hat training course is available for Red Hat Enterprise Linux

2.19. kernel

The kernel packages contain the Linux kernel, the core of any Linux operating system.
  • The Emulex lpfc driver is missing functionality required to support 16 Gb point-to-point configurations for all adapters in Red Hat Enterprise Linux 5. All other currently available 16 Gb lpfc configurations are supported on most adapters available. Specifically, the LPe16000B adapter is not supported for any configuration, and the LPe16000A adapter is supported for all configurations besides a point-to-point configuration.
  • The qla2xxx driver creates optrom and optrom_ctl files in sysfs which are used by some tools such as the scli command line tool from QLogic. However, the functions which implement these pseudo-files have race conditions. As a consequence, a kernel panic occurs when multiple tools use these files at the same time. To work around this problem, make sure only one such process is running at a given point of time.
  • Red Hat Enterprise Linux 5 can become unresponsive or even terminate due to the lack of ticketed spinlocks in the shrink_active_list() function.
  • When USB hardware uses the ACM interface, there is a race condition that can lead to a system deadlock due to the spinlocks not disabling interrupts. This has been noticed through various types of softlockups. To workaround this problem, reboot the machine.
  • If kdump is configured on an i686 system using a non-PAE kernel and memory larger than 4 GB, it creates an elf core header which includes extra unavailable memory range. This causes kdump to become unresponsive.
  • A large number of kernel log messages may flood netconsole while under heavy RX traffic, causing the netconsole kernel module to stop working. To work around this issue, avoid the use of netconsole, or remove the netconsole module using the rmmod netconsole command and re-configure it again using the insmod netconsole command.
  • To update firmware on Mellanox cards, use mstflint which replaces the outdated tvflash utility.
  • The kernel in Red Hat Enterprise Linux 5 does not support Data Center Bridging (DCB). Software-based Fibre Channel over Etherner (FCoE) is a Technology Preview and it is therefore recommended to use Red Hat Enterprise Linux 6 for fully supported software-based FCoE. The following hardware-accelerated FCoE cards are fully supported in Red Hat Enterprise Linux 5: Emulex LPFC, QLogic qla2xxx, Brocade BFA. (BZ#860112)
  • Throughput across machines using IPv6 addresses and with bnx2x interfaces set up can be degraded.
  • The following problems can occur when using Brocade 1010 and 1020 Converged Network Adapters (CNAs):
    • BIOS firmware may not be able to log in the Fibre Channel over Ethernet (FCoE) session when loading a Brocade optional BIOS, which causes the server to be unable to boot and the following error message to appear:
      Adapter 1/0/0 Link initialization failed. Disabling BIOS
      
    • Configuration cannot be saved via serial port of the server. Use a physical console or Brocade HSM software.
    Contact Brocade for additional information on these problems.
  • 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 error messages to continuously appear 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 problem, unload the Brocade BFA driver.
  • Master Boot Record (MBR) or the /boot partition can be installed on an incorrect disk if the server boots from storage area network (SAN) with many Logical Unit Numbers (LUNs) assigned. To work around this problem, partition the space manually so that the operating system uses only the boot LUN as the root (/) and /boot partitions. (BZ#852305)
  • Qemu-kvm does not check if a given CPU flag is really supported by the KVM kernel module. Attempting to enable the "acpi" flag can lead to a kernel panic on guest machines. To work around this problem, do not enable the "acpi" CPU flag in the configuration of a virtual machine. (BZ#838921)
  • Running the ethtool --identify command in a production environment blocks network traffic and certain network configuration operations until ethtool is aborted. To prevent this problem, do not run ethtool --identify in a production environment; this command is supposed for debugging purposes only.
  • Starting with Red Hat Enterprise Linux 5.8, the size of I/O operations allowed by the NFS server has been increased by default. The new default max block size varies depending on RAM size, with a maximum of 1M (1048576 bytes).
    This may cause problems for 32-bit servers configured to use large numbers of nfsd threads. For such servers, we recommend decreasing the number of threads, or decreasing the I/O size by writing to the /proc/fs/nfsd/max_block_size file before starting nfsd. For example, the following command restores the previous default iosize of 32k:
    ~]# echo 32767 >/proc/fs/nfsd/max_block_size
    (BZ#765751 )
  • If the qla4xxx driver fails to discover all iSCSI targets, make sure to Clear Persistent Targets and set up iSCSI again via CTRL+Q in the Qlogic iSCSI option ROM BIOS.
  • The OProfile infrastructure in Red Hat Enterprise Linux 5 does not support the hardware performance counters of the AMD family 0x15 processor family; profiling is only available in timer interrupt mode. When profiling on bare metal, OProfile automatically selects the timer interrupt mode. When running under kernel-xen, due to different CPU family reporting, OProfile must be explicitly configured to use timer interrupt mode. This is possible by adding options oprofile timer=1 to the /etc/modprobe.conf file. (BZ#720587)
  • Red Hat Enterprise Linux 5 may become unresponsive due to the lack of ticketed spinlocks in the shrink_active_list() function. As a result, the spin_lock_irq(&zone->lru_lock) operation disables interrupts, and the following error message is returned when the system hangs:
    NMI Watchdog detected LOCKUP
    
  • Booting a Red Hat Enterprise Linux 5 system with a connected DVD drive and the smartdservice running hangs with the following error messages:
    Starting smartd: hdc: drive_cmd: status=0x58 { DriveReady SeekComplete
    DataRequest }
    ide: failed opcode was: 0xa1
    hdc: status error: status=0x58 { DriveReady SeekComplete DataRequest }
    ide: failed opcode was: unknown
    hdc: drive not ready for command
    hdc: status timeout: status=0xd8 { Busy }
    ide: failed opcode was: unknown
    hdc: drive not ready for command
    hdc: ATAPI reset complete
    hdc: status error: status=0x58 { DriveReady SeekComplete DataRequest }
    ⋮
    
    To work around this issue, disconnect the DVD drive or turn the smartd service off with the following command:
    ~]# chkconfig smartd off
  • The modify SRQ verb is not supported by the eHCA adapter and will fail with an error code when called from an application context.
  • In RHEL 5.8, machine check (MCE) support for Intel Nehalem or newer CPUs (family 6, model >= 26) is disabled. This is a change from RHEL5.6 and earlier where basic MCE support was provided for these CPUs. Uncorrected CPU and memory errors will cause an immediate CPU shut down and system panic.
  • On a Red Hat Enterprise Linux 5.8 system and later, while hand-loading the i386 (32-bit) kernel on z210/z210 SFF with BIOS 1.08, the system may fail to boot. To workaround this issue, please add the following parameter to the boot command line option:
    pci=nosort
    (BZ#703538)
  • Red Hat Enterprise Linux 5.7 has introduced a new multicast snooping feature for the bridge driver used for virtualization (virt-bridge). This feature is disabled by default in order to not break any existing configurations. To enable this feature, please set the following tunnable parameter to 1:
    /sys/class/net/breth0/bridge/multicast_snooping
    Please note that when multicast snooping is enabled, it may cause a regression with certain switches where it causes a break in the multicast forwarding for some peers.
  • By default, libsas defines a wideport based on the attached SAS address, rather than the specification compliant strict definition of also considering the local SAS address. In Red Hat Enterprise Linux 5.8 and later, only the default loose definition is available. The implication is that if an OEM configures an SCU controller to advertise different SAS addresses per PHY, but hooks up a wide target or an expander to those PHYs, libsas will only create one port. The expectation, in the strict case, is that this would result in a single controller multipath configuration.
    It is not possible to use a single controller multipath without the strict_wide_port functionality. Multi-controller multipath should behave as a expected.
    A x8 multipath configuration through a single expander can still be obtained under the following conditions:
    1. Start with an SCU SKU that exposes (2) x4 controllers (total of 8 PHYs)
    2. Assign sas_address1 to all the PHYs on controller1
    3. Assign sas_address2 to all the PHYs on controller2
    4. Hook up the expander across all 8 PHYs
    5. Configure multipath across the two controller instances
    It is critical for controller1 to have a distinct address from controller2, otherwise the expander will be unable to correctly route connection requests to the proper initiator. (BZ#651837)
  • On a Red Hat Enterprise Linux 5 system, it is advisable to update the firmware of the HP ProLiant Generation 6 (G6) controller's firmware to version 5.02 or later. Once the firmware is successfully updated, reboot the system and Kdump will work as expected.
    HP G6 controllers include: P410i, P411, P212, P712, and P812
    In addition, kdump may fail when using the HP Smart Array 5i Controller on a Red Hat Enterprise Linux 5 system. (BZ#695493)
  • On Red Hat Enterprise Linux 5.5 and later, suspending the system with the lpfc driver loaded may crash the system during the resume operation. Therefore, systems using the lpfc driver, either unload the lpfc driver before the system is suspended, or ,if that is not possible, do not suspend the system. (BZ#703631)
  • NUMA class systems should not be booted with a single memory node configuration. Configuration of single node NUMA systems will result in contention for the memory resources on all of the non-local memory nodes. As only one node will have local memory the CPUs on that single node will starve the remaining CPUs for memory allocations, locks, and any kernel data structure access. This contention will lead to the "CPU#n stuck for 10s!" error messages. This configuration can also result in NMI watchdog timeout panics if a spinlock is acquired via spinlock_irq() and held for more than 60 seconds. The system can also hang for indeterminate lengths of time.
    To minimize this problem, NUMA class systems need to have their memory evenly distributed between nodes. NUMA information can be obtained from dmesg output as well as from the numastat command. (BZ#529428)
  • When upgrading from Red Hat Enterprise Linux 5.0, 5.1 or 5.2 to more recent releases, the gfs2-kmod may still be installed on the system. This package must be manually removed or it will override the (newer) version of GFS2 which is built into the kernel. Do not install the gfs2-kmod package on later versions of Red Hat Enterprise Linux. gfs2-kmod is not required since GFS2 is built into the kernel from 5.3 onwards. The content of the gfs2-kmod package is considered a Technology Preview of GFS2, and has not received any updates since Red Hat Enterprise Linux 5.3 was released.
    Note that this note only applies to GFS2 and not to GFS, for which the gfs-kmod package continues to be the only method of obtaining the required kernel module.
  • Issues might be encountered on a system with 8Gb/s LPe1200x HBAs and firmware version 2.00a3 when the Red Hat Enterprise Linux 5.8 kernel is used with the in-box LPFC driver. Such issues include loss of LUNs and/or fiber channel host hangs during fabric faults with multipathing.
    To work around these issues, it is recommended to either:
    • Downgrade the firmware revision of the 8Gb/s LPe1200x HBA to revision 1.11a5, or
    • Modify the LPFC driver’s lpfc_enable_npiv module parameter to zero.
      When loading the LPFC driver from the initrd image (i.e. at system boot time), add the line
      options lpfc_enable_npiv=0
      
      to /etc/modprobe.conf and re-build the initrd image.
      When loading the LPFC driver dynamically, include the lpfc_enable_npiv=0 option in the insmod or modprobe command line.
    For additional information on how to set the LPFC driver module parameters, refer to the Emulex Drivers for Linux User Manual.
  • If AMD IOMMU is enabled in BIOS on ProLiant DL165 G7 systems, the system will reboot automatically when IOMMU attempts to initialize. To work around this issue, either disable IOMMU, or update the BIOS to version 2010.09.06 or later. (BZ#628534)
  • As of Red Hat Enterprise Linux 5.6, the ext4 file system is fully supported. However, provisioning ext4 file systems with the anaconda installer is not supported, and ext4 file systems need to be provisioned manually after the installation. (BZ#563943)
  • In some cases the NFS server fails to notify NFSv4 clients about renames and unlinks done by other clients, or by non-NFS users of the server. An application on a client may then be able to open the file at its old pathname (and read old cached data from it, and perform read locks on it), long after the file no longer exists at that pathname on the server.
    To work around this issue, use NFSv3 instead of NFSv4. Alternatively, turn off support for leases by writing 0 to /proc/sys/fs/leases-enable (ideally on boot, before the nfs server is started). This change prevents NFSv4 delegations from being given out, restore correctness at the expense of some performance.
  • Some laptops may generate continuous events in response to the lid being shut. Consequently, the gnome-power-manager utility will consume CPU resources as it responds to each event. (BZ#660644)
  • A kernel panic may be triggered by the lpfc driver when multiple Emulex OneConnect Universal Converged Network Adapter initiators are included in the same Storage Area Network (SAN) zone. Typically, this kernel panic will present after a cable is pulled or one of the systems is rebooted. To work around this issue, configure the SAN to use single initiator zoning. (BZ#574858)
  • If a Huawei USB modem is unplugged from a system, the device may not be detected when it is attached again. To work around this issue, the usbserial and usb-storage driver modules need to be reloaded, allowing the system to detect the device. Alternatively, the if the system is rebooted, the modem will be detected also. (BZ#517454)
  • Memory on-line is not currently supported with the Boxboro-EX platform. (BZ#515299)
  • Unloading a PF (SR-IOV Physical function) driver from a host when a guest is using a VF (virtual function) from that device can cause a host crash. A PF driver for an SR-IOV device should not be unloaded until after all guest virtual machines with assigned VFs from that SR-IOV device have terminated. (BZ#514360)
  • Data corruption on NFS file systems might be encountered on network adapters without support for error-correcting code (ECC) memory that also have TCP segmentation offloading (TSO) enabled in the driver. Note: data that might be corrupted by the sender still passes the checksum performed by the IP stack of the receiving machine A possible work around to this issue is to disable TSO on network adapters that do not support ECC memory. (BZ#504811)
  • After installation, a System z machine with a large number of memory and CPUs (e.g. 16 CPU's and 200GB of memory) might may fail to IPL. To work around this issue, change the line
    ramdisk=/boot/initrd-2.6.18-<kernel-version-number>.el5.img
    
    to
    ramdisk=/boot/initrd-2.6.18-<kernel-version-number>.el5.img,0x02000000
    
    The command zipl -V should now show 0x02000000 as the starting address for the initial RAM disk (initrd). Stop the logical partition (LPAR), and then manually increase the storage size of the LPAR.
  • On certain hardware configurations the kernel may panic when the Broadcom iSCSI offload driver (bnx2i.ko and cnic.ko) is loaded. To work around this do not manually load the bnx2i or cnic modules, and temporarily disable the iscsi service from starting. To disable the iscsi service, run:
    ~]# chkconfig --del iscsi
    ~]# chkconfig --del iscsid
    On the first boot of your system, the iscsi service may start automatically. To bypass this, during bootup, enter interactive start up and stop the iscsi service from starting.
  • In Red Hat Enterprise Linux 5, invoking the kernel system call "setpriority()" with a "which" parameter of type "PRIO_PROCESS" does not set the priority of child threads. (BZ#472251)
  • A change to the cciss driver in Red Hat Enterprise Linux 5.4 made it incompatible with the echo disk < /sys/power/state suspend-to-disk operation. Consequently, the system will not suspend properly, returning messages such as:
    Stopping tasks:
    ======================================================================
    stopping tasks timed out after 20 seconds (1 tasks remaining):
    cciss_scan00
    Restarting tasks...<6> Strange, cciss_scan00 not stopped
    done
    
    (BZ#513472)
  • The kernel is unable to properly detect whether there is media present in a CD-ROM drive during kickstart installs. The function to check the presence of media incorrectly interprets the "logical unit is becoming ready" sense, returning that the drive is ready when it is not. To work around this issue, wait several seconds between inserting a CD and asking the installer (anaconda) to refresh the CD. (BZ#510632)
  • When a cciss device is under high I/O load, the kdump kernel may panic and the vmcore dump may not be saved successfully. (BZ#509790)
  • Configuring IRQ SMP affinity has no effect on some devices that use message signaled interrupts (MSI) with no MSI per-vector masking capability. Examples of such devices include Broadcom NetXtreme Ethernet devices that use the bnx2 driver.
    If you need to configure IRQ affinity for such a device, disable MSI by creating a file in /etc/modprobe.d/ containing the following line:
    options bnx2 disable_msi=1
    
    Alternatively, you can disable MSI completely using the kernel boot parameter pci=nomsi. (BZ#432451)
  • The smartctl tool cannot properly read SMART parameters from SATA devices. (BZ#429606)
  • IBM T60 laptops will power off completely when suspended and plugged into a docking station. To avoid this, boot the system with the argument acpi_sleep=s3_bios. (BZ#439006)
  • The QLogic iSCSI Expansion Card for the IBM Bladecenter provides both ethernet and iSCSI functions. Some parts on the card are shared by both functions. However, the current qla3xxx and qla4xxx drivers support ethernet and iSCSI functions individually. Both drivers do not support the use of ethernet and iSCSI functions simultaneously.
    Because of this limitation, successive resets (via consecutive ifdown/ifup commands) may hang the device. To avoid this, allow a 10-second interval after an ifup before issuing an ifdown. Also, allow the same 10-second interval after an ifdown before issuing an ifup. This interval allows ample time to stabilize and re-initialize all functions when an ifup is issued. (BZ#276891)
  • Laptops equipped with the Cisco Aironet MPI-350 wireless may hang trying to get a DHCP address during any network-based installation using the wired ethernet port.
    To work around this, use local media for your installation. Alternatively, you can disable the wireless card in the laptop BIOS prior to installation (you can re-enable the wireless card after completing the installation). (BZ#213262)
  • Hardware testing for the Mellanox MT25204 has revealed that an internal error occurs under certain high-load conditions. When the ib_mthca driver reports a catastrophic error on this hardware, it is usually related to an insufficient completion queue depth relative to the number of outstanding work requests generated by the user application.
    Although the driver will reset the hardware and recover from such an event, all existing connections at the time of the error will be lost. This generally results in a segmentation fault in the user application. Further, if opensm is running at the time the error occurs, then you need to manually restart it in order to resume proper operation. (BZ#251934)
  • The IBM T41 laptop model does not enter Suspend Mode properly; as such, Suspend Mode will still consume battery life as normal. This is because Red Hat Enterprise Linux 5 does not yet include the radeonfb module.
    To work around this, add a script named hal-system-power-suspend to /usr/share/hal/scripts/ containing the following lines:
    chvt 1
    radeontool light off
    radeontool dac off
    
    This script will ensure that the IBM T41 laptop enters Suspend Mode properly. To ensure that the system resumes normal operations properly, add the script restore-after-standby to the same directory as well, containing the following lines:
    radeontool dac on
    radeontool light on
    chvt 7
    
    (BZ#227496)
  • If the edac module is loaded, BIOS memory reporting will not work. This is because the edac module clears the register that the BIOS uses for reporting memory errors.
    The current Red Hat Enterprise Linux Driver Update Model instructs the kernel to load all available modules (including the edac module) by default. If you wish to ensure BIOS memory reporting on your system, you need to manually blacklist the edac modules. To do so, add the following lines to /etc/modprobe.conf:
    blacklist edac_mc
    blacklist i5000_edac
    blacklist i3000_edac
    blacklist e752x_edac
    
    (BZ#441329)
  • Due to outstanding driver issues with hardware encryption acceleration, users of Intel WiFi Link 4965, 5100, 5150, 5300, and 5350 wireless cards are advised to disable hardware accelerated encryption using module parameters. Failure to do so may result in the inability to connect to Wired Equivalent Privacy (WEP) protected wireless networks after connecting to WiFi Protected Access (WPA) protected wireless networks.
    To do so, add the following options to /etc/modprobe.conf:
    alias wlan0 iwlagn
    options iwlagn swcrypto50=1 swcrypto=1
    
    where wlan0 is the default interface name of the first Intel WiFi Link device.
    (BZ#468967)
  • A kernel security fix released between Red Hat Enterprise Linux 5.7 and 5.8 may prevent PCI passthrough working and guests starting. Refer to Red Hat Knowledgebase article 66747 for further details.
The following note applies to the PowerPC architecture:
  • The size of the PowerPC kernel image is too large for OpenFirmware to support. Consequently, network booting will fail, resulting in the following error message:
    Please wait, loading kernel...
    /pci@8000000f8000000/ide@4,1/disk@0:2,vmlinux-anaconda: No such file or directory
    boot:
    
    To work around this:
    1. Boot to the OpenFirmware prompt, by pressing the '8' key when the IBM splash screen is displayed.
    2. Run the following command:
      ~]# setenv real-base 2000000
    3. Boot into System Management Services (SMS) with the command:
      ~]# 0> dev /packages/gui obe
    (BZ#462663)