Red Hat ENTERPRISE LINUX 5.4 - TECHNICAL NOTES Manual page 156

Chapter 1. Package Updates
• In previous kernels the tuntap device send path did not have any packet accounting. This meant
that the user-space sender could pin down arbitrary amounts of kernel memory by continuing to
send data to an end-point that was congested. This update adds packet accounting to the tun
driver so that virtio-net gets congestion feedback which is necessary to prevent packet loss for
protocols lacking congestion control (such as UDP) when used in a guest.
• This update adds the virtualization feature VT-d. This feature provides hardware support for directly
assigning physical devices to Xen fully virtualized (HVM) guests or KVM guests. The principal
benefit of the feature is to improve device access performance to be close to native speeds. Please
Red Hat Knowledgebase
refer to the
avoid possible system instability issues.
VT-d support is disabled by default. To enable VT-d one must add intel_iommu=on to the
kernel commandline. Enabling VT-d is required to assign a host's PCI device to a KVM guest.
1028
(BZ#504363)
Additionally, only the assignment of NIC devices from host to guest is supported. Assigning a
block device (hard disk) to a guest is not supported. On hardware platforms that support IOMMU
passthrough it is recommended to also use the iommu=pt kernel commandline option as this will
improve the performance of I/O devices in the host. This parameter has no effect on performance
for devices assigned to guests.
When the iommu=pt mode, if a device is assigned to (and then de-assigned from) a guest, it can no
longer be used in the host until the host has been rebooted. PCI hotplug devices can not be used in
iommu=pt mode
• This update includes a fix for kernel panic encountered when attempting to run a kdump process on
hardware virtual machine (HVM) in an ia64 architecture environment.
• This update corrects softlockup issues encountered when booting earlier kernel versions in a virtual
environment and setting the clocksource to read from the system's Programmable Interval Timer
1030
(PIT). (BZ#427588
• A problem identified with Xen kernels manifested with meminfo reporting an incorrect LowTotal
of 4Tb. A patch applied to the driver alters how highmem pages are handled and corrects the error.
1031
(BZ#428892 )
• When users set LPFC HBA storage to reset in a loop the system would attempt to rediscover
SCSI devices and some of these processes would time-out. The issue was found to be code paths
deleting SCSI devices without setting the device state to SDEV_DEL. A patch included in this update
corrects this behavior(BZ#430170
• The Xen kernel does not currently support the suspend functionality. A fix has been applied to this
release to remove the "Suspend" option from graphical user interface menus.
• This update fixes a race condition when queuing incoming iucv messages by spreading
the message queue spinlock in the message_pending callback across the entire callback
function.This resolves the race condition and enhances system stability.
• This feature fixes hexdump data in s390dbf traces, allowing Red Hat Enterprise Linux to have
complete registered state change notification (RSCN) traces (up to 1024 bytes).
138
1026
before using PCI device assignment with this technology to
(BZ#500901
)
1032
)
(BZ#495863
1027
)
(BZ#418591
(BZ#499626
1025
)
1029
)
1033
(BZ#430928 )
1034
)
1035
(BZ#470618 )