Chapter 5. Important changes to external kernel parameters
This chapter provides system administrators with a summary of significant changes in the kernel distributed with Red Hat Enterprise Linux 9.9. These changes could include, for example, added or updated proc entries, sysctl, and sysfs default values, boot parameters, kernel configuration options, or any noticeable behavior changes.
New kernel parameters
intel_idle.table=
[X86]
Override selected exit-latency and target-residency values in the CPU-specific C-state table.
Format: state:exit_latency:target_residency[,state:exit_latency:target_residency…]
Latency and residency values are expressed in microseconds. Empty fields preserve the existing value. For example, intel_idle.table="C6::400" changes only the target residency of the C6 state. The kernel validates the supplied values before using them.
mpt3sas.issue_scsi_cmd_to_bringup_drive=
[SCSI]
Control whether the mpt3sas driver sends TEST UNIT READY and START UNIT commands while bringing devices to the ready state during unblock processing. The default is 1, which preserves the previous behavior.
mpt3sas.command_retry_count=
[SCSI]
Control the number of retries during device discovery and readiness checks. This setting can improve discovery of slow or transient SAS and PCIe devices.
slider_balance=
[X86]
Override the SoC slider value used by the balanced platform profile for the Intel INT340x thermal driver. The default slider value is 3. A value changed after module load takes effect the next time balanced is written to the platform-profile sysfs attribute. Values that overlap the performance value 0 or the power-saver endpoint are rejected.
md.check_new_feature=
[MD]
Bypass the check that requires the pad3 field in an MD superblock to be zero. This permits assembly of arrays created with newer metadata features that use this field.
check_lps0_constraints=
[ACPI]
Control evaluation of LPS0 _DSM Function 1 and checking of the resulting constraints during suspend-to-idle diagnostics. This avoids potentially problematic firmware evaluation when LPS0 constraint diagnostics are not required.
Changed kernel parameters
rdt=
[X86]
Control Intel telemetry event groups through the rdt= boot parameter. The parameter now accepts perf and energy to force activation of all Intel telemetry event groups of the selected type. Prefix the selector with ! to disable the event groups. A :guid suffix restricts the operation to one event group.
For example:
rdt=perf
rdt=!energy
rdt=perf:<guid>
Telemetry event groups with too few hardware counters for all resource monitoring identifiers (RMIDs) are disabled by default. Explicitly enabling such a group limits the number of monitor groups to the smallest available hardware-counter count.
bt_link_mask_override=
[X86]
Control the Intel Sound Open Firmware Bluetooth link mask. The parameter now accepts 0 so that users can disable a Bluetooth link that firmware reports through the NHLT table.
Removed kernel parameters
intel_idle.preferred_cstates=
[X86]
Control C1/C1E preference on Alder Lake. This parameter affected only Alder Lake, where C1E was already the default, and had no effect on other platforms.
rtl8xxxu.ht40_2g=
[NET]
Control 40 MHz operation in the 2.4 GHz band for the rtl8xxxu driver. 40 MHz operation in the 2.4 GHz band is now enabled by default. To disable it globally, use the existing cfg80211.cfg80211_disable_40mhz_24ghz parameter.
drop_dir_cache=
[CIFS]
Invalidate cached directory entries for the CIFS client. When CONFIG_CIFS_DEBUG is enabled, write 0 to the existing open_dirs procfs entry instead.
New sysctl parameters
vm.thp_thisnode_reclaim
Control whether __GFP_THISNODE transparent huge page allocations enter reclaim after compaction fails. The default is 1, preserving the previous, more aggressive local-node reclaim behavior. Set the value to 0 to avoid reclaiming one NUMA node when memory is available on other nodes.
vm.mem_cgroup_reclaim_retries
Set the maximum number of memory-cgroup reclaim attempts before the out-of-memory path is invoked. The valid range is 1 through 16, and the default is 16. The setting applies to reclaim initiated by memory charging, memory.high, memory.max, memory.reclaim, and forced-empty operations.
vm.dr_prio_drop
Allow direct reclaim to stop before reaching the most aggressive scan priorities. The valid range is 0 through 11; the default is 0, which disables early termination. A nonzero value causes direct reclaim to stop when the priority drops below 12 - vm.dr_prio_drop and proceed to out-of-memory evaluation. The setting does not affect kswapd.
net.vsock.ns_mode
Report whether a network namespace uses the global or local VSOCK namespace mode.
net.vsock.child_ns_mode
Select the VSOCK namespace mode inherited by newly created child network namespaces. child_ns_mode is write-once. A subsequent write of a different value fails with EBUSY. A namespace in local mode cannot select global mode for its children. Writing local in the initial network namespace therefore locks future descendants into local mode. VHOST-VSOCK and loopback transports support these namespaces; virtio-vsock remains outside namespace isolation.