Linux 6.16 changelog.
Summary: This release includes some Ext4 performance improvements; XFS support for large atomic writes; support for USB audio offload; support for zero-copy send TCP payloads from DMABUF memory; various futex improvements; initial support for Intel Trusted Domain Extensions; automatic weighted interleaved memory allocation policy; support for sending coredumps over an AF_UNIX socket, and make easier to build your kernel optimized for your local CPU. As always, there are many other features, new drivers, improvements and fixes. Also, you might be interested in the LWN merge window report: part 1
XFS support for large atomic writes
This release adds support for large atomic writes. These are writes of multiple FS blocks that will be written
atomically - either all the blocks are written, or none.
Recommended LWN article: An update on torn-write protection
USB audio offload support
After years of work, this release incorporates USB offload support for audio devices. This feature offers major power savings on embedded devices where a USB audio stream can continue to flow while the rest of the system is sleeping, something that devices running on battery power really care about. This is a feature that is available on Android under different implementations, now there is an unified implementation that everbody can use
Initial support for Intel Trusted Domain Extensions
This release adds initial support for Intel's Trust Domain Extensions, which protect confidential guest VMs from the host and physical attacks. This is done encrypting the memory of the guest VM. This is similar to AMD's SEV-SNP (already supported).
Documentation: Intel Trust Domain Extensions (TDX)
Intel documentation: Intel® Trust Domain Extensions (Intel® TDX)
Allow to zero-copy send TCP payloads from DMABUF memory
This release adds support for device memory TCP TX path. Device memory TCP (merged in Linux 6.12
Documentation: TX Interface
Recommended LWN article: Direct-to-device networking
Automatic weighted interleaved memory allocation policy
On modern NUMA systems, memory can be allocated from different nodes, some of which can offer better bandwith characteristics than others. Linux allows to configure different weights to each node so that allocations are done as specified by the administrator, but this configuration is complex and does not address memory hotplug events. Round-robin allocations do not take account of bandwith information either. This release adds an automatic auto-tuning policy. In this mode, all node weights are re-calculated and overwritten whenever new bandwidth data is made available during either boot or hotplug events.
Recommended LWN article: Automatic tuning for weighted interleaving
Support for Intel Advanced Performance Extensions
This release adds support for Intel APX, or Advanced Performance Extensions. Intel APX doubles the number of general-purpose registers from 16 to 32, allowing code to contain fewer loads and stores, which leads to better performance and power savings.
Documentation: https://www.intel.com/content/www/us/en/developer/articles/technical/advanced-performance-extensions-apx.html
Add support for sending coredumps over an AF_UNIX socket
This release adds support for sending coredumps over an AF_UNIX socket. This coredump socket allows userspace to not have to rely on usermode helpers for processing coredumps, and provides a safer way to handle them instead of relying on privileged coredumping helpers. This functionality is implemented using another feature that allows to hand out pidfds for reaped peer tasks.
Recommended LWN article: Slowing the flow of core-dump-related CVEs
Futex improvements
This release adds support for process local hash which is only shared by all threads of process (as opossed by the futex hash that by default is system wide and shared by all tasks). This hash will only be used for a
{{{PROCESS_PRIVATE}}} futex operation. There is also support for {{{FUTEX2_NUMA}}} (which feature extends the futex interface to be NUMA-aware) and {{{FUTEX2_MPOL}}} (which feature extends the futex interface to be mempolicy-aware) (cover)
Some Ext4 performance improvements
This release incorporates some performance to the "fast commit"
Línea 72: Línea 64:
Build optimization for the local CPU on x86
For those users who still compile their own kernels, this release adds a {{{CONFIG_X86_NATIVE_CPU}}} option that makes the kernel build with the {{{-march=native}}} option. This makes compilers optimize the compilation for your CPU, which can provide nice performance improvements.