While a "core/entry" pull request may not seem that exciting, this time around for the Linux 5.11 kernel there are two prominent additions.

The two big additions with the core/entry pull request submitted on Monday for Linux 5.11 are:

- TIF_NOTIFY_SIGNAL that replaces the existing and inefficient signal delivery mode of task work. What makes this notable is delivering a huge performance boost to IO_uring, the modern I/O interface for the Linux kernel that has been seeing much praise and adoption. The speed-up with IO_uring on Linux 5.11 is quite commendable along with the other improvements led by Jens Axboe.

HWMON maintainer Guenter Roeck has sent in all of the hardware monitoring changes destined for the Linux 5.11 kernel.

There are several prominent additions to the hardware monitoring subsystem for Linux 5.11 including:

- The AMD SB-TSI sensor driver has been mainlined as a sideband temperature sensor interface for AMD EPYC server platforms. AMD SB-TSI should ultimately prove useful for cases like OpenBMC deployments.

Initially found with "Elkhart Lake" SoCs and likely to be found on further future Intel client SoCs is the integrated memory controller supporting in-band ECC (IBECC). Coming with Linux 5.11 is the "IGEN6" EDAC driver for handling this error detection and correction on Intel SoCs sporting IBECC.

The Linux 5.11 kernel cycle continues to prove to be very exciting. The latest are SECCOMP filters for this secure computing mode yielding a nice speed-up.

The SECCOMP updates for Linux 5.11 include the addition of constant-action bitmaps to reduce the overhead for many real-world syscall filters from O(N) to O(1). The filters benefiting the most are for allow/reject-only system call handling. This constant action bitmaps is also faster than BPF call optimization.

The x86-platform-drivers area of the kernel has a lot of prominent additions with Linux 5.11 for benefiting a variety of AMD and Intel laptops.

The x86-platform-drivers additions have an assortment of Intel/AMD hardware support changes primarily benefiting Linux laptop users. Among the new changes with the ongoing Linux 5.11 merge window are:

- The AMD SoC PMC driver for the power management controller on Raven Ridge, Picasso, Renoir, and Cezanne and is used for handling S2idle transactions. Yes, I'll run some power tests soon.

The cryptography subsystem within the Linux kernel is constantly seeing new hardware drivers and other improvements with the current Linux 5.11 cycle being no different.

Herbert Xu sent in the crypto updates on Monday for Linux 5.11 and has already landed into the mainline kernel.

With the PCI subsystem updates for the in-development Linux 5.11 kernel is the ability to report whether a device is making use of the 64 GT/s link speed allowed by PCI Express 6.0.

With the Linux 5.11 PCI updates is support for reporting a PCI Express link speed of 64 GT/s. The tentative PCI Express 6.0 specification defines a 64.0 GT/s link speed and this current change is about being able to report it to the user rather than as an unknown speed. The PCIe hardware should already be negotiating the best possible speed automatically. The change for Linux 5.11 was contributed by a Synopsys engineer.

The input subsystem changes for the Linux 5.11 kernel have now been submitted and merged. Along related lines, the HID subsystem changes were also submitted with notable updates as well.

On the input side with Linux 5.11 a new feature is the "inhibited" feature to temporarily disregard input from select devices. The use-case for this inhibited input device support is for devices like 2-in-1 laptops where the laptop may be folded underneath the device at times and during that period no input events should reach user-space as it would amount to accidental input. With today's devices there are also other similar setups where at times you may want to avoid any input events from a given device or to prevent it from potentially waking the system. This inhibited input support was spearheaded by Google's Chrome OS engineers.

The previously reported on work for frequency invariance calculations for AMD CPUs with a focus on the AMD EPYC 7002 series has been merged for Linux 5.11 as part of the "sched/core" material.

Following all of the Intel Linux kernel work in recent months around frequency invariance handling for more accurate load tracking and making more accurate frequency scaling decisions, the initial AMD implementation is here with Linux 5.11 as part of the core scheduler updates. In basic terms, the frequency invariance calculation is for addressing the issue of tasks appearing larger if the CPU is running slower so the frequency invariance takes into account the current frequency relative to the maximum possible frequency.

The latest hardware enablement around Intel's Alder Lake for the Linux kernel is audio support.

SUSE's Takashi Iwai already sent in the sound hardware changes for the ongoing Linux 5.11 merge window and that successfully landed. There are some low-level changes like the Intel DSP support now making use of the new Auxiliary Bus, various code clean-ups, fixes thanks to the Undefined Behavior Sanitizer and Kernel Concurrency Sanitizer spotting problems, better USB audio implicit feedback support, new audio quirks, continued work on Intel Sound Open Firmware, and also some HDMI audio fixes.

The ARM64 architecture updates were sent in already for Linux 5.11 along with the various ARM SoC additions, DeviceTree additions for new hardware support, and similar changes. There is a lot of new hardware support as always being brought up by the mainline kernel.

As part of the areas of the kernel overseen by Greg Kroah-Hartman is the USB subsystem. The USB (and Thunderbolt) updates are now in mainline as part of the ongoing Linux 5.11 merge window.

The latest hardware enablement around Intel's Alder Lake for the Linux kernel is audio support.

SUSE's Takashi Iwai already sent in the sound hardware changes for the ongoing Linux 5.11 merge window and that successfully landed. There are some low-level changes like the Intel DSP support now making use of the new Auxiliary Bus, various code clean-ups, fixes thanks to the Undefined Behavior Sanitizer and Kernel Concurrency Sanitizer spotting problems, better USB audio implicit feedback support, new audio quirks, continued work on Intel Sound Open Firmware, and also some HDMI audio fixes.

The Linux 5.11 merge window continues being very active this week with Linus Torvalds hoping kernel maintainers will get in all of their new feature code well before Christmas.

As reported a few months ago, the Linux kernel is looking to drop its WiMAX support. For phasing out their WiMAX support, with Linux 5.11 they have now moved the code into the "staging" area of the kernel. This is intended as a temporary measure while seeing if any actual users remain of this code and any objections occur or anyone takes over maintaining the code. Otherwise in a later kernel release, the WiMAX support will be stripped out entirely from the mainline kernel.

The networking subsystem updates have landed for the in-development Linux 5.11 kernel.

The ACPI / power management and thermal pull requests were all sent out and merged this week for the ongoing Linux 5.11 development.

The power management and thermal updates to the Linux kernel for 5.11 aren't the most exciting in recent time but there still are a few additions worth calling attention to for what will be the first major kernel release of 2021.

Announced back in 2018 by the MIPI Alliance was the I3C Host Controller Interface (HCI) 1.0 specification whereby a common I3C HCI driver could support a range of multi-vendor sensors and other components relying on I3C.

MIPI wrote at the time, "MIPI I3C HCI defines a common set of capabilities for the host controller and the software interface, allowing for the building of class definitions based on a common set of features. The definition allows for vendor-specific extensions and optimizations."

Two years later, a MIPI I3C Host Controller Interface driver for Linux was published over the summer. But for this initial MIPI I3C HCI driver code there was limited testing due to HCI 1.0 hardware still being scarce even while HCI v1.1 and HCI v2.0 standards are already in the works.

This first week of the Linux 5.11 merge window continues to be very active with many of the kernel maintainers looking to land their changes ahead of the Christmas week where they are often taking time off work.

On top of other file-system improvements covered already for Linux 5.11 like a ton of Btrfs improvements and more F2FS encryption/casefolding/compression feature work, the Ceph, OverlayFS, and FUSE components all have some changes worth noting for Linux 5.11.

After undergoing review the past several months, Intel's Platform Monitoring Technology (PMT) is merged with the Linux 5.11 kernel.

Platform Monitoring Technology is hardware telemetry initially exposed with Tiger Lake hardware. This hardware telemetry is driven by customer demand and isn't about reporting hardware data back to Intel but rather collecting data on hardware internally within organizations and making it easier to manage. PMT currently supports Telemetry, Watcher (hardware sampling/tracing), and Crashlog components for various use-cases within organizations.

The KVM (Kernel-based Virtual Machine) changes were sent in today for the Linux 5.11 cycle.

Exciting of these KVM changes for what will be the first major Linux release of 2021 is supporting AMD SEV-ES on the host side. Linux has already seen bring-up work for AMD SEV-ES, the Secure Encrypted Virtualization Encrypted State functionality.

The main feature change for the XFS driver code in Linux 5.11 is adding a new "needs repair" feature flag. When the XFS code marks a file-system as needing repair, it will refuse to mount until the xfs_repair operation is run on it.

The XFS kernel code with Linux 5.11 and later will set the "needs repair" flag on a file-system when it's in a known state needing repair via the xfs_repair user-space utility. The updated xfs_repair utility will in turn clear that feature flag once the operation is complete. When needing a repair, xfs_admin should automatically run the repair operation so ideally the user/administrator will not need to issue the user-space command themselves.

While the Linux 5.11 merge window is only half-way through with prominent pull requests like the DRM / graphics driver updates already have been merged some of the testing has already begun at Phoronix of this new kernel. With the Radeon RX 6800 XT "RDNA 2" graphics continuing to mature, we are seeing slight uplift in some benchmarks when moving from Linux 5.10 stable to Linux 5.11 Git.

When running some quick Linux 5.10 vs. 5.11 Git tests with the RX 5700 XT and RX 6800 XT while running the same Mesa 21.0-devel build, in a few areas the Linux 5.11 performance is providing some measurable speed-ups.

For those running PostgreSQL database servers (and potentially similar workloads) on AMD EPYC servers, Linux 5.11 is bringing a very nice Christmas gift in the form of better performance for at least some 2P server configurations.

We are just half-way through the Linux 5.11 merge window with new code continuing to land, but already I've been running Linux 5.11 Git benchmarks on a number of systems in looking for any prominent performance improvements as a result of the new feature code or any signs of performance regressions... One area where I am seeing definite improvement with Linux 5.11 is on the PostgreSQL database server performance for at least AMD EPYC 2P servers.

Unlike the KVM additions, the Xen hypervisor for the Linux 5.11 merge window doesn't bring any new features but just security fixes for some new vulnerabilities.

The Xen changes for the Linux 5.11 merge window include just a set of patches for addressing two vulnerabilities (XSA-349 and XSA-350).

The Linux hardware monitoring "k10temp" driver is dropping support for reporting CPU voltage and current information for AMD Zen-based processors over lack of documentation for being able to properly support the functionality.

Earlier in 2020 this long-standing AMD hwmon temperature driver added support for core/SoC current and voltage reporting with Zen processors based on the work by the community and some best assumptions around the appropriate registers. But now that support is being dropped for lack of accuracy in some configurations and the possibility it might even damage the hardware.

The previously covered VirtIO memory (VirtIO-MEM) work on its Big Block Mode "BBM" will be landing with Linux 5.11.

The Red Hat led feature work was sent in today as part of Linux 5.11 virtIO updates. This is about overcoming a current limitation that the VirtIO-MEM driver can only support device block sizes up to the size of a single Linux memory block. The Big Block Mode allows for any device block size for any Linux virtual machine even if it's larger than a Linux memory block size of the running system.

Last week the Linux 5.11 power management updates were merged while on Tuesday some additional new material was merged, primarily around Intel's P-State CPU frequency scaling governor when running with the "Schedutil" governor that makes use of the kernel's scheduler utilization data.

With this pull request for P-State is a rework to its passive-mode fast-switch path so it avoids "running some workloads too fast" and will now side with better energy efficiency in select cases while still providing sufficient power for handling the current work. There is also another change to P-State to allow the guaranteed performance value for a given CPU to be increased after boot time.

Linux 5.11's Kconfig build configuration updates are bringing some long overdue improvements to its "Qconf" Qt toolkit user interface option for configuring Linux kernel builds.

Qt 5.0 was released eight years ago and earlier this month Qt 6.0 was christened. Most software was converted from Qt4 to Qt5 years ago but the Linux Kconfig code when firing up the Qt interface has retained support going back to Qt4. But for Linux 5.11 that is being remove to now focus on providing just a Qt5 interface (and presumably Qt 6, in due course) alongside the other Kconfig user interfaces from ncurses to GTK.

A few days ago I noted nice AMD EPYC performance improvements with PostgreSQL when running on Linux 5.11 compared to prior kernels. I've confirmed that for even more AMD EPYC servers now that the PostgreSQL uplift is there, but other workloads are unfortunately regressing for both Ryzen and EPYC. Here's the start of an exciting Christmas benchmarking adventure looking at this change with Linux 5.11...

[...]

What the heck?! A number of the workloads regressing are mostly user-space applications not interacting with the kernel much or even much in the way of I/O... But that ended up jiving with my hypothesis in regards to the Linux 5.11 improvement for PostgreSQL. It's about the CPU frequency invariance support in Linux 5.11 for AMD Zen 2 and newer.

Linux 5.10 as a Long-Term Support (LTS) kernel has been off to a rocky start after an immediate point release due to a RAID issue, some reporting AMDGPU problems, and also a staggering Btrfs performance regression hitting some users.

Making rounds this week was word of a "500~2000% performance regression" knocking Btrfs on Linux 5.10. For a simple test case like extracting a large .tar.zst file could go from taking just around 15 seconds to nearly five minutes or in other cases like from 5 seconds to over 30 seconds. The regression was bisected to a fundamental Btrfs change in Linux 5.10 and reproduced on bare metal while running Btrfs within a virtual machine didn't trigger the major slowdown.

It's not the Grinch in 2020 that stole Christmas, but the Schedutil CPU frequency scaling governor on the in-development Linux 5.11 kernel that is thrashing performance for AMD Zen 2 and newer. Distributions like Ubuntu, Fedora, and Manjaro are beginning to use CPUFreq Schedutil by default on newer kernels and thus leading to a very bad initial/out-of-the-box experience with the current behavior on the early Linux 5.11 code.

As noted a few days ago, PostgreSQL performance was looking very good on AMD EPYC 7002 servers, Upon further testing PostgreSQL was indeed improving nicely across multiple AMD servers but when looking at more workloads, it began looking like a very mixed bag with some workloads often regressing significantly on Linux 5.11 in its current state as we approach the end of the merge window this weekend.

Linux 5.11-rc1 is due for release later today and at least one regression has seen a quick resolution.

Intel's kernel test robot saw a ~1.6% performance regression in one of its scheduler benchmarks. The change causing the regression was trying to ensure only per-CPU KThreads run during hotplug.

The Linux 5.11 merge window has been open the past two weeks following the debut of Linux 5.10 but is set to close today. A lot of new features and exciting improvements were merged for Linux 5.11 although it is somewhat of a bumpy ride at the moment but should be buttoned up and ready for its stable release come February.

Linux 5.11 is very heavy on new feature work for this post-LTS cycle and what will be the first major stable kernel release of 2021. There are a ton of Intel and AMD changes, Syscall User Dispatch for helping some newer Windows games run under Wine, continued IO_uring advancements, many Btrfs file-system improvements, Lenovo contributed ThinkPad palm sensor detection support, and much more.

While the Linux 5.11 merge window has been over for one week where new features are normally added, a power management pull request sent in today for mainline is adding some tardy features including the Dynamic Thermal Power Management (DTPM) framework that in part is designed to help ensure users don't burn themselves with hot devices.

Rafael Wysocki sent in the New Year's Day power management updates for Linux 5.11. The changes include a crash fix for P-State, new C-states table for Intel Snow Ridge processors within the Intel Idle driver, and the new DTPM Framework.

For as great as Linux 5.11 is with its new features, there is also some prominent material that has yet to be upstreamed into the mainline kernel -- some of which is likely to hit in 2021 while other changes have less likely ambitions for mainline.

Yesterday I wrote about the DTPM framework being sent in for Linux 5.11 but ultimately Linus Torvalds has decided not to accept it out of the merge window.

As noted in the prior article, it was sent in a week past the Linux 5.11 merge window. The Dynamic Thermal Power Framework (DTPM) aims to be a higher-level thermal framework for cases like ensuring users don't burn themselves on hot devices and complying with legal requirements that case / exposed device temperatures do not exceed 45 degrees Celsius.