Just as 2025 is starting, we got a new Linux release in mid January, tagged as 6.13. In the spirit of holidays, Linus Torvalds even announced during 6.13-rc6 that he would be building and raffling a guitar pedal for a random kernel developer!
As usual, this release comes with a pack of exciting news done by the kernel community:
-
This release has two important improvements for task scheduling: lazy preemption and proxy execution. The goal with lazy preemption is to find a better balance between throughput and response time. A secondary goal is being able to make it the preferred non-realtime scheduling policy for most cases. Tasks that really need a reschedule in a hurry will use the older
TIF_NEED_RESCHED
flag. A preliminary work for proxy execution was merged, which will let us avoid priority-inversion scenarios when using real time tasks with deadline scheduling, for use cases such as Android. -
New important Rust abstractions arrived, such as VFS data structures and interfaces, and also abstractions for misc devices.
-
Lightweight guard pages: guard pages are used to raise a fatal signal when accessed. This feature had the drawback of having a heavy performance impact, but in this new release the flag
MADV_GUARD_INSTALL
was added for themadvise()
syscall, offering a lightweight way to guard pages.
To know more about the community improvements, check out the summary made by Kernel Newbies.
Now let’s highlight the contributions made by Igalians for this release.
Case-insensitive support for tmpfs
Case sensitivity has been a traditional difference between Linux distros and MS Windows, with the most popular filesystems been in opposite sides: while ext4 is case sensitive, NTFS is case insensitive. This difference proved to be challenging when Windows apps, mainly games, started to be a common use case for Linux distros (thanks to Wine!). For instance, games running through Steam’s Proton would expect that the path assets/player.png
and assets/PLAYER.PNG
would point to be the same file, but this is not the case in ext4. To avoid doing workarounds in userspace, ext4 has support for casefolding since Linux 5.2.
Now, tmpfs joins the group of filesystems with case-insensitive support. This is particularly useful for running games inside containers, like the combination of Wine + Flatpak. In such scenarios, the container shares a subset of the host filesystem with the application, mounting it using tmpfs. To keep the filesystem consistent, with the same expectations of the host filesystem about the mounted one, if the host filesystem is case-insensitive we can do the same thing for the container filesystem too. You can read more about the use case in the patchset cover letter.
While the container frameworks implement proper support for this feature, you can play with it and try it yourself:
$ mount -t tmpfs -o casefold fs_name /mytmpfs
$ cd /mytmpfs # case-sensitive by default, we still need to enable it
$ mkdir a
$ touch a; touch A
$ ls
A a
$ mkdir B; cd b
cd: The directory 'b' does not exist
$ # now let's create a case-insensitive dir
$ mkdir case_dir
$ chattr +F case_dir
$ cd case_dir
$ touch a; touch A
$ ls
a
$ mkdir B; cd b
$ pwd
$ /home/user/mytmpfs/case_dir/B
V3D Super Pages support
As part of Igalia’s effort for enhancing the graphics stack for Raspberry Pi, the V3D DRM driver now has support for Super Pages, improving performance and making memory usage more efficient for Raspberry Pi 4 and 5. Using Linux 6.13, the driver will enable the MMU to allocate not only the default 4KB pages, but also 64KB “Big Pages” and 1MB “Super Pages”.
To measure the difference that Super Pages made to the performance, a series of benchmarks where used, and the highlights are:
- +8.36% of FPS boost for Warzone 2100 in RPi4
- +3.62% of FPS boost for Quake 2 in RPi5
- 10% time reduction for the Mesa CI job
v3dv-rpi5-vk-full:arm64
- Aether SX2 emulator is more fluid to play
You can read a detailed post about this, with all benchmark results, in Maíra’s blog post, including a super cool PlayStation 2 emulation showcase!
New transparent_hugepage_shmem=
command-line parameter
Igalia contributed new kernel command-line parameters to improve the configuration of multi-size Transparent Huge Pages (mTHP) for shmem. These parameters, transparent_hugepage_shmem=
and thp_shmem=
, enable more flexible and fine-grained control over the allocation of huge pages when using shmem.
The transparent_hugepage_shmem=
parameter allows users to set a global default huge page allocation policy for the internal shmem mount. This is particularly valuable for DRM GPU drivers. Just as CPU architectures, GPUs can also take advantage of huge pages, but this is possible only if DRM GEM objects are backed by huge pages.
Since GEM uses shmem to allocate anonymous pageable memory, having control over the default huge page allocation policy allows for the exploration of huge pages use on GPUs that rely on GEM objects backed by shmem.
In addition, the thp_shmem=
parameter provides fine-grained control over the default huge page allocation policy for specific huge page sizes.
By configuring page sizes and policies of huge-page allocations for the internal shmem mount, these changes complement the V3D Super Pages feature, as we can now tailor the size of the huge pages to the needs of our GPUs.
DRM and AMDGPU improvements
As usual in Linux releases, this one collects a list of improvements made by our team in DRM and AMDGPU driver from the last cycle.
Cosmic (the desktop environment behind Pop! OS) users discovered some bugs in the AMD display driver regarding the handling of overlay planes. These issues were pre-existing and came to light with the introduction of cursor overlay mode. They were causing page faults and divide errors. We debugged the issue together with reporters and proposed a set of solutions that were ultimately accepted by AMD developers in time for this release.
In addition, we worked with AMD developers to migrate the driver-specific handling of EDID data to the DRM common code, using drm_edid opaque objects to avoid handling raw EDID data. The first phase was incorporated and allowed the inclusion of new functionality to get EDID from ACPI. However, some dependencies between the AMD the Linux-dependent and OS-agnostic components were left to be resolved in next iterations. It means that next steps will focus on removing the legacy way of handling this data.
Also in the AMD driver, we fixed one out of bounds memory write, fixed one warning on a boot regression and exposed special GPU memory pools via the fdinfo common DRM framework.
In the DRM scheduler code, we added some missing locking, removed a couple of re-lock cycles for slightly reduced command submission overheads and clarified the internal documentation.
In the common dma-fence code, we fixed one memory leak on the failure path and one significant runtime memory leak caused by incorrect merging of fences. The latter was found by the community and was manifesting itself as a system out of memory condition after a few hours of gameplay.
sched_ext
The sched_ext landed in kernel 6.12 to enable the efficient development of BPF-based custom schedulers. During the 6.13 development cycle, the sched_ext community has made efforts to harden the code to make it more reliable and clean up the BPF APIs and documentation for clarity.
Igalia has contributed to hardening the sched_ext core code. We fixed the incorrect use of the scheduler run queue lock, especially during initializing and finalizing the BPF scheduler. Also, we fixed the missing RCU lock protections when the sched_core selects a proper CPU for a task. Without these fixes, the sched_ext core, in the worst case, could crash or raise a kernel oops message.
Other Contributions & Fixes
syzkaller, a kernel fuzzer, has been an important instrument to find kernel bugs. With the help of KASAN, a memory error detector, and syzbot, numerous such bugs have been reported and fixed.
Igalians have contributed to such fixes around a lot of subsystems (like media, network, etc), helping reduce the number of open bugs.
Check the complete list of Igalia’s contributions for the 6.13 release
Authored (70)
André Almeida
- unicode: Fix utf8_load() error path
- MAINTAINERS: Add Unicode tree
- scripts/kernel-doc: Fix build time warnings
- libfs: Create the helper function generic_ci_validate_strict_name()
- ext4: Use generic_ci_validate_strict_name helper
- unicode: Export latest available UTF-8 version number
- unicode: Recreate utf8_parse_version()
- libfs: Export generic_ci_ dentry functions
- tmpfs: Add casefold lookup support
- tmpfs: Add flag FS_CASEFOLD_FL support for tmpfs dirs
- tmpfs: Expose filesystem features via sysfs
- docs: tmpfs: Add casefold options
- libfs: Fix kernel-doc warning in generic_ci_validate_strict_name
- tmpfs: Fix type for sysfs’ casefold attribute
- tmpfs: Initialize sysfs during tmpfs init
Changwoo Min
- sched_ext: Replace rq_lock() to raw_spin_rq_lock() in scx_ops_bypass()
- sched_ext: Clarify sched_ext_ops table for userland scheduler
- sched_ext: add a missing rcu_read_lock/unlock pair at scx_select_cpu_dfl()
- MAINTAINERS: add me as reviewer for sched_ext
Christian Gmeiner
Guilherme G. Piccoli
- Documentation: Improve crash_kexec_post_notifiers description
- wifi: rtlwifi: Drastically reduce the attempts to read efuse in case of failures
Maíra Canal
- drm/v3d: Address race-condition in MMU flush
- drm/v3d: Flush the MMU before we supply more memory to the binner
- drm/v3d: Fix return if scheduler initialization fails
- drm/gem: Create a drm_gem_object_init_with_mnt() function
- drm/v3d: Introduce gemfs
- drm/gem: Create shmem GEM object in a given mountpoint
- drm/v3d: Reduce the alignment of the node allocation
- drm/v3d: Support Big/Super Pages when writing out PTEs
- drm/v3d: Use gemfs/THP in BO creation if available
- drm/v3d: Add modparam for turning off Big/Super Pages
- drm/v3d: Expose Super Pages capability
- drm/vc4: Use
vc4_perfmon_find()
- MAINTAINERS: Add Maíra to VC4 reviewers
- mm: shmem: control THP support through the kernel command line
- mm: move
get_order_from_str()
to internal.h - mm: shmem: override mTHP shmem default with a kernel parameter
- mm: huge_memory: use strscpy() instead of strcpy()
- drm/v3d: Enable Performance Counters before clearing them
- drm/v3d: Ensure job pointer is set to NULL after job completion
Melissa Wen
- drm/amd/display: switch amdgpu_dm_connector to use struct drm_edid
- drm/amd/display: switch to setting physical address directly
- drm/amd/display: always call connector_update when parsing freesync_caps
- drm/amd/display: remove redundant freesync parser for DP
- drm/amd/display: add missing tracepoint event in DM atomic_commit_tail
- drm/amd/display: fix page fault due to max surface definition mismatch
- drm/amd/display: increase MAX_SURFACES to the value supported by hw
- drm/amd/display: fix divide error in DM plane scale calcs
Thadeu Lima de Souza Cascardo
- media: uvcvideo: Require entities to have a non-zero unique ID
- hfsplus: don’t query the device logical block size multiple times
- Bluetooth: btmtk: avoid UAF in btmtk_process_coredump
Tvrtko Ursulin
- drm/v3d: Appease lockdep while updating GPU stats
- drm/sched: Add locking to drm_sched_entity_modify_sched
- Documentation/gpu: Document the situation with unqualified drm-memory-
- drm/amdgpu: Drop unused fence argument from amdgpu_vmid_grab_used
- drm/amdgpu: Use drm_print_memory_stats helper from fdinfo
- drm/amdgpu: Drop impossible condition from amdgpu_job_prepare_job
- drm/amdgpu: Remove the while loop from amdgpu_job_prepare_job
- drm/sched: Optimise drm_sched_entity_push_job
- drm/sched: Stop setting current entity in FIFO mode
- drm/sched: Re-order struct drm_sched_rq members for clarity
- drm/sched: Re-group and rename the entity run-queue lock
- drm/sched: Further optimise drm_sched_entity_push_job
- drm/amd/pm: Vangogh: Fix kernel memory out of bounds write
- drm/amdgpu: Stop reporting special chip memory pools as CPU memory in fdinfo
- drm/amdgpu: Expose special on chip memory pools in fdinfo
- dma-fence: Fix reference leak on fence merge failure path
- dma-fence: Use kernel’s sort for merging fences
- workqueue: Do not warn when cancelling WQ_MEM_RECLAIM work from !WQ_MEM_RECLAIM worker
Reviewed (41)
André Almeida
- futex: Use atomic64_inc_return() in get_inode_sequence_number()
- futex: Use atomic64_try_cmpxchg_relaxed() in get_inode_sequence_number()
- mm: shmem: use signed int for version handling in casefold option
Christian Gmeiner
- drm/vc4: Use
vc4_perfmon_find()
- drm/etnaviv: Request pages from DMA32 zone on addressing_limited
- drm/etnaviv: Use unsigned type to count the number of pages
- drm/etnaviv: Use ‘unsigned’ type to count the number of pages
- drm/etnaviv: Drop the <linux/pm_runtime.h> header
- drm/etnaviv: Fix missing mutex_destroy()
- drm/etnaviv: hold GPU lock across perfmon sampling
- drm/etnaviv: assert GPU lock held in perfmon pipe_*_read functions
- drm/etnaviv: unconditionally enable debug registers
- drm/etnaviv: update hardware headers from rnndb
- drm/etnaviv: take current primitive into account when checking for hung GPU
- drm/etnaviv: always allocate 4K for kernel ringbuffers
- drm/etnaviv: flush shader L1 cache after user commandstream
Iago Toral Quiroga
- drm/v3d: Address race-condition in MMU flush
- drm/v3d: Flush the MMU before we supply more memory to the binner
- drm/v3d: Fix return if scheduler initialization fails
- drm/v3d: Introduce gemfs
- drm/v3d: Reduce the alignment of the node allocation
- drm/v3d: Expose Super Pages capability
- drm/v3d: Enable Performance Counters before clearing them
Jose Maria Casanova Crespo
Juan A. Suarez
Maíra Canal
- drm/v3d: Use v3d_perfmon_find()
- drm/vc4: Run default client setup for all variants.
- drm/vc4: Match drm_dev_enter and exit calls in vc4_hvs_lut_load
- drm/vc4: Match drm_dev_enter and exit calls in vc4_hvs_atomic_flush
- drm/vc4: Correct generation check in vc4_hvs_lut_load
- drm/vkms: Drop unnecessary call to drm_crtc_cleanup()
Tvrtko Ursulin
- drm/gem: Create a drm_gem_object_init_with_mnt() function
- drm/gem: Create shmem GEM object in a given mountpoint
- drm/v3d: Support Big/Super Pages when writing out PTEs
- drm/v3d: Use gemfs/THP in BO creation if available
- drm/v3d: Add modparam for turning off Big/Super Pages
- drm: add DRM_SET_CLIENT_NAME ioctl
- drm: use drm_file client_name in fdinfo
- drm/amdgpu: make drm-memory-* report resident memory
- dma-buf: fix dma_fence_array_signaled v4
- dma-buf: Fix __dma_buf_debugfs_list_del argument for !CONFIG_DEBUG_FS
Tested (1)
Christian Gmeiner
Acked (5)
Changwoo Min
- sched_ext: Rename
scx_bpf_dispatch[_vtime]()
toscx_bpf_dsq_insert[_vtime]()
- sched_ext: Rename
scx_bpf_consume()
toscx_bpf_dsq_move_to_local()
- sched_ext: Rename
scx_bpf_dispatch[_vtime]_from_dsq*()
->scx_bpf_dsq_move[_vtime]*()