News: 0001539942

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Linux 6.15 Fixes Intel Graphics Flickering Issue, Adds AMDGPU DMEM Cgroups Support

([Linux Kernel] 6 Hours Ago DRM Fixes)


With being just one week past the [1]Linux 6.15 merge window, a lot of fixes have been flowing into the mainline tree as is usual for the early stages of the kernel cycle. Merged overnight were a number of Direct Rendering Manager (DRM) fixes that include some notable alterations for the Intel and AMD Radeon kernel graphics drivers.

Ahead of the Linux 6.15-rc2 release on Sunday, there's now a big batch of DRM fixes merged to kernel codebase. On the AMDGPU driver side a notable change this week isn't a "fix" so much as squeezing in a feature. AMDGPU has landed its cgroups implementation around simple managing of discrete video memory (VRAM). In Linux 6.14 the infrastructure was added for [2]device memory "DMEM" cgroup support to allow limiting GPU vRAM and similar uses. Squeezing in post-6.15 merge window is the bit of integration needed for exposing the DMEM cgroups support for the AMDGPU driver with discrete AMD graphics/accelerators.

The AMDGPU driver also has landed some eDP fixes, pinning of VRAM DMA-BUFs if imports can handle peer-to-peer (P2P), workload profile fixes, IP discovery fixes, and various other fixes.

Over on the Intel Xe kernel graphics driver side this week's batch of fixes have added [3]the new 0xE211 device ID for Battlemage for that yet-to-be-announced product. As noted in the aforelinked article, the 0xE211 device ID was also recently added to the Mesa driver code.

The Intel Xe driver has also landed a fix to invalidate L3 read-only cachelines for geometry streams. This is said to fix significant flickering issues on at least Meteor Lake graphics. [4]This bug report outlines those Intel graphics flickering issues.

This week's DRM fixes pull also fixes GDDR+ECC memory type detection for Intel, queue reset fixes for the AMDKFD compute driver, suspend/resume support has been fixed for the Intel NPU (IVPU) accelerator driver, and various other random fixes.

[5]This pull outlines all of the kernel graphics/display drivers that were merged this week ahead of Linux 6.15-rc2.



[1] https://www.phoronix.com/search/Linux+6.15

[2] https://www.phoronix.com/news/DMEM-cgroup-vRAM-Control

[3] https://www.phoronix.com/news/Intel-Battlemage-Mesa-0xe211

[4] https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4460

[5] https://lore.kernel.org/lkml/CAPM=9tzvTUFpUeme8Aom4734ycaAK1b1gJFnjH3hHmg756UMGg@mail.gmail.com/



skeevy420

Couldn't we jury-rig the cat to act as an audio switch, and have it yell
at people to save their core images before logging them out? I'm sure
the cattle prod would be effective in this regard. In any case, a traverse
mounted iguana, while more perverted, gives better traction, not to mention
being easier to stake.