Re: [Devel] [PATCH hci-8.1 0/6] cgroup/dmem: backport fixes and apply a lower dmem.max at once

[Konstantin Khorenko <khorenko@virtuozzo.com>]

Applying with these fixes on my side:
- a Feature: line in every patch, "Feature: fix ms/cgroup" in patches
  1-4 and 6, "Feature: fix ms/drm" in patch 5;
- patch 6: rewrote the rebase note. It now has one paragraph on what is
  taken and why the old write changed nothing, one on what is left out
  and why, and a per-driver list of what a lower dmem.max does now. It
  also says that TTM evicts only when an allocation has no other
  placement, not at every next allocation.

On Mon, 02 Nov 2026 03:51:42 +0400, Mirian Shilakadze wrote:
> Patch 6 is an adapted 747c4bb450ad: a dmem.max below the current usage
> takes effect at once, as upstream, instead of being dropped while the
> write reports success. Its reclaim callback is left out, see the rebase
> note.

Thanks. One more upstream stable fix for the same code applies here:
e773798e14ac0 ("drm/amdgpu: Fix init ordering in
amdgpu_vram_mgr_init()"). amdgpu_vram_mgr_init() registers the dmem
region before the lists and the buddy allocator are set up, so a
failed registration at probe makes amdgpu_vram_mgr_fini() walk
uninitialised lists. I created a bug for it, please send the backport
as a separate patch:
https://virtuozzo.atlassian.net/browse/VSTOR-153702

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel