← Back to team overview

kernel-packages team mailing list archive

[Bug 1534647] Re: Collateral damage due to kernel configuration change enabling CONFIG_ZONE_DEVICE (Kernel 4.4 amd64)

 

** Tags added: bot-stop-nagging kernel-da-key

** Changed in: linux (Ubuntu)
   Importance: Undecided => Medium

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1534647

Title:
  Collateral damage due to kernel configuration change enabling
  CONFIG_ZONE_DEVICE (Kernel 4.4 amd64)

Status in linux package in Ubuntu:
  Confirmed

Bug description:
  Late in the kernel 4.4 RC (Release Candidate) cycle (between rc7 and
  rc8), Ubuntu implemented an amd64 kernel configuration change enabling
  CONFIG_ZONE_DEVICE.

  The related email: https://lists.ubuntu.com/archives/kernel-
  team/2016-January/067683.html

  The commit message does mention collateral damage: "In effect, this precludes devices that can only DMA from memory addresses below 16MB".
  Indeed we already have a complaint about a sound card that doesn't work as of kernel 4.4-rc8.

  If CONFIG_ZONE_DEVICE is enabled, then CONFIG_ZONE_DMA is forced to disabled, which in turn forces these devices, all are sound cards, and their derivative devices, to be disabled:
  Avance Logic ALS300/ALS300+
  ALi M5451 PCI Audio Controller
  Aztech AZF3328 / PCI168
  Emu10k1 (SB Live!, Audigy, E-mu APS)
  Emu10k1X (Dell OEM Version)
  ESS ES1938/1946/1969 (Solo-1)
  ESS ES1968/1978 (Maestro-1/2/2E)
  ICEnsemble ICE1712 (Envy24)
  ESS Allegro/Maestro3
  S3 SonicVibes
  Trident 4D-Wave DX/NX; SiS 7018.

  References: linux/mm/Kconfig; linux/sound/pci/Kconfig

  One test done (suggested by apw on IRC) was to remove the dependency of ZONE_DEVICE on !ZONE_DMA, but that causes:
  "#error ZONES_SHIFT -- too many zones configured adjust calculation", as apw predicted.
  In a mindless way, I tried to allow more zones in linux/include/linux/page-flags-layout.h, but that caused a bunch of "shifting by too many bits type errors. Anyway it seems that MAX_NR_ZONES is, at least partially, hard coded to 4 in some array definitions and such.

  Please do not ask me to do "apport-collect" for this bug report as it
  is not needed, nor relevant.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1534647/+subscriptions


References