dpdk 18.11.4-1ubuntu1 ADT test failure with linux 5.4.0-6.7 => need DPDK 19.11 in Focal

Bug #1853195 reported by Seth Forshee on 2019-11-19
This bug affects 1 person
Affects Status Importance Assigned to Milestone
dpdk (Ubuntu)

Related branches

Seth Forshee (sforshee) on 2019-11-19
tags: added: kernel-adt-failure

Tested on ppc64 with
- vmlinux-5.3.0-23-generic
- vmlinux-5.4.0-6-generic (from ppa:canonical-kernel-team/unstable)

dpdk-igb-uio-dkms -> works fine (as usual)

 - works for 5.3
 - fails for 5.4

Full fail

Main issue:
In file included from ./include/linux/cache.h:5,
                 from ./include/linux/printk.h:9,
                 from ./include/linux/kernel.h:15,
                 from ./include/linux/list.h:9,
                 from ./include/linux/module.h:9,
                 from /usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb_main.c:13:
/usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb_main.c: In function ‘igb_xmit_frame_ring’:
/usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb_main.c:5513:51: error: ‘skb_frag_t’ {aka ‘struct bio_vec’} has no member named ‘size’
 5513 | count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
      | ^
./include/uapi/linux/kernel.h:13:40: note: in definition of macro ‘__KERNEL_DIV_ROUND_UP’
   13 | #define __KERNEL_DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
      | ^
/usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb_main.c:5513:12: note: in expansion of macro ‘TXD_USE_COUNT’
 5513 | count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
      | ^~~~~~~~~~~~~
  CC [M] /var/lib/dkms/dpdk-rte-kni/18.11.4/build/ethtool/igb/e1000_mac.o
  CC [M] /var/lib/dkms/dpdk-rte-kni/18.11.4/build/ethtool/igb/e1000_phy.o
In file included from /usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/e1000_hw.h:16,
                 from /usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/e1000_api.h:16,
                 from /usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb.h:50,
                 from /usr/src/dpdk-rte-kni-18.11.4/ethtool/igb/igb_main.c:39:

Changed in dpdk (Ubuntu):
status: New → Confirmed

As usual in the past, kni dkms was dropped upstream in 19.xx so we need to fix this on our own and provide a fix to the stable trees of 17.11.x and 18.11.x

Arr, I think I have that seen before.
This is similar to a part of bug 1848585.

To chances here:
a) 5.4 got more changes to SKBs we need to adapt to
b) the upstream release of the CVEs has dropped the changes needed

TBH - 19.11 is relased in a few days and there KNI will be dropped there.

So for Focal this will be done with the 19.11 update.
There is no big benefit in getting this resolved before, but quite some work around it.
Therefor I'm postponing this to 19.11.

If it hits also 17.11.x I'll need to fix it there and on the 17.11.x stable branch.
But that will be another bug at another time.

Changed in dpdk (Ubuntu):
status: Confirmed → Triaged
importance: Undecided → Medium
summary: - dpdk 18.11.4-1ubuntu1 ADT test failure with linux 5.4.0-6.7
+ dpdk 18.11.4-1ubuntu1 ADT test failure with linux 5.4.0-6.7 => need DPDK
+ 19.11 in Focal
Seth Forshee (sforshee) wrote :

@Christian: Where do we stand on getting this update for dpdk? We plan to put 5.4 into focal-proposed within the next week, so this is starting to become urgent for us. Thanks!

Hi Seth,
I completed DPDK 19.11 for Debian in git a few hours ago.
My Co-Maintainer will upload that to experimental these days and then we will hit NEW queue.
From there it is unclear how long things will take, but next week seems a bit early.

But if you need something urgent I'll ack anything in regard to this if you want to force-badtest this case.
At the end of the cycle the failing dkms package will be gone from the archive, so no one should care anymore even if it takes a bit to reach that state.

Seth Forshee (sforshee) wrote :

Thanks for the update. We'll see where we are with 5.4 once it hits proposed, if dpdk is the lone remaining test failure we can decide what to do then.

Seth Forshee (sforshee) wrote :

5.4 is in -proposed now. I don't expect everything to be ready for it to migrate until next week, but wanted to give you a heads up.

We might get it to -proposed (still unlikely) until then, but surely not to -release.
Please ignore the dpdk-kni test on the upload of kernel 5.4 and thanks for your work on coordinating all that with package owners every time!

Launchpad Janitor (janitor) wrote :
Download full text (4.0 KiB)

This bug was fixed in the package dpdk - 19.11-2ubuntu1

dpdk (19.11-2ubuntu1) focal; urgency=medium

  * Upload 19.11 to Focal
    - d/control: disable libbpf-dev build dependency while waiting for the
      kernel to build it for LP: 1836708 1826410.
    - d/control: remove librte-pmd-af-xdp20.0 until libbpf-dev is available

dpdk (19.11-2) experimental; urgency=medium

  * dh-missing: ignore examples directory, as they are already installed
    manually. (Closes: #947371)

dpdk (19.11-1) experimental; urgency=medium

  [ Christian Ehrhardt ]
  * New upstream release 19.11; for a full list of changes see:
    - Includes d/p/avoid-as-needed-as-it-causes-overlinking.patch: fix
      overlinking in software using the dpdk pkg-config (LP: #1841759)
    - d/control: bump versions to 19.11
    - d/*.symbols: bump for 19.11
    - rte-kni module support dropped upstream, remove former package
      dpdk-rte-kni-dkms (LP: #1853195)
    - d/p/fix-ppc64-mlx4-build.patch: drop patch now being upstream
    - new Major ABI is set to 20.0 by upstream
    - d/dpdk.lintian-overrides: it is ok that example shell scripts are
      not executable
    - d/control, d/*.symbols: new PMDs for qede pfe nitrox hns3 and
    - d/control, d/*.symbols: new rte libraries rib and fib
    - d/dpdk-dev.install: add new dpdk-test-sad
    - d/rules: global ABI_VERSION now tracked in a file by upstream
    - d/control: add new build-dependencies libisal-dev, libfdt-dev
      and libbpf-dev
    - d/control: drop no more built librte-pmd-ipn3ke20.0
    - d/control: add new PMDs of 19.11 af-xdp and isal
    - d/rules: add ability to handle fully experimental libs to be soname zero
    - d/control: experimental only libs now are on an soname 0.200 (bbdev, bpf,
      compressdev, fib, flow-classify, ipsec, rcu, rib, stack, telemetry)
    - d/control: let libdpdk-dev depend on the new build dependencies
  * fixup package reference in debian/librte-common-octeontx2-19.11.symbols
  * drop duplicate debian/update-control.py helper

dpdk (19.08-1~git1) UNRELEASED; urgency=medium

  [ Christian Ehrhardt ]
  * New upstream release 19.08; for a full list of changes see:
    - d/*.symbols: flat update for 19.08
    - d/control: update to new major version number
    - d/dpdk-dev.install: update new paths to dpdk-test
    - d/control, d/librte-pmd-qede18.11.symbols: drop pmd-qede no more built
      by default
    - d/control, d/*.symbols: new PMDs librte-pmd-octeontx2
      librte-pmd-bbdev-fpga-lte-fec librte-pmd-bbdev-turbo-sw librte-pmd-hinic
      librte-pmd-ice librte-pmd-ipn3ke librte-pmd-memif
    - d/control, d/*.symbols: renamed PMDs
      - librte-pmd-skeleton-rawdev -> librte-rawdev-skeleton
      - librte-pmd-dpaa2-cmdif -> librte-rawdev-dpaa2-cmdif
      - librte-pmd-dpaa2-qdma -> librte-rawdev-dpaa2-qdma
      - librte-pmd-avf -> librte-pmd-iavf
      - librte-pmd-ifpga-rawdev (somewhat indirectly replaced by
    - d/control: add new Homepage referen...


Changed in dpdk (Ubuntu):
status: Triaged → Fix Released

Finally, there it is :-)

@Seth - I'm not sure what to do about HWE kernels for Bionic.
As I mentioned the upstream choice was to remove the module.
Nobody is maintaining it anymore to match newer kernels and I don't see us taking over that, but it is fine with the non-HWE kernel.

OTOH there will be just "one more" HWE kernel for Bionic based on 5.4 then right?

Seth Forshee (sforshee) wrote :

Correct, 5.4 will be the last hwe kernel for bionic. Perhaps the building of that module could be made dependent on being built against a <5.4 kernel version?

Changed in dpdk (Ubuntu Bionic):
importance: Undecided → Low
status: New → Confirmed
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers