Skip to contentSkip to conversation

KVM/arm64 fixes for 6.7, part #2

GIT PULL Oliver Upton 4 participants 8 messages Dec 18, 2023 – Dec 22, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi Paolo,

Here's the second batch of fixes for 6.7. Please note that this pull is based on -rc4 instead of my first fixes tag as the KVM selftests breakage was introduced by one of my changes that went through the perf tree.

Please pull.

Signature · 41 lines
Thanks,
Oliver

The following changes since commit 33cc938e65a98f1d29d0a18403dbbee050dcad9a:

  Linux 6.7-rc4 (2023-12-03 18:52:56 +0900)

are available in the Git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm.git tags/kvmarm-fixes-6.7-2

for you to fetch changes up to 0c12e6c8267f831e491ee64ac6f216601cea3eee:

  KVM: selftests: Ensure sysreg-defs.h is generated at the expected path (2023-12-12 16:49:43 +0000)

----------------------------------------------------------------
KVM/arm64 fixes for 6.7, part #2

 - Ensure a vCPU's redistributor is unregistered from the MMIO bus
   if vCPU creation fails

 - Fix building KVM selftests for arm64 from the top-level Makefile

----------------------------------------------------------------
Marc Zyngier (5):
      KVM: arm64: vgic: Simplify kvm_vgic_destroy()
      KVM: arm64: vgic: Add a non-locking primitive for kvm_vgic_vcpu_destroy()
      KVM: arm64: vgic: Force vcpu vgic teardown on vcpu destroy
      KVM: arm64: vgic: Ensure that slots_lock is held in vgic_register_all_redist_iodevs()
      KVM: Convert comment into an assertion in kvm_io_bus_register_dev()

Oliver Upton (1):
      KVM: selftests: Ensure sysreg-defs.h is generated at the expected path

 arch/arm64/kvm/arm.c                 |  2 +-
 arch/arm64/kvm/vgic/vgic-init.c      | 47 ++++++++++++++++++++++--------------
 arch/arm64/kvm/vgic/vgic-mmio-v3.c   |  4 ++-
 arch/arm64/kvm/vgic/vgic.h           |  1 +
 tools/testing/selftests/kvm/Makefile | 26 ++++++++++++--------
 virt/kvm/kvm_main.c                  |  3 ++-
 6 files changed, 52 insertions(+), 31 deletions(-)
#1 lore raw
Quoted 4 lines · Oliver Upton

Here's the second batch of fixes for 6.7. Please note that this pull is based on -rc4 instead of my first fixes tag as the KVM selftests breakage was introduced by one of my changes that went through the perf tree.

Any news on this? The KVM selftests are still broken in mainline, pending-fixes and -next and the release is getting near.

Oliver, should your tree be in -next?

maintainer
#2 lore raw
Quoted 11 lines · Mark Brown
Quoted 4 lines · Oliver Upton

Here's the second batch of fixes for 6.7. Please note that this pull is based on -rc4 instead of my first fixes tag as the KVM selftests breakage was introduced by one of my changes that went through the perf tree.

Any news on this? The KVM selftests are still broken in mainline, pending-fixes and -next and the release is getting near.

Oliver, should your tree be in -next?

No, we don't have the KVM/arm64 fixes in -next.

And we have another two weeks to release, so ample amount of time until Paolo picks up the PR.

        M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#3 lore raw
Quoted 3 lines · Marc Zyngier
Quoted 1 line

Oliver, should your tree be in -next?

No, we don't have the KVM/arm64 fixes in -next.

I see it's not, I'm asking if it should be - given the latencies involved it seems like it'd be helpful for keeping -next working.

Quoted 2 lines

And we have another two weeks to release, so ample amount of time until Paolo picks up the PR.

Sure, but we do also have the holidays and also the fact that it's a build failure in a configuration used by some of the CIs means that we've got a bunch of testing that simply hasn't been happening for a couple of weeks now.

maintainer
#4 lore raw
Quoted 8 lines · Mark Brown
Quoted 3 lines · Marc Zyngier
Quoted 1 line

Oliver, should your tree be in -next?

No, we don't have the KVM/arm64 fixes in -next.

I see it's not, I'm asking if it should be - given the latencies involved it seems like it'd be helpful for keeping -next working.

This is on purpose. We use -next for, well, the next release, and not as a band-aid for some other purpose. If you think things don't get merged quickly enough, please take it with the person processing the PRs (Paolo).

Quoted 8 lines
Quoted 3 lines

And we have another two weeks to release, so ample amount of time until Paolo picks up the PR.

Sure, but we do also have the holidays and also the fact that it's a build failure in a configuration used by some of the CIs means that we've got a bunch of testing that simply hasn't been happening for a couple of weeks now.

Given that most of the KVM tests are usually more broken than the kernel itself, I'm not losing much sleep over it.

        M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#5 lore raw
Quoted 5 lines · Marc Zyngier
Quoted 3 lines · Mark Brown
Quoted 1 line · Marc Zyngier
Quoted 1 line

Oliver, should your tree be in -next?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

No, we don't have the KVM/arm64 fixes in -next.

I see it's not, I'm asking if it should be - given the latencies involved it seems like it'd be helpful for keeping -next working.

This is on purpose. We use -next for, well, the next release, and not as a band-aid for some other purpose. If you think things

Note that -next includes pending-fixes which is specifically for the purpose of getting coverage for fixes intended to go to mainline (indeed this issue was found and reported before the original problematic patch was sent to mainline, it's not clear to me what went wrong there). As well as the fixes getting coverage through being in -next itself there's a bunch of the CI that specifically looks at pending-fixes, trying to both prioritise keeping mainline stable and catch things like unintended dependencies on things only in -next.

There's also the fact that if mainline is broken and not somehow fixed in -next then that does disrupt the use of -next for testing things aimed at the next release.

Quoted 2 lines

don't get merged quickly enough, please take it with the person processing the PRs (Paolo).

He is on CC here. I'm not sure that it's specifically things not getting merged (well, modulo this one fixing an issue in mainline) - it's a totally reasonable approach to want to give things time to get reviewed and tested before they get merged, but if we're doing that then it seems sensible to take advantage of the coverage from having things in the integration trees. OTOH if things get merged very quickly then it's less clear since there's less time for the integration trees to do their thing. It feels like things aren't lining up well here.

Quoted 10 lines
Quoted 7 lines
Quoted 2 lines

And we have another two weeks to release, so ample amount of time until Paolo picks up the PR.

Sure, but we do also have the holidays and also the fact that it's a build failure in a configuration used by some of the CIs means that we've got a bunch of testing that simply hasn't been happening for a couple of weeks now.

Given that most of the KVM tests are usually more broken than the kernel itself, I'm not losing much sleep over it.

That's suprising to me, other than this release it doesn't match my experiences too well - there is a bit of glitchiness in one of the dirty_log tests on some platforms but otherwise the KVM selftests are generally rather stable on the systems where I pay attention to the results. They're certainly not a testsuite I expect to have to frequently worry about - we might wish for more coverage but that's something different to brokenness.

The build issues during this release cycle have been a bit of an exception.

maintainer
#6 lore raw
Quoted 10 lines · Mark Brown
Quoted 5 lines
Quoted 2 lines

I see it's not, I'm asking if it should be - given the latencies involved it seems like it'd be helpful for keeping -next working.

This is on purpose. We use -next for, well, the next release, and not as a band-aid for some other purpose. If you think things

Note that -next includes pending-fixes which is specifically for the purpose of getting coverage for fixes intended to go to mainline (indeed this issue was found and reported before the original problematic patch was sent to mainline, it's not clear to me what went wrong there).

Indeed most other KVM architectures have a tree included in linux-next's pending-fixes and kvm/master is included in there.

Knowing that KVM/ARM does not have a fixes tree included in linux-next might make me get those in kvm/master a bit faster, but then I'd let them stay in kvm/master, for a day or two of soaking in linux-next. It's never happened to me to send broken or conflicting pull requests after -rc1, as far as I remember, and it's indeed unlikely, but linux-next does provide a little bit of peace of mind.

Quoted 2 lines

He is on CC here. I'm not sure that it's specifically things not getting merged (well, modulo this one fixing an issue in mainline) -

I'm indeed not exactly a speed demon, but in this case I specifically wanted to make sure to include everything posted before Christmas, as this was likely going to be the last PR in the release.

I also wouldn't have minded a review or tested-by for
https://www.spinics.net/lists/kvm/msg335755.html :) but in the end I included it in the pull request anyway.

Paolo

maintainer
#7 lore raw
In reply to Oliver Upton
Quoted 9 lines · Oliver Upton

Hi Paolo,

Here's the second batch of fixes for 6.7. Please note that this pull is based on -rc4 instead of my first fixes tag as the KVM selftests breakage was introduced by one of my changes that went through the perf tree.

Please pull.

Pulled, thanks.

Paolo

Quoted 44 lines

--
Thanks,
Oliver

The following changes since commit 33cc938e65a98f1d29d0a18403dbbee050dcad9a:

Linux 6.7-rc4 (2023-12-03 18:52:56 +0900)

are available in the Git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm.git tags/kvmarm-fixes-6.7-2

for you to fetch changes up to 0c12e6c8267f831e491ee64ac6f216601cea3eee:

KVM: selftests: Ensure sysreg-defs.h is generated at the expected path (2023-12-12 16:49:43 +0000)

----------------------------------------------------------------
KVM/arm64 fixes for 6.7, part #2

- Ensure a vCPU's redistributor is unregistered from the MMIO bus if vCPU creation fails

- Fix building KVM selftests for arm64 from the top-level Makefile

----------------------------------------------------------------
Marc Zyngier (5):
KVM: arm64: vgic: Simplify kvm_vgic_destroy()
KVM: arm64: vgic: Add a non-locking primitive for kvm_vgic_vcpu_destroy()
KVM: arm64: vgic: Force vcpu vgic teardown on vcpu destroy
KVM: arm64: vgic: Ensure that slots_lock is held in vgic_register_all_redist_iodevs()
KVM: Convert comment into an assertion in kvm_io_bus_register_dev()

Oliver Upton (1):
KVM: selftests: Ensure sysreg-defs.h is generated at the expected path

 arch/arm64/kvm/arm.c                 |  2 +-
 arch/arm64/kvm/vgic/vgic-init.c      | 47 ++++++++++++++++++++++--------------
 arch/arm64/kvm/vgic/vgic-mmio-v3.c   |  4 ++-
 arch/arm64/kvm/vgic/vgic.h           |  1 +
 tools/testing/selftests/kvm/Makefile | 26 ++++++++++++--------
 virt/kvm/kvm_main.c                  |  3 ++-
 6 files changed, 52 insertions(+), 31 deletions(-)