mbox series

[0/4] Remove duplicated stats definitions for debugfs

Message ID 20210614025351.365284-1-jingzhangos@google.com (mailing list archive)
Headers show
Series Remove duplicated stats definitions for debugfs | expand

Message

Jing Zhang June 14, 2021, 2:53 a.m. UTC
This is a follow-up patchset to binary stats interface patchset as below:
https://lore.kernel.org/kvm/20210611124624.1404010-1-jingzhangos@google.com

This patchset contains a commit to fix some missing stats and add static
check to make sure we have the right number of stats descriptors and add an
'offset' field in stats descriptor to make sure the or order of stats
descriptors is not relevant to the order of stats in vm/vcpu stats
structure. This will totally avoid the possibility of missing stats and
mismatched stats definitions.

The binary stats interface defines stats in another array of descriptors,
while the original stats debugfs interface uses array of kvm_stats_debugfs
item. To remove the duplicated stats definition, this patchset would
utilize only the stats descriptors to provide stats information to debugfs
interface. This patchset adds a 'mode' flag to support the read/write mode
of stats, which can be used to indicate the file permission of debugfs
stats files. It removes the usage of kvm_stats_debugfs_item and all the
debugfs_entries defined in all archs.

The patch also fixes an issue that read only stats could be cleared in
global level, though not permitted in VM level in the original debugfs
code.

---

Jing Zhang (4):
  KVM: stats: Make sure no missing or mismatched binary stats definition
  KVM: stats: Use binary stats descriptors for debugfs interface
  KVM: stats: Update documentation supporting stats mode and offset
  KVM: selftests: Update binary stats test for stats mode

 Documentation/virt/kvm/api.rst                |  33 +-
 arch/arm64/kvm/guest.c                        |  32 +-
 arch/mips/kvm/mips.c                          |  99 ++----
 arch/powerpc/kvm/book3s.c                     | 101 ++----
 arch/powerpc/kvm/booke.c                      |  79 ++---
 arch/s390/kvm/kvm-s390.c                      | 307 ++++++------------
 arch/x86/kvm/x86.c                            | 120 +++----
 include/linux/kvm_host.h                      | 186 ++++++-----
 include/uapi/linux/kvm.h                      |  11 +-
 .../selftests/kvm/kvm_binary_stats_test.c     |  13 +-
 virt/kvm/kvm_main.c                           | 105 ++++--
 11 files changed, 477 insertions(+), 609 deletions(-)


base-commit: 39be2e28180a2e87af5fbb8d83643812e1a3b371

Comments

Paolo Bonzini June 14, 2021, 9:53 a.m. UTC | #1
On 14/06/21 04:53, Jing Zhang wrote:
> This is a follow-up patchset to binary stats interface patchset as below:
> https://lore.kernel.org/kvm/20210611124624.1404010-1-jingzhangos@google.com
> 
> This patchset contains a commit to fix some missing stats and add static
> check to make sure we have the right number of stats descriptors and add an
> 'offset' field in stats descriptor to make sure the or order of stats
> descriptors is not relevant to the order of stats in vm/vcpu stats
> structure. This will totally avoid the possibility of missing stats and
> mismatched stats definitions.
> 
> The binary stats interface defines stats in another array of descriptors,
> while the original stats debugfs interface uses array of kvm_stats_debugfs
> item. To remove the duplicated stats definition, this patchset would
> utilize only the stats descriptors to provide stats information to debugfs
> interface. This patchset adds a 'mode' flag to support the read/write mode
> of stats, which can be used to indicate the file permission of debugfs
> stats files. It removes the usage of kvm_stats_debugfs_item and all the
> debugfs_entries defined in all archs.
> 
> The patch also fixes an issue that read only stats could be cleared in
> global level, though not permitted in VM level in the original debugfs
> code.

Thanks for putting this together quickly.  I think we can drop the mode, 
see the suggestion in patch 2.

It's probably best to fold the v8 for the stats functionality, these 
patches (squashed appropriately) and Fuad's remars in a single posting. 
  But I'm quite confident that it can make it in 5.14.

Paolo
Jing Zhang June 14, 2021, 1:38 p.m. UTC | #2
On Mon, Jun 14, 2021 at 4:53 AM Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 14/06/21 04:53, Jing Zhang wrote:
> > This is a follow-up patchset to binary stats interface patchset as below:
> > https://lore.kernel.org/kvm/20210611124624.1404010-1-jingzhangos@google.com
> >
> > This patchset contains a commit to fix some missing stats and add static
> > check to make sure we have the right number of stats descriptors and add an
> > 'offset' field in stats descriptor to make sure the or order of stats
> > descriptors is not relevant to the order of stats in vm/vcpu stats
> > structure. This will totally avoid the possibility of missing stats and
> > mismatched stats definitions.
> >
> > The binary stats interface defines stats in another array of descriptors,
> > while the original stats debugfs interface uses array of kvm_stats_debugfs
> > item. To remove the duplicated stats definition, this patchset would
> > utilize only the stats descriptors to provide stats information to debugfs
> > interface. This patchset adds a 'mode' flag to support the read/write mode
> > of stats, which can be used to indicate the file permission of debugfs
> > stats files. It removes the usage of kvm_stats_debugfs_item and all the
> > debugfs_entries defined in all archs.
> >
> > The patch also fixes an issue that read only stats could be cleared in
> > global level, though not permitted in VM level in the original debugfs
> > code.
>
> Thanks for putting this together quickly.  I think we can drop the mode,
> see the suggestion in patch 2.
>
> It's probably best to fold the v8 for the stats functionality, these
> patches (squashed appropriately) and Fuad's remars in a single posting.
>   But I'm quite confident that it can make it in 5.14.
>
> Paolo
>
Thanks. Will drop the mode, squash stats functionality with v8 and fold them
in another single posting (v9).

Jing