[libvirt] [PATCH RESEND v2 0/4] re-introduce <acpi-hotplug-bridge>

Ani Sinha ani at anisinha.ca
Tue Mar 8 16:45:49 UTC 2022

Change log:
v2: rebased the patchset. Laine's response is appended at the end.

I am re-introducing the patchset for <acpi-hotplug-bridge> which got
reverted here few months back:


The reason for the reversal was that there seemed to be some
instability/issues around the use of the qemu commandline which this
patchset tries to support. In particular, some guest operating systems
did not like the way QEMU was trying to disable native hotplug on pcie
root ports. Subsequently, in QEMU 6.2, we have changed our mechanism
using which we disable native hotplug. As I understand, we do not have
any reported issues so far in 6.2 around this area. QEMU will enter a
soft feature freeze in the first week of march in prep for 7.0 release.
Libvirt is also entering a new release cycle phaze. Hence, I am
introducing this patchset early enough in the release cycles so that if
we do see any issues on the qemu side during the rc0, rc1 cycles and if
reversal of this patchset is again required, it can be done in time
before the next libvirt release end of March.

All the patches in this series had been previously reviewed. Some
subsequent fixes were made after my initial patches were pushed. I have
squashed all those fixes and consolidated them into four patches. I have
also updated the documentation to reflect the new changes from the QEMU
side and rebased my changes fixing the tests in the process.

What changed in QEMU post version 6.1 ?

We have made basically two major changes in QEMU. First is this change:

(1) commit 211afe5c69b597acf85fdd577eb497f5be1ffbd8
Author: Julia Suvorova <jusual at redhat.com>
Date:   Fri Nov 12 06:08:56 2021 -0500

    hw/i386/acpi-build: Deny control on PCIe Native Hot-plug in _OSC
    There are two ways to enable ACPI PCI Hot-plug:
            * Disable the Hot-plug Capable bit on PCIe slots.
    This was the first approach which led to regression [1-2], as
    I/O space for a port is allocated only when it is hot-pluggable,
    which is determined by HPC bit.
            * Leave the HPC bit on and disable PCIe Native Hot-plug in _OSC
    This removes the (future) ability of hot-plugging switches with PCIe
    Native hotplug since ACPI PCI Hot-plug only works with cold-plugged
    bridges. If the user wants to explicitely use this feature, they can
    disable ACPI PCI Hot-plug with:
            --global ICH9-LPC.acpi-pci-hotplug-with-bridge-support=off
    Change the bit in _OSC method so that the OS selects ACPI PCI Hot-plug
    instead of PCIe Native.
    [1] https://gitlab.com/qemu-project/qemu/-/issues/641
    [2] https://bugzilla.redhat.com/show_bug.cgi?id=2006409
    Signed-off-by: Julia Suvorova <jusual at redhat.com>
    Signed-off-by: Igor Mammedov <imammedo at redhat.com>
    Message-Id: <20211112110857.3116853-5-imammedo at redhat.com>
    Reviewed-by: Ani Sinha <ani at anisinha.ca>
    Reviewed-by: Michael S. Tsirkin <mst at redhat.com>
    Signed-off-by: Michael S. Tsirkin <mst at redhat.com>

The patch description says it all. Instead of masking out the HPC bit in
pcie slots, we keep them turned on. Instead, we do not advertize native
hotplug capability for PCIE using _OSC control method. See section
6.2.11 in ACPI spec 6.2. At the same time, we turn on ACPI hotplug for
these slots so now the guest OS can select ACPI hotplug instead.

The second change is introduction of a property with which we keep the
existing behavior for pc-q35-6.1 machines. This means HPC bit is masked
and ACPI hotplug is enabled by default for pcie root ports.
The QEMU commit is:

(2) commit c318bef76206c2ecb6016e8e68c4ac6ff9a4c8cb
Author: Julia Suvorova <jusual at redhat.com>
Date:   Fri Nov 12 06:08:54 2021 -0500

    hw/acpi/ich9: Add compat prop to keep HPC bit set for 6.1 machine type
    To solve issues [1-2] the Hot Plug Capable bit in PCIe Slots will be
    turned on, while the switch to ACPI Hot-plug will be done in the
    DSDT table.
    Introducing 'x-keep-native-hpc' property disables the HPC bit only
    in 6.1 and as a result keeps the forced 'reserve-io' on
    pcie-root-ports in 6.1 too.
    [1] https://gitlab.com/qemu-project/qemu/-/issues/641
    [2] https://bugzilla.redhat.com/show_bug.cgi?id=2006409
    Signed-off-by: Julia Suvorova <jusual at redhat.com>
    Signed-off-by: Igor Mammedov <imammedo at redhat.com>
    Message-Id: <20211112110857.3116853-3-imammedo at redhat.com>
    Reviewed-by: Michael S. Tsirkin <mst at redhat.com>
    Signed-off-by: Michael S. Tsirkin <mst at redhat.com>

Lastly, as a related side note, because from QEMU 6.2 onwards, we do not
mask out HPC bit in PCIE, the work done by this patch is no longer

(3) commit e2a6290aab578b2170c1f5909fa556385dc0d820
Author: Marcel Apfelbaum <marcel.apfelbaum at gmail.com>
Date:   Mon Aug 2 12:00:57 2021 +0300

    hw/pcie-root-port: Fix hotplug for PCI devices requiring IO
    Q35 has now ACPI hotplug enabled by default for PCI(e) devices.
    As opposed to native PCIe hotplug, guests like Fedora 34
    will not assign IO range to pcie-root-ports not supporting
    native hotplug, resulting into a regression.
    Reproduce by:
        qemu-bin -M q35 -device pcie-root-port,id=p1 -monitor stdio
        device_add e1000,bus=p1
    In the Guest OS the respective pcie-root-port will have the IO range
    Fix it by setting the "reserve-io" hint capability of the
    pcie-root-ports so the firmware will allocate the IO range instead.
    Acked-by: Igor Mammedov <imammedo at redhat.com>
    Signed-off-by: Marcel Apfelbaum <marcel at redhat.com>
    Message-Id: <20210802090057.1709775-1-marcel at redhat.com>
    Reviewed-by: Michael S. Tsirkin <mst at redhat.com>
    Signed-off-by: Michael S. Tsirkin <mst at redhat.com>

This is what commit (2) alludes to. In pc-q35-6.1 machines we do need
patch (3) since we mask out HPC bit from pcie ports.

I know this is convoluted mess. In fairness I am trying all I can in my
spare time to help from the QEMU side. I am determined to see this
patchset through into libvirt.


Laine's comments ...

My memory isn't completely clear, but I think there was also the issue
that the option claims to enable ACPI hotplug when set to on, but
instead what it actually does (in the Q35 case at least) is to enable
native PCI hotplug when set to off (without actually disabling ACPI
hotplug) and disable native PCI hotplug when set to on, or something
like that. This ends up leaving it up to the guest OS to decide which
type of hotplug to use, meaning its decision could override what's in
the libvirt config, thus confusing everyone. Again, I probably have the
details mixed up, but it was something like this.

I asked mst about this this morning, and he suggested something that
you've already done - Cc'ing the series to qemu-devel and the relevant
maintainers so we can have a discussion with all involved parties about
their opinions on whether we really should expose this existing option
in libvirt, or if we should instead have two new options that are more
orthogonal about enabling/disabling the two types of hotplug, so that
libvirt config can more accurately represent what is being presented to
the guest rather than a "best guess" of what we think the guest is going
to do with what is presented.

(Michael did also say that, with the current flurry of bug reports for
the QEMU rc's, this discusion may not happen until closer to release
when the bug reports die down. I know this doesn't mesh with your desire
to "push now to allow for testing" (which in general would be a good
thing if we were certain that we wanted the option like this and were
just expecting some minor bugs that could be fixed), but my opinion is
that 1) it's possible for anyone interested to test the functionality
using <qemu:commandline>, and 2) we should avoid turning libvirt git
into a revolving door of experiments. The only practical difference
between using <qemu:commandline> and having a dedicated option is that
the use of <qemu:commandline> causes the domain to be tainted, and the
XML is a bit more complicated. But since the people we're talking about
here will already have built their own libvirt binaries, the tainted
status of any guests is irrelevant and the extra complexity of using
<qemu:commandline> is probably trivial to them :-).

Ani Sinha (4):
  qemu: capablities: detect acpi-pci-hotplug-with-bridge-support
  conf: introduce support for acpi-bridge-hotplug feature
  qemu: command: add support for acpi-bridge-hotplug feature
  NEWS: document new acpi pci hotplug config option

 NEWS.rst                                      |  8 ++
 docs/formatdomain.rst                         | 32 +++++++
 docs/schemas/domaincommon.rng                 | 15 ++++
 src/conf/domain_conf.c                        | 89 ++++++++++++++++++-
 src/conf/domain_conf.h                        |  9 ++
 src/qemu/qemu_capabilities.c                  |  4 +
 src/qemu/qemu_capabilities.h                  |  3 +
 src/qemu/qemu_command.c                       | 19 ++++
 src/qemu/qemu_validate.c                      | 42 +++++++++
 .../caps_6.1.0.x86_64.xml                     |  1 +
 .../caps_6.2.0.x86_64.xml                     |  1 +
 .../caps_7.0.0.x86_64.xml                     |  1 +
 ...-hotplug-bridge-disable.aarch64-latest.err |  1 +
 .../aarch64-acpi-hotplug-bridge-disable.xml   | 13 +++
 ...-hotplug-bridge-disable.x86_64-latest.args | 35 ++++++++
 .../pc-i440fx-acpi-hotplug-bridge-disable.xml | 36 ++++++++
 .../pc-i440fx-acpi-hotplug-bridge-enable.xml  | 36 ++++++++
 ...pi-hotplug-bridge-disable.x86_64-6.0.0.err |  1 +
 ...-hotplug-bridge-disable.x86_64-latest.args | 38 ++++++++
 .../q35-acpi-hotplug-bridge-disable.xml       | 53 +++++++++++
 .../q35-acpi-hotplug-bridge-enable.xml        | 53 +++++++++++
 tests/qemuxml2argvtest.c                      |  7 ++
 ...i-hotplug-bridge-disable.x86_64-latest.xml |  1 +
 ...pi-hotplug-bridge-enable.x86_64-latest.xml |  1 +
 ...i-hotplug-bridge-disable.x86_64-latest.xml |  1 +
 ...pi-hotplug-bridge-enable.x86_64-latest.xml |  1 +
 tests/qemuxml2xmltest.c                       |  4 +
 27 files changed, 504 insertions(+), 1 deletion(-)
 create mode 100644 tests/qemuxml2argvdata/aarch64-acpi-hotplug-bridge-disable.aarch64-latest.err
 create mode 100644 tests/qemuxml2argvdata/aarch64-acpi-hotplug-bridge-disable.xml
 create mode 100644 tests/qemuxml2argvdata/pc-i440fx-acpi-hotplug-bridge-disable.x86_64-latest.args
 create mode 100644 tests/qemuxml2argvdata/pc-i440fx-acpi-hotplug-bridge-disable.xml
 create mode 100644 tests/qemuxml2argvdata/pc-i440fx-acpi-hotplug-bridge-enable.xml
 create mode 100644 tests/qemuxml2argvdata/q35-acpi-hotplug-bridge-disable.x86_64-6.0.0.err
 create mode 100644 tests/qemuxml2argvdata/q35-acpi-hotplug-bridge-disable.x86_64-latest.args
 create mode 100644 tests/qemuxml2argvdata/q35-acpi-hotplug-bridge-disable.xml
 create mode 100644 tests/qemuxml2argvdata/q35-acpi-hotplug-bridge-enable.xml
 create mode 120000 tests/qemuxml2xmloutdata/pc-i440fx-acpi-hotplug-bridge-disable.x86_64-latest.xml
 create mode 120000 tests/qemuxml2xmloutdata/pc-i440fx-acpi-hotplug-bridge-enable.x86_64-latest.xml
 create mode 120000 tests/qemuxml2xmloutdata/q35-acpi-hotplug-bridge-disable.x86_64-latest.xml
 create mode 120000 tests/qemuxml2xmloutdata/q35-acpi-hotplug-bridge-enable.x86_64-latest.xml


More information about the libvir-list mailing list