pci: Rename hmp_pcie_aer_inject_error()'s local variable @err

I'd like to use @err for an Error *err.  Rename PCIEAERErr err to
aer_err.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221201121133.3813857-12-armbru@redhat.com>

pci: Inline do_pcie_aer_inject_error() into its only caller

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20221201121133.3813857-11-armbru@redhat.com>

pci: Move HMP command from hw/pci/pcie_aer.c to pci-hmp-cmds.c

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221201121133.3813857-10-armbru@redhat.com>

pci: Fix silent truncation of pcie_aer_inject_error argument

PCI AER error status is 32 bit.  The HMP command supports both
symbolic and numeric error status: anything that isn't a known
symbolic value is parsed as number with strtol().  Issues:

* Empty argument yields value zero.

* Range errors from strtol() are ignored, value is UINT32_MAX.

* Values not representable in uint32_t are silently truncated.

Fix to reject such input by switching to strtoui().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20221201121133.3813857-9-armbru@redhat.com>

pci: Move pcibus_dev_print() to pci-hmp-cmds.c

This method is for HMP command "info qtree".

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221201121133.3813857-8-armbru@redhat.com>

pci: Deduplicate get_class_desc()

pcibus_dev_print() contains a copy of get_class_desc().  Call the
function instead.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20221201121133.3813857-7-armbru@redhat.com>

pci: Build hw/pci/pci-hmp-cmds.c only when CONFIG_PCI

We compile pci-hmp-cmds.c always, but pci-qmp-cmds.c only when
CONFIG_PCI.  hw/pci/pci-stub.c keeps the linker happy when
!CONFIG_PCI.  Build pci-hmp-cmds.c that way, too.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20221201121133.3813857-6-armbru@redhat.com>

pci: Make query-pci stub consistent with the real one

QMP query-pci and HMP info pci can behave differently when there are
no PCI devices.  They can report nothing, like this:

    qemu-system-aarch64 -S -M spitz -display none -monitor stdio
    QEMU 7.1.91 monitor - type 'help' for more information
    (qemu) info pci

Or they can fail, like this:

    qemu-system-microblaze -M petalogix-s3adsp1800 -display none -monitor stdio
    QEMU 7.1.91 monitor - type 'help' for more information
    (qemu) info pci
    PCI devices not supported

They fail when none of the target's machines supports PCI, i.e. when
we're using qmp_query_pci() from hw/pci/pci-stub.c.

The error is not useful, and reporting nothing makes sense, so do that
in pci-stub.c, too.

Now qmp_query_pci() can't fail anymore.  Drop the dead error handling
from hmp_info_pci().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20221201121133.3813857-5-armbru@redhat.com>

pci: Move HMP commands from monitor/ to new hw/pci/pci-hmp-cmds.c

This moves these commands from MAINTAINERS section "Human
Monitor (HMP)" to "PCI".

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20221201121133.3813857-4-armbru@redhat.com>

pci: Move QMP commands to new hw/pci/pci-qmp-cmds.c

This moves these commands from MAINTAINERS section "QMP" to "PCI".

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20221201121133.3813857-3-armbru@redhat.com>
[Commit message improved]

pci: Clean up a few things checkpatch.pl would flag later on

Fix a few style violations so that checkpatch.pl won't complain when I
move this code.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221201121133.3813857-2-armbru@redhat.com>

Merge tag 'pull-loongarch-20221215' of https://gitlab.com/gaosong/qemu into staging

Add cfi01 pflash device

# gpg: Signature made Thu 15 Dec 2022 07:49:03 GMT
# gpg:                using RSA key B8FF1DA0D2FDCB2DA09C6C2C40A2FFF239263EDF
# gpg: Good signature from "Song Gao <m17746591750@163.com>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: B8FF 1DA0 D2FD CB2D A09C  6C2C 40A2 FFF2 3926 3EDF

* tag 'pull-loongarch-20221215' of https://gitlab.com/gaosong/qemu:
  hw/loongarch/virt: Add cfi01 pflash device

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'pull-target-arm-20221216' of https://git.linaro.org/people/pmaydell/qemu-arm into staging

reset refactoring queue:
 * remove uses of qdev_reset_all(), qbus_reset_all(), device_legacy_reset()
 * convert various devices to 3-phase reset, so we can remove their
   uses of device_class_set_parent_reset()

# gpg: Signature made Fri 16 Dec 2022 21:41:11 GMT
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [ultimate]
# gpg:                 aka "Peter Maydell <peter@archaic.org.uk>" [ultimate]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* tag 'pull-target-arm-20221216' of https://git.linaro.org/people/pmaydell/qemu-arm: (36 commits)
  hw/pci-host/pnv_phb3_msi: Convert TYPE_PHB3_MSI to 3-phase reset
  hw/intc/xics: Convert TYPE_ICS to 3-phase reset
  hw/intc/xics: Reset TYPE_ICS objects with device_cold_reset()
  pci: Convert child classes of TYPE_PCIE_ROOT_PORT to 3-phase reset
  pci: Convert TYPE_PCIE_ROOT_PORT to 3-phase reset
  hw/display/virtio-vga: Convert TYPE_VIRTIO_VGA_BASE to 3-phase reset
  hw/virtio: Convert TYPE_VIRTIO_PCI to 3-phase reset
  target/xtensa: Convert to 3-phase reset
  target/tricore: Convert to 3-phase reset
  target/sparc: Convert to 3-phase reset
  target/sh4: Convert to 3-phase reset
  target/rx: Convert to 3-phase reset
  target/riscv: Convert to 3-phase reset
  target/ppc: Convert to 3-phase reset
  target/openrisc: Convert to 3-phase reset
  target/nios2: Convert to 3-phase reset
  target/mips: Convert to 3-phase reset
  target/microblaze: Convert to 3-phase reset
  target/m68k: Convert to 3-phase reset
  target/loongarch: Convert to 3-phase reset
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/pci-host/pnv_phb3_msi: Convert TYPE_PHB3_MSI to 3-phase reset

Convert the TYPE_PHB3_MSI class to 3-phase reset, so we can
avoid using the device_class_set_parent_reset() function.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-8-peter.maydell@linaro.org

hw/intc/xics: Convert TYPE_ICS to 3-phase reset

Convert the TYPE_ICS class to 3-phase reset; this will allow us
to convert the TYPE_PHB3_MSI class which inherits from it.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-7-peter.maydell@linaro.org

hw/intc/xics: Reset TYPE_ICS objects with device_cold_reset()

The realize method for the TYPE_ICS class uses qemu_register_reset()
to register a reset handler, as a workaround for the fact that
currently objects which directly inherit from TYPE_DEVICE don't get
automatically reset.  However, the reset function directly calls
ics_reset(), which is the function that implements the legacy reset
method.  This means that only the parent class's data gets reset, and
a subclass which also needs to handle reset, like TYPE_PHB3_MSI, has
to register its own reset function.

Make the TYPE_ICS reset function call device_cold_reset() instead:
this will handle reset for both the parent class and the subclass,
and will work whether the classes are using legacy reset or 3-phase
reset. This allows us to remove the reset function that the subclass
currently has to set up.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-6-peter.maydell@linaro.org

pci: Convert child classes of TYPE_PCIE_ROOT_PORT to 3-phase reset

Convert the TYPE_CXL_ROOT_PORT and TYPE_PNV_PHB_ROOT_PORT classes to
3-phase reset, so they don't need to use the deprecated
device_class_set_parent_reset() function any more.

We have to do both in the same commit, because they keep the
parent_reset field in their common parent class's class struct.

Note that pnv_phb_root_port_class_init() was pointlessly setting
dc->reset twice, once by calling device_class_set_parent_reset()
and once directly.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-5-peter.maydell@linaro.org

pci: Convert TYPE_PCIE_ROOT_PORT to 3-phase reset

Convert the TYPE_PCIE_ROOT_PORT device to 3-phase reset; this is a
necessary precursor to converting any of its child classes.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-4-peter.maydell@linaro.org

hw/display/virtio-vga: Convert TYPE_VIRTIO_VGA_BASE to 3-phase reset

Convert the TYPE_VIRTIO_VGA_BASE class to 3-phase reset, so we
don't need to use device_class_set_parent_reset() any more.

Note that this is an abstract class itself; none of the subclasses
override its reset method.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-3-peter.maydell@linaro.org

hw/virtio: Convert TYPE_VIRTIO_PCI to 3-phase reset

Convert the TYPE_VIRTIO_PCI class to 3-phase reset.  This is
necessary so that we can convert the subclass TYPE_VIRTIO_VGA_BASE
also to 3-phase reset.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221125115240.3005559-2-peter.maydell@linaro.org

target/xtensa: Convert to 3-phase reset

Convert the xtensa CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-id: 20221124115023.2437291-20-peter.maydell@linaro.org

target/tricore: Convert to 3-phase reset

Convert the tricore CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-id: 20221124115023.2437291-19-peter.maydell@linaro.org

target/sparc: Convert to 3-phase reset

Convert the sparc CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-id: 20221124115023.2437291-18-peter.maydell@linaro.org

target/sh4: Convert to 3-phase reset

Convert the sh4 CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-id: 20221124115023.2437291-17-peter.maydell@linaro.org

target/rx: Convert to 3-phase reset

Convert the rx CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-id: 20221124115023.2437291-16-peter.maydell@linaro.org

target/riscv: Convert to 3-phase reset

Convert the riscv CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-id: 20221124115023.2437291-15-peter.maydell@linaro.org

target/ppc: Convert to 3-phase reset

Convert the ppc CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-id: 20221124115023.2437291-14-peter.maydell@linaro.org

target/openrisc: Convert to 3-phase reset

Convert the openrisc CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-13-peter.maydell@linaro.org

target/nios2: Convert to 3-phase reset

Convert the nios2 CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-12-peter.maydell@linaro.org

target/mips: Convert to 3-phase reset

Convert the mips CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-11-peter.maydell@linaro.org

target/microblaze: Convert to 3-phase reset

Convert the microblaze CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-10-peter.maydell@linaro.org

target/m68k: Convert to 3-phase reset

Convert the m68k CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-9-peter.maydell@linaro.org

target/loongarch: Convert to 3-phase reset

Convert the loongarch CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-8-peter.maydell@linaro.org

target/i386: Convert to 3-phase reset

Convert the i386 CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-7-peter.maydell@linaro.org

target/hexagon: Convert to 3-phase reset

Convert the hexagon CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Message-id: 20221124115023.2437291-6-peter.maydell@linaro.org

target/cris: Convert to 3-phase reset

Convert the cris CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Edgar E. Iglesias <edgar@zeroasic.com>
Message-id: 20221124115023.2437291-5-peter.maydell@linaro.org

target/avr: Convert to 3-phase reset

Convert the avr CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20221124115023.2437291-4-peter.maydell@linaro.org

target/arm: Convert to 3-phase reset

Convert the Arm CPU class to use 3-phase reset, so it doesn't
need to use device_class_set_parent_reset() any more.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-id: 20221124115023.2437291-3-peter.maydell@linaro.org

hw/core/cpu-common: Convert TYPE_CPU class to 3-phase reset

Convert the parent class TYPE_CPU to 3-phase reset. This
is a necessary prerequisite to converting the subclasses.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 20221124115023.2437291-2-peter.maydell@linaro.org

hw/misc: Convert TYPE_MOS6522 subclasses to 3-phase reset

Convert the various subclasses of TYPE_MOS6522 to 3-phase reset.
This removes some uses of device_class_set_parent_reset(), which we
would eventually like to be able to get rid of.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221110143459.3833425-3-peter.maydell@linaro.org

hw/misc/mos6522: Convert TYPE_MOS6522 to 3-phase reset

Convert the TYPE_MOS6522 parent class to use 3-phase reset.  This is
a prerequisite for converting its subclasses.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221110143459.3833425-2-peter.maydell@linaro.org

hw/input/ps2.c: Convert TYPE_PS2_{KBD, MOUSE}_DEVICE to 3-phase reset

Convert the child classes TYPE_PS2_KBD_DEVICE and
TYPE_PS2_MOUSE_DEVICE to the 3-phase reset system.  This allows us to
stop using the old device_class_set_parent_reset() function.

We don't need to register an 'exit' phase function for the
subclasses, because they have no work to do in that phase.  Passing
NULL to resettable_class_set_parent_phases() will result in the
parent class method being called for that phase, so we don't need to
register a function purely to chain to the parent 'exit' phase
function.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221109170009.3498451-3-peter.maydell@linaro.org

hw/input/ps2: Convert TYPE_PS2_DEVICE to 3-phase reset

Convert the parent class TYPE_PS2_DEVICE to 3-phase reset.  Note that
we need an 'exit' phase function as well as the usual 'hold' phase
function, because changing outbound IRQ line state is only permitted
in 'exit'.  (Strictly speaking it's not supposed to be done in a
legacy reset handler either, but you can often get away with it.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221109170009.3498451-2-peter.maydell@linaro.org

hw: Remove device_legacy_reset()

The device_legacy_reset() function is now not used anywhere, so we
can remove the implementation.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

qdev: Remove qdev_reset_all() and qbus_reset_all()

Remove the qdev_reset_all() and qbus_reset_all() functions, now we
have moved all the callers over to the new device_cold_reset() and
bus_cold_reset() functions.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Replace use of qdev_reset_all() with device_cold_reset()

The legacy function qdev_reset_all() performs a recursive reset,
starting from a qdev.  However, it does not permit any of the devices
in the tree to use three-phase reset, because device reset goes
through the device_legacy_reset() function that only calls the single
DeviceClass::reset method.

Switch to using the device_cold_reset() function instead.  This also
performs a recursive reset, where first the children are reset and
then finally the parent, but it uses the new (...in 2020...)
Resettable mechanism, which supports both the old style single-reset
method and also the new 3-phase reset handling.

This commit changes the five remaining uses of this function.

Commit created with:
 sed -i -e 's/qdev_reset_all/device_cold_reset/g' hw/i386/xen/xen_platform.c hw/input/adb.c hw/remote/vfio-user-obj.c hw/s390x/s390-virtio-ccw.c hw/usb/dev-uas.c

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/hyperv/vmbus: Use device_cold_reset() and bus_cold_reset()

In the vmbus code we currently use the legacy functions
qdev_reset_all() and qbus_reset_all().  These perform a recursive
reset, starting from either a qbus or a qdev.  However they do not
permit any of the devices in the tree to use three-phase reset,
because device reset goes through the device_legacy_reset() function
that only calls the single DeviceClass::reset method.

Switch to using the device_cold_reset() and bus_cold_reset()
functions.  These also perform a recursive reset, where first the
children are reset and then finally the parent, but they use the new
(...in 2020...) Resettable mechanism, which supports both the old
style single-reset method and also the new 3-phase reset handling.

This should be a no-behaviour-change commit which just reduces the
use of a deprecated API.

Commit created with:
  sed -i -e 's/qdev_reset_all/device_cold_reset/g;s/qbus_reset_all/bus_cold_reset/g' hw/hyperv/*.c

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

pci: Use device_cold_reset() and bus_cold_reset()

In the PCI subsystem we currently use the legacy function
qdev_reset_all() and qbus_reset_all().  These perform a recursive
reset, starting from either a qbus or a qdev.  However they do not
permit any of the devices in the tree to use three-phase reset,
because device reset goes through the device_legacy_reset() function
that only calls the single DeviceClass::reset method.

Switch to using the device_cold_reset() and bus_cold_reset()
functions.  These also perform a recursive reset, where first the
children are reset and then finally the parent, but they use the new
(...in 2020...) Resettable mechanism, which supports both the old
style single-reset method and also the new 3-phase reset handling.

This should be a no-behaviour-change commit which just reduces the
use of a deprecated API.

Commit created with:
 sed -i -e 's/qdev_reset_all/device_cold_reset/g;s/qbus_reset_all/bus_cold_reset/g' hw/pci/*.c

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/s390x/s390-pci-inst.c: Use device_cold_reset() to reset PCI devices

The semantic difference between the deprecated device_legacy_reset()
function and the newer device_cold_reset() function is that the new
function resets both the device itself and any qbuses it owns,
whereas the legacy function resets just the device itself and nothing
else.

In s390-pci-inst.c we use device_legacy_reset() to reset an
S390PCIBusDevice.  This device doesn't have any child qbuses, so the
functions do the same thing and we can stop using the deprecated one.

Reviewed-by: Matthew Rosato <mjrosato@linux.ibm.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'for-upstream' of https://repo.or.cz/qemu/kevin into staging

Block layer patches

- Code cleanups around block graph modification
- Simplify drain
- coroutine_fn correctness fixes, including splitting generated
  coroutine wrappers into co_wrapper (to be called only from
  non-coroutine context) and co_wrapper_mixed (both coroutine and
  non-coroutine context)
- Introduce a block graph rwlock

# gpg: Signature made Thu 15 Dec 2022 15:08:34 GMT
# gpg:                using RSA key DC3DEB159A9AF95D3D7456FE7F09B272C88F2FD6
# gpg:                issuer "kwolf@redhat.com"
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* tag 'for-upstream' of https://repo.or.cz/qemu/kevin: (50 commits)
  block: GRAPH_RDLOCK for functions only called by co_wrappers
  block: use co_wrapper_mixed_bdrv_rdlock in functions taking the rdlock
  block-coroutine-wrapper.py: introduce annotations that take the graph rdlock
  Mark assert_bdrv_graph_readable/writable() GRAPH_RD/WRLOCK
  graph-lock: TSA annotations for lock/unlock functions
  block: assert that graph read and writes are performed correctly
  block: remove unnecessary assert_bdrv_graph_writable()
  block: wrlock in bdrv_replace_child_noperm
  block: Fix locking in external_snapshot_prepare()
  test-bdrv-drain: Fix incorrrect drain assumptions
  clang-tsa: Add macros for shared locks
  clang-tsa: Add TSA_ASSERT() macro
  Import clang-tsa.h
  async: Register/unregister aiocontext in graph lock list
  graph-lock: Implement guard macros
  graph-lock: Introduce a lock to protect block graph operations
  block: Factor out bdrv_drain_all_begin_nopoll()
  block/dirty-bitmap: convert coroutine-only functions to co_wrapper
  block: convert bdrv_create to co_wrapper
  block-coroutine-wrapper.py: support also basic return types
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'pull-request-2022-12-15' of https://gitlab.com/thuth/qemu into staging

* s390x PCI fixes and improvements (for the ISM device)
* Fix emulated MVCP and MVCS s390x instructions
* Clean-ups for the e1000e qtest
* Enable qtests on Windows
* Update FreeBSD CI to version 12.4
* Check --disable-tcg for ppc64 in the CI
* Improve scripts/make-releases a little bit
* Many other misc small clean-ups and fixes here and there

# gpg: Signature made Thu 15 Dec 2022 15:05:44 GMT
# gpg:                using RSA key 27B88847EEE0250118F3EAB92ED9D774FE702DB5
# gpg:                issuer "thuth@redhat.com"
# gpg: Good signature from "Thomas Huth <th.huth@gmx.de>" [full]
# gpg:                 aka "Thomas Huth <thuth@redhat.com>" [full]
# gpg:                 aka "Thomas Huth <huth@tuxfamily.org>" [full]
# gpg:                 aka "Thomas Huth <th.huth@posteo.de>" [unknown]
# Primary key fingerprint: 27B8 8847 EEE0 2501 18F3  EAB9 2ED9 D774 FE70 2DB5

* tag 'pull-request-2022-12-15' of https://gitlab.com/thuth/qemu: (23 commits)
  tests/qtest/vhost-user-blk-test: don't abort all qtests on missing envar
  .gitlab/issue_templates: Move suggestions into comments
  gitlab-ci: Check building ppc64 without TCG
  FreeBSD: Upgrade to 12.4 release
  tests/qtest: Enable qtest build on Windows
  .gitlab-ci.d/windows.yml: Exclude qTests from 64-bit CI job for now
  .gitlab-ci.d/windows.yml: Keep 64-bit and 32-bit build scripts consistent
  .gitlab-ci.d/windows.yml: Unify the prerequisite packages
  tests/qtest/libqos/e1000e: Correctly group register accesses
  tests/qtest/e1000e-test: De-duplicate constants
  tests/qtest/libqos/e1000e: Remove "other" interrupts
  hw: Include the VMWare devices only in the x86 targets
  MAINTAINERS: Add documentation files to the corresponding sections
  util/oslib-win32: Remove obsolete reference to g_poll code
  util/qemu-config: Fix "query-command-line-options" to provide the right values
  scripts/make-release: Only clone single branches to speed up the script
  scripts/make-release: Add a simple help text for the script
  monitor/misc: Remove superfluous include statements
  target/s390x: The MVCP and MVCS instructions are not privileged
  target/s390x/tcg/mem_helper: Test the right bits in psw_key_valid()
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'pull-target-arm-20221215-1' of https://git.linaro.org/people/pmaydell/qemu-arm into staging

target-arm queue:
 * hw/arm/virt: Add properties to allow more granular
   configuration of use of highmem space
 * target/arm: Add Cortex-A55 CPU
 * hw/intc/arm_gicv3: Fix GICD_TYPER ITLinesNumber advertisement
 * Implement FEAT_EVT
 * Some 3-phase-reset conversions for Arm GIC, SMMU
 * hw/arm/boot: set initrd with #address-cells type in fdt
 * hw/misc: Move some arm-related files from specific_ss into softmmu_ss
 * Restrict arm_cpu_exec_interrupt() to TCG accelerator

# gpg: Signature made Thu 15 Dec 2022 17:38:36 GMT
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [ultimate]
# gpg:                 aka "Peter Maydell <peter@archaic.org.uk>" [ultimate]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* tag 'pull-target-arm-20221215-1' of https://git.linaro.org/people/pmaydell/qemu-arm: (28 commits)
  target/arm: Restrict arm_cpu_exec_interrupt() to TCG accelerator
  hw/misc: Move some arm-related files from specific_ss into softmmu_ss
  hw/arm/boot: set initrd with #address-cells type in fdt
  hw/intc: Convert TYPE_KVM_ARM_ITS to 3-phase reset
  hw/intc: Convert TYPE_ARM_GICV3_ITS to 3-phase reset
  hw/intc: Convert TYPE_ARM_GICV3_ITS_COMMON to 3-phase reset
  hw/intc: Convert TYPE_KVM_ARM_GICV3 to 3-phase reset
  hw/intc: Convert TYPE_ARM_GICV3_COMMON to 3-phase reset
  hw/intc: Convert TYPE_ARM_GIC_KVM to 3-phase reset
  hw/intc: Convert TYPE_ARM_GIC_COMMON to 3-phase reset
  hw/arm: Convert TYPE_ARM_SMMUV3 to 3-phase reset
  hw/arm: Convert TYPE_ARM_SMMU to 3-phase reset
  target/arm: Report FEAT_EVT for TCG '-cpu max'
  target/arm: Implement HCR_EL2.TID4 traps
  target/arm: Implement HCR_EL2.TICAB,TOCU traps
  target/arm: Implement HCR_EL2.TTLBOS traps
  target/arm: Implement HCR_EL2.TTLBIS traps
  target/arm: Allow relevant HCR bits to be written for FEAT_EVT
  hw/intc/arm_gicv3: Fix GICD_TYPER ITLinesNumber advertisement
  target/arm: Add Cortex-A55 CPU
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Restrict arm_cpu_exec_interrupt() to TCG accelerator

When building with --disable-tcg on Darwin we get:

  target/arm/cpu.c:725:16: error: incomplete definition of type 'struct TCGCPUOps'
    cc->tcg_ops->do_interrupt(cs);
    ~~~~~~~~~~~^

Commit 083afd18a9 ("target/arm: Restrict cpu_exec_interrupt()
handler to sysemu") limited this block to system emulation,
but neglected to also limit it to TCG.

Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Message-id: 20221209110823.59495-1-philmd@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/misc: Move some arm-related files from specific_ss into softmmu_ss

The header target/arm/kvm-consts.h checks CONFIG_KVM which is marked as
poisoned in common code, so the files that include this header have to
be added to specific_ss and recompiled for each, qemu-system-arm and
qemu-system-aarch64. However, since the kvm headers are only optionally
used in kvm-constants.h for some sanity checks, we can additionally
check the NEED_CPU_H macro first to avoid the poisoned CONFIG_KVM macro,
so kvm-constants.h can also be used from "common" files (without the
sanity checks - which should be OK since they are still done from other
target-specific files instead). This way, and by adjusting some other
include statements in the related files here and there, we can move some
files from specific_ss into softmmu_ss, so that they only need to be
compiled once during the build process.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221202154023.293614-1-thuth@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

block: GRAPH_RDLOCK for functions only called by co_wrappers

The generated coroutine wrappers already take care to take the lock in
the non-coroutine path, and assume that the lock is already taken in the
coroutine path.

The only thing we need to do for the wrapped function is adding the
GRAPH_RDLOCK annotation. Doing so also allows us to mark the
corresponding callbacks in BlockDriver as GRAPH_RDLOCK_PTR.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-19-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: use co_wrapper_mixed_bdrv_rdlock in functions taking the rdlock

Take the rdlock already, before we add the assertions.

All these functions either read the graph recursively, or call
BlockDriver callbacks that will eventually need to be protected by the
graph rdlock.

Do it now to all functions together, because many of these recursively
call each other.

For example, bdrv_co_truncate calls BlockDriver->bdrv_co_truncate, and
some driver callbacks implement their own .bdrv_co_truncate by calling
bdrv_flush inside. So if bdrv_flush asserts but bdrv_truncate does not
take the rdlock yet, the assertion will always fail.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-18-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-coroutine-wrapper.py: introduce annotations that take the graph rdlock

Add co_wrapper_bdrv_rdlock and co_wrapper_mixed_bdrv_rdlock option to
the block-coroutine-wrapper.py script.

This "_bdrv_rdlock" option takes and releases the graph rdlock when a
coroutine function is created.

This means that when used together with "_mixed", the function marked
with co_wrapper_mixed_bdrv_rdlock will support both coroutine and
non-coroutine case, and in the latter case it will create a coroutine
that takes and releases the rdlock. When called from a coroutine, the
caller must already hold the graph lock.

Example:
void co_wrapper_mixed_bdrv_rdlock bdrv_f1();

Becomes

static void bdrv_co_enter_f1()
{
    bdrv_graph_co_rdlock();
    bdrv_co_function();
    bdrv_graph_co_rdunlock();
}

void bdrv_f1()
{
    if (qemu_in_coroutine) {
        assume_graph_lock();
        bdrv_co_function();
    } else {
        qemu_co_enter(bdrv_co_enter_f1);
        ...
    }
}

When used alone, the function will not work in coroutine context, and
when called in non-coroutine context it will create a new coroutine that
takes care of taking and releasing the rdlock automatically.

Example:
void co_wrapper_bdrv_rdlock bdrv_f1();

Becomes

static void bdrv_co_enter_f1()
{
    bdrv_graph_co_rdlock();
    bdrv_co_function();
    bdrv_graph_co_rdunlock();
}

void bdrv_f1()
{
    assert(!qemu_in_coroutine());
    qemu_co_enter(bdrv_co_enter_f1);
    ...
}

About their usage:
- co_wrapper does not take the rdlock, so it can be used also outside
  the block layer.
- co_wrapper_mixed will be used by many blk_* functions, since the
  coroutine function needs to call blk_wait_while_drained() and
  the rdlock *must* be taken afterwards, otherwise it's a deadlock.
  In the future this annotation will go away, and blk_* will use
  co_wrapper directly.
- co_wrapper_bdrv_rdlock will be used by BlockDriver callbacks, ideally
  by all of them in the future.
- co_wrapper_mixed_bdrv_rdlock will be used by the remaining functions
  that are still called by coroutine and non-coroutine context. In the
  future this annotation will go away, as we will split such mixed
  functions.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-17-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Mark assert_bdrv_graph_readable/writable() GRAPH_RD/WRLOCK

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-16-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

graph-lock: TSA annotations for lock/unlock functions

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-15-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: assert that graph read and writes are performed correctly

Remove the old assert_bdrv_graph_writable, and replace it with
the new version using graph-lock API.

See the function documentation for more information.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-14-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: remove unnecessary assert_bdrv_graph_writable()

We don't protect bdrv->aio_context with the graph rwlock,
so these assertions are not needed

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-13-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: wrlock in bdrv_replace_child_noperm

Protect the main function where graph is modified.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-12-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fix locking in external_snapshot_prepare()

bdrv_img_create() polls internally (when calling bdrv_create(), which is
a co_wrapper), so it can't be called while holding the lock of any
AioContext except the current one without causing deadlocks. Drop the
lock around the call in external_snapshot_prepare().

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-11-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

test-bdrv-drain: Fix incorrrect drain assumptions

The test case assumes that a drain only happens in one specific place
where it drains explicitly. This assumption happened to hold true until
now, but block layer functions may drain interally (any graph
modifications are going to do that through bdrv_graph_wrlock()), so this
is incorrect. Make sure that the test code in .drained_begin only runs
where we actually want it to run.

When scheduling a BH from .drained_begin, we also need to increase the
in_flight counter to make sure that the operation is actually completed
in time before the node that it works on goes away.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-10-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

clang-tsa: Add macros for shared locks

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-8-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

clang-tsa: Add TSA_ASSERT() macro

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-7-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Import clang-tsa.h

This defines macros that allow clang to perform Thread Safety Analysis
based on function and variable annotations that specify the locking
rules. On non-clang compilers, the annotations are ignored.

Imported tsa.h from the original repository with the pthread_mutex_t
wrapper removed:

https://github.com/jhi/clang-thread-safety-analysis-for-c.git

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-6-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

async: Register/unregister aiocontext in graph lock list

Add/remove the AioContext in aio_context_list in graph-lock.c when it is
created/destroyed. This allows using the graph locking operations from
this AioContext.

In order to allow linking util/async.c with binaries that don't include
the block layer, introduce stubs for (un)register_aiocontext().

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-5-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

graph-lock: Implement guard macros

Similar to the implementation in lockable.h, implement macros to
automatically take and release the rdlock.

Create the empty GraphLockable and GraphLockableMainloop structs only to
use it as a type for G_DEFINE_AUTOPTR_CLEANUP_FUNC.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-4-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

graph-lock: Introduce a lock to protect block graph operations

Block layer graph operations are always run under BQL in the main loop.
This is proved by the assertion qemu_in_main_thread() and its wrapper
macro GLOBAL_STATE_CODE.

However, there are also concurrent coroutines running in other iothreads
that always try to traverse the graph. Currently this is protected
(among various other things) by the AioContext lock, but once this is
removed, we need to make sure that reads do not happen while modifying
the graph.

We distinguish between writer (main loop, under BQL) that modifies the
graph, and readers (all other coroutines running in various AioContext),
that go through the graph edges, reading ->parents and->children.

The writer (main loop) has "exclusive" access, so it first waits for any
current read to finish, and then prevents incoming ones from entering
while it has the exclusive access.

The readers (coroutines in multiple AioContext) are free to access the
graph as long the writer is not modifying the graph. In case it is, they
go in a CoQueue and sleep until the writer is done.

If a coroutine changes AioContext, the counter in the original and new
AioContext are left intact, since the writer does not care where the
reader is, but only if there is one.

As a result, some AioContexts might have a negative reader count, to
balance the positive count of the AioContext that took the lock.  This
also means that when an AioContext is deleted it may have a nonzero
reader count. In that case we transfer the count to a global shared
counter so that the writer is always aware of all readers.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-3-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Factor out bdrv_drain_all_begin_nopoll()

Provide a separate function that just quiesces the users of a node to
prevent new requests from coming in, but without waiting for the already
in-flight I/O to complete.

This function can be used in contexts where polling is not allowed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221207131838.239125-2-kwolf@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/dirty-bitmap: convert coroutine-only functions to co_wrapper

bdrv_can_store_new_dirty_bitmap and bdrv_remove_persistent_dirty_bitmap
check if they are running in a coroutine, directly calling the
coroutine callback if it's the case.
Except that no coroutine calls such functions, therefore that check
can be removed, and function creation can be offloaded to
c_w.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-15-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: convert bdrv_create to co_wrapper

This function is never called in coroutine context, therefore
instead of manually creating a new coroutine, delegate it to the
block-coroutine-wrapper script, defining it as co_wrapper.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-14-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-coroutine-wrapper.py: support also basic return types

Extend the regex to cover also return type, pointers included.
This implies that the value returned by the function cannot be
a simple "int" anymore, but the custom return type.
Therefore remove poll_state->ret and instead use a per-function
custom "ret" field.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-13-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-coroutine-wrapper.py: support functions without bs arg

Right now, we take the first parameter of the function to get the
BlockDriverState to pass to bdrv_poll_co(), that internally calls
functions that figure in which aiocontext the coroutine should run.

However, it is useless to pass a bs just to get its own AioContext,
so instead pass it directly, and default to the main loop if no
BlockDriverState is passed as parameter.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-12-eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-coroutine-wrapper.py: introduce co_wrapper

This new annotation starts just a function wrapper that creates
a new coroutine. It assumes the caller is not a coroutine.
It will be the default annotation to be used in the future.

This is much better as c_w_mixed, because it is clear if the caller
is a coroutine or not, and provides the advantage of automating
the code creation. In the future all c_w_mixed functions will be
substituted by co_wrapper.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-11-eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: rename generated_co_wrapper in co_wrapper_mixed

In preparation to the incoming new function specifiers,
rename g_c_w with a more meaningful name and document it.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-10-eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: bdrv_create_file is a coroutine_fn

It is always called in coroutine_fn callbacks, therefore
it can directly call bdrv_co_create().

Rename it to bdrv_co_create_file too.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-9-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: distinguish between bdrv_create running in coroutine and not

Call two different functions depending on whether bdrv_create
is in coroutine or not, following the same pattern as
generated_co_wrapper functions.

This allows to also call the coroutine function directly,
without using CreateCo or relying in bdrv_create().

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-8-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: avoid duplicating filename string in bdrv_create

We know that the string will stay around until the function
returns, and the parameter of drv->bdrv_co_create_opts is const char*,
so it must not be modified either.

Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-7-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/vmdk: add coroutine_fn annotations

These functions end up calling bdrv_create() implemented as generated_co_wrapper
functions.
In addition, they also happen to be always called in coroutine context,
meaning all callers are coroutine_fn.
This means that the g_c_w function will enter the qemu_in_coroutine()
case and eventually suspend (or in other words call qemu_coroutine_yield()).
Therefore we can mark such functions coroutine_fn too.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-6-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-backend: replace bdrv_*_above with blk_*_above

Avoid mixing bdrv_* functions with blk_*, so create blk_* counterparts
for bdrv_block_status_above and bdrv_is_allocated_above.

Note that since blk_co_block_status_above only calls the g_c_w function
bdrv_common_block_status_above and is marked as coroutine_fn, call
directly bdrv_co_common_block_status_above() to avoid using a g_c_w.
Same applies to blk_co_is_allocated_above.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-5-eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd/server.c: add coroutine_fn annotations

These functions end up calling bdrv_*() implemented as generated_co_wrapper
functions.
In addition, they also happen to be always called in coroutine context,
meaning all callers are coroutine_fn.
This means that the g_c_w function will enter the qemu_in_coroutine()
case and eventually suspend (or in other words call qemu_coroutine_yield()).
Therefore we can mark such functions coroutine_fn too.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-4-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-copy: add coroutine_fn annotations

These functions end up calling bdrv_common_block_status_above(), a
generated_co_wrapper function.
In addition, they also happen to be always called in coroutine context,
meaning all callers are coroutine_fn.
This means that the g_c_w function will enter the qemu_in_coroutine()
case and eventually suspend (or in other words call qemu_coroutine_yield()).
Therefore we can mark such functions coroutine_fn too.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-3-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block-io: introduce coroutine_fn duplicates for bdrv_common_block_status_above callers

bdrv_common_block_status_above() is a g_c_w, and it is being called by
many "wrapper" functions like bdrv_is_allocated(),
bdrv_is_allocated_above() and bdrv_block_status_above().

Because we want to eventually split the coroutine from non-coroutine
case in g_c_w, create duplicate wrappers that take care of directly
calling the same coroutine functions called in the g_c_w.

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-Id: <20221128142337.657646-2-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove poll parameter from bdrv_parent_drained_begin_single()

All callers of bdrv_parent_drained_begin_single() pass poll=false now,
so we don't need the parameter any more.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-16-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Don't poll in bdrv_replace_child_noperm()

In order to make sure that bdrv_replace_child_noperm() doesn't have to
poll any more, get rid of the bdrv_parent_drained_begin_single() call.

This is possible now because we can require that the parent is already
drained through the child in question when the function is called and we
don't call the parent drain callbacks more than once.

The additional drain calls needed in callers cause the test case to run
its code in the drain handler too early (bdrv_attach_child() drains
now), so modify it to only enable the code after the test setup has
completed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-15-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop out of coroutine in bdrv_do_drained_begin_quiesce()

The next patch adds a parent drain to bdrv_attach_child_common(), which
shouldn't be, but is currently called from coroutines in some cases (e.g.
.bdrv_co_create implementations generally open new nodes). Therefore,
the assertion that we're not in a coroutine doesn't hold true any more.

We could just remove the assertion because there is nothing in the
function that should be in conflict with running in a coroutine, but
just to be on the safe side, we can reverse the caller relationship
between bdrv_do_drained_begin() and bdrv_do_drained_begin_quiesce() so
that the latter also just drops out of coroutine context and we can
still be certain in the future that any drain code doesn't run in
coroutines.

As a nice side effect, the structure of bdrv_do_drained_begin() is now
symmetrical with bdrv_do_drained_end().

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-14-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove ignore_bds_parents parameter from drain_begin/end.

ignore_bds_parents is now ignored during drain_begin and drain_end, so
we can just remove it there. It is still a valid optimisation for
drain_all in bdrv_drained_poll(), so leave it around there.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-13-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Call drain callbacks only once

We only need to call both the BlockDriver's callback and the parent
callbacks when going from undrained to drained or vice versa. A second
drain section doesn't make a difference for the driver or the parent,
they weren't supposed to send new requests before and after the second
drain.

One thing that gets in the way is the 'ignore_bds_parents' parameter in
bdrv_do_drained_begin_quiesce() and bdrv_do_drained_end(): It means that
bdrv_drain_all_begin() increases bs->quiesce_counter, but does not
quiesce the parent through BdrvChildClass callbacks. If an additional
drain section is started now, bs->quiesce_counter will be non-zero, but
we would still need to quiesce the parent through BdrvChildClass in
order to keep things consistent (and unquiesce it on the matching
bdrv_drained_end(), even though the counter would not reach 0 yet as
long as the bdrv_drain_all() section is still active).

Instead of keeping track of this, let's just get rid of the parameter.
It was introduced in commit 6cd5c9d7b2d as an optimisation so that
during bdrv_drain_all(), we wouldn't recursively drain all parents up to
the root for each node, resulting in quadratic complexity. As it happens,
calling the callbacks only once solves the same problem, so as of this
patch, we'll still have O(n) complexity and ignore_bds_parents is not
needed any more.

This patch only ignores the 'ignore_bds_parents' parameter. It will be
removed in a separate patch.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-12-kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove subtree drains

Subtree drains are not used any more. Remove them.

After this, BdrvChildClass.attach/detach() don't poll any more.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-11-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

stream: Replace subtree drain with a single node drain

The subtree drain was introduced in commit b1e1af394d9 as a way to avoid
graph changes between finding the base node and changing the block graph
as necessary on completion of the image streaming job.

The block graph could change between these two points because
bdrv_set_backing_hd() first drains the parent node, which involved
polling and can do anything.

Subtree draining was an imperfect way to make this less likely (because
with it, fewer callbacks are called during this window). Everyone agreed
that it's not really the right solution, and it was only committed as a
stopgap solution.

This replaces the subtree drain with a solution that simply drains the
parent node before we try to find the base node, and then call a version
of bdrv_set_backing_hd() that doesn't drain, but just asserts that the
parent node is already drained.

This way, any graph changes caused by draining happen before we start
looking at the graph and things stay consistent between finding the base
node and changing the graph.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-10-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Don't use subtree drains in bdrv_drop_intermediate()

Instead of using a subtree drain from the top node (which also drains
child nodes of base that we're not even interested in), use a normal
drain for base, which automatically drains all of the parents, too.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-9-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drain individual nodes during reopen

bdrv_reopen() and friends use subtree drains as a lazy way of covering
all the nodes they touch. Turns out that this lazy way is a lot more
complicated than just draining the nodes individually, even not
accounting for the additional complexity in the drain mechanism itself.

Simplify the code by switching to draining the individual nodes that are
already managed in the BlockReopenQueue anyway.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-8-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fix locking for bdrv_reopen_queue_child()

Callers don't agree whether bdrv_reopen_queue_child() should be called
with the AioContext lock held or not. Standardise on holding the lock
(as done by QMP blockdev-reopen and the replication block driver) and
fix bdrv_reopen() to do the same.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221118174110.55183-7-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Inline bdrv_drain_invoke()

bdrv_drain_invoke() has now two entirely separate cases that share no
code any more and are selected depending on a bool parameter. Each case
has only one caller. Just inline the function.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-6-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove drained_end_counter

drained_end_counter is unused now, nobody changes its value any more. It
can be removed.

In cases where we had two almost identical functions that only differed
in whether the caller passes drained_end_counter, or whether they would
poll for a local drained_end_counter to reach 0, these become a single
function.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20221118174110.55183-5-kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Revert .bdrv_drained_begin/end to non-coroutine_fn

Polling during bdrv_drained_end() can be problematic (and in the future,
we may get cases for bdrv_drained_begin() where polling is forbidden,
and we don't care about already in-flight requests, but just want to
prevent new requests from arriving).

The .bdrv_drained_begin/end callbacks running in a coroutine is the only
reason why we have to do this polling, so make them non-coroutine
callbacks again. None of the callers actually yield any more.

This means that bdrv_drained_end() effectively doesn't poll any more,
even if AIO_WAIT_WHILE() loops are still there (their condition is false
from the beginning). This is generally not a problem, but in
test-bdrv-drain, some additional explicit aio_poll() calls need to be
added because the test case wants to verify the final state after BHs
have executed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-4-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

test-bdrv-drain: Don't yield in .bdrv_co_drained_begin/end()

We want to change .bdrv_co_drained_begin/end() back to be non-coroutine
callbacks, so in preparation, avoid yielding in their implementation.

This does almost the same as the existing logic in bdrv_drain_invoke(),
by creating and entering coroutines internally. However, since the test
case is by far the heaviest user of coroutine code in drain callbacks,
it is preferable to have the complexity in the test case rather than the
drain core, which is already complicated enough without this.

The behaviour for bdrv_drain_begin() is unchanged because we increase
bs->in_flight and this is still polled. However, bdrv_drain_end()
doesn't wait for the spawned coroutine to complete any more. This is
fine, we don't rely on bdrv_drain_end() restarting all operations
immediately before the next aio_poll().

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-3-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qed: Don't yield in bdrv_qed_co_drain_begin()

We want to change .bdrv_co_drained_begin() back to be a non-coroutine
callback, so in preparation, avoid yielding in its implementation.

Because we increase bs->in_flight and bdrv_drained_begin() polls, the
behaviour is unchanged.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-2-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>