On 19/10/2026 13:04, Mark Cave-Ayland wrote:
> This series attempts to improve the modelling of legacy x86 PICs within
> QEMU. The initial aim was to have a qdev device with one output gpio, and
> 16 input gpios representing the legacy IRQ lines, however as I went about
> trying to model this, I found a number of other inconsistencies/bugs: in
> particular there seemed to be some fuzziness around the functionality that
> belongs to an individual i8259 chip, and a cascaded i8259 configuration.
>
> For the purposes of this patch an i8259 device is considered to be a single
> i8259 chip whilst a PIC device is considered to be either a cascaded i8259
> configuration (emulated/KVM) or standalone device (Xen) representing the set
> of 16 legacy IRQs and a single output IRQ.
>
> Due to the updated device names, there is currently one visible change in
> the HMP monitor output for "info irq" where the isa-i8259 device now appears
> as isa-i8259-pic and the kvm-i8259 device appears as kvm-i8259-pic (see
> patch 25).
>
> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
>
> [Patches still requiring review: 18-21, 23]
>
> v3:
> - Rebase onto master
>
> - Add R-B tags from Peter
>
> - Introduce isa_realize() function in patch 18 and use it in patches 19 and
> 21 to fix refcounting for ISA devices within the PIC container device
>
> - Don't pass input IRQ 2 from the primary PIC when implementing the cascaded
> PIC configuration in patch 20
>
> v2:
> - Rebase onto master
>
> - Add R-B tags from Peter
>
> - Remove FIXME from commit message and rename trace event in patch 13
>
> - Introduce a separate I8259State for TYPE_I8259 and KVMI8259State for
> TYPE_KVM_I8259 as suggested by Peter
>
> - Keep but rename the associated classes for TYPE_I8259 and TYPE_KVM_I8259
> and move the DeviceRealize function into each one, instead of keeping it
> in the parent class
>
> - Add patches to make variable names consistent when accessing
> I8259CommonState from child objects
>
> - Update the comment on patch 19 to reflect that implementing the
> cascaded secondary is actually a bugfix
>
>
> Mark Cave-Ayland (28):
> i8259_internal.h: rename PIC_COMMON to I8259_COMMON
> i8259_internal.h: rename TYPE_I8259_COMMON to i8259-common
> i8259_internal.h: rename PICCommonState to I8259CommonState
> i8259_internal.h: rename PICCommonClass to I8259CommonClass
> i8259.c: rename PICClass to I8259Class
> i8259.c: introduce TYPE_I8259 device
> i8259.c: always use cs for I8259CommonState variables
> kvm/i8259.c: rename KVMPICClass to KVMI8259Class
> kvm/i8259.c: introduce TYPE_KVM_I8259 device
> kvm/i8259.c: always use cs for I8259CommonState variables
> i8259_common.c: always use cs for I8259CommonState variables
> i8259_common.c: use i8259/i8259_common prefix for functions
> i8259.c: use i8259 prefix for functions
> kvm/i8259.c: use kvm_i8259 prefix for functions
> i8259_common.c: convert to use DEFINE_TYPES() macro
> i8259.c: convert to use DEFINE_TYPES() macro
> kvm/i8259.c: convert to use DEFINE_TYPES() macro
> isa-bus.c: introduce isa_realize() function
> i8259.c: introduce TYPE_I8259_PIC device
> i8259.c: properly cascade secondary i8259 from primary i8259
> kvm/i8259.c: introduce TYPE_KVM_I8259_PIC device
> xen-hvm.c: rename xen_interrupt_controller_init() to xen_i8259_init()
> xen/xen-hvm.c: introduce TYPE_XEN_I8259_PIC device
> i8259_common.c: remove unused i8259_init_chip() function
> i8259: move InterruptStatsProviderClass get_statistics() from i8259 to
> PIC
> i8259.c: switch isa_pic from I8259State to I8259PICState
> i8259_common.c: remove static irq_count and irq_level variables
> i8259_common.c: add missing 0x prefixes to "info pic" output
>
> include/hw/intc/i8259.h | 11 +-
> include/hw/isa/i8259_internal.h | 19 +-
> include/hw/isa/isa.h | 1 +
> include/hw/xen/xen.h | 2 +-
> hw/i386/kvm/i8259.c | 250 ++++++++++++-----
> hw/i386/pc.c | 2 +-
> hw/i386/xen/xen-hvm.c | 40 ++-
> hw/intc/i8259.c | 470 ++++++++++++++++++++------------
> hw/intc/i8259_common.c | 225 +++++++--------
> hw/isa/isa-bus.c | 5 +
> stubs/xen-hw-stub.c | 2 +-
> hw/intc/trace-events | 8 +-
> 12 files changed, 648 insertions(+), 387 deletions(-)
Thanks for helping review this series, Peter. Now it is fully reviewed,
I'll send a PR for this from my qom-qdev queue later today, unless I
hear any last minute objections from any x86 maintainers.
ATB,
Mark.