Hello, Here is the version 6 of the QEMU models adding support for the XIVE interrupt controller to the sPAPR machine, under TCG and KVM. Support for the PowerNV POWER9 machine will be proposed in a PowerNV patchset sometime next year now.
The most important changes for sPAPR are the removal of the SysBusDevice inheritance, the removal of the KVM classes, support for XIVE structures in big-endian only format and the introduction of a VM change state handler for KVM migration. Thanks, C. Changes in v6 : Common XIVE models : - included documentation in xive.h - removed SysBusDevice inheritance from Xive Sources - set ASSERTED bit in xive_source_lsi_trigger() - renamed XiveFabric in XiveNotifier - reworked XIVE tables accessors, introduce a _write method for words - introduced the source number encoding on PowerBUS in accessors - used fixed big-endian format for XIVE structures - reworked the presenter matching routine - renamed *cam_line helpers sPAPR models : - reworked the 'info pic output - moved the END reset at the sPAPR level - renamed the spapr_xive_irq_enable/disable routine in claim/free - removed the reset_tctx hook - renamed xics_max_server_number() and fixed spapr_irq_init() prototype - removed the use of the xive_router routines in the sPAPR XIVE hcalls - used address_space_map() to validate the EQ - introduced a spapr_xive_reset_tctx() to set the OS CAM line at reset - introduced OV5 defines for the XIVE mode - removed the XIVE classes - enable/disable the XIVE MMIOs depending on the mode - introduced a spapr_rtas_unregister() helper - mixed enhancements KVM : - removed the KVM XIVE models and reworked KVM support with helpers - introduced a VM change state handler to quiesce XIVE before transferring the EQ pages - improved KVM support for the dual machine (removed extra cleanups) PowerNV: - postponed for a PowerNV patchset only Changes in v5 : Common XIVE models : - renamed the XIVE structures to fit the changes of the XIVE architecture documents: IVE, EQD, VPD -> EAS, END, NVT - reworked the monitor ouput to print the EQ contents sPAPR models : - introduced a XIVE Router 'reset' method for the Xive Thread Context to set the OS CAM line of the VCPU - introduced a spapr_irq_init() routine to the sPAPR IRQ backend and reworked the XIVE-only machine to fit mainline QEMU - introduced a reset() method to the sPAPR IRQ backend to handle changes in the interrupt mode after machine reset - introduced a 'dual' machine supporting both interrupt mode KVM : - introduced some more sPAPR NVT and END indexing helpers for KVM support - fixed the virtual LSIs in KVM by using the H_INT_ESB source flag - improved the KVM support with better common classes and cleaner QEMU<->KVM interfaces - improved KVM migration with a better control on the capture sequence. Still some issues with 'ceded' VCPUs - introduced KVM support for the 'dual' machine PowerNV: - introduced address spaces for the IPI and END set translation tables Changes in v4 : See https://lists.gnu.org/archive/html/qemu-devel/2018-06/msg01672.html = XIVE ================================================================= The POWER9 processor comes with a new interrupt controller, called XIVE as "eXternal Interrupt Virtualization Engine". * Overall architecture XIVE Interrupt Controller +------------------------------------+ IPIs | +---------+ +---------+ +--------+ | +-------+ | |VC | |CQ | |PC |----> | CORES | | | esb | | | | |----> | | | | eas | | Bridge | | tctx |----> | | | |SC end | | | | nvt | | | | +------+ | +---------+ +----+----+ +--------+ | +-+-+-+-+ | RAM | +------------------|-----------------+ | | | | | | | | | | | | | | | | | +--------------------v------------------------v-v-v--+ other | <--+ Power Bus +--> chips | esb | +---------+-----------------------+------------------+ | eas | | | | end | +--|------+ | | nvt | +----+----+ | +----+----+ +------+ |SC | | |SC | | | | | | | PQ-bits | | | PQ-bits | | local |-+ | in VC | +---------+ +---------+ PCIe NX,NPU,CAPI SC: Source Controller (aka. IVSE) VC: Virtualization Controller (aka. IVRE) PC: Presentation Controller (aka. IVPE) CQ: Common Queue (Bridge) PQ-bits: 2 bits source state machine (P:pending Q:queued) esb: Event State Buffer (Array of PQ bits in an IVSE) eas: Event Assignment Structure end: Event Notification Descriptor nvt: Notification Virtual Target tctx: Thread interrupt Context The XIVE IC is composed of three sub-engines : - Interrupt Virtualization Source Engine (IVSE), or Source Controller (SC). These are found in PCI PHBs, in the PSI host bridge controller, but also inside the main controller for the core IPIs and other sub-chips (NX, CAP, NPU) of the chip/processor. They are configured to feed the IVRE with events. - Interrupt Virtualization Routing Engine (IVRE) or Virtualization Controller (VC). Its job is to match an event source with an Event Notification Descriptor (END). - Interrupt Virtualization Presentation Engine (IVPE) or Presentation Controller (PC). It maintains the interrupt context state of each thread and handles the delivery of the external exception to the thread. * XIVE internal tables Each of the sub-engines uses a set of tables to redirect exceptions from event sources to CPU threads. +-------+ User or OS | EQ | or +------>|entries| Hypervisor | | .. | Memory | +-------+ | ^ | | +-------------------------------------------------+ | | Hypervisor +------+ +---+--+ +---+--+ +------+ Memory | ESB | | EAT | | ENDT | | NVTT | (skiboot) +----+-+ +----+-+ +----+-+ +------+ ^ | ^ | ^ | ^ | | | | | | | +-------------------------------------------------+ | | | | | | | | | | | | | | +----|--|--------|--|--------|--|-+ +-|-----+ +------+ | | | | | | | | | | tctx| |Thread| IPI or ---+ + v + v + v |---| + .. |-----> | HW events | | | | | | | IVRE | | IVPE | +------+ +---------------------------------+ +-------+ The IVSE have a 2-bits state machine, P for pending and Q for queued, for each source that allows events to be triggered. They are stored in an Event State Buffer (ESB) array and can be controlled by MMIOs. If the event is let through, the IVRE looks up in the Event Assignment Structure (EAS) table for an Event Notification Descriptor (END) configured for the source. Each Event Notification Descriptor defines a notification path to a CPU and an in-memory Event Queue, in which will be enqueued an EQ data for the OS to pull. The IVPE determines if a Notification Virtual Target (NVT) can handle the event by scanning the thread contexts of the VCPUs dispatched on the processor HW threads. It maintains the interrupt context state of each thread in a NVT table. * Overview of the QEMU models for the XIVE sub-engines The XiveSource models the IVSE in general, internal and external. It handles the source ESBs and the MMIO interface to control them. The XiveNotifier is a small helper interface interconnecting the XiveSource to the XiveRouter. The XiveRouter is an abstract model acting as a combined IVRE and IVPE. It routes event notifications using the IVE and EQD tables to the IVPE sub-engine which does a CAM scan to find a CPU to deliver the exception. Storage should be provided by the inheriting classes. XiveENDSource is a special source object. It exposes the EQ ESB MMIOs of the Event Queues which are used for coalescing event notifications and for escalation. Not used on the field, only to sync the EQ cache in OPAL. Finally, the XiveTCTX contains the interrupt state context of a thread, four sets of registers, one for each exception that can be delivered to a CPU. These contexts are scanned by the IVPE to find a matching VP when a notification is triggered. It also models the Thread Interrupt Management Area (TIMA), which exposes the thread context registers to the CPU for interrupt management. * XIVE for sPAPR sPAPRXive models the XIVE interrupt controller of a sPAPR machine. It inherits from the XiveRouter and provisions storage for the IVE and END tables. The NVT table does not need a backend in sPAPR. It owns a XiveSource object for the IPIs and the virtual device interrupts, a memory region for the TIMA and a XiveENDSource to manage the END ESBs. (not used by Linux). These choices were made to have a sPAPR interrupt controller consistent with the one found on baremetal and to facilitate KVM support, the main difficulty being the host memory regions exposed to the guest. The NVT and tbe END indexing needs some care and a set of helpers are defined to ease the conversion between the CPU id as seen by the guest and the XIVE identifiers manipulated by the models. * Integration in the sPAPR machine, xive only and dual A new sPAPR IRQ backend is defined for XIVE. It introduces a couple of new operations to handle the differences in the creation of the device tree and in the allocation of the CPU interrupt controller. A new 'xive' only pseries machine is defined using this XIVE backend. Being able to support both interrupt mode in the same machine requires some more changes. As the machine chooses the interrupt mode at CAS time, it is activated after a reconfiguration done in a reset. This is handled by a new 'dual' sPAPR IRQ backend which is built on top of the XICS and XIVE backend. A new 'dual' pseries machine is defined using this backend. * KVM support Support for KVM introduces a set of specific XIVE models, very much like XICS does, which self-connect to their KVM counterparts in the Linux kernel. Two host memory regions are exposed to the guest and need special care at initialization : - ESB mmios - Thread Interrupt Management Area (TIMA) The models uses KVM accessors to synchronize the QEMU state with KVM. The states are : - the source configuration (EAT) - the END configuration (ENDT) - the OS EQ state (toggle bit and index) - the thread interrupt context registers. Hybrid guest using KVM and an emulated irqchip (kernel_irqchip=off) is supported. Migration under KVM is supported. KVM support for the 'dual' machine required some more changes. Both interrupt mode need to be initialized at the QEMU level to keep the IRQ number space in sync and to allow switching from one mode to another. At the KVM level, the whole initialization of the KVM device, sources and presenters, needs to be done in the reset handler when the interrupt mode is chosen. This is a major change in the KVM models. KVM being initialized at reset, we loose the possiblity to fallback to the QEMU emulated mode in case of failure and failures become fatal to the machine. * PowerNV models The PnvXIVE model uses the XiveRouter abstract model just like sPAPRXive. It provides accessors to the EAS, END and NVT tables which are stored in the QEMU PowerNV machine and not in QEMU anymore. It owns a set of memory regions for the IC registers, the ESBs, the END ESBs, the TIMA, the notification MMIO. Multichip is supported and the available IVSEs are the internal one for the IPIS, the PSI host bridge controller and PHB4. The next interesting step would be to add escalation events and model the VCPU dispatching to support emulated KVM guests. * GitHub trees QEMU sPAPR: https://github.com/legoater/qemu/commits/xive-v6-3.1 QEMU PowerNV: https://github.com/legoater/qemu/commits/powernv-3.1 Linux/KVM: https://github.com/legoater/linux/commits/xive-4.20 OPAL: https://github.com/legoater/skiboot/commits/xive Cédric Le Goater (37): ppc/xive: introduce a XIVE interrupt source model ppc/xive: add support for the LSI interrupt sources ppc/xive: introduce the XiveNotifier interface ppc/xive: introduce the XiveRouter model ppc/xive: introduce the XIVE Event Notification Descriptors ppc/xive: add support for the END Event State buffers ppc/xive: introduce the XIVE interrupt thread context ppc/xive: introduce a simplified XIVE presenter ppc/xive: notify the CPU when the interrupt priority is more privileged spapr/xive: introduce a XIVE interrupt controller spapr/xive: use the VCPU id as a NVT identifier spapr: initialize VSMT before initializing the IRQ backend spapr: introduce a spapr_irq_init() routine spapr: modify the irq backend 'init' method spapr: export and rename the xics_max_server_number() routine spapr: introdude a new machine IRQ backend for XIVE spapr: add hcalls support for the XIVE exploitation interrupt mode spapr: add device tree support for the XIVE exploitation mode spapr: allocate the interrupt thread context under the CPU core spapr: extend the sPAPR IRQ backend for XICS migration spapr: add a 'reset' method to the sPAPR IRQ backend spapr: add a 'pseries-3.1-xive' machine type linux-headers: update to 4.20-rc5 spapr/xive: add KVM support spapr/xive: add state synchronization with KVM spapr/xive: introduce a VM state change handler spapr/xive: add migration support for KVM spapr/xive: fix migration of the XiveTCTX under TCG spapr: set the interrupt presenter at reset spapr/xive: enable XIVE MMIOs at reset spapr: add a 'pseries-3.1-dual' machine type ppc/xics: introduce a icp_kvm_connect() routine spapr/rtas: modify spapr_rtas_register() to remove RTAS handlers sysbus: add a sysbus_mmio_unmap() helper spapr: introduce routines to delete the KVM IRQ device spapr: check for KVM IRQ device activation spapr: add KVM support to the 'dual' machine default-configs/ppc64-softmmu.mak | 3 + include/hw/ppc/spapr.h | 28 +- include/hw/ppc/spapr_cpu_core.h | 2 + include/hw/ppc/spapr_irq.h | 15 +- include/hw/ppc/spapr_xive.h | 79 ++ include/hw/ppc/xics.h | 5 +- include/hw/ppc/xive.h | 453 ++++++++ include/hw/ppc/xive_regs.h | 213 ++++ include/hw/sysbus.h | 1 + linux-headers/asm-powerpc/kvm.h | 46 + linux-headers/linux/kvm.h | 6 + target/ppc/kvm_ppc.h | 6 + hw/core/sysbus.c | 10 + hw/intc/spapr_xive.c | 1519 ++++++++++++++++++++++++++ hw/intc/spapr_xive_kvm.c | 860 +++++++++++++++ hw/intc/xics_kvm.c | 136 ++- hw/intc/xics_spapr.c | 3 +- hw/intc/xive.c | 1686 +++++++++++++++++++++++++++++ hw/ppc/spapr.c | 92 +- hw/ppc/spapr_cpu_core.c | 31 +- hw/ppc/spapr_hcall.c | 13 + hw/ppc/spapr_irq.c | 434 +++++++- hw/ppc/spapr_rtas.c | 2 +- target/ppc/kvm.c | 7 + hw/intc/Makefile.objs | 3 + 25 files changed, 5588 insertions(+), 65 deletions(-) create mode 100644 include/hw/ppc/spapr_xive.h create mode 100644 include/hw/ppc/xive.h create mode 100644 include/hw/ppc/xive_regs.h create mode 100644 hw/intc/spapr_xive.c create mode 100644 hw/intc/spapr_xive_kvm.c create mode 100644 hw/intc/xive.c -- 2.17.2
