Hello, This series adds support for MSI IRQ domains on top of the XICS (P8) and XIVE (P9/P10) IRQ domains for the PowerNV (baremetal) and pSeries (VM) platforms. It should simplify and improve IRQ affinity of PCI MSIs under these PowerPC platforms, specially for drivers distributing multiple RX/TX queues on the different CPUs of the system.
Data locality can still be improved with an interrupt controller node per chip but this requires FW changes. It could be done under OPAL. The patchset has a large impact but it is well contained under the MSI support. Initial tests were done on the P8, P9 and P10 PowerNV and pSeries platforms, under the KVM and PowerVM hypervisor. PCI passthrough was tested on P8/KVM, P9/KVM and P9/pVM with both interrupt modes. P8 passthrough has some optimization to EOI MSIs when under real mode : e3c13e56a471 ("KVM: PPC: Book3S HV: Handle passthrough interrupts in guest") 5d375199ea96 ("KVM: PPC: Book3S HV: Set server for passed-through interrupts") They give us a ~10% bandwidth improvement on some 100G adapters (Thanks Alexey), so it's good to keep but they require access to the low level IRQ domain of the machine. It should be possible to rework the code and use the MSI IRQ domains instead but for now, it's simpler to keep the bypass. That can come later. The P8/CAPI driver is also impacted. Tests were done on a Firestone system with a memory AFU. Thanks, C. Changes since v2 : - Included some CONFIG_IRQ_DOMAIN_HIERARCHY ifdefs - Microwatt fixes for ICS native - Removed irqd_is_started() check when setting the affinity Cédric Le Goater (32): powerpc/pseries/pci: Introduce __find_pe_total_msi() powerpc/pseries/pci: Introduce rtas_prepare_msi_irqs() powerpc/xive: Add support for IRQ domain hierarchy powerpc/xive: Ease debugging of xive_irq_set_affinity() powerpc/pseries/pci: Add MSI domains powerpc/xive: Drop unmask of MSIs at startup powerpc/xive: Remove irqd_is_started() check when setting the affinity powerpc/pseries/pci: Add a domain_free_irqs() handler powerpc/pseries/pci: Add a msi_free() handler to clear XIVE data powerpc/pseries/pci: Add support of MSI domains to PHB hotplug powerpc/powernv/pci: Introduce __pnv_pci_ioda_msi_setup() powerpc/powernv/pci: Add MSI domains KVM: PPC: Book3S HV: Use the new IRQ chip to detect passthrough interrupts KVM: PPC: Book3S HV: XIVE: Change interface of passthrough interrupt routines KVM: PPC: Book3S HV: XIVE: Fix mapping of passthrough interrupts powerpc/xics: Remove ICS list powerpc/xics: Rename the map handler in a check handler powerpc/xics: Give a name to the default XICS IRQ domain powerpc/xics: Add debug logging to the set_irq_affinity handlers powerpc/xics: Add support for IRQ domain hierarchy powerpc/powernv/pci: Customize the MSI EOI handler to support PHB3 powerpc/pci: Drop XIVE restriction on MSI domains powerpc/xics: Drop unmask of MSIs at startup powerpc/pseries/pci: Drop unused MSI code powerpc/powernv/pci: Drop unused MSI code powerpc/powernv/pci: Adapt is_pnv_opal_msi() to detect passthrough interrupt powerpc/xics: Fix IRQ migration powerpc/powernv/pci: Set the IRQ chip data for P8/CXL devices powerpc/powernv/pci: Rework pnv_opal_pci_msi_eoi() KVM: PPC: Book3S HV: XICS: Fix mapping of passthrough interrupts powerpc/xive: Use XIVE domain under xmon and debugfs genirq: Improve "hwirq" output in /proc and /sys/ arch/powerpc/include/asm/kvm_ppc.h | 4 +- arch/powerpc/include/asm/pci-bridge.h | 5 + arch/powerpc/include/asm/pnv-pci.h | 2 +- arch/powerpc/include/asm/xics.h | 3 +- arch/powerpc/include/asm/xive.h | 1 + arch/powerpc/platforms/powernv/pci.h | 6 - arch/powerpc/platforms/pseries/pseries.h | 2 + arch/powerpc/kernel/pci-common.c | 6 + arch/powerpc/kvm/book3s_hv.c | 18 +- arch/powerpc/kvm/book3s_hv_rm_xics.c | 8 +- arch/powerpc/kvm/book3s_xive.c | 18 +- arch/powerpc/platforms/powernv/pci-ioda.c | 256 ++++++++++++++++-- arch/powerpc/platforms/powernv/pci.c | 67 ----- arch/powerpc/platforms/pseries/msi.c | 296 ++++++++++++++++----- arch/powerpc/platforms/pseries/pci_dlpar.c | 4 + arch/powerpc/platforms/pseries/setup.c | 2 + arch/powerpc/sysdev/xics/ics-native.c | 13 +- arch/powerpc/sysdev/xics/ics-opal.c | 40 +-- arch/powerpc/sysdev/xics/ics-rtas.c | 40 +-- arch/powerpc/sysdev/xics/xics-common.c | 129 ++++++--- arch/powerpc/sysdev/xive/common.c | 98 +++++-- kernel/irq/irqdesc.c | 2 +- kernel/irq/irqdomain.c | 1 + kernel/irq/proc.c | 2 +- 24 files changed, 710 insertions(+), 313 deletions(-) -- 2.31.1