[0/13] Fix per-domain IRQ affinity, allow >255 CPUs on x86 without IRQ remapping

Message

David Woodhouse Oct. 5, 2020, 3:28 p.m. UTC

Linux currently refuses to use >255 CPUs on x86 unless it has interrupt
remapping. This is a bit gratuitous because it could use those extra
CPUs just fine; it just can't target external interrupts at them.

The only problem is that our generic IRQ domain code cann't cope with
the concept of domains which can only target a subset of CPUs.

The hyperv-iommu IRQ remapping driver works around this — not by
actually doing any remapping, but just returning -EINVAL if the
affinity is ever set to an unreachable CPU. This almost works, but ends
up being a bit late because irq_set_affinity_locked() doesn't call into
the irqchip driver immediately; the error only happens later.

This patch series implements a per-domain "maximum affinity" set and
uses it for the non-remapped IOAPIC and MSI domains on x86. As well as
allowing more CPUs to be used without interrupt remapping, this also
fixes the case where some IOAPICs or PCI devices aren't actually in
scope of any active IOMMU and are operating without remapping.

While we're at it, recognise that the 8-bit limit is a bit gratuitous
and a hypervisor could offer at least 15 bits of APIC ID in the IOAPIC
RTE and MSI address bits 11-5 without even needing to use remapping.

David Woodhouse (13):
      x86/apic: Use x2apic in guest kernels even with unusable CPUs.
      x86/msi: Only use high bits of MSI address for DMAR unit
      x86/ioapic: Handle Extended Destination ID field in RTE
      x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available
      genirq: Prepare for default affinity to be passed to __irq_alloc_descs()
      genirq: Add default_affinity argument to __irq_alloc_descs()
      irqdomain: Add max_affinity argument to irq_domain_alloc_descs()
      genirq: Add irq_domain_set_affinity()
      x86/irq: Add x86_non_ir_cpumask
      x86/irq: Limit IOAPIC and MSI domains' affinity without IR
      x86/smp: Allow more than 255 CPUs even without interrupt remapping
      iommu/irq_remapping: Kill most of hyperv-iommu.c now it's redundant
      x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID

 Documentation/virt/kvm/cpuid.rst     |   4 +
 arch/x86/include/asm/apic.h          |   1 +
 arch/x86/include/asm/io_apic.h       |   3 +-
 arch/x86/include/asm/mpspec.h        |   2 +
 arch/x86/include/asm/x86_init.h      |   2 +
 arch/x86/include/uapi/asm/kvm_para.h |   1 +
 arch/x86/kernel/apic/apic.c          |  41 +++++++++-
 arch/x86/kernel/apic/io_apic.c       |  23 ++++--
 arch/x86/kernel/apic/msi.c           |  44 +++++++++--
 arch/x86/kernel/apic/x2apic_phys.c   |   9 +++
 arch/x86/kernel/kvm.c                |   6 ++
 arch/x86/kernel/x86_init.c           |   1 +
 drivers/iommu/hyperv-iommu.c         | 149 +----------------------------------
 include/linux/interrupt.h            |   2 +
 include/linux/irq.h                  |  10 ++-
 include/linux/irqdomain.h            |   7 +-
 kernel/irq/devres.c                  |   8 +-
 kernel/irq/ipi.c                     |   2 +-
 kernel/irq/irqdesc.c                 |  29 ++++---
 kernel/irq/irqdomain.c               |  69 ++++++++++++++--
 kernel/irq/manage.c                  |  19 ++++-
 21 files changed, 240 insertions(+), 192 deletions(-)