On an (even slightly) oversubscribed system, spinlocks are quickly
becoming a bottleneck, as some vcpus are spinning, waiting for a
lock to be released, while the vcpu holding the lock may not be
running at all.
This creates contention, and the observed slowdown is 40x for
hackbench. No, this isn't a typo.
The solution is to trap blocking WFEs and tell KVM that we're
now spinning. This ensures that other vpus will get a scheduling
boost, allowing the lock to be released more quickly. Also, using
CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT slightly improves the performance
when the VM is severely overcommited.
Quick test to estimate the performance: hackbench 1 process 1000
2xA15 host (baseline): 1.843s
2xA15 guest w/o patch: 2.083s
4xA15 guest w/o patch: 80.212s
8xA15 guest w/o patch: Could not be bothered to find out
2xA15 guest w/ patch: 2.102s
4xA15 guest w/ patch: 3.205s
8xA15 guest w/ patch: 6.887s
So we go from a 40x degradation to 1.5x in the 2x overcommit case,
which is vaguely more acceptable.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The virtio configuration has recently moved and is now visible everywhere.
Including the file again from KVM as we used to need earlier now causes
dependency problems:
warning: (CAIF_VIRTIO && VIRTIO_PCI && VIRTIO_MMIO && REMOTEPROC && RPMSG)
selects VIRTIO which has unmet direct dependencies (VIRTUALIZATION)
Cc: Christoffer Dall <cdall@cs.columbia.edu>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Commit d21a1c83c7 (ARM: KVM: define KVM_ARM_MAX_VCPUS
unconditionally) changed the Kconfig logic for KVM_ARM_MAX_VCPUS to work around a
build error arising from the use of KVM_ARM_MAX_VCPUS when CONFIG_KVM=n. The
resulting Kconfig logic is a bit awkward and leaves a KVM_ARM_MAX_VCPUS always
defined in the kernel config file.
This change reverts the Kconfig logic back and adds a simple preprocessor
conditional in kvm_host.h to handle when CONFIG_KVM_ARM_MAX_VCPUS is undefined.
Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The CONFIG_KVM_ARM_MAX_VCPUS symbol is needed in order to build the
kernel/context_tracking.c code, which includes the vgic data structures
implictly through the kvm headers. Definining the symbol to zero
on builds without KVM resolves this build error:
In file included from include/linux/kvm_host.h:33:0,
from kernel/context_tracking.c:18:
arch/arm/include/asm/kvm_host.h:28:23: warning: "CONFIG_KVM_ARM_MAX_VCPUS" is not defined [-Wundef]
#define KVM_MAX_VCPUS CONFIG_KVM_ARM_MAX_VCPUS
^
arch/arm/include/asm/kvm_vgic.h:34:24: note: in expansion of macro 'KVM_MAX_VCPUS'
#define VGIC_MAX_CPUS KVM_MAX_VCPUS
^
arch/arm/include/asm/kvm_vgic.h:38:6: note: in expansion of macro 'VGIC_MAX_CPUS'
#if (VGIC_MAX_CPUS > 8)
^
In file included from arch/arm/include/asm/kvm_host.h:41:0,
from include/linux/kvm_host.h:33,
from kernel/context_tracking.c:18:
arch/arm/include/asm/kvm_vgic.h:59:11: error: 'CONFIG_KVM_ARM_MAX_VCPUS' undeclared here (not in a function)
} percpu[VGIC_MAX_CPUS];
^
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Christoffer Dall <cdall@cs.columbia.edu>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
It is now possible to select CONFIG_KVM_ARM_TIMER to enable the
KVM architected timer support.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
It is now possible to select the VGIC configuration option.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
This commit introduces the framework for guest memory management
through the use of 2nd stage translation. Each VM has a pointer
to a level-1 table (the pgd field in struct kvm_arch) which is
used for the 2nd stage translations. Entries are added when handling
guest faults (later patch) and the table itself can be allocated and
freed through the following functions implemented in
arch/arm/kvm/arm_mmu.c:
- kvm_alloc_stage2_pgd(struct kvm *kvm);
- kvm_free_stage2_pgd(struct kvm *kvm);
Each entry in TLBs and caches are tagged with a VMID identifier in
addition to ASIDs. The VMIDs are assigned consecutively to VMs in the
order that VMs are executed, and caches and tlbs are invalidated when
the VMID space has been used to allow for more than 255 simultaenously
running guests.
The 2nd stage pgd is allocated in kvm_arch_init_vm(). The table is
freed in kvm_arch_destroy_vm(). Both functions are called from the main
KVM code.
We pre-allocate page table memory to be able to synchronize using a
spinlock and be called under rcu_read_lock from the MMU notifiers. We
steal the mmu_memory_cache implementation from x86 and adapt for our
specific usage.
We support MMU notifiers (thanks to Marc Zyngier) through
kvm_unmap_hva and kvm_set_spte_hva.
Finally, define kvm_phys_addr_ioremap() to map a device at a guest IPA,
which is used by VGIC support to map the virtual CPU interface registers
to the guest. This support is added by Marc Zyngier.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Targets KVM support for Cortex A-15 processors.
Contains all the framework components, make files, header files, some
tracing functionality, and basic user space API.
Only supported core is Cortex-A15 for now.
Most functionality is in arch/arm/kvm/* or arch/arm/include/asm/kvm_*.h.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Marc Zyngier