| Message ID | 20211105192101.3862492-1-maz@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | KVM: Turn the vcpu array into an xarray | expand |
On 05.11.21 20:20, Marc Zyngier wrote: > The kvm structure is pretty large. A large portion of it is the vcpu > array, which is 4kB on x86_64 and arm64 as they deal with 512 vcpu > VMs. Of course, hardly anyone runs VMs this big, so this is often a > net waste of memory and cache locality. > > A possible approach is to turn the fixed-size array into an xarray, > which results in a net code deletion after a bit of cleanup. > > This series is on top of the current linux/master as it touches the > RISC-V implementation. Only tested on arm64. > > Marc Zyngier (5): > KVM: Move wiping of the kvm->vcpus array to common code > KVM: mips: Use kvm_get_vcpu() instead of open-coded access > KVM: s390: Use kvm_get_vcpu() instead of open-coded access > KVM: x86: Use kvm_get_vcpu() instead of open-coded access > KVM: Convert the kvm->vcpus array to a xarray > > arch/arm64/kvm/arm.c | 10 +--------- > arch/mips/kvm/loongson_ipi.c | 4 ++-- > arch/mips/kvm/mips.c | 23 ++--------------------- > arch/powerpc/kvm/powerpc.c | 10 +--------- > arch/riscv/kvm/vm.c | 10 +--------- > arch/s390/kvm/kvm-s390.c | 26 ++++++-------------------- > arch/x86/kvm/vmx/posted_intr.c | 2 +- > arch/x86/kvm/x86.c | 9 +-------- > include/linux/kvm_host.h | 7 ++++--- > virt/kvm/kvm_main.c | 33 ++++++++++++++++++++++++++------- > 10 files changed, 45 insertions(+), 89 deletions(-) > For x86 you can add my: Tested-by: Juergen Gross <jgross@suse.com> Juergen
On 11/16/21 15:13, Juergen Gross wrote: > On 05.11.21 20:20, Marc Zyngier wrote: >> The kvm structure is pretty large. A large portion of it is the vcpu >> array, which is 4kB on x86_64 and arm64 as they deal with 512 vcpu >> VMs. Of course, hardly anyone runs VMs this big, so this is often a >> net waste of memory and cache locality. >> >> A possible approach is to turn the fixed-size array into an xarray, >> which results in a net code deletion after a bit of cleanup. >> >> This series is on top of the current linux/master as it touches the >> RISC-V implementation. Only tested on arm64. >> >> Marc Zyngier (5): >> KVM: Move wiping of the kvm->vcpus array to common code >> KVM: mips: Use kvm_get_vcpu() instead of open-coded access >> KVM: s390: Use kvm_get_vcpu() instead of open-coded access >> KVM: x86: Use kvm_get_vcpu() instead of open-coded access >> KVM: Convert the kvm->vcpus array to a xarray >> >> arch/arm64/kvm/arm.c | 10 +--------- >> arch/mips/kvm/loongson_ipi.c | 4 ++-- >> arch/mips/kvm/mips.c | 23 ++--------------------- >> arch/powerpc/kvm/powerpc.c | 10 +--------- >> arch/riscv/kvm/vm.c | 10 +--------- >> arch/s390/kvm/kvm-s390.c | 26 ++++++-------------------- >> arch/x86/kvm/vmx/posted_intr.c | 2 +- >> arch/x86/kvm/x86.c | 9 +-------- >> include/linux/kvm_host.h | 7 ++++--- >> virt/kvm/kvm_main.c | 33 ++++++++++++++++++++++++++------- >> 10 files changed, 45 insertions(+), 89 deletions(-) >> > > For x86 you can add my: > > Tested-by: Juergen Gross <jgross@suse.com> Heh, unfortunately x86 is the only one that needs a change in patch 4. I'll Cc you on my version. Paolo
On 16.11.21 15:21, Paolo Bonzini wrote: > On 11/16/21 15:13, Juergen Gross wrote: >> On 05.11.21 20:20, Marc Zyngier wrote: >>> The kvm structure is pretty large. A large portion of it is the vcpu >>> array, which is 4kB on x86_64 and arm64 as they deal with 512 vcpu >>> VMs. Of course, hardly anyone runs VMs this big, so this is often a >>> net waste of memory and cache locality. >>> >>> A possible approach is to turn the fixed-size array into an xarray, >>> which results in a net code deletion after a bit of cleanup. >>> >>> This series is on top of the current linux/master as it touches the >>> RISC-V implementation. Only tested on arm64. >>> >>> Marc Zyngier (5): >>> KVM: Move wiping of the kvm->vcpus array to common code >>> KVM: mips: Use kvm_get_vcpu() instead of open-coded access >>> KVM: s390: Use kvm_get_vcpu() instead of open-coded access >>> KVM: x86: Use kvm_get_vcpu() instead of open-coded access >>> KVM: Convert the kvm->vcpus array to a xarray >>> >>> arch/arm64/kvm/arm.c | 10 +--------- >>> arch/mips/kvm/loongson_ipi.c | 4 ++-- >>> arch/mips/kvm/mips.c | 23 ++--------------------- >>> arch/powerpc/kvm/powerpc.c | 10 +--------- >>> arch/riscv/kvm/vm.c | 10 +--------- >>> arch/s390/kvm/kvm-s390.c | 26 ++++++-------------------- >>> arch/x86/kvm/vmx/posted_intr.c | 2 +- >>> arch/x86/kvm/x86.c | 9 +-------- >>> include/linux/kvm_host.h | 7 ++++--- >>> virt/kvm/kvm_main.c | 33 ++++++++++++++++++++++++++------- >>> 10 files changed, 45 insertions(+), 89 deletions(-) >>> >> >> For x86 you can add my: >> >> Tested-by: Juergen Gross <jgross@suse.com> > > Heh, unfortunately x86 is the only one that needs a change in patch 4. > I'll Cc you on my version. I guess the changes in kvm_main.c are more important for my series. :-) I've replaced patch 4 with your variant and everything is still working. Not sure how relevant that is, though. Juergen
On 11/5/21 20:20, Marc Zyngier wrote: > The kvm structure is pretty large. A large portion of it is the vcpu > array, which is 4kB on x86_64 and arm64 as they deal with 512 vcpu > VMs. Of course, hardly anyone runs VMs this big, so this is often a > net waste of memory and cache locality. > > A possible approach is to turn the fixed-size array into an xarray, > which results in a net code deletion after a bit of cleanup. > > This series is on top of the current linux/master as it touches the > RISC-V implementation. Only tested on arm64. Queued, only locally until I get a review for my replacement of patch 4 (see https://lore.kernel.org/kvm/20211116142205.719375-1-pbonzini@redhat.com/T/). Paolo > Marc Zyngier (5): > KVM: Move wiping of the kvm->vcpus array to common code > KVM: mips: Use kvm_get_vcpu() instead of open-coded access > KVM: s390: Use kvm_get_vcpu() instead of open-coded access > KVM: x86: Use kvm_get_vcpu() instead of open-coded access > KVM: Convert the kvm->vcpus array to a xarray > > arch/arm64/kvm/arm.c | 10 +--------- > arch/mips/kvm/loongson_ipi.c | 4 ++-- > arch/mips/kvm/mips.c | 23 ++--------------------- > arch/powerpc/kvm/powerpc.c | 10 +--------- > arch/riscv/kvm/vm.c | 10 +--------- > arch/s390/kvm/kvm-s390.c | 26 ++++++-------------------- > arch/x86/kvm/vmx/posted_intr.c | 2 +- > arch/x86/kvm/x86.c | 9 +-------- > include/linux/kvm_host.h | 7 ++++--- > virt/kvm/kvm_main.c | 33 ++++++++++++++++++++++++++------- > 10 files changed, 45 insertions(+), 89 deletions(-) >
On Tue, 16 Nov 2021 15:03:40 +0000, Paolo Bonzini <pbonzini@redhat.com> wrote: > > On 11/5/21 20:20, Marc Zyngier wrote: > > The kvm structure is pretty large. A large portion of it is the vcpu > > array, which is 4kB on x86_64 and arm64 as they deal with 512 vcpu > > VMs. Of course, hardly anyone runs VMs this big, so this is often a > > net waste of memory and cache locality. > > > > A possible approach is to turn the fixed-size array into an xarray, > > which results in a net code deletion after a bit of cleanup. > > > > This series is on top of the current linux/master as it touches the > > RISC-V implementation. Only tested on arm64. > > Queued, only locally until I get a review for my replacement of patch > 4 (see > https://lore.kernel.org/kvm/20211116142205.719375-1-pbonzini@redhat.com/T/). In which case, let me send a v2 with the changes that we discussed with Sean. It will still have my version of patch 4, but that's nothing you can't fix. M.