| Message ID | 20240624071422.3473789-1-maobibo@loongson.cn (mailing list archive) |
|---|---|
| Headers | show |
| Series | LoongArch: KVM: VM migration enhancement | expand |
Series applied, thanks. Huacai On Mon, Jun 24, 2024 at 3:14 PM Bibo Mao <maobibo@loongson.cn> wrote: > > This patchset is to solve VM migration issues, the first six patches are > mmu relative, the last patch is relative with vcpu interrupt status. > > It fixes potential issue about tlb flush of secondary mmu and huge page > selection etc. Also it hardens LoongArch kvm mmu module. > > With this patchset, VM successfully migrates on my 3C5000 Dual-Way > machine with 32 cores. > 1. Pass to migrate when unixbench workload runs with 32 vcpus, for > some unixbench testcases there is much IPI sending. > 2. Pass to migrate with kernel compiling with 8 vcpus in VM > 3. Fail to migrate with kernel compiling with 32 vcpus in VM, since > there is to much memory writing operation, also there will be file > system inode inconsistent error after migration. > > --- > v2 ... v3: > 1. Merge patch 7 into this patchset since it is relative with VM > migration bugfix. > 2. Sync pending interrupt when getting ESTAT register, SW ESTAT > register is read after vcpu_put(). > 3. Add notation about smp_wmb() when update pmd entry, to elimate > checkpatch warning. > 4. Remove unnecessary modification about function kvm_pte_huge() > in patch 2. > 5. Add notation about secondary mmu tlb since it is firstly used here. > > v1 ... v2: > 1. Combine seperate patches into one patchset, all are relative with > migration. > 2. Mark page accessed without mmu_lock still, however with page ref > added > --- > Bibo Mao (7): > LoongArch: KVM: Delay secondary mmu tlb flush until guest entry > LoongArch: KVM: Select huge page only if secondary mmu supports it > LoongArch: KVM: Discard dirty page tracking on readonly memslot > LoongArch: KVM: Add memory barrier before update pmd entry > LoongArch: KVM: Add dirty bitmap initially all set support > LoongArch: KVM: Mark page accessed and dirty with page ref added > LoongArch: KVM: Sync pending interrupt when getting ESTAT from user > mode > > arch/loongarch/include/asm/kvm_host.h | 5 ++ > arch/loongarch/include/asm/kvm_mmu.h | 2 +- > arch/loongarch/kvm/main.c | 1 + > arch/loongarch/kvm/mmu.c | 67 ++++++++++++++++++++------- > arch/loongarch/kvm/tlb.c | 5 +- > arch/loongarch/kvm/vcpu.c | 29 ++++++++++++ > 6 files changed, 86 insertions(+), 23 deletions(-) > > > base-commit: 50736169ecc8387247fe6a00932852ce7b057083 > -- > 2.39.3 > >
Huacai, Thanks for the carefulness and efforts. Regards Bibo Mao On 2024/7/5 下午5:33, Huacai Chen wrote: > Series applied, thanks. > > Huacai > > On Mon, Jun 24, 2024 at 3:14 PM Bibo Mao <maobibo@loongson.cn> wrote: >> >> This patchset is to solve VM migration issues, the first six patches are >> mmu relative, the last patch is relative with vcpu interrupt status. >> >> It fixes potential issue about tlb flush of secondary mmu and huge page >> selection etc. Also it hardens LoongArch kvm mmu module. >> >> With this patchset, VM successfully migrates on my 3C5000 Dual-Way >> machine with 32 cores. >> 1. Pass to migrate when unixbench workload runs with 32 vcpus, for >> some unixbench testcases there is much IPI sending. >> 2. Pass to migrate with kernel compiling with 8 vcpus in VM >> 3. Fail to migrate with kernel compiling with 32 vcpus in VM, since >> there is to much memory writing operation, also there will be file >> system inode inconsistent error after migration. >> >> --- >> v2 ... v3: >> 1. Merge patch 7 into this patchset since it is relative with VM >> migration bugfix. >> 2. Sync pending interrupt when getting ESTAT register, SW ESTAT >> register is read after vcpu_put(). >> 3. Add notation about smp_wmb() when update pmd entry, to elimate >> checkpatch warning. >> 4. Remove unnecessary modification about function kvm_pte_huge() >> in patch 2. >> 5. Add notation about secondary mmu tlb since it is firstly used here. >> >> v1 ... v2: >> 1. Combine seperate patches into one patchset, all are relative with >> migration. >> 2. Mark page accessed without mmu_lock still, however with page ref >> added >> --- >> Bibo Mao (7): >> LoongArch: KVM: Delay secondary mmu tlb flush until guest entry >> LoongArch: KVM: Select huge page only if secondary mmu supports it >> LoongArch: KVM: Discard dirty page tracking on readonly memslot >> LoongArch: KVM: Add memory barrier before update pmd entry >> LoongArch: KVM: Add dirty bitmap initially all set support >> LoongArch: KVM: Mark page accessed and dirty with page ref added >> LoongArch: KVM: Sync pending interrupt when getting ESTAT from user >> mode >> >> arch/loongarch/include/asm/kvm_host.h | 5 ++ >> arch/loongarch/include/asm/kvm_mmu.h | 2 +- >> arch/loongarch/kvm/main.c | 1 + >> arch/loongarch/kvm/mmu.c | 67 ++++++++++++++++++++------- >> arch/loongarch/kvm/tlb.c | 5 +- >> arch/loongarch/kvm/vcpu.c | 29 ++++++++++++ >> 6 files changed, 86 insertions(+), 23 deletions(-) >> >> >> base-commit: 50736169ecc8387247fe6a00932852ce7b057083 >> -- >> 2.39.3 >> >>
This patchset is to solve VM migration issues, the first six patches are mmu relative, the last patch is relative with vcpu interrupt status. It fixes potential issue about tlb flush of secondary mmu and huge page selection etc. Also it hardens LoongArch kvm mmu module. With this patchset, VM successfully migrates on my 3C5000 Dual-Way machine with 32 cores. 1. Pass to migrate when unixbench workload runs with 32 vcpus, for some unixbench testcases there is much IPI sending. 2. Pass to migrate with kernel compiling with 8 vcpus in VM 3. Fail to migrate with kernel compiling with 32 vcpus in VM, since there is to much memory writing operation, also there will be file system inode inconsistent error after migration. --- v2 ... v3: 1. Merge patch 7 into this patchset since it is relative with VM migration bugfix. 2. Sync pending interrupt when getting ESTAT register, SW ESTAT register is read after vcpu_put(). 3. Add notation about smp_wmb() when update pmd entry, to elimate checkpatch warning. 4. Remove unnecessary modification about function kvm_pte_huge() in patch 2. 5. Add notation about secondary mmu tlb since it is firstly used here. v1 ... v2: 1. Combine seperate patches into one patchset, all are relative with migration. 2. Mark page accessed without mmu_lock still, however with page ref added --- Bibo Mao (7): LoongArch: KVM: Delay secondary mmu tlb flush until guest entry LoongArch: KVM: Select huge page only if secondary mmu supports it LoongArch: KVM: Discard dirty page tracking on readonly memslot LoongArch: KVM: Add memory barrier before update pmd entry LoongArch: KVM: Add dirty bitmap initially all set support LoongArch: KVM: Mark page accessed and dirty with page ref added LoongArch: KVM: Sync pending interrupt when getting ESTAT from user mode arch/loongarch/include/asm/kvm_host.h | 5 ++ arch/loongarch/include/asm/kvm_mmu.h | 2 +- arch/loongarch/kvm/main.c | 1 + arch/loongarch/kvm/mmu.c | 67 ++++++++++++++++++++------- arch/loongarch/kvm/tlb.c | 5 +- arch/loongarch/kvm/vcpu.c | 29 ++++++++++++ 6 files changed, 86 insertions(+), 23 deletions(-) base-commit: 50736169ecc8387247fe6a00932852ce7b057083