Re: [PATCH v3 0/3] Optimize S2 page splitting

[Leonardo Bras <leo.bras@arm.com>]

On Sun, Jul 12, 2026 at 08:20:28PM +0530, Dev Jain wrote:
> 
> 
> On 08/07/26 7:10 pm, Leonardo Bras wrote:
> > While playing with dirty-bit tracking, I decided to take a look on how page
> > splitting works. Found out all entries are walked, even though we can infer,
> > for instance that:
> > - If a level-3 entry is walked, it means the parent level-2 entry is split
> > - If a split just succeeded in an table entry, it means all children nodes
> >   are already split

Hi Dev, thanks for reviewing!


> 
> That does not happen for force_pte = false right? Or is it guaranteed that
> in your case that force_pte = true always?

IIUC for kvm_pgtable_stage2_split(), force_pte will affect the creation of 
pagetables, and will only happen to be false if the cache used for  
splitting hugepages does not contain enough pages to split the whole 
subtree. In this case, it only maps one page of either table entries or 
ptes. 

I am not sure how often does that happen for the memory cache not being 
able to split everythig, but I guess it's only when memory is tight in the 
system. (Haven't checked, I can be wrong here)

But in any case, the idea of optimizing this was discarted because it ended 
up being faster to just skip by level.


> > 
> > This patches' idea is to introduce new walking flags to skip pagetable
> > levels 0-3.
> > 
> > The idea of skipping child nodes was also tested, but it was marginally
> > slower than just skipping levels, so it was discarted.

^ Here

Thanks!
Leo

> > 
> > Optimization measured on two scenarios involving eager-splitting on a
> > VM with 32 memslot of 2GB (total 64GB), and vcpu per slot:
> > - Scenario 1: No manual protect, whole memslot split at dirty-track enable
> >   (KVM_SET_USER_MEMORY_REGION2 ioctl with KVM_MEM_LOG_DIRTY_PAGES)
> >   - Split happens only once, whole region
> >   - Evalutes improved batch performance of splitting
> > - Scenario 2: Manual protect, split happens during every dirty-bit clean
> >   (KVM_CLEAR_DIRTY_LOG ioctl), average for 2 iterations.
> >   - Split called multiple times, for smaller 64-page sections.
> >   - Evaluate improved performance for multiple calls
> > 
> > Scenario 1, improvement on dirty-track enable ioctl for the memslot:
> > - Memory was already split (4k pages):  -43.18% runtime (stdev 3.87%)
> > - THP backed memory:                    -25.77% runtime (stdev 1.01%)
> > - 64x1GB hugetlb memory:                -25.62% runtime (stdev 1.09%)
> > 
> > Scenario 2, improvement on dirty-log clean ioctl for the memslot:
> > - Memory was already split (4k pages):  -39.88% runtime (stdev 2.71%)
> > - THP backed memory:                    -25.18% runtime (stdev 0.32%)
> > - 64x1GB hugetlb memory:                -49.17% runtime (stdev 1.86%)
> > 
> > For collecting above numbers, the following script was ran in both vanilla
> > and patched kernels, with kernel parameter 'default_hugepagesz=1G', on an
> > TX2 with 128GB RAM.
> > 
> > --- dirty_test.sh
> > #!/bin/bash
> > filename=$(uname -r |cut -d'-' -f 4-)
> > 
> > run_test(){
> >   base_test="./dirty_log_perf_test -b 2G -v 32 -m 6 -m 8"
> > 
> >   # Manual cleaning disable
> >   ${base_test} -g
> >   ${base_test} -g -s anonymous_thp
> >   echo 64 > /proc/sys/vm/nr_hugepages  
> >   ${base_test} -g -s shared_hugetlb
> >   echo 0 > /proc/sys/vm/nr_hugepages
> > 
> >   # Manual cleaning enable
> >   ${base_test}
> >   ${base_test} -s anonymous_thp
> >   echo 64 > /proc/sys/vm/nr_hugepages  
> >   ${base_test} -s shared_hugetlb
> >   echo 0 > /proc/sys/vm/nr_hugepages
> > }
> > 
> > run_test 2>&1 | tee ${filename}
> > ---
> > 
> > Above dirty_log_perf_test command is the standard kvm selftest found in the
> > kernel tree. It tested the following guest modes:
> > Testing guest mode: PA-bits:40,  VA-bits:48,  4K pages
> > Testing guest mode: PA-bits:40,  VA-bits:48, 64K pages
> > (Modes with PA-bits:36 were discarted in this version, given the amount
> > of RAM being used for testing, and the similarity of previous results)
> > 
> > Performance numbers from above modes were used to calculate average and
> > stdev showed in the optimization results.
> > 
> > Changes since v2:
> > - Rebased on top of v7.2-rc1
> > - Improved testing, added more memory, re-tested
> >   - Now: 32 vcpus @ total of 64G
> >   - Before: 1cpu @ 16G
> > 
> > Changes since v1:
> > - Fixed inverted flag verification priority (Sashiko)
> > - Fixed incorrectly skipping POST call if level was skipped (Sashiko), and to that
> > - New pre-patch that changes goto-out -> return to avoid re-testing walk_continue 
> > v1 Link: https://lore.kernel.org/lkml/20260610202112.2695205-2-leo.bras@arm.com/
> > 
> > Changes since RFC:
> > - Changed approach from return value to walk flags (Will Deacon)
> > - Discarted skip_child approach (Oliver Upton)
> > - Measured in real hardware, and from userspace perspective (Marc Zyngier)
> > - Better explanation of what and how numbers were collected
> > RFC Link: https://lore.kernel.org/all/20260515195904.2466381-1-leo.bras@arm.com/
> > 
> > Thanks!
> > Leo
> > 
> > 
> > Leonardo Bras (3):
> >   KVM: arm64: Avoid re-testing walk_continue
> >   KVM: arm64: Introduce KVM_PGTABLE_WALK_SKIP_LEVEL* walk flags
> >   KVM: arm64: Make stage2_split_walker() skip unnecessary walks
> > 
> >  arch/arm64/include/asm/kvm_pgtable.h | 13 +++++++++++++
> >  arch/arm64/kvm/hyp/pgtable.c         | 28 +++++++++++++++++++++-------
> >  2 files changed, 34 insertions(+), 7 deletions(-)
> > 
> > 
> > base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
>