On Wed, Jan 27, 2021 at 8:57 AM SeongJae Park <sjp...@amazon.com> wrote:
>
> On Thu, 24 Dec 2020 08:11:11 +0100 SeongJae Park <sjp...@amazon.com> wrote:
>
> > On Wed, 23 Dec 2020 14:54:02 -0800 Shakeel Butt <shake...@google.com> wrote:
> >
> > > On Wed, Dec 23, 2020 at 8:47 AM SeongJae Park <sjp...@amazon.com> wrote:
> > > >
> > > [snip]
> > > > > [snip]
> > > > > > +
> > > > > > +static bool damon_va_young(struct mm_struct *mm, unsigned long
> > > > > > addr,
> > > > > > + unsigned long *page_sz)
> > > > > > +{
> > > > > > + pte_t *pte = NULL;
> > > > > > + pmd_t *pmd = NULL;
> > > > > > + spinlock_t *ptl;
> > > > > > + bool young = false;
> > > > > > +
> > > > > > + if (follow_pte_pmd(mm, addr, NULL, &pte, &pmd, &ptl))
> > > > > > + return false;
> > > > > > +
> > > > > > + *page_sz = PAGE_SIZE;
> > > > > > + if (pte) {
> > > > > > + young = pte_young(*pte);
> > > > > > + if (!young)
> > > > > > + young = !page_is_idle(pte_page(*pte));
> > > > > > + pte_unmap_unlock(pte, ptl);
> > > > > > + return young;
> > > > > > + }
> > > > > > +
> > > > > > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > > > > > + young = pmd_young(*pmd);
> > > > > > + if (!young)
> > > > > > + young = !page_is_idle(pmd_page(*pmd));
> > > > > > + spin_unlock(ptl);
> > > > > > + *page_sz = ((1UL) << HPAGE_PMD_SHIFT);
> > > > > > +#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
> > > > > > +
> > > > > > + return young;
> > > > >
> > > > > You need mmu_notifier_test_young() here. Hmm I remember mentioning
> > > > > this in some previous version as well.
> > > >
> > > > Your question and my answer was as below:
> > > >
> > > > > Don't you need mmu_notifier_clear_young() here?
> > > >
> > > > I think we don't need it here because we only read the Accessed bit
> > > > and PG_Idle
> > > > if Accessed bit was not set.
> > > >
> > > > I should notice that you mean 'test_young()' but didn't, sorry. I will
> > > > add it
> > > > in the next version.
> > > >
> > >
> > > I should have said mmu_notifier_test_young() instead of
> > > mmu_notifier_clear_young().
> > >
> > > > >
> > > > > BTW have you tested this on a VM?
> > > >
> > > > Yes. Indeed, I'm testing this on a QEMU/KVM environment. You can get
> > > > more
> > > > detail at:
> > > > https://damonitor.github.io/doc/html/latest/vm/damon/eval.html#setup
> > > >
> > >
> > > Hmm without mmu_notifier_test_young() you should be missing the kvm
> > > mmu access updates. Can you please recheck if your eval is correctly
> > > seeing the memory accesses from the VM?
> >
> > Seems I didn't clearly answered, sorry. My test setup installs the
> > DAMON-enabled kernel in a guest VM and run it for workloads in the guest,
> > rather than running DAMON in host to monitor accesses of VMs. The MMU
> > notifier
> > is for latter case, AFAIU, so my test setup didn't see the problem.
>
> Just FYI. I confirmed the mmu_notifier_test_young() added version works for
> the use case. I tested it by running a program accessing 200MB memory in a
> QEMU/KVM guest having 120GB memory and monitoring the qemu process' virtual
> address space from the host using DAMON. The 200MB memory region was clearly
> identifiable.
>
Thanks for confirming. I am still going over the whole series and will
send out the emails in a couple of days.
Shakeel
