Re: [PATCH v2] mm/vmstat: add events for ksm cow

[CGEL]

On Wed, Mar 23, 2022 at 07:43:04PM +0100, David Hildenbrand wrote:
> On 23.03.22 08:57, cgel.zte@xxxxxxxxx wrote:
> > From: Yang Yang <yang.yang29@xxxxxxxxxx>
> > 
> > Users may use ksm by calling madvise(, , MADV_MERGEABLE) when they want
> > to save memory, it's a tradeoff by suffering delay on ksm cow. Users can
> > get to know how much memory ksm saved by reading
> > /sys/kernel/mm/ksm/pages_sharing, but they don't know what's the costs
> > of ksm cow, and this is important of some delay sensitive tasks.
> > 
> > So add ksm cow events to help users evaluate whether or how to use ksm.
> > 
> > Signed-off-by: Yang Yang <yang.yang29@xxxxxxxxxx>
> > Reviewed-by: xu xin <xu.xin16@xxxxxxxxxx>
> > Reviewed-by: Ran Xiaokai <ran.xiaokai@xxxxxxxxxx>
> > ---
> > v2:
> > - fix compile error when CONFIG_KSM is not set
> > ---
> >  include/linux/vm_event_item.h |  2 ++
> >  mm/memory.c                   | 20 +++++++++++++++++---
> >  mm/vmstat.c                   |  2 ++
> >  3 files changed, 21 insertions(+), 3 deletions(-)
> > 
> > diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
> > index 16a0a4fd000b..6f32be04212f 100644
> > --- a/include/linux/vm_event_item.h
> > +++ b/include/linux/vm_event_item.h
> > @@ -131,6 +131,8 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
> >  		SWAP_RA_HIT,
> >  #ifdef CONFIG_KSM
> >  		KSM_SWPIN_COPY,
> > +		KSM_COW_SUCCESS,
> > +		KSM_COW_FAIL,
> >  #endif
> >  #endif
> >  #ifdef CONFIG_X86
> > diff --git a/mm/memory.c b/mm/memory.c
> > index 4111f97c91a0..c24d5f04fab5 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -3257,6 +3257,8 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
> >  	__releases(vmf->ptl)
> >  {
> >  	struct vm_area_struct *vma = vmf->vma;
> > +	vm_fault_t ret = 0;
> > +	bool ksm = 0;
> >  
> >  	if (userfaultfd_pte_wp(vma, *vmf->pte)) {
> >  		pte_unmap_unlock(vmf->pte, vmf->ptl);
> > @@ -3294,6 +3296,7 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
> >  	 */
> >  	if (PageAnon(vmf->page)) {
> >  		struct page *page = vmf->page;
> > +		ksm = PageKsm(page);
> >  
> >  		/*
> >  		 * We have to verify under page lock: these early checks are
> > @@ -3302,7 +3305,7 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
> >  		 *
> >  		 * PageKsm() doesn't necessarily raise the page refcount.
> >  		 */
> > -		if (PageKsm(page) || page_count(page) > 3)
> > +		if (ksm || page_count(page) > 3)
> >  			goto copy;
> >  		if (!PageLRU(page))
> >  			/*
> > @@ -3316,7 +3319,7 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
> >  			goto copy;
> >  		if (PageSwapCache(page))
> >  			try_to_free_swap(page);
> > -		if (PageKsm(page) || page_count(page) != 1) {
> > +		if (ksm || page_count(page) != 1) {
> 
> I think we really want to recheck here, after locking the page.
> Consequently, just do a PageKsm() check below.
> 
> >  			unlock_page(page);
> >  			goto copy;
> >  		}
> > @@ -3339,7 +3342,18 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
> >  	get_page(vmf->page);
> >  
> >  	pte_unmap_unlock(vmf->pte, vmf->ptl);
> > -	return wp_page_copy(vmf);
> > +	ret = wp_page_copy(vmf);
> > +
> > +#ifdef CONFIG_KSM
> > +	if (ksm) {
> > +		if (unlikely(ret & VM_FAULT_ERROR))
> > +			count_vm_event(KSM_COW_FAIL);
> > +		else
> > +			count_vm_event(KSM_COW_SUCCESS);
> > +	}
> > +#endif
> 
> Do we really care if we failed or not? I mean, the failure case will
> usually make your app crash either way ... due to OOM.
>
I think we need failed count. Because ksm cow oom is a little different
from general OOM. User may wonder I am not allocing new memory, why it
cause OOM? And OOM may have big impact for some kind of tasks, so we
better let user know that, do we?
>
> Doing
> 
> #ifdef CONFIG_KSM
> if (PageKsm(page))
> 	count_vm_event(COW_KSM);
> #endif
> 
> before the wp_page_copy(vmf) should be good enough, no?
> 
> Should be good enough IMHO.
> 
> -- 
> Thanks,
> 
> David / dhildenb