Message-ID: <20190813181938.GA4196@hpe.com>
Date:   Tue, 13 Aug 2019 13:19:38 -0500
From:   Dimitri Sivanich <sivanich@....com>
To:     Bharath Vedartham <linux.bhar@...il.com>
Cc:     jhubbard@...dia.com, gregkh@...uxfoundation.org, arnd@...db.de,
        ira.weiny@...el.com, jglisse@...hat.com,
        william.kucharski@...cle.com, hch@....de,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        linux-kernel-mentees@...ts.linuxfoundation.org, sivanich@....com
Subject: Re: [Linux-kernel-mentees][PATCH v5 1/1] sgi-gru: Remove *pte_lookup
 functions, Convert to get_user_page*()

On Tue, Aug 13, 2019 at 10:53:01PM +0530, Bharath Vedartham wrote:
> On Tue, Aug 13, 2019 at 09:50:29AM -0500, Dimitri Sivanich wrote:
> > Bharath,
> > 
> > I do not believe that __get_user_pages_fast will work for the atomic case, as
> > there is no guarantee that the 'current->mm' will be the correct one for the
> > process in question, as the process might have moved away from the cpu that is
> > handling interrupts for it's context.
> So what your saying is, there may be cases where current->mm != gts->ts_mm
> right? __get_user_pages_fast and get_user_pages do assume current->mm.

Correct, in the case of atomic context.

> 
> These changes were inspired a bit from kvm. In kvm/kvm_main.c,
> hva_to_pfn_fast uses __get_user_pages_fast. THe comment above the
> function states it runs in atomic context.
> 
> Just curious, get_user_pages also uses current->mm. Do you think that is
> also an issue? 

Not in non-atomic context.  Notice that it is currently done that way.

> 
> Do you feel using get_user_pages_remote would be a better idea? We can
> specify the mm_struct in get_user_pages_remote?

>From that standpoint maybe, but is it safe in interrupt context?

> 
> Thank you
> Bharath
> > On Sat, Aug 10, 2019 at 01:08:17AM +0530, Bharath Vedartham wrote:
> > > For pages that were retained via get_user_pages*(), release those pages
> > > via the new put_user_page*() routines, instead of via put_page() or
> > > release_pages().
> > > 
> > > This is part a tree-wide conversion, as described in commit fc1d8e7cca2d
> > > ("mm: introduce put_user_page*(), placeholder versions").
> > > 
> > > As part of this conversion, the *pte_lookup functions can be removed and
> > > be easily replaced with get_user_pages_fast() functions. In the case of
> > > atomic lookup, __get_user_pages_fast() is used, because it does not fall
> > > back to the slow path: get_user_pages(). get_user_pages_fast(), on the other
> > > hand, first calls __get_user_pages_fast(), but then falls back to the
> > > slow path if __get_user_pages_fast() fails.
> > > 
> > > Also: remove unnecessary CONFIG_HUGETLB ifdefs.
> > > 
> > > Cc: Ira Weiny <ira.weiny@...el.com>
> > > Cc: John Hubbard <jhubbard@...dia.com>
> > > Cc: Jérôme Glisse <jglisse@...hat.com>
> > > Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
> > > Cc: Dimitri Sivanich <sivanich@....com>
> > > Cc: Arnd Bergmann <arnd@...db.de>
> > > Cc: William Kucharski <william.kucharski@...cle.com>
> > > Cc: Christoph Hellwig <hch@....de>
> > > Cc: linux-kernel@...r.kernel.org
> > > Cc: linux-mm@...ck.org
> > > Cc: linux-kernel-mentees@...ts.linuxfoundation.org
> > > Reviewed-by: Ira Weiny <ira.weiny@...el.com>
> > > Reviewed-by: John Hubbard <jhubbard@...dia.com>
> > > Reviewed-by: William Kucharski <william.kucharski@...cle.com>
> > > Signed-off-by: Bharath Vedartham <linux.bhar@...il.com>
> > > ---
> > > This is a fold of the 3 patches in the v2 patch series.
> > > The review tags were given to the individual patches.
> > > 
> > > Changes since v3
> > > 	- Used gup flags in get_user_pages_fast rather than
> > > 	boolean flags.
> > > Changes since v4
> > > 	- Updated changelog according to John Hubbard.
> > > ---
> > >  drivers/misc/sgi-gru/grufault.c | 112 +++++++++-------------------------------
> > >  1 file changed, 24 insertions(+), 88 deletions(-)
> > > 
> > > diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
> > > index 4b713a8..304e9c5 100644
> > > --- a/drivers/misc/sgi-gru/grufault.c
> > > +++ b/drivers/misc/sgi-gru/grufault.c
> > > @@ -166,96 +166,20 @@ static void get_clear_fault_map(struct gru_state *gru,
> > >  }
> > >  
> > >  /*
> > > - * Atomic (interrupt context) & non-atomic (user context) functions to
> > > - * convert a vaddr into a physical address. The size of the page
> > > - * is returned in pageshift.
> > > - * 	returns:
> > > - * 		  0 - successful
> > > - * 		< 0 - error code
> > > - * 		  1 - (atomic only) try again in non-atomic context
> > > - */
> > > -static int non_atomic_pte_lookup(struct vm_area_struct *vma,
> > > -				 unsigned long vaddr, int write,
> > > -				 unsigned long *paddr, int *pageshift)
> > > -{
> > > -	struct page *page;
> > > -
> > > -#ifdef CONFIG_HUGETLB_PAGE
> > > -	*pageshift = is_vm_hugetlb_page(vma) ? HPAGE_SHIFT : PAGE_SHIFT;
> > > -#else
> > > -	*pageshift = PAGE_SHIFT;
> > > -#endif
> > > -	if (get_user_pages(vaddr, 1, write ? FOLL_WRITE : 0, &page, NULL) <= 0)
> > > -		return -EFAULT;
> > > -	*paddr = page_to_phys(page);
> > > -	put_page(page);
> > > -	return 0;
> > > -}
> > > -
> > > -/*
> > > - * atomic_pte_lookup
> > > + * mmap_sem is already helod on entry to this function. This guarantees
> > > + * existence of the page tables.
> > >   *
> > > - * Convert a user virtual address to a physical address
> > >   * Only supports Intel large pages (2MB only) on x86_64.
> > > - *	ZZZ - hugepage support is incomplete
> > > - *
> > > - * NOTE: mmap_sem is already held on entry to this function. This
> > > - * guarantees existence of the page tables.
> > > + *	ZZZ - hugepage support is incomplete.
> > >   */
> > > -static int atomic_pte_lookup(struct vm_area_struct *vma, unsigned long vaddr,
> > > -	int write, unsigned long *paddr, int *pageshift)
> > > -{
> > > -	pgd_t *pgdp;
> > > -	p4d_t *p4dp;
> > > -	pud_t *pudp;
> > > -	pmd_t *pmdp;
> > > -	pte_t pte;
> > > -
> > > -	pgdp = pgd_offset(vma->vm_mm, vaddr);
> > > -	if (unlikely(pgd_none(*pgdp)))
> > > -		goto err;
> > > -
> > > -	p4dp = p4d_offset(pgdp, vaddr);
> > > -	if (unlikely(p4d_none(*p4dp)))
> > > -		goto err;
> > > -
> > > -	pudp = pud_offset(p4dp, vaddr);
> > > -	if (unlikely(pud_none(*pudp)))
> > > -		goto err;
> > > -
> > > -	pmdp = pmd_offset(pudp, vaddr);
> > > -	if (unlikely(pmd_none(*pmdp)))
> > > -		goto err;
> > > -#ifdef CONFIG_X86_64
> > > -	if (unlikely(pmd_large(*pmdp)))
> > > -		pte = *(pte_t *) pmdp;
> > > -	else
> > > -#endif
> > > -		pte = *pte_offset_kernel(pmdp, vaddr);
> > > -
> > > -	if (unlikely(!pte_present(pte) ||
> > > -		     (write && (!pte_write(pte) || !pte_dirty(pte)))))
> > > -		return 1;
> > > -
> > > -	*paddr = pte_pfn(pte) << PAGE_SHIFT;
> > > -#ifdef CONFIG_HUGETLB_PAGE
> > > -	*pageshift = is_vm_hugetlb_page(vma) ? HPAGE_SHIFT : PAGE_SHIFT;
> > > -#else
> > > -	*pageshift = PAGE_SHIFT;
> > > -#endif
> > > -	return 0;
> > > -
> > > -err:
> > > -	return 1;
> > > -}
> > > -
> > >  static int gru_vtop(struct gru_thread_state *gts, unsigned long vaddr,
> > >  		    int write, int atomic, unsigned long *gpa, int *pageshift)
> > >  {
> > >  	struct mm_struct *mm = gts->ts_mm;
> > >  	struct vm_area_struct *vma;
> > >  	unsigned long paddr;
> > > -	int ret, ps;
> > > +	int ret;
> > > +	struct page *page;
> > >  
> > >  	vma = find_vma(mm, vaddr);
> > >  	if (!vma)
> > > @@ -263,21 +187,33 @@ static int gru_vtop(struct gru_thread_state *gts, unsigned long vaddr,
> > >  
> > >  	/*
> > >  	 * Atomic lookup is faster & usually works even if called in non-atomic
> > > -	 * context.
> > > +	 * context. get_user_pages_fast does atomic lookup before falling back to
> > > +	 * slow gup.
> > >  	 */
> > >  	rmb();	/* Must/check ms_range_active before loading PTEs */
> > > -	ret = atomic_pte_lookup(vma, vaddr, write, &paddr, &ps);
> > > -	if (ret) {
> > > -		if (atomic)
> > > +	if (atomic) {
> > > +		ret = __get_user_pages_fast(vaddr, 1, write, &page);
> > > +		if (!ret)
> > >  			goto upm;
> > > -		if (non_atomic_pte_lookup(vma, vaddr, write, &paddr, &ps))
> > > +	} else {
> > > +		ret = get_user_pages_fast(vaddr, 1, write ? FOLL_WRITE : 0, &page);
> > > +		if (!ret)
> > >  			goto inval;
> > >  	}
> > > +
> > > +	paddr = page_to_phys(page);
> > > +	put_user_page(page);
> > > +
> > > +	if (unlikely(is_vm_hugetlb_page(vma)))
> > > +		*pageshift = HPAGE_SHIFT;
> > > +	else
> > > +		*pageshift = PAGE_SHIFT;
> > > +
> > >  	if (is_gru_paddr(paddr))
> > >  		goto inval;
> > > -	paddr = paddr & ~((1UL << ps) - 1);
> > > +	paddr = paddr & ~((1UL << *pageshift) - 1);
> > >  	*gpa = uv_soc_phys_ram_to_gpa(paddr);
> > > -	*pageshift = ps;
> > > +
> > >  	return VTOP_SUCCESS;
> > >  
> > >  inval:
> > > -- 
> > > 2.7.4
> > > 

Powered by blists - more mailing lists