Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 1 Sep 2022 19:26:54 -0400
From:   Peter Xu <peterx@redhat.com>
To:     Yang Shi <shy828301@gmail.com>
Cc:     david@redhat.com, kirill.shutemov@linux.intel.com,
        jhubbard@nvidia.com, jgg@nvidia.com, hughd@google.com,
        akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm: gup: fix the fast GUP race against THP collapse
Message-ID: <YxE/vuQlWJCJMuG2@xz-m1.local>
References: <20220901222707.477402-1-shy828301@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20220901222707.477402-1-shy828301@gmail.com>
Precedence: bulk

Hi, Yang,

On Thu, Sep 01, 2022 at 03:27:07PM -0700, Yang Shi wrote:
> Since general RCU GUP fast was introduced in commit 2667f50e8b81 ("mm:
> introduce a general RCU get_user_pages_fast()"), a TLB flush is no longer
> sufficient to handle concurrent GUP-fast in all cases, it only handles
> traditional IPI-based GUP-fast correctly.

If TLB flush (or, IPI broadcasts) used to work to protect against gup-fast,
I'm kind of confused why it's not sufficient even if with RCU gup?  Isn't
that'll keep working as long as interrupt disabled (which current fast-gup
will still do)?

IIUC the issue is you suspect not all archs correctly implemented
pmdp_collapse_flush(), or am I wrong?

> On architectures that send
> an IPI broadcast on TLB flush, it works as expected.  But on the
> architectures that do not use IPI to broadcast TLB flush, it may have
> the below race:
> 
>    CPU A                                          CPU B
> THP collapse                                     fast GUP
>                                               gup_pmd_range() <-- see valid pmd
>                                                   gup_pte_range() <-- work on pte
> pmdp_collapse_flush() <-- clear pmd and flush
> __collapse_huge_page_isolate()
>     check page pinned <-- before GUP bump refcount
>                                                       pin the page
>                                                       check PTE <-- no change
> __collapse_huge_page_copy()
>     copy data to huge page
>     ptep_clear()
> install huge pmd for the huge page
>                                                       return the stale page
> discard the stale page
> 
> The race could be fixed by checking whether PMD is changed or not after
> taking the page pin in fast GUP, just like what it does for PTE.  If the
> PMD is changed it means there may be parallel THP collapse, so GUP
> should back off.

Could the race also be fixed by impl pmdp_collapse_flush() correctly for
the archs that are missing? Do you know which arch(s) is broken with it?

It's just not clear to me whether this patch is an optimization or a fix,
if it's a fix whether the IPI broadcast in ppc pmdp_collapse_flush() would
still be needed.

Thanks,

> 
> Also update the stale comment about serializing against fast GUP in
> khugepaged.
> 
> Fixes: 2667f50e8b81 ("mm: introduce a general RCU get_user_pages_fast()")
> Signed-off-by: Yang Shi <shy828301@gmail.com>
> ---
>  mm/gup.c        | 30 ++++++++++++++++++++++++------
>  mm/khugepaged.c | 10 ++++++----
>  2 files changed, 30 insertions(+), 10 deletions(-)
> 
> diff --git a/mm/gup.c b/mm/gup.c
> index f3fc1f08d90c..4365b2811269 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -2380,8 +2380,9 @@ static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start,
>  }
>  
>  #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
> -static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
> -			 unsigned int flags, struct page **pages, int *nr)
> +static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
> +			 unsigned long end, unsigned int flags,
> +			 struct page **pages, int *nr)
>  {
>  	struct dev_pagemap *pgmap = NULL;
>  	int nr_start = *nr, ret = 0;
> @@ -2423,7 +2424,23 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
>  			goto pte_unmap;
>  		}
>  
> -		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
> +		/*
> +		 * THP collapse conceptually does:
> +		 *   1. Clear and flush PMD
> +		 *   2. Check the base page refcount
> +		 *   3. Copy data to huge page
> +		 *   4. Clear PTE
> +		 *   5. Discard the base page
> +		 *
> +		 * So fast GUP may race with THP collapse then pin and
> +		 * return an old page since TLB flush is no longer sufficient
> +		 * to serialize against fast GUP.
> +		 *
> +		 * Check PMD, if it is changed just back off since it
> +		 * means there may be parallel THP collapse.
> +		 */
> +		if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) ||
> +		    unlikely(pte_val(pte) != pte_val(*ptep))) {
>  			gup_put_folio(folio, 1, flags);
>  			goto pte_unmap;
>  		}
> @@ -2470,8 +2487,9 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
>   * get_user_pages_fast_only implementation that can pin pages. Thus it's still
>   * useful to have gup_huge_pmd even if we can't operate on ptes.
>   */
> -static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
> -			 unsigned int flags, struct page **pages, int *nr)
> +static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
> +			 unsigned long end, unsigned int flags,
> +			 struct page **pages, int *nr)
>  {
>  	return 0;
>  }
> @@ -2791,7 +2809,7 @@ static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned lo
>  			if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
>  					 PMD_SHIFT, next, flags, pages, nr))
>  				return 0;
> -		} else if (!gup_pte_range(pmd, addr, next, flags, pages, nr))
> +		} else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr))
>  			return 0;
>  	} while (pmdp++, addr = next, addr != end);
>  
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 2d74cf01f694..518b49095db3 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1049,10 +1049,12 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
>  
>  	pmd_ptl = pmd_lock(mm, pmd); /* probably unnecessary */
>  	/*
> -	 * After this gup_fast can't run anymore. This also removes
> -	 * any huge TLB entry from the CPU so we won't allow
> -	 * huge and small TLB entries for the same virtual address
> -	 * to avoid the risk of CPU bugs in that area.
> +	 * This removes any huge TLB entry from the CPU so we won't allow
> +	 * huge and small TLB entries for the same virtual address to
> +	 * avoid the risk of CPU bugs in that area.
> +	 *
> +	 * Parallel fast GUP is fine since fast GUP will back off when
> +	 * it detects PMD is changed.
>  	 */
>  	_pmd = pmdp_collapse_flush(vma, address, pmd);
>  	spin_unlock(pmd_ptl);
> -- 
> 2.26.3
> 

-- 
Peter Xu