Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 9 Oct 2019 18:27:37 +0300
From:   "Kirill A. Shutemov" <kirill@shutemov.name>
To:     Thomas =?utf-8?Q?Hellstr=C3=B6m_=28VMware=29?= 
        <thomas_os@shipmail.org>
Cc:     linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        torvalds@linux-foundation.org,
        Thomas Hellstrom <thellstrom@vmware.com>,
        Matthew Wilcox <willy@infradead.org>,
        Will Deacon <will.deacon@arm.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Rik van Riel <riel@surriel.com>,
        Minchan Kim <minchan@kernel.org>,
        Michal Hocko <mhocko@suse.com>,
        Huang Ying <ying.huang@intel.com>,
        =?utf-8?B?SsOpcsO0bWU=?= Glisse <jglisse@redhat.com>
Subject: Re: [PATCH v4 3/9] mm: pagewalk: Don't split transhuge pmds when a
 pmd_entry is present
Message-ID: <20191009152737.p42w7w456zklxz72@box>
References: <20191008091508.2682-1-thomas_os@shipmail.org>
 <20191008091508.2682-4-thomas_os@shipmail.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20191008091508.2682-4-thomas_os@shipmail.org>
User-Agent: NeoMutt/20180716
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Oct 08, 2019 at 11:15:02AM +0200, Thomas Hellstr?m (VMware) wrote:
> From: Thomas Hellstrom <thellstrom@vmware.com>
> 
> The pagewalk code was unconditionally splitting transhuge pmds when a
> pte_entry was present. However ideally we'd want to handle transhuge pmds
> in the pmd_entry function and ptes in pte_entry function. So don't split
> huge pmds when there is a pmd_entry function present, but let the callback
> take care of it if necessary.

Do we have any current user that expect split_huge_pmd() in this scenario.

> 
> In order to make sure a virtual address range is handled by one and only
> one callback, and since pmd entries may be unstable, we introduce a
> pmd_entry return code that tells the walk code to continue processing this
> pmd entry rather than to move on. Since caller-defined positive return
> codes (up to 2) are used by current callers, use a high value that allows a
> large range of positive caller-defined return codes for future users.
> 
> Cc: Matthew Wilcox <willy@infradead.org>
> Cc: Will Deacon <will.deacon@arm.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Rik van Riel <riel@surriel.com>
> Cc: Minchan Kim <minchan@kernel.org>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: Huang Ying <ying.huang@intel.com>
> Cc: J?r?me Glisse <jglisse@redhat.com>
> Cc: Kirill A. Shutemov <kirill@shutemov.name>
> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
> Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
> ---
>  include/linux/pagewalk.h |  8 ++++++++
>  mm/pagewalk.c            | 28 +++++++++++++++++++++-------
>  2 files changed, 29 insertions(+), 7 deletions(-)
> 
> diff --git a/include/linux/pagewalk.h b/include/linux/pagewalk.h
> index bddd9759bab9..c4a013eb445d 100644
> --- a/include/linux/pagewalk.h
> +++ b/include/linux/pagewalk.h
> @@ -4,6 +4,11 @@
>  
>  #include <linux/mm.h>
>  
> +/* Highest positive pmd_entry caller-specific return value */
> +#define PAGE_WALK_CALLER_MAX     (INT_MAX / 2)
> +/* The handler did not handle the entry. Fall back to the next level */
> +#define PAGE_WALK_FALLBACK       (PAGE_WALK_CALLER_MAX + 1)
> +

That's hacky.

Maybe just use an error code for this? -EAGAIN?

>  struct mm_walk;
>  
>  /**
> @@ -16,6 +21,9 @@ struct mm_walk;
>   *			this handler is required to be able to handle
>   *			pmd_trans_huge() pmds.  They may simply choose to
>   *			split_huge_page() instead of handling it explicitly.
> + *                      If the handler did not handle the PMD, or split the
> + *                      PMD and wants it handled by the PTE handler, it
> + *                      should return PAGE_WALK_FALLBACK.

Indentation is broken. Use tabs.

>   * @pte_entry:		if set, called for each non-empty PTE (4th-level) entry
>   * @pte_hole:		if set, called for each hole at all levels
>   * @hugetlb_entry:	if set, called for each hugetlb entry
> diff --git a/mm/pagewalk.c b/mm/pagewalk.c
> index 83c0b78363b4..f844c2a2aa60 100644
> --- a/mm/pagewalk.c
> +++ b/mm/pagewalk.c
> @@ -50,10 +50,18 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
>  		 * This implies that each ->pmd_entry() handler
>  		 * needs to know about pmd_trans_huge() pmds
>  		 */
> -		if (ops->pmd_entry)
> +		if (ops->pmd_entry) {
>  			err = ops->pmd_entry(pmd, addr, next, walk);
> -		if (err)
> -			break;
> +			if (!err)
> +				continue;
> +			else if (err <= PAGE_WALK_CALLER_MAX)
> +				break;
> +			WARN_ON(err != PAGE_WALK_FALLBACK);
> +			err = 0;
> +			if (pmd_trans_unstable(pmd))
> +				goto again;
> +			/* Fall through */
> +		}
>  
>  		/*
>  		 * Check this here so we only break down trans_huge
> @@ -61,8 +69,8 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
>  		 */
>  		if (!ops->pte_entry)
>  			continue;
> -
> -		split_huge_pmd(walk->vma, pmd, addr);
> +		if (!ops->pmd_entry)
> +			split_huge_pmd(walk->vma, pmd, addr);
>  		if (pmd_trans_unstable(pmd))
>  			goto again;
>  		err = walk_pte_range(pmd, addr, next, walk);
> @@ -281,11 +289,17 @@ static int __walk_page_range(unsigned long start, unsigned long end,
>   *
>   *  - 0  : succeeded to handle the current entry, and if you don't reach the
>   *         end address yet, continue to walk.
> - *  - >0 : succeeded to handle the current entry, and return to the caller
> - *         with caller specific value.
> + *  - >0, and <= PAGE_WALK_CALLER_MAX : succeeded to handle the current entry,
> + *         and return to the caller with caller specific value.
>   *  - <0 : failed to handle the current entry, and return to the caller
>   *         with error code.
>   *
> + * For pmd_entry(), a value <= PAGE_WALK_CALLER_MAX indicates that the entry
> + * was handled by the callback. PAGE_WALK_FALLBACK indicates that the entry
> + * could not be handled by the callback and should be re-checked. If the
> + * callback needs the entry to be handled by the next level, it should
> + * split the entry and then return PAGE_WALK_FALLBACK.
> + *
>   * Before starting to walk page table, some callers want to check whether
>   * they really want to walk over the current vma, typically by checking
>   * its vm_flags. walk_page_test() and @ops->test_walk() are used for this
> -- 
> 2.21.0
> 

-- 
 Kirill A. Shutemov