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:   Wed, 22 Jun 2022 10:52:10 -0700
From:   Zach O'Keefe <zokeefe@google.com>
To:     Andrew Morton <akpm@linux-foundation.org>
Cc:     Yang Shi <shy828301@gmail.com>, vbabka@suse.cz,
        kirill.shutemov@linux.intel.com, willy@infradead.org,
        linmiaohe@huawei.com, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [v5 PATCH 4/7] mm: thp: kill transparent_hugepage_active()
Message-ID: <YrNWyqxIKsQq1ky5@google.com>
References: <20220616174840.1202070-1-shy828301@gmail.com>
 <20220616174840.1202070-5-shy828301@gmail.com>
 <YrIU2iP0H5LQbV7R@google.com>
 <20220621175412.02d90df5f381de0cd676adf8@linux-foundation.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220621175412.02d90df5f381de0cd676adf8@linux-foundation.org>
Precedence: bulk

On 21 Jun 17:54, Andrew Morton wrote:
> On Tue, 21 Jun 2022 11:58:34 -0700 "Zach O'Keefe" <zokeefe@google.com> wrote:
> 
> > > -bool transparent_hugepage_active(struct vm_area_struct *vma)
> > > +bool hugepage_vma_check(struct vm_area_struct *vma,
> > > +			unsigned long vm_flags,
> > > +			bool smaps)
> > >  {
> > > -	/* The addr is used to check if the vma size fits */
> > > -	unsigned long addr = (vma->vm_end & HPAGE_PMD_MASK) - HPAGE_PMD_SIZE;
> > > +	if (!transhuge_vma_enabled(vma, vm_flags))
> > > +		return false;
> > > +
> > 
> > During testing my work on top this patch, I found a small bug here.
> > 
> > Namely, transhuge_vma_enabled() will check vma->vm_mm->flags (to see if
> > MMF_DISABLE_THP is set); however, for vDSO vmas, vma->vm_mm is NULL.
> > 
> > Previously, transparent_hugepage_active() in smaps path would check
> > transhuge_vma_suitable() before checking these flags, which would fail for vDSO
> > vma since we'd take the !vma_is_anonymous() branch and find the vma (most
> > likely) wasn't suitably aligned (by chance ?).
> > 
> > Anyways, I think we need to check vma->vm_mm.
> 
> Like this?
> 
> --- a/mm/huge_memory.c~mm-thp-kill-transparent_hugepage_active-fix
> +++ a/mm/huge_memory.c
> @@ -73,6 +73,9 @@ bool hugepage_vma_check(struct vm_area_s
>  			unsigned long vm_flags,
>  			bool smaps)
>  {
> +	if (!vma->vm_mm)
> +		return false;
> +
>  	if (!transhuge_vma_enabled(vma, vm_flags))
>  		return false;
>  
> _
> 

Hey Andrew,

In principle, yes that would fix this. I don't know precisely how this fix will
be applied, but note that the subsequent patch "mm: thp: kill
__transhuge_page_enabled()" won't apply on top of this automatically. Also, I
wonder if we should add a comment for future travellers who wonder what kind of
vmas don't have an associated mm (it was news to me); though, I'm not sure if
vDSO is the only such case (though show_map_vma() seems to think so), or if this
just asking for stale comments down the road. Maybe it's fine as is.

Thanks,
Zach