Received-SPF: pass (google.com: domain of linux-kernel+bounces-184680-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Tue, 21 May 2024 10:06:20 +0100
From: Vincent Donnefort <vdonnefort@google.com>
To: David Hildenbrand <david@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	Andrew Morton <akpm@linux-foundation.org>,
	Dan Williams <dan.j.williams@intel.com>, rostedt@goodmis.org
Subject: Re: [PATCH v1 1/2] mm/memory: cleanly support zeropage in
 vm_insert_page*(), vm_map_pages*() and vmf_insert_mixed()
Message-ID: <ZkxkDPnPiQzPEm-0@google.com>
References: <20240430204044.52755-1-david@redhat.com>
 <20240430204044.52755-2-david@redhat.com>
 <Zkdys7YKC5pe1vAu@google.com>
 <3decc6c8-9035-44d6-89c6-8d42a5e0bc40@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <3decc6c8-9035-44d6-89c6-8d42a5e0bc40@redhat.com>

On Tue, May 21, 2024 at 10:25:43AM +0200, David Hildenbrand wrote:
> On 17.05.24 17:07, Vincent Donnefort wrote:
> > Hi David,
> > 
> > [...]
> > 
> > > -static int validate_page_before_insert(struct page *page)
> > > +static bool vm_mixed_zeropage_allowed(struct vm_area_struct *vma)
> > > +{
> > > +	VM_WARN_ON_ONCE(vma->vm_flags & VM_PFNMAP);
> > > +	/*
> > > +	 * Whoever wants to forbid the zeropage after some zeropages
> > > +	 * might already have been mapped has to scan the page tables and
> > > +	 * bail out on any zeropages. Zeropages in COW mappings can
> > > +	 * be unshared using FAULT_FLAG_UNSHARE faults.
> > > +	 */
> > > +	if (mm_forbids_zeropage(vma->vm_mm))
> > > +		return false;
> > > +	/* zeropages in COW mappings are common and unproblematic. */
> > > +	if (is_cow_mapping(vma->vm_flags))
> > > +		return true;
> > > +	/* Mappings that do not allow for writable PTEs are unproblematic. */
> > > +	if (!(vma->vm_flags & (VM_WRITE | VM_MAYWRITE)))
> > > +		return false;
> > 
> > Shouldn't we return true here?
> 
> Indeed, thanks! I wish we would have user in the tree already that could
> exercise that code path.

I have a patch ready to use this path from the memory map tracing! I can either
send it once this one is picked-up or you can add it to your series?

> 
> [...]
> 
> > > @@ -2043,7 +2085,7 @@ static int insert_page_in_batch_locked(struct vm_area_struct *vma, pte_t *pte,
> > >   	if (!page_count(page))
> > >   		return -EINVAL;
> > 
> > This test here prevents inserting the zero-page.
> 
> You mean the existing page_count() check? or the (wrong) vma->vm_flags check
> in vm_mixed_zeropage_allowed() ?

I meant this page_count() here. As a quick test, I removed that check (also fixed
the vm_flags above) and the zero-page was properly mapped!

> 
> -- 
> Cheers,
> 
> David / dhildenb
>