[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v2 5/9] mm/huge_memory: mark PMD mappings of the huge zero folio special


  • To: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
  • From: David Hildenbrand <david@xxxxxxxxxx>
  • Date: Thu, 17 Jul 2025 22:31:28 +0200
  • Autocrypt: addr=david@xxxxxxxxxx; keydata= xsFNBFXLn5EBEAC+zYvAFJxCBY9Tr1xZgcESmxVNI/0ffzE/ZQOiHJl6mGkmA1R7/uUpiCjJ dBrn+lhhOYjjNefFQou6478faXE6o2AhmebqT4KiQoUQFV4R7y1KMEKoSyy8hQaK1umALTdL QZLQMzNE74ap+GDK0wnacPQFpcG1AE9RMq3aeErY5tujekBS32jfC/7AnH7I0v1v1TbbK3Gp XNeiN4QroO+5qaSr0ID2sz5jtBLRb15RMre27E1ImpaIv2Jw8NJgW0k/D1RyKCwaTsgRdwuK Kx/Y91XuSBdz0uOyU/S8kM1+ag0wvsGlpBVxRR/xw/E8M7TEwuCZQArqqTCmkG6HGcXFT0V9 PXFNNgV5jXMQRwU0O/ztJIQqsE5LsUomE//bLwzj9IVsaQpKDqW6TAPjcdBDPLHvriq7kGjt WhVhdl0qEYB8lkBEU7V2Yb+SYhmhpDrti9Fq1EsmhiHSkxJcGREoMK/63r9WLZYI3+4W2rAc UucZa4OT27U5ZISjNg3Ev0rxU5UH2/pT4wJCfxwocmqaRr6UYmrtZmND89X0KigoFD/XSeVv jwBRNjPAubK9/k5NoRrYqztM9W6sJqrH8+UWZ1Idd/DdmogJh0gNC0+N42Za9yBRURfIdKSb B3JfpUqcWwE7vUaYrHG1nw54pLUoPG6sAA7Mehl3nd4pZUALHwARAQABzSREYXZpZCBIaWxk ZW5icmFuZCA8ZGF2aWRAcmVkaGF0LmNvbT7CwZgEEwEIAEICGwMGCwkIBwMCBhUIAgkKCwQW AgMBAh4BAheAAhkBFiEEG9nKrXNcTDpGDfzKTd4Q9wD/g1oFAmgsLPQFCRvGjuMACgkQTd4Q 9wD/g1o0bxAAqYC7gTyGj5rZwvy1VesF6YoQncH0yI79lvXUYOX+Nngko4v4dTlOQvrd/vhb 02e9FtpA1CxgwdgIPFKIuXvdSyXAp0xXuIuRPQYbgNriQFkaBlHe9mSf8O09J3SCVa/5ezKM OLW/OONSV/Fr2VI1wxAYj3/Rb+U6rpzqIQ3Uh/5Rjmla6pTl7Z9/o1zKlVOX1SxVGSrlXhqt kwdbjdj/csSzoAbUF/duDuhyEl11/xStm/lBMzVuf3ZhV5SSgLAflLBo4l6mR5RolpPv5wad GpYS/hm7HsmEA0PBAPNb5DvZQ7vNaX23FlgylSXyv72UVsObHsu6pT4sfoxvJ5nJxvzGi69U s1uryvlAfS6E+D5ULrV35taTwSpcBAh0/RqRbV0mTc57vvAoXofBDcs3Z30IReFS34QSpjvl Hxbe7itHGuuhEVM1qmq2U72ezOQ7MzADbwCtn+yGeISQqeFn9QMAZVAkXsc9Wp0SW/WQKb76 FkSRalBZcc2vXM0VqhFVzTb6iNqYXqVKyuPKwhBunhTt6XnIfhpRgqveCPNIasSX05VQR6/a OBHZX3seTikp7A1z9iZIsdtJxB88dGkpeMj6qJ5RLzUsPUVPodEcz1B5aTEbYK6428H8MeLq NFPwmknOlDzQNC6RND8Ez7YEhzqvw7263MojcmmPcLelYbfOwU0EVcufkQEQAOfX3n0g0fZz Bgm/S2zF/kxQKCEKP8ID+Vz8sy2GpDvveBq4H2Y34XWsT1zLJdvqPI4af4ZSMxuerWjXbVWb T6d4odQIG0fKx4F8NccDqbgHeZRNajXeeJ3R7gAzvWvQNLz4piHrO/B4tf8svmRBL0ZB5P5A 2uhdwLU3NZuK22zpNn4is87BPWF8HhY0L5fafgDMOqnf4guJVJPYNPhUFzXUbPqOKOkL8ojk CXxkOFHAbjstSK5Ca3fKquY3rdX3DNo+EL7FvAiw1mUtS+5GeYE+RMnDCsVFm/C7kY8c2d0G NWkB9pJM5+mnIoFNxy7YBcldYATVeOHoY4LyaUWNnAvFYWp08dHWfZo9WCiJMuTfgtH9tc75 7QanMVdPt6fDK8UUXIBLQ2TWr/sQKE9xtFuEmoQGlE1l6bGaDnnMLcYu+Asp3kDT0w4zYGsx 5r6XQVRH4+5N6eHZiaeYtFOujp5n+pjBaQK7wUUjDilPQ5QMzIuCL4YjVoylWiBNknvQWBXS lQCWmavOT9sttGQXdPCC5ynI+1ymZC1ORZKANLnRAb0NH/UCzcsstw2TAkFnMEbo9Zu9w7Kv AxBQXWeXhJI9XQssfrf4Gusdqx8nPEpfOqCtbbwJMATbHyqLt7/oz/5deGuwxgb65pWIzufa N7eop7uh+6bezi+rugUI+w6DABEBAAHCwXwEGAEIACYCGwwWIQQb2cqtc1xMOkYN/MpN3hD3 AP+DWgUCaCwtJQUJG8aPFAAKCRBN3hD3AP+DWlDnD/4k2TW+HyOOOePVm23F5HOhNNd7nNv3 Vq2cLcW1DteHUdxMO0X+zqrKDHI5hgnE/E2QH9jyV8mB8l/ndElobciaJcbl1cM43vVzPIWn 01vW62oxUNtEvzLLxGLPTrnMxWdZgxr7ACCWKUnMGE2E8eca0cT2pnIJoQRz242xqe/nYxBB /BAK+dsxHIfcQzl88G83oaO7vb7s/cWMYRKOg+WIgp0MJ8DO2IU5JmUtyJB+V3YzzM4cMic3 bNn8nHjTWw/9+QQ5vg3TXHZ5XMu9mtfw2La3bHJ6AybL0DvEkdGxk6YHqJVEukciLMWDWqQQ RtbBhqcprgUxipNvdn9KwNpGciM+hNtM9kf9gt0fjv79l/FiSw6KbCPX9b636GzgNy0Ev2UV m00EtcpRXXMlEpbP4V947ufWVK2Mz7RFUfU4+ETDd1scMQDHzrXItryHLZWhopPI4Z+ps0rB CQHfSpl+wG4XbJJu1D8/Ww3FsO42TMFrNr2/cmqwuUZ0a0uxrpkNYrsGjkEu7a+9MheyTzcm vyU2knz5/stkTN2LKz5REqOe24oRnypjpAfaoxRYXs+F8wml519InWlwCra49IUSxD1hXPxO WBe5lqcozu9LpNDH/brVSzHCSb7vjNGvvSVESDuoiHK8gNlf0v+epy5WYd7CGAgODPvDShGN g3eXuA==
  • Cc: linux-kernel@xxxxxxxxxxxxxxx, linux-mm@xxxxxxxxx, xen-devel@xxxxxxxxxxxxxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx, nvdimm@xxxxxxxxxxxxxxx, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>, Juergen Gross <jgross@xxxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>, Dan Williams <dan.j.williams@xxxxxxxxx>, Matthew Wilcox <willy@xxxxxxxxxxxxx>, Jan Kara <jack@xxxxxxx>, Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>, Christian Brauner <brauner@xxxxxxxxxx>, "Liam R. Howlett" <Liam.Howlett@xxxxxxxxxx>, Vlastimil Babka <vbabka@xxxxxxx>, Mike Rapoport <rppt@xxxxxxxxxx>, Suren Baghdasaryan <surenb@xxxxxxxxxx>, Michal Hocko <mhocko@xxxxxxxx>, Zi Yan <ziy@xxxxxxxxxx>, Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>, Nico Pache <npache@xxxxxxxxxx>, Ryan Roberts <ryan.roberts@xxxxxxx>, Dev Jain <dev.jain@xxxxxxx>, Barry Song <baohua@xxxxxxxxxx>, Jann Horn <jannh@xxxxxxxxxx>, Pedro Falcato <pfalcato@xxxxxxx>, Hugh Dickins <hughd@xxxxxxxxxx>, Oscar Salvador <osalvador@xxxxxxx>, Lance Yang <lance.yang@xxxxxxxxx>
  • Delivery-date: Thu, 17 Jul 2025 20:31:46 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 17.07.25 20:29, Lorenzo Stoakes wrote:
On Thu, Jul 17, 2025 at 01:52:08PM +0200, David Hildenbrand wrote:
The huge zero folio is refcounted (+mapcounted -- is that a word?)
differently than "normal" folios, similarly (but different) to the ordinary
shared zeropage.

Yeah, I sort of wonder if we shouldn't just _not_ do any of that with zero
pages?

I wish we could get rid of the weird refcounting of the huge zero folio and get rid of the shrinker. But as long as the shrinker exists, I'm afraid that weird per-process refcounting must stay.


But for some reason the huge zero page wants to exist or not exist based on
usage for one. Which is stupid to me.

Yes, I will try at some point (once we have the static huge zero folio) to remove the shrinker part and make it always static. Well, at least for reasonable architectures.



For this reason, we special-case these pages in
vm_normal_page*/vm_normal_folio*, and only allow selected callers to
still use them (e.g., GUP can still take a reference on them).

vm_normal_page_pmd() already filters out the huge zero folio. However,
so far we are not marking it as special like we do with the ordinary
shared zeropage. Let's mark it as special, so we can further refactor
vm_normal_page_pmd() and vm_normal_page().

While at it, update the doc regarding the shared zero folios.

Hmm I wonder how this will interact with the static PMD series at [0]?

No, it shouldn't.


I wonder if more use of that might result in some weirdness with refcounting
etc.?

I don't think so.


Also, that series was (though I reviewed against it) moving stuff that
references the huge zero folio out of there, but also generally allows
access and mapping of this folio via largest_zero_folio() so not only via
insert_pmd().

So we're going to end up with mappings of this that are not marked special
that are potentially going to have refcount/mapcount manipulation that
contradict what you're doing here perhaps?

I don't think so. It's just like having the existing static (small) shared zeropage where the same rules about refcounting+mapcounting apply.


[0]: 
https://lore.kernel.org/all/20250707142319.319642-1-kernel@xxxxxxxxxxxxxxxx/


Reviewed-by: Oscar Salvador <osalvador@xxxxxxx>
Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>

I looked thorugh places that use vm_normal_page_pm() (other than decl of
function):

fs/proc/task_mmu.c - seems to handle NULL page correctly + still undertsands 
zero page
mm/pagewalk.c - correctly handles NULL page + huge zero page
mm/huge_memory.c - can_change_pmd_writable() correctly returns false.

And all seems to work wtih this change.

Overall, other than concerns above + nits below LGTM, we should treat all
the zero folios the same in this regard, so:

Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>

Thanks!


---
  mm/huge_memory.c |  5 ++++-
  mm/memory.c      | 14 +++++++++-----
  2 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index db08c37b87077..3f9a27812a590 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1320,6 +1320,7 @@ static void set_huge_zero_folio(pgtable_t pgtable, struct 
mm_struct *mm,
  {
        pmd_t entry;
        entry = folio_mk_pmd(zero_folio, vma->vm_page_prot);
+       entry = pmd_mkspecial(entry);
        pgtable_trans_huge_deposit(mm, pmd, pgtable);
        set_pmd_at(mm, haddr, pmd, entry);
        mm_inc_nr_ptes(mm);
@@ -1429,7 +1430,9 @@ static vm_fault_t insert_pmd(struct vm_area_struct *vma, 
unsigned long addr,
        if (fop.is_folio) {
                entry = folio_mk_pmd(fop.folio, vma->vm_page_prot);

-               if (!is_huge_zero_folio(fop.folio)) {
+               if (is_huge_zero_folio(fop.folio)) {
+                       entry = pmd_mkspecial(entry);
+               } else {
                        folio_get(fop.folio);
                        folio_add_file_rmap_pmd(fop.folio, &fop.folio->page, 
vma);
                        add_mm_counter(mm, mm_counter_file(fop.folio), 
HPAGE_PMD_NR);
diff --git a/mm/memory.c b/mm/memory.c
index 92fd18a5d8d1f..173eb6267e0ac 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -537,7 +537,13 @@ static void print_bad_pte(struct vm_area_struct *vma, 
unsigned long addr,
   *
   * "Special" mappings do not wish to be associated with a "struct page" 
(either
   * it doesn't exist, or it exists but they don't want to touch it). In this
- * case, NULL is returned here. "Normal" mappings do have a struct page.
+ * case, NULL is returned here. "Normal" mappings do have a struct page and
+ * are ordinarily refcounted.
+ *
+ * Page mappings of the shared zero folios are always considered "special", as
+ * they are not ordinarily refcounted. However, selected page table walkers
+ * (such as GUP) can still identify these mappings and work with the
+ * underlying "struct page".

I feel like we need more detail or something more explicit about what 'not
ordinary' refcounting constitutes. This is a bit vague.

Hm, I am not sure this is the correct place to document that. But let me see if I can come up with something reasonable

(like, the refcount and mapcount of these folios is never adjusted when mapping them into page tables)


   *
   * There are 2 broad cases. Firstly, an architecture may define a 
pte_special()
   * pte bit, in which case this function is trivial. Secondly, an architecture
@@ -567,9 +573,8 @@ static void print_bad_pte(struct vm_area_struct *vma, 
unsigned long addr,
   *
   * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
   * page" backing, however the difference is that _all_ pages with a struct
- * page (that is, those where pfn_valid is true) are refcounted and considered
- * normal pages by the VM. The only exception are zeropages, which are
- * *never* refcounted.
+ * page (that is, those where pfn_valid is true, except the shared zero
+ * folios) are refcounted and considered normal pages by the VM.
   *
   * The disadvantage is that pages are refcounted (which can be slower and
   * simply not an option for some PFNMAP users). The advantage is that we
@@ -649,7 +654,6 @@ struct page *vm_normal_page_pmd(struct vm_area_struct *vma, 
unsigned long addr,

You know I"m semi-ashamed to admit I didn't even know this function
exists. But yikes that we have a separate function like this just for PMDs.

It's a bit new-ish :)


--
Cheers,

David / dhildenb




 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.