1 .. SPDX-License-Identifier: GPL-2.0
3 ====================================================
4 pin_user_pages() and related calls
5 ====================================================
12 This document describes the following functions::
16 pin_user_pages_remote()
18 Basic description of FOLL_PIN
19 =============================
21 FOLL_PIN and FOLL_LONGTERM are flags that can be passed to the get_user_pages*()
22 ("gup") family of functions. FOLL_PIN has significant interactions and
23 interdependencies with FOLL_LONGTERM, so both are covered here.
25 FOLL_PIN is internal to gup, meaning that it should not appear at the gup call
26 sites. This allows the associated wrapper functions (pin_user_pages*() and
27 others) to set the correct combination of these flags, and to check for problems
30 FOLL_LONGTERM, on the other hand, *is* allowed to be set at the gup call sites.
31 This is in order to avoid creating a large number of wrapper functions to cover
32 all combinations of get*(), pin*(), FOLL_LONGTERM, and more. Also, the
33 pin_user_pages*() APIs are clearly distinct from the get_user_pages*() APIs, so
34 that's a natural dividing line, and a good point to make separate wrapper calls.
35 In other words, use pin_user_pages*() for DMA-pinned pages, and
36 get_user_pages*() for other cases. There are five cases described later on in
37 this document, to further clarify that concept.
39 FOLL_PIN and FOLL_GET are mutually exclusive for a given gup call. However,
40 multiple threads and call sites are free to pin the same struct pages, via both
41 FOLL_PIN and FOLL_GET. It's just the call site that needs to choose one or the
42 other, not the struct page(s).
44 The FOLL_PIN implementation is nearly the same as FOLL_GET, except that FOLL_PIN
45 uses a different reference counting technique.
47 FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying that is,
48 FOLL_LONGTERM is a specific case, more restrictive case of FOLL_PIN.
50 Which flags are set by each wrapper
51 ===================================
53 For these pin_user_pages*() functions, FOLL_PIN is OR'd in with whatever gup
54 flags the caller provides. The caller is required to pass in a non-null struct
55 pages* array, and the function then pins pages by incrementing each by a special
56 value: GUP_PIN_COUNTING_BIAS.
58 For large folios, the GUP_PIN_COUNTING_BIAS scheme is not used. Instead,
59 the extra space available in the struct folio is used to store the
62 This approach for large folios avoids the counting upper limit problems
63 that are discussed below. Those limitations would have been aggravated
64 severely by huge pages, because each tail page adds a refcount to the
65 head page. And in fact, testing revealed that, without a separate pincount
66 field, refcount overflows were seen in some huge page stress tests.
68 This also means that huge pages and large folios do not suffer
69 from the false positives problem that is mentioned below.::
73 pin_user_pages FOLL_PIN is always set internally by this function.
74 pin_user_pages_fast FOLL_PIN is always set internally by this function.
75 pin_user_pages_remote FOLL_PIN is always set internally by this function.
77 For these get_user_pages*() functions, FOLL_GET might not even be specified.
78 Behavior is a little more complex than above. If FOLL_GET was *not* specified,
79 but the caller passed in a non-null struct pages* array, then the function
80 sets FOLL_GET for you, and proceeds to pin pages by incrementing the refcount
85 get_user_pages FOLL_GET is sometimes set internally by this function.
86 get_user_pages_fast FOLL_GET is sometimes set internally by this function.
87 get_user_pages_remote FOLL_GET is sometimes set internally by this function.
89 Tracking dma-pinned pages
90 =========================
92 Some of the key design constraints, and solutions, for tracking dma-pinned
95 * An actual reference count, per struct page, is required. This is because
96 multiple processes may pin and unpin a page.
98 * False positives (reporting that a page is dma-pinned, when in fact it is not)
99 are acceptable, but false negatives are not.
101 * struct page may not be increased in size for this, and all fields are already
104 * Given the above, we can overload the page->_refcount field by using, sort of,
105 the upper bits in that field for a dma-pinned count. "Sort of", means that,
106 rather than dividing page->_refcount into bit fields, we simple add a medium-
107 large value (GUP_PIN_COUNTING_BIAS, initially chosen to be 1024: 10 bits) to
108 page->_refcount. This provides fuzzy behavior: if a page has get_page() called
109 on it 1024 times, then it will appear to have a single dma-pinned count.
110 And again, that's acceptable.
112 This also leads to limitations: there are only 31-10==21 bits available for a
113 counter that increments 10 bits at a time.
115 * Callers must specifically request "dma-pinned tracking of pages". In other
116 words, just calling get_user_pages() will not suffice; a new set of functions,
117 pin_user_page() and related, must be used.
119 FOLL_PIN, FOLL_GET, FOLL_LONGTERM: when to use which flags
120 ==========================================================
122 Thanks to Jan Kara, Vlastimil Babka and several other -mm people, for describing
125 CASE 1: Direct IO (DIO)
126 -----------------------
127 There are GUP references to pages that are serving
128 as DIO buffers. These buffers are needed for a relatively short time (so they
129 are not "long term"). No special synchronization with page_mkclean() or
130 munmap() is provided. Therefore, flags to set at the call site are: ::
134 ...but rather than setting FOLL_PIN directly, call sites should use one of
135 the pin_user_pages*() routines that set FOLL_PIN.
139 There are GUP references to pages that are serving as DMA
140 buffers. These buffers are needed for a long time ("long term"). No special
141 synchronization with page_mkclean() or munmap() is provided. Therefore, flags
142 to set at the call site are: ::
144 FOLL_PIN | FOLL_LONGTERM
146 NOTE: Some pages, such as DAX pages, cannot be pinned with longterm pins. That's
147 because DAX pages do not have a separate page cache, and so "pinning" implies
148 locking down file system blocks, which is not (yet) supported in that way.
150 CASE 3: MMU notifier registration, with or without page faulting hardware
151 -------------------------------------------------------------------------
152 Device drivers can pin pages via get_user_pages*(), and register for mmu
153 notifier callbacks for the memory range. Then, upon receiving a notifier
154 "invalidate range" callback , stop the device from using the range, and unpin
155 the pages. There may be other possible schemes, such as for example explicitly
156 synchronizing against pending IO, that accomplish approximately the same thing.
158 Or, if the hardware supports replayable page faults, then the device driver can
159 avoid pinning entirely (this is ideal), as follows: register for mmu notifier
160 callbacks as above, but instead of stopping the device and unpinning in the
161 callback, simply remove the range from the device's page tables.
163 Either way, as long as the driver unpins the pages upon mmu notifier callback,
164 then there is proper synchronization with both filesystem and mm
165 (page_mkclean(), munmap(), etc). Therefore, neither flag needs to be set.
167 CASE 4: Pinning for struct page manipulation only
168 -------------------------------------------------
169 If only struct page data (as opposed to the actual memory contents that a page
170 is tracking) is affected, then normal GUP calls are sufficient, and neither flag
173 CASE 5: Pinning in order to write to the data within the page
174 -------------------------------------------------------------
175 Even though neither DMA nor Direct IO is involved, just a simple case of "pin,
176 write to a page's data, unpin" can cause a problem. Case 5 may be considered a
177 superset of Case 1, plus Case 2, plus anything that invokes that pattern. In
178 other words, if the code is neither Case 1 nor Case 2, it may still require
179 FOLL_PIN, for patterns like this:
181 Correct (uses FOLL_PIN calls):
183 write to the data within the pages
186 INCORRECT (uses FOLL_GET calls):
188 write to the data within the pages
191 page_maybe_dma_pinned(): the whole point of pinning
192 ===================================================
194 The whole point of marking pages as "DMA-pinned" or "gup-pinned" is to be able
195 to query, "is this page DMA-pinned?" That allows code such as page_mkclean()
196 (and file system writeback code in general) to make informed decisions about
197 what to do when a page cannot be unmapped due to such pins.
199 What to do in those cases is the subject of a years-long series of discussions
200 and debates (see the References at the end of this document). It's a TODO item
201 here: fill in the details once that's worked out. Meanwhile, it's safe to say
202 that having this available: ::
204 static inline bool page_maybe_dma_pinned(struct page *page)
206 ...is a prerequisite to solving the long-running gup+DMA problem.
208 Another way of thinking about FOLL_GET, FOLL_PIN, and FOLL_LONGTERM
209 ===================================================================
211 Another way of thinking about these flags is as a progression of restrictions:
212 FOLL_GET is for struct page manipulation, without affecting the data that the
213 struct page refers to. FOLL_PIN is a *replacement* for FOLL_GET, and is for
214 short term pins on pages whose data *will* get accessed. As such, FOLL_PIN is
215 a "more severe" form of pinning. And finally, FOLL_LONGTERM is an even more
216 restrictive case that has FOLL_PIN as a prerequisite: this is for pages that
217 will be pinned longterm, and whose data will be accessed.
223 tools/testing/selftests/mm/gup_test.c
225 has the following new calls to exercise the new pin*() wrapper functions:
227 * PIN_FAST_BENCHMARK (./gup_test -a)
228 * PIN_BASIC_TEST (./gup_test -b)
230 You can monitor how many total dma-pinned pages have been acquired and released
231 since the system was booted, via two new /proc/vmstat entries: ::
233 /proc/vmstat/nr_foll_pin_acquired
234 /proc/vmstat/nr_foll_pin_released
236 Under normal conditions, these two values will be equal unless there are any
237 long-term [R]DMA pins in place, or during pin/unpin transitions.
239 * nr_foll_pin_acquired: This is the number of logical pins that have been
240 acquired since the system was powered on. For huge pages, the head page is
241 pinned once for each page (head page and each tail page) within the huge page.
242 This follows the same sort of behavior that get_user_pages() uses for huge
243 pages: the head page is refcounted once for each tail or head page in the huge
244 page, when get_user_pages() is applied to a huge page.
246 * nr_foll_pin_released: The number of logical pins that have been released since
247 the system was powered on. Note that pages are released (unpinned) on a
248 PAGE_SIZE granularity, even if the original pin was applied to a huge page.
249 Becaused of the pin count behavior described above in "nr_foll_pin_acquired",
250 the accounting balances out, so that after doing this::
252 pin_user_pages(huge_page);
253 for (each page in huge_page)
254 unpin_user_page(page);
256 ...the following is expected::
258 nr_foll_pin_released == nr_foll_pin_acquired
260 (...unless it was already out of balance due to a long-term RDMA pin being in
266 dump_page() has been enhanced slightly to handle these new counting
267 fields, and to better report on large folios in general. Specifically,
268 for large folios, the exact pincount is reported.
273 * `Some slow progress on get_user_pages() (Apr 2, 2019) <https://lwn.net/Articles/784574/>`_
274 * `DMA and get_user_pages() (LPC: Dec 12, 2018) <https://lwn.net/Articles/774411/>`_
275 * `The trouble with get_user_pages() (Apr 30, 2018) <https://lwn.net/Articles/753027/>`_
276 * `LWN kernel index: get_user_pages() <https://lwn.net/Kernel/Index/#Memory_management-get_user_pages>`_
278 John Hubbard, October, 2019