1.. SPDX-License-Identifier: GPL-2.0+
2
3======
4XArray
5======
6
7:Author: Matthew Wilcox
8
9Overview
10========
11
12The XArray is an abstract data type which behaves like a very large array
13of pointers.  It meets many of the same needs as a hash or a conventional
14resizable array.  Unlike a hash, it allows you to sensibly go to the
15next or previous entry in a cache-efficient manner.  In contrast to a
16resizable array, there is no need to copy data or change MMU mappings in
17order to grow the array.  It is more memory-efficient, parallelisable
18and cache friendly than a doubly-linked list.  It takes advantage of
19RCU to perform lookups without locking.
20
21The XArray implementation is efficient when the indices used are densely
22clustered; hashing the object and using the hash as the index will not
23perform well.  The XArray is optimised for small indices, but still has
24good performance with large indices.  If your index can be larger than
25``ULONG_MAX`` then the XArray is not the data type for you.  The most
26important user of the XArray is the page cache.
27
28Each non-``NULL`` entry in the array has three bits associated with
29it called marks.  Each mark may be set or cleared independently of
30the others.  You can iterate over entries which are marked.
31
32Normal pointers may be stored in the XArray directly.  They must be 4-byte
33aligned, which is true for any pointer returned from :c:func:`kmalloc` and
34:c:func:`alloc_page`.  It isn't true for arbitrary user-space pointers,
35nor for function pointers.  You can store pointers to statically allocated
36objects, as long as those objects have an alignment of at least 4.
37
38You can also store integers between 0 and ``LONG_MAX`` in the XArray.
39You must first convert it into an entry using :c:func:`xa_mk_value`.
40When you retrieve an entry from the XArray, you can check whether it is
41a value entry by calling :c:func:`xa_is_value`, and convert it back to
42an integer by calling :c:func:`xa_to_value`.
43
44Some users want to store tagged pointers instead of using the marks
45described above.  They can call :c:func:`xa_tag_pointer` to create an
46entry with a tag, :c:func:`xa_untag_pointer` to turn a tagged entry
47back into an untagged pointer and :c:func:`xa_pointer_tag` to retrieve
48the tag of an entry.  Tagged pointers use the same bits that are used
49to distinguish value entries from normal pointers, so each user must
50decide whether they want to store value entries or tagged pointers in
51any particular XArray.
52
53The XArray does not support storing :c:func:`IS_ERR` pointers as some
54conflict with value entries or internal entries.
55
56An unusual feature of the XArray is the ability to create entries which
57occupy a range of indices.  Once stored to, looking up any index in
58the range will return the same entry as looking up any other index in
59the range.  Setting a mark on one index will set it on all of them.
60Storing to any index will store to all of them.  Multi-index entries can
61be explicitly split into smaller entries, or storing ``NULL`` into any
62entry will cause the XArray to forget about the range.
63
64Normal API
65==========
66
67Start by initialising an XArray, either with :c:func:`DEFINE_XARRAY`
68for statically allocated XArrays or :c:func:`xa_init` for dynamically
69allocated ones.  A freshly-initialised XArray contains a ``NULL``
70pointer at every index.
71
72You can then set entries using :c:func:`xa_store` and get entries
73using :c:func:`xa_load`.  xa_store will overwrite any entry with the
74new entry and return the previous entry stored at that index.  You can
75use :c:func:`xa_erase` instead of calling :c:func:`xa_store` with a
76``NULL`` entry.  There is no difference between an entry that has never
77been stored to, one that has been erased and one that has most recently
78had ``NULL`` stored to it.
79
80You can conditionally replace an entry at an index by using
81:c:func:`xa_cmpxchg`.  Like :c:func:`cmpxchg`, it will only succeed if
82the entry at that index has the 'old' value.  It also returns the entry
83which was at that index; if it returns the same entry which was passed as
84'old', then :c:func:`xa_cmpxchg` succeeded.
85
86If you want to only store a new entry to an index if the current entry
87at that index is ``NULL``, you can use :c:func:`xa_insert` which
88returns ``-EEXIST`` if the entry is not empty.
89
90You can enquire whether a mark is set on an entry by using
91:c:func:`xa_get_mark`.  If the entry is not ``NULL``, you can set a mark
92on it by using :c:func:`xa_set_mark` and remove the mark from an entry by
93calling :c:func:`xa_clear_mark`.  You can ask whether any entry in the
94XArray has a particular mark set by calling :c:func:`xa_marked`.
95
96You can copy entries out of the XArray into a plain array by calling
97:c:func:`xa_extract`.  Or you can iterate over the present entries in
98the XArray by calling :c:func:`xa_for_each`.  You may prefer to use
99:c:func:`xa_find` or :c:func:`xa_find_after` to move to the next present
100entry in the XArray.
101
102Calling :c:func:`xa_store_range` stores the same entry in a range
103of indices.  If you do this, some of the other operations will behave
104in a slightly odd way.  For example, marking the entry at one index
105may result in the entry being marked at some, but not all of the other
106indices.  Storing into one index may result in the entry retrieved by
107some, but not all of the other indices changing.
108
109Sometimes you need to ensure that a subsequent call to :c:func:`xa_store`
110will not need to allocate memory.  The :c:func:`xa_reserve` function
111will store a reserved entry at the indicated index.  Users of the normal
112API will see this entry as containing ``NULL``.  If you do not need to
113use the reserved entry, you can call :c:func:`xa_release` to remove the
114unused entry.  If another user has stored to the entry in the meantime,
115:c:func:`xa_release` will do nothing; if instead you want the entry to
116become ``NULL``, you should use :c:func:`xa_erase`.
117
118If all entries in the array are ``NULL``, the :c:func:`xa_empty` function
119will return ``true``.
120
121Finally, you can remove all entries from an XArray by calling
122:c:func:`xa_destroy`.  If the XArray entries are pointers, you may wish
123to free the entries first.  You can do this by iterating over all present
124entries in the XArray using the :c:func:`xa_for_each` iterator.
125
126Allocating XArrays
127------------------
128
129If you use :c:func:`DEFINE_XARRAY_ALLOC` to define the XArray, or
130initialise it by passing ``XA_FLAGS_ALLOC`` to :c:func:`xa_init_flags`,
131the XArray changes to track whether entries are in use or not.
132
133You can call :c:func:`xa_alloc` to store the entry at any unused index
134in the XArray.  If you need to modify the array from interrupt context,
135you can use :c:func:`xa_alloc_bh` or :c:func:`xa_alloc_irq` to disable
136interrupts while allocating the ID.
137
138Using :c:func:`xa_store`, :c:func:`xa_cmpxchg` or :c:func:`xa_insert`
139will mark the entry as being allocated.  Unlike a normal XArray, storing
140``NULL`` will mark the entry as being in use, like :c:func:`xa_reserve`.
141To free an entry, use :c:func:`xa_erase` (or :c:func:`xa_release` if
142you only want to free the entry if it's ``NULL``).
143
144You cannot use ``XA_MARK_0`` with an allocating XArray as this mark
145is used to track whether an entry is free or not.  The other marks are
146available for your use.
147
148Memory allocation
149-----------------
150
151The :c:func:`xa_store`, :c:func:`xa_cmpxchg`, :c:func:`xa_alloc`,
152:c:func:`xa_reserve` and :c:func:`xa_insert` functions take a gfp_t
153parameter in case the XArray needs to allocate memory to store this entry.
154If the entry is being deleted, no memory allocation needs to be performed,
155and the GFP flags specified will be ignored.
156
157It is possible for no memory to be allocatable, particularly if you pass
158a restrictive set of GFP flags.  In that case, the functions return a
159special value which can be turned into an errno using :c:func:`xa_err`.
160If you don't need to know exactly which error occurred, using
161:c:func:`xa_is_err` is slightly more efficient.
162
163Locking
164-------
165
166When using the Normal API, you do not have to worry about locking.
167The XArray uses RCU and an internal spinlock to synchronise access:
168
169No lock needed:
170 * :c:func:`xa_empty`
171 * :c:func:`xa_marked`
172
173Takes RCU read lock:
174 * :c:func:`xa_load`
175 * :c:func:`xa_for_each`
176 * :c:func:`xa_find`
177 * :c:func:`xa_find_after`
178 * :c:func:`xa_extract`
179 * :c:func:`xa_get_mark`
180
181Takes xa_lock internally:
182 * :c:func:`xa_store`
183 * :c:func:`xa_store_bh`
184 * :c:func:`xa_store_irq`
185 * :c:func:`xa_insert`
186 * :c:func:`xa_erase`
187 * :c:func:`xa_erase_bh`
188 * :c:func:`xa_erase_irq`
189 * :c:func:`xa_cmpxchg`
190 * :c:func:`xa_store_range`
191 * :c:func:`xa_alloc`
192 * :c:func:`xa_alloc_bh`
193 * :c:func:`xa_alloc_irq`
194 * :c:func:`xa_reserve`
195 * :c:func:`xa_reserve_bh`
196 * :c:func:`xa_reserve_irq`
197 * :c:func:`xa_destroy`
198 * :c:func:`xa_set_mark`
199 * :c:func:`xa_clear_mark`
200
201Assumes xa_lock held on entry:
202 * :c:func:`__xa_store`
203 * :c:func:`__xa_insert`
204 * :c:func:`__xa_erase`
205 * :c:func:`__xa_cmpxchg`
206 * :c:func:`__xa_alloc`
207 * :c:func:`__xa_reserve`
208 * :c:func:`__xa_set_mark`
209 * :c:func:`__xa_clear_mark`
210
211If you want to take advantage of the lock to protect the data structures
212that you are storing in the XArray, you can call :c:func:`xa_lock`
213before calling :c:func:`xa_load`, then take a reference count on the
214object you have found before calling :c:func:`xa_unlock`.  This will
215prevent stores from removing the object from the array between looking
216up the object and incrementing the refcount.  You can also use RCU to
217avoid dereferencing freed memory, but an explanation of that is beyond
218the scope of this document.
219
220The XArray does not disable interrupts or softirqs while modifying
221the array.  It is safe to read the XArray from interrupt or softirq
222context as the RCU lock provides enough protection.
223
224If, for example, you want to store entries in the XArray in process
225context and then erase them in softirq context, you can do that this way::
226
227    void foo_init(struct foo *foo)
228    {
229        xa_init_flags(&foo->array, XA_FLAGS_LOCK_BH);
230    }
231
232    int foo_store(struct foo *foo, unsigned long index, void *entry)
233    {
234        int err;
235
236        xa_lock_bh(&foo->array);
237        err = xa_err(__xa_store(&foo->array, index, entry, GFP_KERNEL));
238        if (!err)
239            foo->count++;
240        xa_unlock_bh(&foo->array);
241        return err;
242    }
243
244    /* foo_erase() is only called from softirq context */
245    void foo_erase(struct foo *foo, unsigned long index)
246    {
247        xa_lock(&foo->array);
248        __xa_erase(&foo->array, index);
249        foo->count--;
250        xa_unlock(&foo->array);
251    }
252
253If you are going to modify the XArray from interrupt or softirq context,
254you need to initialise the array using :c:func:`xa_init_flags`, passing
255``XA_FLAGS_LOCK_IRQ`` or ``XA_FLAGS_LOCK_BH``.
256
257The above example also shows a common pattern of wanting to extend the
258coverage of the xa_lock on the store side to protect some statistics
259associated with the array.
260
261Sharing the XArray with interrupt context is also possible, either
262using :c:func:`xa_lock_irqsave` in both the interrupt handler and process
263context, or :c:func:`xa_lock_irq` in process context and :c:func:`xa_lock`
264in the interrupt handler.  Some of the more common patterns have helper
265functions such as :c:func:`xa_store_bh`, :c:func:`xa_store_irq`,
266:c:func:`xa_erase_bh` and :c:func:`xa_erase_irq`.
267
268Sometimes you need to protect access to the XArray with a mutex because
269that lock sits above another mutex in the locking hierarchy.  That does
270not entitle you to use functions like :c:func:`__xa_erase` without taking
271the xa_lock; the xa_lock is used for lockdep validation and will be used
272for other purposes in the future.
273
274The :c:func:`__xa_set_mark` and :c:func:`__xa_clear_mark` functions are also
275available for situations where you look up an entry and want to atomically
276set or clear a mark.  It may be more efficient to use the advanced API
277in this case, as it will save you from walking the tree twice.
278
279Advanced API
280============
281
282The advanced API offers more flexibility and better performance at the
283cost of an interface which can be harder to use and has fewer safeguards.
284No locking is done for you by the advanced API, and you are required
285to use the xa_lock while modifying the array.  You can choose whether
286to use the xa_lock or the RCU lock while doing read-only operations on
287the array.  You can mix advanced and normal operations on the same array;
288indeed the normal API is implemented in terms of the advanced API.  The
289advanced API is only available to modules with a GPL-compatible license.
290
291The advanced API is based around the xa_state.  This is an opaque data
292structure which you declare on the stack using the :c:func:`XA_STATE`
293macro.  This macro initialises the xa_state ready to start walking
294around the XArray.  It is used as a cursor to maintain the position
295in the XArray and let you compose various operations together without
296having to restart from the top every time.
297
298The xa_state is also used to store errors.  You can call
299:c:func:`xas_error` to retrieve the error.  All operations check whether
300the xa_state is in an error state before proceeding, so there's no need
301for you to check for an error after each call; you can make multiple
302calls in succession and only check at a convenient point.  The only
303errors currently generated by the XArray code itself are ``ENOMEM`` and
304``EINVAL``, but it supports arbitrary errors in case you want to call
305:c:func:`xas_set_err` yourself.
306
307If the xa_state is holding an ``ENOMEM`` error, calling :c:func:`xas_nomem`
308will attempt to allocate more memory using the specified gfp flags and
309cache it in the xa_state for the next attempt.  The idea is that you take
310the xa_lock, attempt the operation and drop the lock.  The operation
311attempts to allocate memory while holding the lock, but it is more
312likely to fail.  Once you have dropped the lock, :c:func:`xas_nomem`
313can try harder to allocate more memory.  It will return ``true`` if it
314is worth retrying the operation (i.e. that there was a memory error *and*
315more memory was allocated).  If it has previously allocated memory, and
316that memory wasn't used, and there is no error (or some error that isn't
317``ENOMEM``), then it will free the memory previously allocated.
318
319Internal Entries
320----------------
321
322The XArray reserves some entries for its own purposes.  These are never
323exposed through the normal API, but when using the advanced API, it's
324possible to see them.  Usually the best way to handle them is to pass them
325to :c:func:`xas_retry`, and retry the operation if it returns ``true``.
326
327.. flat-table::
328   :widths: 1 1 6
329
330   * - Name
331     - Test
332     - Usage
333
334   * - Node
335     - :c:func:`xa_is_node`
336     - An XArray node.  May be visible when using a multi-index xa_state.
337
338   * - Sibling
339     - :c:func:`xa_is_sibling`
340     - A non-canonical entry for a multi-index entry.  The value indicates
341       which slot in this node has the canonical entry.
342
343   * - Retry
344     - :c:func:`xa_is_retry`
345     - This entry is currently being modified by a thread which has the
346       xa_lock.  The node containing this entry may be freed at the end
347       of this RCU period.  You should restart the lookup from the head
348       of the array.
349
350   * - Zero
351     - :c:func:`xa_is_zero`
352     - Zero entries appear as ``NULL`` through the Normal API, but occupy
353       an entry in the XArray which can be used to reserve the index for
354       future use.  This is used by allocating XArrays for allocated entries
355       which are ``NULL``.
356
357Other internal entries may be added in the future.  As far as possible, they
358will be handled by :c:func:`xas_retry`.
359
360Additional functionality
361------------------------
362
363The :c:func:`xas_create_range` function allocates all the necessary memory
364to store every entry in a range.  It will set ENOMEM in the xa_state if
365it cannot allocate memory.
366
367You can use :c:func:`xas_init_marks` to reset the marks on an entry
368to their default state.  This is usually all marks clear, unless the
369XArray is marked with ``XA_FLAGS_TRACK_FREE``, in which case mark 0 is set
370and all other marks are clear.  Replacing one entry with another using
371:c:func:`xas_store` will not reset the marks on that entry; if you want
372the marks reset, you should do that explicitly.
373
374The :c:func:`xas_load` will walk the xa_state as close to the entry
375as it can.  If you know the xa_state has already been walked to the
376entry and need to check that the entry hasn't changed, you can use
377:c:func:`xas_reload` to save a function call.
378
379If you need to move to a different index in the XArray, call
380:c:func:`xas_set`.  This resets the cursor to the top of the tree, which
381will generally make the next operation walk the cursor to the desired
382spot in the tree.  If you want to move to the next or previous index,
383call :c:func:`xas_next` or :c:func:`xas_prev`.  Setting the index does
384not walk the cursor around the array so does not require a lock to be
385held, while moving to the next or previous index does.
386
387You can search for the next present entry using :c:func:`xas_find`.  This
388is the equivalent of both :c:func:`xa_find` and :c:func:`xa_find_after`;
389if the cursor has been walked to an entry, then it will find the next
390entry after the one currently referenced.  If not, it will return the
391entry at the index of the xa_state.  Using :c:func:`xas_next_entry` to
392move to the next present entry instead of :c:func:`xas_find` will save
393a function call in the majority of cases at the expense of emitting more
394inline code.
395
396The :c:func:`xas_find_marked` function is similar.  If the xa_state has
397not been walked, it will return the entry at the index of the xa_state,
398if it is marked.  Otherwise, it will return the first marked entry after
399the entry referenced by the xa_state.  The :c:func:`xas_next_marked`
400function is the equivalent of :c:func:`xas_next_entry`.
401
402When iterating over a range of the XArray using :c:func:`xas_for_each`
403or :c:func:`xas_for_each_marked`, it may be necessary to temporarily stop
404the iteration.  The :c:func:`xas_pause` function exists for this purpose.
405After you have done the necessary work and wish to resume, the xa_state
406is in an appropriate state to continue the iteration after the entry
407you last processed.  If you have interrupts disabled while iterating,
408then it is good manners to pause the iteration and reenable interrupts
409every ``XA_CHECK_SCHED`` entries.
410
411The :c:func:`xas_get_mark`, :c:func:`xas_set_mark` and
412:c:func:`xas_clear_mark` functions require the xa_state cursor to have
413been moved to the appropriate location in the xarray; they will do
414nothing if you have called :c:func:`xas_pause` or :c:func:`xas_set`
415immediately before.
416
417You can call :c:func:`xas_set_update` to have a callback function
418called each time the XArray updates a node.  This is used by the page
419cache workingset code to maintain its list of nodes which contain only
420shadow entries.
421
422Multi-Index Entries
423-------------------
424
425The XArray has the ability to tie multiple indices together so that
426operations on one index affect all indices.  For example, storing into
427any index will change the value of the entry retrieved from any index.
428Setting or clearing a mark on any index will set or clear the mark
429on every index that is tied together.  The current implementation
430only allows tying ranges which are aligned powers of two together;
431eg indices 64-127 may be tied together, but 2-6 may not be.  This may
432save substantial quantities of memory; for example tying 512 entries
433together will save over 4kB.
434
435You can create a multi-index entry by using :c:func:`XA_STATE_ORDER`
436or :c:func:`xas_set_order` followed by a call to :c:func:`xas_store`.
437Calling :c:func:`xas_load` with a multi-index xa_state will walk the
438xa_state to the right location in the tree, but the return value is not
439meaningful, potentially being an internal entry or ``NULL`` even when there
440is an entry stored within the range.  Calling :c:func:`xas_find_conflict`
441will return the first entry within the range or ``NULL`` if there are no
442entries in the range.  The :c:func:`xas_for_each_conflict` iterator will
443iterate over every entry which overlaps the specified range.
444
445If :c:func:`xas_load` encounters a multi-index entry, the xa_index
446in the xa_state will not be changed.  When iterating over an XArray
447or calling :c:func:`xas_find`, if the initial index is in the middle
448of a multi-index entry, it will not be altered.  Subsequent calls
449or iterations will move the index to the first index in the range.
450Each entry will only be returned once, no matter how many indices it
451occupies.
452
453Using :c:func:`xas_next` or :c:func:`xas_prev` with a multi-index xa_state
454is not supported.  Using either of these functions on a multi-index entry
455will reveal sibling entries; these should be skipped over by the caller.
456
457Storing ``NULL`` into any index of a multi-index entry will set the entry
458at every index to ``NULL`` and dissolve the tie.  Splitting a multi-index
459entry into entries occupying smaller ranges is not yet supported.
460
461Functions and structures
462========================
463
464.. kernel-doc:: include/linux/xarray.h
465.. kernel-doc:: lib/xarray.c
466