xref: /openbmc/qemu/docs/devel/memory.rst (revision ffe98631)
1==============
2The memory API
3==============
4
5The memory API models the memory and I/O buses and controllers of a QEMU
6machine.  It attempts to allow modelling of:
7
8- ordinary RAM
9- memory-mapped I/O (MMIO)
10- memory controllers that can dynamically reroute physical memory regions
11  to different destinations
12
13The memory model provides support for
14
15- tracking RAM changes by the guest
16- setting up coalesced memory for kvm
17- setting up ioeventfd regions for kvm
18
19Memory is modelled as an acyclic graph of MemoryRegion objects.  Sinks
20(leaves) are RAM and MMIO regions, while other nodes represent
21buses, memory controllers, and memory regions that have been rerouted.
22
23In addition to MemoryRegion objects, the memory API provides AddressSpace
24objects for every root and possibly for intermediate MemoryRegions too.
25These represent memory as seen from the CPU or a device's viewpoint.
26
27Types of regions
28----------------
29
30There are multiple types of memory regions (all represented by a single C type
31MemoryRegion):
32
33- RAM: a RAM region is simply a range of host memory that can be made available
34  to the guest.
35  You typically initialize these with memory_region_init_ram().  Some special
36  purposes require the variants memory_region_init_resizeable_ram(),
37  memory_region_init_ram_from_file(), or memory_region_init_ram_ptr().
38
39- MMIO: a range of guest memory that is implemented by host callbacks;
40  each read or write causes a callback to be called on the host.
41  You initialize these with memory_region_init_io(), passing it a
42  MemoryRegionOps structure describing the callbacks.
43
44- ROM: a ROM memory region works like RAM for reads (directly accessing
45  a region of host memory), and forbids writes. You initialize these with
46  memory_region_init_rom().
47
48- ROM device: a ROM device memory region works like RAM for reads
49  (directly accessing a region of host memory), but like MMIO for
50  writes (invoking a callback).  You initialize these with
51  memory_region_init_rom_device().
52
53- IOMMU region: an IOMMU region translates addresses of accesses made to it
54  and forwards them to some other target memory region.  As the name suggests,
55  these are only needed for modelling an IOMMU, not for simple devices.
56  You initialize these with memory_region_init_iommu().
57
58- container: a container simply includes other memory regions, each at
59  a different offset.  Containers are useful for grouping several regions
60  into one unit.  For example, a PCI BAR may be composed of a RAM region
61  and an MMIO region.
62
63  A container's subregions are usually non-overlapping.  In some cases it is
64  useful to have overlapping regions; for example a memory controller that
65  can overlay a subregion of RAM with MMIO or ROM, or a PCI controller
66  that does not prevent card from claiming overlapping BARs.
67
68  You initialize a pure container with memory_region_init().
69
70- alias: a subsection of another region. Aliases allow a region to be
71  split apart into discontiguous regions. Examples of uses are memory
72  banks used when the guest address space is smaller than the amount
73  of RAM addressed, or a memory controller that splits main memory to
74  expose a "PCI hole". You can also create aliases to avoid trying to
75  add the original region to multiple parents via
76  `memory_region_add_subregion`.
77
78  Aliases may point to any type of region, including other aliases,
79  but an alias may not point back to itself, directly or indirectly.
80  You initialize these with memory_region_init_alias().
81
82- reservation region: a reservation region is primarily for debugging.
83  It claims I/O space that is not supposed to be handled by QEMU itself.
84  The typical use is to track parts of the address space which will be
85  handled by the host kernel when KVM is enabled.  You initialize these
86  by passing a NULL callback parameter to memory_region_init_io().
87
88It is valid to add subregions to a region which is not a pure container
89(that is, to an MMIO, RAM or ROM region). This means that the region
90will act like a container, except that any addresses within the container's
91region which are not claimed by any subregion are handled by the
92container itself (ie by its MMIO callbacks or RAM backing). However
93it is generally possible to achieve the same effect with a pure container
94one of whose subregions is a low priority "background" region covering
95the whole address range; this is often clearer and is preferred.
96Subregions cannot be added to an alias region.
97
98Migration
99---------
100
101Where the memory region is backed by host memory (RAM, ROM and
102ROM device memory region types), this host memory needs to be
103copied to the destination on migration. These APIs which allocate
104the host memory for you will also register the memory so it is
105migrated:
106
107- memory_region_init_ram()
108- memory_region_init_rom()
109- memory_region_init_rom_device()
110
111For most devices and boards this is the correct thing. If you
112have a special case where you need to manage the migration of
113the backing memory yourself, you can call the functions:
114
115- memory_region_init_ram_nomigrate()
116- memory_region_init_rom_nomigrate()
117- memory_region_init_rom_device_nomigrate()
118
119which only initialize the MemoryRegion and leave handling
120migration to the caller.
121
122The functions:
123
124- memory_region_init_resizeable_ram()
125- memory_region_init_ram_from_file()
126- memory_region_init_ram_from_fd()
127- memory_region_init_ram_ptr()
128- memory_region_init_ram_device_ptr()
129
130are for special cases only, and so they do not automatically
131register the backing memory for migration; the caller must
132manage migration if necessary.
133
134Region names
135------------
136
137Regions are assigned names by the constructor.  For most regions these are
138only used for debugging purposes, but RAM regions also use the name to identify
139live migration sections.  This means that RAM region names need to have ABI
140stability.
141
142Region lifecycle
143----------------
144
145A region is created by one of the memory_region_init*() functions and
146attached to an object, which acts as its owner or parent.  QEMU ensures
147that the owner object remains alive as long as the region is visible to
148the guest, or as long as the region is in use by a virtual CPU or another
149device.  For example, the owner object will not die between an
150address_space_map operation and the corresponding address_space_unmap.
151
152After creation, a region can be added to an address space or a
153container with memory_region_add_subregion(), and removed using
154memory_region_del_subregion().
155
156Various region attributes (read-only, dirty logging, coalesced mmio,
157ioeventfd) can be changed during the region lifecycle.  They take effect
158as soon as the region is made visible.  This can be immediately, later,
159or never.
160
161Destruction of a memory region happens automatically when the owner
162object dies.
163
164If however the memory region is part of a dynamically allocated data
165structure, you should call object_unparent() to destroy the memory region
166before the data structure is freed.  For an example see VFIOMSIXInfo
167and VFIOQuirk in hw/vfio/pci.c.
168
169You must not destroy a memory region as long as it may be in use by a
170device or CPU.  In order to do this, as a general rule do not create or
171destroy memory regions dynamically during a device's lifetime, and only
172call object_unparent() in the memory region owner's instance_finalize
173callback.  The dynamically allocated data structure that contains the
174memory region then should obviously be freed in the instance_finalize
175callback as well.
176
177If you break this rule, the following situation can happen:
178
179- the memory region's owner had a reference taken via memory_region_ref
180  (for example by address_space_map)
181
182- the region is unparented, and has no owner anymore
183
184- when address_space_unmap is called, the reference to the memory region's
185  owner is leaked.
186
187
188There is an exception to the above rule: it is okay to call
189object_unparent at any time for an alias or a container region.  It is
190therefore also okay to create or destroy alias and container regions
191dynamically during a device's lifetime.
192
193This exceptional usage is valid because aliases and containers only help
194QEMU building the guest's memory map; they are never accessed directly.
195memory_region_ref and memory_region_unref are never called on aliases
196or containers, and the above situation then cannot happen.  Exploiting
197this exception is rarely necessary, and therefore it is discouraged,
198but nevertheless it is used in a few places.
199
200For regions that "have no owner" (NULL is passed at creation time), the
201machine object is actually used as the owner.  Since instance_finalize is
202never called for the machine object, you must never call object_unparent
203on regions that have no owner, unless they are aliases or containers.
204
205
206Overlapping regions and priority
207--------------------------------
208Usually, regions may not overlap each other; a memory address decodes into
209exactly one target.  In some cases it is useful to allow regions to overlap,
210and sometimes to control which of an overlapping regions is visible to the
211guest.  This is done with memory_region_add_subregion_overlap(), which
212allows the region to overlap any other region in the same container, and
213specifies a priority that allows the core to decide which of two regions at
214the same address are visible (highest wins).
215Priority values are signed, and the default value is zero. This means that
216you can use memory_region_add_subregion_overlap() both to specify a region
217that must sit 'above' any others (with a positive priority) and also a
218background region that sits 'below' others (with a negative priority).
219
220If the higher priority region in an overlap is a container or alias, then
221the lower priority region will appear in any "holes" that the higher priority
222region has left by not mapping subregions to that area of its address range.
223(This applies recursively -- if the subregions are themselves containers or
224aliases that leave holes then the lower priority region will appear in these
225holes too.)
226
227For example, suppose we have a container A of size 0x8000 with two subregions
228B and C. B is a container mapped at 0x2000, size 0x4000, priority 2; C is
229an MMIO region mapped at 0x0, size 0x6000, priority 1. B currently has two
230of its own subregions: D of size 0x1000 at offset 0 and E of size 0x1000 at
231offset 0x2000. As a diagram::
232
233        0      1000   2000   3000   4000   5000   6000   7000   8000
234        |------|------|------|------|------|------|------|------|
235  A:    [                                                      ]
236  C:    [CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC]
237  B:                  [                          ]
238  D:                  [DDDDD]
239  E:                                [EEEEE]
240
241The regions that will be seen within this address range then are::
242
243  [CCCCCCCCCCCC][DDDDD][CCCCC][EEEEE][CCCCC]
244
245Since B has higher priority than C, its subregions appear in the flat map
246even where they overlap with C. In ranges where B has not mapped anything
247C's region appears.
248
249If B had provided its own MMIO operations (ie it was not a pure container)
250then these would be used for any addresses in its range not handled by
251D or E, and the result would be::
252
253  [CCCCCCCCCCCC][DDDDD][BBBBB][EEEEE][BBBBB]
254
255Priority values are local to a container, because the priorities of two
256regions are only compared when they are both children of the same container.
257This means that the device in charge of the container (typically modelling
258a bus or a memory controller) can use them to manage the interaction of
259its child regions without any side effects on other parts of the system.
260In the example above, the priorities of D and E are unimportant because
261they do not overlap each other. It is the relative priority of B and C
262that causes D and E to appear on top of C: D and E's priorities are never
263compared against the priority of C.
264
265Visibility
266----------
267The memory core uses the following rules to select a memory region when the
268guest accesses an address:
269
270- all direct subregions of the root region are matched against the address, in
271  descending priority order
272
273  - if the address lies outside the region offset/size, the subregion is
274    discarded
275  - if the subregion is a leaf (RAM or MMIO), the search terminates, returning
276    this leaf region
277  - if the subregion is a container, the same algorithm is used within the
278    subregion (after the address is adjusted by the subregion offset)
279  - if the subregion is an alias, the search is continued at the alias target
280    (after the address is adjusted by the subregion offset and alias offset)
281  - if a recursive search within a container or alias subregion does not
282    find a match (because of a "hole" in the container's coverage of its
283    address range), then if this is a container with its own MMIO or RAM
284    backing the search terminates, returning the container itself. Otherwise
285    we continue with the next subregion in priority order
286
287- if none of the subregions match the address then the search terminates
288  with no match found
289
290Example memory map
291------------------
292
293::
294
295  system_memory: container@0-2^48-1
296   |
297   +---- lomem: alias@0-0xdfffffff ---> #ram (0-0xdfffffff)
298   |
299   +---- himem: alias@0x100000000-0x11fffffff ---> #ram (0xe0000000-0xffffffff)
300   |
301   +---- vga-window: alias@0xa0000-0xbffff ---> #pci (0xa0000-0xbffff)
302   |      (prio 1)
303   |
304   +---- pci-hole: alias@0xe0000000-0xffffffff ---> #pci (0xe0000000-0xffffffff)
305
306  pci (0-2^32-1)
307   |
308   +--- vga-area: container@0xa0000-0xbffff
309   |      |
310   |      +--- alias@0x00000-0x7fff  ---> #vram (0x010000-0x017fff)
311   |      |
312   |      +--- alias@0x08000-0xffff  ---> #vram (0x020000-0x027fff)
313   |
314   +---- vram: ram@0xe1000000-0xe1ffffff
315   |
316   +---- vga-mmio: mmio@0xe2000000-0xe200ffff
317
318  ram: ram@0x00000000-0xffffffff
319
320This is a (simplified) PC memory map. The 4GB RAM block is mapped into the
321system address space via two aliases: "lomem" is a 1:1 mapping of the first
3223.5GB; "himem" maps the last 0.5GB at address 4GB.  This leaves 0.5GB for the
323so-called PCI hole, that allows a 32-bit PCI bus to exist in a system with
3244GB of memory.
325
326The memory controller diverts addresses in the range 640K-768K to the PCI
327address space.  This is modelled using the "vga-window" alias, mapped at a
328higher priority so it obscures the RAM at the same addresses.  The vga window
329can be removed by programming the memory controller; this is modelled by
330removing the alias and exposing the RAM underneath.
331
332The pci address space is not a direct child of the system address space, since
333we only want parts of it to be visible (we accomplish this using aliases).
334It has two subregions: vga-area models the legacy vga window and is occupied
335by two 32K memory banks pointing at two sections of the framebuffer.
336In addition the vram is mapped as a BAR at address e1000000, and an additional
337BAR containing MMIO registers is mapped after it.
338
339Note that if the guest maps a BAR outside the PCI hole, it would not be
340visible as the pci-hole alias clips it to a 0.5GB range.
341
342MMIO Operations
343---------------
344
345MMIO regions are provided with ->read() and ->write() callbacks,
346which are sufficient for most devices. Some devices change behaviour
347based on the attributes used for the memory transaction, or need
348to be able to respond that the access should provoke a bus error
349rather than completing successfully; those devices can use the
350->read_with_attrs() and ->write_with_attrs() callbacks instead.
351
352In addition various constraints can be supplied to control how these
353callbacks are called:
354
355- .valid.min_access_size, .valid.max_access_size define the access sizes
356  (in bytes) which the device accepts; accesses outside this range will
357  have device and bus specific behaviour (ignored, or machine check)
358- .valid.unaligned specifies that the *device being modelled* supports
359  unaligned accesses; if false, unaligned accesses will invoke the
360  appropriate bus or CPU specific behaviour.
361- .impl.min_access_size, .impl.max_access_size define the access sizes
362  (in bytes) supported by the *implementation*; other access sizes will be
363  emulated using the ones available.  For example a 4-byte write will be
364  emulated using four 1-byte writes, if .impl.max_access_size = 1.
365- .impl.unaligned specifies that the *implementation* supports unaligned
366  accesses; if false, unaligned accesses will be emulated by two aligned
367  accesses.
368
369API Reference
370-------------
371
372.. kernel-doc:: include/exec/memory.h
373