4 The memory API models the memory and I/O buses and controllers of a QEMU
5 machine. It attempts to allow modelling of:
8 - memory-mapped I/O (MMIO)
9 - memory controllers that can dynamically reroute physical memory regions
10 to different destinations
12 The memory model provides support for
14 - tracking RAM changes by the guest
15 - setting up coalesced memory for kvm
16 - setting up ioeventfd regions for kvm
18 Memory is modelled as an acyclic graph of MemoryRegion objects. Sinks
19 (leaves) are RAM and MMIO regions, while other nodes represent
20 buses, memory controllers, and memory regions that have been rerouted.
22 In addition to MemoryRegion objects, the memory API provides AddressSpace
23 objects for every root and possibly for intermediate MemoryRegions too.
24 These represent memory as seen from the CPU or a device's viewpoint.
29 There are four types of memory regions (all represented by a single C type
32 - RAM: a RAM region is simply a range of host memory that can be made available
35 - MMIO: a range of guest memory that is implemented by host callbacks;
36 each read or write causes a callback to be called on the host.
38 - container: a container simply includes other memory regions, each at
39 a different offset. Containers are useful for grouping several regions
40 into one unit. For example, a PCI BAR may be composed of a RAM region
43 A container's subregions are usually non-overlapping. In some cases it is
44 useful to have overlapping regions; for example a memory controller that
45 can overlay a subregion of RAM with MMIO or ROM, or a PCI controller
46 that does not prevent card from claiming overlapping BARs.
48 - alias: a subsection of another region. Aliases allow a region to be
49 split apart into discontiguous regions. Examples of uses are memory banks
50 used when the guest address space is smaller than the amount of RAM
51 addressed, or a memory controller that splits main memory to expose a "PCI
52 hole". Aliases may point to any type of region, including other aliases,
53 but an alias may not point back to itself, directly or indirectly.
55 It is valid to add subregions to a region which is not a pure container
56 (that is, to an MMIO, RAM or ROM region). This means that the region
57 will act like a container, except that any addresses within the container's
58 region which are not claimed by any subregion are handled by the
59 container itself (ie by its MMIO callbacks or RAM backing). However
60 it is generally possible to achieve the same effect with a pure container
61 one of whose subregions is a low priority "background" region covering
62 the whole address range; this is often clearer and is preferred.
63 Subregions cannot be added to an alias region.
68 Regions are assigned names by the constructor. For most regions these are
69 only used for debugging purposes, but RAM regions also use the name to identify
70 live migration sections. This means that RAM region names need to have ABI
76 A region is created by one of the constructor functions (memory_region_init*())
77 and destroyed by the destructor (memory_region_destroy()). In between,
78 a region can be added to an address space by using memory_region_add_subregion()
79 and removed using memory_region_del_subregion(). Region attributes may be
80 changed at any point; they take effect once the region becomes exposed to the
83 Overlapping regions and priority
84 --------------------------------
85 Usually, regions may not overlap each other; a memory address decodes into
86 exactly one target. In some cases it is useful to allow regions to overlap,
87 and sometimes to control which of an overlapping regions is visible to the
88 guest. This is done with memory_region_add_subregion_overlap(), which
89 allows the region to overlap any other region in the same container, and
90 specifies a priority that allows the core to decide which of two regions at
91 the same address are visible (highest wins).
92 Priority values are signed, and the default value is zero. This means that
93 you can use memory_region_add_subregion_overlap() both to specify a region
94 that must sit 'above' any others (with a positive priority) and also a
95 background region that sits 'below' others (with a negative priority).
97 If the higher priority region in an overlap is a container or alias, then
98 the lower priority region will appear in any "holes" that the higher priority
99 region has left by not mapping subregions to that area of its address range.
100 (This applies recursively -- if the subregions are themselves containers or
101 aliases that leave holes then the lower priority region will appear in these
104 For example, suppose we have a container A of size 0x8000 with two subregions
105 B and C. B is a container mapped at 0x2000, size 0x4000, priority 1; C is
106 an MMIO region mapped at 0x0, size 0x6000, priority 2. B currently has two
107 of its own subregions: D of size 0x1000 at offset 0 and E of size 0x1000 at
108 offset 0x2000. As a diagram:
110 0 1000 2000 3000 4000 5000 6000 7000 8000
111 |------|------|------|------|------|------|------|-------|
113 C: [CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC]
118 The regions that will be seen within this address range then are:
119 [CCCCCCCCCCCC][DDDDD][CCCCC][EEEEE][CCCCC]
121 Since B has higher priority than C, its subregions appear in the flat map
122 even where they overlap with C. In ranges where B has not mapped anything
125 If B had provided its own MMIO operations (ie it was not a pure container)
126 then these would be used for any addresses in its range not handled by
127 D or E, and the result would be:
128 [CCCCCCCCCCCC][DDDDD][BBBBB][EEEEE][BBBBB]
130 Priority values are local to a container, because the priorities of two
131 regions are only compared when they are both children of the same container.
132 This means that the device in charge of the container (typically modelling
133 a bus or a memory controller) can use them to manage the interaction of
134 its child regions without any side effects on other parts of the system.
135 In the example above, the priorities of D and E are unimportant because
136 they do not overlap each other. It is the relative priority of B and C
137 that causes D and E to appear on top of C: D and E's priorities are never
138 compared against the priority of C.
142 The memory core uses the following rules to select a memory region when the
143 guest accesses an address:
145 - all direct subregions of the root region are matched against the address, in
146 descending priority order
147 - if the address lies outside the region offset/size, the subregion is
149 - if the subregion is a leaf (RAM or MMIO), the search terminates, returning
151 - if the subregion is a container, the same algorithm is used within the
152 subregion (after the address is adjusted by the subregion offset)
153 - if the subregion is an alias, the search is continued at the alias target
154 (after the address is adjusted by the subregion offset and alias offset)
155 - if a recursive search within a container or alias subregion does not
156 find a match (because of a "hole" in the container's coverage of its
157 address range), then if this is a container with its own MMIO or RAM
158 backing the search terminates, returning the container itself. Otherwise
159 we continue with the next subregion in priority order
160 - if none of the subregions match the address then the search terminates
166 system_memory: container@0-2^48-1
168 +---- lomem: alias@0-0xdfffffff ---> #ram (0-0xdfffffff)
170 +---- himem: alias@0x100000000-0x11fffffff ---> #ram (0xe0000000-0xffffffff)
172 +---- vga-window: alias@0xa0000-0xbfffff ---> #pci (0xa0000-0xbffff)
175 +---- pci-hole: alias@0xe0000000-0xffffffff ---> #pci (0xe0000000-0xffffffff)
179 +--- vga-area: container@0xa0000-0xbffff
181 | +--- alias@0x00000-0x7fff ---> #vram (0x010000-0x017fff)
183 | +--- alias@0x08000-0xffff ---> #vram (0x020000-0x027fff)
185 +---- vram: ram@0xe1000000-0xe1ffffff
187 +---- vga-mmio: mmio@0xe2000000-0xe200ffff
189 ram: ram@0x00000000-0xffffffff
191 This is a (simplified) PC memory map. The 4GB RAM block is mapped into the
192 system address space via two aliases: "lomem" is a 1:1 mapping of the first
193 3.5GB; "himem" maps the last 0.5GB at address 4GB. This leaves 0.5GB for the
194 so-called PCI hole, that allows a 32-bit PCI bus to exist in a system with
197 The memory controller diverts addresses in the range 640K-768K to the PCI
198 address space. This is modelled using the "vga-window" alias, mapped at a
199 higher priority so it obscures the RAM at the same addresses. The vga window
200 can be removed by programming the memory controller; this is modelled by
201 removing the alias and exposing the RAM underneath.
203 The pci address space is not a direct child of the system address space, since
204 we only want parts of it to be visible (we accomplish this using aliases).
205 It has two subregions: vga-area models the legacy vga window and is occupied
206 by two 32K memory banks pointing at two sections of the framebuffer.
207 In addition the vram is mapped as a BAR at address e1000000, and an additional
208 BAR containing MMIO registers is mapped after it.
210 Note that if the guest maps a BAR outside the PCI hole, it would not be
211 visible as the pci-hole alias clips it to a 0.5GB range.
216 Various region attributes (read-only, dirty logging, coalesced mmio, ioeventfd)
217 can be changed during the region lifecycle. They take effect once the region
218 is made visible (which can be immediately, later, or never).
223 MMIO regions are provided with ->read() and ->write() callbacks; in addition
224 various constraints can be supplied to control how these callbacks are called:
226 - .valid.min_access_size, .valid.max_access_size define the access sizes
227 (in bytes) which the device accepts; accesses outside this range will
228 have device and bus specific behaviour (ignored, or machine check)
229 - .valid.aligned specifies that the device only accepts naturally aligned
230 accesses. Unaligned accesses invoke device and bus specific behaviour.
231 - .impl.min_access_size, .impl.max_access_size define the access sizes
232 (in bytes) supported by the *implementation*; other access sizes will be
233 emulated using the ones available. For example a 4-byte write will be
234 emulated using four 1-byte writes, if .impl.max_access_size = 1.
235 - .impl.valid specifies that the *implementation* only supports unaligned
236 accesses; unaligned accesses will be emulated by two aligned accesses.
237 - .old_portio and .old_mmio can be used to ease porting from code using
238 cpu_register_io_memory() and register_ioport(). They should not be used