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 banks 72 used when the guest address space is smaller than the amount of RAM 73 addressed, or a memory controller that splits main memory to expose a "PCI 74 hole". Aliases may point to any type of region, including other aliases, 75 but an alias may not point back to itself, directly or indirectly. 76 You initialize these with memory_region_init_alias(). 77 78- reservation region: a reservation region is primarily for debugging. 79 It claims I/O space that is not supposed to be handled by QEMU itself. 80 The typical use is to track parts of the address space which will be 81 handled by the host kernel when KVM is enabled. You initialize these 82 by passing a NULL callback parameter to memory_region_init_io(). 83 84It is valid to add subregions to a region which is not a pure container 85(that is, to an MMIO, RAM or ROM region). This means that the region 86will act like a container, except that any addresses within the container's 87region which are not claimed by any subregion are handled by the 88container itself (ie by its MMIO callbacks or RAM backing). However 89it is generally possible to achieve the same effect with a pure container 90one of whose subregions is a low priority "background" region covering 91the whole address range; this is often clearer and is preferred. 92Subregions cannot be added to an alias region. 93 94Migration 95--------- 96 97Where the memory region is backed by host memory (RAM, ROM and 98ROM device memory region types), this host memory needs to be 99copied to the destination on migration. These APIs which allocate 100the host memory for you will also register the memory so it is 101migrated: 102 103- memory_region_init_ram() 104- memory_region_init_rom() 105- memory_region_init_rom_device() 106 107For most devices and boards this is the correct thing. If you 108have a special case where you need to manage the migration of 109the backing memory yourself, you can call the functions: 110 111- memory_region_init_ram_nomigrate() 112- memory_region_init_rom_nomigrate() 113- memory_region_init_rom_device_nomigrate() 114 115which only initialize the MemoryRegion and leave handling 116migration to the caller. 117 118The functions: 119 120- memory_region_init_resizeable_ram() 121- memory_region_init_ram_from_file() 122- memory_region_init_ram_from_fd() 123- memory_region_init_ram_ptr() 124- memory_region_init_ram_device_ptr() 125 126are for special cases only, and so they do not automatically 127register the backing memory for migration; the caller must 128manage migration if necessary. 129 130Region names 131------------ 132 133Regions are assigned names by the constructor. For most regions these are 134only used for debugging purposes, but RAM regions also use the name to identify 135live migration sections. This means that RAM region names need to have ABI 136stability. 137 138Region lifecycle 139---------------- 140 141A region is created by one of the memory_region_init*() functions and 142attached to an object, which acts as its owner or parent. QEMU ensures 143that the owner object remains alive as long as the region is visible to 144the guest, or as long as the region is in use by a virtual CPU or another 145device. For example, the owner object will not die between an 146address_space_map operation and the corresponding address_space_unmap. 147 148After creation, a region can be added to an address space or a 149container with memory_region_add_subregion(), and removed using 150memory_region_del_subregion(). 151 152Various region attributes (read-only, dirty logging, coalesced mmio, 153ioeventfd) can be changed during the region lifecycle. They take effect 154as soon as the region is made visible. This can be immediately, later, 155or never. 156 157Destruction of a memory region happens automatically when the owner 158object dies. 159 160If however the memory region is part of a dynamically allocated data 161structure, you should call object_unparent() to destroy the memory region 162before the data structure is freed. For an example see VFIOMSIXInfo 163and VFIOQuirk in hw/vfio/pci.c. 164 165You must not destroy a memory region as long as it may be in use by a 166device or CPU. In order to do this, as a general rule do not create or 167destroy memory regions dynamically during a device's lifetime, and only 168call object_unparent() in the memory region owner's instance_finalize 169callback. The dynamically allocated data structure that contains the 170memory region then should obviously be freed in the instance_finalize 171callback as well. 172 173If you break this rule, the following situation can happen: 174 175- the memory region's owner had a reference taken via memory_region_ref 176 (for example by address_space_map) 177 178- the region is unparented, and has no owner anymore 179 180- when address_space_unmap is called, the reference to the memory region's 181 owner is leaked. 182 183 184There is an exception to the above rule: it is okay to call 185object_unparent at any time for an alias or a container region. It is 186therefore also okay to create or destroy alias and container regions 187dynamically during a device's lifetime. 188 189This exceptional usage is valid because aliases and containers only help 190QEMU building the guest's memory map; they are never accessed directly. 191memory_region_ref and memory_region_unref are never called on aliases 192or containers, and the above situation then cannot happen. Exploiting 193this exception is rarely necessary, and therefore it is discouraged, 194but nevertheless it is used in a few places. 195 196For regions that "have no owner" (NULL is passed at creation time), the 197machine object is actually used as the owner. Since instance_finalize is 198never called for the machine object, you must never call object_unparent 199on regions that have no owner, unless they are aliases or containers. 200 201 202Overlapping regions and priority 203-------------------------------- 204Usually, regions may not overlap each other; a memory address decodes into 205exactly one target. In some cases it is useful to allow regions to overlap, 206and sometimes to control which of an overlapping regions is visible to the 207guest. This is done with memory_region_add_subregion_overlap(), which 208allows the region to overlap any other region in the same container, and 209specifies a priority that allows the core to decide which of two regions at 210the same address are visible (highest wins). 211Priority values are signed, and the default value is zero. This means that 212you can use memory_region_add_subregion_overlap() both to specify a region 213that must sit 'above' any others (with a positive priority) and also a 214background region that sits 'below' others (with a negative priority). 215 216If the higher priority region in an overlap is a container or alias, then 217the lower priority region will appear in any "holes" that the higher priority 218region has left by not mapping subregions to that area of its address range. 219(This applies recursively -- if the subregions are themselves containers or 220aliases that leave holes then the lower priority region will appear in these 221holes too.) 222 223For example, suppose we have a container A of size 0x8000 with two subregions 224B and C. B is a container mapped at 0x2000, size 0x4000, priority 2; C is 225an MMIO region mapped at 0x0, size 0x6000, priority 1. B currently has two 226of its own subregions: D of size 0x1000 at offset 0 and E of size 0x1000 at 227offset 0x2000. As a diagram:: 228 229 0 1000 2000 3000 4000 5000 6000 7000 8000 230 |------|------|------|------|------|------|------|------| 231 A: [ ] 232 C: [CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC] 233 B: [ ] 234 D: [DDDDD] 235 E: [EEEEE] 236 237The regions that will be seen within this address range then are:: 238 239 [CCCCCCCCCCCC][DDDDD][CCCCC][EEEEE][CCCCC] 240 241Since B has higher priority than C, its subregions appear in the flat map 242even where they overlap with C. In ranges where B has not mapped anything 243C's region appears. 244 245If B had provided its own MMIO operations (ie it was not a pure container) 246then these would be used for any addresses in its range not handled by 247D or E, and the result would be:: 248 249 [CCCCCCCCCCCC][DDDDD][BBBBB][EEEEE][BBBBB] 250 251Priority values are local to a container, because the priorities of two 252regions are only compared when they are both children of the same container. 253This means that the device in charge of the container (typically modelling 254a bus or a memory controller) can use them to manage the interaction of 255its child regions without any side effects on other parts of the system. 256In the example above, the priorities of D and E are unimportant because 257they do not overlap each other. It is the relative priority of B and C 258that causes D and E to appear on top of C: D and E's priorities are never 259compared against the priority of C. 260 261Visibility 262---------- 263The memory core uses the following rules to select a memory region when the 264guest accesses an address: 265 266- all direct subregions of the root region are matched against the address, in 267 descending priority order 268 269 - if the address lies outside the region offset/size, the subregion is 270 discarded 271 - if the subregion is a leaf (RAM or MMIO), the search terminates, returning 272 this leaf region 273 - if the subregion is a container, the same algorithm is used within the 274 subregion (after the address is adjusted by the subregion offset) 275 - if the subregion is an alias, the search is continued at the alias target 276 (after the address is adjusted by the subregion offset and alias offset) 277 - if a recursive search within a container or alias subregion does not 278 find a match (because of a "hole" in the container's coverage of its 279 address range), then if this is a container with its own MMIO or RAM 280 backing the search terminates, returning the container itself. Otherwise 281 we continue with the next subregion in priority order 282 283- if none of the subregions match the address then the search terminates 284 with no match found 285 286Example memory map 287------------------ 288 289:: 290 291 system_memory: container@0-2^48-1 292 | 293 +---- lomem: alias@0-0xdfffffff ---> #ram (0-0xdfffffff) 294 | 295 +---- himem: alias@0x100000000-0x11fffffff ---> #ram (0xe0000000-0xffffffff) 296 | 297 +---- vga-window: alias@0xa0000-0xbffff ---> #pci (0xa0000-0xbffff) 298 | (prio 1) 299 | 300 +---- pci-hole: alias@0xe0000000-0xffffffff ---> #pci (0xe0000000-0xffffffff) 301 302 pci (0-2^32-1) 303 | 304 +--- vga-area: container@0xa0000-0xbffff 305 | | 306 | +--- alias@0x00000-0x7fff ---> #vram (0x010000-0x017fff) 307 | | 308 | +--- alias@0x08000-0xffff ---> #vram (0x020000-0x027fff) 309 | 310 +---- vram: ram@0xe1000000-0xe1ffffff 311 | 312 +---- vga-mmio: mmio@0xe2000000-0xe200ffff 313 314 ram: ram@0x00000000-0xffffffff 315 316This is a (simplified) PC memory map. The 4GB RAM block is mapped into the 317system address space via two aliases: "lomem" is a 1:1 mapping of the first 3183.5GB; "himem" maps the last 0.5GB at address 4GB. This leaves 0.5GB for the 319so-called PCI hole, that allows a 32-bit PCI bus to exist in a system with 3204GB of memory. 321 322The memory controller diverts addresses in the range 640K-768K to the PCI 323address space. This is modelled using the "vga-window" alias, mapped at a 324higher priority so it obscures the RAM at the same addresses. The vga window 325can be removed by programming the memory controller; this is modelled by 326removing the alias and exposing the RAM underneath. 327 328The pci address space is not a direct child of the system address space, since 329we only want parts of it to be visible (we accomplish this using aliases). 330It has two subregions: vga-area models the legacy vga window and is occupied 331by two 32K memory banks pointing at two sections of the framebuffer. 332In addition the vram is mapped as a BAR at address e1000000, and an additional 333BAR containing MMIO registers is mapped after it. 334 335Note that if the guest maps a BAR outside the PCI hole, it would not be 336visible as the pci-hole alias clips it to a 0.5GB range. 337 338MMIO Operations 339--------------- 340 341MMIO regions are provided with ->read() and ->write() callbacks, 342which are sufficient for most devices. Some devices change behaviour 343based on the attributes used for the memory transaction, or need 344to be able to respond that the access should provoke a bus error 345rather than completing successfully; those devices can use the 346->read_with_attrs() and ->write_with_attrs() callbacks instead. 347 348In addition various constraints can be supplied to control how these 349callbacks are called: 350 351- .valid.min_access_size, .valid.max_access_size define the access sizes 352 (in bytes) which the device accepts; accesses outside this range will 353 have device and bus specific behaviour (ignored, or machine check) 354- .valid.unaligned specifies that the *device being modelled* supports 355 unaligned accesses; if false, unaligned accesses will invoke the 356 appropriate bus or CPU specific behaviour. 357- .impl.min_access_size, .impl.max_access_size define the access sizes 358 (in bytes) supported by the *implementation*; other access sizes will be 359 emulated using the ones available. For example a 4-byte write will be 360 emulated using four 1-byte writes, if .impl.max_access_size = 1. 361- .impl.unaligned specifies that the *implementation* supports unaligned 362 accesses; if false, unaligned accesses will be emulated by two aligned 363 accesses. 364