Based on kernel version 4.3. Page generated on 2015-11-02 12:51 EST.
1 The x86 kvm shadow mmu 2 ====================== 3 4 The mmu (in arch/x86/kvm, files mmu.[ch] and paging_tmpl.h) is responsible 5 for presenting a standard x86 mmu to the guest, while translating guest 6 physical addresses to host physical addresses. 7 8 The mmu code attempts to satisfy the following requirements: 9 10 - correctness: the guest should not be able to determine that it is running 11 on an emulated mmu except for timing (we attempt to comply 12 with the specification, not emulate the characteristics of 13 a particular implementation such as tlb size) 14 - security: the guest must not be able to touch host memory not assigned 15 to it 16 - performance: minimize the performance penalty imposed by the mmu 17 - scaling: need to scale to large memory and large vcpu guests 18 - hardware: support the full range of x86 virtualization hardware 19 - integration: Linux memory management code must be in control of guest memory 20 so that swapping, page migration, page merging, transparent 21 hugepages, and similar features work without change 22 - dirty tracking: report writes to guest memory to enable live migration 23 and framebuffer-based displays 24 - footprint: keep the amount of pinned kernel memory low (most memory 25 should be shrinkable) 26 - reliability: avoid multipage or GFP_ATOMIC allocations 27 28 Acronyms 29 ======== 30 31 pfn host page frame number 32 hpa host physical address 33 hva host virtual address 34 gfn guest frame number 35 gpa guest physical address 36 gva guest virtual address 37 ngpa nested guest physical address 38 ngva nested guest virtual address 39 pte page table entry (used also to refer generically to paging structure 40 entries) 41 gpte guest pte (referring to gfns) 42 spte shadow pte (referring to pfns) 43 tdp two dimensional paging (vendor neutral term for NPT and EPT) 44 45 Virtual and real hardware supported 46 =================================== 47 48 The mmu supports first-generation mmu hardware, which allows an atomic switch 49 of the current paging mode and cr3 during guest entry, as well as 50 two-dimensional paging (AMD's NPT and Intel's EPT). The emulated hardware 51 it exposes is the traditional 2/3/4 level x86 mmu, with support for global 52 pages, pae, pse, pse36, cr0.wp, and 1GB pages. Work is in progress to support 53 exposing NPT capable hardware on NPT capable hosts. 54 55 Translation 56 =========== 57 58 The primary job of the mmu is to program the processor's mmu to translate 59 addresses for the guest. Different translations are required at different 60 times: 61 62 - when guest paging is disabled, we translate guest physical addresses to 63 host physical addresses (gpa->hpa) 64 - when guest paging is enabled, we translate guest virtual addresses, to 65 guest physical addresses, to host physical addresses (gva->gpa->hpa) 66 - when the guest launches a guest of its own, we translate nested guest 67 virtual addresses, to nested guest physical addresses, to guest physical 68 addresses, to host physical addresses (ngva->ngpa->gpa->hpa) 69 70 The primary challenge is to encode between 1 and 3 translations into hardware 71 that support only 1 (traditional) and 2 (tdp) translations. When the 72 number of required translations matches the hardware, the mmu operates in 73 direct mode; otherwise it operates in shadow mode (see below). 74 75 Memory 76 ====== 77 78 Guest memory (gpa) is part of the user address space of the process that is 79 using kvm. Userspace defines the translation between guest addresses and user 80 addresses (gpa->hva); note that two gpas may alias to the same hva, but not 81 vice versa. 82 83 These hvas may be backed using any method available to the host: anonymous 84 memory, file backed memory, and device memory. Memory might be paged by the 85 host at any time. 86 87 Events 88 ====== 89 90 The mmu is driven by events, some from the guest, some from the host. 91 92 Guest generated events: 93 - writes to control registers (especially cr3) 94 - invlpg/invlpga instruction execution 95 - access to missing or protected translations 96 97 Host generated events: 98 - changes in the gpa->hpa translation (either through gpa->hva changes or 99 through hva->hpa changes) 100 - memory pressure (the shrinker) 101 102 Shadow pages 103 ============ 104 105 The principal data structure is the shadow page, 'struct kvm_mmu_page'. A 106 shadow page contains 512 sptes, which can be either leaf or nonleaf sptes. A 107 shadow page may contain a mix of leaf and nonleaf sptes. 108 109 A nonleaf spte allows the hardware mmu to reach the leaf pages and 110 is not related to a translation directly. It points to other shadow pages. 111 112 A leaf spte corresponds to either one or two translations encoded into 113 one paging structure entry. These are always the lowest level of the 114 translation stack, with optional higher level translations left to NPT/EPT. 115 Leaf ptes point at guest pages. 116 117 The following table shows translations encoded by leaf ptes, with higher-level 118 translations in parentheses: 119 120 Non-nested guests: 121 nonpaging: gpa->hpa 122 paging: gva->gpa->hpa 123 paging, tdp: (gva->)gpa->hpa 124 Nested guests: 125 non-tdp: ngva->gpa->hpa (*) 126 tdp: (ngva->)ngpa->gpa->hpa 127 128 (*) the guest hypervisor will encode the ngva->gpa translation into its page 129 tables if npt is not present 130 131 Shadow pages contain the following information: 132 role.level: 133 The level in the shadow paging hierarchy that this shadow page belongs to. 134 1=4k sptes, 2=2M sptes, 3=1G sptes, etc. 135 role.direct: 136 If set, leaf sptes reachable from this page are for a linear range. 137 Examples include real mode translation, large guest pages backed by small 138 host pages, and gpa->hpa translations when NPT or EPT is active. 139 The linear range starts at (gfn << PAGE_SHIFT) and its size is determined 140 by role.level (2MB for first level, 1GB for second level, 0.5TB for third 141 level, 256TB for fourth level) 142 If clear, this page corresponds to a guest page table denoted by the gfn 143 field. 144 role.quadrant: 145 When role.cr4_pae=0, the guest uses 32-bit gptes while the host uses 64-bit 146 sptes. That means a guest page table contains more ptes than the host, 147 so multiple shadow pages are needed to shadow one guest page. 148 For first-level shadow pages, role.quadrant can be 0 or 1 and denotes the 149 first or second 512-gpte block in the guest page table. For second-level 150 page tables, each 32-bit gpte is converted to two 64-bit sptes 151 (since each first-level guest page is shadowed by two first-level 152 shadow pages) so role.quadrant takes values in the range 0..3. Each 153 quadrant maps 1GB virtual address space. 154 role.access: 155 Inherited guest access permissions in the form uwx. Note execute 156 permission is positive, not negative. 157 role.invalid: 158 The page is invalid and should not be used. It is a root page that is 159 currently pinned (by a cpu hardware register pointing to it); once it is 160 unpinned it will be destroyed. 161 role.cr4_pae: 162 Contains the value of cr4.pae for which the page is valid (e.g. whether 163 32-bit or 64-bit gptes are in use). 164 role.nxe: 165 Contains the value of efer.nxe for which the page is valid. 166 role.cr0_wp: 167 Contains the value of cr0.wp for which the page is valid. 168 role.smep_andnot_wp: 169 Contains the value of cr4.smep && !cr0.wp for which the page is valid 170 (pages for which this is true are different from other pages; see the 171 treatment of cr0.wp=0 below). 172 role.smap_andnot_wp: 173 Contains the value of cr4.smap && !cr0.wp for which the page is valid 174 (pages for which this is true are different from other pages; see the 175 treatment of cr0.wp=0 below). 176 role.smm: 177 Is 1 if the page is valid in system management mode. This field 178 determines which of the kvm_memslots array was used to build this 179 shadow page; it is also used to go back from a struct kvm_mmu_page 180 to a memslot, through the kvm_memslots_for_spte_role macro and 181 __gfn_to_memslot. 182 gfn: 183 Either the guest page table containing the translations shadowed by this 184 page, or the base page frame for linear translations. See role.direct. 185 spt: 186 A pageful of 64-bit sptes containing the translations for this page. 187 Accessed by both kvm and hardware. 188 The page pointed to by spt will have its page->private pointing back 189 at the shadow page structure. 190 sptes in spt point either at guest pages, or at lower-level shadow pages. 191 Specifically, if sp1 and sp2 are shadow pages, then sp1->spt[n] may point 192 at __pa(sp2->spt). sp2 will point back at sp1 through parent_pte. 193 The spt array forms a DAG structure with the shadow page as a node, and 194 guest pages as leaves. 195 gfns: 196 An array of 512 guest frame numbers, one for each present pte. Used to 197 perform a reverse map from a pte to a gfn. When role.direct is set, any 198 element of this array can be calculated from the gfn field when used, in 199 this case, the array of gfns is not allocated. See role.direct and gfn. 200 root_count: 201 A counter keeping track of how many hardware registers (guest cr3 or 202 pdptrs) are now pointing at the page. While this counter is nonzero, the 203 page cannot be destroyed. See role.invalid. 204 parent_ptes: 205 The reverse mapping for the pte/ptes pointing at this page's spt. If 206 parent_ptes bit 0 is zero, only one spte points at this pages and 207 parent_ptes points at this single spte, otherwise, there exists multiple 208 sptes pointing at this page and (parent_ptes & ~0x1) points at a data 209 structure with a list of parent_ptes. 210 unsync: 211 If true, then the translations in this page may not match the guest's 212 translation. This is equivalent to the state of the tlb when a pte is 213 changed but before the tlb entry is flushed. Accordingly, unsync ptes 214 are synchronized when the guest executes invlpg or flushes its tlb by 215 other means. Valid for leaf pages. 216 unsync_children: 217 How many sptes in the page point at pages that are unsync (or have 218 unsynchronized children). 219 unsync_child_bitmap: 220 A bitmap indicating which sptes in spt point (directly or indirectly) at 221 pages that may be unsynchronized. Used to quickly locate all unsychronized 222 pages reachable from a given page. 223 mmu_valid_gen: 224 Generation number of the page. It is compared with kvm->arch.mmu_valid_gen 225 during hash table lookup, and used to skip invalidated shadow pages (see 226 "Zapping all pages" below.) 227 clear_spte_count: 228 Only present on 32-bit hosts, where a 64-bit spte cannot be written 229 atomically. The reader uses this while running out of the MMU lock 230 to detect in-progress updates and retry them until the writer has 231 finished the write. 232 write_flooding_count: 233 A guest may write to a page table many times, causing a lot of 234 emulations if the page needs to be write-protected (see "Synchronized 235 and unsynchronized pages" below). Leaf pages can be unsynchronized 236 so that they do not trigger frequent emulation, but this is not 237 possible for non-leafs. This field counts the number of emulations 238 since the last time the page table was actually used; if emulation 239 is triggered too frequently on this page, KVM will unmap the page 240 to avoid emulation in the future. 241 242 Reverse map 243 =========== 244 245 The mmu maintains a reverse mapping whereby all ptes mapping a page can be 246 reached given its gfn. This is used, for example, when swapping out a page. 247 248 Synchronized and unsynchronized pages 249 ===================================== 250 251 The guest uses two events to synchronize its tlb and page tables: tlb flushes 252 and page invalidations (invlpg). 253 254 A tlb flush means that we need to synchronize all sptes reachable from the 255 guest's cr3. This is expensive, so we keep all guest page tables write 256 protected, and synchronize sptes to gptes when a gpte is written. 257 258 A special case is when a guest page table is reachable from the current 259 guest cr3. In this case, the guest is obliged to issue an invlpg instruction 260 before using the translation. We take advantage of that by removing write 261 protection from the guest page, and allowing the guest to modify it freely. 262 We synchronize modified gptes when the guest invokes invlpg. This reduces 263 the amount of emulation we have to do when the guest modifies multiple gptes, 264 or when the a guest page is no longer used as a page table and is used for 265 random guest data. 266 267 As a side effect we have to resynchronize all reachable unsynchronized shadow 268 pages on a tlb flush. 269 270 271 Reaction to events 272 ================== 273 274 - guest page fault (or npt page fault, or ept violation) 275 276 This is the most complicated event. The cause of a page fault can be: 277 278 - a true guest fault (the guest translation won't allow the access) (*) 279 - access to a missing translation 280 - access to a protected translation 281 - when logging dirty pages, memory is write protected 282 - synchronized shadow pages are write protected (*) 283 - access to untranslatable memory (mmio) 284 285 (*) not applicable in direct mode 286 287 Handling a page fault is performed as follows: 288 289 - if the RSV bit of the error code is set, the page fault is caused by guest 290 accessing MMIO and cached MMIO information is available. 291 - walk shadow page table 292 - check for valid generation number in the spte (see "Fast invalidation of 293 MMIO sptes" below) 294 - cache the information to vcpu->arch.mmio_gva, vcpu->arch.access and 295 vcpu->arch.mmio_gfn, and call the emulator 296 - If both P bit and R/W bit of error code are set, this could possibly 297 be handled as a "fast page fault" (fixed without taking the MMU lock). See 298 the description in Documentation/virtual/kvm/locking.txt. 299 - if needed, walk the guest page tables to determine the guest translation 300 (gva->gpa or ngpa->gpa) 301 - if permissions are insufficient, reflect the fault back to the guest 302 - determine the host page 303 - if this is an mmio request, there is no host page; cache the info to 304 vcpu->arch.mmio_gva, vcpu->arch.access and vcpu->arch.mmio_gfn 305 - walk the shadow page table to find the spte for the translation, 306 instantiating missing intermediate page tables as necessary 307 - If this is an mmio request, cache the mmio info to the spte and set some 308 reserved bit on the spte (see callers of kvm_mmu_set_mmio_spte_mask) 309 - try to unsynchronize the page 310 - if successful, we can let the guest continue and modify the gpte 311 - emulate the instruction 312 - if failed, unshadow the page and let the guest continue 313 - update any translations that were modified by the instruction 314 315 invlpg handling: 316 317 - walk the shadow page hierarchy and drop affected translations 318 - try to reinstantiate the indicated translation in the hope that the 319 guest will use it in the near future 320 321 Guest control register updates: 322 323 - mov to cr3 324 - look up new shadow roots 325 - synchronize newly reachable shadow pages 326 327 - mov to cr0/cr4/efer 328 - set up mmu context for new paging mode 329 - look up new shadow roots 330 - synchronize newly reachable shadow pages 331 332 Host translation updates: 333 334 - mmu notifier called with updated hva 335 - look up affected sptes through reverse map 336 - drop (or update) translations 337 338 Emulating cr0.wp 339 ================ 340 341 If tdp is not enabled, the host must keep cr0.wp=1 so page write protection 342 works for the guest kernel, not guest guest userspace. When the guest 343 cr0.wp=1, this does not present a problem. However when the guest cr0.wp=0, 344 we cannot map the permissions for gpte.u=1, gpte.w=0 to any spte (the 345 semantics require allowing any guest kernel access plus user read access). 346 347 We handle this by mapping the permissions to two possible sptes, depending 348 on fault type: 349 350 - kernel write fault: spte.u=0, spte.w=1 (allows full kernel access, 351 disallows user access) 352 - read fault: spte.u=1, spte.w=0 (allows full read access, disallows kernel 353 write access) 354 355 (user write faults generate a #PF) 356 357 In the first case there are two additional complications: 358 - if CR4.SMEP is enabled: since we've turned the page into a kernel page, 359 the kernel may now execute it. We handle this by also setting spte.nx. 360 If we get a user fetch or read fault, we'll change spte.u=1 and 361 spte.nx=gpte.nx back. 362 - if CR4.SMAP is disabled: since the page has been changed to a kernel 363 page, it can not be reused when CR4.SMAP is enabled. We set 364 CR4.SMAP && !CR0.WP into shadow page's role to avoid this case. Note, 365 here we do not care the case that CR4.SMAP is enabled since KVM will 366 directly inject #PF to guest due to failed permission check. 367 368 To prevent an spte that was converted into a kernel page with cr0.wp=0 369 from being written by the kernel after cr0.wp has changed to 1, we make 370 the value of cr0.wp part of the page role. This means that an spte created 371 with one value of cr0.wp cannot be used when cr0.wp has a different value - 372 it will simply be missed by the shadow page lookup code. A similar issue 373 exists when an spte created with cr0.wp=0 and cr4.smep=0 is used after 374 changing cr4.smep to 1. To avoid this, the value of !cr0.wp && cr4.smep 375 is also made a part of the page role. 376 377 Large pages 378 =========== 379 380 The mmu supports all combinations of large and small guest and host pages. 381 Supported page sizes include 4k, 2M, 4M, and 1G. 4M pages are treated as 382 two separate 2M pages, on both guest and host, since the mmu always uses PAE 383 paging. 384 385 To instantiate a large spte, four constraints must be satisfied: 386 387 - the spte must point to a large host page 388 - the guest pte must be a large pte of at least equivalent size (if tdp is 389 enabled, there is no guest pte and this condition is satisfied) 390 - if the spte will be writeable, the large page frame may not overlap any 391 write-protected pages 392 - the guest page must be wholly contained by a single memory slot 393 394 To check the last two conditions, the mmu maintains a ->write_count set of 395 arrays for each memory slot and large page size. Every write protected page 396 causes its write_count to be incremented, thus preventing instantiation of 397 a large spte. The frames at the end of an unaligned memory slot have 398 artificially inflated ->write_counts so they can never be instantiated. 399 400 Zapping all pages (page generation count) 401 ========================================= 402 403 For the large memory guests, walking and zapping all pages is really slow 404 (because there are a lot of pages), and also blocks memory accesses of 405 all VCPUs because it needs to hold the MMU lock. 406 407 To make it be more scalable, kvm maintains a global generation number 408 which is stored in kvm->arch.mmu_valid_gen. Every shadow page stores 409 the current global generation-number into sp->mmu_valid_gen when it 410 is created. Pages with a mismatching generation number are "obsolete". 411 412 When KVM need zap all shadow pages sptes, it just simply increases the global 413 generation-number then reload root shadow pages on all vcpus. As the VCPUs 414 create new shadow page tables, the old pages are not used because of the 415 mismatching generation number. 416 417 KVM then walks through all pages and zaps obsolete pages. While the zap 418 operation needs to take the MMU lock, the lock can be released periodically 419 so that the VCPUs can make progress. 420 421 Fast invalidation of MMIO sptes 422 =============================== 423 424 As mentioned in "Reaction to events" above, kvm will cache MMIO 425 information in leaf sptes. When a new memslot is added or an existing 426 memslot is changed, this information may become stale and needs to be 427 invalidated. This also needs to hold the MMU lock while walking all 428 shadow pages, and is made more scalable with a similar technique. 429 430 MMIO sptes have a few spare bits, which are used to store a 431 generation number. The global generation number is stored in 432 kvm_memslots(kvm)->generation, and increased whenever guest memory info 433 changes. This generation number is distinct from the one described in 434 the previous section. 435 436 When KVM finds an MMIO spte, it checks the generation number of the spte. 437 If the generation number of the spte does not equal the global generation 438 number, it will ignore the cached MMIO information and handle the page 439 fault through the slow path. 440 441 Since only 19 bits are used to store generation-number on mmio spte, all 442 pages are zapped when there is an overflow. 443 444 Unfortunately, a single memory access might access kvm_memslots(kvm) multiple 445 times, the last one happening when the generation number is retrieved and 446 stored into the MMIO spte. Thus, the MMIO spte might be created based on 447 out-of-date information, but with an up-to-date generation number. 448 449 To avoid this, the generation number is incremented again after synchronize_srcu 450 returns; thus, the low bit of kvm_memslots(kvm)->generation is only 1 during a 451 memslot update, while some SRCU readers might be using the old copy. We do not 452 want to use an MMIO sptes created with an odd generation number, and we can do 453 this without losing a bit in the MMIO spte. The low bit of the generation 454 is not stored in MMIO spte, and presumed zero when it is extracted out of the 455 spte. If KVM is unlucky and creates an MMIO spte while the low bit is 1, 456 the next access to the spte will always be a cache miss. 457 458 459 Further reading 460 =============== 461 462 - NPT presentation from KVM Forum 2008 463 http://www.linux-kvm.org/wiki/images/c/c8/KvmForum2008%24kdf2008_21.pdf