MemoryManager.cpp 52 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223
  1. /*
  2. * Copyright (c) 2018-2022, Andreas Kling <kling@serenityos.org>
  3. *
  4. * SPDX-License-Identifier: BSD-2-Clause
  5. */
  6. #include <AK/Assertions.h>
  7. #include <AK/Memory.h>
  8. #include <AK/NonnullRefPtrVector.h>
  9. #include <AK/StringView.h>
  10. #include <Kernel/Arch/CPU.h>
  11. #include <Kernel/Arch/InterruptDisabler.h>
  12. #include <Kernel/Arch/PageDirectory.h>
  13. #include <Kernel/Arch/PageFault.h>
  14. #include <Kernel/Arch/RegisterState.h>
  15. #include <Kernel/BootInfo.h>
  16. #include <Kernel/FileSystem/Inode.h>
  17. #include <Kernel/Heap/kmalloc.h>
  18. #include <Kernel/KSyms.h>
  19. #include <Kernel/Memory/AnonymousVMObject.h>
  20. #include <Kernel/Memory/MemoryManager.h>
  21. #include <Kernel/Memory/PageDirectory.h>
  22. #include <Kernel/Memory/PhysicalRegion.h>
  23. #include <Kernel/Memory/SharedInodeVMObject.h>
  24. #include <Kernel/Multiboot.h>
  25. #include <Kernel/Panic.h>
  26. #include <Kernel/Prekernel/Prekernel.h>
  27. #include <Kernel/Process.h>
  28. #include <Kernel/Sections.h>
  29. #include <Kernel/StdLib.h>
  30. extern u8 start_of_kernel_image[];
  31. extern u8 end_of_kernel_image[];
  32. extern u8 start_of_kernel_text[];
  33. extern u8 start_of_kernel_data[];
  34. extern u8 end_of_kernel_bss[];
  35. extern u8 start_of_ro_after_init[];
  36. extern u8 end_of_ro_after_init[];
  37. extern u8 start_of_unmap_after_init[];
  38. extern u8 end_of_unmap_after_init[];
  39. extern u8 start_of_kernel_ksyms[];
  40. extern u8 end_of_kernel_ksyms[];
  41. extern multiboot_module_entry_t multiboot_copy_boot_modules_array[16];
  42. extern size_t multiboot_copy_boot_modules_count;
  43. namespace Kernel::Memory {
  44. ErrorOr<FlatPtr> page_round_up(FlatPtr x)
  45. {
  46. if (x > (explode_byte(0xFF) & ~0xFFF)) {
  47. return Error::from_errno(EINVAL);
  48. }
  49. return (((FlatPtr)(x)) + PAGE_SIZE - 1) & (~(PAGE_SIZE - 1));
  50. }
  51. // NOTE: We can NOT use Singleton for this class, because
  52. // MemoryManager::initialize is called *before* global constructors are
  53. // run. If we do, then Singleton would get re-initialized, causing
  54. // the memory manager to be initialized twice!
  55. static MemoryManager* s_the;
  56. MemoryManager& MemoryManager::the()
  57. {
  58. return *s_the;
  59. }
  60. bool MemoryManager::is_initialized()
  61. {
  62. return s_the != nullptr;
  63. }
  64. static UNMAP_AFTER_INIT VirtualRange kernel_virtual_range()
  65. {
  66. size_t kernel_range_start = kernel_mapping_base + 2 * MiB; // The first 2 MiB are used for mapping the pre-kernel
  67. return VirtualRange { VirtualAddress(kernel_range_start), KERNEL_PD_END - kernel_range_start };
  68. }
  69. MemoryManager::GlobalData::GlobalData()
  70. : region_tree(kernel_virtual_range())
  71. {
  72. }
  73. UNMAP_AFTER_INIT MemoryManager::MemoryManager()
  74. : m_global_data(LockRank::None)
  75. {
  76. s_the = this;
  77. parse_memory_map();
  78. activate_kernel_page_directory(kernel_page_directory());
  79. protect_kernel_image();
  80. // We're temporarily "committing" to two pages that we need to allocate below
  81. auto committed_pages = commit_physical_pages(2).release_value();
  82. m_shared_zero_page = committed_pages.take_one();
  83. // We're wasting a page here, we just need a special tag (physical
  84. // address) so that we know when we need to lazily allocate a page
  85. // that we should be drawing this page from the committed pool rather
  86. // than potentially failing if no pages are available anymore.
  87. // By using a tag we don't have to query the VMObject for every page
  88. // whether it was committed or not
  89. m_lazy_committed_page = committed_pages.take_one();
  90. }
  91. UNMAP_AFTER_INIT MemoryManager::~MemoryManager() = default;
  92. UNMAP_AFTER_INIT void MemoryManager::protect_kernel_image()
  93. {
  94. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  95. // Disable writing to the kernel text and rodata segments.
  96. for (auto const* i = start_of_kernel_text; i < start_of_kernel_data; i += PAGE_SIZE) {
  97. auto& pte = *ensure_pte(kernel_page_directory(), VirtualAddress(i));
  98. pte.set_writable(false);
  99. }
  100. if (Processor::current().has_nx()) {
  101. // Disable execution of the kernel data, bss and heap segments.
  102. for (auto const* i = start_of_kernel_data; i < end_of_kernel_image; i += PAGE_SIZE) {
  103. auto& pte = *ensure_pte(kernel_page_directory(), VirtualAddress(i));
  104. pte.set_execute_disabled(true);
  105. }
  106. }
  107. }
  108. UNMAP_AFTER_INIT void MemoryManager::unmap_prekernel()
  109. {
  110. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  111. auto start = start_of_prekernel_image.page_base().get();
  112. auto end = end_of_prekernel_image.page_base().get();
  113. for (auto i = start; i <= end; i += PAGE_SIZE)
  114. release_pte(kernel_page_directory(), VirtualAddress(i), i == end ? IsLastPTERelease::Yes : IsLastPTERelease::No);
  115. flush_tlb(&kernel_page_directory(), VirtualAddress(start), (end - start) / PAGE_SIZE);
  116. }
  117. UNMAP_AFTER_INIT void MemoryManager::protect_readonly_after_init_memory()
  118. {
  119. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  120. // Disable writing to the .ro_after_init section
  121. for (auto i = (FlatPtr)&start_of_ro_after_init; i < (FlatPtr)&end_of_ro_after_init; i += PAGE_SIZE) {
  122. auto& pte = *ensure_pte(kernel_page_directory(), VirtualAddress(i));
  123. pte.set_writable(false);
  124. flush_tlb(&kernel_page_directory(), VirtualAddress(i));
  125. }
  126. }
  127. void MemoryManager::unmap_text_after_init()
  128. {
  129. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  130. auto start = page_round_down((FlatPtr)&start_of_unmap_after_init);
  131. auto end = page_round_up((FlatPtr)&end_of_unmap_after_init).release_value_but_fixme_should_propagate_errors();
  132. // Unmap the entire .unmap_after_init section
  133. for (auto i = start; i < end; i += PAGE_SIZE) {
  134. auto& pte = *ensure_pte(kernel_page_directory(), VirtualAddress(i));
  135. pte.clear();
  136. flush_tlb(&kernel_page_directory(), VirtualAddress(i));
  137. }
  138. dmesgln("Unmapped {} KiB of kernel text after init! :^)", (end - start) / KiB);
  139. }
  140. UNMAP_AFTER_INIT void MemoryManager::protect_ksyms_after_init()
  141. {
  142. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  143. auto start = page_round_down((FlatPtr)start_of_kernel_ksyms);
  144. auto end = page_round_up((FlatPtr)end_of_kernel_ksyms).release_value_but_fixme_should_propagate_errors();
  145. for (auto i = start; i < end; i += PAGE_SIZE) {
  146. auto& pte = *ensure_pte(kernel_page_directory(), VirtualAddress(i));
  147. pte.set_writable(false);
  148. flush_tlb(&kernel_page_directory(), VirtualAddress(i));
  149. }
  150. dmesgln("Write-protected kernel symbols after init.");
  151. }
  152. IterationDecision MemoryManager::for_each_physical_memory_range(Function<IterationDecision(PhysicalMemoryRange const&)> callback)
  153. {
  154. return m_global_data.with([&](auto& global_data) {
  155. VERIFY(!global_data.physical_memory_ranges.is_empty());
  156. for (auto& current_range : global_data.physical_memory_ranges) {
  157. IterationDecision decision = callback(current_range);
  158. if (decision != IterationDecision::Continue)
  159. return decision;
  160. }
  161. return IterationDecision::Continue;
  162. });
  163. }
  164. UNMAP_AFTER_INIT void MemoryManager::register_reserved_ranges()
  165. {
  166. m_global_data.with([&](auto& global_data) {
  167. VERIFY(!global_data.physical_memory_ranges.is_empty());
  168. ContiguousReservedMemoryRange range;
  169. for (auto& current_range : global_data.physical_memory_ranges) {
  170. if (current_range.type != PhysicalMemoryRangeType::Reserved) {
  171. if (range.start.is_null())
  172. continue;
  173. global_data.reserved_memory_ranges.append(ContiguousReservedMemoryRange { range.start, current_range.start.get() - range.start.get() });
  174. range.start.set((FlatPtr) nullptr);
  175. continue;
  176. }
  177. if (!range.start.is_null()) {
  178. continue;
  179. }
  180. range.start = current_range.start;
  181. }
  182. if (global_data.physical_memory_ranges.last().type != PhysicalMemoryRangeType::Reserved)
  183. return;
  184. if (range.start.is_null())
  185. return;
  186. global_data.reserved_memory_ranges.append(ContiguousReservedMemoryRange { range.start, global_data.physical_memory_ranges.last().start.get() + global_data.physical_memory_ranges.last().length - range.start.get() });
  187. });
  188. }
  189. bool MemoryManager::is_allowed_to_read_physical_memory_for_userspace(PhysicalAddress start_address, size_t read_length) const
  190. {
  191. // Note: Guard against overflow in case someone tries to mmap on the edge of
  192. // the RAM
  193. if (start_address.offset_addition_would_overflow(read_length))
  194. return false;
  195. auto end_address = start_address.offset(read_length);
  196. return m_global_data.with([&](auto& global_data) {
  197. for (auto const& current_range : global_data.reserved_memory_ranges) {
  198. if (current_range.start > start_address)
  199. continue;
  200. if (current_range.start.offset(current_range.length) < end_address)
  201. continue;
  202. return true;
  203. }
  204. return false;
  205. });
  206. }
  207. UNMAP_AFTER_INIT void MemoryManager::parse_memory_map()
  208. {
  209. // Register used memory regions that we know of.
  210. m_global_data.with([&](auto& global_data) {
  211. global_data.used_memory_ranges.ensure_capacity(4);
  212. #if ARCH(I386) || ARCH(X86_64)
  213. global_data.used_memory_ranges.append(UsedMemoryRange { UsedMemoryRangeType::LowMemory, PhysicalAddress(0x00000000), PhysicalAddress(1 * MiB) });
  214. #endif
  215. global_data.used_memory_ranges.append(UsedMemoryRange { UsedMemoryRangeType::Kernel, PhysicalAddress(virtual_to_low_physical((FlatPtr)start_of_kernel_image)), PhysicalAddress(page_round_up(virtual_to_low_physical((FlatPtr)end_of_kernel_image)).release_value_but_fixme_should_propagate_errors()) });
  216. if (multiboot_flags & 0x4) {
  217. auto* bootmods_start = multiboot_copy_boot_modules_array;
  218. auto* bootmods_end = bootmods_start + multiboot_copy_boot_modules_count;
  219. for (auto* bootmod = bootmods_start; bootmod < bootmods_end; bootmod++) {
  220. global_data.used_memory_ranges.append(UsedMemoryRange { UsedMemoryRangeType::BootModule, PhysicalAddress(bootmod->start), PhysicalAddress(bootmod->end) });
  221. }
  222. }
  223. auto* mmap_begin = multiboot_memory_map;
  224. auto* mmap_end = multiboot_memory_map + multiboot_memory_map_count;
  225. struct ContiguousPhysicalVirtualRange {
  226. PhysicalAddress lower;
  227. PhysicalAddress upper;
  228. };
  229. Vector<ContiguousPhysicalVirtualRange> contiguous_physical_ranges;
  230. for (auto* mmap = mmap_begin; mmap < mmap_end; mmap++) {
  231. // We have to copy these onto the stack, because we take a reference to these when printing them out,
  232. // and doing so on a packed struct field is UB.
  233. auto address = mmap->addr;
  234. auto length = mmap->len;
  235. ArmedScopeGuard write_back_guard = [&]() {
  236. mmap->addr = address;
  237. mmap->len = length;
  238. };
  239. dmesgln("MM: Multiboot mmap: address={:p}, length={}, type={}", address, length, mmap->type);
  240. auto start_address = PhysicalAddress(address);
  241. switch (mmap->type) {
  242. case (MULTIBOOT_MEMORY_AVAILABLE):
  243. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::Usable, start_address, length });
  244. break;
  245. case (MULTIBOOT_MEMORY_RESERVED):
  246. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::Reserved, start_address, length });
  247. break;
  248. case (MULTIBOOT_MEMORY_ACPI_RECLAIMABLE):
  249. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::ACPI_Reclaimable, start_address, length });
  250. break;
  251. case (MULTIBOOT_MEMORY_NVS):
  252. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::ACPI_NVS, start_address, length });
  253. break;
  254. case (MULTIBOOT_MEMORY_BADRAM):
  255. dmesgln("MM: Warning, detected bad memory range!");
  256. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::BadMemory, start_address, length });
  257. break;
  258. default:
  259. dbgln("MM: Unknown range!");
  260. global_data.physical_memory_ranges.append(PhysicalMemoryRange { PhysicalMemoryRangeType::Unknown, start_address, length });
  261. break;
  262. }
  263. if (mmap->type != MULTIBOOT_MEMORY_AVAILABLE)
  264. continue;
  265. // Fix up unaligned memory regions.
  266. auto diff = (FlatPtr)address % PAGE_SIZE;
  267. if (diff != 0) {
  268. dmesgln("MM: Got an unaligned physical_region from the bootloader; correcting {:p} by {} bytes", address, diff);
  269. diff = PAGE_SIZE - diff;
  270. address += diff;
  271. length -= diff;
  272. }
  273. if ((length % PAGE_SIZE) != 0) {
  274. dmesgln("MM: Got an unaligned physical_region from the bootloader; correcting length {} by {} bytes", length, length % PAGE_SIZE);
  275. length -= length % PAGE_SIZE;
  276. }
  277. if (length < PAGE_SIZE) {
  278. dmesgln("MM: Memory physical_region from bootloader is too small; we want >= {} bytes, but got {} bytes", PAGE_SIZE, length);
  279. continue;
  280. }
  281. for (PhysicalSize page_base = address; page_base <= (address + length); page_base += PAGE_SIZE) {
  282. auto addr = PhysicalAddress(page_base);
  283. // Skip used memory ranges.
  284. bool should_skip = false;
  285. for (auto& used_range : global_data.used_memory_ranges) {
  286. if (addr.get() >= used_range.start.get() && addr.get() <= used_range.end.get()) {
  287. should_skip = true;
  288. break;
  289. }
  290. }
  291. if (should_skip)
  292. continue;
  293. if (contiguous_physical_ranges.is_empty() || contiguous_physical_ranges.last().upper.offset(PAGE_SIZE) != addr) {
  294. contiguous_physical_ranges.append(ContiguousPhysicalVirtualRange {
  295. .lower = addr,
  296. .upper = addr,
  297. });
  298. } else {
  299. contiguous_physical_ranges.last().upper = addr;
  300. }
  301. }
  302. }
  303. for (auto& range : contiguous_physical_ranges) {
  304. global_data.physical_regions.append(PhysicalRegion::try_create(range.lower, range.upper).release_nonnull());
  305. }
  306. for (auto& region : global_data.physical_regions)
  307. global_data.system_memory_info.physical_pages += region.size();
  308. register_reserved_ranges();
  309. for (auto& range : global_data.reserved_memory_ranges) {
  310. dmesgln("MM: Contiguous reserved range from {}, length is {}", range.start, range.length);
  311. }
  312. initialize_physical_pages();
  313. VERIFY(global_data.system_memory_info.physical_pages > 0);
  314. // We start out with no committed pages
  315. global_data.system_memory_info.physical_pages_uncommitted = global_data.system_memory_info.physical_pages;
  316. for (auto& used_range : global_data.used_memory_ranges) {
  317. dmesgln("MM: {} range @ {} - {} (size {:#x})", UserMemoryRangeTypeNames[to_underlying(used_range.type)], used_range.start, used_range.end.offset(-1), used_range.end.as_ptr() - used_range.start.as_ptr());
  318. }
  319. for (auto& region : global_data.physical_regions) {
  320. dmesgln("MM: User physical region: {} - {} (size {:#x})", region.lower(), region.upper().offset(-1), PAGE_SIZE * region.size());
  321. region.initialize_zones();
  322. }
  323. });
  324. }
  325. UNMAP_AFTER_INIT void MemoryManager::initialize_physical_pages()
  326. {
  327. m_global_data.with([&](auto& global_data) {
  328. // We assume that the physical page range is contiguous and doesn't contain huge gaps!
  329. PhysicalAddress highest_physical_address;
  330. for (auto& range : global_data.used_memory_ranges) {
  331. if (range.end.get() > highest_physical_address.get())
  332. highest_physical_address = range.end;
  333. }
  334. for (auto& region : global_data.physical_memory_ranges) {
  335. auto range_end = PhysicalAddress(region.start).offset(region.length);
  336. if (range_end.get() > highest_physical_address.get())
  337. highest_physical_address = range_end;
  338. }
  339. // Calculate how many total physical pages the array will have
  340. m_physical_page_entries_count = PhysicalAddress::physical_page_index(highest_physical_address.get()) + 1;
  341. VERIFY(m_physical_page_entries_count != 0);
  342. VERIFY(!Checked<decltype(m_physical_page_entries_count)>::multiplication_would_overflow(m_physical_page_entries_count, sizeof(PhysicalPageEntry)));
  343. // Calculate how many bytes the array will consume
  344. auto physical_page_array_size = m_physical_page_entries_count * sizeof(PhysicalPageEntry);
  345. auto physical_page_array_pages = page_round_up(physical_page_array_size).release_value_but_fixme_should_propagate_errors() / PAGE_SIZE;
  346. VERIFY(physical_page_array_pages * PAGE_SIZE >= physical_page_array_size);
  347. // Calculate how many page tables we will need to be able to map them all
  348. auto needed_page_table_count = (physical_page_array_pages + 512 - 1) / 512;
  349. auto physical_page_array_pages_and_page_tables_count = physical_page_array_pages + needed_page_table_count;
  350. // Now that we know how much memory we need for a contiguous array of PhysicalPage instances, find a memory region that can fit it
  351. PhysicalRegion* found_region { nullptr };
  352. Optional<size_t> found_region_index;
  353. for (size_t i = 0; i < global_data.physical_regions.size(); ++i) {
  354. auto& region = global_data.physical_regions[i];
  355. if (region.size() >= physical_page_array_pages_and_page_tables_count) {
  356. found_region = &region;
  357. found_region_index = i;
  358. break;
  359. }
  360. }
  361. if (!found_region) {
  362. dmesgln("MM: Need {} bytes for physical page management, but no memory region is large enough!", physical_page_array_pages_and_page_tables_count);
  363. VERIFY_NOT_REACHED();
  364. }
  365. VERIFY(global_data.system_memory_info.physical_pages >= physical_page_array_pages_and_page_tables_count);
  366. global_data.system_memory_info.physical_pages -= physical_page_array_pages_and_page_tables_count;
  367. if (found_region->size() == physical_page_array_pages_and_page_tables_count) {
  368. // We're stealing the entire region
  369. global_data.physical_pages_region = global_data.physical_regions.take(*found_region_index);
  370. } else {
  371. global_data.physical_pages_region = found_region->try_take_pages_from_beginning(physical_page_array_pages_and_page_tables_count);
  372. }
  373. global_data.used_memory_ranges.append({ UsedMemoryRangeType::PhysicalPages, global_data.physical_pages_region->lower(), global_data.physical_pages_region->upper() });
  374. // Create the bare page directory. This is not a fully constructed page directory and merely contains the allocators!
  375. m_kernel_page_directory = PageDirectory::must_create_kernel_page_directory();
  376. {
  377. // Carve out the whole page directory covering the kernel image to make MemoryManager::initialize_physical_pages() happy
  378. FlatPtr start_of_range = ((FlatPtr)start_of_kernel_image & ~(FlatPtr)0x1fffff);
  379. FlatPtr end_of_range = ((FlatPtr)end_of_kernel_image & ~(FlatPtr)0x1fffff) + 0x200000;
  380. MUST(global_data.region_tree.place_specifically(*MUST(Region::create_unbacked()).leak_ptr(), VirtualRange { VirtualAddress(start_of_range), end_of_range - start_of_range }));
  381. }
  382. // Allocate a virtual address range for our array
  383. // This looks awkward, but it basically creates a dummy region to occupy the address range permanently.
  384. auto& region = *MUST(Region::create_unbacked()).leak_ptr();
  385. MUST(global_data.region_tree.place_anywhere(region, RandomizeVirtualAddress::No, physical_page_array_pages * PAGE_SIZE));
  386. auto range = region.range();
  387. // Now that we have our special m_physical_pages_region region with enough pages to hold the entire array
  388. // try to map the entire region into kernel space so we always have it
  389. // We can't use ensure_pte here because it would try to allocate a PhysicalPage and we don't have the array
  390. // mapped yet so we can't create them
  391. // Create page tables at the beginning of m_physical_pages_region, followed by the PhysicalPageEntry array
  392. auto page_tables_base = global_data.physical_pages_region->lower();
  393. auto physical_page_array_base = page_tables_base.offset(needed_page_table_count * PAGE_SIZE);
  394. auto physical_page_array_current_page = physical_page_array_base.get();
  395. auto virtual_page_array_base = range.base().get();
  396. auto virtual_page_array_current_page = virtual_page_array_base;
  397. for (size_t pt_index = 0; pt_index < needed_page_table_count; pt_index++) {
  398. auto virtual_page_base_for_this_pt = virtual_page_array_current_page;
  399. auto pt_paddr = page_tables_base.offset(pt_index * PAGE_SIZE);
  400. auto* pt = reinterpret_cast<PageTableEntry*>(quickmap_page(pt_paddr));
  401. __builtin_memset(pt, 0, PAGE_SIZE);
  402. for (size_t pte_index = 0; pte_index < PAGE_SIZE / sizeof(PageTableEntry); pte_index++) {
  403. auto& pte = pt[pte_index];
  404. pte.set_physical_page_base(physical_page_array_current_page);
  405. pte.set_user_allowed(false);
  406. pte.set_writable(true);
  407. if (Processor::current().has_nx())
  408. pte.set_execute_disabled(false);
  409. pte.set_global(true);
  410. pte.set_present(true);
  411. physical_page_array_current_page += PAGE_SIZE;
  412. virtual_page_array_current_page += PAGE_SIZE;
  413. }
  414. unquickmap_page();
  415. // Hook the page table into the kernel page directory
  416. u32 page_directory_index = (virtual_page_base_for_this_pt >> 21) & 0x1ff;
  417. auto* pd = reinterpret_cast<PageDirectoryEntry*>(quickmap_page(boot_pd_kernel));
  418. PageDirectoryEntry& pde = pd[page_directory_index];
  419. // FIXME: port quickmap_page to aarch64
  420. #if !ARCH(AARCH64)
  421. VERIFY(!pde.is_present()); // Nothing should be using this PD yet
  422. #endif
  423. // We can't use ensure_pte quite yet!
  424. pde.set_page_table_base(pt_paddr.get());
  425. pde.set_user_allowed(false);
  426. pde.set_present(true);
  427. pde.set_writable(true);
  428. pde.set_global(true);
  429. unquickmap_page();
  430. flush_tlb_local(VirtualAddress(virtual_page_base_for_this_pt));
  431. }
  432. // We now have the entire PhysicalPageEntry array mapped!
  433. m_physical_page_entries = (PhysicalPageEntry*)range.base().get();
  434. for (size_t i = 0; i < m_physical_page_entries_count; i++)
  435. new (&m_physical_page_entries[i]) PageTableEntry();
  436. // Now we should be able to allocate PhysicalPage instances,
  437. // so finish setting up the kernel page directory
  438. m_kernel_page_directory->allocate_kernel_directory();
  439. // Now create legit PhysicalPage objects for the page tables we created.
  440. virtual_page_array_current_page = virtual_page_array_base;
  441. for (size_t pt_index = 0; pt_index < needed_page_table_count; pt_index++) {
  442. VERIFY(virtual_page_array_current_page <= range.end().get());
  443. auto pt_paddr = page_tables_base.offset(pt_index * PAGE_SIZE);
  444. auto physical_page_index = PhysicalAddress::physical_page_index(pt_paddr.get());
  445. auto& physical_page_entry = m_physical_page_entries[physical_page_index];
  446. auto physical_page = adopt_lock_ref(*new (&physical_page_entry.allocated.physical_page) PhysicalPage(MayReturnToFreeList::No));
  447. // NOTE: This leaked ref is matched by the unref in MemoryManager::release_pte()
  448. (void)physical_page.leak_ref();
  449. virtual_page_array_current_page += (PAGE_SIZE / sizeof(PageTableEntry)) * PAGE_SIZE;
  450. }
  451. dmesgln("MM: Physical page entries: {}", range);
  452. });
  453. }
  454. PhysicalPageEntry& MemoryManager::get_physical_page_entry(PhysicalAddress physical_address)
  455. {
  456. auto physical_page_entry_index = PhysicalAddress::physical_page_index(physical_address.get());
  457. VERIFY(physical_page_entry_index < m_physical_page_entries_count);
  458. return m_physical_page_entries[physical_page_entry_index];
  459. }
  460. PhysicalAddress MemoryManager::get_physical_address(PhysicalPage const& physical_page)
  461. {
  462. PhysicalPageEntry const& physical_page_entry = *reinterpret_cast<PhysicalPageEntry const*>((u8 const*)&physical_page - __builtin_offsetof(PhysicalPageEntry, allocated.physical_page));
  463. size_t physical_page_entry_index = &physical_page_entry - m_physical_page_entries;
  464. VERIFY(physical_page_entry_index < m_physical_page_entries_count);
  465. return PhysicalAddress((PhysicalPtr)physical_page_entry_index * PAGE_SIZE);
  466. }
  467. PageTableEntry* MemoryManager::pte(PageDirectory& page_directory, VirtualAddress vaddr)
  468. {
  469. VERIFY_INTERRUPTS_DISABLED();
  470. VERIFY(page_directory.get_lock().is_locked_by_current_processor());
  471. u32 page_directory_table_index = (vaddr.get() >> 30) & 0x1ff;
  472. u32 page_directory_index = (vaddr.get() >> 21) & 0x1ff;
  473. u32 page_table_index = (vaddr.get() >> 12) & 0x1ff;
  474. auto* pd = quickmap_pd(const_cast<PageDirectory&>(page_directory), page_directory_table_index);
  475. PageDirectoryEntry const& pde = pd[page_directory_index];
  476. if (!pde.is_present())
  477. return nullptr;
  478. return &quickmap_pt(PhysicalAddress((FlatPtr)pde.page_table_base()))[page_table_index];
  479. }
  480. PageTableEntry* MemoryManager::ensure_pte(PageDirectory& page_directory, VirtualAddress vaddr)
  481. {
  482. VERIFY_INTERRUPTS_DISABLED();
  483. VERIFY(page_directory.get_lock().is_locked_by_current_processor());
  484. u32 page_directory_table_index = (vaddr.get() >> 30) & 0x1ff;
  485. u32 page_directory_index = (vaddr.get() >> 21) & 0x1ff;
  486. u32 page_table_index = (vaddr.get() >> 12) & 0x1ff;
  487. auto* pd = quickmap_pd(page_directory, page_directory_table_index);
  488. auto& pde = pd[page_directory_index];
  489. if (pde.is_present())
  490. return &quickmap_pt(PhysicalAddress(pde.page_table_base()))[page_table_index];
  491. bool did_purge = false;
  492. auto page_table_or_error = allocate_physical_page(ShouldZeroFill::Yes, &did_purge);
  493. if (page_table_or_error.is_error()) {
  494. dbgln("MM: Unable to allocate page table to map {}", vaddr);
  495. return nullptr;
  496. }
  497. auto page_table = page_table_or_error.release_value();
  498. if (did_purge) {
  499. // If any memory had to be purged, ensure_pte may have been called as part
  500. // of the purging process. So we need to re-map the pd in this case to ensure
  501. // we're writing to the correct underlying physical page
  502. pd = quickmap_pd(page_directory, page_directory_table_index);
  503. VERIFY(&pde == &pd[page_directory_index]); // Sanity check
  504. VERIFY(!pde.is_present()); // Should have not changed
  505. }
  506. pde.set_page_table_base(page_table->paddr().get());
  507. pde.set_user_allowed(true);
  508. pde.set_present(true);
  509. pde.set_writable(true);
  510. pde.set_global(&page_directory == m_kernel_page_directory.ptr());
  511. // NOTE: This leaked ref is matched by the unref in MemoryManager::release_pte()
  512. (void)page_table.leak_ref();
  513. return &quickmap_pt(PhysicalAddress(pde.page_table_base()))[page_table_index];
  514. }
  515. void MemoryManager::release_pte(PageDirectory& page_directory, VirtualAddress vaddr, IsLastPTERelease is_last_pte_release)
  516. {
  517. VERIFY_INTERRUPTS_DISABLED();
  518. VERIFY(page_directory.get_lock().is_locked_by_current_processor());
  519. u32 page_directory_table_index = (vaddr.get() >> 30) & 0x1ff;
  520. u32 page_directory_index = (vaddr.get() >> 21) & 0x1ff;
  521. u32 page_table_index = (vaddr.get() >> 12) & 0x1ff;
  522. auto* pd = quickmap_pd(page_directory, page_directory_table_index);
  523. PageDirectoryEntry& pde = pd[page_directory_index];
  524. if (pde.is_present()) {
  525. auto* page_table = quickmap_pt(PhysicalAddress((FlatPtr)pde.page_table_base()));
  526. auto& pte = page_table[page_table_index];
  527. pte.clear();
  528. if (is_last_pte_release == IsLastPTERelease::Yes || page_table_index == 0x1ff) {
  529. // If this is the last PTE in a region or the last PTE in a page table then
  530. // check if we can also release the page table
  531. bool all_clear = true;
  532. for (u32 i = 0; i <= 0x1ff; i++) {
  533. if (!page_table[i].is_null()) {
  534. all_clear = false;
  535. break;
  536. }
  537. }
  538. if (all_clear) {
  539. get_physical_page_entry(PhysicalAddress { pde.page_table_base() }).allocated.physical_page.unref();
  540. pde.clear();
  541. }
  542. }
  543. }
  544. }
  545. UNMAP_AFTER_INIT void MemoryManager::initialize(u32 cpu)
  546. {
  547. ProcessorSpecific<MemoryManagerData>::initialize();
  548. if (cpu == 0) {
  549. new MemoryManager;
  550. kmalloc_enable_expand();
  551. }
  552. }
  553. Region* MemoryManager::kernel_region_from_vaddr(VirtualAddress address)
  554. {
  555. if (is_user_address(address))
  556. return nullptr;
  557. return MM.m_global_data.with([&](auto& global_data) {
  558. return global_data.region_tree.find_region_containing(address);
  559. });
  560. }
  561. Region* MemoryManager::find_user_region_from_vaddr(AddressSpace& space, VirtualAddress vaddr)
  562. {
  563. return space.find_region_containing({ vaddr, 1 });
  564. }
  565. void MemoryManager::validate_syscall_preconditions(Process& process, RegisterState const& regs)
  566. {
  567. bool should_crash = false;
  568. char const* crash_description = nullptr;
  569. int crash_signal = 0;
  570. auto unlock_and_handle_crash = [&](char const* description, int signal) {
  571. should_crash = true;
  572. crash_description = description;
  573. crash_signal = signal;
  574. };
  575. process.address_space().with([&](auto& space) -> void {
  576. VirtualAddress userspace_sp = VirtualAddress { regs.userspace_sp() };
  577. if (!MM.validate_user_stack(*space, userspace_sp)) {
  578. dbgln("Invalid stack pointer: {}", userspace_sp);
  579. return unlock_and_handle_crash("Bad stack on syscall entry", SIGSEGV);
  580. }
  581. VirtualAddress ip = VirtualAddress { regs.ip() };
  582. auto* calling_region = MM.find_user_region_from_vaddr(*space, ip);
  583. if (!calling_region) {
  584. dbgln("Syscall from {:p} which has no associated region", ip);
  585. return unlock_and_handle_crash("Syscall from unknown region", SIGSEGV);
  586. }
  587. if (calling_region->is_writable()) {
  588. dbgln("Syscall from writable memory at {:p}", ip);
  589. return unlock_and_handle_crash("Syscall from writable memory", SIGSEGV);
  590. }
  591. if (space->enforces_syscall_regions() && !calling_region->is_syscall_region()) {
  592. dbgln("Syscall from non-syscall region");
  593. return unlock_and_handle_crash("Syscall from non-syscall region", SIGSEGV);
  594. }
  595. });
  596. if (should_crash) {
  597. handle_crash(regs, crash_description, crash_signal);
  598. }
  599. }
  600. Region* MemoryManager::find_region_from_vaddr(VirtualAddress vaddr)
  601. {
  602. if (auto* region = kernel_region_from_vaddr(vaddr))
  603. return region;
  604. auto page_directory = PageDirectory::find_current();
  605. if (!page_directory)
  606. return nullptr;
  607. VERIFY(page_directory->address_space());
  608. return find_user_region_from_vaddr(*page_directory->address_space(), vaddr);
  609. }
  610. PageFaultResponse MemoryManager::handle_page_fault(PageFault const& fault)
  611. {
  612. auto faulted_in_range = [&fault](auto const* start, auto const* end) {
  613. return fault.vaddr() >= VirtualAddress { start } && fault.vaddr() < VirtualAddress { end };
  614. };
  615. if (faulted_in_range(&start_of_ro_after_init, &end_of_ro_after_init))
  616. PANIC("Attempt to write into READONLY_AFTER_INIT section");
  617. if (faulted_in_range(&start_of_unmap_after_init, &end_of_unmap_after_init)) {
  618. auto const* kernel_symbol = symbolicate_kernel_address(fault.vaddr().get());
  619. PANIC("Attempt to access UNMAP_AFTER_INIT section ({:p}: {})", fault.vaddr(), kernel_symbol ? kernel_symbol->name : "(Unknown)");
  620. }
  621. if (faulted_in_range(&start_of_kernel_ksyms, &end_of_kernel_ksyms))
  622. PANIC("Attempt to access KSYMS section");
  623. if (Processor::current_in_irq()) {
  624. dbgln("CPU[{}] BUG! Page fault while handling IRQ! code={}, vaddr={}, irq level: {}",
  625. Processor::current_id(), fault.code(), fault.vaddr(), Processor::current_in_irq());
  626. dump_kernel_regions();
  627. return PageFaultResponse::ShouldCrash;
  628. }
  629. dbgln_if(PAGE_FAULT_DEBUG, "MM: CPU[{}] handle_page_fault({:#04x}) at {}", Processor::current_id(), fault.code(), fault.vaddr());
  630. auto* region = find_region_from_vaddr(fault.vaddr());
  631. if (!region) {
  632. return PageFaultResponse::ShouldCrash;
  633. }
  634. return region->handle_fault(fault);
  635. }
  636. ErrorOr<NonnullOwnPtr<Region>> MemoryManager::allocate_contiguous_kernel_region(size_t size, StringView name, Region::Access access, Region::Cacheable cacheable)
  637. {
  638. VERIFY(!(size % PAGE_SIZE));
  639. OwnPtr<KString> name_kstring;
  640. if (!name.is_null())
  641. name_kstring = TRY(KString::try_create(name));
  642. auto vmobject = TRY(AnonymousVMObject::try_create_physically_contiguous_with_size(size));
  643. auto region = TRY(Region::create_unplaced(move(vmobject), 0, move(name_kstring), access, cacheable));
  644. TRY(m_global_data.with([&](auto& global_data) { return global_data.region_tree.place_anywhere(*region, RandomizeVirtualAddress::No, size); }));
  645. TRY(region->map(kernel_page_directory()));
  646. return region;
  647. }
  648. ErrorOr<NonnullOwnPtr<Memory::Region>> MemoryManager::allocate_dma_buffer_page(StringView name, Memory::Region::Access access, RefPtr<Memory::PhysicalPage>& dma_buffer_page)
  649. {
  650. dma_buffer_page = TRY(allocate_physical_page());
  651. // Do not enable Cache for this region as physical memory transfers are performed (Most architectures have this behaviour by default)
  652. return allocate_kernel_region(dma_buffer_page->paddr(), PAGE_SIZE, name, access, Region::Cacheable::No);
  653. }
  654. ErrorOr<NonnullOwnPtr<Memory::Region>> MemoryManager::allocate_dma_buffer_page(StringView name, Memory::Region::Access access)
  655. {
  656. RefPtr<Memory::PhysicalPage> dma_buffer_page;
  657. return allocate_dma_buffer_page(name, access, dma_buffer_page);
  658. }
  659. ErrorOr<NonnullOwnPtr<Memory::Region>> MemoryManager::allocate_dma_buffer_pages(size_t size, StringView name, Memory::Region::Access access, NonnullRefPtrVector<Memory::PhysicalPage>& dma_buffer_pages)
  660. {
  661. VERIFY(!(size % PAGE_SIZE));
  662. dma_buffer_pages = TRY(allocate_contiguous_physical_pages(size));
  663. // Do not enable Cache for this region as physical memory transfers are performed (Most architectures have this behaviour by default)
  664. return allocate_kernel_region(dma_buffer_pages.first().paddr(), size, name, access, Region::Cacheable::No);
  665. }
  666. ErrorOr<NonnullOwnPtr<Memory::Region>> MemoryManager::allocate_dma_buffer_pages(size_t size, StringView name, Memory::Region::Access access)
  667. {
  668. VERIFY(!(size % PAGE_SIZE));
  669. NonnullRefPtrVector<Memory::PhysicalPage> dma_buffer_pages;
  670. return allocate_dma_buffer_pages(size, name, access, dma_buffer_pages);
  671. }
  672. ErrorOr<NonnullOwnPtr<Region>> MemoryManager::allocate_kernel_region(size_t size, StringView name, Region::Access access, AllocationStrategy strategy, Region::Cacheable cacheable)
  673. {
  674. VERIFY(!(size % PAGE_SIZE));
  675. OwnPtr<KString> name_kstring;
  676. if (!name.is_null())
  677. name_kstring = TRY(KString::try_create(name));
  678. auto vmobject = TRY(AnonymousVMObject::try_create_with_size(size, strategy));
  679. auto region = TRY(Region::create_unplaced(move(vmobject), 0, move(name_kstring), access, cacheable));
  680. TRY(m_global_data.with([&](auto& global_data) { return global_data.region_tree.place_anywhere(*region, RandomizeVirtualAddress::No, size); }));
  681. TRY(region->map(kernel_page_directory()));
  682. return region;
  683. }
  684. ErrorOr<NonnullOwnPtr<Region>> MemoryManager::allocate_kernel_region(PhysicalAddress paddr, size_t size, StringView name, Region::Access access, Region::Cacheable cacheable)
  685. {
  686. VERIFY(!(size % PAGE_SIZE));
  687. auto vmobject = TRY(AnonymousVMObject::try_create_for_physical_range(paddr, size));
  688. OwnPtr<KString> name_kstring;
  689. if (!name.is_null())
  690. name_kstring = TRY(KString::try_create(name));
  691. auto region = TRY(Region::create_unplaced(move(vmobject), 0, move(name_kstring), access, cacheable));
  692. TRY(m_global_data.with([&](auto& global_data) { return global_data.region_tree.place_anywhere(*region, RandomizeVirtualAddress::No, size, PAGE_SIZE); }));
  693. TRY(region->map(kernel_page_directory()));
  694. return region;
  695. }
  696. ErrorOr<NonnullOwnPtr<Region>> MemoryManager::allocate_kernel_region_with_vmobject(VMObject& vmobject, size_t size, StringView name, Region::Access access, Region::Cacheable cacheable)
  697. {
  698. VERIFY(!(size % PAGE_SIZE));
  699. OwnPtr<KString> name_kstring;
  700. if (!name.is_null())
  701. name_kstring = TRY(KString::try_create(name));
  702. auto region = TRY(Region::create_unplaced(vmobject, 0, move(name_kstring), access, cacheable));
  703. TRY(m_global_data.with([&](auto& global_data) { return global_data.region_tree.place_anywhere(*region, RandomizeVirtualAddress::No, size); }));
  704. TRY(region->map(kernel_page_directory()));
  705. return region;
  706. }
  707. ErrorOr<CommittedPhysicalPageSet> MemoryManager::commit_physical_pages(size_t page_count)
  708. {
  709. VERIFY(page_count > 0);
  710. auto result = m_global_data.with([&](auto& global_data) -> ErrorOr<CommittedPhysicalPageSet> {
  711. if (global_data.system_memory_info.physical_pages_uncommitted < page_count) {
  712. dbgln("MM: Unable to commit {} pages, have only {}", page_count, global_data.system_memory_info.physical_pages_uncommitted);
  713. return ENOMEM;
  714. }
  715. global_data.system_memory_info.physical_pages_uncommitted -= page_count;
  716. global_data.system_memory_info.physical_pages_committed += page_count;
  717. return CommittedPhysicalPageSet { {}, page_count };
  718. });
  719. if (result.is_error()) {
  720. Process::for_each([&](Process const& process) {
  721. size_t amount_resident = 0;
  722. size_t amount_shared = 0;
  723. size_t amount_virtual = 0;
  724. process.address_space().with([&](auto& space) {
  725. amount_resident = space->amount_resident();
  726. amount_shared = space->amount_shared();
  727. amount_virtual = space->amount_virtual();
  728. });
  729. dbgln("{}({}) resident:{}, shared:{}, virtual:{}",
  730. process.name(),
  731. process.pid(),
  732. amount_resident / PAGE_SIZE,
  733. amount_shared / PAGE_SIZE,
  734. amount_virtual / PAGE_SIZE);
  735. return IterationDecision::Continue;
  736. });
  737. }
  738. return result;
  739. }
  740. void MemoryManager::uncommit_physical_pages(Badge<CommittedPhysicalPageSet>, size_t page_count)
  741. {
  742. VERIFY(page_count > 0);
  743. m_global_data.with([&](auto& global_data) {
  744. VERIFY(global_data.system_memory_info.physical_pages_committed >= page_count);
  745. global_data.system_memory_info.physical_pages_uncommitted += page_count;
  746. global_data.system_memory_info.physical_pages_committed -= page_count;
  747. });
  748. }
  749. void MemoryManager::deallocate_physical_page(PhysicalAddress paddr)
  750. {
  751. return m_global_data.with([&](auto& global_data) {
  752. // Are we returning a user page?
  753. for (auto& region : global_data.physical_regions) {
  754. if (!region.contains(paddr))
  755. continue;
  756. region.return_page(paddr);
  757. --global_data.system_memory_info.physical_pages_used;
  758. // Always return pages to the uncommitted pool. Pages that were
  759. // committed and allocated are only freed upon request. Once
  760. // returned there is no guarantee being able to get them back.
  761. ++global_data.system_memory_info.physical_pages_uncommitted;
  762. return;
  763. }
  764. PANIC("MM: deallocate_physical_page couldn't figure out region for page @ {}", paddr);
  765. });
  766. }
  767. RefPtr<PhysicalPage> MemoryManager::find_free_physical_page(bool committed)
  768. {
  769. RefPtr<PhysicalPage> page;
  770. m_global_data.with([&](auto& global_data) {
  771. if (committed) {
  772. // Draw from the committed pages pool. We should always have these pages available
  773. VERIFY(global_data.system_memory_info.physical_pages_committed > 0);
  774. global_data.system_memory_info.physical_pages_committed--;
  775. } else {
  776. // We need to make sure we don't touch pages that we have committed to
  777. if (global_data.system_memory_info.physical_pages_uncommitted == 0)
  778. return;
  779. global_data.system_memory_info.physical_pages_uncommitted--;
  780. }
  781. for (auto& region : global_data.physical_regions) {
  782. page = region.take_free_page();
  783. if (!page.is_null()) {
  784. ++global_data.system_memory_info.physical_pages_used;
  785. break;
  786. }
  787. }
  788. });
  789. VERIFY(!page.is_null());
  790. return page;
  791. }
  792. NonnullRefPtr<PhysicalPage> MemoryManager::allocate_committed_physical_page(Badge<CommittedPhysicalPageSet>, ShouldZeroFill should_zero_fill)
  793. {
  794. auto page = find_free_physical_page(true);
  795. if (should_zero_fill == ShouldZeroFill::Yes) {
  796. InterruptDisabler disabler;
  797. auto* ptr = quickmap_page(*page);
  798. memset(ptr, 0, PAGE_SIZE);
  799. unquickmap_page();
  800. }
  801. return page.release_nonnull();
  802. }
  803. ErrorOr<NonnullRefPtr<PhysicalPage>> MemoryManager::allocate_physical_page(ShouldZeroFill should_zero_fill, bool* did_purge)
  804. {
  805. return m_global_data.with([&](auto&) -> ErrorOr<NonnullRefPtr<PhysicalPage>> {
  806. auto page = find_free_physical_page(false);
  807. bool purged_pages = false;
  808. if (!page) {
  809. // We didn't have a single free physical page. Let's try to free something up!
  810. // First, we look for a purgeable VMObject in the volatile state.
  811. for_each_vmobject([&](auto& vmobject) {
  812. if (!vmobject.is_anonymous())
  813. return IterationDecision::Continue;
  814. auto& anonymous_vmobject = static_cast<AnonymousVMObject&>(vmobject);
  815. if (!anonymous_vmobject.is_purgeable() || !anonymous_vmobject.is_volatile())
  816. return IterationDecision::Continue;
  817. if (auto purged_page_count = anonymous_vmobject.purge()) {
  818. dbgln("MM: Purge saved the day! Purged {} pages from AnonymousVMObject", purged_page_count);
  819. page = find_free_physical_page(false);
  820. purged_pages = true;
  821. VERIFY(page);
  822. return IterationDecision::Break;
  823. }
  824. return IterationDecision::Continue;
  825. });
  826. }
  827. if (!page) {
  828. // Second, we look for a file-backed VMObject with clean pages.
  829. for_each_vmobject([&](auto& vmobject) {
  830. if (!vmobject.is_inode())
  831. return IterationDecision::Continue;
  832. auto& inode_vmobject = static_cast<InodeVMObject&>(vmobject);
  833. if (auto released_page_count = inode_vmobject.try_release_clean_pages(1)) {
  834. dbgln("MM: Clean inode release saved the day! Released {} pages from InodeVMObject", released_page_count);
  835. page = find_free_physical_page(false);
  836. VERIFY(page);
  837. return IterationDecision::Break;
  838. }
  839. return IterationDecision::Continue;
  840. });
  841. }
  842. if (!page) {
  843. dmesgln("MM: no physical pages available");
  844. return ENOMEM;
  845. }
  846. if (should_zero_fill == ShouldZeroFill::Yes) {
  847. auto* ptr = quickmap_page(*page);
  848. memset(ptr, 0, PAGE_SIZE);
  849. unquickmap_page();
  850. }
  851. if (did_purge)
  852. *did_purge = purged_pages;
  853. return page.release_nonnull();
  854. });
  855. }
  856. ErrorOr<NonnullRefPtrVector<PhysicalPage>> MemoryManager::allocate_contiguous_physical_pages(size_t size)
  857. {
  858. VERIFY(!(size % PAGE_SIZE));
  859. size_t page_count = ceil_div(size, static_cast<size_t>(PAGE_SIZE));
  860. auto physical_pages = TRY(m_global_data.with([&](auto& global_data) -> ErrorOr<NonnullRefPtrVector<PhysicalPage>> {
  861. // We need to make sure we don't touch pages that we have committed to
  862. if (global_data.system_memory_info.physical_pages_uncommitted < page_count)
  863. return ENOMEM;
  864. for (auto& physical_region : global_data.physical_regions) {
  865. auto physical_pages = physical_region.take_contiguous_free_pages(page_count);
  866. if (!physical_pages.is_empty()) {
  867. global_data.system_memory_info.physical_pages_uncommitted -= page_count;
  868. global_data.system_memory_info.physical_pages_used += page_count;
  869. return physical_pages;
  870. }
  871. }
  872. dmesgln("MM: no contiguous physical pages available");
  873. return ENOMEM;
  874. }));
  875. {
  876. auto cleanup_region = TRY(MM.allocate_kernel_region(physical_pages[0].paddr(), PAGE_SIZE * page_count, {}, Region::Access::Read | Region::Access::Write));
  877. memset(cleanup_region->vaddr().as_ptr(), 0, PAGE_SIZE * page_count);
  878. }
  879. return physical_pages;
  880. }
  881. void MemoryManager::enter_process_address_space(Process& process)
  882. {
  883. process.address_space().with([](auto& space) {
  884. enter_address_space(*space);
  885. });
  886. }
  887. void MemoryManager::enter_address_space(AddressSpace& space)
  888. {
  889. auto* current_thread = Thread::current();
  890. VERIFY(current_thread != nullptr);
  891. activate_page_directory(space.page_directory(), current_thread);
  892. }
  893. void MemoryManager::flush_tlb_local(VirtualAddress vaddr, size_t page_count)
  894. {
  895. Processor::flush_tlb_local(vaddr, page_count);
  896. }
  897. void MemoryManager::flush_tlb(PageDirectory const* page_directory, VirtualAddress vaddr, size_t page_count)
  898. {
  899. Processor::flush_tlb(page_directory, vaddr, page_count);
  900. }
  901. PageDirectoryEntry* MemoryManager::quickmap_pd(PageDirectory& directory, size_t pdpt_index)
  902. {
  903. VERIFY_INTERRUPTS_DISABLED();
  904. VirtualAddress vaddr(KERNEL_QUICKMAP_PD_PER_CPU_BASE + Processor::current_id() * PAGE_SIZE);
  905. size_t pte_index = (vaddr.get() - KERNEL_PT1024_BASE) / PAGE_SIZE;
  906. auto& pte = boot_pd_kernel_pt1023[pte_index];
  907. auto pd_paddr = directory.m_directory_pages[pdpt_index]->paddr();
  908. if (pte.physical_page_base() != pd_paddr.get()) {
  909. pte.set_physical_page_base(pd_paddr.get());
  910. pte.set_present(true);
  911. pte.set_writable(true);
  912. pte.set_user_allowed(false);
  913. flush_tlb_local(vaddr);
  914. }
  915. return (PageDirectoryEntry*)vaddr.get();
  916. }
  917. PageTableEntry* MemoryManager::quickmap_pt(PhysicalAddress pt_paddr)
  918. {
  919. VERIFY_INTERRUPTS_DISABLED();
  920. VirtualAddress vaddr(KERNEL_QUICKMAP_PT_PER_CPU_BASE + Processor::current_id() * PAGE_SIZE);
  921. size_t pte_index = (vaddr.get() - KERNEL_PT1024_BASE) / PAGE_SIZE;
  922. auto& pte = ((PageTableEntry*)boot_pd_kernel_pt1023)[pte_index];
  923. if (pte.physical_page_base() != pt_paddr.get()) {
  924. pte.set_physical_page_base(pt_paddr.get());
  925. pte.set_present(true);
  926. pte.set_writable(true);
  927. pte.set_user_allowed(false);
  928. flush_tlb_local(vaddr);
  929. }
  930. return (PageTableEntry*)vaddr.get();
  931. }
  932. u8* MemoryManager::quickmap_page(PhysicalAddress const& physical_address)
  933. {
  934. VERIFY_INTERRUPTS_DISABLED();
  935. auto& mm_data = get_data();
  936. mm_data.m_quickmap_previous_interrupts_state = mm_data.m_quickmap_in_use.lock();
  937. VirtualAddress vaddr(KERNEL_QUICKMAP_PER_CPU_BASE + Processor::current_id() * PAGE_SIZE);
  938. u32 pte_idx = (vaddr.get() - KERNEL_PT1024_BASE) / PAGE_SIZE;
  939. auto& pte = ((PageTableEntry*)boot_pd_kernel_pt1023)[pte_idx];
  940. if (pte.physical_page_base() != physical_address.get()) {
  941. pte.set_physical_page_base(physical_address.get());
  942. pte.set_present(true);
  943. pte.set_writable(true);
  944. pte.set_user_allowed(false);
  945. flush_tlb_local(vaddr);
  946. }
  947. return vaddr.as_ptr();
  948. }
  949. void MemoryManager::unquickmap_page()
  950. {
  951. VERIFY_INTERRUPTS_DISABLED();
  952. auto& mm_data = get_data();
  953. VERIFY(mm_data.m_quickmap_in_use.is_locked());
  954. VirtualAddress vaddr(KERNEL_QUICKMAP_PER_CPU_BASE + Processor::current_id() * PAGE_SIZE);
  955. u32 pte_idx = (vaddr.get() - KERNEL_PT1024_BASE) / PAGE_SIZE;
  956. auto& pte = ((PageTableEntry*)boot_pd_kernel_pt1023)[pte_idx];
  957. pte.clear();
  958. flush_tlb_local(vaddr);
  959. mm_data.m_quickmap_in_use.unlock(mm_data.m_quickmap_previous_interrupts_state);
  960. }
  961. bool MemoryManager::validate_user_stack(AddressSpace& space, VirtualAddress vaddr) const
  962. {
  963. if (!is_user_address(vaddr))
  964. return false;
  965. auto* region = find_user_region_from_vaddr(space, vaddr);
  966. return region && region->is_user() && region->is_stack();
  967. }
  968. void MemoryManager::unregister_kernel_region(Region& region)
  969. {
  970. VERIFY(region.is_kernel());
  971. m_global_data.with([&](auto& global_data) { global_data.region_tree.remove(region); });
  972. }
  973. void MemoryManager::dump_kernel_regions()
  974. {
  975. dbgln("Kernel regions:");
  976. #if ARCH(I386)
  977. char const* addr_padding = "";
  978. #else
  979. char const* addr_padding = " ";
  980. #endif
  981. dbgln("BEGIN{} END{} SIZE{} ACCESS NAME",
  982. addr_padding, addr_padding, addr_padding);
  983. m_global_data.with([&](auto& global_data) {
  984. for (auto& region : global_data.region_tree.regions()) {
  985. dbgln("{:p} -- {:p} {:p} {:c}{:c}{:c}{:c}{:c}{:c} {}",
  986. region.vaddr().get(),
  987. region.vaddr().offset(region.size() - 1).get(),
  988. region.size(),
  989. region.is_readable() ? 'R' : ' ',
  990. region.is_writable() ? 'W' : ' ',
  991. region.is_executable() ? 'X' : ' ',
  992. region.is_shared() ? 'S' : ' ',
  993. region.is_stack() ? 'T' : ' ',
  994. region.is_syscall_region() ? 'C' : ' ',
  995. region.name());
  996. }
  997. });
  998. }
  999. void MemoryManager::set_page_writable_direct(VirtualAddress vaddr, bool writable)
  1000. {
  1001. SpinlockLocker page_lock(kernel_page_directory().get_lock());
  1002. auto* pte = ensure_pte(kernel_page_directory(), vaddr);
  1003. VERIFY(pte);
  1004. if (pte->is_writable() == writable)
  1005. return;
  1006. pte->set_writable(writable);
  1007. flush_tlb(&kernel_page_directory(), vaddr);
  1008. }
  1009. CommittedPhysicalPageSet::~CommittedPhysicalPageSet()
  1010. {
  1011. if (m_page_count)
  1012. MM.uncommit_physical_pages({}, m_page_count);
  1013. }
  1014. NonnullRefPtr<PhysicalPage> CommittedPhysicalPageSet::take_one()
  1015. {
  1016. VERIFY(m_page_count > 0);
  1017. --m_page_count;
  1018. return MM.allocate_committed_physical_page({}, MemoryManager::ShouldZeroFill::Yes);
  1019. }
  1020. void CommittedPhysicalPageSet::uncommit_one()
  1021. {
  1022. VERIFY(m_page_count > 0);
  1023. --m_page_count;
  1024. MM.uncommit_physical_pages({}, 1);
  1025. }
  1026. void MemoryManager::copy_physical_page(PhysicalPage& physical_page, u8 page_buffer[PAGE_SIZE])
  1027. {
  1028. auto* quickmapped_page = quickmap_page(physical_page);
  1029. memcpy(page_buffer, quickmapped_page, PAGE_SIZE);
  1030. unquickmap_page();
  1031. }
  1032. ErrorOr<NonnullOwnPtr<Memory::Region>> MemoryManager::create_identity_mapped_region(PhysicalAddress address, size_t size)
  1033. {
  1034. auto vmobject = TRY(Memory::AnonymousVMObject::try_create_for_physical_range(address, size));
  1035. auto region = TRY(Memory::Region::create_unplaced(move(vmobject), 0, {}, Memory::Region::Access::ReadWriteExecute));
  1036. Memory::VirtualRange range { VirtualAddress { (FlatPtr)address.get() }, size };
  1037. region->m_range = range;
  1038. TRY(region->map(MM.kernel_page_directory()));
  1039. return region;
  1040. }
  1041. ErrorOr<NonnullOwnPtr<Region>> MemoryManager::allocate_unbacked_region_anywhere(size_t size, size_t alignment)
  1042. {
  1043. auto region = TRY(Region::create_unbacked());
  1044. TRY(m_global_data.with([&](auto& global_data) { return global_data.region_tree.place_anywhere(*region, RandomizeVirtualAddress::No, size, alignment); }));
  1045. return region;
  1046. }
  1047. MemoryManager::SystemMemoryInfo MemoryManager::get_system_memory_info()
  1048. {
  1049. return m_global_data.with([&](auto& global_data) {
  1050. auto physical_pages_unused = global_data.system_memory_info.physical_pages_committed + global_data.system_memory_info.physical_pages_uncommitted;
  1051. VERIFY(global_data.system_memory_info.physical_pages == (global_data.system_memory_info.physical_pages_used + physical_pages_unused));
  1052. return global_data.system_memory_info;
  1053. });
  1054. }
  1055. }