1 @node Project 3--Virtual Memory, Project 4--File Systems, Project 2--User Programs, Top
2 @chapter Project 3: Virtual Memory
4 By now you should be familiar with the inner workings of Pintos.
5 You've already come a long way: your OS can properly handle multiple
6 threads of execution with proper synchronization, and can load
7 multiple user programs at once. However, when loading user programs,
8 your OS is limited by how much main memory the simulated machine has.
9 In this assignment, you will remove that limitation.
11 You will be using the @file{vm} directory for this project. The
12 @file{vm} directory contains only the @file{Makefile}s. The only
13 change from @file{userprog} is that this new @file{Makefile} turns on
14 the setting @option{-DVM}. All code you write will either be newly
15 generated files (e.g.@: if you choose to implement your paging code in
16 their own source files), or will be modifications to pre-existing code
17 (e.g.@: you will change the behavior of @file{process.c}
20 There are only a couple of source files you will probably be
21 encountering for the first time:
26 Provides access to the physical disk, abstracting away the rather
30 You will be building this assignment on the last one. It will benefit
31 you to get your project 2 in good working order before this assignment
32 so those bugs don't keep haunting you.
34 All the test programs from the previous project should also work with
35 this project. You should also write programs to test the new features
36 introduced in this project.
38 Your submission should define @code{THREAD_JOIN_IMPLEMENTED} in
39 @file{constants.h} (@pxref{Conditional Compilation}).
44 * Disk as Backing Store::
45 * Memory Mapped Files::
47 * Problem 3-1 Page Table Management::
48 * Problem 3-2 Paging To and From Disk::
49 * Problem 3-3 Memory Mapped Files::
50 * Virtual Memory FAQ::
54 @section A Word about Design
56 It is important for you to note that in addition to getting virtual
57 memory working, this assignment is also meant to be an open-ended
58 design problem. We will expect you to come up with a design that
59 makes sense. You will have the freedom to choose how to handle page
60 faults, how to organize the swap disk, how to implement paging, etc.
61 In each case, we will expect you to provide a defensible justification
62 in your design documentation as to why your choices are reasonable.
63 You should evaluate your design on all the available criteria: speed
64 of handling a page fault, space overhead in memory, minimizing the
65 number of page faults, simplicity, etc.
67 In keeping with this, you will find that we are going to say as little
68 as possible about how to do things. Instead we will focus on what end
69 functionality we require your OS to support.
74 For the last assignment, whenever a context switch occurred, the new
75 process would install its own page table into the machine. The page
76 table contained all the virtual-to-physical translations for the
77 process. Whenever the processor needed to look up a translation, it
78 consulted the page table. As long as the process only accessed
79 memory that it didn't own, all was well. If the process accessed
80 memory it didn't own, it ``page faulted'' and @func{page_fault}
81 terminated the process.
83 When we implement virtual memory, the rules have to change. A page
84 fault is no longer necessarily an error, since it might only indicate
85 that the page must be brought in from a disk file or from swap. You
86 will have to implement a more sophisticated page fault handler to
89 On the 80@var{x}86, the page table format is fixed by hardware. We
90 have provided code for managing page tables for you to use in
91 @file{userprog/pagedir.c}. The functions in there should provide an
92 abstract interface to all the page table functionality that you need
93 to complete the project. However, you may still find it worthwhile to
94 understand a little about the hardware page table format, so we'll go
95 into a little of detail about that in this section.
97 The top-level paging data structure is a 4 kB page called the ``page
98 directory'' (PD) arranged as an array of 1,024 32-bit page directory
99 entries (PDEs), each of which represents 4 MB of virtual memory. Each
100 PDE may point to the physical address of another 4 kB page called a
101 ``page table'' (PT) arranged in the same fashion as an array of 1,024
102 32-bit page table entries (PTEs), each of which translates a single 4
103 kB virtual page into physical memory.
105 Thus, translation of a virtual address into a physical address follows
106 the three-step process illustrated in the diagram
107 below:@footnote{Actually, virtual to physical translation on the
108 80@var{x}86 architecture happens via an intermediate ``linear
109 address,'' but Pintos (and most other 80@var{x}86 OSes) set up the CPU
110 so that linear and virtual addresses are one and the same, so that you
111 can effectively ignore this CPU feature.}
115 The top 10 bits of the virtual address (bits 22:32) are used to index
116 into the page directory. If the PDE is marked ``present,'' the
117 physical address of a page table is read from the PDE thus obtained.
118 If the PDE is marked ``not present'' then a page fault occurs.
121 The next 10 bits of the virtual address (bits 12:22) are used to index
122 into the page table. If the PTE is marked ``present,'' the physical
123 address of a data page is read from the PTE thus obtained. If the PTE
124 is marked ``not present'' then a page fault occurs.
128 The bottom 12 bits of the virtual address (bits 0:12) are added to the
129 data page's physical base address, producing the final physical
136 +--------------------------------------------------------------------+
137 | Page Directory Index | Page Table Index | Page Offset |
138 +--------------------------------------------------------------------+
140 _______/ _______/ _____/
142 / Page Directory / Page Table / Data Page
143 / .____________. / .____________. / .____________.
144 |1,023|____________| |1,023|____________| | |____________|
145 |1,022|____________| |1,022|____________| | |____________|
146 |1,021|____________| |1,021|____________| \__\|____________|
147 |1,020|____________| |1,020|____________| /|____________|
150 | | . | /| . | \ | . |
151 \____\| . |_ | . | | | . |
152 /| . | \ | . | | | . |
153 | . | | | . | | | . |
155 |____________| | |____________| | |____________|
156 4|____________| | 4|____________| | |____________|
157 3|____________| | 3|____________| | |____________|
158 2|____________| | 2|____________| | |____________|
159 1|____________| | 1|____________| | |____________|
160 0|____________| \__\0|____________| \____\|____________|
165 Header @file{threads/mmu.h} has useful functions for various
166 operations on virtual addresses. You should look over the header
167 yourself, but its most important functions include these:
170 @item pd_no(@var{va})
171 Returns the page directory index in virtual address @var{va}.
173 @item pt_no(@var{va})
174 Returns the page table index in virtual address @var{va}.
176 @item pg_ofs(@var{va})
177 Returns the page offset in virtual address @var{va}.
179 @item pg_round_down(@var{va})
180 Returns @var{va} rounded down to the nearest page boundary, that is,
181 @var{va} but with its page offset set to 0.
183 @item pg_round_up(@var{va})
184 Returns @var{va} rounded up to the nearest page boundary.
187 @node Disk as Backing Store
188 @section Disk as Backing Store
190 In VM systems, since memory is less plentiful than disk, you will
191 effectively use memory as a cache for disk. Looking at it from
192 another angle, you will use disk as a backing store for memory. This
193 provides the abstraction of an (almost) unlimited virtual memory size.
194 Part of your task in this project is to do this, with the additional
195 constraint that your performance should be close to that provided by
196 physical memory. You will use the page tables' ``dirty'' bits to
197 denote whether pages need to be written back to disk when they're
198 evicted from main memory and the ``accessed'' bit for page replacement
199 algorithms. Whenever the hardware writes memory, it sets the dirty
200 bit, and if it reads or writes to the page, it sets the accessed bit.
202 As with any caching system, performance depends on the policy used to
203 decide which things are kept in memory and which are only stored on
204 disk. On a page fault, the kernel must decide which page to replace.
205 Ideally, it will throw out a page that will not be referenced for a
206 long time, keeping in memory those pages that are soon to be
207 referenced. Another consideration is that if the replaced page has
208 been modified, the page must be first saved to disk before the needed
209 page can be brought in. Many virtual memory systems avoid this extra
210 overhead by writing modified pages to disk in advance, so that later
211 page faults can be completed more quickly.
213 @node Memory Mapped Files
214 @section Memory Mapped Files
216 The traditional way to access the file system is via @code{read} and
217 @code{write} system calls, but that requires an extra level of copying
218 between the kernel and the user level. A secondary interface is
219 simply to ``map'' the file into the virtual address space. The
220 program can then use load and store instructions directly on the file
221 data. (An alternative way of viewing the file system is as ``durable
222 memory.'' Files just store data structures. If you access data
223 structures in memory using load and store instructions, why not access
224 data structures in files the same way?)
226 Memory mapped files are typically implemented using system calls. One
227 system call maps the file to a particular part of the address space.
228 For example, one might conceptually map the file @file{foo}, which is
230 long, starting at address 5000. Assuming that nothing else is already
231 at virtual addresses 5000@dots{}6000, any memory accesses to these
232 locations will access the corresponding bytes of @file{foo}.
234 A consequence of memory mapped files is that address spaces are
235 sparsely populated with lots of segments, one for each memory mapped
236 file (plus one each for code, data, and stack). You will implement
237 memory mapped files in problem 3-3. You should
238 design your solutions to problems 3-1 and 3-2 to anticipate this.
243 In project 2, the stack was a single page at the top of the user
244 virtual address space. The stack's location does not change in this
245 project, but your kernel should allocate additional pages to the stack
246 on demand. That is, if the stack grows past its current bottom, the
247 system should allocate additional pages for the stack as necessary
248 (unless those pages are unavailable because they are in use by another
251 It is impossible to predict how large the stack will grow at compile
252 time, so we must allocate pages as necessary. You should only
253 allocate additional pages if they ``appear'' to be stack accesses.
254 You must devise a heuristic that attempts to distinguish stack
255 accesses from other accesses. Document and explain the heuristic in
256 your design documentation.
258 The first stack page need not be loaded lazily. You can initialize it
259 with the command line at load time, with no need to wait for it to be
260 faulted in. Even if you did wait, the very first instruction in the
261 user program is likely to be one that faults in the page.
263 @node Problem 3-1 Page Table Management
264 @section Problem 3-1: Page Table Management
266 Implement page directory and page table management to support virtual
267 memory. You will need data structures to accomplish the following
272 Some way of translating in software from virtual page frames to
273 physical page frames. Consider using a hash table (@pxref{Hash
277 Some way of translating from physical page frames back to virtual page
278 frames, so that when you evict a physical page from its frame, you can
279 invalidate its translation(s).
281 It is possible to do this translation without adding a new data
282 structure, by modifying the code in @file{userprog/pagedir.c}. However,
283 if you do that you'll need to carefully study and understand section 3.7
284 in @bibref{IA32-v3}, and in practice it is probably easier to add a new
288 Some way of finding a page on disk if it is not in memory. You won't
289 need this data structure until problem 3-2, but planning ahead is a
293 The page fault handler, @func{page_fault} in
294 @file{threads/exception.c}, needs to do roughly the following:
298 Locate the page backing the virtual
299 address that faulted. It might be in the file system, in swap,
300 or it might be an invalid virtual address.
301 If you implement sharing, it might even
302 already be in physical memory and just not set up in the page table,
304 If the virtual address is invalid, that is, if there's nothing
305 assigned to go there, or if the virtual address is above
306 @code{PHYS_BASE}, meaning that it belongs to the kernel instead of the
307 user, then the process's memory access must be disallowed. You should
308 terminate the process at this point, being sure to free all of its
312 If the page is not in physical memory, fetch it by appropriate means.
313 If necessary to make room, first evict some other page from memory.
314 (When you do that you need to first remove references to the page from
315 any page table that refers to it.)
318 Point the page table entry for the faulting virtual address to the
319 physical page. You can use the functions in @file{userprog/pagedir.c}.
322 You'll need to modify the ELF loader in @file{userprog/process.c} to
323 do page table management according to your new design. As supplied,
324 it reads all the process's pages from disk and initializes the page
325 tables for them at the same time. For testing purposes, you'll
326 probably want to leave the code that reads the pages from disk, but
327 use your new page table management code to construct the page tables
328 only as page faults occur for them.
330 You should use the @func{palloc_get_page} function to get the page
331 frames that you use for storing user virtual pages. Be sure to pass
332 the @code{PAL_USER} flag to this function when you do so, because that
333 allocates pages from a ``user pool'' separate from the ``kernel pool''
334 that other calls to @func{palloc_get_page} make.
336 There are many possible ways to implement virtual memory. The above
337 is simply an outline of our suggested implementation.
339 @node Problem 3-2 Paging To and From Disk
340 @section Problem 3-2: Paging To and From Disk
342 Implement paging to and from files and the swap disk. You may use the
343 disk on interface @code{hd1:1} as the swap disk, using the disk
344 interface prototyped in @code{devices/disk.h}.
346 You will need routines to move a page from memory to disk and from
347 disk to memory, where ``disk'' is either a file or the swap disk. If
348 you do everything correctly, your VM should still work when you
349 implement your own file system for the next assignment.
351 You will need a way to track pages which are used by a process but
352 which are not in physical memory, to fully handle page faults. Pages
353 that you write to swap should not be constrained to be in sequential
354 order. You will also need a way to track all of the physical memory
355 pages, to find an unused one when needed, or to evict a page
356 when memory is needed but no empty pages are available. The data
357 structures that you designed for problem 3-1 should do most of the work for
360 You will need a page replacement algorithm. The hardware sets the
361 accessed and dirty bits when it accesses memory. You can gain access
362 to this information using the functions prototyped in
363 @file{userprog/pagedir.h}. You should be able to take advantage of
364 this information to implement some algorithm which attempts to achieve
365 LRU-type behavior. We expect that your algorithm perform at least as
366 well as a reasonable implementation of the second-chance (clock)
367 algorithm. You will need to show in your test cases the value of your
368 page replacement algorithm by demonstrating for some workload that it
369 pages less frequently using your algorithm than using some inferior
370 page replacement policy. The canonical example of a poor page
371 replacement policy is random replacement.
373 Since you will already be paging from disk, you should implement a
374 ``lazy'' loading scheme for new processes. When a process is created,
375 it will not run immediately. Therefore, it doesn't make sense to load
376 all its code, data, and stack into memory when the process is created,
377 since it might incur additional disk accesses to do so (if it gets
378 paged out before it runs). When loading a new process, you should
379 leave most pages on disk, and bring them in as demanded when the
380 program begins running. Your VM system should also use the executable
381 file itself as backing store for read-only segments, since these
382 segments won't change.
384 There are a few special cases. Look at the loop in
385 @func{load_segment} in @file{userprog/process.c}. Each time
386 around the loop, @code{read_bytes} represents the number of bytes to
387 read from the executable file and @code{zero_bytes} represents the number
388 of bytes to initialize to zero following the bytes read. The two
389 always sum to @code{PGSIZE}. The page handling depends on these
394 If @code{read_bytes} equals @code{PGSIZE}, the page should be demand
395 paged from disk on its first access.
398 If @code{zero_bytes} equals @code{PGSIZE}, the page does not need to
399 be read from disk at all because it is all zeroes. You should handle
400 such pages by creating a new page consisting of all zeroes at the
404 If neither @code{read_bytes} nor @code{zero_bytes} equals
405 @code{PGSIZE}, then part of the page is to be read from disk and the
406 remainder zeroed. This is a special case. You are allowed to handle
407 it by reading the partial page from disk at executable load time and
408 zeroing the rest of the page. This is the only case in which we will
409 allow you to load a page in a non-``lazy'' fashion. Many real OSes
410 such as Linux do not load partial pages lazily.
413 Incidentally, if you have trouble handling the third case above, you
414 can eliminate it temporarily by linking the test programs with a
415 special ``linker script.'' Read @file{tests/userprog/Makefile} for
416 details. We will not test your submission with this special linker
417 script, so the code you turn in must properly handle all cases.
419 For extra credit, you may implement sharing: when multiple processes
420 are created that use the same executable file, share read-only pages
421 among those processes instead of creating separate copies of read-only
422 segments for each process. If you carefully designed your data
423 structures in problem 3-1, sharing of read-only pages should not make this
424 part significantly harder.
426 @node Problem 3-3 Memory Mapped Files
427 @section Problem 3-3: Memory Mapped Files
429 Implement memory mapped files.
431 You will need to implement the following system calls:
435 @itemx bool mmap (int @var{fd}, void *@var{addr}, unsigned @var{length})
437 Maps the file open as @var{fd} into the process's address space
438 starting at @var{addr} for @var{length} bytes. Returns true if
439 successful, false on failure. Failure cases include the following:
443 @var{addr} is not page-aligned.
446 @var{length} is not positive.
449 The range of pages mapped overlaps any existing set of mapped pages,
450 including the stack or pages mapped at executable load time.
453 @var{length} is treated as if it were rounded up to the nearest
454 multiple of the page size, that is, as if the first statement in the
455 system call's implementation were
457 length = ROUND_UP (length, PGSIZE);
459 (The @code{ROUND_UP} macro is defined in @file{<round.h>}.)
460 The remainder of this description assumes that this has been done.
462 If @var{length} is less than @var{fd}'s length, you should only map
463 the first @var{length} bytes of the file. If @var{length} is greater
464 than @var{fd}'s length, when the file's length is also rounded up to a
465 page multiple, the call should fail. Ideally it would extend the
466 file, but our file system does not yet support growing files.
468 If @var{length} is greater than @var{fd}'s (unrounded) length, then some
469 bytes in the final mapped page ``stick out'' beyond the end of the
470 file. Set these bytes to zero when the page is faulted in from
471 disk, and discard them when the page is written back to disk.
473 Your VM system should use the @code{mmap}'d file itself as
474 backing store for the mapped segment. That is, to evict a page mapped by
475 @code{mmap} must be evicted, write it to the file it was mapped from.
476 (In fact, you may choose to implement executable mappings as a special
477 case of file mappings.)
480 @itemx bool munmap (void *addr, unsigned length)
482 Unmaps @var{length} bytes starting at @var{addr}. Returns true on
483 success, false on failure. Failure cases include the following:
487 @var{addr} is not page-aligned.
490 @var{length} is not positive.
493 One or more pages within the range to be unmapped were not mapped
494 using the @code{mmap} system call.
497 As with @code{mmap}, @var{length} is treated as if it were rounded up
498 to the nearest multiple of the page size.
500 It is valid to unmap only some of the pages that were mapped in a
501 previous system call.
504 All mappings are implicitly unmapped when a process exits, whether via
505 @code{exit} or by any other means. When a file is unmapped, whether
506 implicitly or explicitly, all outstanding changes are written to the
507 file, and the pages are removed from the process's list of used
510 @node Virtual Memory FAQ
515 @b{Do we need a working HW 2 to implement HW 3?}
521 @b{How do I use the hash table provided in @file{lib/kernel/hash.c}?}
523 First, you need to embed a @code{hash_elem} object as a member of the
524 object that the hash table will contain. Each @code{hash_elem} allows
525 the object to a member of at most one hash table at a given time. All
526 the hash table functions that deal with hash table items actually use
527 the address of a @code{hash_elem}. You can convert a pointer to a
528 @code{hash_elem} member into a pointer to the structure in which
529 member is embedded using the @code{hash_entry} macro.
531 Second, you need to decide on a key type. The key should be something
532 that is unique for each object, because a given hash table may not
533 contain two objects with equal keys. Then you need to write two
534 functions. The first is a @dfn{hash function} that converts a key
535 into an integer. Some sample hash functions that you can use or just
536 examine are given in @file{lib/kernel/hash.c}. The second function
537 needed is a @dfn{comparison function} that compares a pair and returns
538 true if the first is less than the second. These two functions have
539 to be compatible with the prototypes for @code{hash_hash_func} and
540 @code{hash_less_func} in @file{lib/kernel/hash.h}.
542 Here's a quick example. Suppose you want to put @struct{thread}s
543 in a hash table. First, add a @code{hash_elem} to the thread
544 structure by adding a line to its definition:
547 hash_elem h_elem; /* Hash table element. */
550 We'll choose the @code{tid} member in @struct{thread} as the key,
551 and write a hash function and a comparison function:
554 /* Returns a hash for E. */
556 thread_hash (const hash_elem *e, void *aux UNUSED)
558 struct thread *t = hash_entry (e, struct thread, h_elem);
559 return hash_int (t->tid);
562 /* Returns true if A's tid is less than B's tid. */
564 thread_less (const hash_elem *a_, const hash_elem *b_,
567 struct thread *a = hash_entry (a_, struct thread, h_elem);
568 struct thread *b = hash_entry (b_, struct thread, h_elem);
569 return a->tid < b->tid;
573 Then we can create a hash table like this:
578 hash_init (&threads, thread_hash, thread_less, NULL);
581 Finally, if @code{@var{t}} is a pointer to a @struct{thread},
582 then we can insert it into the hash table with:
585 hash_insert (&threads, &@var{t}->h_elem);
588 If you have any other questions about hash tables, the CS109
589 and CS161 textbooks have good chapters on them, or you can come
590 to any of the TA's office hours for further clarification.
593 @b{What are the @var{aux} parameters to the hash table functions good
596 In simple cases you won't have any need for the @var{aux} parameters.
597 In these cases you can just pass a null pointer to @func{hash_init}
598 for @var{aux} and ignore the values passed to the hash function and
599 comparison functions. (You'll get a compiler warning if you don't use
600 the @var{aux} parameter, but you can turn that off with the
601 @code{UNUSED} macro, as shown above, or you can just ignore it.)
603 @var{aux} is useful when you have some property of the data in the
604 hash table that's both constant and needed for hashing or comparisons,
605 but which is not stored in the data items themselves. For example, if
606 the items in a hash table contain fixed-length strings, but the items
607 themselves don't indicate what that fixed length is, you could pass
608 the length as an @var{aux} parameter.
611 @b{The current implementation of the hash table does not do something
612 that we need it to do. What gives?}
614 You are welcome to modify it. It is not used by any of the code we
615 provided, so modifying it won't affect any code but yours. Do
616 whatever it takes to make it work the way you want.
619 @b{What controls the layout of user programs?}
621 The linker is responsible for the layout of a user program in
622 memory. The linker is directed by a ``linker script'' which tells it
623 the names and locations of the various program segments. You can
624 learn more about linker scripts by reading the ``Scripts'' chapter in
625 the linker manual, accessible via @samp{info ld}.
629 * Problem 3-1 and 3-2 FAQ::
630 * Problem 3-3 Memory Mapped File FAQ::
633 @node Problem 3-1 and 3-2 FAQ
634 @subsection Problem 3-1 and 3-2 FAQ
638 @b{Does the virtual memory system need to support growth of the data
641 No. The size of the data segment is determined by the linker. We
642 still have no dynamic allocation in Pintos (although it is possible to
643 ``fake'' it at the user level by using memory-mapped files). However,
644 implementing it would add little additional complexity to a
645 well-designed system.
648 @b{Why do I need to pass @code{PAL_USER} to @func{palloc_get_page}
649 when I allocate physical page frames?}@anchor{Why PAL_USER?}
651 You can layer some other allocator on top of @func{palloc_get_page}
652 if you like, but it should be the underlying mechanism, directly or
653 indirectly, for two reasons. First, running out of pages in the user
654 pool just causes user programs to page, but running out of pages in
655 the kernel pool will cause all kinds of problems, because many kernel
656 functions depend on being able to allocate memory. Second, you can
657 use the @option{-ul} option to @command{pintos} to limit the size of
658 the user pool, which makes it easy to test your VM implementation with
659 various user memory sizes.
662 @node Problem 3-3 Memory Mapped File FAQ
663 @subsection Problem 3-3: Memory Mapped File FAQ
667 @b{How do we interact with memory-mapped files?}
669 Let's say you want to map a file called @file{foo} into your address
670 space at address @t{0x10000000}. You open the file, determine its
671 length, and then use @code{mmap}:
679 void *addr = (void *) 0x10000000;
680 int fd = open ("foo");
681 int length = filesize (fd);
682 if (mmap (fd, addr, length))
683 printf ("success!\n");
687 Suppose @file{foo} is a text file and you want to print the first 64
688 bytes on the screen (assuming, of course, that the length of the file
689 is at least 64). Without @code{mmap}, you'd need to allocate a
690 buffer, use @code{read} to get the data from the file into the buffer,
691 and finally use @code{write} to put the buffer out to the display. But
692 with the file mapped into your address space, you can directly address
696 write (addr, 64, STDOUT_FILENO);
699 Similarly, if you wanted to replace the first byte of the file,
700 all you need to do is:
706 When you're done using the memory-mapped file, you simply unmap
710 munmap (addr, length);
714 @b{What if two processes memory-map the same file?}
716 There is no requirement in Pintos that the two processes see
717 consistent data. Unix handles this by making the processes share the
718 same physical page, but the @code{mmap} system call also has an
719 argument allowing the client to specify whether the page is shared or
720 private (i.e.@: copy-on-write).
723 @b{What happens if a user removes a @code{mmap}'d file?}
725 You should follow the Unix convention and the mapping should still be
726 valid. @xref{Removing an Open File}, for more information.
729 @b{What if a process writes to a page that is memory-mapped, but the
730 location written to in the memory-mapped page is past the end
731 of the memory-mapped file?}
733 Can't happen. @code{mmap} checks that the mapped region is within the
734 file's length and Pintos provides no way to shorten a file. (Until
735 project 4, there's no way to extend a file either.) You can remove a
736 file, but the mapping remains valid (see the previous question).
739 @b{Do we have to handle memory mapping @code{stdin} or @code{stdout}?}
741 No. Memory mapping implies that a file has a length and that a user
742 can seek to any location in the file. Since the console device has
743 neither of these properties, @code{mmap} should return false when the
744 user attempts to memory map a file descriptor for the console device.
747 @b{What happens when a process exits with mapped files?}
749 When a process finishes, each of its mapped files is implicitly
750 unmapped. When a process @code{mmap}s a file and then writes into the
751 area for the file it is making the assumption the changes will be
755 @b{If a user closes a mapped file, should it be automatically
758 No, once created the mapping is valid until @code{munmap} is called
759 or the process exits.