5 - In Project 2, we're missing tests that pass arguments to system calls
6 that span multiple pages, where some are mapped and some are not.
7 An implementation that only checks the first page, rather than all pages
8 that can be touched during a call to read()/write() passes all tests.
10 - In Project 2, we're missing a test that would fail if they assumed
11 that contiguous user-virtual addresses are laid out contiguously
12 in memory. The loading code should ensure that non-contiguous
13 physical pages are allocated for the data segment (at least.)
15 - Need some tests that test that illegal accesses lead to process
16 termination. I have written some, will add them. In P2, obviously,
17 this would require that the students break this functionality since
18 the page directory is initialized for them, still it would be good
21 - There does not appear to be a test that checks that they close all
22 fd's on exit. Idea: add statistics & self-diagnostics code to palloc.c
23 and malloc.c. Self-diagnostics code could be used for debugging.
24 The statistics code would report how much kernel memory is free.
25 Add a system call "get_kernel_memory_information". User programs
26 could engage in a variety of activities and notice leaks by checking
27 the kernel memory statistics.
31 From: "Godmar Back" <godmar@gmail.com>
32 Subject: priority donation tests
33 To: "Ben Pfaff" <blp@cs.stanford.edu>
34 Date: Fri, 3 Mar 2006 11:02:08 -0500
38 it seems the priority donation tests are somewhat incomplete and allow
39 incorrect implementations to pass with a perfect score.
41 We are seeing the following wrong implementations pass all tests:
43 - Implementations that assume locks are released in the opposite order
44 in which they're acquired. The students implement this by
45 popping/pushing on the donation list.
47 - Implementations that assume that the priority of a thread waiting on
48 a semaphore or condition variable cannot change between when the
49 thread was blocked and when it is unblocked. The students implement
50 this by doing an insert into an ordered list on block, rather than
51 picking the maximum thread on unblock.
53 Neither of these two cases is detected; do you currently check for
54 these mistakes manually?
56 I wrote a test that checks for the first case; it is here:
57 http://people.cs.vt.edu/~gback/pintos/priority-donate-multiple-2.patch
61 I also wrote a test case for the second scenario:
62 http://people.cs.vt.edu/~gback/pintos/priority-donate-sema.c
63 http://people.cs.vt.edu/~gback/pintos/priority-donate-sema.ck
65 I put the other tests up here:
66 http://people.cs.vt.edu/~gback/pintos/priority-donate-multiple2.c
67 http://people.cs.vt.edu/~gback/pintos/priority-donate-multiple2.ck
69 From: "Godmar Back" <godmar@gmail.com>
70 Subject: multiple threads waking up at same clock tick
71 To: "Ben Pfaff" <blp@cs.stanford.edu>
72 Date: Wed, 1 Mar 2006 08:14:47 -0500
74 Greg Benson points out another potential TODO item for P1.
79 The alarm tests do not test to see if multiple threads are woken up if
80 their timers have expired. That is, students can write a solution
81 that just wakes up the first thread on the sleep queue rather than
82 check for additional threads. Of course, the next thread will be
83 woken up on the next tick. Also, this might be hard to test.
86 Way to test this: (from Godmar Back)
88 Thread A with high priority spins until 'ticks' changes, then calls to
89 timer_sleep(X), Thread B with lower priority is then resumed, calls
90 set_priority to make its priority equal to that of thread A, then
91 calls timer_sleep(X), all of that before the next clock interrupt
94 On wakeup, each thread records wake-up time and calls yield
95 immediately, forcing the scheduler to switch to the other
96 equal-priority thread. Both wake-up times must be the same (and match
97 the planned wake-up time.)
100 I actually tested it and it's hard to pass with the current ips setting.
101 The bounds on how quickly a thread would need to be able to return after
102 sleep appear too tight. Need another idea.
104 From: "Godmar Back" <godmar@gmail.com>
106 For reasons I don't currently understand, some of our students seem
107 hesitant to include each thread in a second "all-threads" list and are
108 looking for ways to implement the advanced scheduler without one.
110 Currently, I believe, all tests for the mlfqs are such that all
111 threads are either ready or sleeping in timer_sleep(). This allows for
112 an incorrect implementation in which recent-cpu and priorities are
113 updated only for those threads that are on the alarm list or the ready
116 The todo item would be a test where a thread is blocked on a
117 semaphore, lock or condition variable and have its recent_cpu decay to
118 zero, and check that it's scheduled right after the unlock/up/signal.
120 From: "Godmar Back" <godmar@gmail.com>
121 Subject: set_priority & donation - a TODO item
122 To: "Ben Pfaff" <blp@cs.stanford.edu>
123 Date: Mon, 20 Feb 2006 22:20:26 -0500
127 it seems that there are currently no tests that check the proper
128 behavior of thread_set_priority() when called by a thread that is
129 running under priority donation. The proper behavior, I assume, is to
130 temporarily drop the donation if the set priority is higher, and to
131 reassume the donation should the thread subsequently set its own
132 priority again to a level that's lower than a still active donation.
136 From: Godmar Back <godmar@gmail.com>
137 Subject: project 4 question/comment regarding caching inode data
138 To: Ben Pfaff <blp@cs.stanford.edu>
139 Date: Sat, 14 Jan 2006 15:59:33 -0500
143 in section 6.3.3 in the P4 FAQ, you write:
145 "You can store a pointer to inode data in struct inode, if you want,"
147 Should you point out that if they indeed do that, they likely wouldn't
148 be able to support more than 64 open inodes systemwide at any given
151 (This seems like a rather strong limitation; do your current tests
152 open more than 64 files?
153 It would also point to an obvious way to make the projects harder by
154 specifically disallowing that inode data be locked in memory during
155 the entire time an inode is kept open.)
159 From: Godmar Back <godmar@gmail.com>
160 Subject: on caching in project 4
161 To: Ben Pfaff <blp@cs.stanford.edu>
162 Date: Mon, 9 Jan 2006 20:58:01 -0500
164 here's an idea for future semesters.
166 I'm in the middle of project 4, I've started by implementing a buffer
167 cache and plugging it into the existing filesystem. Along the way I
168 was wondering how we could test the cache.
170 Maybe one could adopt a similar testing strategy as in project 1 for
171 the MLQFS scheduler: add a function that reads "get_cache_accesses()"
172 and a function "get_cache_hits()". Then create a version of pintos
173 that creates access traces for a to-be-determined workload. Run an
174 off-line analysis that would determine how many hits a perfect cache
175 would have (MAX), and how much say an LRU strategy would give (MIN).
176 Then add a fudge factor to account for different index strategies and
177 test that the reported number of cache hits/accesses is within (MIN,
178 MAX) +/- fudge factor.
180 (As an aside - I am curious why you chose to use a clock-style
181 algorithm rather than the more straightforward LRU for your buffer
182 cache implementation in your sample solution. Is there a reason for
183 that? I was curious to see if it made a difference, so I implemented
184 LRU for your cache implementation and ran the test workload of project
185 4 and printed cache hits/accesses.
186 I found that for that workload, the clock-based algorithm performs
187 almost identical to LRU (within about 1%, but I ran nondeterministally
188 with QEMU). I then reduced the cache size to 32 blocks and found again
189 the same performance, which raises the suspicion that the test
190 workload might not force any cache replacement, so the eviction
191 strategy doesn't matter.)
193 Godmar Back <godmar@gmail.com> writes:
195 > in your sample solution to P4, dir_reopen does not take any locks when
196 > changing a directory's open_cnt. This looks like a race condition to
197 > me, considering that dir_reopen is called from execute_process without
198 > any filesystem locks held.
200 * Get rid of rox--causes more trouble than it's worth
202 * Reconsider command line arg style--confuses everyone.
204 * Finish writing tour.
208 * Get rid of mmap syscall, add sbrk.
210 * page-linear, page-shuffle VM tests do not use enough memory to force
211 eviction. Should increase memory consumption.
213 * Add FS persistence test(s).
215 * process_death test needs improvement
219 * Improve automatic interpretation of exception messages.
223 - Mark read-only pages as actually read-only in the page table. Or,
224 since this was consistently rated as the easiest project by the
225 students, require them to do it.
227 - Don't provide per-process pagedir implementation but only
228 single-process implementation and require students to implement
229 the separation? This project was rated as the easiest after all.
230 Alternately we could just remove the synchronization on pid
231 selection and check that students fix it.
235 - Need a better way to measure performance improvement of buffer
236 cache. Some students reported that their system was slower with
237 cache--likely, Bochs doesn't simulate a disk with a realistic
242 - Add "Digging Deeper" sections that describe the nitty-gritty x86
243 details for the benefit of those interested.
245 - Add explanations of what "real" OSes do to give students some
252 . Low-level x86 stuff, like paged page tables.
254 . Specifics on how to implement sbrk, malloc.
258 . opendir/readdir/closedir
260 . everything needed for getcwd()
262 To add partition support:
264 - Find four partition types that are more or less unused and choose to
265 use them for Pintos. (This is implemented.)
267 - Bootloader reads partition tables of all BIOS devices to find the
268 first that has the "Pintos kernel" partition type. (This is
269 implemented.) Ideally the bootloader would make sure there is
270 exactly one such partition, but I didn't implement that yet.
272 - Bootloader reads kernel into memory at 1 MB using BIOS calls. (This
275 - Kernel arguments have to go into a separate sector because the
276 bootloader is otherwise too big to fit now? (I don't recall if I
277 did anything about this.)
279 - Kernel at boot also scans partition tables of all the disks it can
280 find to find the ones with the four Pintos partition types (perhaps
281 not all exist). After that, it makes them available to the rest of
282 the kernel (and doesn't allow access to other devices, for safety).
284 - "pintos" and "pintos-mkdisk" need to write a partition table to the
285 disks that they create. "pintos-mkdisk" will need to take a new
286 parameter specifying the type. (I might have partially implemented
287 this, don't remember.)
289 - "pintos" should insist on finding a partition header on disks handed
292 - Need some way for "pintos" to assemble multiple disks or partitions
293 into a single image that can be copied directly to a USB block
294 device. (I don't know whether I came up with a good solution yet or
295 not, or whether I implemented any of it.)
299 - Needs to be able to scan PCI bus for UHCI controller. (I
300 implemented this partially.)
302 - May want to be able to initialize USB controllers over CardBus
303 bridges. I don't know whether this requires additional work or if
304 it's useful enough to warrant extra work. (It's of special interest
305 for me because I have a laptop that only has USB via CardBus.)
307 - There are many protocol layers involved: SCSI over USB-Mass Storage
308 over USB over UHCI over PCI. (I may be forgetting one.) I don't
309 know yet whether it's best to separate the layers or to merge (some
310 of) them. I think that a simple and clean organization should be a
313 - VMware can likely be used for testing because it can expose host USB
314 devices as guest USB devices. This is safer and more convenient
315 than using real hardware for testing.
317 - Should test with a variety of USB keychain devices because there
318 seems to be wide variation among them, especially in the SCSI
319 protocols they support. Should try to use a "lowest-common
320 denominator" SCSI protocol if any such thing really exists.
322 - Might want to add a feature whereby kernel arguments can be given
323 interactively, rather than passed on-disk. Needs some though.