1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
|
[[meta copyright="Copyright © 2008 Free Software Foundation, Inc."]]
[[meta license="""[[toggle id="license" text="GFDL 1.2+"]][[toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license
is included in the section entitled
[[GNU_Free_Documentation_License|/fdl]]."]]"""]]
Some projects:
# Minor
## New hash function
The current hash function in libhurd-ihash results in a lot of
collisions when the hash table is 80% full. To overcome this, we keep
hash tables at most 30% full. This represents a fair amount of
overhead. Find a better algorithm. There can either be one that is
appropriate in the general case or one that works well in a relevant,
specific case, e.g., viengoos/object.c uses a hash to find the object
corresponding to a frame, which is keyed on its physical address.
# Major
## Address Space Management
In Viengoos, a process's address space is managed entirely in user
space by the process itself. This creates two interesting problems:
dealing with circular dependencies resulting from having to manage the
address space data structures and accessing and manipulating the
address space data structures.
First, managing the address space requires resources, which in turn
may require address space (e.g., data structures require memory which
require address space, etc.). We currently break this circular
dependency by trying to keep enough resources in reserve that
allocating resources for managing the address space never requires
more resources than are minimally in the reserve. The reserve is
currently chosen in an ad-hoc fashion. It would be nice to determine
it more systematically. Moreover, it would be nice to reduce the
cases in which a reserve is required. This may be possible by
restructuring some of the code.
Second, the address space data structures are protected using a single
lock. This not only means that only a single thread can be updating
the address space at a time, but that if a thread faults and the
address space is locked, then the process dead locks! It should be
possible to at least walk the address space using lock-free
techniques. This requires updating the address space construction
code such that all addresses remain valid during any given
manipulation. Second, to avoid the mentioned dead-lock problem, we
try to ensure that accessing the data structures will never result in
a fault. This means protecting the stack. An alternative approach is
to use undo buffers.
# Thesis
## Capability aware compiler
Modify, e.g., gcc to understand capability semantics and teach gcc how
to optimize it, e.g., how to batch and combine calls.
|
