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
|
[[!meta copyright="Copyright © 2013 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]]."]]"""]]
[[!tag open_issue_glibc open_issue_gnumach open_issue_hurd]]
Evaluate whether or not usage of vDSOs (virtual dynamically linked shared
objects; [[!wikipedia vDSO]]) can be useful in a GNU Hurd system.
Explanation and example for the Linux kernel: [Creating a vDSO: the Colonel's
Other
Chicken](http://www.linuxjournal.com/content/creating-vdso-colonels-other-chicken),
Matt Davis, 2012-02-06. The *Resources* given are also worth reading.
Basically, this is useful for exporting data from the kernel (generally, or
given a process context ([[Unix]]), or task/thread context, and so on).
On a GNU Hurd system, parts of the data that makes up a process context doesn't
actually live inside the kernel, but instead is directly held in glibc. For
example `sysdeps/mach/hurd/getpid.c:__getpid` does a mere `return _hurd_pid`.
For this reason, vDSOs might not be as useful on GNU Hurd as they are with the
Linux kernel. Or, put another way, as GNU Hurd system doesn't have many
[[system_call]]s, also there aren't many that could be replaced.
Generally only *real* [[system_call]]s should be candidates for implementation
with vDSO code, because otherwise that'd break the ([[RPC]]) system's inherent
[[/virtualization]] capabilities.
Having vDSO code might be useful for:
* `mach_*_self`: `mach_host_self`, `mach_task_self`, `mach_thread_self`?
* [[mapped-time_interface|microkernel/mach/gnumach/interface/device/time]]
Every application can then use that via the regular
`gettimeofday`/`clock_gettime` and similar calls instead of using the
special [[hurd/libshouldbeinlibc]]'s `<maptime.h>` interface.
Can implement [[`clock_gettime` stuff|clock_gettime]] more easily that way,
for example for nanosecond precision?
Now, the [[mapped-time_interface]] is virtualizable -- the question is
whether there is a way so that we can make a compromise here?
|
