| 1 |
Hi everyone, |
| 2 |
|
| 3 |
I'm going to submit a bug for this, but the linux-headers-2.4.26-r1 |
| 4 |
require another small patch. This specifically effects the |
| 5 |
asm-arm/unistd.h file. |
| 6 |
|
| 7 |
The syscall for waitpid is not actually defined for the arm, instead it |
| 8 |
should always be changed to wait4 via the static inline waitpid function. |
| 9 |
|
| 10 |
When building glibc-2.3.2, however, there is a check made for |
| 11 |
__NR_waitpid (system call # 7). That should _not_ be defined in the |
| 12 |
linux-headers. If so, then any glibc-linked (static or dynamic) |
| 13 |
application will return errno 38 (ENOSYS/Function not implemented) when |
| 14 |
waitpid is invoked. |
| 15 |
|
| 16 |
This can be verified if my instructions are followed below for creating |
| 17 |
a cross-toolchain based on linux-headers-2.4.26-r1 and glibc-2.3.2-r12. |
| 18 |
That combination will fail for waitpid calls. With the |
| 19 |
linux-headers-2.4.26-r2 ebuild in my overlay, waitpid works properly. |
| 20 |
|
| 21 |
http://gentoo-wiki.com/Gentoo_for_the_TS72xx_(Old_Toolchain) |
| 22 |
|
| 23 |
In my overlay, I've created a separate ebuild for glibc-2.3.2 so it |
| 24 |
behaves better with the more recent glibc ebuild structure, and also a |
| 25 |
separate ebuild called linux-headers-2.4.26-r2 which has my very tiny |
| 26 |
patch as well. |
| 27 |
|
| 28 |
A simple test program is attached, which can be shown to provide the |
| 29 |
output below: |
| 30 |
|
| 31 |
### Incorrect output |
| 32 |
./waitpid_demo; echo $? |
| 33 |
This is the parent, pid 4106 |
| 34 |
|
| 35 |
waitpid: 38: Function not implemented |
| 36 |
|
| 37 |
|
| 38 |
This is the child, pid 4107 |
| 39 |
child sleeping |
| 40 |
|
| 41 |
child sleeping |
| 42 |
|
| 43 |
child sleeping |
| 44 |
|
| 45 |
child woke up! |
| 46 |
|
| 47 |
### Correct output, with my patch applied to the headers |
| 48 |
./waitpid_demo |
| 49 |
|
| 50 |
This is the child, pid 4179 |
| 51 |
|
| 52 |
child sleeping |
| 53 |
|
| 54 |
This is the parent, pid 4178 |
| 55 |
|
| 56 |
parent sleeping |
| 57 |
|
| 58 |
child sleeping |
| 59 |
|
| 60 |
parent sleeping |
| 61 |
|
| 62 |
child sleeping |
| 63 |
|
| 64 |
parent sleeping |
| 65 |
|
| 66 |
child woke up! |
| 67 |
|
| 68 |
child 4179 returned |
| 69 |
|
| 70 |
The patch is also attached. |
| 71 |
|
| 72 |
This obviously has a massive effect on any program that uses waitpid - |
| 73 |
and for gnu systems that means _lots_ of libraries and applications |
| 74 |
would otherwise not work. |
| 75 |
|
| 76 |
I'll file the bug momentarily. |
| 77 |
|
| 78 |
Cheers, |
| 79 |
|
| 80 |
Chris |
Archives