THREAD_GROUP and linux thread

THREAD_GROUP and linux thread

[Hua Zhong]

Hi:
 
I am making some kernel changes to keep track of threads. Basically I
want to maintain a linked list of all threads.
 
The linux thread library on my system doesn't make use of the
CLONE_THREAD flag, so the thread_group list is not being used. I can
just add a new field in task_struct and maintain the list in
fork.c/exec.c/exit.c exactly as what the kernel does to the thread_group
list, but I am just wondering if I could just change the kernel to treat
CLONE_VM the same way as CLONE_THREAD which is a much simpler change.
 
So my questions are:
 
1. Is it safe to do so? [ in other word, is it a requirement to use
CLONE_VM together with CLONE_THREAD ]
2. Which version of pthread uses the CLONE_THREAD flag?
 
I couldn't find the documentation about CLONE_THREAD usages, so your
reply is very much appreciated.
 
Hua

Re: THREAD_GROUP and linux thread

[Jamie Lokier]

Hua Zhong wrote:
> but I am just wondering if I could just change the kernel to treat
> CLONE_VM the same way as CLONE_THREAD which is a much simpler change.

No.  There are a good many programs which use CLONE_VM without
CLONE_THREAD, and I'm sure they will be surprised to find signals
suddenly being shared, which is implied by CLONE_THREAD.

> 2. Which version of pthread uses the CLONE_THREAD flag?

NPTL.

Enjoy,
-- Jamie

RE: THREAD_GROUP and linux thread

[Hua Zhong]

> Hua Zhong wrote:
> > but I am just wondering if I could just change the kernel to treat
> > CLONE_VM the same way as CLONE_THREAD which is a much 
> simpler change.
> 
> No.  There are a good many programs which use CLONE_VM without
> CLONE_THREAD, and I'm sure they will be surprised to find signals
> suddenly being shared

Thank you, that's exactly what I want to know.

> which is implied by CLONE_THREAD.

The most accurate description is "which is implied by thread_group
list"....

So I guess I'll live with having a separate list.

> > 2. Which version of pthread uses the CLONE_THREAD flag?
> 
> NPTL.
> 
> Enjoy,
> -- Jamie
>