In preparation for switching linuxulator to the use the native 1:1 threads add hook for cleaning thread resources before the thread die.

[dchagin]

As discussed with kib@ this change is incorrect. At this point, the execing process does not have the address space.
Possible direction is to put a sv_thread_detach hook in to the proc_exec handler.

[dchagin]

I have not found a suitable solution, so this is the best way to handle single threading.
execve() case to be handled by eventhadler.

[dchagin]

reclaim

[trasz]

Konstantin, is the problem still there?

[kib]

In D1038#9, @trasz wrote:

Konstantin, is the problem still there?

What problem ?

[trasz]

Quoting the first comment above, "As discussed with kib@ this change is incorrect. At this point, the execing process does not have the address space. Possible direction is to put a sv_thread_detach hook in to the proc_exec handler". So, my question is, is this patch still incorrect, or has it been reworked to make it work properly?

[kib]

In D1038#13, @trasz wrote:

Quoting the first comment above, "As discussed with kib@ this change is incorrect. At this point, the execing process does not have the address space. Possible direction is to put a sv_thread_detach hook in to the proc_exec handler". So, my question is, is this patch still incorrect, or has it been reworked to make it work properly?

I have no information about patch rewrite. Isn't Dmitry supposed to know more about state of the work on _his_ patch ?

[trasz]

Hm, okay... Dmitry, is the problem previously pointed out by Konstantin still there, or is the patch correct now?

[dchagin]

whoops, sorry for the delay, im so busy these days.. ok, let me to clarify.

for Linux thread we need a proper way to do some cleanup work at thread exit time,
some of this work require access to the user space - robust futex cleanup code (sx used here).

Konstantin quite right pointed out me that this code could not work for exec() case.
As suspended threads call thread_suspend_check() from ast() or sleep() and at this time old
proc vmspace does not exist, so my sv_thread_detach() would fail.

ie, this method works only for exit1(), and in linux_exec() I must finish all threads before do_execve() call.

[dchagin]

D1039 updated, linux_common_execve() in linux_emul.c