[Lldb-commits] [PATCH] D38938: Logging: provide a way to safely disable logging in a forked process

[Jim Ingham via lldb-commits]

This thread behavior over fork is the same on Mach, BTW.

Apple extended posix_spawnp to take a “Stop at first instruction” attribute, basically just running PT_TRACEME for us.  That has ended up being very handy because we get all the nice cleanup behavior that pthread_spawn does, but still get the one thing the debugger needs fork for…  Don’t know whether it does something useful for locks, however.

Jim


> On Oct 16, 2017, at 2:06 PM, Tamas Berghammer via lldb-commits <lldb-commits at lists.llvm.org> wrote:
> 
> On linux when you call fork the new process will only have the thread what called fork. Other threads will be ignored with leaving whatever dirty state they had left in the new process. Regarding execve it doesn't do fork so we would have to do fork & execve what have the same issue (actually we are using execve as of now but it isn't different from exec in this regard).
> 
> Tamas
> 
> On Mon, Oct 16, 2017 at 1:57 PM Zachary Turner via lldb-commits <lldb-commits at lists.llvm.org <mailto:lldb-commits at lists.llvm.org>> wrote:
> On Sun, Oct 15, 2017 at 3:15 PM Pavel Labath <labath at google.com <mailto:labath at google.com>> wrote:
> On 15 October 2017 at 23:04, Zachary Turner <zturner at google.com <mailto:zturner at google.com>> wrote:
> > Doesn’t DisableAllLogChannels acquire a scoped lock? If so wouldn’t it just
> > release at the end of the function?
> 
> 
> The thing is, it is unable to acquire the lock in the first place,
> because the mutex is already "locked". So, the sequence of events is
> process 1, thread 1: acquire lock
> process 1, thread 2: fork(), process 2 is created
> process 1 thread 1: release lock
> 
> Suppose process 1 thread 1 had been executing this code:
> ```
> lock();
> logSomething();  // thread 2 forks when thread 1 is here.
> unlock();
> ```
> 
> Doesn't thread 2 in the child process resume running from the same point?  If so, it seems that both the child and parent would both gracefully unlock the lock.
> 
> I'm sure I'm wrong about this, but hopefully you can clarify what I'm missing.
> 
> As a follow-up question, why can't we use execve instead of fork / exec?
> _______________________________________________
> lldb-commits mailing list
> lldb-commits at lists.llvm.org <mailto:lldb-commits at lists.llvm.org>
> http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits <http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits>
> _______________________________________________
> lldb-commits mailing list
> lldb-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20171016/a2b30ea7/attachment.html>