Re: [Xen-devel] [PATCH 0/2] question on libxl ao [and 1 more messages]

["Bamvor Jian Zhang" <bjzhang@xxxxxxxx>]

 >>>Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx> wrote: 
> Bamvor Jian Zhang writes ("[PATCH 0/2] question on libxl ao"): 
>> there are four choices for using ao_how in libvirt. 
>> (a), ao without callback and set libxl_sigchld_owner_mainloop 
>> (b), ao without callback and set libxl_sigchld_owner_libxl_always 
>> (c), ao with callback and set libxl_sigchld_owner_mainloop 
>> (d), ao with callback and set libxl_sigchld_owner_libxl_always 
>>  
>> i got some trouble when i try (b) and (c). 
>> i still not decided which one is better choice for implement in 
>> libvirt. it would be easier if i could try above four choice.
>  
> I'm not sure I follow.  I think the right answer for libvirt is 
> probably aos specifying callback, and libxl_sigchld_owner_mainloop. 
> I'm assuming that libvirt has something for handling children 
> already.  But I haven't looked.  If it doesn't then 
> libxl_sigchld_owner_always would probably be best.
AFAICS, no child handle in libvirt. i will try your patches.

thanks.

bamvor
>  
> Bamvor Jian Zhang writes ("[PATCH 1/2] expose child fd in order to handle  
> child exit in libvirt"): 
>> libvirt could handle fd and timeout event through 
>> libxl_osevent_hooks. either of these will not inform the child 
>> exit if libvirt set the libxl_sigchld_owner_libxl_always.
>  
> I don't exactly follow, but I think you have found a bug.
>  
> If libvirt specifies libxl_sigchld_owner_libxl_always then libxl will 
> install a SIGCHLD handler and to the self-pipe trick.  The self-pipe 
> fd should be registered with the fd event machinery in the normal way. 
> However, this doesn't actually appear to happen.  Instead the 
> beforepoll/afterpoll functions handle that fd ad-hoc.  The result is 
> that, probably, nothing will notice the selfpipe becoming writable.
>  
> I will see about writing a patch to fix this.
>  
>> add a function for returning the sigchld_selfpipe in order to 
>> handle the child exit in libvirt. 
>> meanwhile, there is only one pipe in ctx, it seems that it is 
>> not worth to add this to libxl_osevent_hooks.
>  
> This approach is not right, though.  There is no need to add 
> additional interfaces to libxl.  The bug just needs to be fixed.
>  
> Thanks, 
> Ian.
>  
> _______________________________________________ 
> Xen-devel mailing list 
> Xen-devel@xxxxxxxxxxxxx 
> http://lists.xen.org/xen-devel 



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel