[ros-dev] re: ps/kill.c change

[Alex Ionescu]

Hartmut Birr wrote:

>Alex Ionescu wrote:
>
>  
>
>>Hi,
>>
>>Sorry but I accidentally lost 60 emails including the one I'm supposed
>>to reply to.
>>
>>Hartmut, you have done a recent change in kill.c in which you change
>>the APC inside PspTerminateThreadByPointer to a user-mode APC. This is
>>incorrect and I don't understand why it was done... furthermore, I've
>>even documented a source of information which proves my code was
>>correct (on top of the function header), so why did you change it?
>>
>>    
>>
>Hi,
>
>the thread termination APC must be a user mode APC. The kernel routine
>(PsExitSpecialApc) calls PspExitThread. PspExitThread can only be called
>if no cleanup is necessary like freeing pool memory or dereferencing
>kernel mode objects. The only point for the thread terminating is the
>return to user mode after a system call or at the end of an exception or
>hardware interrupt. PspExitThread can not be called from a kernel mode
>APC, because the apc is delivered at any point where the irql drops
>below APC_LEVEL. The thread may have referenced some kernel objects at
>this point, may have allocated pool memory or may have locked the kernel
>address space and so on. Currently there exist some situations where it
>isn't possible to terminate a thread if the thread runs in kernel mode.
>Such a situation is if the thread waits on an object which will not be
>signaled.
>
>- Hartmut
>  
>
Hi,

Your reply makes perfect sense and now that I look at it, I see that 
there is a certain part of the code that I forgot to write (I wrote a 
comment on it)...however I insist that the first APC queued is Kernel, 
you can see it for yourself in the NT source code snippet...but in the 
end it does become a user-mode APC...I'll write up a patch.

Best regards,
Alex Ionescu