Write a version for pthread_get_name_np

[Willem Jan Withagen]

On 13/08/2018 02:24, Yuri Pankov wrote:
> Konstantin Belousov wrote:
>> On Mon, Aug 13, 2018 at 01:36:28AM +0200, Willem Jan Withagen wrote:
>>> On 13/08/2018 00:57, Konstantin Belousov wrote:
>>>> On Mon, Aug 13, 2018 at 12:01:03AM +0200, Willem Jan Withagen wrote:
>>>>> On 12/08/2018 22:58, Konstantin Belousov wrote:
>>>>>> On Sun, Aug 12, 2018 at 10:33:06PM +0200, Willem Jan Withagen wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> For some porting I'm trying to write a pthread_get_name_np(), but
>>>>>>> I have run into a snag....
>>>>>>> Tried some the the procstat code I would come up with:
>>>>>>>
>>>>>>> int pthread_get_name_np(pthread_t *thread,
>>>>>>>                             char *name, size_t len){
>>>>>>>         struct procstat *procstat;
>>>>>>>         int pid = getpid();
>>>>>>>         unsigned int count, i;
>>>>>>>         struct kinfo_proc *kip, *kipp;
>>>>>>>
>>>>>>>         procstat = procstat_open_sysctl();
>>>>>>>         kip = procstat_getprocs(procstat, KERN_PROC_PID | 
>>>>>>> KERN_PROC_INC_THREAD,
>>>>>>>                 pid, &count);
>>>>>>>         for (i = 0; i < count; i++) {
>>>>>>>             kipp = &kip[i];
>>>>>>>             printf("thread: %ld(0x%lx), %ld(0x%lx)\n", thread, 
>>>>>>> thread,
>>>>>>> kipp->ki_tid, kipp->ki_tid);
>>>>>>>             if (thread == kipp->ki_tid) {
>>>>>>>                 kinfo_proc_thread_name(kipp, name, len);
>>>>>>>             }
>>>>>>>         }
>>>>>>>         return 0;
>>>>>>> }
>>>>>>>
>>>>>>> Problem only is that the TID value in what procstat_getprocs 
>>>>>>> returns in
>>>>>>> ki_tid is nowhere near
>>>>>>> what pthread_self() returns.
>>>>>>> But both manual page and the include file describe the value as 
>>>>>>> Thread ID.
>>>>>>> Only in sys/user.h the type is
>>>>>>>         lwpid_t ki_tid;                 /* XXXKSE thread id */
>>>>>>>
>>>>>>> What do I need to translate one tid into the other so I can really
>>>>>>> compare them?
>>>>>>
>>>>>> You need to cast pthread_t * to struct thread *, this is an internal
>>>>>> libthr structure which represents the thread in userspace.  The
>>>>>> structure field tid gives you the tid you want to pass to sysctl.
>>>>>
>>>>> mmmm,
>>>>>
>>>>> I'm afraid I'm not quit able to followup on your answer.
>>>>>
>>>>> Trying to find what to include to be able to cast this, I can only 
>>>>> find
>>>>> a small bit defined in /usr/src/lib/libthr/thread/thr_private.h.
>>>>> /*
>>>>>     * lwpid_t is 32bit but kernel thr API exports tid as long type
>>>>>     * to preserve the ABI for M:N model in very early date (r131431).
>>>>>     */
>>>>> #define TID(thread)     ((uint32_t) ((thread)->tid))
>>>>>
>>>>> But that I cannot just "include" that file without a lot of mess, and
>>>>> then still it does not compile.
>>>>>
>>>>> So could you point me to where this private part of struct thread is
>>>>> hidding?
>>>> I do not understand your confusion. The thr_private.h header is only
>>>> for use by libthr, the FreeBSD threading library implementation. The
>>>> function that you are trying to implement, requires understanding of 
>>>> the
>>>> libthr internals and can only be implemented as part of libthr.
>>>>
>>>> Any other attempt to translate libthr handle for thread into tid needs
>>>> same access to the struct pthread.
>>>
>>> Oke,
>>>
>>> I don't have more knowledge of (p)threads than just how to use it.
>>> I was trying to do this as a piece of code in the compat.cc of my 
>>> ceph port.
>>>
>>> But trying to reread what you write is boils down to: there is no way to
>>> link a value returned by (struct pthread *)pthread_self() to the kd_tid
>>> value that can be found in the (struct kinfo_proc *) describing that
>>> same thread.
>> No, this is not what I said.
>>
>> I am claiming that this functionality must be implemented in libthr.
>> So if you want this function, it needs to be added to libthr, and it is
>> quite trivial to do.
>>
>> You can code it, or you might provide (draft of the) man page, and I
>> will code it according to the spec.
> 
> How about the attached patch for man page part?

Thanx,

I guess that'll work.

Although I was following the procstat way and when there was no name set 
we get a '-'. But if Konstatin is going to write it, it is his prerogative.
And I would expect it to be equivalent partner of thr_{set,get}_name as 
well. But that could be added later in the man page.

--WjW