jstack reports wrong thread priorities‏

[David Holmes]

Hi Dmytro,

Native priorities are a bit of a mess as you can tell. I like your 
suggestion of reporting native priority plus Java priority distinctly, 
and without this complex mapping scheme. But I think part of the problem 
is that for some threads the priority is only manipulated at the native 
level and so the Java priority never gets updated.

The patch for os.cpp get_priority seems reasonable.

David

On 28/08/2012 1:46 AM, Dmytro Sheyko wrote:
>
>
>
>
> Hi,
>
> I noticed that jstack on Linux for all threads always shows "prio=10" (which is wrong especially for threads that have NORM_PRIORITY, i.e. 5).
> It seems that os::get_priority() mistakenly assumes that for native priorities greater value corresponds to higher priority.
> This is not true for niceness, which is used as native priority for linux threads (where less value corresponds to higher priority).
> Thus os::get_priority() incorrectly translates native to java priority.
>
> Following patch fixes the problem (on Linux):
>
> diff -r de2aa86e037d src/share/vm/runtime/os.cpp
> --- a/src/share/vm/runtime/os.cpp       Thu Aug 23 12:27:33 2012 -0700
> +++ b/src/share/vm/runtime/os.cpp       Mon Aug 27 17:52:09 2012 +0300
> @@ -208,7 +208,12 @@
>     OSReturn ret = get_native_priority(thread,&os_prio);
>     if (ret != OS_OK) return ret;
>
> -  for (p = MaxPriority; p>  MinPriority&&  java_to_os_priority[p]>  os_prio; p--) ;
> +  if (java_to_os_priority[MaxPriority]>  java_to_os_priority[MinPriority]) {
> +    for (p = MaxPriority; p>  MinPriority&&  java_to_os_priority[p]>  os_prio; p--) ;
> +  } else {
> +    // niceness values are in reverse order
> +    for (p = MaxPriority; p>  MinPriority&&  java_to_os_priority[p]<  os_prio; p--) ;
> +  }
>     priority = (ThreadPriority)p;
>     return OS_OK;
>   }
>
>
> However jstack is still inaccurate about priorities on Windows. It shows "prio=6" for threads that have NORM_PRIORITY.
> The cause of such inaccuracy is that several java priorities are mapped to the same native priority. (E.g. 5 and 6 are mapped to THREAD_PRIORITY_NORMAL)
> Therefore I believe that instead of trying to figure out java priority by native priority we should better:
> 1. report native priority "as is" for all threads
> 2. report java priority only for java threads using "priority" field.
>
> Propose following patch:
>
> diff -r de2aa86e037d src/share/vm/runtime/thread.cpp
> --- a/src/share/vm/runtime/thread.cpp   Thu Aug 23 12:27:33 2012 -0700
> +++ b/src/share/vm/runtime/thread.cpp   Mon Aug 27 17:52:09 2012 +0300
> @@ -820,7 +820,11 @@
>   void Thread::print_on(outputStream* st) const {
>     // get_priority assumes osthread initialized
>     if (osthread() != NULL) {
> -    st->print("prio=%d tid=" INTPTR_FORMAT " ", get_priority(this), this);
> +    int os_prio;
> +    if (os::get_native_priority(this,&os_prio) == OS_OK) {
> +      st->print("os_prio=%d ", os_prio);
> +    }
> +    st->print("tid=" INTPTR_FORMAT " ", this);
>       osthread()->print_on(st);
>     }
>     debug_only(if (WizardMode) print_owned_locks_on(st);)
> @@ -2710,7 +2714,11 @@
>   void JavaThread::print_on(outputStream *st) const {
>     st->print("\"%s\" ", get_thread_name());
>     oop thread_oop = threadObj();
> -  if (thread_oop != NULL&&  java_lang_Thread::is_daemon(thread_oop))  st->print("daemon ");
> +  if (thread_oop != NULL) {
> +    st->print("#%lld ", java_lang_Thread::thread_id(thread_oop));
> +    if (java_lang_Thread::is_daemon(thread_oop))  st->print("daemon ");
> +    st->print("prio=%d ", java_lang_Thread::priority(thread_oop));
> +  }
>     Thread::print_on(st);
>     // print guess for valid stack memory region (assume 4K pages); helps lock debugging
>     st->print_cr("[" INTPTR_FORMAT "]", (intptr_t)last_Java_sp()&  ~right_n_bits(12));
>
> PS
> just filled bug report for the issue:
> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7194254
>
> Thanks,
> Dmytro
>
>