RFR(XS): 8204652: [aix] fix thread stack allocation, too big stack sizes not recognized

[David Holmes]

On 11/06/2018 6:07 PM, Lindenmaier, Goetz wrote:
> Hi,
> 
> Yes, reporting the pthread_create is a bit misleading. But it is the same
> as reported on linux x86_64:
> 
> bin/java -XX:CompilerThreadStackSize=8000000000000
> [0.271s][warning][os,thread] Failed to start thread - pthread_create failed (EAGAIN) for attributes: stacksize: 8000000000000k, guardsize: 0k, detached.
> Error occurred during initialization of VM
> java.lang.OutOfMemoryError: unable to create native thread: possibly out of memory or process/resource limits reached

But on linux you _do_ call pthread_create and it did fail!

> As we know linux is bad at implementing the pthread functionality here:
> It accepts setting the big value as stack size, and only when you call create
> you get the failure.

Does POSIX require otherwise?

> So it's not true that the VM starts with any other value, it also crashes on
> linux.  And you get the same output as on linux. So I realy don't want to
> start implementing any clever tracing, or even implement other stack
> handling (setting defaults) as on linux.  I would just like to keep it as is.
> Currently we get a SIGSEGV which is hard to debug on aix.

The point is the error should be reported accurately in the logging 
information. It's not coming from pthread_create in your AIX case so 
don't report that it is.

David

> Best,
>    Goetz
> 
>> -----Original Message-----
>> From: Thomas Stüfe [mailto:thomas.stuefe at gmail.com]
>> Sent: Montag, 11. Juni 2018 06:26
>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
>> Cc: hotspot-runtime-dev at openjdk.java.net
>> Subject: Re: RFR(XS): 8204652: [aix] fix thread stack allocation, too big stack
>> sizes not recognized
>>
>> Hi Goetz,
>>
>> On Sun, Jun 10, 2018 at 9:11 PM, Lindenmaier, Goetz
>> <goetz.lindenmaier at sap.com> wrote:
>>> Hi Thomas,
>>>
>>> The stack is started with 192k.  See the output:
>>>
>>
>> Ah you are right. With 64K pages this would not be enough to establish
>> red/yellow zones and still have breathing space left.
>>
>>> bash-4.4$ ~/oJ/vms/aix_j-1/bin/java -
>> XX:CompilerThreadStackSize=8000000000000
>>> [1.482s][warning][os,thread] The compiler thread stack size specified is
>> invalid: 8000000000000k
>>> [1.483s][warning][os,thread] Failed to start thread - pthread_create failed
>> (22=EINVAL) for attributes: stacksize: 192k, guardsize: 4k, detached.
>>> Error occurred during initialization of VM
>>> java.lang.OutOfMemoryError: unable to create native thread: possibly out
>> of memory or process/resource limits reached
>>>
>>> I think the output is not that bad ...
>>
>> See my mail to David: I think the output is misleading, since
>> pthread_create() did not even run. So, "pthread_create failed" is
>> wrong.
>>
>> Best Regards, Thomas
>>
>>>
>>> Best,
>>>    Goetz.
>>>
>>>> -----Original Message-----
>>>> From: Thomas Stüfe <thomas.stuefe at gmail.com>
>>>> Sent: Sunday, June 10, 2018 7:33 PM
>>>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
>>>> Cc: hotspot-runtime-dev at openjdk.java.net
>>>> Subject: Re: RFR(XS): 8204652: [aix] fix thread stack allocation, too big
>> stack
>>>> sizes not recognized
>>>>
>>>> Hi Goetz,
>>>>
>>>> I see what you intent to do, but I think we need to think about error
>>>> handling a bit more. The pattern here (which was broken you try to fix
>>>> it with this patch) is that we try, sequentially:
>>>>
>>>> - pthread_attr_setstacksize()
>>>>
>>>> - if ok-so-far pthread_attr_setguardsize
>>>>
>>>> - if ok-so-far pthread_ceate
>>>>
>>>> - logging and error handling
>>>>
>>>>
>>>> Problem with that is that in the end, we log information about a
>>>> failing thread creation, printing errno as if pthread_create() failed,
>>>> but that errno may have come actually from one of the preceeding OS
>>>> calls. So the error message would be misleading.
>>>>
>>>> I think it would be better to handle errors of each system call explicitly:
>>>>
>>>> pthread_attr_setstacksize()
>>>> if failed
>>>>     - log out "pthread_attr_setsize failed" and errno
>>>>     - [ also: optionally decide what to do: failing thread creation?
>>>> retry with a default stack size? ]
>>>>     - return false
>>>>
>>>> pthread_attr_setguardsize()
>>>> if failed
>>>>     - log out "pthread_attr_setguardsize failed" and errno
>>>>     - decide what to do - ignore? assert? fail thread creation? Should
>>>> we even care?
>>>>     - return false
>>>>
>>>> and then call pthread_create with the existing error handling as follow up.
>>>>
>>>> What do you think?`
>>>>
>>>> (Also, if pthread_attr_setstacksize() fails, should the thread not be
>>>> created with the pthread default stack size, which should be ~2MB? Why
>>>> is that stack tiny?)
>>>>
>>>> Best Regards, Thomas
>>>>
>>>>
>>>> On Sun, Jun 10, 2018 at 5:33 PM, Lindenmaier, Goetz
>>>> <goetz.lindenmaier at sap.com> wrote:
>>>>> Hi,
>>>>>
>>>>> This fixes a crash in thread creation with too big stack sizes
>>>>> http://cr.openjdk.java.net/~goetz/wr18/8204652-
>> aixHugeStacksCrash/01/
>>>>> Please review.
>>>>>
>>>>> For Java and Compiler threads the return code of
>>>> pthread_attr_setstacksize
>>>>>   is overwritten. If pthread_attr_setstacksize does not succeed, a tiny
>>>>>   stack size with not enough space for the yellow/red pages is allocated.
>>>>>   VM crashes with SIGSEGV right away.
>>>>>
>>>>> Best regards,
>>>>>    Goetz