Re: Random "child_info_fork::abort:"

[Evgeny Grin <k2k at yandex dot ru>]

On 13.10.2016 18:24, Evgeny Grin wrote:
> On 13.10.2016 10:11, Evgeny Grin wrote:
>> On 12.10.2016 21:45, Evgeny Grin wrote:
>>> On 12.10.2016 13:59, Evgeny Grin wrote:
>>>> On 12.10.2016 8:59, Brian Inglis wrote:
>>>>> On 2016-10-11 15:32, Evgeny Grin wrote:
>>>>>> I'm using Windows Insider (slow ring, prerelease). After recent update
>>>>>> to build 14931, cygwin keeps randomly fail on fork. This happens not
>>>>>> every fork, but frequent enough. Simplest way to trigger it is to run
>>>>>> mandb.
>>>>>> At variable delay I got something like
>>>>>> child_info_fork::abort: T:\cygwin64\bin\cygman-2-7-5.dll: Loaded to
>>>>>> different address: parent(0x3FEAB0000) != child(0x1C0000)
>>>>>> Dll name and child address may vary.
>>>>>> I updated cygwin to latest version, but problem remains.
>>>>>> Exactly the same problem with Msys2, but additionally from time to time
>>>>>> I got different error:
>>>>>> *** fatal error - cygheap base mismatch detected - 0x1802FE408/0x106E408.
>>>>>> This problem is probably due to....
>>>>>
>>>>>> I set variable CYGWIN=detect_bloda, but nothing is detected.
>>>>>> What else could I check?
>>>>>
>>>>> Run rebase -is to check for DLL conflicts.
>>>>> After the update did you run "rebase-trigger full" and setup,
>>>>> with NO Cygwin processes running, to remap the system DLLs
>>>>> and rebase the Cygwin DLLs?
>>>>
>>>> Done. No conflicts.
>>>> Setup run fine, problem is not solved.
>>>>
>>>> Observation: Cygwin errors appears more often than Msys2 errors.
>>>> Even got initial loading on incorrect address:
>>>>
>>>> child_info_fork::abort: T:\cygwin64\bin\cygman-2-7-5.dll: Loaded to
>>>> different address: parent(0x190000) != child(0x3FB700000)
>>>>
>>>>> It's likely the Insider debug builds dynamically enable/disable
>>>>> features and functions or run alternate system DLLs which gather
>>>>> info by acting as BLODAs.
>>>>> MS can mess around with your systems to enable new stuff (possibly
>>>>> in different combinations) and see which systems they cause problems
>>>>> on.
>>>>> Hopefully they can also dynamically revert new releases causing
>>>>> problems.
>>>>> Your systems are the canaries for their Continuous Delivery QA.
>>>>>
>>>>> Make sure you continously back up any work on those systems and
>>>>> don't ignore warnings, especially build expiries (you are meant
>>>>> to be allowed 3 reboots after expiry before the buildrefuses
>>>>> to boot!)
>>>>
>>>> Thanks!
>>>> Not sure that situation is just "test" or "temporary". I'm using
>>>> prerelease track so this can go to release.
>>>
>>> Additionally tested under VM.
>>> Clear install of Windows 10 x64 14931 (US English), VM guest tools and
>>> Cygwin x64. First run of man-db - same error.
>>> Need new workarounds?
>>> Seems that will be in all new builds.
>>
>> Same happens with new Windows build 14942.
>> Both Cygwin and Msys2 is not usable anymore.
>>
>> MS is pushing us to "Ubuntu on Windows"?
>>
> 
> When trying to actively use cygwin, got same (as in Msys2) errors:
> 
>  *** fatal error - cygheap base mismatch detected - 0x180302408/0x1082408.
> 
> Is it because of cygwin.dll loaded on incorrect address?
> 

What I got from strace (typical output):

 649705 [main] mandb 6288 child_info::sync: n 2, waiting for
subproc_ready(0x280) and child process(0x1CC)
   6358 [main] mandb 7888 child_info_fork::abort:
T:\cygwin64\bin\cyggdbm-4.dll: Loaded to different address:
parent(0x1FFFC510000) != child(0x150000)
--- Process 7888 created
--- Process 7888 loaded C:\Windows\System32\ntdll.dll at 00007FFE2FF70000
--- Process 7888 loaded C:\Windows\System32\kernel32.dll at 00007FFE2D600000
--- Process 7888 loaded C:\Windows\System32\KernelBase.dll at
00007FFE2C730000
--- Process 7888 thread 912 created
--- Process 7888 thread 10792 created
--- Process 7888 loaded T:\cygwin64\bin\cygmandb-2-7-5.dll at
000001FFFB8F0000
--- Process 7888 loaded T:\cygwin64\bin\cygwin1.dll at 0000000180040000
--- Process 7888 loaded T:\cygwin64\bin\cygintl-8.dll at 00000003FBB90000
--- Process 7888 thread 14024 created
--- Process 7888 loaded T:\cygwin64\bin\cygiconv-2.dll at 00000003FBBF0000
--- Process 7888 loaded T:\cygwin64\bin\cygpipeline-1.dll at
000001FFFB470000
--- Process 7888 loaded T:\cygwin64\bin\cygman-2-7-5.dll at 000001FFFB900000
--- Process 7888 loaded T:\cygwin64\bin\cyggdbm-4.dll at 0000000000150000
--- Process 7888 loaded T:\cygwin64\bin\cygman-2-7-5.dll at 0000000000170000
--- Process 7888 unloaded DLL at 0000000000170000
--- Process 7888 loaded T:\cygwin64\bin\cygz.dll at 000001FFFACF0000
--- Process 7888 loaded T:\cygwin64\bin\cyggdbm-4.dll at 000001FFFC510000
--- Process 7888 unloaded DLL at 000001FFFC510000
   6358 [main] mandb 7888 child_info_fork::abort:
T:\cygwin64\bin\cyggdbm-4.dll: Loaded to different address:
parent(0x1FFFC510000) != child(0x150000)
--- Process 7888 thread 14024 exited with status 0x800000
--- Process 7888 thread 10144 exited with status 0x800000
--- Process 7888 thread 10792 exited with status 0x800000
--- Process 7888 exited with status 0x800000
 663649 [main] mandb 6288 child_info::sync: pid 7888, WFMO returned 1,
exit_code 0x800000, res 0
 663743 [main] mandb 6288 sig_send: sendsig 0xD8, pid 6288, signal -73,
its_me 1
 663771 [main] mandb 6288 sig_send: wakeup 0x2EC
 663800 [main] mandb 6288 sig_send: Waiting for pack.wakeup 0x2EC
 663840 [sig] mandb 6288 wait_sig: signalling pack.wakeup 0x2EC
 663886 [main] mandb 6288 sig_send: returning 0x0 from sending signal -73
mandb: fork failed: Resource temporarily unavailable


The cyggdbm-4.dll is loaded two times: at 0x150000 and at 0x1FFFC510000,
but for some reason unloaded from 0x1FFFC510000.
This looks like a bug in cygwin.

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple