Cygwin 3.4.3 and 3.5.0... hangs in make, top, procps, ls /proc/PID/...

[Brian Inglis]

On 2023-01-16 08:18, Corinna Vinschen via Cygwin wrote:
> Hi Takashi,
> 
> On Jan 16 23:45, Takashi Yano via Cygwin wrote:
>> On Mon, 16 Jan 2023 11:23:54 +0100
>> Corinna Vinschen wrote:
>>> On Jan 16 18:02, Takashi Yano via Cygwin wrote:
>>> [...]
>>>> Errors seem to be three types: (null), cygpid.xxx and shared.5.
>>>> I'm not sure what is happening and why at all, however, this
>>>> did not seem to happen before the commit 30add3e6b3e3.
>>>
>>> I'll try to reproduce this issue.  But the weird thing is certainly
>>> this: The affected shared mem regions are apparently not the cygheap.
>>> Rather, they are the "shared" and "cygpid" shared mem regions, which
>>> should not at all collide with the cygheap.  I guess we need more
>>> debug output in the api_fatal call inside open_shared...
>>
>> I am now trying the test case with reverting the commit 60675f1a7eb2
>> "Cygwin: decouple shared mem regions from Cygwin DLL", and the issue
>> does not happen for several hours so far. I guess this most likely
>> is the direct cause of the problem.
> 
> Actually, I' running your testcase on two machines in parallel now for
> quite some time, which only one hunk of 60675f1a7eb2 reverted, i.e.
> 
> diff --git a/winsup/cygwin/mm/shared.cc b/winsup/cygwin/mm/shared.cc
> index 893b20d289b4..6d6d2940b6d4 100644
> --- a/winsup/cygwin/mm/shared.cc
> +++ b/winsup/cygwin/mm/shared.cc
> @@ -140,7 +140,11 @@ open_shared (const WCHAR *name, int n, HANDLE& shared_h, DWORD size,
>     if (*m == SH_JUSTCREATE || *m == SH_JUSTOPEN)
>       addr = NULL;
>     else
> -    addr = (void *) region_address[*m];
> +    {
> +      addr = (void *) region_address[*m];
> +      VirtualFree (addr, 0, MEM_RELEASE);
> +    }
> +
>   
>     WCHAR map_buf[MAX_PATH];
>     WCHAR *mapname = NULL;
> 
> So far (and knocking on wood madly while writing this) the mapping
> problem didn't show up once.  Maybe you'd like to try the same?

After struggling to build and release a new ncurses test package yesterday, with 
many hangs including unrelated hourly cron jobs, make, top, ps, and using Cygwin 
/bin/ps and /bin/kill -f combos a lot to make progress, I upgraded last night to 
latest including *cygwin test 3.5.0-0.107.g8bd56ec87345*, and an ncurses 
rebuild, and another package rebuild, with top monitoring under X and usual 
hourly cron jobs running had no issues for the first time in weeks!

I also have a couple of cygcheck -cd cygwin loops running, and damn, *just* got 
a failure on one, and stackdump is zero length:

$       0 [main] cygcheck (16228) C:\usr\local\cygwin64\bin\cygcheck.exe: *** 
fatal error -
  MapViewOfFileEx '(null)'(0x108), Win32 error 487.  Terminating.
    1260 [main] cygcheck (16228) cygwin_exception::open_stackdumpfile: Dumping 
stack trace to cygcheck.exe.stackdump
$ ls -glo *.stackdump
-rwxr-xr-x 1 0 Jan 16 12:16 cygcheck.exe.stackdump
-- 
Take care. Thanks, Brian Inglis			Calgary, Alberta, Canada

La perfection est atteinte			Perfection is achieved
non pas lorsqu'il n'y a plus rien à ajouter	not when there is no more to add
mais lorsqu'il n'y a plus rien à retirer	but when there is no more to cut
			-- Antoine de Saint-Exupéry