#538 After running dosemu -t terminal is unsane

[Andrew Bird]

Hi all,
With current devel I find that after running "dosemu -t" the terminal is not working properly. If I type blind 'stty sane' then normal function returns. I'm sure this was working a little while back and I'm wondering if it's fallout from fixing my PIT timing bug as these messages now appear at exit.

ERROR: HLT on detached thread
ERROR: leavedos called recursively, forgetting the graceful exit!

[Stas Sergeev]

Hmm, I can't reproduce.
Is the error msg right on startup, or on exit?
How to reproduce?

[Andrew Bird]

Hi Stas,
It seems to only occur when running a command non interactively i.e. on the dosemu command line

bash-4.2$ bin/dosemu.bin -n --Fimagedir tmp-image -t -I "cpuemu off video{vga}" "dir /w"
ERROR: SYSCALL ERROR: 13, *Permission denied* in file mouseint.c, line 527: expr=
        open(mice->dev, mode)
ERROR: Cannot open internal mouse device /dev/input/mice
ALSA lib rawmidi_hw.c:233:(snd_rawmidi_hw_open) open /dev/snd/midiC0D0 failed: No such file or directory

ERROR: HLT on detached thread
ERROR: leavedos called recursively, forgetting the graceful exit!

If I run the same interactively then the terminal settings are restored correctly.

Hope it helps,

Andrew

[Stas Sergeev]

-D9+hg please

[Andrew Bird]

Hi Stas,
Here's the extra log. I just had a thought though, in autoexec.bat I wrap the
unix -e call with callgrind instrumentation on/off calls. At the moment there's no callgrind running so callgrind_control might return non zero. Would that make a difference?

e.g. in autoexec.bat

prompt $P$G                                                                     
path c:\bin;c:\gnu;c:\dosemu                                                    
unix -s DOSEMU_VERSION                                                          
unix callgrind_control --instr=on                                               
unix -e                                                                         
unix callgrind_control --instr=off

[Andrew Bird]

Just to answer my own question. I commented the 'unix callgrind_control' lines and the terminal insane problem persists.

[Stas Sergeev]

OK, my optimization seems to opened a race
condition. Now I have to re-introduce the
"unoptimized" version of vm86() too. :(
I still can't reproduce, so please try the
attached patch.

[Andrew Bird]

Hi Stas,
I now get an abort. I've attached the new boot_-D9+hg.log

DOSEMU 1.4.0.8-707-g26afabe (2014-06-08), configured: 2014-06-10 14:37:47 +0100
Please test against a recent version before reporting bugs and problems.
Submit Bugs & Patches to linux-msdos@vger.kernel.org or via http://dosemu.org.
FreeDOS kernel build 2036 cvs [version Aug 18 2006 compiled Aug 18 2006]
Kernel compatibility 7.10 - WATCOMC - 80386 CPU required - FAT32 support

(C) Copyright 1995-2006 Pasquale J. Villani and The FreeDOS Project.
All Rights Reserved. This is free software and comes with ABSOLUTELY NO
WARRANTY; you can redistribute it and/or modify it under the terms of the
GNU General Public License as published by the Free Software Foundation;
either version 2, or (at your option) any later version.
C: HD1, Pri[ 1], CHS=    0-1-1, start=     0 MB, size=  2000 MB

FreeCom version 0.84-pre2 XMS_Swap [Aug 28 2006 00:29:00]
C:\>prompt $P$G
C:\>path c:\bin;c:\gnu;c:\dosemu
C:\>unix -s DOSEMU_VERSION
C:\>rem unix callgrind_control --instr=on
C:\>unix -e
About to Execute : dir/w
dosemu.bin: do_vm86.c:445: vm86_helper: Assertion `in_dpmi_dos_int' failed.
Aborted

[Stas Sergeev]

A slightly better one.

[Andrew Bird]

Hi Stas,
That fixes it, although it seems to have a slight performance cost. My test program took 158 secs with current devel and 160 secs after applying the patch.

I've attached the log again

[Stas Sergeev]

Applied, thanks.
No, I don't think this can give a
performance regression. The new helper
only gets called from leavedos().
Maybe some other recent patch did, but
not this one.

[Andrew Bird]

Hi Stas,
Later, when profiling under callgrind, I saw this with the latest devel:

dosemu.bin: coopth.c:794: ensure_single: Assertion `thr->cur_thr == 1' failed.
/home/ajb/run.sh: line 17: 10981 Killed                  /opt/valgrind/bin/valgrind --tool=callgrind --log-file=run.log --dump-every-bb=10000000 --instr-atstart=no --read-var-info=yes --separate-callers=5 --separate-recs=10 -v bin/dosemu.bin -n --Fimagedir tmp-image -t -I "cpuemu off video{vga} dpmi off" "test 1"

I'm not sure if it's reproduceable whilst the dosemu logging is running

[Stas Sergeev]

Could you please reproduce the problem
with the attached patch?
Hmm, this coopthreads are becoming nasty.

[Andrew Bird]

Hi Stas,
A shorter test exited with thr->cur_thr=2. I attached the log also

[Stas Sergeev]

-D9+hg please.

[Andrew Bird]

Hi Stas,
Here you go.

[Stas Sergeev]

Please use the attached logging patch.

[Andrew Bird]

Hi Stas,
Here's a thought. Are you sure the thread ID should always be 1? I was playing with helgrind on dosemu a couple of days ago and I seem to remember that the threads were initialised in different orders sometimes?

[Stas Sergeev]

This is not an ID, this is a recursion
counter. For some reason you have recursion
level of inte6 = 2. This is likely a bug
somewhere. I don't think inte6 should ever
re-enter.

[Andrew Bird]

Stas,
Here's the new log.

[Stas Sergeev]

Thanks.
I added a bit more logging to git.
Please pull and re-do the log
Together with the patch that was here.
But please add a missing \n to the
a.diff that I forgot to add.

[Andrew Bird]

Hi Stas,
Here's the new log after pull

[Stas Sergeev]

Should now be fixed in git.
There are still coopth-unsafe places around,
perhaps. Only a good testing can locate all
of them, even though nothing have popped up
for quite some time.

[Andrew Bird]

Hi Stas,
That seems to be okay now. I'll let you know if I come across any others.

Many thanks,

Andrew

[Stas Sergeev]

Thanks for testing.
dosemu "dir /p"
or
dosemu exitemu
were still unsafe.
Should now be fixed completely.