Re: [PATCH 0/5] linux-user/s390x: Fix psw.mask handling in signals

[Alex Bennée]

Alex Bennée <alex.bennee@linaro.org> writes:

> Richard Henderson <richard.henderson@linaro.org> writes:
>
>> The PSW_MASK_CC component of psw.mask was not handled properly
>> in the creation or restoration of signal frames.
>
> Still seeing issues running on s390x machine:
>
>   05:00:29 [ajb@qemu01:~/l/q/b/debug] s390x/signal-fixes|… 38 + retry.py -n 
> 100 -c -- ./qemu-s390x ./tests/tcg/s390x-linux-user/signals
>   ...
>   ...
>   Results summary:
>   0: 62 times (62.00%), avg time 2.253 (0.00 varience/0.00 deviation)
>   -11: 38 times (38.00%), avg time 0.251 (0.00 varience/0.00 deviation)
>   Ran command 100 times, 62 passes
>
> I don't get much from the backtrace, maybe the atomic triggered the seg?
>
>   #0  0x0000000001016244 in vfprintf ()
>   [Current thread is 1 (Thread 0x4001001910 (LWP 27308))]
>   (gdb) bt
>   #0  0x0000000001016244 in vfprintf ()
>   #1  0x000000000101d484 in printf ()
>   #2  0x0000000001000b2e in background_thread_func (arg=0x10a3620) at 
> /home/ajb/lsrc/qemu.git/tests/tcg/multiarch/signals.c:65
>   #3  0x0000000001003150 in start_thread (arg=0x4001001910) at 
> pthread_create.c:463
>   #4  0x0000000001035b40 in thread_start ()
>   (gdb) frame 2
>   #2  0x0000000001000b2e in background_thread_func (arg=0x10a3620) at 
> /home/ajb/lsrc/qemu.git/tests/tcg/multiarch/signals.c:65
>   65          printf("thread%d: started\n", job->number);
>   (gdb) info locals
>   job = 0x10a3620
>   (gdb) p job
>   $1 = (ThreadJob *) 0x10a3620
>   (gdb) p job->number
>   $2 = 0
>   (gdb) frame 0
>   #0  0x0000000001016244 in vfprintf ()
>   (gdb) x/5i $pc
>   => 0x1016244 <vfprintf+1316>:   asi     224(%r11),1
>      0x101624a <vfprintf+1322>:   cgijne  %r1,0,0x1017570 <vfprintf+6224>
>      0x1016250 <vfprintf+1328>:   lg      %r1,336(%r11)
>      0x1016256 <vfprintf+1334>:   lghi    %r3,37
>      0x101625a <vfprintf+1338>:   aghik   %r6,%r1,1
>   (gdb) p/x $r11
>   $3 = 0x4001000708
>   (gdb) p/x $r11 + 224
>   $4 = 0x40010007e8
>   (gdb) x/1g $4
>   0x40010007e8:   0x0000000000000000
>   (gdb)
>
> However running on x86 backend everything seems to be fine.
>
>   Results summary:
>   0: 200 times (100.00%), avg time 2.255 (0.00 varience/0.00 deviation)
>   Ran command 200 times, 200 passes

It's hard to desegregate the SEGVs we get during normal runs but a pass
followed by a fail shows:

  qemu: SIGSEGV pc=0x3fff473bb1e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff473bb1e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff473bb1e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff473bb1e address=40007ff000 w=1 oldset=0x00000000
  shutting down after: 2000 signals
  qemu: SIGSEGV pc=0x3fff484bd1e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff4848f1a address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff47477f2 address=40007ff000 w=1 oldset=0x00000000
  [Thread 0x3fffdff1780 (LWP 32928) exited]
  [Inferior 1 (process 32928) exited normally]
  (gdb) r
  Starting program: /home/ajb/lsrc/qemu.git/builds/debug/qemu-s390x 
./tests/tcg/s390x-linux-user/signals
  [Thread debugging using libthread_db enabled]
  Using host libthread_db library "/lib/s390x-linux-gnu/libthread_db.so.1".
  [New Thread 0x3fffceff910 (LWP 32964)]
  qemu: SIGSEGV pc=0x3fff4752f2a address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff471fa1e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff475c49c address=40007ff000 w=1 oldset=0x00000000
  [New Thread 0x3fffdff0910 (LWP 32965)]
  qemu: SIGSEGV pc=0x3fff4703b18 address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fffd812efe address=4001000000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff471081e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff4715ee6 address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff471a02a address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff472491e address=40007ff000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff4725d1e address=4001000000 w=1 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff4730536 address=4001000000 w=0 oldset=0x00000000
  qemu: SIGSEGV pc=0x3fff473171e address=40007ff000 w=1 oldset=0x00000000
  [Thread 0x3fffdff0910 (LWP 32965) exited]
  [Thread 0x3fffceff910 (LWP 32964) exited]

  Program terminated with signal SIGSEGV, Segmentation fault.
  The program no longer exists.
  (gdb)

So something is going astray to 4001000000 which otherwise doesn't.

>
>>
>>
>> r~
>>
>>
>> Richard Henderson (5):
>>   target/s390x: Expose load_psw and get_psw_mask to cpu.h
>>   target/s390x: Do not modify cpu state in s390_cpu_get_psw_mask
>>   target/s390x: Improve s390_cpu_dump_state vs cc_op
>>   target/s390x: Use s390_cpu_{set_psw,get_psw_mask} in gdbstub
>>   linux-user/s390x: Save and restore psw.mask properly
>>
>>  target/s390x/cpu.h         |   3 ++
>>  target/s390x/internal.h    |   5 --
>>  linux-user/s390x/signal.c  |  37 ++++++++++++--
>>  target/s390x/cc_helper.c   |   2 +-
>>  target/s390x/excp_helper.c |  28 +++++-----
>>  target/s390x/gdbstub.c     |  15 +-----
>>  target/s390x/helper.c      | 101 ++++++++++++++++++++-----------------
>>  target/s390x/sigp.c        |   3 +-
>>  8 files changed, 110 insertions(+), 84 deletions(-)

-- 
Alex Bennée