[Starpu-devel] segfault in StarPU memory allocator

[Alfredo Buttari <alfredo.buttari@enseeiht.fr>]

Dear StarPU developers/users,
I need some help to debug a code which segfaults in a worker when the
starpu_shutdown() routine is called with the following backtrace:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffe0b52700 (LWP 166825)]
starpu_free_on_node_flags (dst_node=1, addr=140735338284336, size=0,
flags=6) at datawizard/malloc.c:1061
1061 int block = ((addr - chunk->base) / CHUNK_ALLOC_MIN) + 1,
prevblock, nextblock;
Missing separate debuginfos, use: debuginfo-install
glibc-2.17-196.el7_4.2.x86_64 libxml2-2.9.1-6.el7_2.3.x86_64
numactl-libs-2.0.9-6.el7_2.x86_64 xz-libs-5.1.2-12alpha.el7.x86_64
zlib-1.2.7-15.el7.x86_64
(gdb) backtrace
#0  starpu_free_on_node_flags (dst_node=1, addr=140735338284336,
size=0, flags=6) at datawizard/malloc.c:1061
#1  0x00007ffff74810f3 in free_memory_on_node (mc=<optimized out>,
node=<optimized out>) at datawizard/memalloc.c:390
#2  flush_memchunk_cache (node=1, reclaim=140735338284336) at
datawizard/memalloc.c:845
#3  0x00007ffff74811c6 in _starpu_memory_reclaim_generic (node=1,
force=2144896304, reclaim=0) at datawizard/memalloc.c:971
#4  0x00007ffff74b3a86 in _starpu_cuda_driver_deinit (worker_set=0x1)
at drivers/cuda/driver_cuda.c:1027
#5  0x00007ffff74b27df in _starpu_cuda_worker (_arg=0x1) at
drivers/cuda/driver_cuda.c:1070
#6  0x00007ffff3883e25 in start_thread () from /lib64/libpthread.so.0
#7  0x00007fffef88f34d in clone () from /lib64/libc.so.6


Unfortunately valgrind does not like CUDA code:

==1292== Memcheck, a memory error detector
==1292== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==1292== Using Valgrind-3.10.0 and LibVEX; rerun with -h for copyright info
==1292== Command: ./dqrm_test
==1292==
vex amd64->IR: unhandled instruction bytes: 0xF 0xAE 0x64 0x24 0x40
0x48 0x8B 0x73
vex amd64->IR:   REX=0 REX.W=0 REX.R=0 REX.X=0 REX.B=0
vex amd64->IR:   VEX=0 VEX.L=0 VEX.nVVVV=0x0 ESC=0F
vex amd64->IR:   PFX.66=0 PFX.F2=0 PFX.F3=0
==1292== valgrind: Unrecognised instruction at address 0x4015f48.
==1292==    at 0x4015F48: _dl_runtime_resolve_xsave (in /usr/lib64/ld-2.17.so)
[cut]


and cuda-memcheck reports no errors.

Do you have any idea on how to proceed?



ciao
Alfredo




-----------------------------------------
Alfredo Buttari, PhD
CNRS-IRIT
2 rue Camichel, 31071 Toulouse, France
http://buttari.perso.enseeiht.fr