[gdal-dev] core dump on dir info

Sun Feb 4 23:44:08 PST 2024

Hi Mike

Out of curiosity, are run running it in a virtual machine?
A few year ago I had problems running a program in a virtual machine
(virtualbox, but I read it happens in others) due to a missing SSE
instruction. The solution there was to "enable" the missing instructions in
the virtual machine configuration (that I don't know why it was not the
default).

Cheers

On Mon, 5 Feb 2024, 02:04 Even Rouault via gdal-dev, <
gdal-dev at lists.osgeo.org> wrote:

> ghcr.io/osgeo/gdal:ubuntu-full-latest has been regenerated with the
> rebuild of TileDB without AVX2. I've also enabled the
> drivers-with-external-depencies-built-as-plugin GDAL build mode, so it is
> easy to just remove a given plugin by deleting the corresponding .so in
> /usr/lib/x86_64-linux-gnu/gdalplugins
>
> Even
> Le 04/02/2024 à 22:51, Michael Sumner a écrit :
>
> indeed there's no avx2:
>
> cat /proc/cpuinfo|grep sse|head -n 1
> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
> cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt pdpe1gb
> rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq
> ssse3 fma cx16 sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx
> f16c hypervisor lahf_lm cmp_legacy svm cr8_legacy abm sse4a misalignsse
> 3dnowprefetch osvw xop fma4 tbm perfctr_core ssbd ibpb vmmcall tsc_adjust
> bmi1 virt_ssbd arat npt nrip_save arch_capabilities
>
> Cheers, Mike
>
>
>
> On Sun, Feb 4, 2024 at 10:55 PM Even Rouault <even.rouault at spatialys.com>
> wrote:
>
>> ok, so I believe this is the AVX2 issue I was talking about, as I realize
>> that enabling AVX2 is the default mode when TileDB is built from source
>> (which the Docker image does), and must be explicitly disabled with
>> "./bootstrap --disable-avx2" (I've just changed the build recipe to include
>> that, will take effect next time the images are refreshed)
>>
>> To confirm, can you send or just check the output of : cat
>> /proc/cpuinfo|grep sse|head -n 1
>>
>> If there is no "avx2" in it, this is at 99.9% the reason of the issue.
>>
>> Even
>> Le 04/02/2024 à 06:20, Michael Sumner a écrit :
>>
>> skipping TileDB does fix:
>>
>> ogr2ogr /tmp/newdir
>> https://github.com/SymbolixAU/geojsonsf/raw/master/inst/examples/geo_melbourne.geojson -f
>> "ESRI Shapefile"
>> export GDAL_SKIP=TileDB
>> ogrinfo /tmp/newdir/
>> INFO: Open of `/tmp/newdir/'
>>       using driver `ESRI Shapefile' successful.
>> 1: geo_melbourne (Polygon)
>>
>> unset GDAL_SKIP
>> ogrinfo /tmp/newdir/
>> Illegal instruction (core dumped)
>>
>> I failed to explain that I'm using gdal containers from the repo:
>>
>> docker run --rm -ti ghcr.io/osgeo/gdal:ubunt
>>
>> u-full-latest
>>
>> apt update
>> apt install -y gdb
>>
>> Here's the output of under gdb as you suggested, there was a lot so I put
>> it on a gist:
>> https://gist.github.com/mdsumner/839ae6e05ededf640b65bfee3a20a4c0
>>
>> gdb --args ogrinfo /tmp/newdir/
>> > run
>> > thread apply all bt
>>
>> Thanks!
>>
>>
>>
>>
>>
>> On Sat, Feb 3, 2024 at 7:49 PM Even Rouault <even.rouault at spatialys.com>
>> wrote:
>>
>>> - When it crashes under gdb, type "thread apply all bt" to get the stack
>>> trace of all threads
>>>
>>> - I suspect there is a connection with
>>> https://github.com/OSGeo/gdal/pull/9170 , but that pull request
>>> wouldn't help here as "/tmp/newdir" could be a valid connection to TileDB
>>>
>>> - how did you get TileDB installed? It looks to be packaged? Which
>>> distribution do you use?
>>>
>>> - SIGILL reminds me of issues with some TileDB builds using the AVX2
>>> instruction set by default, which could cause some crash on host CPUs that
>>> don't have AVX2 (unlikely on recent hardware though)
>>>
>>> - Setting GDAL_SKIP=TileDB should be a workaround
>>>
>>>
>>> Le 03/02/2024 à 07:15, Michael Sumner a écrit :
>>>
>>> Thanks Even, so there's something about tiledb under gdb (or maybe I am
>>> mangling the context,  I will try variants of the host I'm using). Run with
>>> valgrind included below.
>>>
>>> gdb --args ogrinfo /tmp/newdir/
>>> ...
>>> (gdb) run
>>> Starting program: /usr/local/bin/ogrinfo /tmp/newdir/
>>> [Thread debugging using libthread_db enabled]
>>> Using host libthread_db library
>>> "/lib/x86_64-linux-gnu/libthread_db.so.1".
>>> [New Thread 0x7fffe7757640 (LWP 988)]
>>> [New Thread 0x7fffe6f56640 (LWP 989)]
>>> [New Thread 0x7fffde755640 (LWP 990)]
>>> [New Thread 0x7fffd5f54640 (LWP 991)]
>>> [New Thread 0x7fffc5753640 (LWP 992)]
>>> [New Thread 0x7fffc4f52640 (LWP 993)]
>>> [New Thread 0x7fffb4751640 (LWP 994)]
>>> [New Thread 0x7fffabf50640 (LWP 995)]
>>> [New Thread 0x7fffab74f640 (LWP 996)]
>>> [New Thread 0x7fffa2f4e640 (LWP 997)]
>>> [New Thread 0x7fff9a74d640 (LWP 998)]
>>> [New Thread 0x7fff91f4c640 (LWP 999)]
>>> [New Thread 0x7fff8974b640 (LWP 1000)]
>>> [New Thread 0x7fff78f4a640 (LWP 1001)]
>>> [New Thread 0x7fff78749640 (LWP 1002)]
>>> [New Thread 0x7fff6f5ff640 (LWP 1003)]
>>>
>>> Thread 1 "ogrinfo" received signal SIGILL, Illegal instruction.
>>> 0x00007ffff3773c9e in tiledb::common::ThreadPool::ThreadPool(unsigned
>>> long) () from /lib/x86_64-linux-gnu/libtiledb.so.2.16
>>>
>>>
>>>
>>>
>>> valgrind -s ogrinfo /tmp/newdir
>>> ==704== Memcheck, a memory error detector
>>> ==704== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
>>> ==704== Using Valgrind-3.18.1 and LibVEX; rerun with -h for copyright
>>> info
>>> ==704== Command: ogrinfo /tmp/newdir
>>> ==704==
>>> INFO: Open of `/tmp/newdir'
>>>       using driver `ESRI Shapefile' successful.
>>> 1: geo_melbourne (Polygon)
>>> ==704==
>>> ==704== HEAP SUMMARY:
>>> ==704==     in use at exit: 25,486 bytes in 216 blocks
>>> ==704==   total heap usage: 15,761 allocs, 15,545 frees, 2,390,169 bytes
>>> allocated
>>> ==704==
>>> ==704== LEAK SUMMARY:
>>> ==704==    definitely lost: 0 bytes in 0 blocks
>>> ==704==    indirectly lost: 0 bytes in 0 blocks
>>> ==704==      possibly lost: 544 bytes in 1 blocks
>>> ==704==    still reachable: 24,942 bytes in 215 blocks
>>> ==704==         suppressed: 0 bytes in 0 blocks
>>> ==704== Rerun with --leak-check=full to see details of leaked memory
>>> ==704==
>>> ==704== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
>>>
>>>
>>> ogrinfo /tmp/newdir
>>> Illegal instruction (core dumped)
>>>
>>> Cheers, Mike
>>>
>>>
>>>
>>>
>>> On Sat, Feb 3, 2024 at 12:46 PM Even Rouault <even.rouault at spatialys.com>
>>> wrote:
>>>
>>>> Michael,
>>>>
>>>> I'm wondering if there not might be something wrong with your build or
>>>> runtime environment. Or there's something subtle, because that works fine
>>>> for me with my dev build or in the
>>>> ghcr.io/osgeo/gdal:alpine-normal-3.8.3 Docker image
>>>>
>>>> Try running "valgrind ogrinfo /tmp/newdir/" or "gdb --args ogrinfo
>>>> /tmp/newdir/" (type "run") to get more useful information
>>>>
>>>> Even
>>>> Le 03/02/2024 à 02:35, Michael Sumner via gdal-dev a écrit :
>>>>
>>>> I'm getting Illegal instruction / core dumped on ogrinfo of a
>>>> directory:
>>>>
>>>> ogr2ogr /tmp/newdir
>>>> https://github.com/SymbolixAU/geojsonsf/raw/master/inst/examples/geo_melbourne.geojson
>>>> -f "ESRI Shapefile"
>>>>
>>>> ogrinfo /tmp/newdir/
>>>> Illegal instruction (core dumped)
>>>>
>>>> I've worked back through some docker images and it wasn't a problem in
>>>> 3.6.0, but I'm getting it since 3.7.0 - or I'm doing something wrong
>>>> entirely.
>>>>
>>>> Cheers, Mike
>>>>
>>>>
>>>> --
>>>> Michael Sumner
>>>> Software and Database Engineer
>>>> Australian Antarctic Division
>>>> Hobart, Australia
>>>> e-mail: mdsumner at gmail.com
>>>>
>>>> _______________________________________________
>>>> gdal-dev mailing listgdal-dev at lists.osgeo.orghttps://lists.osgeo.org/mailman/listinfo/gdal-dev
>>>>
>>>> -- http://www.spatialys.com
>>>> My software is free, but my time generally not.
>>>>
>>>>
>>>
>>> --
>>> Michael Sumner
>>> Software and Database Engineer
>>> Australian Antarctic Division
>>> Hobart, Australia
>>> e-mail: mdsumner at gmail.com
>>>
>>> -- http://www.spatialys.com
>>> My software is free, but my time generally not.
>>>
>>>
>>
>> --
>> Michael Sumner
>> Software and Database Engineer
>> Australian Antarctic Division
>> Hobart, Australia
>> e-mail: mdsumner at gmail.com
>>
>> -- http://www.spatialys.com
>> My software is free, but my time generally not.
>>
>>
>
> --
> Michael Sumner
> Software and Database Engineer
> Australian Antarctic Division
> Hobart, Australia
> e-mail: mdsumner at gmail.com
>
> -- http://www.spatialys.com
> My software is free, but my time generally not.
>
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/gdal-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20240205/9a888f3b/attachment-0001.htm>