RFR: 8226238: Improve error output and fix elf issues in os::dll_load

[Langer, Christoph]

Hi Matthias,

your change looks good overall.

I only have a few style nits:

src/hotspot/os/linux/os_linux.cpp, line 1751 (new):

Can you convert

unsigned char current_endianness = ELFDATA2MSB; // BE
#if defined(VM_LITTLE_ENDIAN)
  current_endianness = ELFDATA2LSB; // LE
#endif

to

#if defined(VM_LITTLE_ENDIAN)
  unsigned char current_endianness = ELFDATA2LSB; // LE
#else
  unsigned char current_endianness = ELFDATA2MSB; // BE
#endif

And the same in line 1611 of src/hotspot/os/solaris/os_solaris.cpp.

src/hotspot/os/linux/os_linux.cpp, line 1802: you could fix the indentation of ELFDATA2LSB for EM_ARM
same for line 1580 of src/hotspot/os/solaris/os_solaris.cpp.

Best regards
Christoph

> -----Original Message-----
> From: hotspot-dev <hotspot-dev-bounces at openjdk.java.net> On Behalf Of
> Baesken, Matthias
> Sent: Dienstag, 25. Juni 2019 09:43
> To: 'hotspot-dev at openjdk.java.net' <hotspot-dev at openjdk.java.net>
> Subject: RE: RFR: 8226238: Improve error output and fix elf issues in
> os::dll_load
> 
> Ping ...
> 
> Best regards, Matthias
> 
> 
> From: Baesken, Matthias
> Sent: Dienstag, 18. Juni 2019 13:24
> To: 'hotspot-dev at openjdk.java.net' <hotspot-dev at openjdk.java.net>
> Subject: RE: RFR: 8226238: Improve error output and fix elf issues in
> os::dll_load
> 
> Hello,  I noticed  the solaris   os::dll_load  coding has similar  issues.
> New webrev  ( handling linux + solaris ) :
> 
> http://cr.openjdk.java.net/~mbaesken/webrevs/8226238.3/
> 
> Best regards, Matthias
> 
> 
> 
> Hello,
> please  review  this change to linux    os::dll_load .
> os::dll_load  includes some coding to report  error details  in cases  where  the
> lib loading fails .
> For this ,  we ELF  header of the shared-object we failed to load   is evaluated
> .
> 
> However the code   currently has some issues .
> - it misses to convert endianness of elf_head.e_machine (ELF data entry) in
> case the platform we run on, and the lib we try to load  have different
> endianness
> - the error message could show the real elf architecture name in mismatch
> cases (but does not)
> 
> 
> Adjusted error message looks like this :
> example from renaissance db-shootout benchmark   that  runs  on linuxs390x
> and attempts to load an x86_64 JNI lib "by mistake"
> message without the change :
>  .... ELF file data encoding not big-endian (Possible cause: endianness
> mismatch)]
> 
> 
> with change :
>  .... ELF file data encoding not big-endian (Possible cause: can't load AMD 64
> .so on a IBM System/390 platform)
> 
> 
> bug/webrev :
> 
> https://bugs.openjdk.java.net/browse/JDK-8226238
> http://cr.openjdk.java.net/~mbaesken/webrevs/8226238.2/