[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Crash-utility] [ANNOUNCE] crash gcore command, version 1.3.0-rc is released




----- Original Message -----
> 
> 
> ----- Original Message -----
> > From: Dave Anderson <anderson redhat com>
> > Subject: Re: [Crash-utility] [ANNOUNCE] crash gcore command, version
> > 1.3.0-rc
> > is released
> > Date: Wed, 15 Oct 2014 15:11:44 -0400
> > 
> > > 
> > > 
> > > ----- Original Message -----
> > >> 
> > >> 
> > >> (2014/10/14 22:34), Dave Anderson wrote:
> > >> >
> > >> > Hello Daisuke,
> > >> >
> > >> > Thanks for the update -- I will post my results later.
> > >> >
> > >> > However, I note that you did not make the small ARM64 update
> > >> > that I emailed to you last week:
> > >> >
> > >> > This should be changed:
> > >> >
> > >> >    #ifdef ARM64
> > >> >    #define ELF_EXEC_PAGESIZE 4096
> > >> >
> > >> > to:
> > >> >
> > >> >    #ifdef ARM64
> > >> >    #define ELF_EXEC_PAGESIZE PAGESIZE()
> > >> >
> > >> > Please make that change.
> > >> >
> > >> 
> > >> I'll merge this in the next rc release.
> > > 
> > > OK good -- thanks.
> > > 
> > > I have successfully tested this version (with the fix above) on ARM64
> > > kernels configured with 4K and 64K pages, and also when the gcore module
> > > is built and run on an x86_64 host with an ARM64 target.
> > > 
> > 
> > Thanks.
> > 
> > > Another question: another engineer here at Red Hat has recently
> > > created a crash-gcore-command patch that implements support for PPC64,
> > > both big-endian and little-endian.  May we post that patch for inclusion
> > > in version 1.3.0?
> > > 
> > 
> > Sure. It's no problem to merge PPC64 patche. Please post it.
> > 
> > However, as you've understood, I have no PPC64 system. I cannot test
> > PPC64 arch at all.
> > 
> > Also, who will maintain the PPC64 code? The engineer? or you'll do it
> > in addition to ARM64?
> 
> Correct, support for both architectures will come from here (Red Hat)
> since they eventually will be RHEL packages.  IBM would also be interested
> in the PPC64 flavors.
> 
> Thanks,
>   Dave

While sanity-checking the proposed PPC64 patch, I see that there is a small bug
in the current crash-gcore-command-1.3.0-rc code w/respect to the ELF header
contents:

The four architectures currently have these #defines:

  libgcore/gcore_defs.h       ELF_DATA               40 #define ELF_DATA ELFDATA2LSB
  libgcore/gcore_defs.h       ELF_DATA               60 #define ELF_DATA ELFDATA2LSB
  libgcore/gcore_defs.h       ELF_DATA               80 #define ELF_DATA ELFDATA2LSB
  libgcore/gcore_defs.h       ELF_DATA              100 #define ELF_DATA ELFDATA2LSB

But the ELF_DATA #define is not used, but rather the two ELF header fill functions
hardwire it to ELFDATA2LSB: 

  libgcore/gcore_elf_struct.c elf64_fill_elf_header  47 e->e_ident[EI_DATA] = ELFDATA2LSB;
  libgcore/gcore_elf_struct.c elf32_fill_elf_header 220 e->e_ident[EI_DATA] = ELFDATA2LSB;

That's OK for the other 4 architectures, but the PPC64 architecture will need to be 
either be ELFDATA2LSB or ELFDATA2MSB based upon whether it's ppc64le or not.

Dave


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]