RE: [PATCHv4 00/10] x86/Linux Target Description Changes

[Andrew Burgess <aburgess@redhat.com>]

"Willgerodt, Felix" <felix.willgerodt@intel.com> writes:

>> -----Original Message-----
>> From: Andrew Burgess <aburgess@redhat.com>
>> Sent: Freitag, 5. April 2024 14:33
>> To: gdb-patches@sourceware.org
>> Cc: Andrew Burgess <aburgess@redhat.com>
>> Subject: [PATCHv4 00/10] x86/Linux Target Description Changes
>> 
>> In v4:
>> 
>>   - I tried merging V3, but it turned out I broke pretty much
>>     everything that wasn't x86 based when configured with
>>     --enable-targets=all,
>> 
>>   - The problem was a failure to correctly split the shared code
>>     between the gdb/arch/ and gdb/nat/ directories, as a consequence,
>>     code which is needed on a non x86 based host to support x86 based
>>     targets wasn't available to the compilation, and the build failed,
>> 
>>   - In V4 I've gone through every patch and resplit the code in a way
>>     which I now believe is correct, I've done the following tests:
>> 
>>     + On a non x86 host I've built GDB to support only the current
>>     host as a target, to support all targets, and to support x86-64
>>     and i386 linux targets,
>> 
>>     + On an i386 virtual machine I built GDB only for the host as a
>>     target, and for all targets.  I regression tested the all targets
>>     build for unix, native-gdbserver, and native-extended-gdbserver,
>> 
>>     + On an x86-64 machine I've built GDB for only the current host as
>>     a target, and for all targets.  I regression tested the all targets
>>     build for unix, native-gdbserver, and native-extended-gdbserver.
>> 
>>   - Only patches 6, 8, and 10 require significant review.  All of the
>>     other patches are pretty trivial (though reviews always welcome).
>> 
>>   - I think there's more improvements that can be made to the x86
>>     target description creation/lookup/caching.  This series only
>>     changes the Linux lookup, and we still cache i386/amd64/x32
>>     separately.
>> 
>>     In the future I think we can merge all x86 target description
>>     caching into a single data structure, this would be for all OS
>>     variants and all ABI variants.
>> 
>>     Though making that "grand unification" will certainly require some
>>     of the code in this series to change, I think the bulk of it will
>>     remain, and trying to do everything in one series is just going to
>>     result in an even larger series.  I'd prefer to get these first
>>     patches merged, then come back to build on this work once this is
>>     merged and we know there's no problems with it.
>> 
>> In v3:
>> 
>>   - Rebased.  Nasty merge conflict with 4bb20a6244b7091 which I think
>>     I've resolved, but am unable to test.  Reposting so the author of
>>     that other commit can validate.
>> 
>>   - Initial testing looks good.  Full tests are still running.
>> 
>> In v2:
>> 
>>   - Rebase to current upstream/master, no merge conflicts,
>> 
>>   - Retested.
>> 
>> ---
>> 
>> Andrew Burgess (10):
>>   gdbserver/ipa/x86: remove unneeded declarations
>>   gdbserver: convert have_ptrace_getregset to a tribool
>>   gdb/x86: move reading of cs and ds state into gdb/nat directory
>>   gdbserver/x86: move no-xml code earlier in x86_linux_read_description
>>   gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition
>>   gdb/gdbserver: share some code relating to target description creation
>>   gdb/arch: assert that X86_XSTATE_MPX is not set for x32
>>   gdbserver: update target description creation for x86/linux
>>   gdb: move xcr0 == 0 check into i386_linux_core_read_description
>>   gdb/gdbserver: share x86/linux tdesc caching
>> 
>>  gdb/Makefile.in                     |   8 +
>>  gdb/amd64-linux-tdep.c              |  32 +---
>>  gdb/amd64-linux-tdep.h              |   6 -
>>  gdb/arch/amd64-linux-tdesc.c        |  61 +++++++
>>  gdb/arch/amd64-linux-tdesc.h        |  30 ++++
>>  gdb/arch/amd64.c                    |   8 +-
>>  gdb/arch/i386-linux-tdesc.c         |  51 ++++++
>>  gdb/arch/i386-linux-tdesc.h         |  29 ++++
>>  gdb/arch/x86-linux-tdesc-features.c | 247 ++++++++++++++++++++++++++++
>>  gdb/arch/x86-linux-tdesc-features.h |  62 +++++++
>>  gdb/arch/x86-linux-tdesc.h          |  37 +++++
>>  gdb/configure.nat                   |   8 +-
>>  gdb/configure.tgt                   |  11 +-
>>  gdb/i386-linux-tdep.c               |  42 ++---
>>  gdb/i386-linux-tdep.h               |  23 ---
>>  gdb/nat/x86-linux-tdesc.c           | 120 ++++++++++++++
>>  gdb/nat/x86-linux-tdesc.h           |  60 +++++++
>>  gdb/nat/x86-linux.c                 |  47 ++++++
>>  gdb/nat/x86-linux.h                 |  28 ++++
>>  gdb/x86-linux-nat.c                 | 123 ++------------
>>  gdbserver/configure.srv             |  11 ++
>>  gdbserver/linux-amd64-ipa.cc        |  50 ++----
>>  gdbserver/linux-arm-low.cc          |   6 +-
>>  gdbserver/linux-i386-ipa.cc         |  26 +--
>>  gdbserver/linux-low.cc              |   2 +-
>>  gdbserver/linux-low.h               |   2 +-
>>  gdbserver/linux-x86-low.cc          | 192 +++++++--------------
>>  gdbserver/linux-x86-tdesc.cc        | 142 +---------------
>>  gdbserver/linux-x86-tdesc.h         |  56 -------
>>  gdbsupport/x86-xstate.h             |  20 +++
>>  30 files changed, 949 insertions(+), 591 deletions(-)
>>  create mode 100644 gdb/arch/amd64-linux-tdesc.c
>>  create mode 100644 gdb/arch/amd64-linux-tdesc.h
>>  create mode 100644 gdb/arch/i386-linux-tdesc.c
>>  create mode 100644 gdb/arch/i386-linux-tdesc.h
>>  create mode 100644 gdb/arch/x86-linux-tdesc-features.c
>>  create mode 100644 gdb/arch/x86-linux-tdesc-features.h
>>  create mode 100644 gdb/arch/x86-linux-tdesc.h
>>  create mode 100644 gdb/nat/x86-linux-tdesc.c
>>  create mode 100644 gdb/nat/x86-linux-tdesc.h
>>  delete mode 100644 gdbserver/linux-x86-tdesc.h
>> 
>> 
>> base-commit: 16810e455feb26ef826a3ed876d6d7e6d24818b0
>> --
>> 2.25.4
>
>
> Hi Andrew,
>
> Thanks for doing this, I like what this does.
> I am still in the middle of reviewing this. Though in first testing I saw
> that I can't build a 32-bit gdbserver on Ubuntu 22.04.
> I am building like this: 
>
> export CFLAGS="-g -O0 -m32"
> export CXXFLAGS="${CFLAGS} -m32"
> export LDFLAGS="-m32"
>
> $src/gdb/configure \                                                            
>      --build="x86_64-linux-gnu" \                                                
>      --host="i686-linux-gnu" \                                                   
>      --disable-gdb
>
> make -s -j$(nproc)  all-gdbserver
>
> Not sure if there is an easier way of doing this, but this works for master.
> With your series I see this:
>
>   CXX    nat/x86-linux-tdesc.o
> /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c: In function ?const target_desc*
> x86_linux_tdesc_for_tid(int, tribool*, gdb::function_view<void(long long unsigned int)>, const char*, uint64_t*)?:
> /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:54:7: error: ?have_ptrace_getfpxregs? was not declared in this scope;
> did you mean ?have_ptrace_getregset??
>    54 |   if (have_ptrace_getfpxregs == -1)
>       |       ^~~~~~~~~~~~~~~~~~~~~~
>       |       have_ptrace_getregset
> /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:56:7: error: ?elf_fpxregset_t? was not declared in this scope
>    56 |       elf_fpxregset_t fpxregs;
>       |       ^~~~~~~~~~~~~~~
> In file included from /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:29:
> /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:58:53: error: ?fpxregs? was not declared in this scope
>    58 |       if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0)
>       |                                                     ^~~~~~~
> /gdb/gdbserver/../gdb/nat/gdb_ptrace.h:142:57: note: in definition of macro ?ptrace?
>   142 |          ptrace ((PTRACE_TYPE_ARG1) request, pid, addr, data)
>       |                                                         ^~~~
> make[1]: *** [Makefile:577: nat/x86-linux-tdesc.o] Error 1
> make: *** [Makefile:13074: install-gdbserver] Error 2
>
> I didn't investigate further yet, sorry.

Thanks for bringing this to my attention.  I'll investigate this
tomorrow.

Thanks,
Andrew


>
> Regards,
> Felix
> Intel Deutschland GmbH
> Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
> Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de>
> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva  
> Chairperson of the Supervisory Board: Nicole Lau
> Registered Office: Munich
> Commercial Register: Amtsgericht Muenchen HRB 186928