[Lldb-commits] [PATCH] Refactor POSIXThread::GetRegisterContext and use i386 register context

[Todd Fiala]

  More info:

  I started suspecting that the PTRACE_USERDATA offset we were looking at was just wrong.  I now have some data that would seem to suggest that hypothesis is correct.

  I modified the linux process monitor register read/write operations to print off the PTRACE_PEEKUSER offsets being used.  Here is the new lldb run with that information printed out in an [offset = ...] section for the debug breakpoint control registers.

  tfiala at ub-x86-13:~/play/hello$ ~/lldb/work/build-debug/bin/lldb ./hello
  Current executable set to './hello' (i386).
  (lldb) log enable linux registers
  (lldb) run
  Process 11609 launching
  operation ProcessMonitor::Execute() write reg dr6 [offset 0x02a8]: 0xffffffff
  operation ProcessMonitor::Execute() write reg dr7 [offset 0x02ac]: 0xffffffff
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() read reg dr6 [offset 0x02a8]: 0x2ac
  operation ProcessMonitor::Execute() write reg dr6 [offset 0x02a8]: 0xffffffff
  operation ProcessMonitor::Execute() read reg dr7 [offset 0x02ac]: 0x2a8
  operation ProcessMonitor::Execute() read reg dr2 [offset 0x0298]: 0x2a8
  lldb: /home/tfiala/lldb/work/llvm/tools/lldb/source/Plugins/Process/POSIX/POSIXThread.cpp:526: void POSIXThread::WatchNotify(const ProcessMessage&): Assertion `wp_sp.get() && "No watchpoint found"' failed.
  Aborted (core dumped)

  Here is a little program I wrote that should be printing out the actual offsets that should be used, based on sys/user.h on a Ubuntu 13.10 x86 (32-bit) box:



    #include <stdio.h>
    #include <sys/user.h>

    int main(int argc, char **argv)
    {
      const int debug_reg_count = 8;

      for (int i = 0; i < debug_reg_count; ++i)
        {
          printf("DR%d: 0x%04lx\n", i, ((unsigned long) &((struct user *)0)->u_debugreg[i]));
        }

      return 0;
    }


  This prints out the following on that 32-bit box:

  tfiala at ub-x86-13:~/play/debug_regs$ g++ -g -o print_regs print_regs.cpp
  tfiala at ub-x86-13:~/play/debug_regs$ ./print_regs
  DR0: 0x00fc
  DR1: 0x0100
  DR2: 0x0104
  DR3: 0x0108
  DR4: 0x010c
  DR5: 0x0110
  DR6: 0x0114
  DR7: 0x0118

  These appear to be significantly different.  If this DR register calculation code is correct, then the x86 32-bit registers are way off on Linux 32-bit.

  Investigating further now.

http://llvm-reviews.chandlerc.com/D2765