powerpc-linux biarch corefile support
Mark Kettenis
mark.kettenis@xs4all.nl
Mon Jan 22 17:12:00 GMT 2007
> Date: Tue, 23 Jan 2007 03:13:08 +1030
> From: Alan Modra <amodra@bigpond.net.au>
>
> This patch updates the powerpc-linux backend to use the newer struct
> regset collect_regset facility. Along with the BFD change in
> http://sources.redhat.com/ml/binutils/2006-12/msg00205.html, this
> gives better support for debugging 32-bit powerpc programs under a
> 64-bit gdb. A powerpc64-linux gdb running the gdb testsuite with -m32
> shows the following improvements without any regressions. No
> regressions on powerpc-linux either.
>
> -FAIL: gdb.base/gcore.exp: where in corefile (pattern 1)
> -FAIL: gdb.base/gcore.exp: corefile restored general registers
> -FAIL: gdb.base/gcore.exp: capture_command_output failed on print array_func::local_array.
> -FAIL: gdb.base/gcore.exp: corefile restored stack array
> -FAIL: gdb.base/gcore.exp: corefile restored backtrace
> -FAIL: gdb.base/multi-forks.exp: follow parent, print pids (timeout)
> -FAIL: gdb.base/readline.exp: print 42
> -FAIL: gdb.threads/gcore-thread.exp: a corefile thread is executing thread2
> -FAIL: gdb.threads/gcore-thread.exp: thread2 is current thread in corefile
>
> We do lose one feature of the old code, which took some pains to write
> the whole register field in a gregset_t when gdb's idea of the
> register size was smaller than space in the gregset_t. I figure this
> is unimportant since the current gdb code doesn't make use of writing
> a single field as far as I can tell. Even if it did, it would
> presumably be writing into a buffer that had been initialised at some
> point with a full gregset. When writing the full regset, we now clear
> the buffer beforehand.
>
> OK to apply?
Unfortunately not:
1. You use C99 structure initialization syntax; GDB still uses C90.
2. The use of memset() in ppc_collect_gregset()/ppc_collect_fpregset()
is wrong. These functions should leave the contents of the buffer
you're collecting the registers in alone, except for the registers
that are actually being collected. The special value -1, means
"all registers GDB supports" not "all registers the operating
system supports". In some cases the OS will store some additional
bits in the data structure used by ptrace(PT_GETREGS, ...) and we
want to pass those back unchanged in the matching
ptrace(PT_SETREGS, ...). I guess you need this for zero-extending
the GPR's to 64-bits, but I think you should do that explicitly for
each register, even if REGNUM isn't -1.
Also, I'd probably use gpr_size instead of gpr_step, since that
expresses more clearly that the registers in the set are really that
size.
Mark
P.S. Could you do me a favour and send unified diffs? That's seems to
be what everybody does these days, and I really have lost the
ability to read context diffs :(.
> * ppc-linux-nat.c (supply_gregset): Use ppc_supply_gregset.
> (right_fill_reg): Delete.
> (fill_gregset): Use ppc_collect_gregset.
> (supply_fpregset): Use ppc_supply_fpregset.
> (fill_fpregset): Use ppc_collect_fpregset.
> * ppc-linux-tdep.c (PPC_LINUX_PT_*): Don't define.
> (right_supply_register, ppc_linux_supply_gregset): Delete.
> (ppc32_linux_supply_gregset, ppc64_linux_supply_gregset): Delete.
> (ppc_linux_supply_fpregset): Delete.
> (ppc32_linux_reg_offsets, ppc64_linux_reg_offsets): New.
> (ppc64_32_linux_reg_offsets): New.
> (ppc32_linux_gregset, ppc64_linux_gregset): Update to use reg offsets,
> ppc_supply_gregset, and ppc_collect_gregset.
> (ppc64_32_linux_gregset): New.
> (ppc_linux_fpregset): Rename to ppc32_linux_fpregset and update.
> (ppc_linux_gregset, ppc_linux_fpregset): New functions.
> (ppc_linux_regset_from_core_section): Update.
> * ppc-tdep.h (ppc_linux_gregset, ppc_linux_fpregset): Declare.
> (ppc_linux_supply_gregset, ppc_linux_supply_fpregset): Delete.
> (struct ppc_reg_offsets): Add "gpr_step" field.
> * ppcnbsd-tdep.c (_initialize_ppcnbsd_tdep): Init gpr_step.
> * ppcobsd-tdep.c (_initialize_ppcobsd_tdep): Likewise.
> * rs6000-tdep.c (ppc_supply_gregset): Heed gpr_step. Fix xer reg
> offset typo.
> (ppc_collect_gregset): Heed gpr_step. Clear entire gregset before
> filling if transferring all regs.
> (ppc_collect_fpregset): Clear entire fpregset before filling if
> transferring all regs.
More information about the Gdb-patches
mailing list