Re: [patch] gdbarch_syscall_pc_increment

[Yao Qi <yao@codesourcery.com>]

On 12/12/2012 10:23 PM, Aleksandar Ristovski wrote:
> This patch fixes stepping over system call instruction on architectures
> that use software single stepping and also may increment PC upon return
> from the system call to communicate error. (e.g. this is what we do on
> Neutrino).
>
> Note: I could only test this on Neutrino.
>

Hi,
I don't understand why do you add a gdbarch hook, but use it only in a 
target-specific part?  The goal of gdbarch hooks is about hiding the 
difference of ports and giving a common interface to the common part of 
GDB.  If your issue is arm specific, we don't need this new gdbarch hook 
at all.

If I understand your problem correctly, you have to define your own 
function 'arm_neutrino_syscall_next_pc' in your file 
arm-neutrino-tdep.c, and install it on function pointer 
'syscall_next_pc' (in 'struct gdbarch_tdep' in arm-tdep.h) in 
'arm_neutrino_init_abi'.  Please have a look on how 'syscall_next_pc' is 
set in arm-linux-tdep.c.  Then you can compute the pc for your own os in 
'arm_neutrino_syscall_next_pc'.  Hope it helps.

>
> ChangeLog:
>
>         * gdbarch.sh (syscall_pc_increment): New function.
>         * gdbarch.h, gdbarch.c: Regenerated.
>         * arm-tdep.c (arm_software_single_step): Use
>         gdbarch_syscall_pc_increment and if provided, insert second
>         single step breakpoint at the incremented address.

I guess the reason you need the 2nd single step breakpoint is that GDB 
computes the 'next pc' by mistake, so you need the 2nd single step 
breakpoint setting on the 'right' address of 'next pc'.  In other words, 
do we really need the 2nd single step breakpoint if the address of 'next 
pc' is computed correctly?

-- 
Yao (齐尧)