Re: [PATCH] gdbarch: Add pc_signed field and use it when adjusting BP addresses

[Andreas Schwab <schwab@suse.de>]

On Mär 15 2018, Vlad Ivanov <vlad.ivanov@lab-systems.ru> wrote:

> 15.03.2018, 14:33, "Andreas Schwab" <schwab@suse.de>:
>> On Mär 15 2018, vlad.ivanov@lab-systems.ru wrote:
>>
>>>  diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
>>>  index 454fda7684..247ec34857 100644
>>>  --- a/gdb/breakpoint.c
>>>  +++ b/gdb/breakpoint.c
>>>  @@ -6999,7 +6999,10 @@ adjust_breakpoint_address (struct gdbarch *gdbarch,
>>>             adjusted_bpaddr = gdbarch_adjust_breakpoint_address (gdbarch, bpaddr);
>>>           }
>>>
>>>  - adjusted_bpaddr = address_significant (gdbarch, adjusted_bpaddr);
>>>  + /* Don't cut out "insignificant" address bits on targets with
>>>  + signed PC. */
>>>  + if (!gdbarch_pc_signed (gdbarch))
>>>  + adjusted_bpaddr = address_significant (gdbarch, adjusted_bpaddr);
>>
>> Shouldn't it be sign-extended instead?
>>
>> Andreas.
>>
>
> MIPS backend already returns a sign-extended value, and address_significant 
> cuts out bits 63 to 32. This makes breakpoint address comparison in step 
> routines to misbehave.

What happens if you start with a non-canonical address?  The call to
address_significant wouldn't be needed if adjusted_bpaddr were always
canonical to begin with.

Andreas.

-- 
Andreas Schwab, SUSE Labs, schwab@suse.de
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."