Re: RFA: Generate normal DWARF DW_LOC descriptors for non integer mode pointers

[Jeff Law <law@redhat.com>]

On 05/16/2016 08:12 AM, Nick Clifton wrote:
> Hi Guys,
>
>   Currently dwarf2out.c:mem_loc_descriptor() has some special case
>   code to handle the situation where an address is held in a register
>   whose mode is not of type MODE_INT.  It generates a
>   DW_OP_GNU_regval_type expression which may later on be converted into
>   a frame pointer based expression.  This is a problem for targets which
>   use a partial integer mode for their pointers (eg the msp430).  In
>   such cases the conversion to a frame pointer based expression could
>   be wrong if the frame pointer is not being used.
>
>   For example the GDB testfile gdb/testsuite/gdb.base/advance.c contains
>   this code fragment:
>
>     int
>     main ()
>     {
>       int result;
>       int b, c;
>       c = 5;
>       b = 3;    /* advance this location */
>
>       func (c); /* stop here after leaving current frame */
>
>   which compiles to these instructions:
>
> 	suba	#6,	r1
> 	mov	#5,	4(r1)	
> 	mov	#3,	2(r1)	
> 	mov	4(r1),	r12	
> 	calla	#0		;<func>
>
>   (Note that only r1 - the stack pointer - is used.  r4 - the frame
>   pointer - is not).
>
>   The debug information produced for the "c" local variable looks like
>   this:
>
>     Abbrev Number: 3 (DW_TAG_variable)
>      DW_AT_name        : c
>      DW_AT_decl_file   : 1
>      DW_AT_decl_line   : 40
>      DW_AT_type        : <0x37>
>      DW_AT_location    : 5 byte block: f5 4 21 32 1c 	(DW_OP_GNU_regval_type: 4 (r4) <0x21>; DW_OP_lit2; DW_OP_minus)
>
>   ie it says that "c" is stored in memory location "r4 - 2", which is
>   wrong since register r4 is not even used in this function.
>
>   The patch below addresses this problem by allowing the normal,
>   register based descriptor to be produced when the mode is Pmode.
>
>   With this patch applied the unexpected failure count in the GDB
>   testsuite for the MSP430's -mlarge multilib changes from 2253 to 367.
>   There are no regressions, for MSP430 or x86_64, and no changes to
>   the GCC testsuite results for either target.
>
>   OK to apply ?
>
> Cheers
>   Nick
>
> gcc/ChangeLog
> 2016-05-16  Nick Clifton  <nickc@redhat.com>
>
> 	* dwarf2out.c (mem_loc_descriptor): Convert REG based addresses
> 	whose mode is Pmode into basereg descriptors even if Pmode is
>         not an integer mode.
I'm not real familiar with dwarf, so if one of other maintainers steps 
in and says this is OK, then ignore my comments/questions.

I may be missing something, but isn't it the transition to an FP 
relative address rather than a SP relative address that's the problem 
here?  Where does that happen?  Is it possible we've got the wrong 
DECL_RTL or somesuch?

Jeff