Re: MIPS: Handle manual calls of MIPS16 functions with a call stub

["Maciej W. Rozycki" <macro@mips.com>]

On Fri, 1 Feb 2008, Daniel Jacobowitz wrote:

> >  The other way round -- the minimal symbol points to the actual entry 
> > point, but the stub precedes it and is included in the DWARF-2 block 
> > together with the MIPS16 function body.  Here's an example that triggers a 
> > failure in the test suite (generated from gdb.base/call-ar-st.c by GCC 
> > 4.2.2):
> 
> Then why aren't we calling the instruction at the start of the block,
> i.e. the stub?  In which case not using the MIPS16 convention is
> correct.  I don't see why you'd want to call
> __fn_stub_print_ten_doubles as a MIPS16 function.

 Well, GDB does not ever seem to call the stub.  I have not written code 
responsible for this, but I can see two possible reasons:

1. Simplicity -- depending on the callers of the function in question 
   there may be no stub.  If there are no standard MIPS callers, then the 
   stub is stripped out by the linker.

2. Performance -- the stub is a couple of additional instructions to 
   execute which buy you nothing when called from GDB as it may load the 
   correct argument registers according to the ABI in the first place.

And as I wrote the block associated with print_ten_doubles() does not span 
__fn_stub_print_ten_doubles() -- I may have not been clear enough about 
this being the case for the DWARF-2 record.

 This is what GDB has to say about the function (with the fix applied):

(gdb) print print_ten_doubles
$1 = {void (double, double, double, double, double, double, double, double,
    double, double)} 0x80020a91 <print_ten_doubles>
(gdb) print __fn_stub_print_ten_doubles
$2 = {<text variable, no debug info>} 0x800283d0 <__fn_stub_print_ten_doubles>

 And this is what the relevant DWARF-2 record holds:

 <1><de6>: Abbrev Number: 16 (DW_TAG_subprogram)
  <de7>     DW_AT_external    : 1
  <de8>     DW_AT_name        : print_ten_doubles
  <dfa>     DW_AT_decl_file   : 1
  <dfb>     DW_AT_decl_line   : 664
  <dfd>     DW_AT_low_pc      : 0x80020a90
  <e01>     DW_AT_high_pc     : 0x80020b00
  <e05>     DW_AT_frame_base  : 0x306   (location list)
  <e09>     DW_AT_sibling     : <e9d>

 I have done a little more research of this matter now and it looks like 
the reason this is happening is a likely bug somewhere in GAS.  For 
comparison, here are the unrelocated DWARF-2 records for 
print_ten_doubles() and a nearby function that has no stub:

 <1><bab>: Abbrev Number: 16 (DW_TAG_subprogram)
    <bac>   DW_AT_external    : 1
    <bad>   DW_AT_name        : init_small_structs
    <bc0>   DW_AT_decl_file   : 1
    <bc1>   DW_AT_decl_line   : 613
    <bc3>   DW_AT_low_pc      : 0x790
    <bc7>   DW_AT_high_pc     : 0x900
    <bcb>   DW_AT_frame_base  : 0x2db   (location list)
    <bcf>   DW_AT_sibling     : <0xcaf>

 <1><caf>: Abbrev Number: 16 (DW_TAG_subprogram)
    <cb0>   DW_AT_external    : 1
    <cb1>   DW_AT_name        : print_ten_doubles
    <cc3>   DW_AT_decl_file   : 1
    <cc4>   DW_AT_decl_line   : 664
    <cc6>   DW_AT_low_pc      : 0x900
    <cca>   DW_AT_high_pc     : 0x97c
    <cce>   DW_AT_frame_base  : 0x306   (location list)
    <cd2>   DW_AT_sibling     : <0xd66>

And here are the relevant relocation records:

00000bc3  00003c02 R_MIPS_32         00000790   .LFB23
00000bc7  00000202 R_MIPS_32         00000000   .text

00000cc6  00000202 R_MIPS_32         00000000   .text
00000cca  00000202 R_MIPS_32         00000000   .text

Notice that the DWARF-2 record at 0xbc3 is relocated against .LFB23 and 
one at 0xcc6 -- against .text, rather than .LFB20 as it should be.  I 
presume this is because of the section switch happening inbetween.  Or 
could it be because of ".set nomips16" actually preceding the section 
switch?  Thiemo, can you perhaps make any comments about this?

 I do not know how long this bug has been there in GAS, but it may still 
be worth handling broken binaries people may have.  Then again -- maybe 
not.  But we have no fix for GAS as yet.  Regardless I have not made a 
strong opinion either way.

  Maciej