[gdb/testsuite] Fix gdb.dwarf2/dw2-entry-points.exp on ppc64le

  On ppc64le-linux, I run into:
...
(gdb) bt^M
 #0  0x00000000100006dc in foobar (J=2)^M
 #1  0x000000001000070c in prog ()^M
(gdb) FAIL: gdb.dwarf2/dw2-entry-points.exp: bt foo
...

The test-case attemps to emulate additional entry points of a function, with
function bar having entry points foo and foobar:
...
(gdb) p bar
$1 = {void (int, int)} 0x1000064c <bar>
(gdb) p foo
$2 = {void (int, int)} 0x10000698 <foo>
(gdb) p foobar
$3 = {void (int)} 0x100006d0 <foobar>
...

However, when setting a breakpoint on the entry point foo:
...
(gdb) b foo
Breakpoint 1 at 0x100006dc
...
it ends up in foobar instead of in foo, due to prologue skipping, and
consequently the backtrace show foobar instead foo.

The problem is that the test-case does not emulate an actual prologue at each
entry point.

Fix this by disabling the prologue skipping when setting a breakpoint, using
"break *foo".

Tested on ppc64le-linux and x86_64-linux.

PR testsuite/31232
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31232
---
 gdb/testsuite/gdb.dwarf2/dw2-entry-points.exp | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)


base-commit: 4ece39c56cfdd5647d4061f3c084b9de6f9e443c
  

On 11/01/2024 13:42, Tom de Vries wrote:
> On ppc64le-linux, I run into:
> ...
> (gdb) bt^M
>   #0  0x00000000100006dc in foobar (J=2)^M
>   #1  0x000000001000070c in prog ()^M
> (gdb) FAIL: gdb.dwarf2/dw2-entry-points.exp: bt foo
> ...
>
> The test-case attemps to emulate additional entry points of a function, with
> function bar having entry points foo and foobar:
> ...
> (gdb) p bar
> $1 = {void (int, int)} 0x1000064c <bar>
> (gdb) p foo
> $2 = {void (int, int)} 0x10000698 <foo>
> (gdb) p foobar
> $3 = {void (int)} 0x100006d0 <foobar>
> ...
>
> However, when setting a breakpoint on the entry point foo:
> ...
> (gdb) b foo
> Breakpoint 1 at 0x100006dc
> ...
> it ends up in foobar instead of in foo, due to prologue skipping, and
> consequently the backtrace show foobar instead foo.
>
> The problem is that the test-case does not emulate an actual prologue at each
> entry point.
>
> Fix this by disabling the prologue skipping when setting a breakpoint, using
> "break *foo".

Sorry if my question doesn't make sense, but wouldn't we also want to 
verify that prologue skipping works correctly in a function with 
multiple entrypoints?

I agree that this looks like a testsuite fail rather than a real GDB 
bug, but I feel like we'll be losing a valuable test if we don't try it 
at any point, especially now that it is easy to explicitly mark an 
instruction as EPILOGUE_END in the dwarf assembler.
  

On 1/11/24 13:55, Guinevere Larsen wrote:
> On 11/01/2024 13:42, Tom de Vries wrote:
>> On ppc64le-linux, I run into:
>> ...
>> (gdb) bt^M
>>   #0  0x00000000100006dc in foobar (J=2)^M
>>   #1  0x000000001000070c in prog ()^M
>> (gdb) FAIL: gdb.dwarf2/dw2-entry-points.exp: bt foo
>> ...
>>
>> The test-case attemps to emulate additional entry points of a 
>> function, with
>> function bar having entry points foo and foobar:
>> ...
>> (gdb) p bar
>> $1 = {void (int, int)} 0x1000064c <bar>
>> (gdb) p foo
>> $2 = {void (int, int)} 0x10000698 <foo>
>> (gdb) p foobar
>> $3 = {void (int)} 0x100006d0 <foobar>
>> ...
>>
>> However, when setting a breakpoint on the entry point foo:
>> ...
>> (gdb) b foo
>> Breakpoint 1 at 0x100006dc
>> ...
>> it ends up in foobar instead of in foo, due to prologue skipping, and
>> consequently the backtrace show foobar instead foo.
>>
>> The problem is that the test-case does not emulate an actual prologue 
>> at each
>> entry point.
>>
>> Fix this by disabling the prologue skipping when setting a breakpoint, 
>> using
>> "break *foo".
> 
> Sorry if my question doesn't make sense, but wouldn't we also want to 
> verify that prologue skipping works correctly in a function with 
> multiple entrypoints?
> 

Agreed, ideally we'd also verify it in this test-case.

However, the commit introducing support for this also adds a test-case 
gdb.fortran/entry-point.exp which does verify that part of the 
functionality with compiler-generated entry points and prologues.

> I agree that this looks like a testsuite fail rather than a real GDB 
> bug, but I feel like we'll be losing a valuable test if we don't try it 
> at any point, especially now that it is easy to explicitly mark an 
> instruction as EPILOGUE_END in the dwarf assembler.
> 

I suppose you mean prologue_end (and I hadn't thought of that, thanks 
for the suggestion), but yes, we could do this.  It would require adding 
a .debug_line section in the dwarf assembly, something that is currently 
missing.

But given that the other test-case already tests this functionality, I'm 
not sure if it's worth the effort.

Thanks,
- Tom
  

On 11/01/2024 14:08, Tom de Vries wrote:
> On 1/11/24 13:55, Guinevere Larsen wrote:
>> On 11/01/2024 13:42, Tom de Vries wrote:
>>> On ppc64le-linux, I run into:
>>> ...
>>> (gdb) bt^M
>>>   #0  0x00000000100006dc in foobar (J=2)^M
>>>   #1  0x000000001000070c in prog ()^M
>>> (gdb) FAIL: gdb.dwarf2/dw2-entry-points.exp: bt foo
>>> ...
>>>
>>> The test-case attemps to emulate additional entry points of a 
>>> function, with
>>> function bar having entry points foo and foobar:
>>> ...
>>> (gdb) p bar
>>> $1 = {void (int, int)} 0x1000064c <bar>
>>> (gdb) p foo
>>> $2 = {void (int, int)} 0x10000698 <foo>
>>> (gdb) p foobar
>>> $3 = {void (int)} 0x100006d0 <foobar>
>>> ...
>>>
>>> However, when setting a breakpoint on the entry point foo:
>>> ...
>>> (gdb) b foo
>>> Breakpoint 1 at 0x100006dc
>>> ...
>>> it ends up in foobar instead of in foo, due to prologue skipping, and
>>> consequently the backtrace show foobar instead foo.
>>>
>>> The problem is that the test-case does not emulate an actual 
>>> prologue at each
>>> entry point.
>>>
>>> Fix this by disabling the prologue skipping when setting a 
>>> breakpoint, using
>>> "break *foo".
>>
>> Sorry if my question doesn't make sense, but wouldn't we also want to 
>> verify that prologue skipping works correctly in a function with 
>> multiple entrypoints?
>>
>
> Agreed, ideally we'd also verify it in this test-case.
>
> However, the commit introducing support for this also adds a test-case 
> gdb.fortran/entry-point.exp which does verify that part of the 
> functionality with compiler-generated entry points and prologues.
>
>> I agree that this looks like a testsuite fail rather than a real GDB 
>> bug, but I feel like we'll be losing a valuable test if we don't try 
>> it at any point, especially now that it is easy to explicitly mark an 
>> instruction as EPILOGUE_END in the dwarf assembler.
>>
>
> I suppose you mean prologue_end (and I hadn't thought of that, thanks 
> for the suggestion), but yes, we could do this.  It would require 
> adding a .debug_line section in the dwarf assembly, something that is 
> currently missing.
oops, yes sorry, prologue_end. Working on the epilogue_begin stuff has 
me mixing the 2 words all the time hahaha
>
> But given that the other test-case already tests this functionality, 
> I'm not sure if it's worth the effort.

Yeah, I think I agree with you here that if a testcase already verifies 
that, going through the trouble of adding the whole lines section is not 
worth it.

FWIW, patch doesn't introduce any regressions on my x86, f39 machine,

Tested-By: Guinevere Larsen <blarsen@redhat.com>
  

Tom de Vries <tdevries@suse.de> wrote:

>The test-case attemps to emulate additional entry points of a function, with
>function bar having entry points foo and foobar:
[...]
>Fix this by disabling the prologue skipping when setting a breakpoint, using
>"break *foo".

Normally I don't like to introduce "break *..." because it ignores
the ppc64le ELFv2 local/global entry point distinction.  But in
this particular case, this distinction doesn't apply anyway as
the extra entry points are synthetic and don't have local versions.

So I think this should be OK.

Thanks,
Ulrich