[PATCH] gdb: make "start" breakpoint inferior-specific

[Simon Marchi]

On 8/31/22 10:03, Bruno Larsen via Gdb-patches wrote:
> On 04/08/2022 19:40, Simon Marchi via Gdb-patches wrote:
>> I saw this failure on a CI:
>>
>>      (gdb) add-inferior
>>      [New inferior 2]
>>      Added inferior 2
>>      (gdb) PASS: gdb.threads/vfork-multi-inferior.exp: method=non-stop: add-inferior
>>      inferior 2
>>      [Switching to inferior 2 [<null>] (<noexec>)]
>>      (gdb) PASS: gdb.threads/vfork-multi-inferior.exp: method=non-stop: inferior 2
>>      kill
>>      The program is not being run.
>>      (gdb) file /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior-sleep
>>      Reading symbols from /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior-sleep...
>>      (gdb) run &
>>      Starting program: /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior-sleep
>>      (gdb) PASS: gdb.threads/vfork-multi-inferior.exp: method=non-stop: run inferior 2
>>      inferior 1
>>      [Switching to inferior 1 [<null>] (<noexec>)]
>>      (gdb) PASS: gdb.threads/vfork-multi-inferior.exp: method=non-stop: inferior 1
>>      kill
>>      The program is not being run.
>>      (gdb) file /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior
>>      Reading symbols from /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior...
>>      (gdb) break should_break_here
>>      Breakpoint 1 at 0x11b1: file /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/src/binutils-gdb/gdb/testsuite/gdb.threads/vfork-multi-inferior.c, line 25.
>>      (gdb) PASS: gdb.threads/vfork-multi-inferior.exp: method=non-stop: break should_break_here
>>      [Thread debugging using libthread_db enabled]
>>      Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>>      start
>>      Temporary breakpoint 2 at 0x11c0: -qualified main. (2 locations)
>>      Starting program: /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/tmp/tmp.GYATAXR8Ku/gdb/testsuite/outputs/gdb.threads/vfork-multi-inferior/vfork-multi-inferior
>>      [Thread debugging using libthread_db enabled]
>>      Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>>
>>      Thread 2.1 "vfork-multi-inf" hit Temporary breakpoint 2, main () at /home/jenkins/workspace/binutils-gdb_master_linuxbuild/platform/jammy-amd64/target_board/unix/src/binutils-gdb/gdb/testsuite/gdb.threads/vfork-multi-inferior-sleep.c:23
>>      23      sleep (30);
>>      (gdb) FAIL: gdb.threads/vfork-multi-inferior.exp: method=non-stop: start inferior 1
>>
>> What happens is:
>>
>>   1. We start inferior 2 with "run&", it runs very slowly, takes time to
>>      get to main
>>   2. We switch to inferior 1, and run "start"
>>   3. The temporary breakpoint inserted by "start" applies to all inferiors
>>   4. Inferior 2 hits that breakpoint and GDB reports that hit
>>
>> To avoid this, breakpoints inserted by "start" should be
>> inferior-specific.  However, we don't have a nice way to make
>> inferior-specific breakpoints yet.  It's possible to make
>> pspace-specific breakpoints (for example how the internal_breakpoint
>> constructor does) by creating a symtab_and_line manually.  However,
>> inferiors can share program spaces (usually on particular embedded
>> targets), so we could have a situation where two inferiors run the same
>> code in the same program space.  In that case, it would just not be
>> possible to insert a breakpoint in one inferior but not the other.
>>
>> A simple solution that should work all the time is to add a condition to
>> the breakpoint inserted by "start", to check the inferior reporting the
>> hit is the expected one.  This is what this patch implements.
>>
>> Add a test that does:
>>
>>   - start a background inferior 1 that calls main in a loop
>>   - start aninferior 2 with the "start" command
>>   - validate that we hit the breakpoint in inferior 2
>>
>> Without the fix, we hit the breakpoint in inferior 1 pretty much all the
>> time.
>>
>> Change-Id: Ib0148498a476bfa634ed62353c95f163623c686a
> 
> Hi Simon,
> 
> This change makes a lot of sense, I just wonder if it would make sense to make breakpoints inferior-specific. I could see a situation where someone would run the same program twice, one with an ok input while the other has a buggy input for instance, to see the difference between them, and inferior-specific breakpoints could come in handy in a situation like this.
> 
> If you think it would be too much work for a usecase that is too niche, this patch is pretty straight forward, I'd suggest you approve your patch.

I certainly don't want to add a new user-facing feature just for this
fix, as that's a lot of work (figuring out the syntax, doc, NEWS, tests,
etc).  We could enhance the internal API to allow passing an inferior to
make a breakpoint inferior-specific, but I don't think it should be a
prerequisite for this fix.  It would end up doing functionally
identical.

Would you like to to add your Reviewed-By on this patch?

Simon