Re: x86 watchpoints bug (Re: ping: Re: PATCH : allow to set length of hw watchpoints (e.g. for Valgrind gdbserver))

[Pedro Alves <pedro@codesourcery.com>]

On Tuesday 31 May 2011 23:15:13, Philippe Waroquiers wrote:
> 
> > +gdb_test_no_output "set breakpoint always-inserted on"
> 
> I confirm that the bug of the DR register kept busy is also present
> on a native GDB debugging.

Thanks.  I'm trying a different fix, but I don't think
I'll be able to finish it tonight.

> 
> Note the bug is slightly more difficult to trigger without
> the  "set breakpoint always-inserted on"
> as it seems to depend on the order in which watchpoints are inserted.
> 
> If you just do:
>   break s.c:24
>   run
>   watch s1
>   watch s2
>   watch s4
>   watch s3
>   c
> you do not trigger the busy bug as gdb inserts them
> in the order :  s3
>                      s4
>                      s1
>                      s2
> and then there is no busy register remaining.
> 
> I was able to trigger the bug with the following sequence: s1 s2 s4 s16 (that gdb inserts
> in the order s4 s1 s2 s16. See below the full trace of the bug with this order.
> 
> Note that it would be less user confusing if gdb would always insert the watchpoints in the
> order the user entered them. No idea if this is easy to do.
> Of course, the confusion only happens with limited hw watchpoint (with Valgrind gdbserver,
> there will be no confusion :).
> E.g. this is what you can obtain on amd64 with the Valgrind gdbserver (and the patched gdb
> allowing unlimited length):
> (gdb) info watch
> Num     Type           Disp Enb Address            What
> 1       hw watchpoint  keep y                      s1
> 2       hw watchpoint  keep y                      s2
> 3       hw watchpoint  keep y                      s3
> 4       hw watchpoint  keep y                      s4
> 5       hw watchpoint  keep y                      s5
> 6       hw watchpoint  keep y                      s6
> 7       hw watchpoint  keep y                      s7
> 8       hw watchpoint  keep y                      s8
> 9       hw watchpoint  keep y                      s16
> 10      hw watchpoint  keep y                      s32
> 11      hw watchpoint  keep y                      s64
> 12      hw watchpoint  keep y                      s128
> 13      hw watchpoint  keep y                      s1000
>  breakpoint already hit 1 time
> (gdb) 
> (the above is an advertisement  for the Valgrind gdbserver + patched gdb to allow unlimited length :)
> 
> Philippe
> 
> 
> ########################## bug without always-inserted, using s1 s2 s4 s16
> ...
> stopped_data_addr:
>  CONTROL (DR7): 0000000000000000          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000000000, ref.count=0  DR1: addr=0x0000000000000000, ref.count=0
>  DR2: addr=0x0000000000000000, ref.count=0  DR3: addr=0x0000000000000000, ref.count=0
> 
> Breakpoint 1, main () at s.c:24
> 24       p[i] = 1;
> (gdb) watch s1
> Hardware watchpoint 2: s1
> (gdb) watch s2
> Hardware watchpoint 3: s2
> (gdb) watch s4
> Hardware watchpoint 4: s4
> (gdb) watch s16
> Hardware watchpoint 5: s16
> (gdb) c
> Continuing.
> stopped_data_addr:
>  CONTROL (DR7): 0000000000000000          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000000000, ref.count=0  DR1: addr=0x0000000000000000, ref.count=0
>  DR2: addr=0x0000000000000000, ref.count=0  DR3: addr=0x0000000000000000, ref.count=0
> insert_watchpoint (addr=606ac0, len=4, type=data-write):
>  CONTROL (DR7): 00000000000d0101          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000606ac0, ref.count=1  DR1: addr=0x0000000000000000, ref.count=0
>  DR2: addr=0x0000000000000000, ref.count=0  DR3: addr=0x0000000000000000, ref.count=0
> insert_watchpoint (addr=609248, len=1, type=data-write):
>  CONTROL (DR7): 00000000001d0105          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000606ac0, ref.count=1  DR1: addr=0x0000000000609248, ref.count=1
>  DR2: addr=0x0000000000000000, ref.count=0  DR3: addr=0x0000000000000000, ref.count=0
> insert_watchpoint (addr=60d128, len=2, type=data-write):
>  CONTROL (DR7): 00000000051d0115          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000606ac0, ref.count=1  DR1: addr=0x0000000000609248, ref.count=1
>  DR2: addr=0x000000000060d128, ref.count=1  DR3: addr=0x0000000000000000, ref.count=0
> insert_watchpoint (addr=60d1c0, len=16, type=data-write):
>  CONTROL (DR7): 00000000951d0155          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000606ac0, ref.count=1  DR1: addr=0x0000000000609248, ref.count=1
>  DR2: addr=0x000000000060d128, ref.count=1  DR3: addr=0x000000000060d1c0, ref.count=1
> Warning:
> Could not insert hardware watchpoint 5.
> Could not insert hardware breakpoints:
> You may have requested too many hardware breakpoints/watchpoints.
> 
> remove_watchpoint (addr=606ac0, len=4, type=data-write):
>  CONTROL (DR7): 00000000951d0154          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000000000, ref.count=0  DR1: addr=0x0000000000609248, ref.count=1
>  DR2: addr=0x000000000060d128, ref.count=1  DR3: addr=0x000000000060d1c0, ref.count=1
> remove_watchpoint (addr=609248, len=1, type=data-write):
>  CONTROL (DR7): 00000000951d0150          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000000000, ref.count=0  DR1: addr=0x0000000000000000, ref.count=0
>  DR2: addr=0x000000000060d128, ref.count=1  DR3: addr=0x000000000060d1c0, ref.count=1
> remove_watchpoint (addr=60d128, len=2, type=data-write):
>  CONTROL (DR7): 00000000951d0140          STATUS (DR6): 00000000ffff4ff0
>  DR0: addr=0x0000000000000000, ref.count=0  DR1: addr=0x0000000000000000, ref.count=0
>  DR2: addr=0x0000000000000000, ref.count=0  DR3: addr=0x000000000060d1c0, ref.count=1
> 0x0000000000400494 in main () at s.c:24
> 24       p[i] = 1;
> (gdb) 
> 
> 

-- 
Pedro Alves