[PATCH] Map registers to remote numbers when encoding an ax_reg or ax_reg_mask operation

[PATCH] Map registers to remote numbers when encoding an ax_reg or ax_reg_mask operation

[Antoine Tremblay]

When encoding the agent expression operation ax_reg or ax_reg_mask, the
register number used is internal to GDB. However GDBServer expects a tdesc
based number.

This usually does not cause a problem since at the moment, for raw
registers GDBServer R trace action ignores the register mask and just
collects all registers.

It can be a problem, however with pseudo registers on some platforms if the
tdesc number doesn't match the GDB internal register number.

This is the case with ARM, the upcoming ARM tracepoint support, fails
these test cases without this patch:

gdb.trace/collection.exp: collect register locals collectively:*

GDBSever would exit with: unhandled register size
Since the register number is not mapped.

This patch fixes these issues by calling gdbarch_remote_register_number
before encoding the register number in the ax_reg or ax_reg_mask operation.

Tested on x86 native-gdbserver no regressions observed.
---
 gdb/ax-general.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/gdb/ax-general.c b/gdb/ax-general.c
index 30f90e8..eb0e6fc 100644
--- a/gdb/ax-general.c
+++ b/gdb/ax-general.c
@@ -308,6 +308,9 @@ ax_reg (struct agent_expr *x, int reg)
     }
   else
     {
+      /* Get the remote register number.  */
+      reg = gdbarch_remote_register_number (x->gdbarch, reg);
+
       /* Make sure the register number is in range.  */
       if (reg < 0 || reg > 0xffff)
         error (_("GDB bug: ax-general.c (ax_reg): "
@@ -456,7 +459,11 @@ ax_reg_mask (struct agent_expr *ax, int reg)
     }
   else
     {
-      int byte = reg / 8;
+      int byte = 0;
+
+      /* Get the remote register number.  */
+      reg = gdbarch_remote_register_number (ax->gdbarch, reg);
+      byte = reg / 8;
 
       /* Grow the bit mask if necessary.  */
       if (byte >= ax->reg_mask_len)
-- 
2.6.4

Re: [PATCH] Map registers to remote numbers when encoding an ax_reg or ax_reg_mask operation

[Pedro Alves]

On 02/25/2016 01:25 PM, Antoine Tremblay wrote:
> When encoding the agent expression operation ax_reg or ax_reg_mask, the
> register number used is internal to GDB. However GDBServer expects a tdesc
> based number.
> 
> This usually does not cause a problem since at the moment, for raw
> registers GDBServer R trace action ignores the register mask and just
> collects all registers.
> 
> It can be a problem, however with pseudo registers on some platforms if the
> tdesc number doesn't match the GDB internal register number.
> 
> This is the case with ARM, the upcoming ARM tracepoint support, fails
> these test cases without this patch:
> 
> gdb.trace/collection.exp: collect register locals collectively:*
> 
> GDBSever would exit with: unhandled register size
> Since the register number is not mapped.
> 
> This patch fixes these issues by calling gdbarch_remote_register_number
> before encoding the register number in the ax_reg or ax_reg_mask operation.
> 
> Tested on x86 native-gdbserver no regressions observed.

>    else
>      {
> -      int byte = reg / 8;
> +      int byte = 0;

No need to zero initialize.

> +
> +      /* Get the remote register number.  */
> +      reg = gdbarch_remote_register_number (ax->gdbarch, reg);
> +      byte = reg / 8;
>  
>        /* Grow the bit mask if necessary.  */
>        if (byte >= ax->reg_mask_len)
> 

OK with ChangeLog entry.

Thanks,
Pedro Alves

Re: [PATCH] Map registers to remote numbers when encoding an ax_reg or ax_reg_mask operation

[Antoine Tremblay]

Pedro Alves writes:

> On 02/25/2016 01:25 PM, Antoine Tremblay wrote:
>> When encoding the agent expression operation ax_reg or ax_reg_mask, the
>> register number used is internal to GDB. However GDBServer expects a tdesc
>> based number.
>> 
>> This usually does not cause a problem since at the moment, for raw
>> registers GDBServer R trace action ignores the register mask and just
>> collects all registers.
>> 
>> It can be a problem, however with pseudo registers on some platforms if the
>> tdesc number doesn't match the GDB internal register number.
>> 
>> This is the case with ARM, the upcoming ARM tracepoint support, fails
>> these test cases without this patch:
>> 
>> gdb.trace/collection.exp: collect register locals collectively:*
>> 
>> GDBSever would exit with: unhandled register size
>> Since the register number is not mapped.
>> 
>> This patch fixes these issues by calling gdbarch_remote_register_number
>> before encoding the register number in the ax_reg or ax_reg_mask operation.
>> 
>> Tested on x86 native-gdbserver no regressions observed.
>
>>    else
>>      {
>> -      int byte = reg / 8;
>> +      int byte = 0;
>
> No need to zero initialize.

OK.

>
>> +
>> +      /* Get the remote register number.  */
>> +      reg = gdbarch_remote_register_number (ax->gdbarch, reg);
>> +      byte = reg / 8;
>>  
>>        /* Grow the bit mask if necessary.  */
>>        if (byte >= ax->reg_mask_len)
>> 
>
> OK with ChangeLog entry.

Oops, yes I'll add a ChangeLog sorry about that.

Thanks,
Antoine

Re: [PATCH] Map registers to remote numbers when encoding an ax_reg or ax_reg_mask operation

[Antoine Tremblay]

Antoine Tremblay writes:

> Pedro Alves writes:
>
>> On 02/25/2016 01:25 PM, Antoine Tremblay wrote:
>>> When encoding the agent expression operation ax_reg or ax_reg_mask, the
>>> register number used is internal to GDB. However GDBServer expects a tdesc
>>> based number.
>>> 
>>> This usually does not cause a problem since at the moment, for raw
>>> registers GDBServer R trace action ignores the register mask and just
>>> collects all registers.
>>> 
>>> It can be a problem, however with pseudo registers on some platforms if the
>>> tdesc number doesn't match the GDB internal register number.
>>> 
>>> This is the case with ARM, the upcoming ARM tracepoint support, fails
>>> these test cases without this patch:
>>> 
>>> gdb.trace/collection.exp: collect register locals collectively:*
>>> 
>>> GDBSever would exit with: unhandled register size
>>> Since the register number is not mapped.
>>> 
>>> This patch fixes these issues by calling gdbarch_remote_register_number
>>> before encoding the register number in the ax_reg or ax_reg_mask operation.
>>> 
>>> Tested on x86 native-gdbserver no regressions observed.
>>
>>>    else
>>>      {
>>> -      int byte = reg / 8;
>>> +      int byte = 0;
>>
>> No need to zero initialize.
>
> OK.
>
>>
>>> +
>>> +      /* Get the remote register number.  */
>>> +      reg = gdbarch_remote_register_number (ax->gdbarch, reg);
>>> +      byte = reg / 8;
>>>  
>>>        /* Grow the bit mask if necessary.  */
>>>        if (byte >= ax->reg_mask_len)
>>> 
>>
>> OK with ChangeLog entry.
>
> Oops, yes I'll add a ChangeLog sorry about that.
>

Pushed.