Translate memory addresses

Translate memory addresses

[Michael Eager]

Hi --

I'm working with a target which needs to translate
a virtual memory addresses to a physical address
before accessing the target's memory or the core file.

I added a new field to target_ops to specify the
translation function and added the following to
target_xfer_partial():

   /* Translate address if needed.  */
   if (ops->to_translate_address)
     offset = ops->to_translate_address (ops, offset);

So far, so good.  I can add this to the target definition
for the live targets.  But this doesn't work for core files.

Core_ops is defined in corelow.c, and there doesn't appear
to be an obvious hook for the target code to modify it to
add the translation routine.

Any suggestions on the best way to add a hook or how to specify
a target routine to do address translations for core files?

-- 
Michael Eager	 eager@eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077

Re: Translate memory addresses

[Tom Tromey]

>>>>> "Michael" == Michael Eager <eager@eagerm.com> writes:

Michael> I'm working with a target which needs to translate
Michael> a virtual memory addresses to a physical address
Michael> before accessing the target's memory or the core file.

Michael> Core_ops is defined in corelow.c, and there doesn't appear
Michael> to be an obvious hook for the target code to modify it to
Michael> add the translation routine.

Michael> Any suggestions on the best way to add a hook or how to specify
Michael> a target routine to do address translations for core files?

I am not an expert in this area, but couldn't you make a new
arch_stratum target that does the translation?

With this approach maybe you don't even need a hook -- just do the
translation and call the target beneath.

Tom

Re: Translate memory addresses

[Michael Eager]

Tom Tromey wrote:
>>>>>> "Michael" == Michael Eager <eager@eagerm.com> writes:
> 
> Michael> I'm working with a target which needs to translate
> Michael> a virtual memory addresses to a physical address
> Michael> before accessing the target's memory or the core file.
> 
> Michael> Core_ops is defined in corelow.c, and there doesn't appear
> Michael> to be an obvious hook for the target code to modify it to
> Michael> add the translation routine.
> 
> Michael> Any suggestions on the best way to add a hook or how to specify
> Michael> a target routine to do address translations for core files?
> 
> I am not an expert in this area, but couldn't you make a new
> arch_stratum target that does the translation?
> 
> With this approach maybe you don't even need a hook -- just do the
> translation and call the target beneath.

Interesting idea.  I didn't think of adding a stratum for translation.

I took a clue from the way that sol-thread.c prevents corelow.c
from calling add_target to install core_ops.  I added the following
to corelow.c:

   CORE_ADDR (*target_translate_address) (struct target_ops *, CORE_ADDR addr);

and in init_core_ops():

   core_ops.to_translate_address = target_translate_address;

The target code contains a definition for target_translate_address
which is initialized to point to the translation routine.  For other
targets, this will be zero.

It's inelegant, what can I say, but functional.


-- 
Michael Eager	 eager@eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077

Re: Translate memory addresses

[Mark Kettenis]

> Date: Tue, 08 Mar 2011 08:47:53 -0800
> From: Michael Eager <eager@eagerm.com>
> 
> Tom Tromey wrote:
> >>>>>> "Michael" == Michael Eager <eager@eagerm.com> writes:
> > 
> > Michael> I'm working with a target which needs to translate
> > Michael> a virtual memory addresses to a physical address
> > Michael> before accessing the target's memory or the core file.
> > 
> > Michael> Core_ops is defined in corelow.c, and there doesn't appear
> > Michael> to be an obvious hook for the target code to modify it to
> > Michael> add the translation routine.
> > 
> > Michael> Any suggestions on the best way to add a hook or how to specify
> > Michael> a target routine to do address translations for core files?
> > 
> > I am not an expert in this area, but couldn't you make a new
> > arch_stratum target that does the translation?
> > 
> > With this approach maybe you don't even need a hook -- just do the
> > translation and call the target beneath.
> 
> Interesting idea.  I didn't think of adding a stratum for translation.
> 
> I took a clue from the way that sol-thread.c prevents corelow.c
> from calling add_target to install core_ops.  I added the following
> to corelow.c:
> 
>    CORE_ADDR (*target_translate_address) (struct target_ops *, CORE_ADDR addr);
> 
> and in init_core_ops():
> 
>    core_ops.to_translate_address = target_translate_address;
> 
> The target code contains a definition for target_translate_address
> which is initialized to point to the translation routine.  For other
> targets, this will be zero.
> 
> It's inelegant, what can I say, but functional.

Did you look at how the BSD kvm target does things?  It does a
physical to virtual address translation (by using the system-provided
libkvm), and seems to get away with this without having such a hook.

Re: Translate memory addresses

[Michael Eager]

Mark Kettenis wrote:
>> Date: Tue, 08 Mar 2011 08:47:53 -0800
>> From: Michael Eager <eager@eagerm.com>
>>
>> Tom Tromey wrote:
>>>>>>>> "Michael" == Michael Eager <eager@eagerm.com> writes:
>>> Michael> I'm working with a target which needs to translate
>>> Michael> a virtual memory addresses to a physical address
>>> Michael> before accessing the target's memory or the core file.
>>>
>>> Michael> Core_ops is defined in corelow.c, and there doesn't appear
>>> Michael> to be an obvious hook for the target code to modify it to
>>> Michael> add the translation routine.
>>>
>>> Michael> Any suggestions on the best way to add a hook or how to specify
>>> Michael> a target routine to do address translations for core files?
>>>
>>> I am not an expert in this area, but couldn't you make a new
>>> arch_stratum target that does the translation?
>>>
>>> With this approach maybe you don't even need a hook -- just do the
>>> translation and call the target beneath.
>> Interesting idea.  I didn't think of adding a stratum for translation.
>>
>> I took a clue from the way that sol-thread.c prevents corelow.c
>> from calling add_target to install core_ops.  I added the following
>> to corelow.c:
>>
>>    CORE_ADDR (*target_translate_address) (struct target_ops *, CORE_ADDR addr);
>>
>> and in init_core_ops():
>>
>>    core_ops.to_translate_address = target_translate_address;
>>
>> The target code contains a definition for target_translate_address
>> which is initialized to point to the translation routine.  For other
>> targets, this will be zero.
>>
>> It's inelegant, what can I say, but functional.
> 
> Did you look at how the BSD kvm target does things?  It does a
> physical to virtual address translation (by using the system-provided
> libkvm), and seems to get away with this without having such a hook.

How would this help?  The kernel isn't running when debugging
a hardware target and you can't execute target code when debugging
a core file.


-- 
Michael Eager	 eager@eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077