[ECOS] Port problem

[ECOS] Port problem

[Cristiano Ligieri Pereira]

Hi all,

I'm porting eCos to a XScale based platform. The board have Cygmon in its
Flash memory, which is invoked at reset time. I can load an eCos
application in it (I'm modifying the port to IQ80310). So far I have
modified the file PLATFORM_SETUP1, in which I change the memory map and
initialize (or re-initialize since cygmon initialize them already) all the
necessary components (caches, BTB, MMU, etc...). All this seems to work
fine. The memory mapping is changed so that I have SDRAM at address 0x0.

Now I'm a little confused on how to make the serial communication work. I
can download an application using GDB. So obviously there is a connection
between the target and the GDB in the host. What should I do to keep using
this connection to send debugging messages to the host? Or should I break
this connection and restart it somehow?

Thanks in advance,
Cristiano.

------------------------------------------------------------
Cristiano Ligieri Pereira - http://www.ics.uci.edu/~cpereira

Re: [ECOS] Port problem

[Cristiano Ligieri Pereira]

Hi again!

I don't know whether my previous question made no sense or if it was too
simple... since I've got no answer. :-|

I'm working on the port of an XScale based platform. So far I have done a
reasonable progress (in my opinion...) but now I'm kind of stuck. :(

I started the port copying the IQ80310 and modifying it's files since both
platforms are quite different (except the processor, obviously). Then the
next thing I did was to clean up the macro PLATFORM_SETUP1 so that the
specific IQ80310 initializations were removed and the specific
initializations of my platform were inserted (those I got from CygMon
source code). After this I've changed the memory map (that I actually sent
to this list yesterday) so that I have RAM at 0x0.

After this I've been trying two things: load a Hello World eCos
application configured to execute on RAM (by means of GDB/Cygmon) and load
RedBoot configured to execute at RAM as well. For both I load the
application at the original RAM address (since for cygmon this is still
the RAM memory address) and for both I'm executing the code in
PLATFORM_SETUP1, which remaps the memory.

When I try to execute the Hello World application the code crashes when
calling diag_printf function in the main body. For the RAM/RedBoot code it
is crashing when cleaning the BSS (I've checked the addresses I'm hitting
and they fall within the original memory addresses 0xC0000000).

how should I do about the serial communication with the host? Can I reuse
the communication already established with CygMon? Do I need to break it
and restart a new connection?

Am I on the right track? I was feeling good about it till two days
ago. But now that I'm stuck I getting frustrated instead. :-|

Thanks again,
Cristiano.

> Hi all,
> 
> I'm porting eCos to a XScale based platform. The board have Cygmon in its
> Flash memory, which is invoked at reset time. I can load an eCos
> application in it (I'm modifying the port to IQ80310). So far I have
> modified the file PLATFORM_SETUP1, in which I change the memory map and
> initialize (or re-initialize since cygmon initialize them already) all the
> necessary components (caches, BTB, MMU, etc...). All this seems to work
> fine. The memory mapping is changed so that I have SDRAM at address 0x0.
> 
> Now I'm a little confused on how to make the serial communication work. I
> can download an application using GDB. So obviously there is a connection
> between the target and the GDB in the host. What should I do to keep using
> this connection to send debugging messages to the host? Or should I break
> this connection and restart it somehow?
> 
> Thanks in advance,
> Cristiano.
> 
> ------------------------------------------------------------
> Cristiano Ligieri Pereira - http://www.ics.uci.edu/~cpereira

Re: [ECOS] Port problem

[Jesper Skov]

>>>>> "Cristiano" == Cristiano Ligieri Pereira <cpereira@ics.uci.edu> writes:

Cristiano> After this I've been trying two things: load a Hello World
Cristiano> eCos application configured to execute on RAM (by means of
Cristiano> GDB/Cygmon) and load RedBoot configured to execute at RAM
Cristiano> as well. For both I load the application at the original
Cristiano> RAM address (since for cygmon this is still the RAM memory
Cristiano> address) and for both I'm executing the code in
Cristiano> PLATFORM_SETUP1, which remaps the memory.

Diddling the MMU when running from RAM is not going to work unless you
are really careful!

Cristiano> how should I do about the serial communication with the
Cristiano> host? Can I reuse the communication already established
Cristiano> with CygMon? Do I need to break it and restart a new
Cristiano> connection?

The problem is probably that the eCos application will try to use
virtual vectors for writing to the serial port: but these may not be
compatible with CygMon.

Running RedBoot from RAM should work though (as it will initialize the
virtual vectors and use its own serial driver).

Jesper

Re: [ECOS] Port problem

[Mark Salter]

>>>>> Jesper Skov writes:

>>>>> "Cristiano" == Cristiano Ligieri Pereira <cpereira@ics.uci.edu> writes:
Cristiano> After this I've been trying two things: load a Hello World
Cristiano> eCos application configured to execute on RAM (by means of
Cristiano> GDB/Cygmon) and load RedBoot configured to execute at RAM
Cristiano> as well. For both I load the application at the original
Cristiano> RAM address (since for cygmon this is still the RAM memory
Cristiano> address) and for both I'm executing the code in
Cristiano> PLATFORM_SETUP1, which remaps the memory.

> Diddling the MMU when running from RAM is not going to work unless you
> are really careful!

Cristiano> how should I do about the serial communication with the
Cristiano> host? Can I reuse the communication already established
Cristiano> with CygMon? Do I need to break it and restart a new
Cristiano> connection?

> The problem is probably that the eCos application will try to use
> virtual vectors for writing to the serial port: but these may not be
> compatible with CygMon.

> Running RedBoot from RAM should work though (as it will initialize the
> virtual vectors and use its own serial driver).

And PLATFORM_SETUP1 is a no-op for RAM-based RedBoots, at least
for iq80310.

--Mark

Re: [ECOS] Port problem

[Cristiano Ligieri Pereira]

> And PLATFORM_SETUP1 is a no-op for RAM-based RedBoots, at least
> for iq80310.

Yes, it is. But I'm forcing to execute it so that I'm able to remmap
memory. So shouldn't I remmap the memory when executing RedBoot from RAM
then?

Thanks,
Cristiano.

Re: [ECOS] Port problem

[Gary Thomas]

On 17-Aug-2001 Cristiano Ligieri Pereira wrote:
> 
> Hi again!
> 
> I don't know whether my previous question made no sense or if it was too
> simple... since I've got no answer. :-|
> 
> I'm working on the port of an XScale based platform. So far I have done a
> reasonable progress (in my opinion...) but now I'm kind of stuck. :(
> 
> I started the port copying the IQ80310 and modifying it's files since both
> platforms are quite different (except the processor, obviously). Then the
> next thing I did was to clean up the macro PLATFORM_SETUP1 so that the
> specific IQ80310 initializations were removed and the specific
> initializations of my platform were inserted (those I got from CygMon
> source code). After this I've changed the memory map (that I actually sent
> to this list yesterday) so that I have RAM at 0x0.
> 

You built this CygMon yourself?  If so, what was the source code base?

Note: we really only support RedBoot on the XScale, not CygMon.

> After this I've been trying two things: load a Hello World eCos
> application configured to execute on RAM (by means of GDB/Cygmon) and load
> RedBoot configured to execute at RAM as well. For both I load the
> application at the original RAM address (since for cygmon this is still
> the RAM memory address) and for both I'm executing the code in
> PLATFORM_SETUP1, which remaps the memory.
> 
> When I try to execute the Hello World application the code crashes when
> calling diag_printf function in the main body. For the RAM/RedBoot code it
> is crashing when cleaning the BSS (I've checked the addresses I'm hitting
> and they fall within the original memory addresses 0xC0000000).
> 

This tells me that you may have a memory mapping problem.

eCos programs expect the memory map to be somewhat different than the hardware
map.  Most particularly, DRAM gets moved to 0x00000000 and the FLASH at 0
gets moved around as well.  On some XScale platforms, this is even trickier,
but the basic idea is the same.

> how should I do about the serial communication with the host? Can I reuse
> the communication already established with CygMon? Do I need to break it
> and restart a new connection?
> 

No.

> Am I on the right track? I was feeling good about it till two days
> ago. But now that I'm stuck I getting frustrated instead. :-|
> 

I think you should look into getting RedBoot working as your base debug 
environment.

> Thanks again,
> Cristiano.
> 
>> Hi all,
>> 
>> I'm porting eCos to a XScale based platform. The board have Cygmon in its
>> Flash memory, which is invoked at reset time. I can load an eCos
>> application in it (I'm modifying the port to IQ80310). So far I have
>> modified the file PLATFORM_SETUP1, in which I change the memory map and
>> initialize (or re-initialize since cygmon initialize them already) all the
>> necessary components (caches, BTB, MMU, etc...). All this seems to work
>> fine. The memory mapping is changed so that I have SDRAM at address 0x0.
>> 
>> Now I'm a little confused on how to make the serial communication work. I
>> can download an application using GDB. So obviously there is a connection
>> between the target and the GDB in the host. What should I do to keep using
>> this connection to send debugging messages to the host? Or should I break
>> this connection and restart it somehow?
>> 
>> Thanks in advance,
>> Cristiano.
>> 
>> ------------------------------------------------------------
>> Cristiano Ligieri Pereira - http://www.ics.uci.edu/~cpereira

Re: [ECOS] Port problem

[Cristiano Ligieri Pereira]

> You built this CygMon yourself?  If so, what was the source code base?
> 
> Note: we really only support RedBoot on the XScale, not CygMon.

I didn't build it but I have the sources from Intel. The addressesare the
following (file elf-lrh-rom.specs):

*link:
-T bsp.ld%s -Ttext 0x00000000 -Tdata 0xc0000000 %(old_link)

> This tells me that you may have a memory mapping problem.
> 
> eCos programs expect the memory map to be somewhat different than the hardware
> map.  Most particularly, DRAM gets moved to 0x00000000 and the FLASH at 0
> gets moved around as well.  On some XScale platforms, this is even trickier,
> but the basic idea is the same.

That's why I'm forcing to execute PLATFORM_SETUP1 and remapping the memory
there according to what you've just said. But it looks like I'm missing
something then.

> > how should I do about the serial communication with the host? Can I reuse
> > the communication already established with CygMon? Do I need to break it
> > and restart a new connection?
> > 
> 
> No.

The No refers to which question? :-)

> I think you should look into getting RedBoot working as your base debug 
> environment.

Ok. I'll do that. 

Cristiano.

Re: [ECOS] Port problem

[Gary Thomas]

On 17-Aug-2001 Cristiano Ligieri Pereira wrote:
>> > how should I do about the serial communication with the host? Can I reuse
>> > the communication already established with CygMon? Do I need to break it
>> > and restart a new connection?
>> > 
>> 
>> No.
> 

It should not be necessary to restart any communications channel when you
run an eCos program.  It should just work.

> The No refers to which question? :-)
> 
>> I think you should look into getting RedBoot working as your base debug 
>> environment.
> 
> Ok. I'll do that. 
> 

Start with this.  Once you have RedBoot working for the platform, you should
be in much better shape.