Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Joao Vicente]

Hi

I am trying to boot the ppc kernel on a MPC8260 based custom board.
I am using the denx distribution linuxppc_2_4_devel, and I am configuring it using the TQM8260 default configuration, customising some CPM settings for UART and FEC.

I have managed to port the u-boot loader for this board, and am at a stage where I am decompressing the kernel image onto the top of the RAM.
I have debug enabled on the u-boot loader and the last line I see is

## Transferring control to Linux (at address 00000000) ...

And then the board resets when allowed to run free, past this point.

One of the doubts that I have is that when I look into the system.map file, it shows that the start address is at 0xc0000000 (as shown below)

c0000000 T _start
c0000000 T _stext
c000000c T __start
c000002c t __after_mmu_off
c0000060 t turn_on_mmu
c00000c4 T __secondary_hold
c0000100 t Reset
...

To the best of my knowledge the kernel image is not relocatable, so I am wondering if this could be my problem or just one of them ...
Should the map file be showing

00000000 T _start
...

If so, where should I change the .text address so that that happens ?

I am also trying to debug the code from that address with gdb/BDI2000, but I am not getting very far.
Basically I can see the program counter incrementing when I single step, but I cannot see the source even though I am trying to load the symbol table at 0x00000000 with the  'add-symbol-file vmlinux 0x00000000' gdb command, after invoking gdb from the kernel base directory (where vmlinux is)
I also have tried to load the code with 'File|Open Program' command (I am using DDD), and then tried to execute the gdb 'target remote 10.121.10.22:2001' command, and I can see the code (head.S file) on my DDD code window, but the lines are not bound to any address, and I cannot follow the code when I try to step though it.

Again does this not work because the code is not relocatable, or fo I need to use any additional compiler switches that are not switched on the default kernel configuration, or by any other reason I fail to see ?

I would appreciate any help on this issue

Thanks in advance

Joao Vicente,
Software Engineer.
www.spectel.com


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Wolfgang Denk]

Dear Joao,

in message <DEF39A0710293E489D45B10E06645CBD0243B1A7@dub-msx1.spectelcorp.com> you wrote:
>
> I am trying to boot the ppc kernel on a MPC8260 based custom board.
...
> One of the doubts that I have is that when I look into the system.map file, it shows that the start address is at 0xc0000000 (as shown below)
>
> c0000000 T _start

Yes, this is the correct kernel virtual address.

> To the best of my knowledge the kernel image is not relocatable, so I am wondering if this could be my problem or just one of them ...
> Should the map file be showing
>
> 00000000 T _start

No!!!

> If so, where should I change the .text address so that that happens ?

Don't change anything!

> I am also trying to debug the code from that address with gdb/BDI2000, but I am not getting very far.
> Basically I can see the program counter incrementing when I single step, but I cannot see the source even though I am trying to load the symbol table at 0x00000000 with the  'add-symbol-file vmlinux 0x00000000' gdb command, after invoking gdb from the k
> ernel base directory (where vmlinux is)

Just use the vmlinux file _without_ relocating  any  addresses.  Make
sure  to  enable  "MMU  XLAT"  and define "PTBASE" in your BDI config
file.


Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
"If people are good only because they fear punishment, and  hope  for
reward, then we are a sorry lot indeed."            - Albert Einstein

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Wolfgang Denk]

Dear Joao,

in message <DEF39A0710293E489D45B10E06645CBD0243B3B2@dub-msx1.spectelcorp.com> you wrote:
>
> on u-boot IMMR was set to 0x04700000, which is acceptable as far as
> u-boot is concerned, but not allowed once the kernel is initialised.
>
> The other issue was that the Chip Select was not configured properly on
> u-boot (2MB instead of 32MB), which was not noticed in u-boot since the
> u-boot never tried to access memory above that region, but caused the
> kernel to reset the target.
...
> I am using the denx distribution linuxppc_2_4_devel, and I am
> configuring it using the TQM8260 default configuration, customising some
> CPM settings for UART and FEC.

I would like to point out that you certainly did NOT pick up  such  a
bogus memory map from the TQM8260 default configuration, nor from any
other board maintened by DENX.

Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Old programmers never die, they just branch to a new address.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Joao Vicente]

Hi all

Just want to post the resolution to the problem(s) causing the kernel
code to reset the target.

on u-boot IMMR was set to 0x04700000, which is acceptable as far as
u-boot is concerned, but not allowed once the kernel is initialised.

The other issue was that the Chip Select was not configured properly on
u-boot (2MB instead of 32MB), which was not noticed in u-boot since the
u-boot never tried to access memory above that region, but caused the
kernel to reset the target.

I was still not able to debug the kernel (with gdb) before the kernel
turned on the MMU, but since the kernel code is running past the MMU
initialisation, I am now able to debug with source.

I would also like to take this opportunity to thank all the people who
replied to this post.

Best regards

Joao Vicente

-----Original Message-----
From: Joao Vicente
Sent: Thursday, October 09, 2003 10:52 AM
To: linuxppc-embedded@lists.linuxppc.org
Subject: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom
MPC8260 board



Hi

I am trying to boot the ppc kernel on a MPC8260 based custom board.
I am using the denx distribution linuxppc_2_4_devel, and I am
configuring it using the TQM8260 default configuration, customising some
CPM settings for UART and FEC.

I have managed to port the u-boot loader for this board, and am at a
stage where I am decompressing the kernel image onto the top of the RAM.
I have debug enabled on the u-boot loader and the last line I see is

## Transferring control to Linux (at address 00000000) ...

And then the board resets when allowed to run free, past this point.

One of the doubts that I have is that when I look into the system.map
file, it shows that the start address is at 0xc0000000 (as shown below)

c0000000 T _start
c0000000 T _stext
c000000c T __start
c000002c t __after_mmu_off
c0000060 t turn_on_mmu
c00000c4 T __secondary_hold
c0000100 t Reset
...

To the best of my knowledge the kernel image is not relocatable, so I am
wondering if this could be my problem or just one of them ...
Should the map file be showing

00000000 T _start
...

If so, where should I change the .text address so that that happens ?

I am also trying to debug the code from that address with gdb/BDI2000,
but I am not getting very far.
Basically I can see the program counter incrementing when I single step,
but I cannot see the source even though I am trying to load the symbol
table at 0x00000000 with the  'add-symbol-file vmlinux 0x00000000' gdb
command, after invoking gdb from the kernel base directory (where
vmlinux is)
I also have tried to load the code with 'File|Open Program' command (I
am using DDD), and then tried to execute the gdb 'target remote
10.121.10.22:2001' command, and I can see the code (head.S file) on my
DDD code window, but the lines are not bound to any address, and I
cannot follow the code when I try to step though it.

Again does this not work because the code is not relocatable, or fo I
need to use any additional compiler switches that are not switched on
the default kernel configuration, or by any other reason I fail to see ?

I would appreciate any help on this issue

Thanks in advance

Joao Vicente,
Software Engineer.
www.spectel.com


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Joao Vicente]

OK

Does that mean that I cannot debug the source until after the MMU is turned on?

-----Original Message-----
From: Wolfgang Denk [mailto:wd@denx.de]
Sent: Friday, October 10, 2003 2:01 PM
To: Joao Vicente
Cc: linuxppc-embedded@lists.linuxppc.org
Subject: Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a
custom MPC8260 board


Dear Joao,

in message <DEF39A0710293E489D45B10E06645CBD0243B232@dub-msx1.spectelcorp.com> you wrote:
>
> Once inside gdb, I am typing 'target remote 10.121.10.22:2001' after
> which I am obtaing the following output:
> 0x00000000 in ?? ()

OK.

> One thing I cant get my head around is that if the  system.map file
> shows _start address at 0xc000000, how can the gdb relate the symbols on
> vmlinux, assuming they are are expected to be at 0xc0000000?

0xC....... is a kernel virtual address which will be valid as soon as
Linux  turns  on  the  MMU  (which   it   does   very   early   durin
initialization).

> I have also tried to load the symbol file at 0x00000000 with the
> 'add-symbol-file vmlinux 0x00000000' followed by 'target remote
> 10.121.10.22:2001'. Does this make any sense if the code is not
> relocatable ?

No, this makes no sense.

> My intention here is not to have this build to work with my board, but
> to just be able to display the symbols and trace the flow of the kernel,
> so that I can then start customising it for the custom hardware.

Ummm... In my experience a better approach would be to figure out the
necessary changes _before_ even attempting to boot the kernel. Inter-
preting the effects of a misconfigured kernel it not exactly easy  as
funny things might happen.

> In order to customise the build for my board, what would be your
> sugestion. Should I modify config.in in order to launch an independent

Yes. Create a new configuration option for your board.

> config on the build, or use one of the given ones and hard code a
> XXX_CONFIG specific for my board onto include/linux/autoconf.h or some

Modifying include/linux/autoconf.h makes no  sense.  This  file  gets
automagically created dureing the kernel config & build steps.

> For instance, I have noticed that there is specific CONFIG_ADS8260
> #ifdefs throughout the ppc arch source, but no standard configuration is
> available from 'make menuconfig'. Is this because this board's port is
> not complete, or should a default config be selected on menuconfig, and
> the CONFIG_ADS8260 added into a top header file ?

My guess is that this port was only completed in more recent versions
of the (linuxppc_2_4_devel) kernel tree.

> The u-boot README file is an excellent source of documentation for
> porting the u-boot for custom boards, which has been extremelly helpful
> on my exercise on porting u-boot (along with your kind guiding help).
> Do you know of any similar documentation as far as customising the
> kernel for a custom board ?

Except for the source code, no.

Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Philosophy is a game with objectives and no rules.
Mathematics is a game with rules and no objectives.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Wolfgang Denk]

Dear Joao,

in message <DEF39A0710293E489D45B10E06645CBD0243B232@dub-msx1.spectelcorp.com> you wrote:
>
> Once inside gdb, I am typing 'target remote 10.121.10.22:2001' after
> which I am obtaing the following output:
> 0x00000000 in ?? ()

OK.

> One thing I cant get my head around is that if the  system.map file
> shows _start address at 0xc000000, how can the gdb relate the symbols on
> vmlinux, assuming they are are expected to be at 0xc0000000?

0xC....... is a kernel virtual address which will be valid as soon as
Linux  turns  on  the  MMU  (which   it   does   very   early   durin
initialization).

> I have also tried to load the symbol file at 0x00000000 with the
> 'add-symbol-file vmlinux 0x00000000' followed by 'target remote
> 10.121.10.22:2001'. Does this make any sense if the code is not
> relocatable ?

No, this makes no sense.

> My intention here is not to have this build to work with my board, but
> to just be able to display the symbols and trace the flow of the kernel,
> so that I can then start customising it for the custom hardware.

Ummm... In my experience a better approach would be to figure out the
necessary changes _before_ even attempting to boot the kernel. Inter-
preting the effects of a misconfigured kernel it not exactly easy  as
funny things might happen.

> In order to customise the build for my board, what would be your
> sugestion. Should I modify config.in in order to launch an independent

Yes. Create a new configuration option for your board.

> config on the build, or use one of the given ones and hard code a
> XXX_CONFIG specific for my board onto include/linux/autoconf.h or some

Modifying include/linux/autoconf.h makes no  sense.  This  file  gets
automagically created dureing the kernel config & build steps.

> For instance, I have noticed that there is specific CONFIG_ADS8260
> #ifdefs throughout the ppc arch source, but no standard configuration is
> available from 'make menuconfig'. Is this because this board's port is
> not complete, or should a default config be selected on menuconfig, and
> the CONFIG_ADS8260 added into a top header file ?

My guess is that this port was only completed in more recent versions
of the (linuxppc_2_4_devel) kernel tree.

> The u-boot README file is an excellent source of documentation for
> porting the u-boot for custom boards, which has been extremelly helpful
> on my exercise on porting u-boot (along with your kind guiding help).
> Do you know of any similar documentation as far as customising the
> kernel for a custom board ?

Except for the source code, no.

Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Philosophy is a game with objectives and no rules.
Mathematics is a game with rules and no objectives.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Joao Vicente]

Hi Dong-In

I do have -g compiler flag, but my gcc does not recognize -gdb.

Thanks

Joao

-----Original Message-----
From: DONG-IN KANG [mailto:dkang@east.isi.edu]
Sent: Thursday, October 09, 2003 8:03 PM
Subject: RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

 Did you set the compiler flag "-g -gdb" when you configure the kernel?

-----Original Message-----
From: Joao Vicente
Sent: Thursday, October 09, 2003 10:55 AM
Subject: RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

I am affraid I am still unable to debug the kernel.

I have tried to define the line
MMU  XLAT
and defined the PTBASE as shown below

MMU         XLAT        ;translate effective to physical address
PTBASE      0x000000f0  ;address where kernel/user stores pointer to page
table

As you can see I defined PTBASE as 0x000000f0.
Is this the right address for this build ?

I tried it to debug the target with this configuration file but was
still unable to get the symbols.

I notice the BDI user manual suggests to use the following line
WM32    0x000000f0      0x00000000      ;Invalidate page table base.

which I also tried without success.

I also got a trial version of 5.1.1 which I believe you recommended to
fix a similar problem.

"Wolfgang was able to point me to a trial version of gdb which reports
itself as 5.1.1 but which has a patch applied that fixes the problem.
I have attached this version to this email so you can try it for
yourself."

I am not unsure if this is relevant but this is the method I am using to
build the kernel

make TQM8260_config
make menuconfig (where I am changing FCC, MDIO and SCC parameters only)
make oldconfig
make dep
make uImage

Do you reckon is there anything else I could try to fix this problem.

-----Original Message-----
From: Wolfgang Denk [mailto:wd@denx.de]
Sent: Thursday, October 09, 2003 1:38 PM
Subject: Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

in message <DEF39A0710293E489D45B10E06645CBD0243B1A7@dub-msx1.spectelcorp.com> you wrote:
>
> I am trying to boot the ppc kernel on a MPC8260 based custom board.
...
> One of the doubts that I have is that when I look into the system.map
> file, it shows that the start address is at 0xc0000000 (as shown
> below)
>
> c0000000 T _start

Yes, this is the correct kernel virtual address.

> To the best of my knowledge the kernel image is not relocatable, so
> I am wondering if this could be my problem or just one of them ...
> Should the map file be showing
>
> 00000000 T _start

No!!!

> If so, where should I change the .text address so that that happens ?

Don't change anything!

> I am also trying to debug the code from that address with gdb/BDI2000,
> but I am not getting very far. Basically I can see the program counter
> incrementing when I single step, but I cannot see the source even
> though I am trying to load the symbol table at 0x00000000 with the
> 'add-symbol-file vmlinux 0x00000000' gdb command, after invoking gdb
> from the kernel base directory (where vmlinux is)

Just use the vmlinux file _without_ relocating  any  addresses.  Make
sure  to  enable  "MMU  XLAT"  and define "PTBASE" in your BDI config
file.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[DONG-IN KANG]

 Did you set the compiler flag "-g -gdb" when you configure the kernel?

 Dong-In

-----Original Message-----
From: Joao Vicente
Sent: Thursday, October 09, 2003 10:55 AM
Subject: RE: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

I am affraid I am still unable to debug the kernel.

I have tried to define the line
MMU  XLAT
and defined the PTBASE as shown below

MMU         XLAT        ;translate effective to physical address
PTBASE      0x000000f0  ;address where kernel/user stores pointer to page
table

As you can see I defined PTBASE as 0x000000f0.
Is this the right address for this build ?

I tried it to debug the target with this configuration file but was
still unable to get the symbols.

I notice the BDI user manual suggests to use the following line
WM32    0x000000f0      0x00000000      ;Invalidate page table base.

which I also tried without success.

I also got a trial version of 5.1.1 which I believe you recommended to
fix a similar problem.

"Wolfgang was able to point me to a trial version of gdb which reports
itself as 5.1.1 but which has a patch applied that fixes the problem.
I have attached this version to this email so you can try it for
yourself."

I am not unsure if this is relevant but this is the method I am using to
build the kernel

make TQM8260_config
make menuconfig (where I am changing FCC, MDIO and SCC parameters only)
make oldconfig
make dep
make uImage

Do you reckon is there anything else I could try to fix this problem.

-----Original Message-----
From: Wolfgang Denk [mailto:wd@denx.de]
Sent: Thursday, October 09, 2003 1:38 PM
Subject: Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom
MPC8260 board

in message <DEF39A0710293E489D45B10E06645CBD0243B1A7@dub-msx1.spectelcorp.com> you wrote:
>
> I am trying to boot the ppc kernel on a MPC8260 based custom board.
...
> One of the doubts that I have is that when I look into the system.map
> file, it shows that the start address is at 0xc0000000 (as shown below)
>
> c0000000 T _start

Yes, this is the correct kernel virtual address.

> To the best of my knowledge the kernel image is not relocatable, so
> I am wondering if this could be my problem or just one of them ...
> Should the map file be showing
>
> 00000000 T _start

No!!!

> If so, where should I change the .text address so that that happens ?

Don't change anything!

> I am also trying to debug the code from that address with gdb/BDI2000,
> but I am not getting very far. Basically I can see the program counter
> incrementing when I single step, but I cannot see the source even
> though I am trying to load the symbol table at 0x00000000 with the
> 'add-symbol-file vmlinux 0x00000000' gdb command, after invoking gdb
> from the kernel base directory (where vmlinux is)

Just use the vmlinux file _without_ relocating  any  addresses.  Make
sure  to  enable  "MMU  XLAT"  and define "PTBASE" in your BDI config
file.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Problem debugging ppc kernel with DDD/GDB/BDI2000 on a custom MPC8260 board

[Wolfgang Denk]

Dear Joao,

in message <DEF39A0710293E489D45B10E06645CBD0243B1E7@dub-msx1.spectelcorp.com> you wrote:
>
> I am affraid I am still unable to debug the kernel.

Just to be sure: did you add "-g" to  your  CFLAGS_KERNEL  definition
(in the toplevel Makefile) ?

> I am not unsure if this is relevant but this is the method I am using to
> build the kernel
>
> make TQM8260_config

Don't use TQM8260_config unless you either have a  TQM8260  board  or
you know _exactly_ what you're doing.

> make menuconfig (where I am changing FCC, MDIO and SCC parameters only)

Which means your kernel will  not  work  unless  you  really  have  a
TQM8260 board.

> make oldconfig
> make dep
> make uImage

This looks ok to me.

> VECTOR      CATCH       ;catch unhandled exceptions

Did you try out nocatch as recommended earlier?

Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
I've already got a female to worry about. Her name is the Enterprise.
	-- Kirk, "The Corbomite Maneuver", stardate 1514.0

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/