[U-Boot-Users] Why exceptions are enabled before exception vector code is copied

[U-Boot-Users] Why exceptions are enabled before exception vector code is copied

[Cedric VINCENT]

Dear Wolfgang.

Can I ask you why exceptions are enabled before exception vector code
is copied to the low RAM pages ? (I am talking about U-Boot/PPC)

On my board, U-Boot crashes before relocation (when debugging) since
Xilinx Microprocessor Debugger (XMD) makes a spurious "Machine Check
Exception". For other reasons, such an exception could arrived before
u-boot starts.

Finally, it is a bug or a very strange coding behaviour...

Cedric VINCENT.

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Wolfgang Denk]

In message <c8f10050706004556bcca54@mail.gmail.com> you wrote:
> 
> Can I ask you why exceptions are enabled before exception vector code
> is copied to the low RAM pages ? (I am talking about U-Boot/PPC)

No exceptions are supposed to happen until then.

> On my board, U-Boot crashes before relocation (when debugging) since
> Xilinx Microprocessor Debugger (XMD) makes a spurious "Machine Check
> Exception". For other reasons, such an exception could arrived before
> u-boot starts.

Well, then fix  this  problem.  What  do  you  suggest  hoiw  such  a
situation should be handled while running from flash?

> Finally, it is a bug or a very strange coding behaviour...

It does not happen on other systems.

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The universe does not have laws - it has habits, and  habits  can  be
broken.

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Cedric VINCENT]

> No exceptions are supposed to happen until then.

> What  do  you  suggest  hoiw  such  a
> situation should be handled while running from flash?

There is no reason to enable them before exception vector code is
copied. So exceptions should be enabled *after* exception vector code
is copied.

> It does not happen on other systems.

The spurious "Machine Check Excpetion" is well known on Virtex/PPC.
Most Xilinx FPGA boards are affected (tested on Virtex-II Pro and
Virtex 4).

regards, Cedric VINCENT.

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Wolfgang Denk]

In message <c8f1005070601532cb85bff@mail.gmail.com> you wrote:
> > No exceptions are supposed to happen until then.
> 
> > What  do  you  suggest  hoiw  such  a
> > situation should be handled while running from flash?
> 
> There is no reason to enable them before exception vector code is
> copied. So exceptions should be enabled *after* exception vector code
> is copied.

Feel free to submit a patch.

But what would happen in your case of a Machine Check?  What  do  you
suggest to recover from that?

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The optimum committee has no members.
                                                   - Norman Augustine

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Cedric VINCENT]

> Feel free to submit a patch.

OK, maybe the next week.

> But what would happen in your case of a Machine Check?  What  do  you
> suggest to recover from that?

When this spurious "Machine Check Exception" occured, the MSR didn't
contain any known causes. Maybe the function "MachineCheckException"
(from cpu/ppc4xx/traps.c)  may be modified with something like this :
---------
  #if (CONFIG_COMMANDS & CFG_CMD_KGDB)
      if (debugger_exception_handler && (*debugger_exception_handler)(regs))
          return;
  #endif
 
+ #if defined(CONFIG_USE_XMD)
+     if( SPURIOUS_MCE == (regs->msr & 0x000F0000) )
+         return;
+ #endif

    printf("Machine check in kernel mode.\n");
    printf("Caused by (from msr): ");
    printf("regs %p ",regs);
    switch( regs->msr & 0x000F0000) {
---------
where CONFIG_USE_XMD is a macro (to use with care !) defined by user
when he uses XMD (during debugging stage)  and SPURIOUS_MCE is the MSR
unknown cause.

This (ugly) stuff makes sense to enable exception after relocation. I
didn't test this, since I disable Machine Check Exception when
debugging my Virtex Board. Maybe it will not work (if this spurious
exception occured more than once). I will try to test next week.

Regards,
Cedric VINCENT.

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Cedric VINCENT]

Here comes a patch to enable exceptions after exceptions vector code
is copied (i.e after relocation). I tested it with my board (PPC 405),
I also changed 440 and IOP480 target but without testing...

I run the MAKEALL script : there are no errors, warnings was not due
to this patch.

CHANGELOG : * cpu/ppc4xx/start.S : exceptions are enabled after relocation. 


About XMD, as I thought the spurious exception comes infinitely... So
there is no easy way to fix it. If you apply this patch (this is an
easy way to detect this spurious exception), you should add at the end
of doc/README.ml300 :
========================
6. Warning
----------
 
It was reported that XMD makes (sometimes) a spurious "Machine Check Exception".
When debugging, if you are sure that this exception doesn't come from
your system,
comment this line into cpu/ppc4xx/start.S :

addi	r7,r0,0x1000		/* set ME bit (Machine Exceptions) */
 
========================

Regards,
Cedric VINCENT
-------------- next part --------------
A non-text attachment was scrubbed...
Name: u-boot_exception.diff
Type: application/octet-stream
Size: 2467 bytes
Desc: not available
Url : http://lists.denx.de/pipermail/u-boot/attachments/20050706/1b1e86c7/attachment.obj 

[U-Boot-Users] Re: Why exceptions are enabled before exception vector code is copied

[Wolfgang Denk]

In message <c8f1005070606506c278f6b@mail.gmail.com> you wrote:
> 
> Here comes a patch to enable exceptions after exceptions vector code
> is copied (i.e after relocation). I tested it with my board (PPC 405),
> I also changed 440 and IOP480 target but without testing...
> 
> I run the MAKEALL script : there are no errors, warnings was not due
> to this patch.
> 
> CHANGELOG : * cpu/ppc4xx/start.S : exceptions are enabled after relocation.

Added, thanks.

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Backed up the system lately?