On Mar 07, 2011, at 17:26, Graeme Russ wrote:
On Tue, Mar 8, 2011 at 9:06 AM, Moffett, Kyle D Kyle.D.Moffett@boeing.com wrote:
On Mar 07, 2011, at 16:54, Graeme Russ wrote:
This part does not make much sense - If the CPU is in 'a bad state' then it will probably be lights out anyway. As I understand it, an emergency restart is a restart not initiated by the user (divide by zero, unhandled exception etc), in which case i386 will make use of it
I was considering unhandled exceptions, etc. to be "a bad state" :-D.
Maybe I didn't explain it well enough in the patch summary, but basically the default for "__arch_emergency_restart()" is to just call "__arch_restart()". Since the i386 "__arch_restart()" function should work fine even when U-Boot is in trouble, that architecture does not need to override the default.
Hopefully I am making sense now? Should I reword it from "when the CPU is in a bad state" to "when U-Boot is in trouble", or is there something else that would be easier to understand?
I understand what you are doing now, thanks.
I think you can scrap this part of the description and I will have i386 start using __arch_emergency_restart() for 'Internal U-Boot Errors' such as divide by zero, unhandled exception, general protection faults etc
I don't particularly like the 'emergency' naming - It's like if we don't do it things will blow up :) I think 'automatic' might be a closer term
The name "emergency_restart()" was borrowed from the Linux kernel; the kernel shuts down disks and network interfaces prior to a regular restart but not during an emergency restart. The best analogy is to the "EMERG" log level (KERN_EMERG inside the Linux kernel).
Furthermore, you should *not* directly call __arch_emergency_restart(), that is simply the architecture hook you provide to the generic code. Your exception handlers should call "emergency_restart()" to allow board-specific code to override the reboot, for example by triggering an onboard watchdog to reset other hardware.
EG:
=> some_i386_trap_handler() => emergency_restart() => __board_emergency_restart() => __arch_emergency_restart()
=> do_reset() [The new, generic version] => system_restart() => __board_restart() => __arch_restart()
The __{board,arch}_restart(), etc are just predefined weak functions that can be overridden by the implementation.
For example, my HWW-1U-1A board file effectively does this:
int __board_restart(void) { while (poll_external_software()) { if (ctrlc()) return -1; } return 0; }
void __board_emergency_restart(void) { while (poll_external_software_uninterruptible()) ; }
During a normal restart, I allow the polling to be interrupted with Ctrl-C, but during an emergency restart I don't. In both cases, my function just *returns* to allow the default MPC85xx code to actually perform the hardware-level restart (by poking at the "reset request" bit in a CPU register).
Is there anywhere yet where the code paths for the emergency and non emergency variants differ in the way they end up resetting the board?
There are several U-Boot board ports whose do_reset() functions (now called "__board_restart()") just perform an unconditional jump to "_start" or a FLASH "soft-reset" address. If the system has experienced an unexpected exception or other problem then that is not safe, and it would be better to just hang().
Cheers, Kyle Moffett