On Wed, May 15, 2013 at 3:26 PM, Albert ARIBAUD albert.u.boot@aribaud.net wrote:
Hi Rob,
On Wed, 15 May 2013 14:56:07 -0500, Rob Herring robherring2@gmail.com wrote:
From: Rob Herring rob.herring@calxeda.com
interrupt_init also sets up the abort stack, but is not setup before relocation. So any aborts during relocation will hang and not print out any useful information. Fix this by moving the interrupt_init to after the stack setup in board_init_f.
Signed-off-by: Rob Herring rob.herring@calxeda.com
arch/arm/lib/board.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/arch/arm/lib/board.c b/arch/arm/lib/board.c index 09ab4ad..6dbe7e2 100644 --- a/arch/arm/lib/board.c +++ b/arch/arm/lib/board.c @@ -447,6 +447,7 @@ void board_init_f(ulong bootflag) addr_sp += 128; /* leave 32 words for abort-stack */ gd->irq_sp = addr_sp; #endif
interrupt_init(); debug("New Stack Pointer is: %08lx\n", addr_sp);I fail to understand how this is even supposed to work through relocation: exception vectors are not relocated, so if they work before relocation, then they won't work any more afterward unless some code is added to relocate them.
They work before and after in my testing. The vectors are relocated along with the rest of u-boot and the vector base is updated by c_runtime_cpu_setup. I'm simply setting up the abort stack earlier. An alternative would be to setup a different abort handler and stack before relocation. I'm open to suggestions, but as it stands now aborts before or during relocation will simply hang without this.
Also: if this patch is moving interrupt_init(), then where is the line where a call to interrupt_init() is removed?
Yes, the later call to interrupt_init should be removed assuming we keep this approach.
Rob