Le 28/02/2014 09:37, Sascha Hauer a écrit :
On Fri, Feb 28, 2014 at 09:21:40AM +0100, Philippe Rétornaz wrote:
Le 27/02/2014 21:36, Sascha Hauer a écrit :
On Thu, Feb 27, 2014 at 03:03:45PM +0100, Philippe Rétornaz wrote:
+void __bare_init __naked barebox_arm_reset_vector(void)
+{
+ uint32_t r;
+
+ arm_cpu_lowlevel_init();
+
+ /* Enable IPU Display interface */
+ writel(1 << 6, MX31_IPU_CTRL_BASE_ADDR);
+
+ writel(0x074B0BF5, MX31_CCM_BASE_ADDR + MX31_CCM_CCMR);
+
+ asm volatile("1:\n\t"
+ "SUBS %0, %0, #1 \n\t"
+ "BNE 1b \n\t"
+ : : "r" (0x4000) : "cc");
You can write a delay loop in c with:
volatile int c;
for (c = 0; c < 0x4000; c++)
Well, no, at least not on my toolchain. Because the volatile ask gcc to
not optimize the variable, it then put it on the stack. But the stack
pointer is not yet initialized, so it crashes. I've tried with a
barrier() instead of the volatile, but it leads to the same assembly
(which is not surprising). Here is the compiled code with your suggestion:
ldr r2, .L9+8
b .L2
.L3:
ldr r3, [sp, #4]
add r3, r3, #1
str r3, [sp, #4]
.L2:
ldr r3, [sp, #4]
cmp r3, r2
ble .L3
With L9+8:
.word 16383
But the stack pointer is initialised only in barebox_arm_entry() which
is called later. So I decided that a two instructions assembly loop was
the simplest solution.
This may happen because the function gets too complex and gcc starts
using the stack in this case.
Try rewriting the lowlevel stuff as:
static void __noinline mx31_moboard_startup(void)
{
/* Put setup here */
}
void __bare_init __naked barebox_arm_reset_vector(void)
{
arm_cpu_lowlevel_init();
arm_setup_stack(MX31_IRAM_BASE_ADDR + MX31_IRAM_SIZE - 12);
mx31_moboard_startup();
}
With this you can use the stack and should be on the safe side.
Ok, I tried and it works fine so I will do this.
BTW, any reason for the -12, why not -8 ?
Thanks,
Philippe
_______________________________________________
barebox mailing list
barebox@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/barebox
