I'm working on an armv6 core and have an FIQ hander that works great when I do all of my work in it. However I need to branch to some additional code that's too large for the FIQ memory area.
The FIQ handler gets copied from fiq_start to fiq_end to 0xFFFF001C when registered
static void test_fiq_handler(void)
{
asm volatile("\
.global fiq_start\n\
fiq_start:");
// clear gpio irq
asm("ldr r10, GPIO_BASE_ISR");
asm("ldr r9, [r10]");
asm("orr r9, #0x04");
asm("str r9, [r10]");
// clear force register
asm("ldr r10, AVIC_BASE_INTFRCH");
asm("ldr r9, [r10]");
asm("mov r9, #0");
asm("str r9, [r10]");
// prepare branch register
asm(" ldr r11, fiq_handler");
// save all registers, build sp and branch to C
asm(" adr r9, regpool");
asm(" stmia r9, {r0 - r8, r14}");
asm(" adr sp, fiq_sp");
asm(" ldr sp, [sp]");
asm(" add lr, pc,#4");
asm(" mov pc, r11");
#if 0
asm("ldr r10, IOMUX_ADDR12");
asm("ldr r9, [r10]");
asm("orr r9, #0x08 @ top/vertex LED");
asm("str r9,[r10] @turn on LED");
asm("bic r9, #0x08 @ top/vertex LED");
asm("str r9,[r10] @turn on LED");
#endif
asm(" adr r9, regpool");
asm(" ldmia r9, {r0 - r8, r14}");
// return
asm("subs pc, r14, #4");
asm("IOMUX_ADDR12: .word 0xFC2A4000");
asm("AVIC_BASE_INTCNTL: .word 0xFC400000");
asm("AVIC_BASE_INTENNUM: .word 0xFC400008");
asm("AVIC_BASE_INTDISNUM: .word 0xFC40000C");
asm("AVIC_BASE_FIVECSR: .word 0xFC400044");
asm("AVIC_BASE_INTFRCH: .word 0xFC400050");
asm("GPIO_BASE_ISR: .word 0xFC2CC018");
asm(".globl fiq_handler");
asm("fiq_sp: .long fiq_stack+120");
asm("fiq_handler: .long 0");
asm("regpool: .space 40");
asm(".pool");
asm(".align 5");
asm("fiq_stack: .space 124");
asm(".global fiq_end");
asm("fiq_end:");
}
fiq_hander gets set to the following function:
static void fiq_flip_pins(void)
{
asm("ldr r10, IOMUX_ADDR12_k");
asm("ldr r9, [r10]");
asm("orr r9, #0x08 @ top/vertex LED");
asm("str r9,[r10] @turn on LED");
asm("bic r9, #0x08 @ top/vertex LED");
asm("str r9,[r10] @turn on LED");
asm("IOMUX_ADDR12_k: .word 0xFC2A4000");
}
EXPORT_SYMBOL(fiq_flip_pins);
I know that since the FIQ handler operates outside of any normal kernel API's and that it is a rather high priority interrupt I must ensure that whatever I call is already swapped into memory. I do this by having the fiq_flip_pins function defined in the monolithic kernel and not as a module which gets vmalloc.
If I don't branch to the fiq_flip_pins function, and instead do the work in the test_fiq_handler function everything works as expected. It's the branching that's causing me problems at the moment. Right after branching I get a kernel panic about a paging request. I don't understand why I'm getting the paging request.
fiq_flip_pins is in the kernel at:
c00307ec t fiq_flip_pins
Unable to handle kernel paging request at virtual address 736e6f63
pgd = c3dd0000
[736e6f63] *pgd=00000000
Internal error: Oops: 5 [#1] PREEMPT
Modules linked in: hello_1
CPU: 0 Not tainted (2.6.31-207-g7286c01-svn4 #122)
PC is at strnlen+0x10/0x28
LR is at string+0x38/0xcc
pc : [<c016b004>] lr : [<c016c754>] psr: a00001d3
sp : c3817ea0 ip : 736e6f63 fp : 00000400
r10: c03cab5c r9 : c0339ae0 r8 : 736e6f63
r7 : c03caf5c r6 : c03cab6b r5 : ffffffff r4 : 00000000
r3 : 00000004 r2 : 00000000 r1 : ffffffff r0 : 736e6f63
Flags: NzCv IRQs off FIQs off Mode SVC_32 ISA ARM Segment user
Control: 00c5387d Table: 83dd0008 DAC: 00000015
Process sh (pid: 1663, stack limit = 0xc3816268)
Stack: (0xc3817ea0 to 0xc3818000)
Since there are no API calls in my code I have to assume that something is going wrong in the C call and back. Any help solving this is appreciated.