[Qemu-devel] Cortex m3 initialisation

[Qemu-devel] Cortex m3 initialisation

[Benjamin Bagland]

[-- Attachment #1.1: Type: text/plain, Size: 742 bytes --]

Hi,

I'm having an issue while trying to use the lm3s811 board emulation with
qemu 0.12.3 or git. It works fine however with qemu-0.11.1.

The problem comes from hw/armv7m.c, around line 231, the initial PC and SP
are set by reading from physical memory, however the rom data is not copied
in physical memory until the reset later on in the initialisation (comes
from commit 45a50b1668822c23afc I think). Since the initial PC is invalid,
the cpu goes straight into UsageFault instead of going to the proper reset
routine.

I've made a small patch to read from rom instead of physical memory, that
fixes the issue with my test program but I'm not sure if it's the proper
thing to do, or even if it works with other arm boards.

Rgds,
Benjamin

[-- Attachment #1.2: Type: text/html, Size: 816 bytes --]

[-- Attachment #2: init.patch --]
[-- Type: application/octet-stream, Size: 573 bytes --]

diff --git a/hw/armv7m.c b/hw/armv7m.c
index 034323d..674a8b4 100644
--- a/hw/armv7m.c
+++ b/hw/armv7m.c
@@ -231,8 +231,10 @@ qemu_irq *armv7m_init(int flash_size, int sram_size,
        regular ROM image and perform the normal CPU reset sequence.
        Otherwise jump directly to the entry point.  */
     if (lowaddr == 0) {
-	env->regs[13] = ldl_phys(0);
-	pc = ldl_phys(4);
+        /* roms are not loaded in physical memory until reset */
+        uint32_t *rom = rom_ptr(0);
+        env->regs[13] = rom[0];
+        pc = rom[1];
     } else {
 	pc = entry;
     }

Re: [Qemu-devel] Cortex m3 initialisation

[andrzej zaborowski]

On 14 March 2010 15:56, Benjamin Bagland <bagland@ashesoftime.net> wrote:
> Hi,
>
> I'm having an issue while trying to use the lm3s811 board emulation with
> qemu 0.12.3 or git. It works fine however with qemu-0.11.1.
>
> The problem comes from hw/armv7m.c, around line 231, the initial PC and SP
> are set by reading from physical memory, however the rom data is not copied
> in physical memory until the reset later on in the initialisation (comes
> from commit 45a50b1668822c23afc I think). Since the initial PC is invalid,
> the cpu goes straight into UsageFault instead of going to the proper reset
> routine.
>
> I've made a small patch to read from rom instead of physical memory, that
> fixes the issue with my test program but I'm not sure if it's the proper
> thing to do, or even if it works with other arm boards.

The patch seems to be generated on top of some code that is not in the
repository, but in short, I think both loading the kernel, roms, and
setting of pc and sp should be done only on reset.  There's probably a
bootloader somewhere that sets the pc and sp on real hw.

Cheers