[SLOVED] Aarch64 - "stp q0, q0, [x8, #224]" hangs my Pi3

[Schnoogle]

Hi there,

while playing around with aarch64 mode I've seen my Raspberry Pi hangs at a specific instruction.
The compiler is generating:

Code: Select all

stp	q0, q0, [x8, #224]

and I've proven with handwritten code at a specific known location that this code hangs my Pi.

Does anyone could point me into a direction what might be a reason for this?
I've checked that the actual core is running in EL1 and that I've enabled to use FP/NEON instructions with cpacr_el1 register.

I've also cross-checked behaviour in lates QEMU version and there the code gets executed fine (no exception thrown or the like)

Thanks in advance for your support.
BR
Schnoogle

[DoubleCheck]

I suppose you haven't enabled the MMU?

It's an alignment problem.

[LdB]

What value is in x8? It should to be 16 byte aligned.

[Schnoogle]

I suppose you haven't enabled the MMU?

It's an alignment problem.

Hi, I checked with and without MMU beeing enabled. Some behaviour.

What value is in x8? It should to be 16 byte aligned.

Well actually this is the stackpointer. The compiler generates code like this:

Code: Select all

sub sp, sp, #0x130
    add x8, sp, #0x8

to set X8 to a specific part of the current stack frame.

However, you pointed me into the right direction (thanks for that )and I disabled the alignment checks with this:

Code: Select all

mrs		x1, sctlr_el1
    mov     x2, #(1 << 1 | 1 << 3 | 1 << 4) 
	bic     x1, x1, x2  // clear SA0, SA and A bits
	msr     sctlr_el1, x1
	isb

And the code does work now, BUT only when MMU is enabled. So I guess only the enabled MMU can handle the unaligned access ???

[LdB]

Yes but more specifically you have broken the AAPCS64 calling convention somehow

AARCH64 makes an absolute requirement that when calling external functions the stack must be 16byte aligned
GCC will default -mpreferred-stack-boundary=4 meaning all its stack stuff is 16byte aligned

So what are you doing to the stack that it is getting so upset about?
Specifically I am querying are you inlining 64bit code assembler on a 32 bit version of linux which will have 8 byte stack alignment

[Schnoogle]

So what are you doing to the stack that it is getting so upset about?
Specifically I am querying are you inlining 64bit code assembler on a 32 bit version of linux which will have 8 byte stack alignment

Hey,

well, interestingly I'm dooing complete bare-metal stuff in Aarch64. So there is no Linux running nor is it a mixture of 32bit and 64bit. I'm using ARM's Aarch64-elf cross compiler toolchain to build the RasPi bare metal kernel on my Windows machine.

The only "speciality" might be, that I'm not compiling from C but rather using Rust and I'm not sure whether I could pass the "preferred-stack-boundary" flag to the rust compiler. I'll check in the Rust forum as well

[Schnoogle]

Hi there,

it turned out to really be an issue with the rust compiler. Adding a strict alignment flag solved the issue by generating code that complies to the required alignment rules...

https://users.rust-lang.org/t/rust-comp ... talestudio