I've been writing optimized bare metal code for the Raspberry Pi 4 for years without issues. The source code in my case is in Rust, but that shouldn't make a difference. I don't have a Raspberry Pi 3, and since I'm currently writing bare metal code for the 5, that's the one I'm testing this on. I found your conclusion weird, so I wrote some assembly code to make a 32 bit read with an offset to an SIMD register, from an identity mapped hardware register in virtualized device memory that is not 128 bit aligned, to make sure that this is not an architecture oddity, and didn't experience anything unexpected.
Below is the code that I wrote:
And below is a debug session where I tested it:
So, as you can see, at least on the Cortex-A76, a 32 bit read from a 128 bit unaligned hardware register with an offset to a SIMD register, works as expected. Also while I didn't show it, as that would require showing the translation table as well, the MAIR byte corresponding to the memory mapped to that address is 0x00, meaning Device-nGnRnE memory.
How are you reading the register values for debugging?
Below is the code that I wrote:
Code:
unsafe { asm!( "b .", // Spin waiting for the debugger. "ldr {reg}, =0x107D001000", // Always-on UART's base address. "ldr {vreg:s}, [{reg}, #0x18]", // Load value from the flags register. "ldr {reg:w}, [{reg}, #0x18]", // Load value from the flags register. reg = out (reg) _, vreg = out (vreg) _, options (nomem, nostack, preserves_flags) ); }Code:
(lldb) process connect connect://0:3333Process 1 stopped* thread #1, stop reason = signal SIGINT frame #0: 0x000000000008208c-> 0x8208c: b 0x8208c 0x82090: ldr x9, #0x168 0x82094: ldr s0, [x9, #0x18] 0x82098: ldr w9, [x9, #0x18]Target 0: (No executable module.) stopped.(lldb) register write pc 0x82090(lldb) print/x *(int *) 0x107D001018(int) 0x00000090(lldb) nextProcess 1 stopped* thread #1, stop reason = instruction step over frame #0: 0x0000000000082094-> 0x82094: ldr s0, [x9, #0x18] 0x82098: ldr w9, [x9, #0x18] 0x8209c: adrp x9, 11 0x820a0: ldrb w10, [x9, #0x8]Target 0: (No executable module.) stopped.(lldb) register read/x s0 s0 = 0x00000000(lldb) nextProcess 1 stopped* thread #1, stop reason = instruction step over frame #0: 0x0000000000082098-> 0x82098: ldr w9, [x9, #0x18] 0x8209c: adrp x9, 11 0x820a0: ldrb w10, [x9, #0x8] 0x820a4: tbnz w10, #0x0, 0x820c4Target 0: (No executable module.) stopped.(lldb) register read/x s0 s0 = 0x00000090(lldb) register read w9 w9 = 0x7d001000(lldb) nextProcess 1 stopped* thread #1, stop reason = instruction step over frame #0: 0x000000000008209c-> 0x8209c: adrp x9, 11 0x820a0: ldrb w10, [x9, #0x8] 0x820a4: tbnz w10, #0x0, 0x820c4 0x820a8: mov w10, #0xaTarget 0: (No executable module.) stopped.(lldb) register read w9 w9 = 0x00000090(lldb) process detachProcess 1 detachedHow are you reading the register values for debugging?
Statistics: Posted by Fridux — Fri May 17, 2024 12:55 am