A simple (and not optimal) library to fulfill memory access requirements for BCM2711 PCIe

memcpy_unaligned.c

#include <stddef.h>
#include <stdio.h>
#include <stdint.h>
//#include <stdlib.h>

void* memcpy(void* dst, const void* src, size_t n){
	volatile char* vc_src = (char*)src;
	volatile char* vc_dst = (char*)dst;
	// copy byte by byte, hopefully avoiding any alignment issues
	size_t pos = 0;
	
	while(pos < n){
	        if((size_t)(dst+pos) % 4 != 0 || (size_t)(src+pos) % 4 != 0){
        	    // one of the addresses isn't aligned
        	    *(vc_dst+pos) = *(vc_src+pos);
        	    pos++;
        	}
        	if((size_t)(dst+pos) % 4 == 0 && (size_t)(src+pos) % 4 == 0 && n-pos >= 4){
        	    // both are aligned
        	    *(uint32_t*)(dst + pos) = *(uint32_t*)(src + pos);
        	    pos += 4;
        	}
        	
        	if(n-pos < 4 && n-pos != 0){
	            *(vc_dst+pos) = *(vc_src+pos);
	            pos++;
	        }
	}
	return dst;
}

void* memmove(void* dst, const void* src, size_t n){
	void* tmp = __builtin_malloc(n);
	memcpy(tmp,src,n);
	memcpy(dst,tmp,n);
	__builtin_free(tmp);
	return dst;
	if(dst > src){	// start at the last byte
		volatile char* vc_src = (char*)src;
        	volatile char* vc_dst = (char*)dst;
	        // copy byte by byte, hopefully avoiding any alignment issues
	        size_t pos = n; 
        	while(pos > 0){
                        *(vc_dst+pos-1) = *(vc_src+pos-1);  // copy single byte
                	pos--;
	        }
	} else {	// start at the first byte
		volatile char* vc_src = (char*)src;
	        volatile char* vc_dst = (char*)dst;
        	// copy byte by byte, hopefully avoiding any alignment issues
	        size_t pos = 0; 
        	while(pos < n){
                        *(vc_dst+pos) = *(vc_src+pos);  // copy single byte
        	        pos++;
	        }
	}
	return dst;
}

void* memset(void* s, int c, size_t n){
	volatile unsigned char* chr_ptr = (unsigned char*)s;
	unsigned char dat = (unsigned char)c;
	for(size_t i = 0; i < n; i++){
		*(chr_ptr+i) = dat;
	}
	return s;
}

gpu-debug-5.10.y has my current code now. I'll transfer the changes to a newer kernel version once I clean them up.

@Coreforge - That's awesome, thanks! And don't worry about printks/commented code, I just wanted to be able to see it in action, and also be able to pull down and build locally to test on my own 5950 at some point. I hope to do an update video on the graphics cards now that you have one partially working and someone else got the SM750 somewhat working too!

Leaving instructions for use here (for convenience):

wget https://gist.githubusercontent.com/Coreforge/91da3d410ec7eb0ef5bc8dee24b91359/raw/1b72d428b2fe1cba459d5ae7f73663483743ff55/memcpy_unaligned.c

gcc -shared -fPIC -o memcpy.so memcpy_unaligned.c
sudo mv memcpy.so /usr/local/lib/memcpy.so
sudo nano /etc/ld.so.preload

# Put the following line inside ld.so.preload:
/usr/local/lib/memcpy.so