_gpu_dxg_r4d_alloc3.nx source

1// _gpu_dxg_r4d_alloc3.nx -- SOVEREIGN-GPU ladder R4d ATTEMPT 3: provide the per-alloc OUT priv_drv_data buffer. 2// flags=0x10000 reached the host (0xC000000D INVALID_PARAMETER). WDDM standard allocations RETURN per-alloc 3// driver private data -> the caller must supply entry.priv_drv_data@16 + priv_drv_data_size@24 (NULL there can 4// itself be INVALID_PARAMETER). Sweep: EXISTINGHEAP/CROSSADAPTER x {OUT buf 1024/4096} x {sysmem on/off} x 5// {global priv_drv_data_size 24/0} x {global private_runtime_data on}. Print ret + entry.allocation@0 + 6// entry.priv_drv_data_size@24 readback (host may write REQUIRED size) + gpu_va@40. raw syscalls only. license_tier: ORIGINAL 7import "nx_syscalls.nx" 8 9const ENUM2_CODE: i64 = 0xC0104714 10const QAI_CODE: i64 = 0xC0184709 11const OAFL_CODE: i64 = 0xC00C4701 12const CDEV_CODE: i64 = 0xC0404702 13const CALLOC_CODE: i64 = 0xC0484706 14 15func p(s: *u8) -> i64 { var nn: i64=0; while s[nn]!=(0 as u8){nn=nn+1} sys_write(1,s,nn); return 0 } 16func n(v: i64) -> i64 { let bb: *u8=sys_mmap(28); var m: i64=v; if m<0{m=0-m;sys_write(1,"-" as *u8,1)}; let t: *u8=sys_mmap(28); var k: i64=0; if m==0{t[0]=48;k=1}; while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1}; var i: i64=0; while i<k{bb[i]=t[k-1-i];i=i+1}; sys_write(1,bb,k); return 0 } 17func x(v: i64) -> i64 { p("0x" as *u8); let bb:*u8=sys_mmap(20); var k:i64=0; var m:i64=v; if m==0{bb[0]=48;k=1}; while m>0{ let d:i64=m&15; if d<10{bb[k]=(48+d) as u8}else{bb[k]=(87+d) as u8}; m=(m>>4); k=k+1 } var i:i64=0; let o:*u8=sys_mmap(20); while i<k{o[i]=bb[k-1-i];i=i+1} sys_write(1,o,k); return 0 } 18func rd32(buf: *u8, off: i64) -> i64 { return (buf[off] as i64)|((buf[off+1] as i64)<<8)|((buf[off+2] as i64)<<16)|((buf[off+3] as i64)<<24) } 19func rd64(buf: *u8, off: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; return q[0] } 20func setu32(buf: *u8, off: i64, val: i64) -> i64 { let w: *u8 = (buf as i64 + off) as *u8; w[0]=(val&0xff) as u8; w[1]=((val>>8)&0xff) as u8; w[2]=((val>>16)&0xff) as u8; w[3]=((val>>24)&0xff) as u8; return 0 } 21func setu64(buf: *u8, off: i64, val: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; q[0]=val; return 0 } 22 23func query_type(fd: i64, handle: i64, qtype: i64, psize: i64, outv: *i64) -> i64 { 24 let priv: *u8 = sys_mmap(64); var zz: i64 = 0; while zz < 64 { priv[zz] = 0 as u8; zz = zz + 1 } 25 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 26 let h: *i64 = (req as i64 + 0) as *i64; h[0] = (handle & 0xffffffff) | ((qtype & 0xffffffff) << 32) 27 let pd: *i64 = (req as i64 + 8) as *i64; pd[0] = priv as i64 28 let pds: *i64 = (req as i64 + 16) as *i64; pds[0] = psize & 0xffffffff 29 let ret: i64 = sys_ioctl(fd, QAI_CODE, req as i64); outv[0] = rd32(priv, 0); return ret 30} 31func open_from_luid(fd: i64, luid_lo: i64, luid_hi: i64, outh: *i64) -> i64 { 32 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 33 let lo: *i64 = (req as i64 + 0) as *i64; lo[0] = (luid_lo & 0xffffffff) | ((luid_hi & 0xffffffff) << 32) 34 let ret: i64 = sys_ioctl(fd, OAFL_CODE, req as i64); outh[0] = rd32(req, 8); return ret 35} 36func create_device(fd: i64, adapter: i64, outd: *i64) -> i64 { 37 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 38 setu32(buf, 0, adapter & 0xffffffff) 39 let ret: i64 = sys_ioctl(fd, CDEV_CODE, buf as i64); outd[0] = rd32(buf, 12); return ret 40} 41 42// one configurable EXISTINGHEAP/CROSSADAPTER attempt. dumps ret + key entry readbacks. 43func attempt(fd: i64, device: i64, sysmem: *u8, dtype: i64, use_sysmem: i64, privbuf: *u8, privsize: i64, gpsize: i64, rt: *u8, rtsize: i64) -> i64 { 44 let desc: *u8 = sys_mmap(64); var dz: i64 = 0; while dz < 64 { desc[dz] = 0 as u8; dz = dz + 1 } 45 setu32(desc, 0, dtype); setu64(desc, 8, 65536) // type, existing_heap.size = 64KB 46 let entry: *u8 = sys_mmap(256); var ez: i64 = 0; while ez < 256 { entry[ez] = 0 as u8; ez = ez + 1 } 47 if use_sysmem == 1 { setu64(entry, 8, sysmem as i64) } // sysmem @8 48 if privsize > 0 { setu64(entry, 16, privbuf as i64); setu32(entry, 24, privsize) } // OUT priv_drv_data @16 + size @24 49 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 50 setu32(buf, 0, device & 0xffffffff) 51 if rtsize > 0 { setu64(buf, 16, rt as i64); setu32(buf, 24, rtsize) } // global private_runtime_data @16 + size @24 52 setu64(buf, 32, desc as i64); setu32(buf, 40, gpsize) // standard_allocation @32, priv_drv_data_size @40 53 setu32(buf, 44, 1); setu64(buf, 48, entry as i64) // alloc_count @44, allocation_info @48 54 setu32(buf, 56, 0x10000) // flags: standard_allocation 55 let ret: i64 = sys_ioctl(fd, CALLOC_CODE, buf as i64) 56 p(" ret=" as *u8); n(ret); p(" alloc=" as *u8); x(rd32(entry,0)); p(" priv_sz_rb=" as *u8); n(rd32(entry,24)); p(" gpu_va=" as *u8); x(rd64(entry,40)) 57 if ret == 0 { if rd32(entry,0) != 0 { p(" <== REAL ALLOCATION!!" as *u8) } } 58 p("\n" as *u8) 59 return ret 60} 61 62func main() -> i64 { 63 p("=== R4d ATTEMPT 3: per-alloc OUT priv_drv_data buffer + type/sysmem/size sweep (flags=0x10000) ===\n" as *u8) 64 let fd: i64 = sys_openat_rd("/dev/dxg" as *u8) 65 if fd < 0 { p("open(/dev/dxg) failed\n" as *u8); sys_exit(1); return 1 } 66 let ainfo: *u8 = sys_mmap(4096); var z: i64 = 0; while z < 4096 { ainfo[z] = 0 as u8; z = z + 1 } 67 let ereq: *u8 = sys_mmap(64); ereq[0] = 8 as u8; let r8: *i64 = (ereq as i64 + 8) as *i64; r8[0] = ainfo as i64 68 sys_ioctl(fd, ENUM2_CODE, ereq as i64); let nc: i64 = rd32(ereq, 0) 69 let v: *i64 = sys_mmap(16); var disc_lo: i64 = 0; var disc_hi: i64 = 0 70 var ai: i64 = 0 71 while ai < nc { 72 let base: i64 = ai * 20; let eh: i64 = rd32(ainfo, base) 73 let qret: i64 = query_type(fd, eh, 15, 4, v); let t: i64 = v[0] 74 if qret == 0 { if ((t>>4)&1) == 1 { if ((t>>2)&1) == 0 { disc_lo = rd32(ainfo, base+4); disc_hi = rd32(ainfo, base+8) } } } 75 ai = ai + 1 76 } 77 let ah: *i64 = sys_mmap(16); open_from_luid(fd, disc_lo, disc_hi, ah) 78 let dh: *i64 = sys_mmap(16); create_device(fd, ah[0], dh); let device: i64 = dh[0] 79 p(" device=" as *u8); x(device); p("\n" as *u8) 80 81 let sysmem: *u8 = sys_mmap(65536); var sz: i64 = 0; while sz < 65536 { sysmem[sz] = 0 as u8; sz = sz + 1 } 82 let pbuf: *u8 = sys_mmap(8192); var pz: i64 = 0; while pz < 8192 { pbuf[pz] = 0 as u8; pz = pz + 1 } 83 let rtbuf: *u8 = sys_mmap(4096); var rz: i64 = 0; while rz < 4096 { rtbuf[rz] = 0 as u8; rz = rz + 1 } 84 let nullp: *u8 = 0 as *u8 85 86 p(" [1] EH sysmem=1 outbuf=1024 gpsize=24:" as *u8); attempt(fd, device, sysmem, 1, 1, pbuf, 1024, 24, nullp, 0) 87 p(" [2] EH sysmem=1 outbuf=4096 gpsize=24:" as *u8); attempt(fd, device, sysmem, 1, 1, pbuf, 4096, 24, nullp, 0) 88 p(" [3] EH sysmem=1 outbuf=1024 gpsize=0 :" as *u8); attempt(fd, device, sysmem, 1, 1, pbuf, 1024, 0, nullp, 0) 89 p(" [4] EH sysmem=0 outbuf=1024 gpsize=24:" as *u8); attempt(fd, device, sysmem, 1, 0, pbuf, 1024, 24, nullp, 0) 90 p(" [5] CROSSADAPTER sysmem=0 outbuf=1024:" as *u8); attempt(fd, device, sysmem, 2, 0, pbuf, 1024, 24, nullp, 0) 91 p(" [6] EH sysmem=1 outbuf=1024 +runtime :" as *u8); attempt(fd, device, sysmem, 1, 1, pbuf, 1024, 24, rtbuf, 1024) 92 sys_close(fd) 93 p("=== R4d ATTEMPT 3 DONE ===\n" as *u8) 94 return 0 95}