_gpu_dxg_r5a_probe.nx source

1// _gpu_dxg_r5a_probe.nx -- R5a SUBMIT offset probe (SAFE): find broadcast_context_count + broadcast_context[0] 2// offsets. With command_length=0 (no-op submit: well-formed, the GPU executes 0 bytes => no TDR), a submit whose 3// count is valid AND context resolves returns NON-(-22) (host accepts the empty submit); a wrong offset leaves 4// count=0 or ctx unresolved => -22. Sweep (count_off, ctx_off) candidates. Then do the REAL semaphore submit 5// (length=24) with the working offsets and read the target CPU-side. raw syscalls only. license_tier: ORIGINAL 6import "nx_syscalls.nx" 7 8const ENUM2_CODE: i64 = 0xC0104714 9const QAI_CODE: i64 = 0xC0184709 10const OAFL_CODE: i64 = 0xC00C4701 11const CDEV_CODE: i64 = 0xC0404702 12const CCV_CODE: i64 = 0xC0284704 13const CPQ_CODE: i64 = 0xC0204707 14const CALLOC_CODE: i64 = 0xC0484706 15const MAPVA_CODE: i64 = 0xC068470C 16const MKRES_CODE: i64 = 0xC030470B 17const SUBMIT_CODE: i64 = 0xC180470F 18 19func p(s: *u8) -> i64 { var nn: i64=0; while s[nn]!=(0 as u8){nn=nn+1} sys_write(1,s,nn); return 0 } 20func 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 } 21func 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 } 22func 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) } 23func rd64(buf: *u8, off: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; return q[0] } 24func 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 } 25func setu64(buf: *u8, off: i64, val: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; q[0]=val; return 0 } 26 27func query_type(fd: i64, handle: i64, qtype: i64, psize: i64, outv: *i64) -> i64 { 28 let priv: *u8 = sys_mmap(64); var zz: i64 = 0; while zz < 64 { priv[zz] = 0 as u8; zz = zz + 1 } 29 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 30 let h: *i64 = (req as i64 + 0) as *i64; h[0] = (handle & 0xffffffff) | ((qtype & 0xffffffff) << 32) 31 let pd: *i64 = (req as i64 + 8) as *i64; pd[0] = priv as i64 32 let pds: *i64 = (req as i64 + 16) as *i64; pds[0] = psize & 0xffffffff 33 let ret: i64 = sys_ioctl(fd, QAI_CODE, req as i64); outv[0] = rd32(priv, 0); return ret 34} 35func open_from_luid(fd: i64, luid_lo: i64, luid_hi: i64, outh: *i64) -> i64 { 36 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 37 let lo: *i64 = (req as i64 + 0) as *i64; lo[0] = (luid_lo & 0xffffffff) | ((luid_hi & 0xffffffff) << 32) 38 let ret: i64 = sys_ioctl(fd, OAFL_CODE, req as i64); outh[0] = rd32(req, 8); return ret 39} 40func create_device(fd: i64, adapter: i64, outd: *i64) -> i64 { 41 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 42 setu32(buf, 0, adapter & 0xffffffff) 43 let ret: i64 = sys_ioctl(fd, CDEV_CODE, buf as i64); outd[0] = rd32(buf, 12); return ret 44} 45func create_ctx(fd: i64, device: i64, outc: *i64) -> i64 { 46 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 47 setu32(buf, 0, device & 0xffffffff) 48 let ret: i64 = sys_ioctl(fd, CCV_CODE, buf as i64); outc[0] = rd32(buf, 32); return ret 49} 50func create_pq(fd: i64, device: i64, outpq: *i64, outfva: *i64) -> i64 { 51 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 52 setu32(buf, 0, device & 0xffffffff) 53 let ret: i64 = sys_ioctl(fd, CPQ_CODE, buf as i64); outpq[0] = rd32(buf, 8); outfva[0] = rd64(buf, 16); return ret 54} 55func alloc_eh(fd: i64, device: i64, sysmem: *u8, outh: *i64) -> i64 { 56 let desc: *u8 = sys_mmap(64); var dz: i64 = 0; while dz < 64 { desc[dz] = 0 as u8; dz = dz + 1 } 57 setu32(desc, 0, 1); setu64(desc, 8, 65536) 58 let entry: *u8 = sys_mmap(256); var ez: i64 = 0; while ez < 256 { entry[ez] = 0 as u8; ez = ez + 1 } 59 setu64(entry, 8, sysmem as i64) 60 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 61 setu32(buf, 0, device & 0xffffffff); setu64(buf, 32, desc as i64); setu32(buf, 40, 0) 62 setu32(buf, 44, 1); setu64(buf, 48, entry as i64); setu32(buf, 56, 0x10020) 63 let ret: i64 = sys_ioctl(fd, CALLOC_CODE, buf as i64); outh[0] = rd32(entry, 0); return ret 64} 65func map_va(fd: i64, pq: i64, alloc: i64, outva: *i64) -> i64 { 66 let buf: *u8 = sys_mmap(256); var bz: i64 = 0; while bz < 256 { buf[bz] = 0 as u8; bz = bz + 1 } 67 setu32(buf, 0, pq & 0xffffffff); setu64(buf, 24, 0x10000000000) 68 setu32(buf, 32, alloc & 0xffffffff); setu64(buf, 48, 16); setu64(buf, 56, 1) 69 let ret: i64 = sys_ioctl(fd, MAPVA_CODE, buf as i64); outva[0] = rd64(buf, 88); return ret 70} 71func makeresident(fd: i64, pq: i64, alloc_handle: i64) -> i64 { 72 let alist: *u8 = sys_mmap(64); setu32(alist, 0, alloc_handle & 0xffffffff) 73 let plist: *u8 = sys_mmap(64); setu32(plist, 0, 0) 74 let buf: *u8 = sys_mmap(256); var bz: i64 = 0; while bz < 256 { buf[bz] = 0 as u8; bz = bz + 1 } 75 setu32(buf, 0, pq & 0xffffffff); setu32(buf, 4, 1); setu64(buf, 8, alist as i64); setu64(buf, 16, plist as i64) 76 return sys_ioctl(fd, MKRES_CODE, buf as i64) 77} 78// submit with count at count_off and context at ctx_off, given command_buffer + length. 79func submit(fd: i64, cmdbuf: i64, len: i64, count_off: i64, ctx_off: i64, context: i64) -> i64 { 80 let sb: *u8 = sys_mmap(1024); var sz: i64 = 0; while sz < 1024 { sb[sz] = 0 as u8; sz = sz + 1 } 81 setu64(sb, 0, cmdbuf); setu32(sb, 8, len) 82 setu32(sb, count_off, 1); setu32(sb, ctx_off, context & 0xffffffff) 83 return sys_ioctl(fd, SUBMIT_CODE, sb as i64) 84} 85 86func main() -> i64 { 87 p("=== R5a SUBMIT offset probe (length=0 safe) ===\n" as *u8) 88 let fd: i64 = sys_openat_rd("/dev/dxg" as *u8) 89 if fd < 0 { p("open fail\n" as *u8); sys_exit(1); return 1 } 90 let ainfo: *u8 = sys_mmap(4096); var z: i64 = 0; while z < 4096 { ainfo[z] = 0 as u8; z = z + 1 } 91 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 92 sys_ioctl(fd, ENUM2_CODE, ereq as i64); let nc: i64 = rd32(ereq, 0) 93 let v: *i64 = sys_mmap(16); var disc_lo: i64 = 0; var disc_hi: i64 = 0 94 var ai: i64 = 0 95 while ai < nc { 96 let base: i64 = ai * 20; let eh: i64 = rd32(ainfo, base) 97 let qret: i64 = query_type(fd, eh, 15, 4, v); let t: i64 = v[0] 98 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) } } } 99 ai = ai + 1 100 } 101 let ah: *i64 = sys_mmap(16); open_from_luid(fd, disc_lo, disc_hi, ah) 102 let dh: *i64 = sys_mmap(16); create_device(fd, ah[0], dh); let device: i64 = dh[0] 103 let chh: *i64 = sys_mmap(16); create_ctx(fd, device, chh); let context: i64 = chh[0] 104 let pqh: *i64 = sys_mmap(16); let fvah: *i64 = sys_mmap(16); create_pq(fd, device, pqh, fvah); let pq: i64 = pqh[0] 105 let sm_pb: *u8 = sys_mmap(65536); var a1: i64 = 0; while a1 < 65536 { sm_pb[a1] = 0 as u8; a1 = a1 + 1 } 106 let pbh: *i64 = sys_mmap(16); alloc_eh(fd, device, sm_pb, pbh); let pb_alloc: i64 = pbh[0] 107 let pbva: *i64 = sys_mmap(16); map_va(fd, pq, pb_alloc, pbva); let pb_va: i64 = pbva[0] 108 makeresident(fd, pq, pb_alloc) 109 p(" device=" as *u8); x(device); p(" context=" as *u8); x(context); p(" pb_va=" as *u8); x(pb_va); p("\n" as *u8) 110 p(" probe (count_off,ctx_off) submit len=0 -> ret (!=-22 means context resolved):\n" as *u8) 111 p(" (0x18,0x1c): " as *u8); n(submit(fd, pb_va, 0, 0x18, 0x1c, context)); p("\n" as *u8) 112 p(" (0x18,0x20): " as *u8); n(submit(fd, pb_va, 0, 0x18, 0x20, context)); p("\n" as *u8) 113 p(" (0x14,0x18): " as *u8); n(submit(fd, pb_va, 0, 0x14, 0x18, context)); p("\n" as *u8) 114 p(" (0x1c,0x20): " as *u8); n(submit(fd, pb_va, 0, 0x1c, 0x20, context)); p("\n" as *u8) 115 p(" (0x10,0x14): " as *u8); n(submit(fd, pb_va, 0, 0x10, 0x14, context)); p("\n" as *u8) 116 sys_close(fd) 117 p("=== R5a PROBE DONE ===\n" as *u8) 118 return 0 119}