_gpu_dxg_r5_submit_size.nx source

1// _gpu_dxg_r5_submit_size.nx -- R5 SAFE size-sweep for LX_DXSUBMITCOMMAND (nr=0x0f). The submit validation 2// requires broadcast_context_count in [1,64]; with count=0 the RECOGNIZED size returns -EINVAL (BEFORE any GPU 3// submit => no TDR risk), wrong sizes return -ENOTTY. So sweep sizes with broadcast_context_count@0x18=0 to pin 4// the struct size. struct d3dkmt_submitcommand: command_buffer@0, command_length@8, flags@0x0c, 5// present_history_token@0x10, broadcast_context_count@0x18, broadcast_context[64]@0x20 ... 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 12 13func p(s: *u8) -> i64 { var nn: i64=0; while s[nn]!=(0 as u8){nn=nn+1} sys_write(1,s,nn); return 0 } 14func 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 } 15func 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 } 16func iowr(nr: i64, size: i64) -> i64 { return (3 << 30) | ((size & 0x3fff) << 16) | (0x47 << 8) | (nr & 0xff) } 17func 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) } 18func 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 } 19 20func query_type(fd: i64, handle: i64, qtype: i64, psize: i64, outv: *i64) -> i64 { 21 let priv: *u8 = sys_mmap(64); var zz: i64 = 0; while zz < 64 { priv[zz] = 0 as u8; zz = zz + 1 } 22 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 23 let h: *i64 = (req as i64 + 0) as *i64; h[0] = (handle & 0xffffffff) | ((qtype & 0xffffffff) << 32) 24 let pd: *i64 = (req as i64 + 8) as *i64; pd[0] = priv as i64 25 let pds: *i64 = (req as i64 + 16) as *i64; pds[0] = psize & 0xffffffff 26 let ret: i64 = sys_ioctl(fd, QAI_CODE, req as i64); outv[0] = rd32(priv, 0); return ret 27} 28func open_from_luid(fd: i64, luid_lo: i64, luid_hi: i64, outh: *i64) -> i64 { 29 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 30 let lo: *i64 = (req as i64 + 0) as *i64; lo[0] = (luid_lo & 0xffffffff) | ((luid_hi & 0xffffffff) << 32) 31 let ret: i64 = sys_ioctl(fd, OAFL_CODE, req as i64); outh[0] = rd32(req, 8); return ret 32} 33func create_device(fd: i64, adapter: i64, outd: *i64) -> i64 { 34 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 35 setu32(buf, 0, adapter & 0xffffffff) 36 let ret: i64 = sys_ioctl(fd, CDEV_CODE, buf as i64); outd[0] = rd32(buf, 12); return ret 37} 38 39func main() -> i64 { 40 p("=== R5 SUBMITCOMMAND SAFE size-sweep (nr=0x0f, broadcast_context_count=0 => -EINVAL no TDR) ===\n" as *u8) 41 let fd: i64 = sys_openat_rd("/dev/dxg" as *u8) 42 if fd < 0 { p("open fail\n" as *u8); sys_exit(1); return 1 } 43 let ainfo: *u8 = sys_mmap(4096); var z: i64 = 0; while z < 4096 { ainfo[z] = 0 as u8; z = z + 1 } 44 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 45 sys_ioctl(fd, ENUM2_CODE, ereq as i64); let nc: i64 = rd32(ereq, 0) 46 let v: *i64 = sys_mmap(16); var disc_lo: i64 = 0; var disc_hi: i64 = 0 47 var ai: i64 = 0 48 while ai < nc { 49 let base: i64 = ai * 20; let eh: i64 = rd32(ainfo, base) 50 let qret: i64 = query_type(fd, eh, 15, 4, v); let t: i64 = v[0] 51 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) } } } 52 ai = ai + 1 53 } 54 let ah: *i64 = sys_mmap(16); open_from_luid(fd, disc_lo, disc_hi, ah) 55 let dh: *i64 = sys_mmap(16); create_device(fd, ah[0], dh) 56 p(" sweeping sizes (broadcast_context_count@0x18=0):\n" as *u8) 57 let sizes: *i64 = sys_mmap(256) 58 sizes[0]=256; sizes[1]=320; sizes[2]=352; sizes[3]=376; sizes[4]=384; sizes[5]=388; sizes[6]=392; sizes[7]=396; sizes[8]=400; sizes[9]=408; sizes[10]=416; sizes[11]=448; sizes[12]=512; sizes[13]=0 59 var si: i64 = 0 60 while sizes[si] != 0 { 61 let sz: i64 = sizes[si] 62 let buf: *u8 = sys_mmap(1024); var bz: i64 = 0; while bz < 1024 { buf[bz] = 0 as u8; bz = bz + 1 } 63 // command_buffer@0=0, broadcast_context_count@0x18=0 -> safe (count check fails before submit) 64 let ret: i64 = sys_ioctl(fd, iowr(0x0f, sz), buf as i64) 65 p(" size=" as *u8); n(sz); p(" code=" as *u8); x(iowr(0x0f, sz)); p(" ret=" as *u8); n(ret) 66 if ret != 0 - 25 { p(" <- RECOGNIZED (non-ENOTTY)" as *u8) } 67 p("\n" as *u8) 68 si = si + 1 69 } 70 sys_close(fd) 71 p("=== R5 SUBMIT SIZE DONE ===\n" as *u8) 72 return 0 73}