_gpu_dxg_r4f_recon.nx source

1// _gpu_dxg_r4f_recon.nx -- SOVEREIGN-GPU ladder R4f RECON: make the R4d allocation GPU-RESIDENT on the 5080. 2// LX_DXMAKERESIDENT = _IOWR(0x47,0x0b,48) = 0xC030470B. struct d3dddi_makeresident (natural-aligned, 48B): 3// paging_queue@0; alloc_count@4; allocation_list@8 (ptr to array of d3dkmthandle); priority_list@16 (ptr to 4// array of u32 priorities); flags@24; paging_fence_value@32 (OUT); num_bytes_to_trim@40 (OUT). Async (fence). 5// Chain: device -> paging queue -> EXISTINGHEAP alloc -> makeresident. Try priority_list set vs NULL, a couple 6// priority values; print ret + paging_fence_value + num_bytes_to_trim. 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 CPQ_CODE: i64 = 0xC0204707 14const CALLOC_CODE: i64 = 0xC0484706 15const MKRES_CODE: i64 = 0xC030470B 16 17func p(s: *u8) -> i64 { var nn: i64=0; while s[nn]!=(0 as u8){nn=nn+1} sys_write(1,s,nn); return 0 } 18func 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 } 19func 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 } 20func 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) } 21func rd64(buf: *u8, off: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; return q[0] } 22func 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 } 23func setu64(buf: *u8, off: i64, val: i64) -> i64 { let q: *i64 = (buf as i64 + off) as *i64; q[0]=val; return 0 } 24 25func query_type(fd: i64, handle: i64, qtype: i64, psize: i64, outv: *i64) -> i64 { 26 let priv: *u8 = sys_mmap(64); var zz: i64 = 0; while zz < 64 { priv[zz] = 0 as u8; zz = zz + 1 } 27 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 28 let h: *i64 = (req as i64 + 0) as *i64; h[0] = (handle & 0xffffffff) | ((qtype & 0xffffffff) << 32) 29 let pd: *i64 = (req as i64 + 8) as *i64; pd[0] = priv as i64 30 let pds: *i64 = (req as i64 + 16) as *i64; pds[0] = psize & 0xffffffff 31 let ret: i64 = sys_ioctl(fd, QAI_CODE, req as i64); outv[0] = rd32(priv, 0); return ret 32} 33func open_from_luid(fd: i64, luid_lo: i64, luid_hi: i64, outh: *i64) -> i64 { 34 let req: *u8 = sys_mmap(64); var y: i64 = 0; while y < 64 { req[y] = 0 as u8; y = y + 1 } 35 let lo: *i64 = (req as i64 + 0) as *i64; lo[0] = (luid_lo & 0xffffffff) | ((luid_hi & 0xffffffff) << 32) 36 let ret: i64 = sys_ioctl(fd, OAFL_CODE, req as i64); outh[0] = rd32(req, 8); return ret 37} 38func create_device(fd: i64, adapter: i64, outd: *i64) -> i64 { 39 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 40 setu32(buf, 0, adapter & 0xffffffff) 41 let ret: i64 = sys_ioctl(fd, CDEV_CODE, buf as i64); outd[0] = rd32(buf, 12); return ret 42} 43func create_pq(fd: i64, device: i64, outpq: *i64, outfva: *i64) -> i64 { 44 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 45 setu32(buf, 0, device & 0xffffffff) 46 let ret: i64 = sys_ioctl(fd, CPQ_CODE, buf as i64); outpq[0] = rd32(buf, 8); outfva[0] = rd64(buf, 16); return ret 47} 48func alloc_eh(fd: i64, device: i64, sysmem: *u8, outh: *i64) -> i64 { 49 let desc: *u8 = sys_mmap(64); var dz: i64 = 0; while dz < 64 { desc[dz] = 0 as u8; dz = dz + 1 } 50 setu32(desc, 0, 1); setu64(desc, 8, 65536) 51 let entry: *u8 = sys_mmap(256); var ez: i64 = 0; while ez < 256 { entry[ez] = 0 as u8; ez = ez + 1 } 52 setu64(entry, 8, sysmem as i64) 53 let buf: *u8 = sys_mmap(512); var bz: i64 = 0; while bz < 512 { buf[bz] = 0 as u8; bz = bz + 1 } 54 setu32(buf, 0, device & 0xffffffff); setu64(buf, 32, desc as i64); setu32(buf, 40, 0) 55 setu32(buf, 44, 1); setu64(buf, 48, entry as i64); setu32(buf, 56, 0x10020) 56 let ret: i64 = sys_ioctl(fd, CALLOC_CODE, buf as i64); outh[0] = rd32(entry, 0); return ret 57} 58// makeresident; OUT paging_fence@32 -> outf[0], num_bytes_to_trim@40 -> outt[0]. 59func makeresident(fd: i64, pq: i64, alloc_list: *u8, prio_list: i64, flags: i64, outf: *i64, outt: *i64) -> i64 { 60 let buf: *u8 = sys_mmap(256); var bz: i64 = 0; while bz < 256 { buf[bz] = 0 as u8; bz = bz + 1 } 61 setu32(buf, 0, pq & 0xffffffff); setu32(buf, 4, 1) // paging_queue, alloc_count=1 62 setu64(buf, 8, alloc_list as i64) // allocation_list @8 63 setu64(buf, 16, prio_list) // priority_list @16 64 setu32(buf, 24, flags) // flags @24 65 let ret: i64 = sys_ioctl(fd, MKRES_CODE, buf as i64) 66 outf[0] = rd64(buf, 32); outt[0] = rd64(buf, 40); return ret 67} 68 69func main() -> i64 { 70 p("=== R4f RECON: LX_DXMAKERESIDENT -> make the allocation GPU-resident on the 5080 ===\n" as *u8) 71 let fd: i64 = sys_openat_rd("/dev/dxg" as *u8) 72 if fd < 0 { p("open fail\n" as *u8); sys_exit(1); return 1 } 73 let ainfo: *u8 = sys_mmap(4096); var z: i64 = 0; while z < 4096 { ainfo[z] = 0 as u8; z = z + 1 } 74 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 75 sys_ioctl(fd, ENUM2_CODE, ereq as i64); let nc: i64 = rd32(ereq, 0) 76 let v: *i64 = sys_mmap(16); var disc_lo: i64 = 0; var disc_hi: i64 = 0 77 var ai: i64 = 0 78 while ai < nc { 79 let base: i64 = ai * 20; let eh: i64 = rd32(ainfo, base) 80 let qret: i64 = query_type(fd, eh, 15, 4, v); let t: i64 = v[0] 81 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) } } } 82 ai = ai + 1 83 } 84 let ah: *i64 = sys_mmap(16); open_from_luid(fd, disc_lo, disc_hi, ah) 85 let dh: *i64 = sys_mmap(16); create_device(fd, ah[0], dh); let device: i64 = dh[0] 86 let pqh: *i64 = sys_mmap(16); let fvah: *i64 = sys_mmap(16); create_pq(fd, device, pqh, fvah); let pq: i64 = pqh[0] 87 let sysmem: *u8 = sys_mmap(65536); var sz: i64 = 0; while sz < 65536 { sysmem[sz] = 0 as u8; sz = sz + 1 } 88 let alh: *i64 = sys_mmap(16); alloc_eh(fd, device, sysmem, alh); let alloc: i64 = alh[0] 89 p(" device=" as *u8); x(device); p(" pq=" as *u8); x(pq); p(" alloc=" as *u8); x(alloc); p("\n" as *u8) 90 91 let alist: *u8 = sys_mmap(64); setu32(alist, 0, alloc) // allocation_list[0] = alloc handle 92 let plist: *u8 = sys_mmap(64); setu32(plist, 0, 0) // priority_list[0] = 0 93 let of: *i64 = sys_mmap(16); let ot: *i64 = sys_mmap(16) 94 95 p(" [1] prio_list=[0] flags=0:" as *u8); let r1: i64 = makeresident(fd, pq, alist, plist as i64, 0, of, ot); p(" ret=" as *u8); n(r1); p(" fence=" as *u8); n(of[0]); p(" trim=" as *u8); n(ot[0]); p("\n" as *u8) 96 p(" [2] prio_list=NULL flags=0:" as *u8); let r2: i64 = makeresident(fd, pq, alist, 0, 0, of, ot); p(" ret=" as *u8); n(r2); p(" fence=" as *u8); n(of[0]); p(" trim=" as *u8); n(ot[0]); p("\n" as *u8) 97 setu32(plist, 0, 0x78000000) 98 p(" [3] prio_list=[hi] flags=0:" as *u8); let r3: i64 = makeresident(fd, pq, alist, plist as i64, 0, of, ot); p(" ret=" as *u8); n(r3); p(" fence=" as *u8); n(of[0]); p(" trim=" as *u8); n(ot[0]); p("\n" as *u8) 99 sys_close(fd) 100 p("=== R4f RECON DONE ===\n" as *u8) 101 return 0 102}