"""Is the sub=0,N=0 variance real uninitialized memory, or just first-call-vs-rest?
Give every repetition a FRESH slot and compare. Also record ta_ret per rep.
"""
import ctypes,struct,sys
from tinygrad import Device
from tinygrad.runtime.autogen.am import am
from tinygrad.helpers import lo32,hi32
d=Device["AMD"]; adev=d.iface.dev_impl; psp=adev.psp
SZ=ctypes.sizeof(am.struct_psp_gfx_cmd_resp); P='/Users/user/dev/amd-psp-re/psp_out/'; S=0x8000; MARK=0x5A
def alloc(n):
    pa=adev.mm.palloc(n,align=0x1000,zero=True); return pa,adev.paddr2mc(pa)
def load(cmd,blob):
    ap,amc=alloc(0x40000); cp,cmc=alloc(S)
    adev.vram.view(ap,len(blob))[:]=blob; adev.gmc.flush_hdp()
    adev.vram.view(psp.cmd_paddr,SZ)[:]=b"\x00"*SZ
    c=am.struct_psp_gfx_cmd_resp(cmd_id=cmd); t=c.cmd.cmd_load_ta
    t.app_phy_addr_hi,t.app_phy_addr_lo=hi32(amc),lo32(amc); t.app_len=len(blob)
    t.cmd_buf_phy_addr_hi,t.cmd_buf_phy_addr_lo=hi32(cmc),lo32(cmc); t.cmd_buf_len=S
    return psp._ring_submit(c).resp.session_id, cp
def inv(sid,shm,buf,tc):
    adev.vram.view(shm,S)[:]=bytes(buf); adev.gmc.flush_hdp()
    adev.vram.view(psp.cmd_paddr,SZ)[:]=b"\x00"*SZ
    ic=am.struct_psp_gfx_cmd_resp(cmd_id=am.GFX_CMD_ID_INVOKE_CMD); iv=ic.cmd.cmd_invoke_cmd
    iv.session_id,iv.ta_cmd_id=sid,tc
    try: psp._ring_submit(ic)
    except RuntimeError: pass
    except TimeoutError: return None
    return bytes(adev.vram.view(shm,S)[:])
asd,_=load(am.GFX_CMD_ID_LOAD_ASD,open(P+'asd_full.bin','rb').read())
dtm,dsh=load(am.GFX_CMD_ID_LOAD_TA,open(P+'dtm_full.bin','rb').read())
hdcp,hsh=load(am.GFX_CMD_ID_LOAD_TA,open(P+'hdcp_full.bin','rb').read())
def topo(h):
    b=bytearray(S); struct.pack_into('<I',b,0,2); struct.pack_into('<10I',b,0x10,h,1,1,0,0,0,0,0,0,0)
    o=inv(dtm,dsh,b,2); return struct.unpack_from('<I',o,8)[0] if o else None
print("bring-up:",[topo(h) for h in range(8)])
def open_slot():
    b=bytearray(S); struct.pack_into('<I',b,0,0x0E); struct.pack_into('<I',b,0xC,0)
    o=inv(hdcp,hsh,b,0x0E); return struct.unpack_from('<I',o,4)[0], struct.unpack_from('<I',o,0x1410)[0]
def run(sub,N,slot):
    ln=5*N+25+1; b=bytearray(S)
    for i in range(0x1400,S): b[i]=MARK
    struct.pack_into('<I',b,0,0x0F); struct.pack_into('<I',b,0xC,slot)
    struct.pack_into('<I',b,0x10,sub); struct.pack_into('<I',b,0x14,ln)
    b[0x28]=sub; b[0x29]=0; b[0x2A]=0; b[0x2B]=N&0xFF; b[0x2C]=0
    for k in range(0x32,min(0x28+ln+1,0x1400)): b[k]=0x41
    o=inv(hdcp,hsh,b,0x0F)
    return (struct.unpack_from('<I',o,4)[0], o[0x1400:0x1480]) if o else (None,None)
print("\nFRESH SLOT per repetition, sub=0 N=0:")
outs=[]
for rep in range(8):
    g,slot=open_slot()
    if g!=0: print("  rep %d: could not open slot (ret=%s) - slots exhausted"%(rep,g)); break
    r,pay=run(0,0,slot)
    outs.append(pay)
    print("  rep %-2d slot=%-3d ta_ret=%-3s payload[0x10:0x30]=%s"%(rep,slot,r,pay[0x10:0x30].hex()))
if len(outs)>=3:
    vary=[i for i in range(len(outs[0])) if len(set(x[i] for x in outs))>1]
    print("\nvarying byte offsets across fresh-slot reps: %d %s"%(len(vary),[hex(0x1400+i) for i in vary[:12]]))
    print("VERDICT:", "NON-DETERMINISTIC (possible uninit disclosure)" if vary else "fully deterministic -> earlier variance was the first-call/later-call state transition, NOT a leak")
print("sos_alive:",psp.is_sos_alive())