Each started context had two host threads, polling every 10 ms of wall
time: one sent queued packets, received, and ran hellos, keepalives and
timeouts, and the other handed events to the callbacks. A single
CoreTiming event now does both for every running context, every 1 ms of
emulated time, and Stop does what the threads did on their way out.
Matching no longer depends on host scheduling or wall-clock delays, so an
accept's ESTABLISHED comes 1-10 ms after SelectTarget at any emulation
speed, as on a PSP (pspautotests net/adhoc/pair/matchingaccept).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A PSP reports the pool size less 0x20 and a maximum of 0 until anything
is allocated (net/adhoc/matching, recorded on a PSP). With a context the
faked half-pool maximum stays, since that isn't measured.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Fixes#22331
A child's GetMembers starts with itself, then switches to the parent
first one keepalive interval after it was accepted
(net/adhoc/pair/matchingorder, recorded on two PSPs). We kept it self
first. Bomberman's client checks that the first member is the host
before it marks itself ready, so its lobby stayed empty and the host's
start failed with "failed to connect to the game".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CancelTarget on an incoming P2P request marked the peer timed out. It
only escaped a TIMEOUT if one of its hellos came in before the next
timeout check, which it no longer does now that timeouts are checked
all the time. On a PSP it's a peer known from its hellos again, with no
TIMEOUT (pspautotests net/adhoc/pair/matching3).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Where pspautotests net/adhoc/timing or net/adhoc/pair/timing measured
a value, use it:
- Call costs: PdpSend 21 us (10 broadcast), PtpSend 122, PtpFlush 196,
GetPdpStat 1, GetPtpStat 2, PollSocket 18, GetScanInfo 2, GetPeerList
and MatchingGetMembers 4 (no longer a 100 us reschedule),
MatchingStart 416 instead of 30 ms, sceNetAdhocctlTerm 154,
sceNetAdhocTerm 10 and sceNetTerm 209 instead of 10 ms each,
GameModeUpdateMaster 80, UpdateReplica 9.
- Game mode data goes out every 12 ms, not every 0.5 ms.
- DISCONNECT comes 640 ms after ExitGameMode, not 105 ms.
- CONNECT after a Join no longer waits for a peer: a PSP delivers it
before the host shows up in the peer list.
Several of the old values were tuned for particular games (BattleZone's
frame rate, Fate Unlimited Codes and Bleach through PollSocket), and
may have been making up for other differences that are fixed now.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A PSP checks a PTP connection's oldest unacknowledged data every
rexmt_int, counting from when the connection was established, and drops
the connection at the first check where that data is older than
(rexmt_cnt + 1) * rexmt_int. A matching peer times out keepalive_int *
init_count after its last keepalive, checked all the time instead of on
the console's own keepalive tick, which could add up to an interval.
Both measured with several settings in pspautotests
net/adhoc/pair/vanishtiming; DATA_TIMEOUT already matched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The vanish pair test (recorded on two PSPs) now passes on both sides,
over the relay:
- From the Disconnect call, a console's PDP and PTP sends succeed and go
nowhere, and nothing reaches it.
- A PTP connection to a peer that has left the group fails with
DISCONNECTED after rexmt_int * rexmt_cnt. Relay sockets now get their
retry settings (and are zeroed), and an accepted one counts as
connected.
- Matching data to a peer that has left gets DATA_TIMEOUT after the
retransmissions instead of a DATA_ACK.
- The matching timeout for a silent peer is keepalive_int * init_count,
plus MATCHING_TIMEOUT_EXTRA_US (0) to tune for internet latency. It
was (keepalive + rexmt) * count + 500ms, later than a PSP's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The rule that dropped the first join after the hello data changed came
from a race in pspautotests net/adhoc/pair/matching2, where the host
started looking for the REQUEST after it had come. A PSP reports the
join at once however soon after SetHelloOpt it arrives, also when the
host changes its hello data every second (net/adhoc/pair/matchingjoin).
Ridge Racer does that every 5 seconds, so its clients could never enter
a race.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The matching, matching2 and matching3 pair tests (recorded on two PSPs)
now pass on both sides:
- Event and input messages are queued in order. They were stacks, so a
burst came out reversed (PONG and PANG crossed), and two places spawned
their messages backwards to compensate.
- A peer stays busy until its DATA_ACK is delivered (DATA_BUSY), and
AbortSendData discards that DATA_ACK.
- A host's SelectTarget while its accept is in progress gives
REQUEST_IN_PROGRESS.
- A child that leaves, a cancelled request and a denied request (client
and P2P) go back to a known non-member peer: no TIMEOUT, hellos keep
coming, and it can ask again; a cancelled request gives
TARGET_NOT_READY. A client says BYE to a host it has left.
- After SetHelloOpt the next join from an unknown peer reaches the host
without a REQUEST.
- GetMembers lists the console itself first, also on a client.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A stack size too small to be one (where a partition was passed) gives
STACKSIZE_TOO_SHORT, and a partition outside 1-6 then INVALID_ID
(pspautotests net/adhoc/matching). How small a stack still works isn't
known, so only sizes below 0x100 are refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fake matching thread was started through sceKernelStartThread,
which switched threads inside the call, and the game got the new
thread's ID back. On a PSP, Start returns 0 (pspautotests
net/adhoc/matching). Create and start it like adhocctl's thread instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A child the host accepts right after another one learns about that one
twice: from the sibling list in its accept packet, and from the BIRTH
packet the host sends about it. It added it both times, so with three or
four players the last one to join had an extra member (pspautotests
net/adhoc/group/matching, not recorded on PSPs yet).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Several comments called PSP behaviour a workaround for one game: PDP
send and receive returning 0 and writing the length back, Scan while
connected reporting ALREADY_CONNECTED, PtpOpen connecting by itself,
connecting a connected socket returning 0, PtpAccept's optional peer
address and port, the malloc stat's pool size, and matching dropping
hellos from a host mid-join. The pspautotests recorded on PSPs show the
PSP does all of these, so say that, and keep the games as examples.
Also bumps pspautotests for the new adhocctl Disconnect events and the
pair/matching3 test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CancelTarget on a MAC that isn't a peer, and Start on a context ID that
doesn't exist, faked success for every game. On the PSP they fail with
UNKNOWN_TARGET and INVALID_ID (pspautotests net/adhoc/matching). Only
the games they were added for keep them now:
MatchingCancelUnknownTargetSucceeds for Soulcalibur: Broken Destiny, and
MatchingStartUnknownIdSucceeds for GTA: Vice City Stories.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From the net/adhoc/matching and limits pspautotests, recorded on PSPs:
- Create: a bad mode gives INVALID_MODE, rxbuflen has to be at least
6 * maxnum + 4, host and P2P contexts need nonzero hello and keep-alive
intervals, init_count can't be negative and rexmt_int can't be 0. At
most 15 contexts exist at once.
- Start on a running context gives IS_RUNNING. Start2 takes each
thread's partition after its stack size, not before.
- Stop and Delete on unknown IDs give INVALID_ID, and Delete on a running
context IS_RUNNING (internal cleanup still stops and deletes).
- GetHelloOpt fails before Start, copies at most the game's buffer and
leaves the length alone. It used to write the whole hello past the
buffer.
CancelTarget on an unknown peer still succeeds, as Soul Calibur needs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
HLEFunc and ReplaceFunc now take the MIPSState they run on. The dispatch
paths (both interpreters and every JIT backend) pass in the state they
belong to, and CallSyscall*, hleFinishSyscall and the HLE mips call
machinery use it instead of currentMIPS.
The wrappers in FunctionWrappers.h take the function as `auto` and call it
through HLECall, which also accepts an implementation with an extra leading
MIPSState * parameter and passes the state to it. Any other signature still
fails to compile, as before. PARAM/RETURN now refer to a local `mips`.
The HLE functions (and replacement functions) that used currentMIPS now take
the state that way. Callers outside a syscall's own frame - CoreTiming
events, the adhoc matching thread, dialogs, PPGe, the debugger UIs - pass
currentMIPS or &mipsr4k explicitly. The kernel's thread internals still use
currentMIPS.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
NetAdhocMatching_Delete unlinked the context before calling Stop, which then
couldn't find it, so the input and event threads were never joined and
deleting the context destroyed joinable std::threads (std::terminate). Seen
on Android when shutting down a game with matching running.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Loading a savestate kept the current session's netAdhocInited,
netAdhocctlInited and matching-inited flags rather than the saved ones.
Loaded into a fresh boot, a state saved with adhocctl initialized (but not
connected) then failed every adhocctl call with NOT_INITIALIZED. MGS Peace
Walker shuts ad hoc down before a cutscene and retried
sceNetAdhocctlDisconnect forever, leaving the screen black.
Take the flags from the state, and bring the host side in line: start the
friendFinder and log in again when the state has adhocctl up and we don't,
stop it in the opposite case. Also save the product code, which that login
sends.
Also add a game.state.save debugger command, used to reproduce this.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Matching packet handlers added untrusted payload lengths to fixed header
sizes before comparing them with the datagram length. Large values could
overflow and permit out-of-bounds reads. Compare against the remaining
packet size instead, and reject negative ACCEPT sibling counts.
With WLAN on and no server reachable, sceNetAdhocctlInit keeps its thread
waiting for the login. The load dropped the request, and the wait ended in
BUSY, which Init can't retry. Splinter Cell then ran its failure path with
a deleted event flag. Also stop freeing matching event buffers into the
restored allocator, and take the event lock when clearing.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
