mirror of
https://github.com/phantomptr/ps5upload.git
synced 2026-10-22 03:42:55 -05:00
phantomptrandClaude Opus 5.5
c14782944d
Upload, install, convert and manage PS5 game packages over the network. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
75 lines
3.6 KiB
C
75 lines
3.6 KiB
C
#ifndef PS5UPLOAD2_WAKE_WATCHDOG_H
|
|
#define PS5UPLOAD2_WAKE_WATCHDOG_H
|
|
|
|
#include <stdint.h>
|
|
|
|
/*
|
|
* Rest-mode wake watchdog.
|
|
*
|
|
* After the PS5 wakes from rest mode (suspend-to-RAM), the kernel resets
|
|
* process credentials. Our payload's ucred elevation becomes stale —
|
|
* Sony APIs silently fail and the fan threshold pin is lost.
|
|
*
|
|
* The watchdog treats a long delay as a possible wake event only when both
|
|
* REALTIME and MONOTONIC clocks advance by the threshold. Requiring two
|
|
* independent signals prevents NTP, manual time changes, and client clock
|
|
* sync from turning a wall-clock adjustment into permission for kernel
|
|
* writes. On detection we:
|
|
*
|
|
* 1. Check if a BigApp (game) is running — if so, skip. The PS5
|
|
* throttles background threads when a game is foreground, which
|
|
* can cause a 5s sleep to overrun 60s without any actual suspend.
|
|
* A real resume from rest kills all games first, so a live BigApp
|
|
* means we were never actually suspended.
|
|
* 2. Force re-elevation of ucred credentials.
|
|
* 3. Re-apply the persisted fan threshold.
|
|
* 4. Fire a toast so the user sees the re-activation on screen.
|
|
*
|
|
* Mount reconciliation is deliberately excluded: getmntinfo can hang on
|
|
* some firmware/loader combinations and must not run from this detached
|
|
* heuristic thread.
|
|
*
|
|
* MEASURED 2026-08-31, and it bounds what this thread can ever be worth:
|
|
* on FW 5.10 and FW 9.60 the payload does NOT survive rest mode at all. A
|
|
* real standby/wake cycle leaves the ELF loader (9021) listening and the
|
|
* helper's own ports (9113/9114) closed — the process is gone, so no
|
|
* re-escalation can run and the desktop's auto-loader re-sends the payload
|
|
* instead. On those firmwares every branch this thread can reach is
|
|
* therefore a FALSE positive. That asymmetry — unproven upside, proven
|
|
* downside — is why the gate below is deliberately hard to satisfy.
|
|
*
|
|
* A consequence worth stating plainly: if some firmware does keep the
|
|
* process across suspend but does not advance CLOCK_MONOTONIC over it,
|
|
* this gate never fires and recovery defers to that same payload re-send.
|
|
* That is the intended trade. Do not loosen the gate to "fix" it without
|
|
* first proving, on hardware, that a suspended payload survives.
|
|
*
|
|
* Algorithm adapted from elf-arsenal's wake_watchdog_thread (sys.c:1855).
|
|
* Threshold is 60s (raised from elf-arsenal's original 15s for the same
|
|
* reason: PS5 thread throttling under game load).
|
|
*
|
|
* The thread is detached and self-contained — safe to start once at
|
|
* payload boot and never join.
|
|
*/
|
|
|
|
/* Launch the wake watchdog as a detached thread. Safe to call once
|
|
* from main() after the fan threshold restore block. The thread runs
|
|
* for the lifetime of the payload process. */
|
|
void start_wake_watchdog(void);
|
|
|
|
/* Pure decision gate used by the watchdog and its host self-test.
|
|
*
|
|
* A wall-clock jump alone is not evidence of rest mode: Sony NTP, the new
|
|
* clock-sync feature, or a manual date correction can all move REALTIME by
|
|
* hours while the process has only slept five seconds. Re-running kernel
|
|
* ucred writes on that weak signal is unsafe, especially on new firmware.
|
|
* Require an independent monotonic suspend-sized gap and a successful proof
|
|
* that no BigApp is running. `bigapp_state`: 0=no app, 1=app, -1=unknown. */
|
|
static inline int wake_watchdog_should_recover(int64_t realtime_elapsed,
|
|
int64_t monotonic_elapsed,
|
|
int bigapp_state) {
|
|
return realtime_elapsed >= 60 && monotonic_elapsed >= 60 && bigapp_state == 0;
|
|
}
|
|
|
|
#endif /* PS5UPLOAD2_WAKE_WATCHDOG_H */
|