Repository navigation
Returnal (PPSA01284) #17930
snoozedDev
started this conversation in
Game bring-up
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Existing thread
Returnal,PPSA01284andPPSA01285; the only earlier reports are the closed issues libkernel: pthread_attr_setsolosched_np stops Returnal (PPSA01285); fix branch ready #3148 and Returnal (PPSA01285) on Windows: intermittent fault in a frame-pointer stack walk at app.exe+0x337aa9d (UE4 Main Thread, read of 0x8) #4239, both Windows.)Title
Returnal (PPSA01284, US; #4239 reported the EU release, PPSA01285)
Content version tested
02.000.003 (
sce_sys/param.json; sdkVersion 0x0200000000000000)AnyPS5 commit
main39a1728 (2026-11-09, includes #16974), nothing on top. Runs earlier the same day onmaine37e9fa (before #16974) are under "Stops seen and their state" below.System tested
Linux 7.2.8 (Arch-based), GCC 16.2.1, Release with
-DANYPS5_ENABLE_SPIRV_TOOLS=ON, RTX 3090 (NVIDIA 615.78.08), 64 GiB RAM. Two runs on the same host's integrated GPU (RADV, RAPHAEL_MENDOCINO). Headless gamescope 3.16.31 at 2304x1296, 60 Hz; the title's swapchain is 1381x777.Furthest point
Logos or loading screen
What is shown
The PlayStation Studios intro and logo and the Housemarque logo, at about 50 fps, for the first 25 s (frames read back on the GPU with
APS5_DUMP_FRAMESshow the intro at 7 s, the Studios logo at 11 s, the Housemarque logo at 14 to 16 s). Every presented frame read back from 17 s on is black (frames 17 to 32 of the frame dump are a single colour, value 0), and from about 30 s on the RTX 3090 the title presents 14 to 17 frames per 10 s with an empty shader cache (27 to 28 from about 870 s on) and 27 to 29 per 10 s with a warm one, until the 900 s cap. The last[present]line of the warm run:GPU busy 18.6 ms, idle gaps 342.3 msper frame. Pad presses sent withtools/pad_input.pyat 121 s (cross), 151 s (options), 181 s (cross, 600 ms), 211 s (down), 225 s (cross), 261 s (triangle) and 291 s (circle) changed neither the present rate nor the frames. The driver's[texture] N storage imagescount grows by about 70 per 10 s for the whole run (34,150 at 900 s cold; 61,300 at 900 s warm, which the[texture]line gives as 94,053 MiB), while the process held about 3.5 GiB of the card. The title's only output is the lineXYXYXY XYXYXY Clearnupworld <address> -> 0, repeated about as often as it presents (2,422 lines in the 900 s warm run).has mixed DCC keysis logged 32 times in each run that kept presenting past the logos (runs 1, 3, 4, 5 and 7 below). Audio and input were not checked beyond the pad presses above. The title did not callsceFontOpenFontSet.Current blocker
No abort on
main39a1728; the blocker is the black frames after the Housemarque logo. No issue opened: the frames are black in the driver's read-back as well as in gamescope and the title does not react to input, so there is no failing line to report yet. Two intermittent events at the same point, 35 to 40 s in, right after[SAVEDATA:native] dirNameSearch user=268435456 pattern=(all) hit=1 set=1 -> 0x00000000(the save profile the first run created): in 1 of 5 RTX 3090 runs the title stopped presenting (no[present]line after 25 s, no title line after 35 s, alive to the cap), and in 1 of 2 RADV runs it segfaulted (kernel:segfault at 38 ip 00000000057731f1 sp 0000000200601ee0 error 4 in app.elf[53771f1,400000+5fce000]). The other five runs continued. Not investigated.How it was run
sce_sys/playgo-chunk.datnorsce_sys/playgo-scenario.json. Tried on e37e9fa: unset, the title stops at 0.3 s with the documentedscePlayGo: ... chunk ID 1 cannot be checked; set APS5_PLAYGO_UNLISTED_CHUNKS=...error.installedreaches thePS5 Out of Memorypath of Returnal (PPSA01285) on Windows: intermittent fault in a frame-pointer stack walk at app.exe+0x337aa9d (UE4 Main Thread, read of 0x8) #4239 (see Stops).absentand a list0,…,255behave the same;absentwas used for every run below.anyps5-fonts/held the Noto substitutes; not used by this title.tourist/saved/config/crashreportclient/ue4cc-ps5-<id>/crashreportclient.iniunder/app0at 0.3 s of every start, so the dump was mounted read-only (bwrap --ro-bind); the open fails and the title continues._sd/SaveProfile; later runs find it (dirNameSearch ... hit=1).shader_cache/; the NVIDIA driver cache was left in place.Runs on
main39a1728 (2026-11-09)APS5_DUMP_FRAMES=40 APS5_DUMP_FRAMES_EVERY=60, pad pressesHost report (run 4; paths replaced, the input-method variable left out)
Stops seen and their state
sceAjmInitializerefused, fixed onmainby fix(libSceAjm.native): stop checking sceAjmInitialize's reserved #16974. Onmaine37e9fa (before fix(libSceAjm.native): stop checking sceAjmInitialize's reserved #16974) every run with a bounded PlayGo answer (absentor a list; 9 of 9, cold and warm, NVIDIA and RADV, with and withoutAPS5_MIMALLOC) ended 2.6 to 5.7 s in on the main thread withint $0x45(exit 139). A first-parent bisect over cb1d6e5..e65fe03 withAPS5_LAZY_SHADER_PREPARE=1named 6daf9a1 (fix(libSceAjm.native): refuse a nonzero reserved in sceAjmInitialize #15906); perf(agc): wait at submit only for the preparations of the programs a submission binds #6867 had only removed the submit-time shader wait that kept older builds from reaching that point. A temporaryfprintfinsceAjmInitializeshowed the title calls it withreserved = 0x200000000and, on thatmain, gets0x80930005; with the check bypassed the same build ran on. That value is a third one next to the0x300000000and0x20recorded on sceAjmInitialize: a retail title passes reserved = 0x300000000 and fails since #15906 #17659 and fix(libSceAjm.native): stop checking sceAjmInitialize's reserved #16974. On 39a1728 the stop is gone (runs 1 to 7).PS5 Out of Memorypath. Reached on e37e9fa only withAPS5_PLAYGO_UNLISTED_CHUNKS=installed(2 of 2, at 19 and 29 s):PS5 Out of Memory: AvailablePhysical: 5005180928 UsedPhysical: 9280552960, thenLowLevelFatalError [File:Unknown] [Line: 188] Ran out of memory allocating 5565054976 bytes with alignment 65536. The title's stack print now ends at a host frame (libkernel.prxAps5CallGuestEntry_nid_no_patch+0x10) and the process exits through the title's libcint $0x45, so the walker fault of Returnal (PPSA01285) on Windows: intermittent fault in a frame-pointer stack walk at app.exe+0x337aa9d (UE4 Main Thread, read of 0x8) #4239 did not occur in these runs (fix(libkernel): end guest thread, fiber and context frame chains at a zero record #7123 is on thatmain). Withabsent(the fix(libScePlayGo): answer chunk ids from the package layout #4491 answer) the path is not taken.The numbers above are measured. The reading of the stack print, what the bisect means and the value of
reservedare my analysis of the logs and of a probe build.AI-assisted: yes (Claude Code, Claude Fable 5.1; the runs are the title's output). I am accountable for this report and will answer questions on it.
All reactions