[ntamas94]

What

On the console the kernel socket calls and libSceNet share one descriptor space: sceNetGetSockInfo describes a socket made with socket() exactly like one made with sceNetSocket, and lists it when s is below 0. AnyPS5 answered EBADF for such a descriptor and left it out of the list, because libSceNet only looked in its own table. Knowledge is Power: Decades (CUSA12435) calls sceNetGetSockInfo on its PlayLink server socket, which it makes with socket(), right after binding it, and reads from it only when recv_queue_length is not 0 (#9932).

libkernel now exports two helpers next to KernelLookupSocketNative_nid_no_patch: KernelDescribeSocket_nid_no_patch returns one of its sockets' native handle together with the owner that keeps it open, its family, type and non-blocking mode, and KernelListSockets_nid_no_patch lists its descriptors. sceNetGetSockInfo reads its own table first and the kernel's otherwise, and merges both into the sorted list. The socket is not adopted into libSceNet's table, so closing it with close() leaves nothing stale behind. Other sceNet calls on such a socket still answer EBADF; the debt entry says so.

sock_buffer_size now takes the record's source (native handle and configured sizes) instead of a Sock; #8362 calls it with a Sock in its new fill_sock_info6, so whichever of the two lands second needs that call adapted.

guest_sce_net gains rows for a UDP socket made with socket_nid_postfix and bound to 127.0.0.1: its record (descriptor, type 2, address, port, SELF flag, buffer sizes), its place in the list, and EBADF after close_nid_postfix. With the old Export.cpp, SocketPoll.hpp and Udp.cpp the test aborts at the first new row.

Tested

Windows 11, MinGW-w64 GCC (winlibs), Ninja: guest_sce_net passed on the final head (659962c, main 895f47f), python tools/check_conventions.py --base origin/main passes. Measured on a PS4 Pro (firmware 10.71) with a homebrew probe: a bound UDP socket from socket() and one from sceNetSocket both return 1 with a full record, and sceNetGetSockInfo(-1, NULL, 0, 0) counts both:

sceNet s=15 n=1 flags=0 ret=1 (0x1) errno=123 written=160
posix s=16 n=1 flags=0 ret=1 (0x1) errno=123 written=160
count s=-1 info=NULL ret=2 errno=123
all s=-1 n=4 flags=0 ret=2 (0x2) errno=123 written=320

Not run with a title.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: fix(libSceNet): report the receive queue length in sceNetGetSockInfo #9932
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

[MarcZX10]

I reviewed the descriptor lookup/listing changes and the new guest_sce_net cases. Holding the kernel socket owner while reading its native handle, and keeping the POSIX descriptor separate from the libSceNet table, matches the intended socket() behavior; the list is sorted after combining both tables. The declared dependency on #9932 is appropriate. The existing hosted eligibility, conventions, intake and report checks pass; hosted product-test jobs are skipped.

The PR reports guest_sce_net passing on Windows at this head. Since the implementation also adds the Linux ioctl(FIONREAD) path and the product jobs did not run, could you run the focused guest_sce_net test on Linux as well? That would verify both the new kernel-descriptor path and the receive-queue field on the other native socket API. I have not run a local C++ test on this head.

[github-actions]

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 5021	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 5836	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 7088	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7090	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7201	core/libs/prx/libSceNet/Export.cpp core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7485	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 7490	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp
PR 7507	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 8448	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 8463	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 8605	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9231	core/libs/prx/libSceNet/Export.cpp
PR 9509	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 9515	core/libs/prx/libSceNet/Export.cpp core/libs/prx/libkernel/Socket/include/SocketPoll.hpp core/libs/prx/libkernel/Socket/src/Udp.cpp core/libs/tests/GuestSceNet.cpp
PR 9543	core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 9578	core/libs/prx/libSceNet/Export.cpp core/libs/prx/libkernel/Socket/include/SocketPoll.hpp core/libs/prx/libkernel/Socket/src/Udp.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 9713	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10097	docs/dev/debt/libSceNet.md
PR 10502	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 10544	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10621	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10628	core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 10629	core/libs/prx/libSceNet/Export.cpp docs/dev/debt/libSceNet.md
PR 10645	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10767	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 10881	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 10949	core/libs/prx/libSceNet/Export.cpp core/libs/prx/libkernel/Socket/src/Udp.cpp docs/dev/debt/libSceNet.md
PR 11037	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 11134	core/libs/prx/libSceNet/Export.cpp
PR 11135	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11141	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11155	docs/dev/debt/libSceNet.md
PR 11194	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11219	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11223	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11242	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 11243	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp
PR 11247	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11272	core/libs/prx/libSceNet/Export.cpp core/libs/tests/GuestSceNet.cpp docs/dev/debt/libSceNet.md
PR 11281	core/libs/prx/libkernel/Socket/src/Udp.cpp
PR 11341	core/libs/prx/libSceNet/Export.cpp

[Queuse]

I tested the combined network changes on main b7a451d62b: #9523 a559f1e0be, #9542 ff4412c7e6, #9525 435a41bb82, #9543 aedc74347a, #9930 c079a98a41, #9931 fea0994737, #9932 d174eb7d04 and this PR 659962c821.

The reproducible integration tree is c9c12937ba. I resolved the shared getter hunks by retaining every option path and moving the general minimum-length check after TCP_NODELAY while preserving short reuse/broadcast reads. All test registrations and the neighboring debt entries are retained.

After rebuilding libc, libkernel, libSceNet and the eight test targets:

• Windows 11, WinLibs MinGW-w64 GCC 15.2.0-r7, Ninja, Release: 8/8 passed, 3.42 s.
• Linux under WSL2, Ubuntu 26.04.1, GCC 15.2.0, Ninja, Release: 8/8 passed, 1.48 s.

The tests were guest_net_socket_state, guest_net_protocol, guest_net_buffer_validation, guest_net_boolean_readback, guest_sce_net_keepalive, guest_net_oob_inline, guest_sce_net and guest_net_nodelay_readback. Conventions and whitespace checks pass too.

This supplies combined runtime evidence for the agreed ordering, including your kernel-created socket records. It is a focused local integration run; full hosted CI, a title and console hardware were not run. The remaining #6044/#5836 coordination is unchanged.

[MarcZX10]

The new integration evidence answers my earlier Linux test request: the reported Windows 8/8 and WSL2 8/8 focused runs include this PR alongside #9932, #9930, #9931 and the other listed network changes, and retain the kernel-created socket record coverage. I also reran current-main whitespace and conventions checks; both pass.\n\nOne remaining integration gap: the branch is 5 commits ahead and 264 behind current main (merge base 895f47f4), while current main has since advanced libSceNet/Export.cpp and docs/dev/debt/libSceNet.md, both files changed here. The PR’s integration run is based on b7a451d, not current main. Please refresh on current main, preserve the already-merged queue-length and network changes in Export.cpp, and rerun the focused socket-info tests on the refreshed head. The KernelDescribeSocket owner lifetime and descriptor listing logic otherwise look sound in this pass; I found no separate defect.

[MarcZX10]

Please rebase/merge current main and rerun the focused guest_sce_net coverage. The branch is 264 commits behind (merge base 895f47f4), and current main has intervening edits in both libSceNet/Export.cpp and docs/dev/debt/libSceNet.md, which this PR changes. The reported 8/8 Windows + WSL2 integration run is useful and covers the dependency network changes, but used main b7a451d, so it does not validate this PR integrated with today’s sceNetGetSockInfo queue-length and related network updates. Current-main diff-check and conventions pass; this request is specifically for resolving those shared-file updates and retesting the final head.

[ntamas94]

Thanks for the review. I merged current main 0a0ed72 (through the refreshed #9932 head 645bb29, which this PR is stacked on); the only conflict was the shared sceNetGetSockInfo debt entry, resolved by keeping #9932's new datagram notes and this PR's socket() sentence (998 characters). On the merge head 5125258, Windows 11, MinGW-w64 GCC, Ninja, after a fresh configure (main added the 3rdparty/ogg and 3rdparty/vorbis submodules):

python tools/check_conventions.py --base origin/main
All conventions checks passed.
ctest --test-dir build -R '^guest_sce_net$' --output-on-failure
1/1 Test #1422: guest_sce_net ....................   Passed    1.54 sec
100% tests passed, 0 tests failed out of 1

Thanks also for counting Queuse's Windows and WSL2 integration run as the Linux evidence.

[ntamas94]

Updated to 1d95b47: merged the refreshed #9932 (641b92a, with main ef0d8ae), which now records the console's datagram queue count. To keep both sceNetGetSockInfo debt entries under 1000 characters, this PR's socket() note is now its own entry. check_conventions --base origin/main and guest_sce_net pass on 1d95b47 (Windows 11, MinGW-w64 GCC, fresh build).