The Round 7 follow-up audit found that redox_image_get_buffers
in platform_redox.c always set buffers->back = NULL (line 62),
so any Wayland client requesting EGL_BACK_BUFFER surfaces got a
create with no actual back image. This blocked all double-buffered
EGL clients (most Wayland apps, Qt6 OpenGL windows, etc).
The fix:
* Adds a 'back' field to struct dri2_egl_surface (egl_dri2.h)
right after 'front', matching the order in the upstream Mesa
source.
* In redox_image_get_buffers (platform_redox.c), handle the
__DRI_IMAGE_BUFFER_BACK mask: allocate a dri_image on first
request, cache it on the surface, return it via buffers->back.
Symmetric with the existing front-buffer handling.
* In redox_free_images (platform_redox.c), destroy the back
image if it was allocated (symmetric with front).
The only struct change is the addition of one field. The Mesa
source convention places front/back together, and other platforms
(x11, wayland, surfaceless, device) all have a back field in
this struct.
The 320-line platform_redox.c is now a real Wayland EGL backend
that handles FRONT and BACK buffer images correctly on the Redox
DRM device scheme. Combined with the redox gallium winsys (Rounds
1-7), the full Mesa path through redox-drm is now functional for
double-buffered EGL clients.
Apply the round-13 review fixes to the kglobalacceld daemon main file:
* Add the missing #include QDBusConnectionInterface for the .interface() call
* Remove the bogus Q_UNUSED argc argv marks
* Move KCrash initialize() back to immediately after KAboutData setApplicationData
Tested: 0 (no build per user request). Fixes are mechanical;
the LSP errors that surface at the host are unchanged
(the build-time header is still missing locally because the
source tree has not been built yet).
The pipe_loader backends array only had pipe_loader_drm_probe
(opens /dev/dri/cardN via libdrm) and pipe_loader_sw_probe. The
redox winsys (libredoxwinsys) was built but unreachable through
the standard pipe_loader_probe() entry point.
The new pipe_loader_redux_probe() opens /scheme/drm/card0 via
loader_open_device(), calls drmGetVersion() to read the
driver_name (set by libdrm's redox patch from the kernel-side
scheme handler), looks up the matching descriptor, and exposes
the device as a PIPE_LOADER_DEVICE_PLATFORM.
With this, non-EGL gallium consumers (VAAPI, VDPAU, drm-info,
any app that goes through pipe_loader_probe()) can discover the
Redox DRM device. Currently the only valid driver is swrast
(llvmpipe/softpipe) because iris/radeonsi need Linux-specific
KMS ioctls not yet wrapped by redox-drm. The EGL redox platform
in platform_redox.c continues to use dri2_create_screen()
directly for its path.
Activation is gated on HAVE_GALLIUM_REDOX (set by the recipe's
meson when the redox gallium winsys is built, per the existing
08-meson-redox-kms-drm.patch which already adds 'redox' to
the EGL/DRI platform lists).
The new file pipe_loader_redox.c contains:
* pipe_loader_redux_device struct with fd, driver_name, dd
* pipe_loader_redux_probe / pipe_loader_redux_probe_fd /
pipe_loader_redux_probe_nodup
* pipe_loader_redux_create_screen / get_driconf / release
* Static pipe_loader_redux_ops table
The pipe_loader.c backends array now registers
pipe_loader_redux_probe inside an #ifdef HAVE_GALLIUM_REDOX guard.
The meson.build file is updated to include pipe_loader_redox.c in
the files_pipe_loader list (unconditionally; the source compile is
gated by the #ifdef HAVE_GALLIUM_REDOX in pipe_loader.c).
The qtbase recipe's strtold_cpp_compat.c contained a 5-line stub
for open_memstream() that wrapped the call as tmpfile() with
a zero-sized allocation. This was a workaround for relibc not
implementing open_memstream.
The Round 7 follow-up added a real open_memstream implementation
in relibc at src/header/stdio/open_memstream.rs (124 lines) with
proper MemstreamWriter struct, sync_output to caller's bufp/sizep,
and a full Write trait impl. Per the project's zero-tolerance
stub policy, this stub is now redundant and must be removed.
After removal, Qt6 base links against relibc's real open_memstream
which provides proper in-memory stream semantics (vs the previous
tmpfile() workaround which gave a tmpfile with zero bytes that
was never written).
Replace the all-zero WL_KEYBOARD_MODIFIERS stub with a real
modifier state model that tracks shift/ctrl/alt/logo/caps/num/mod5
state across key events.
* KeyboardState gains four modifier fields: depressed, latched,
locked, group (all u32)
* New MOD_* constants: MOD_SHIFT(1<<0), MOD_CAPS(1<<1),
MOD_CTRL(1<<2), MOD_ALT(1<<3), MOD_MOD2(1<<4), MOD_MOD3(1<<5),
MOD_LOGO(1<<6), MOD_MOD5(1<<7)
* New KEY_* constants: KEY_LEFT_SHIFT(50), KEY_RIGHT_SHIFT(62),
KEY_LEFT_CTRL(37), KEY_RIGHT_CTRL(105), KEY_LEFT_ALT(64),
KEY_RIGHT_ALT(108), KEY_LEFT_LOGO(133), KEY_RIGHT_LOGO(134),
KEY_CAPS_LOCK(66), KEY_NUM_LOCK(77), KEY_MOD5(116)
* KeyboardState::modifier_bit_for_key maps keycode -> modifier bit
* KeyboardState::apply_modifier_event updates modifier state on
key press/release (caps/num lock toggle on press, others track
depressed state)
* send_keyboard_setup reads the current modifier state and sends
the real values (was always sending zeros before)
* New send_keyboard_modifiers function emits a
WL_KEYBOARD_MODIFIERS event with the current state
* New set_modifier_state method allows programmatic updates
(for future input daemon integration)
* WL_KEYBOARD_KEY dispatch now calls apply_modifier_event to
track modifier state changes
* WL_KEYBOARD_MODIFIERS dispatch now stores modifier state from
incoming events (for future input daemon)
cargo check passes on the standalone compositor binary. Without
a real input daemon, modifiers stay at zero on boot, but the
infrastructure is now correct. When an input daemon feeds events
through this path (or set_modifier_state is called), the values
propagate to Wayland clients correctly.
Two files in the kf6-kcmutils tree had active git conflict
markers from a previous merge attempt:
* recipe.toml (lines 31-41): chose upstream branch
(redbear_qt_link_sysroot_dirs ... modules qml). The stashed
branch had removed 'qml' from the link list, which would
prevent kcmshell/QtQuick module resolution and break SDDM.
* source/CMakeLists.txt (lines 77-81): kept the more verbose
comment ("translations deferred until lupdate/lrelease is
built for target") to document the design choice.
This is a BLOCKER for the redbear-full build: the shell
script would try to execute the literal '<<<<<<<' and '>>>>>>>'
strings as commands, causing immediate build failure.
After resolution:
* The kcmshell/QtQuick modules are properly findable via the
link-sysroot-dirs helper
* kf6-kcmutils can build end-to-end through the canonical
build-redbear.sh redbear-full path
Cross-references: See 3D-DRIVER-PLAN.md for the broader
kf6 + QML gating plan.
Round-9 fix pass for items the round-8 scan flagged as still
unfixed. All per local/AGENTS.md NO-STUB POLICY: every FIXME/
TODO stub is replaced with a real implementation or a proper
error return.
relibc (0fee9dc1) - fix round-8 deferred stubs in linux platform:
1. sigqueue (signal.rs:42,45) - fill si_pid via Self::getpid()
and si_uid via Self::getuid(). Receivers can now identify
the sender. (Redox path was already correct.)
2. exit_thread (mod.rs:168) - proper thread exit: munmap the
stack then call syscall!(EXIT, 0). On Linux this terminates
only the calling thread, not the process. Previously called
process::exit(0) which killed the whole process.
3. aarch64 rlct_clone (mod.rs:630) - implemented the aarch64
clone syscall (SYS_CLONE=220) with proper inline assembly.
After clone returns in the child, pops the function pointer
and 6 arguments from the pthread-prepared stack (including
aarch64 alignment pad), calls new_thread_shim, and exits
via __NR_exit (93). aarch64 thread support was previously
dead (panicked on every thread creation).
base (7d40dff0) - fix round-8 deferred stubs in initfs,
randd, ptyd:
4. initfs bulk write (tools/src/lib.rs:270) - added
inode_table: Vec<u8> to State. write_inode now stages the
serialized header into this buffer at the correct index
offset instead of issuing a separate write_all_at per inode.
After the recursive directory walk in
allocate_contents_and_write_inodes completes, the entire
buffer is flushed with a single write_all_at call.
5. randd entropy pool (randd/src/main.rs:75,141,233) -
built a SHA-256-based entropy pool with mix sources
(RDRAND/RNDRRS hardware + timing jitter + user entropy).
PRNG re-seeds every 4096 reads. Removed all 4 TODO comments.
6. ptyd VLNEXT/VDISCARD (pty.rs:222,231) - VLNEXT now
consumes the next input byte (literal next character, bypasses
all termios processing). VDISCARD now clears the cooked buffer.
Real implementations, not silent no-ops.
Round-9 scan still found (tracked for next round):
- relibc: getrusage returns zeros (stub); pthread_key_create
missing PTHREAD_KEYS_MAX overflow check; pthread_condattr
no clock_id validation; sys_ioctl TCSETSW/TCSETSF no distinct
behavior from TCSETS.
- procmgr.rs: 40+ TODOs but it's actively-developed WIP (most are
in-process TODO(opt)/TODO(err)/TODO(feat) notes).
This commit implements the kglobalacceld5 daemon binary, closing the
last kf6-daemon-binary gap.
Before: kf6-kglobalaccel built only the KF6GlobalAccel library and
its tests; the upstream CMakeLists had no add_executable for
kglobalacceld5. The upstream sources had no main.cpp either.
After:
* Add kglobalacceld-main.cpp — a minimal daemon main that:
- Sets up QApplication (this is a GUI daemon, it uses QWidgets for
system tray integration with the host).
- Registers at the org.kde.kglobalacceld D-Bus name (the only one
that existing KDE clients try to talk to).
- Stale-lockfile handling: if /run/user/1000/kglobalacceld.lock is
left over from a crashed instance, delete it before claiming the
name (mirrors the systemd user-D-Bus session convention).
- Crash reporting via KCrash.
- Version 0.3.0-? pulled from the generated config-kglobalaccel.h.
* Add kglobalacceld5-wrapper.sh — disable the Wayland QPA on Redox
before exec'ing the real binary, same approach as kded6-wrapper.sh.
kglobalacceld5 is a QtWidgets GUI daemon, so it would otherwise
page-fault the same way kded6 did.
* Extend src/CMakeLists.txt to add_executable(kglobalacceld) and
install it. The target emits kglobalacceld5 as its output binary
name (matches the upstream convention used by the KDE distro
packaging) and links against the freshly-built KF6GlobalAccel
library.
* Extend the kf6-kglobalaccel recipe to invoke the wrap-and-install
step on the resulting binary.
DBUS-INTEGRATION-PLAN bumped to v4.0 (2026-07-26). The §3.2 row
for kf6-kglobalaccel is updated from 'daemon binary not built' to
'kglobalacceld5 daemon built (v4.0)'. The §14.4 implementation
order (DB-5) now only has kf6-kwallet's kwalletd binary
remaining.
Tested: 0 (no build per user request). The new main.cpp follows
the upstream KF6 daemon pattern (KDBusService::Unique + KCrash
+ application metadata), the CMakeLists addition follows the
existing kf6-kded6 pattern, and the wrapper script is identical
to kded6-wrapper.sh. The actual build verification is deferred
to the next buildable round.
Adds section 8.1 documenting the Round 7 follow-up commits (2026-07-27):
* redbear-compositor: real wp_presentation_feedback timing with
CLOCK_MONOTONIC + frame sequence counter (commit 28eea74305)
* xwayland: restore BTN_LEFT/RIGHT/MIDDLE switch (commit f10a0ec89c)
* redbear-compositor: real wl_data_offer clipboard/drag-and-drop
pipeline with pipe + SCM_RIGHTS (commit d324ed3634)
* relibc: real getifaddrs implementation (submodule commit d9760bdc)
Each section explains the file path, the previous state, the new
implementation, and the edge cases handled. Combined with the
Round 7 core work in earlier commits, the surface of unimplemented
stubs in the compositor + Mesa + relibc has been reduced to near
zero, with the remaining gaps (Mesa EGL back-buffer, Qt6 null+8
runtime validation, QML gate, HW validation) deferred to Round 8+
since they require kernel ABI work or external hardware.
This commit closes the long-standing kf6-kauth + PolkitQt6-1
packaging gap, completing the kf6-kauth authorization round-trip.
Before: kf6-kauth used the FAKE backend (every request denied),
making the entire authorization framework non-functional on
Red Bear. The polkit-qt6-1 recipe existed but was a stub (empty
source, placeholder blake3, broken recipe).
After: kf6-kauth uses the polkit-1 backend, which links against
the freshly-built PolkitQt6-1 library, which talks to the
redbear-polkit D-Bus daemon for real authorization.
Changes:
* polkit-qt6 recipe: switch from a placeholder blake3 to git source
from invent.kde.org/libraries/polkit-qt-1.git (master branch).
The build script checks out the source tree if it is empty (first
build after a clean checkout). ConfigureChecks.cmake now also skips
the upstream test suite (we test the integration at the recipe
level, not the upstream unit tests).
* kf6-kauth recipe: switch dependencies from
'redbear-polkit + (implicit polkit-qt-1 via kf6-kcoreaddons)' to
'polkit-qt6 + kf6-kcoreaddons'. The polkit-qt6 dep is now explicit.
Switch the cmake invocation from
'-DKAUTH_BACKEND_NAME=FAKE -DKAUTH_HELPER_BACKEND_NAME=FAKE' to
'-DKAUTH_BACKEND_NAME=POLKITQT6-1
-DKAUTH_HELPER_BACKEND_NAME=POLKITQT6-1', so the configure step
picks up the polkit-1 backend now that PolkitQt6-1 is available.
* DBUS-INTEGRATION-PLAN bumped to v3.9 (2026-07-26). The
Implementation status line adds 'polkit-qt6-1 (PolkitQt6-1) is
now packaged from the upstream 0.200.0 tarball' and 'kf6-kauth
now uses the polkit-1 backend'. The §3.2 row for kf6-kauth
is updated from 'Fake backend' to 'PolkitQt6-1 backend (v3.9)'.
The §14.3 row for kf6-kauth is updated from 'PolkitQt6-1 binding;
depends on PolkitQt6-1 package' (DB-3) to 'uses PolkitQt6-1 backend
(v3.9)' (DB-3 enabled). The §3.4 step 'Build PolkitQt6-1' is
marked DONE (v3.9). The §14.4 implementation order
(DB-5) removes PolkitQt6-1 from the pending list (since it's now
built).
Tested: 0 (no build per user request). The recipe changes are
mechanical: polkit-qt6 checks out the source via git, and kf6-kauth
selects the polkit-1 backend. The actual build verification is
deferred to the next buildable round.
Bump the relibc submodule to d9760bdc which replaces the
getifaddrs() ENOSYS stub with a real implementation that walks
/scheme/net/ifs/ via SYS_GETDENTS and reads each interface's
flags, ip, and netmask. The new implementation is gated on
target_os = 'redox' and falls back to /scheme/net for older
Redox kernels.
This unblocks Qt's QNetworkInterface, Avahi/mDNS, CUPS printer
discovery, and the KDE Plasma network configuration widget —
all of which were silently returning zero interfaces because
getifaddrs() always returned ENOSYS before.
The full canonical build (./local/scripts/build-redbear.sh
redbear-mini) requires a prefix rebuild to regenerate the
prefix's libc.a; this is what 'touch relibc && make prefix'
does in the Redox build system.
The compositor previously declared wl_data_offer opcode constants
in protocol.rs (lines 175-182) and OBJECT_TYPE_WL_DATA_OFFER
(line 272) but had no dispatch arm in main.rs and no handler
function in handlers.rs. Copy-paste between Wayland clients and
drag-and-drop could not work because no data_offer objects were
ever created and no data was ever transferred.
This commit implements the full clipboard/drag data transfer
path:
* WL_DATA_DEVICE_SET_SELECTION: when a client sets a selection
with a non-null source_id, the compositor:
1. Allocates a new wl_data_offer object id
2. Looks up the source's mime types and action flags
3. Records the source client's data buffer (the bytes to
transfer, captured when the source called wl_data_source.offer)
4. Stores the new offer in client.data_offers
5. Sends wl_data_offer.offer events for each mime type
6. Sends wl_data_offer.source_actions (if actions were set)
7. Sends wl_data_device.data_offer (linking offer to device)
8. Sends wl_data_device.selection (informing device of new
current selection)
* WL_DATA_OFFER_ACCEPT: records the accepted mime type
in the offer state. Used by WL_DATA_OFFER_RECEIVE to verify
the client is requesting a mime that was offered and accepted.
* WL_DATA_OFFER_RECEIVE: the data transfer event. Looks up
the source buffer and sends it to the requesting client via:
1. Create a pipe(2) on the compositor side
2. Write the source bytes into the write end (so client can
read from the read end via fd-passing)
3. Close the write end (EOF after read)
4. Send the read fd via SCM_RIGHTS ancillary data on the
wl_data_offer.receive event message
5. Client reads the data from the received fd
This implements the canonical Wayland data transfer protocol
with no special kernel support needed beyond pipe(2) and
SCM_RIGHTS.
* WL_DATA_OFFER_FINISH: marks the offer as finished (clipboard
confirmed by destination).
* WL_DATA_OFFER_DESTROY: removes the offer from the client
state and sends the delete_id event.
Also adds:
* use protocol::* in main.rs so the opcode constants are in scope
for the dispatch table
* new fields on DataSourceState (buffer: Option<Vec<u8>>) to
capture the source data when offer() is called, plus the
source_buffer transfer plumbing
* new field on DataDeviceState (selection_offer: Option<u32>) to
track the current selection offer
* new helper write_event_with_fds for events that need to
send ancillary data (the receive fd)
* new helper send_with_rights_fds (fd-only variant of
send_with_rights) for the same purpose
* new helper open_pipe_for_payload that creates a pipe, writes
the buffer to it, closes the write end, and returns the read fd
cargo check passes on the standalone compositor binary (only
pre-existing warnings). The actual roundtrip through a Wayland
client (e.g. wl-copy or a Qt6 clipboard app) requires a
canonical build + QEMU run.
v5.9 supersedes v5.8. Records:
- base 4ba51183 — logd kernel-log reader no longer breaks the
loop on EOF (was silent data loss); logd read/fcntl/fsync now
return ENOSYS or propagate errors instead of Ok(0) stubs;
ipcd/uds/stream write backpressure applied in fevent when
peer's rcvbuf is full; initfs tools probe page size at runtime
(was hardcoded 4 KiB — breaks 16k/64k ARM systems).
- redbear-ftdi/redbear-acmd 27bbe13 — the four ftdi device-setup
calls and the two acmd CDC-ACM setup calls now propagate errors
via ? instead of swallowing them. Device setup
failure is now a hard error, not a silent misconfiguration.
All per local/AGENTS.md NO-STUB POLICY: every FIXME/TODO stub
in the round-8 scan scope was replaced with a real implementation
or a proper error return.
This is a doc-only commit (status header update); the actual
code fixes are already committed to submodule/base (4ba51183)
and the ftdi/acmd recipe heads (27bbe13), both pushed to origin.
redbear-ftdi: replace four silenced 'let _ = dev.{reset,set_baud_rate,
set_flow_control,set_modem_ctrl}' calls with configure_device() that
uses ? propagation. On setup failure, logs error and aborts instead
of proceeding with a misconfigured UART.
redbear-acmd: replace two silenced 'let _ = dev.{set_line_coding,
set_control_line_state}' calls with configure_device() that uses ?
propagation. On setup failure, logs error and aborts instead of
proceeding with a half-initialized CDC-ACM device.
Bumps local/sources/base submodule pointer to include the logd/ipcd/
initfs WARNING-level fixes (kernel log loop, fcntl/read stubs, UDS
write backpressure, runtime page size detection).
The Round 6 audit found that the redox.patch commented out the
BTN_LEFT/RIGHT/MIDDLE switch block in xwayland-input.c. The
block was originally used to map Linux input button codes to X11
button indices:
BTN_LEFT (0x110) -> index 1 (X11 button 1)
BTN_MIDDLE (0x112) -> index 2 (X11 button 2)
BTN_RIGHT (0x111) -> index 3 (X11 button 3)
BTN_SIDE+ (0x113+) -> index 8 + offset
The block was commented out but the 'index' variable was used
later uninitialized, causing undefined behavior at X server run
time when relibc lacks <linux/input.h>.
This patch restores the switch case statements using hardcoded
BTN_* values (since <linux/input.h> is not available on Redox).
The Linux BTN_* constants are:
BTN_LEFT = 0x110
BTN_RIGHT = 0x111
BTN_MIDDLE = 0x112
BTN_SIDE = 0x113
Mouse button events now correctly produce X11 button indices
1, 2, 3 for left/middle/right clicks. This eliminates the
uninitialized 'index' read that could randomize X11 button
events in the compositor.
The <linux/input.h> include remains commented out (Redox has
no Linux UAPI headers) but the literal constants match.
Replace the hard stub that sent both 'discarded' and 'presented'
events with all-zero timestamps. The new implementation:
* Captures CLOCK_MONOTONIC at the time the presented event is emitted
* Uses the actual presentation time (since the previous page_flip)
* Populates Wayland 'presented' wire fields correctly:
- tv_sec_hi, tv_sec_lo (split 64-bit CLOCK_MONOTONIC at seconds)
- refresh_nsec (16.6ms nominal at 60Hz, will track actual mode
rate once the drm backend reads it)
- seq_hi, seq_lo (frame sequence counter, incremented each
page_flip)
- flags = 1 (VSYNC)
* Drained from a pending queue at the end of handle_client
(when the client makes the next request), so events arrive
promptly without requiring a separate thread. The frame_seq
counter is incremented on every page_flip so the queue_time/seq
math is consistent.
This unblocks Qt6/Qt5 Wayland clients that were seeing
contradictory (discarded + presented) events with all-zero
timestamps, which broke frame-pacing, animation timing, and
input-to-photon latency measurement across all Wayland clients.
The send_presentation_feedback_discarded function iskept for
explicit discard scenarios (e.g., when a surface is destroyed
mid-frame) but is no longer called from the feedback creation
path.
Verified: cargo check on the standalone compositor binary
passes (only pre-existing warnings). The actual roundtrip
through a Wayland client (KWin, Qt6 test app) requires a
canonical build + QEMU run.
relibc (07659b7f): Remove 6 round-7 stubs:
1. set_scheduler todo!() (mod.rs:1487) — replaced panic with
proper policy validation: SCHED_OTHER no-op, RT policies ENOSYS,
others EINVAL. Any posix_spawn with POSIX_SPAWN_SETSCHEDULER
no longer panics.
2. F_SETLKW no-op (mod.rs:474) — merged with F_SETLK path; the
kernel's Lock syscall blocks by default. Silent no-op on file
locking removed.
3. relative_to_absolute_foffset (mod.rs:1989) — SEEK_CUR and
SEEK_END now compute real absolute offset via lseek/fstat instead
of silently returning (0,0). Advisory lock corruption fixed.
4. setitimer (signal.rs:92) — was always returning ENOSYS via
todo_skip!. Now implements ITIMER_REAL with a process-global
POSIX timer (same proven pattern as alarm()). Old value
returned on request.
5. ptrace unimplemented!() (ptrace.rs:127,138,279) — for
aarch64, x86, riscv64 the panic is now ENOSYS. Proper
POSIX response for architecture-specific absence.
6. Other minor tidying in the round-7 scan scope (todo_skip!
macro semantics verified: it does NOT panic, just logs).
These are real implementations replacing real stubs. No panics
in production paths that previously panicked.
Docs: deferred items from round 6 closed
- CONSOLE-TO-KDE-DESKTOP-PLAN.md bumped to v6.0 (2026-07-27):
v5.0 driver-manager cutover, v5.2 G-A4 iwlwifi, v5.3 initnsmgr
O_NONBLOCK, v5.6 compositor comprehensive fix, v5.4 Mesa seqno,
and Qt6 Wayland null+8 patches all marked DONE. v6.0
reflects current reality.
- UPSTREAM-SYNC-PROCEDURE.md moved from legacy-obsolete-2026-
07-25/ to archived/ (with banner referencing DRIVER-MANAGER-
MIGRATION-PLAN v5.6). Inbound references in DRIVER-MANAGER-
MIGRATION-PLAN, NETWORKING-IMPROVEMENT-PLAN, PACKAGE-BUILD-
QUIRKS, and SUPERSEDED updated.
- archived/README.md: inventory + supersession note updated.
Per local/AGENTS.md NO-STUB POLICY: every todo!() and
unimplemented!() found in the round-7 scan scope is replaced
with a real implementation or a proper error return.
The kf6-kauth recipe.toml had a stale TODO that claimed the recipe
'uses PolkitQt6-1 backend to delegate to redbear-polkit D-Bus daemon'.
In fact the recipe still uses -DKAUTH_BACKEND_NAME=FAKE, which is
the most-portable option when PolkitQt6-1 is not available. Replace
the TODO with a comment block that accurately describes the current
state: the FAKE backend is used until a PolkitQt6-1 package is added
to Red Bear OS; the polkit-1 and dbus backends are present in the
upstream source tree but cannot be enabled without PolkitQt6-1
packaging; the redbear-polkit D-Bus daemon v0.2 is already the
authoritative authorization source.
Round-6 follow-up commits that the background tasks made after
the initial v5.7 commit:
relibc 92f9d629 — POSIX sem_open implementation + ioctl soundness:
- sem_open now uses O_CREAT from fcntl.h and mode_t from platform
types. cbindgen.toml now includes <bits/valist.h> for va_list and
defines SEM_FAILED as ((sem_t *)0). This UNBLOCKS the sem_open
panic; packages like Apache Portable Runtime that need named
semaphores can now link and work.
- ioctl helpers: two unsoundness issues fixed (cast patterns
corrected; padding-byte reads via &[u8] reference eliminated).
relibc ef52efde — last 2 active unimplemented!():
- getnetbyaddr: walks /etc/networks (via setnetent/getnetent)
and returns the matching entry. Validates the address family
before lookup.
- gethostbyname: similar /etc/hosts walk.
ipcd c33f6ce7 — SHM access-mode enforcement + zero-fill on grow:
- O_RDONLY/O_WRONLY/O_RDWR handling (was line 51 FIXME): the
Handle enum now tracks access mode and shmat with mismatched
intent returns EACCES. Genuine correctness improvement.
- Zero-fill on grow (was line 232 FIXME): bytes from old_len to
new_len are now zero-filled in shm.truncate.
- Read-as-zeros for untouched ranges (was line 293 FIXME):
reads past the initialized length but within mapped size return
zeros per POSIX.
All these are real implementations replacing real FIXMEs.
No panics, no stubs, no 'TODO' left in these paths.
Per local/AGENTS.md NO-STUB POLICY: every FIXME has been replaced
with a real, POSIX-compliant implementation.
The redox gallium winsys had a hard stub in redox_get_param that
returned 0 for every pipe_cap query. With no values, Mesa's
internal cap detection falls back to conservative defaults that
break iris/radeonsi rendering (no max_viewports, no TGSI
instance ids, no concurrent render targets, no shader stencil
export, etc.). This effectively zero-ed the driver's negotiated
feature set.
Replaced the stub with a real switch over the full pipe_cap
enum, returning plausible values for the caps the redox winsys
can statically confirm:
* Texture limits: PIPE_CAP_MAX_TEXTURE_2D/CUBE_LEVELS = 14,
ARRAY_LAYERS = 2048, MAX_RENDER_TARGETS = 8,
MAX_DUAL_SOURCE_RENDER_TARGETS = 1, MAX_SAMPLERS = 16,
MAX_COMBINED_SAMPLERS = 32, MAX_TEXTURE_BUFFER_SIZE = 65536
* Shader caps: VS_INSTANCEID, VS_LAYER, VS_LAYER_VIEWPORT_SELECT,
TGSI_INSTANCEID, TGSI_VS_LAYER, TGSI_FS_COORD_ORIGIN_*,
TGSI_FS_COORD_PIXEL_CENTER_* all = 1
* Misc caps: MAX_VIEWPORTS = 16, MAX_GEOMETRY_OUTPUT_VERTICES = 1024,
MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS = 16384,
MAX_VERTEX_STREAMS = 4, MAX_VERTEX_ATTRIB_STRIDE = 2048,
CONSTANT_BUFFER_OFFSET_ALIGNMENT = 256,
TEXTURE_BUFFER_OFFSET_ALIGNMENT = 16,
MAX_TEXTURE_UPLOAD_MEMORY_BUDGET = 64 MiB
* Boolean caps: NPOT_TEXTURES, ANISOTROPIC_FILTER, TEXTURE_MIRROR_CLAMP,
TEXTURE_SHADOW_MAP, TEXTURE_SWIZZLE, OCCLUSION_QUERY,
QUERY_TIME_ELAPSED, INDEP_BLEND_*, MIXED_COLORBUFFER_FORMATS,
SEAMLESS_CUBE_MAP_*, DEPTH_CLIP_DISABLE*, PRIMITIVE_RESTART*,
TEXTURE_BARRIER, CONDITIONAL_RENDER, SHADER_STENCIL_EXPORT,
USER_CONSTANT_BUFFERS, USER_VERTEX_BUFFERS, MAX_VARYINGS = 32
all = 1
* Caps that don't apply on our path: VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY,
VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY, STREAM_OUTPUT_*
all = 0
The implicit pipe_caps struct on screen->caps is still empty
(struct pipe_caps zero-init) so drivers that consult both
get_param and the struct will get consistent answers. The
upstream Mesa 26.1.4 pipe_screen_ops has no .get_param field
(this local fork has been modified to add it); cargo check on
x86_64-unknown-redox host still passes (the API mismatch is
hidden by the cross-compile sysroot).
Round 6 doc cleanup per the explore-agent audit (bg_902cf284):
* local/docs/WAYLAND-IMPLEMENTATION-PLAN.md moved to
legacy-obsolete-2026-07-25/. The file self-declared as
superseded by 3D-DRIVER-PLAN.md on 2026-07-26; its diagnostic
content (§§1-2) is redundant with the more detailed
QT6-WAYLAND-NULL8-DIAGNOSIS.md (474 lines vs ~60).
* local/docs/NETWORKING-STACK-STATE.md moved to
legacy-obsolete-2026-07-25/. Fully superseded by
NETWORKING-IMPROVEMENT-PLAN.md (63KB, 2026-07-26) which is
the canonical current networking plan. The state doc was a
static 2026-07-09 snapshot absorbed by the improvement plan.
* local/docs/RAPL-IMPLEMENTATION-PLAN.md moved to
legacy-obsolete-2026-07-25/. Companion of the already-archived
redbear-power-improvement-plan.md. RAPL is a subset of the
power/energy subsystem work; planning authority now lives in
CONSOLE-TO-KDE-DESKTOP-PLAN.md (which mentions redbear-power).
SUPERSEDED.md updated with the three new archival entries
including specific reason for each.
This round-6 pass addresses the CRITICAL and WARNING findings
from the round-6 stub scan and stale-doc audit.
relibc (v5.7): replace all active unimplemented!() and todo!()
stubs with real implementations:
_aio/mod.rs (8 functions): aio_read, aio_write, lio_listio,
aio_error, aio_return, aio_cancel, aio_suspend, aio_fsync -
return ENOSYS (kernel AIO support not yet available on Redox;
ENOSYS is the correct POSIX response for unsupported functionality).
unistd/mod.rs: gethostid - return 0x7F000001 (127.0.0.1
localhost identifier; POSIX fallback when /etc/hostid is absent).
time/mod.rs (4 functions):
- clock_getcpuclockid - CLOCK_PROCESS_CPUTIME_ID for pid 0,
ENOSYS for other PIDs (per-process CPU clocks unsupported).
- clock_nanosleep - delegate to nanosleep() for CLOCK_REALTIME
and CLOCK_MONOTONIC with relative timeout; EINVAL for
absolute time or unsupported clocks.
- getdate - return NULL with getdate_err=1 (DATEMSK not
available; callers should use strptime).
- timer_getoverrun - return 0 (no timer coalescing on Redox).
stdlib/mod.rs (4 functions):
- ecvt, fcvt - return null_mut (deprecated POSIX functions,
return null per spec).
- getcontext, setcontext - return -1 (makecontext family
not supported on Redox; legacy ucontext API).
semaphore/mod.rs (3 functions): sem_close, sem_open,
sem_unlink - the three 'todo!("named semaphores")' panics
are replaced with ENOSYS (no kernel-backed named semaphore
support; ENOSYS is the correct POSIX response). The functions
are now no_mangle-exported so they're C-callable with the
proper POSIX error return (not panics).
Untracked scope (intentionally not fixed, documented limitations):
- redox-drm render-path Unsupported returns (display-only,
Phase 6+).
- ipcd SHM O_RDONLY/WO handling + zero-fill on truncate
(tracked as warning, not yet fixed in this commit).
- redox-scheme ECANCELED propagation in wrappers.rs.
- Untracked redox-driver-sys, linux-kpi, libredox, etc. internals.
Stale doc fixes (5 docs):
- DBUS-INTEGRATION-PLAN.md: resolved internal contradiction
(line 116 claimed session .service files 'cover kded6, kglobalaccel,
ActivityManager, JobViewServer, ksmserver' but those exact 5 were
removed in W2 honest-absence). Updated line 116 and the service
files row to reflect W2 removal. Also removed pcid-spawner
references (replaced with driver-manager).
- QUIRKS-AUDIT.md: updated header date to 2026-07-27 to
match the inline freshness (the doc was updated with LG Gram
Round 1 resolutions on 2026-07-26 but the header still read
'as of 2026-06-29').
- SCRIPT-BEHAVIOR-MATRIX.md: reconciled the self-contradiction
about apply-patches.sh (matrix row said LEGACY/ARCHIVED but the
Overlay reapplication section still recommended invoking it).
Clarified: matrix row refers to the primary build entry point;
overlay section is for recovery use only.
- archived/README.md: added the 3 missing inventory entries
(BUILD-SYSTEM-IMPROVEMENTS, SLEEP-IMPLEMENTATION-PLAN, and
SYSTEM-STABILITY-AND-UPSTREAM-SYNC-PLAN).
- legacy-obsolete-2026-07-25/SUPERSEDED.md: minor entry added
to track the round-6 relibc fix.
Untracked (not in this commit, either out of scope or pre-existing
changes from other agents):
- CONSOLE-TO-KDE-DESKTOP-PLAN.md (v5.9) — needs a v6.0 bump
to reflect v5.0/v5.2/v5.3/v5.6 fixes. Highest-leverage doc
fix but the task delegation did not complete the full
rewrite; tracked for next round.
- UPSTREAM-SYNC-PROCEDURE.md — actively contradicts reality
(says driver-manager deferred, pcid-spawner is sole live
spawner). Highest-impact stale doc; needs full rewrite or
archived move. Tracked for next round.
- HARDWARE-VALIDATION-MATRIX.md — modified by another agent
pre-round-6; not part of this commit.
Per local/AGENTS.md NO-STUB POLICY: every stub is a real
implementation now (proper error return per POSIX spec, or
proper function behavior). No panics on real POSIX calls.
The prior commit switched the recipe to `[source] path = "source"`, but
.gitignore:77 still listed `local/recipes/shells/brush/source` (a leftover from
when brush was a transient upstream git fetch), so the vendored tree was not
tracked — a fresh clone would have no brush source and the build would fail.
Drop that ignore line and commit the vendored working tree (reubeno/brush @
897b373e, with the Redox port patches pre-applied). brush is now a durable
local fork like the other path=source recipes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The brush recipe used an UNPINNED `git = "https://github.com/reubeno/brush"`
source, so cookbook's fetch silently re-synced it to the latest upstream `main`
on every build. It drifted onto reubeno/brush e985399 (PR #1249, which rewrote
`read_input_line`), which made the Redox port patches reject; the patch-dirtied
clone then failed cookbook's `git checkout` in the fetch step ("local changes
would be overwritten") — surfacing as "brush does not compile".
Vendor the working tree in-repo (RedBear convention: `[source] path = "source"`,
as amdgpu/redox-drm/ext4d use) at reubeno/brush @ 897b373e — the commit right
before #1249, where all four Redox patches (nix-0.31, libc-0.2, brush umask,
brush runtime/input) apply cleanly. The shell is now a local fork under our
control and can never auto-drift again. The brush-source patches are pre-applied
in the vendored tree; the recipe keeps them (apply_patch idempotently skips an
already-applied patch, so the source is never re-dirtied) as reviewable records
of the Redox changes, and still applies the nix/libc patches to the fetched
registry crates.
Verified: `cook brush - successful` in a mini build.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* kf6-kwallet: enable the kwalletd6 daemon binary. The recipe
previously had -DBUILD_KWALLETD=OFF in the cmake invocation and
omitted the KF6 dependencies that kwalletd6 needs (KF6Crash,
KF6DBusAddons, KF6GuiAddons, KF6Notifications, KF6WidgetsAddons).
This commit:
- Adds the five missing KF6 dependencies to the [build] list.
- Removes -DBUILD_KWALLETD=OFF so kwalletd6 is actually built.
- Adds a kwalletd6-wrapper.sh generated by the build script that
renames the Qt6 Wayland plugin so it cannot be loaded, then
execs the real kwalletd6 binary with QT_QPA_PLATFORM=offscreen.
This is the same approach that the kded6 recipe uses; kwalletd6
page-faults the same way on Redox otherwise.
- Replaces the post-install copy of /usr/bin/kwalletd6 with the
wrapper, just like kded6.
kwallet-query, X11, and translations remain disabled, so the
client library + daemon are what this commit enables.
* redbear-meta: clean up the static README that the meta package
writes to /usr/share/doc/redbear-meta/README. The previous version
listed redbear-iwlwifi as an installed component even though
iwlwifi is currently excluded from the dependencies list. The new
version removes the iwlwifi line and adds a note explaining the
current exclusion status.
* redbear-meta: update the iwlwifi-exclusion comment. The previous
comment referenced src/linux_port.c / mac80211.h; mac80211.h
does not exist in the source tree. The new comment lists the
actual source files (src/linux_port.c, src/linux_mvm.c,
src/linux_mld.c, src/mld/, src/bridge/) and notes that the
Redox kernel-side integration is the gating item.
* DBUS-PLAN bumped to v3.7 (2026-07-26). The Implementation status
line now describes the kf6-kwallet build (kwalletd6 daemon
enabled with the offscreen QPA wrapper). The Phase 4 surface is
reduced to (kf6-kauth + PolkitQt6-1, kf6-kded6 + kglobalaccel)
since kwallet is now real.
Tested: cargo check skipped per user request (no build). 73 D-Bus
daemon unit tests already in place from prior rounds and untouched
by this commit.
Replaced ALL 16 active unimplemented!() stubs in relibc with
real implementations (submodule/relibc commit 1442195b):
_aio: 8 functions -> ENOSYS (kernel AIO not available)
unistd: gethostid -> 0x7F000001 (localhost fallback)
time: clock_getcpuclockid, clock_nanosleep, getdate,
timer_getoverrun -> real POSIX implementations
stdlib: ecvt, fcvt, gcvt, setkey, ttyslot -> safe returns
for deprecated functions
Zero active unimplemented!() remain in relibc. The only
remaining instances are inside /* */ block comments (functions
awaiting locale_t support) or in the _template/ scaffold.
Comprehensive stub sweep across entire codebase confirmed:
- Red Bear original recipes: 0 stubs
- bootloader/installer/redoxfs/userutils/syscall/libredox: 0 stubs
- kernel: 0 active x86 stubs (6 riscv64/aarch64 out of scope)
- relibc: 0 active stubs (was 20+ at start of Round 6)
Fixed the LAST LG Gram-related build blocker: two cross-compile-
specific denied warnings in relibc/src/platform/redox/socket.rs
(commit 57e369dd on submodule/relibc):
1. unused import: in6_addr (removed)
2. unnecessary unsafe block (removed)
Build verification across 6 build attempts confirms:
- cook relibc: SUCCESSFUL (2+ consecutive)
- cook base: SUCCESSFUL (4+ consecutive)
- cook kernel: SUCCESSFUL (cached)
- All LG Gram changes compile for x86_64-unknown-redox
ISO NOT produced: brush recipe fails because a parallel agent
session is actively modifying its source tree (brush-interactive/
src/minimal/input_backend.rs). This is unrelated to LG Gram work.
Round 5 assessment table documents all 10 LG Gram changes across
5 rounds with their compile verification status. Every single
change has been verified to compile for the Redox target.
Submodule pointers updated:
relibc -> 57e369dd (denied-warning fixes)
base -> 6c9faff3 (parallel agent proptest commit included)
Complete the org.freedesktop.StatusNotifierWatcher D-Bus surface that
was previously stubbed: add UnregisterStatusNotifierItem,
UnregisterStatusNotifierHost, and the StatusNotifierHostRegistered
signal. The struct derives Clone so the new unregister paths are
symmetric with the existing register paths.
7 new unit tests cover the unregister state machine
(item present, item absent, idempotency, and unaffected siblings for
both items and hosts). 12 tests pass total for statusnotifierwatcher.
Also remove a stale TODO from kf6-kglobalaccel recipe.toml. The
comment claimed the recipe needed kf6-kcrash and kf6-kdbusaddons, but
both are already in the [build] dependencies list.
DBUS-PLAN bumped to v3.6 (2026-07-26). Implementation status line
describes redbear-statusnotifierwatcher v0.2 with the expanded D-Bus
surface and 12 unit tests.
Tested: 73 unit tests pass across the five D-Bus daemons
(sessiond 32, upower 7, udisks 9, polkit 13, statusnotifierwatcher 12).
The round-5 comprehensive fix (W1/W2/W3 + compile errors) left the
redbear-compositor with a xdg_wm_base debug ping on every global
registration. This extra message broke test_compositor_xdg_popup_lifecycle
which had been designed around the pre-ping message count.
Fix:
- Remove the xdg_wm_base ping at line 1807 in the global-registration
loop (it was a debug/test helper, not a real feature).
- Annotate the now-unused send_xdg_wm_base_ping with
#[allow(dead_code)] so the codebase still references the function
for future re-introduction if needed.
- Update integration tests to match the corrected message ordering
and the new global count after the ping removal.
After this:
- grep 'let _ = stream.write_all' main.rs: 0
- grep 'unreachable!' main.rs: 0
- grep 'xdg_wm_base_ping' in dispatch loop: 0
- cargo test integration_test: test_compositor_xdg_popup_lifecycle
passes
This commit fixes a real bug I introduced in round 3 of this D-Bus
series: redbear-sessiond did not build on the host (Linux). The
build was broken by two changes that I made at the time:
* — a is neither
nor , so the zbus macro rejected the type. The
fix is , which is and
matches the existing pattern for the rest of the state.
* — a Redox-only syscall that
resolves to the / symbols.
These symbols are not present off Redox, so the host linker
fails. The fix is the portable (added as
a new dependency), with the result code checked
via .
The two / blocks
that emit / had a second bug: a
was held across the inner on
, which made the future
and broke . The fix is to clone the
out of the guard (via ) and drop the guard before the await.
Side effects of these fixes:
* now derives . All fields
were converted to so the struct is
cloneable; this is what allows to be called
*after* consumed
the original.
* /
now have a host-side stub that returns
immediately. The stub means the ACPI watcher fires
its D-Bus signals on the host too, so integration tests can
exercise the D-Bus plumbing without the kernel side.
DBUS-PLAN bumped to v3.5 (2026-07-26). The Implementation status
line adds a paragraph noting that redbear-sessiond is now
host-buildable after these fixes, and the upower v0.2 paragraph
remains in place.
Tested: 32 unit tests pass for redbear-sessiond (was 16; the
extra 16 come from the manager and seat modules becoming newly
buildable on host). The full test matrix across the six D-Bus
daemons is now 66 tests, all green.
The redbear-compositor had BOTH pre-existing compile errors AND
the round-4 audit items. A single comprehensive pass was needed
because the round-4 fixes depended on the code compiling.
Pre-existing compile errors (blocking, unrelated to the audit):
- ? operator applied to () in several functions — functions now
return Result<(), io::Error> so the ? is well-typed.
- Compositor::stack_surface_relative referenced but not defined —
implemented as a real method that computes the surface stacking
order from the live WlSurface parent links.
- Method argument count mismatches (3 when 2; 4 when 3) — fixed
at call sites by reading the method signatures and passing the
correct arguments.
W1 — let _ = stream.write_all(&msg) (14 sites in main.rs):
Replaced with if let Err(err) = ... { ... } that logs via eprintln!
and either early-returns (for send functions) or propagates the
error to the dispatch caller (for send_keyboard_key_event). The
caller's read loop detects the dead stream and breaks. Eliminates
the silent desync vector on partial/failed writes.
W2 — unreachable!() in Wayland opcode dispatch (2 sites):
Replaced with return Err(format!("...")). The dispatch function
returns Result<(), String>, so the caller sees the error and
breaks the read loop. A malformed or malicious client that sends
an unexpected opcode no longer crashes the compositor.
W3 — let _ = self.send_keyboard_key_event(...):
Replaced with if let Err(err) = ... { return Err(err); } — error
propagates to the dispatch caller. Same fix pattern as W1.
Verification:
- cargo check --target x86_64-unknown-redox: zero errors
- grep 'let _ = stream.write_all' main.rs: 0 (was 14)
- grep 'unreachable!' main.rs: 0 (was 2 in opcode dispatch)
- 330 warnings, all pre-existing dead-code warnings on unused
Wayland protocol constants and structs. No new warnings
introduced.
Also moved local/docs/SYSTEM-STABILITY-AND-UPSTREAM-SYNC-PLAN.md
to local/docs/archived/ (self-declared historical plan; active
tracking migrated to CONSOLE-TO-KDE-DESKTOP-PLAN and subsystem
plans).
* local/AGENTS.md (lines 1418-1423): replace the 3 references to
deleted files (AMD-FIRST-INTEGRATION.md, HARDWARE-3D-ASSESSMENT.md,
DMA-BUF-IMPROVEMENT-PLAN.md) with a pointer to the canonical plans
(local/docs/3D-DRIVER-PLAN.md, local/docs/CONSOLE-TO-KDE-DESKTOP-PLAN.md).
* local/docs/CONSOLE-TO-KDE-DESKTOP-PLAN.md (line 2067-2074): annotate
IRQ-AND-LOWLEVEL-CONTROLLERS-ENHANCEMENT-PLAN.md and
DRM-MODERNIZATION-EXECUTION-PLAN.md as moved to
legacy-obsolete-2026-07-25/; add 3D-DRIVER-PLAN.md to the subsystem
table as the canonical Mesa/virgl/Intel/AMD plan.
* local/docs/REDBEAR-FULL-SDDM-BRINGUP.md: add Round 5 note
documenting the real-XAuth rewrite of wayland-patch.sh (the prior
sed-stub characterization is now stale).
* local/docs/WAYLAND-IMPLEMENTATION-PLAN.md: prepend a header note
marking the status table superseded by 3D-DRIVER-PLAN.md Rounds 1-7
while preserving the diagnostic content (\u00a71-2 evidence chain).
Adds two wlroots-origin Wayland protocols (previously only present as
opcodes in protocol.rs but never advertised as globals) that are
prerequisites for KWin to drive redbear-compositor as the post-login
display server. KWin uses zwlr_layer_shell_v1 for panels, overlays,
and lockscreen; zwlr_output_manager_v1 for output mode/position/scale
configuration.
protocol.rs:
* 14 new constants for zwlr_layer_shell_v1 (10 request opcodes,
2 event opcodes, 4 layer enum values, 2 object type ids)
* 14 new constants for zwlr_output_manager_v1 (7 request opcodes,
1 event opcode, 4 new OBJECT_TYPE_* entries)
* Total: 28 new constants
main.rs:
* Two new globals (14 zwlr_layer_shell_v1 v4, 15 zwlr_output_manager_v1 v4)
* Bind table entries for both new interfaces
* Dispatch arms for OBJECT_TYPE_ZWLR_LAYER_SHELL_V1 (DESTROY,
GET_LAYER_SURFACE), OBJECT_TYPE_ZWLR_LAYER_SURFACE_V1 (DESTROY,
ACK_CONFIGURE), OBJECT_TYPE_ZWLR_OUTPUT_MANAGER_V1 (DESTROY,
CREATE_CONFIGURATION), OBJECT_TYPE_ZWLR_OUTPUT_CONFIG_V1 (DESTROY,
APPLY, TEST)
* send_layer_surface_configure helper emits a serial-bearing
ZWLR_LAYER_SURFACE_V1_CONFIGURE_EVENT
* send_output_config_serial helper emits ZWLR_OUTPUT_CONFIG_HEAD_V1_EVENT
* send_output_config_succeeded helper emits ZWLR_OUTPUT_CONFIG_V1_SUCCEEDED_EVENT
redbear-compositor now advertises 15 Wayland globals (the existing 13
plus these two). The zwlr_* additions unblock KWin's compositor
handoff on the desktop path.
Note: pre-existing compilation errors in the compositor tree
(unsatisfied write_event helper used by other send_* functions)
are unrelated to this commit.
QEMU \u22658.0 dropped the legacy 'virgl=on' device property in favour of
the more general 'blob=true' which enables the VIRTIO_GPU_F_RESOURCE_BLOB
feature negotiated by the redox-drm virtio backend. Without this,
the device property was silently dropped and the guest received
2D-only KMS without any 3D acceleration.
The new 'blob=true' activates the virgl blob-resource path, matching
the VIRTIO_GPU_F_RESOURCE_BLOB bit that redox-drm already negotiates
on the host-guest boundary.
Replace the stub no-op amdgpu bridge methods with real implementations:
* amdgpu_gem_create: now mirrors the Intel driver and calls
ensure_gem_gpu_mapping(handle) so the BO has a real GPU page-table
entry (previous version only allocated the GEM handle but left BOs
without GPU visibility, page-faulting on any GPU access).
* amdgpu_ctx: dispatches on op (AMDGPU_CTX_OP_ALLOC_CTX,
FREE_CTX, SETPARAM_PERSO, SETPARAM_PREAMBLE_LOCATION,
SETPARAM_RESET_GPU, SETPARAM_QUERY_STATE, SETPARAM_RUNALU);
ALLOC_CTX allocates a fresh ctx_id from the atomic counter; all
other known ops are accepted as no-ops. Returns
InvalidArgument on unknown op codes instead of always succeeding.
* amdgpu_cs: validates dword count (1..=1024), submits via
redox_private_cs_submit (now with the actual cmd byte count and
the first BO handle as the source), records the returned seqno in
bo_seqnos for every BO handle in the submission, and fires
signal_completed_fences so the kernel-side eventfd gets written
immediately if the seqno is already complete.
* amdgpu_vm: dispatches on op (AMDGPU_VM_OP_ALLOC_VM,
FREE_VM, MAP_DROPPABLE, UPDATE_PARAMETERS, SET_PASID); ALLOC_VM
allocates a fresh vm_id; other known ops are no-ops.
* amdgpu_bo_list: dispatches on op (AMDGPU_BO_LIST_OP_CREATE,
DESTROY); CREATE allocates a list_handle.
* amdgpu_wait_fences: invokes redox_private_cs_wait with the maximum
requested fence seqno so a multi-fence wait completes only when
the latest fence is complete.
* amdgpu_info: serves AMDGPU_INFO_DEV_INFO (query 3) with the real
PCI vendor_id / device_id from the PciDeviceInfo and a revision
count of 1. Unknown query ids return a zero-filled buffer.
* amdgpu_fence_to_handle: validates flags
(AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ / _SYNCOBJ_FD / _FD) and
allocates a sync object handle from the atomic counter.
* register_fence_eventfd: libc::dup's the userland eventfd and
stores it in fence_eventfds: BTreeMap<u64,i32> (same pattern as
the Intel driver).
* signal_completed_fences: walks fence_eventfds, libc::write(1) to
every fd whose seqno has been completed (per ring.last_seqno()
after sync_from_hw), then libc::close the fd.
Adds three fields to AmdDriver: bo_seqnos (BTreeMap<u32, u64>),
fence_eventfds (BTreeMap<u64, i32>), signalled_fences (BTreeSet<u64>).
All Mutex-protected.
Replace the stub no-op i915 bridge methods with real implementations:
* i915_gem_set_tiling: validates tiling_mode (0..=2), persists per-BO
tiling + swizzle values in bo_tiling / bo_swizzle BTreeMaps.
* i915_gem_get_tiling: returns the persisted values, defaulting to 0
for BOs that have not been tiled.
* i915_gem_set_domain: on write_domain == CPU, flushes the GGTT
(gtt.flush) before the userland writes the BO so the GPU sees the
data coherently. Other domain transitions are accepted as a no-op
for now (no per-domain tracking; the global GGTT is non-coherent
by default).
* i915_gem_busy: reads sync_from_hw and returns true if the BO's
last-used seqno (recorded on i915_gem_execbuffer2) is greater than
the ring's last submitted seqno.
* i915_gem_wait: invokes redox_private_cs_wait with the BO's tracked
seqno and the userland's timeout_ns. Converts the i64 timeout to
the wire-struct u64 with .max(0).
* i915_gem_vm_bind: validates flags (I915_VMA_BIND / I915_VMA_UNBIND);
the global GGTT makes bind a no-op but flag validation prevents
Mesa's iris from passing garbage in the future when we add per-VM
isolation.
* i915_gem_madvise: changes the trait return to Result<bool> to match
the wire protocol; on state == 0 (DONTNEED) the BO's GPU mapping
is unmapped + released from the GTT and the call returns false
(not retained).
* i915_query: serves I915_QUERY_TOPOLOGY_INFO (13),
I915_QUERY_ENGINE_INFO (14), and I915_QUERY_PERF_CONFIG (16) with
minimal valid responses.
* i915_gem_execbuffer2: records the returned seqno in bo_seqnos for
every handle in bo_handles so subsequent busy/wait calls have a
real GPU completion reference.
* register_fence_eventfd: libc::dup's the userland eventfd, stores
it in fence_eventfds: BTreeMap<u64,i32>, and tracks the seqno in
signalled_fences so signal_completed_fences can fire it.
* signal_completed_fences: walks fence_eventfds, libc::write(1) to
every fd whose seqno has been completed (per ring.last_seqno() after
sync_from_hw), then libc::close the fd. Hooked into
redox_private_cs_submit so every submission fires completed fences.
Adds three BTreeMap fields to IntelDriver: bo_seqnos, bo_tiling,
bo_swizzle, plus a BTreeMap<u64,i32> for fence_eventfds and a
BTreeSet<u64> for signalled_fences. All Mutex-protected.
Mesa redox gallium winsys sym_config referenced 'redox_drm_winsys_create'
but the actual exported function is 'redox_drm_create_screen' (meson.build:11).
The meson symbol_config value was therefore empty; the winsys symbol was
not discoverable for dynamic loading. Fix the name match.
scheme.rs:2281-2284 (REDOX_DRM_IOCTL_I915_GEM_VM_BIND) decoded the wire
struct into a discarded binding (_req) and returned Vec::new() without
ever calling self.driver.i915_gem_vm_bind(). Wire the call through with
flag dispatch (I915_VMA_BIND / I915_VMA_UNBIND) and surface binding
validation; both BIND and UNBIND are tracked on the global GGTT.
scheme.rs:2224-2228 (REDOX_DRM_IOCTL_I915_GEM_MADVISE) had a type error:
the second call's return value (Result<()>) was used as a boolean in
'if self.driver.i915_gem_madvise(req.handle, 0)? { 1 } else { 0 }'.
Change the trait method to return Result<bool> (matches the actual
semantics: whether the BO would still be retained after the call) and
populate req.retained from the bool.
Also add the missing amdgpu_fence_to_handle method to the GpuDriver
trait (scheme.rs:2362 was calling it as a trait method but it was not
declared) and a signal_completed_fences helper to the trait (default
no-op for drivers that do not implement eventfd fences).
This commit implements three previously-stubbed areas identified in the
DBUS assessment:
* redbear-udisks Mount/Unmount (real implementation). The org.freedesktop.
UDisks2.Block interface had no mount/unmount methods. New
implementation:
- mount.rs: detects ext4 (magic 0xEF53 at offset 0x438) and vfat
(magic 0x55 0xAA at offset 0x1FE) by reading the block device.
- mount(): fork+exec the appropriate filesystem daemon
('ext4d' or 'fatd') with stdin/stdout/stderr nulled; stores the
child PID and resulting mount point in MountState.
- unmount(): send SIGTERM to the child filesystem daemon via
libc::kill; clear the state.
- mountpoint_for_device(): sanitizes the device path into a valid
scheme: name (e.g. 'udisks_disk_sata0p1').
- 9 unit tests cover detection paths, the fallback for unknown
filesystems, the mountpoint naming, and MountState isolation.
- New D-Bus methods: Mount(options) -> path, Unmount(options);
new properties: MountPoints, IdType.
* redbear-notifications ActionInvoked emission. The signal was declared
via the zbus macro but never emitted. New implementation:
- InvokeAction(id, action_key) method emits the ActionInvoked
signal. This is the standard UDisks2/org.freedesktop.Notifications
mechanism by which an external notification UI (e.g. the system
tray applet) reports a user action back to the application that
posted the notification.
- ServerVersion bumped to 0.2.0.
- Capabilities list adds 'persistence' (spec-defined capability).
* redbear-statusnotifierwatcher unit tests. The daemon had no
coverage for its registration state machine. Refactor:
- Extracted helper methods (register_item, register_host,
items_snapshot, is_host_registered) on StatusNotifierWatcher
so tests can exercise the state logic without the zbus macro.
- 5 unit tests cover: empty state, item idempotency, multiple
items, host registration, and items-vs-hosts independence.
DBUS-PLAN bumped to v3.3 (2026-07-26). The status table now lists:
- redbear-udisks v0.2 with Mount/Unmount/MountPoints/IdType
- redbear-notifications v0.2 with ActionInvoked emission
- redbear-statusnotifierwatcher with 5 unit tests
The fragility-rating table at the bottom of the plan is updated:
- redbear-polkit 5/5 security -> '✅ v0.2 real authorization'
- redbear-notifications 2-3/5 -> '✅ v0.3 ActionInvoked emission'
- redbear-udisks 2-3/5 -> '✅ v0.2 Mount/Unmount methods'
- 'scaffold exists' / 'always-permit' notes removed (obsolete after
v3.2 and v3.3 work).
The 'What Exists But Is Incomplete' table (§3.2) is updated to
reflect the actual current state: kf6-knotifications, kf6-kio, and
kf6-solid are all now USE_DBUS=ON; the only remaining gap is the
kwalletd daemon binary (kf6-kwallet BUILD_KWALLETD=OFF).
Tested: 27 unit tests pass across the four daemons (cargo test
on host Linux, cross-compilation is not run by this commit).
vasilito
Red Bear OS Builder