Fix/cpu bugs - #52
Merged
Merged
Fix/cpu bugs#52
Conversation
Two low-risk fixes targeted at CPU-0 saturation under multi-core desktop load. serial: the transmit-empty check was stubbed out with `(void)` so every character was written to COM1 unconditionally. At console spam / boot-log rates the guest outruns the host UART and bytes are silently dropped. Add a bounded THRE busy-wait before each outb so back-pressure is respected without any risk of wedging a caller that runs with IRQs masked. framebuffer: the legacy present path always called framebuffer_vmware_sync after every update. That helper both rings the SVGA doorbell (SYNC reg) and spins reading SVGA_REG_BUSY until the host has drained the command FIFO. Because framebuffer_present_region runs under the global execution lock with local IRQs masked, that busy-wait became the dominant per-frame stall: one core spent each present inside SVGA_REG_BUSY polling, and every other core's syscall blocked behind the global ticket. Change it so the fast path only kicks the doorbell (new framebuffer_vmware_kick, non-blocking), and the synchronous drain happens only when the update fails because the ring is full — which is exactly when we must wait to make progress before retrying. The newer SVGA backend already rings its own doorbell at packet publish (fifo.c:85), so the change is a strict reduction of work on that path. The global execution lock itself is not touched here; that is the next step and needs VM feedback to sequence safely.
Reconcile the architecture documents against what the tree actually does now. Date-stamped status/audit/progress/handoff records are left as historical snapshots and not rewritten. - EEVDF_SCHEDULER.md / EXECUTION_LOCK.md: the claim `SMP_USER_SCHEDULER_ENABLED=0` / "AP scheduler disabled" / "active CPU = 1" is stale. smp.c has no such flag and every AP runs userland_schedule_from_frame -> arch_enter_user_frame. Rewrite to state APs schedule user tasks, while keeping the two real, still accurate limits: all device IRQ stay on the BSP, and the global execution ticket lock serialises cross-core syscalls. - DESKTOP_PERFORMANCE.md: document the 2026-09-24 asynchronous present change (doorbell-only on the fast path, synchronous FIFO drain only on backpressure) and why it mattered under the execution lock. - DRIVERS.md: list the ac97 and es1371 loadable modules that already build (DRIVER_NAMES in mk/boot.mk) instead of only mouse/serial/e1000. - RPR.md: reflect that `make pages` now also emits the generated Documentation section (docs/ rendered by tools/build/md2html.awk) and that verify-pages.sh rejects raw Markdown leaks. - README.md (docs index): add pointers to the build-system, RPR, scheduler, execution-lock and desktop-performance documents that existed but were not listed.
Second pass over every non-timestamped doc, with every claim re-verified against the tree before editing: - ABI.md: rewrite Time/Device/Driver/Appearance/Auth/Machine-Identity/ Network/Audio/Service sections around the real interfaces (OpenRC leonos-ntp/leonos-dhcp + /run state files, PAM + /etc/passwd accounts, devmand/sessiond AF_UNIX services, LEONOS_NET_CONTROL_IOCTL on AF_INET fds, /dev/driverctl control ioctl, sysfs DMI identity, OSS-only audio); devfs node list now matches storage_vfs.c exactly (drop phantom /dev/net0 and /dev/audio0, add gpu/rtc/driverctl/kmsg/shm0/aliases). - SYSCALLS.md: replace the removed private ioctl-multiplexer catalogue with the actual per-node device ABI; document the real socket syscall dispatch, udhcpc lease flow (no kernel DHCP; netctl/oobe exception), and rcctl-based service control; correct GPU dispatch to /dev/gpu. - unix-ipc-protocol.md: remove the never-shipped authd/netmand sections (accounts are PAM, network is kernel control ioctl + OpenRC), drop the fictional serviced/ hosting, fix the migration table file inventory. - FILESYSTEM.md: correct the /dev node list and input-method retirement to /run/leonos/input-method.sock. - ABI_MIGRATION.md: dirfd-relative openat, signal handler frames, Unix98 hangup, windowd GUI migration, device list and input-method rows now match implemented state. - KERNEL_DEBUG.md: probe registry is 41 syscall entries, ioctl probes empty (removed with the private ioctls). - BOOT_AND_INTEGRITY.md: runtime root is ext2, not exFAT. - BUILDSYSTEM.md: components.toml stage/entry/sdk/api fields, loader under generated/boot, APK version 2.<epoch>-r0, real smoke markers. - APK_PREPARATION.md / ROOTFS / UPSTREAM_TOOLS / THIRD_PARTY / BUILD_AND_INSTALLER: dead `build.py` invocations and obsolete build/ paths replaced with make targets and out/x86_64/release paths; ownership groups reference configs/apk-ownership.json, staged by tools/build/rootfs-stage.sh. - SOURCEHUT_SYNC.md: secret is SOURCEHUT_REPOSITORIES (plural, multi- line). BROWSER.md: litehtml pin is b9e89f0; no gen_ninja.py. - APP_REGISTRY.md: manifest example includes the generated system= key. - INSTALLER_ACCOUNTS.md: component inventory comes from the APK database and system=1 markers, not the phantom components.list.
Replace the single aggregate CPU polyline with one colored line per core in the same plot, each core keeping a stable color from a 32-entry palette (cycled for larger machines). Per-core samples ride the shared history ring so all graphs advance together, offline cores leave gaps, and the header still shows aggregate utilization. The per-core usage list gains a matching color swatch so the line-to-core mapping stays readable without an in-plot legend.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.