Linehu and Cursor
8554a82677
feat(panic): FUCK [panic] mode=strict|fallback + realtest image variant
...
- UTSM panic_full returns in fallback mode (logs [PANIC:FALLBACK] + code + hit
count); panic_halt keeps always-halt sites (CPU exception, VMM-E11 LA57,
VMM-E15 DSK jump, UTSM-E42 test hook). Call sites given safe continuations.
- Mode passed to DSK/FirstInit/login/desktop via DSK_BOOT_FLAG_PANIC_FALLBACK;
DKM kernel_api.panic / kern.log.panic return to driver; ext4 fatal aborts the
op via setjmp/longjmp and DSK falls back to FAT32 ESP.
- build.ps1/Makefile: dev=strict, release=fallback, new realtest (release
options + strict); default -Variant all; serialized iso packing under -j.
- Fix '*/' inside comments (dkm.h, kernel_api.h, bus.c) that broke the build.
- Includes pending work: dkm_bus service bus, mt7920/wlan drivers, desktop
files/startmenu/prank, pe-tools, diskinfo, docs.
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-10-12 23:57:19 +08:00
Linehu
d95ae091e7
fix(dkm): support R_X86_64_PLT32 relocations in ELF loader (DKM-E37)
...
Multi-TU drivers produce PLT32 for cross-object calls (lld keeps a
relocation for potentially-interposable global symbols). ath9k.drv
(ath9k.o + wpa2_crypto.o) has 22 such relocations, which tripped
"DKM-E37 UNSUPPORTED RELOCATION" at boot.
PLT32 has the same semantics as PC32 in a static relocatable
(patch = S + A - P), and is a 4-byte write. Added the constant, the
patch computation, and switched the write-width branch to
"not R_X86_64_64 -> u32" so all 4-byte types are handled uniformly.
Verified: all 28 .drv files now use only
{R_X86_64_32S, R_X86_64_64, R_X86_64_PC32, R_X86_64_PLT32}.
2026-10-06 21:37:37 +08:00
Linehu
174af71b39
feat: exhaustive STUB instrumentation - 241 panic sites across VMM/kernel + whitelist alignment
...
- VMM cluster: XJ3-E/LIN-E/VMM-E/VMX-E/EPT-E/IPC-E/VVB-E (124 sites)
- kernel cluster: TST-E/MM-E/UTRW-E/LAPIC-E/IDT-E/SAS-E/PRB-E/NETST-E/DKM-E (117 sites)
- whitelist fixes: vmm_init env probes E10-E14 fallback, disk_log read-only block path, DSL-E05 non-FAT32 device 0 probe
- QEMU smoke: wizard -> USER.CONF -> login -> desktop ready, zero spurious panic
2026-10-03 21:26:04 +08:00
Linehu
3aa15e7b21
feat: D4 DKM colored log prefixes - every driver has its own temper
...
- log.c: color context (log_set_color / log_color_enable); ANSI only goes to serial, BOOTLOG.TXT buffer stays clean plain text
- kernel_api.c: dkm_log_set_driver() — driver-name FNV-1a hash into an 8-color bright ANSI palette, deterministic per driver
- load sites (dkm_core builtin + boot_modules external) set color around driver_init and clear after
- FUCK [debug] color_log gate (dev=1, release=0)
- QEMU verified: 1150 colored lines, per-driver stable colors (.build_tmp/qemu_serial_d4color.log)
2026-10-02 19:48:54 +08:00
xyou
efa626ed17
feat: Phase 1 网络端到端打通 + FirstInit/Login/DevMode 完善
...
- 用户态共享网络协议栈 CODE/tools/net_stack.h（eth/ARP/IPv4/ICMP/UDP-DNS/TCP static inline，轮询模型）
- shell 内建 ping/curl/run 命令；ping.elf/curl.elf 对齐 net_stack.h 真实协议栈
- shell dev_mode 串口镜像（g_serial_mirror）便于自动化验证
- DSK 启动流程：firstInit.txt 双行格式（firstInit + dev_mode），dev_mode=1 触发 shell 自动回归测试
- FirstInit 完成→持久化 USER.CONF + 翻转 firstInit 0->1（保留 dev_mode）
- firstInit=1 时 DSK 加载 login.elf 进行用户选择 + 密码校验后再进 desktop
- 新增 CODE/login/ 登录应用、CODE/DKM/dkm_shared.h、Logo.png
- ahci/e1000/virtio_net/fat32/console_fb 驱动更新；utsm serial/block/dsk_loader 调整
- build.ps1 打包 ping.elf/curl.elf 进 SATA FAT32 根目录
- 移除 CODE/tools/bash（与 shell.elf 重复）
2026-07-26 21:07:36 +08:00
xyou
f1a2a110d7
Record networking and DMA development progress
...
Add the current OS source tree, build scripts, system image layout, and development notes covering DMA page allocation, e1000 bring-up, net API dispatch, and DHCP work in progress.
2026-07-02 16:17:09 +08:00
