kernel: ship blob .cmd stubs + gate open_pm + add dev_warn shim (cv200/av100 neo unblock)#171
Merged
Merged
Conversation
Linux 7.0 modpost (scripts/mod/modpost.c) calls read_text_file() on the
.<basename>.o.cmd sibling of every .o referenced from a module's .mod
list, and exits non-zero if the file is missing. For prebuilt blob .o
files (cv200's hi3518e_*.o, av100's hi3516a_*.o) there's no kbuild rule
that generates the .cmd — they're committed binaries with no build
metadata.
cv500 already has these .cmd files committed (31 of them, dating back
to the cv500 mainline-kernel bring-up). cv200 and av100 didn't — which
worked fine on 4.9 lite (modpost was lax) and on standalone CI builds
(actions/checkout fetches dotfiles), but breaks on firmware-tarball
neo builds where the github-archive tarball strips them.
Each .cmd is a one-line stub:
savedcmd_kernel/obj/<chiparch>/<obj>.o := :
satisfying modpost's read_text_file() without claiming any CRC info
(no #SYMVER lines). Linker symbol resolution is unaffected.
47 files total (21 cv200 + 26 av100). Unblocks cv200_neo and
av100_neo firmware builds end-to-end.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two follow-ups discovered while exercising the hi3516av100_neo firmware build for the first time (cv200 was unblocked by #170; av100 hit two further issues this commit closes): - kernel/hi3516av100.kbuild: gate open_pm.ko build behind DISABLE_PM (opensdk already passes DISABLE_PM=1 by default; matching the gate makes the directive effective). The vendor blob hi3516a_pm.o references dev_pm_opp_init_cpufreq_table, cpufreq_table_validate_and_show, cpufreq_generic_attr — all removed in mainline. Even with CONFIG_CPU_FREQ + REGULATOR enabled, modpost still fails. open_pm.ko is the AVS / CPU-DVFS blob; not needed for boot. Lite (4.9) is unaffected — the DISABLE_PM=1 default already applied there and the previous unconditional obj-m made it a no-op anyway. - osal_v2_shim/cv200_shim.c, osal_v2a_shim/av100_shim.c: add dev_warn export. Same shape as the dev_err shim landed in #170 — macro in <linux/dev_printk.h> that funnels through _dev_err; variadic wrapper via dev_vprintk_emit at KERN_WARNING (loglevel digit 4). av100 blobs reference dev_warn from open_pm (gated out above) and from other vendor sources; safer to ship the shim for symmetry. Verified end-to-end: hi3516av100_neo now builds and boots to login prompt with eth0 up via the new mainline higmac driver (openipc/linux feat/av100-mainline), DHCP from SLIRP, ping 10.0.2.2 succeeds. cv200_neo continues to meet the same DoD. cv200_lite and av100_lite (4.9 kernel) remain byte-equivalent to nightly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
widgetii
force-pushed
the
feat/v2-shim-expand-modpost-syms
branch
from
ea1f8c1 to
73d45f4
Compare
This was referenced May 22, 2026
Merged
widgetii
added a commit
that referenced
this pull request
May 24, 2026
* kernel/hi3516cv100: V1 OSAL shim + mmz source patches for Linux 7.0 Brings the cv100 (V1, ARM926EJ-S) module set up against the openipc/linux 7.0 base used by cv200_neo / av100_neo. cv100 has no OSAL abstraction — the vendor .o blobs in kernel/obj/hi3516cv100/ embed kernel API calls verbatim against the 3.0.8 SDK base. Two-layer fix mirrors the cv200 #162 pattern: * `kernel/osal_v1_shim/cv100_shim.c` — re-exports legacy kernel symbols the cv100 blobs reference (do_gettimeofday, register_sysctl_table, init_timer_key, strlcpy, _cond_resched, __kmalloc, kmem_cache_alloc, __memzero, printk, del_timer, vmalloc, sched_setscheduler). Gated to >=5.0 so on the cv100_lite production target (3.0.8) the module is a benign no-op shell. Strict subset of v2_shim coverage; symbols cv200 needs but cv100 doesn't (PDE_DATA, dev_err/warn, jiffies_to_msecs, register_sysctl_paths) are omitted. * `kernel/mmz/hi3516cv100/{media-mem,kcom,mmz-userdev}.c` — strip mach/hardware.h and asm/system.h (removed in modern ARM); guard asm/outercache.h with CONFIG_OUTER_CACHE; add the 3.10 proc-fs version branch (proc_create + COMPAT_USE_PROC_OPS) for media-mem's /proc entry; switch mmz_write_proc signature for the new fops/proc_ops .write slot. * `kernel/cipher/hi3516cv100/src/cipher_intf.c` — drop asm/system.h. * `kernel/hi3516cv100.kbuild` — wire v1_shim ahead of mmz with the same CFLAGS_*=-D_cond_resched=...unused_inline trick cv200 uses to suppress the static-inline definition from <linux/sched.h>. * kernel/{rtc,ir}/hi3516cv100: drop IO_ADDRESS, switch to ioremap_nocache Both files keyed their MMIO macros off `IO_ADDRESS(phys_addr)`, the mach/hardware.h static virtual-mapping table that's gone on modern ARM (removed alongside mach support cleanup). cv200's equivalent files had already been refactored to ioremap-backed globals pre-#162 — cv100's source predates that work. * `rtc/hi_rtc.c`: hold ioremap_nocache results for the CRG (0x20030000) and SPI (0x20060000) windows in `rtc_{crg,spi}_base_addr` globals, redefine CRG_BASE_ADDR / RTC_SPI_BASE_ADDR to point to them. All writel/readl call sites get a `(void *)` cast (modern __iomem-typed prototype). The temperature_detection timer callback gets the COMPAT_TIMER_SETUP signature swap from kernel_compat.h — same dual- signature pattern as cv200 hi_rtc.c. iounmap both windows in rtc_exit(). * `ir/hiir.c`: ioremap_nocache the 0x40-byte IR register window (0x20070000) and the single-word IOCONFIG cell (0x200f01F8) at init, redefine IR_REG_BASE / IOCONFIG as casts of the resulting void __iomem* pointers so WRITE_REG/READ_REG keep their offset arithmetic. iounmap on exit. Both modules' writel/readl indirection through ((volatile u32 *)addr) in WRITE_REG/READ_REG works equally well with the new pointer-typed base addresses cast through unsigned long. * kernel/obj/hi3516cv100: ship .cmd stubs for blob .o files Mirrors PR #171 for cv200/av100 — modern kbuild's modpost requires a companion .<file>.o.cmd file alongside each .o. The .o blobs in kernel/obj/hi3516cv100/ were extracted from vendor SDK V1.0.B.0 with no .cmd metadata, so a stub `savedcmd_kernel/obj/.../foo.o := :` satisfies modpost on >= 5.x kernels without changing the build flow on 3.0.8 (modpost on older kernels doesn't consult the .cmd files). Force-added past .gitignore (which blanket-excludes .*.cmd because buildroot's normal flow regenerates them from .o builds). * kernel/osal_v1_shim: extend with V1-specific kernel-API shims cv100 blobs (V1, Hi3518 SDK V1.0.B.0, kernel 3.0.8 base) reference a broader set of removed-since-3.0 kernel symbols than cv200 does — surface via modpost when linking against ARMv5 Linux 7.0. Add re-exports for: * __arm_ioremap / __iounmap — pre-3.10 ARM ioremap/iounmap helpers; wrap modern ioremap/iounmap. * create_proc_entry — removed 3.10; wrap proc_create_data(..., NULL). Blob post-call writes to ->proc_fops / ->write_proc become no-ops (struct fields gone) but procfs entry exists; acceptable since cv100 procfs nodes are diagnostic. * __bug — BUG() arch backend; panic() with file:line for clean fail. * do_mmap_pgoff / do_munmap — kernel-internal mmap helpers, unexported in 4.x for security. Wrap vm_mmap / vm_munmap which carry security hooks; SELinux is off in OpenIPC's kernel config so semantics match for the cv100 mmz blob's MMZ-region user-mapping path. * __copy_to_user / __copy_from_user — renamed to arm_copy_{to,from}_user on modern ARM. Signature-compatible alias. * osal_v1_shim: rename static-inline __copy_*_user via Kbuild -D Modern <linux/uaccess.h> defines __copy_to_user / __copy_from_user as static-inline wrappers around raw_copy_*_user — but the cv100 .o blobs expect them as exported symbols. Use the same Kbuild macro-rename trick the shim already uses for _cond_resched: pass -D__copy_to_user=...unused_inline so the header inline is emitted under a throwaway name, then #undef inside the shim TU and provide our own function under the legacy symbol name. * osal_v1_shim: shim msecs_to_jiffies / __get_free_pages / module_put / __udelay Round 2 of cv100_neo modpost unresolved symbols (after the first batch covering __copy_*_user, __arm_ioremap, __iounmap, create_proc_entry, __bug, do_mmap_pgoff, do_munmap): * msecs_to_jiffies — static inline; alias to __msecs_to_jiffies. * __get_free_pages — was a function pre-6.x; alias to get_free_pages_noprof. * module_put — static inline with no exported helper for the put path; no-op stub since cv100 blob modules don't unload at runtime under QEMU. * __udelay — now a macro dispatching arm_delay_ops.udelay; wrap the function-pointer call with the same 0x10C7 pre-scale the macro applies. Two more Kbuild -D macro renames (msecs_to_jiffies, module_put) to suppress the static-inline definitions in their respective headers. * kernel/init/hi3516cv100: re-export blob __ksymtab entries via DECLARE_BLOB_FUNC The cv100 .o blobs (V1, Hi3518 SDK V1.0.B.0) carry __ksymtab_* entries in the legacy 8-byte struct kernel_symbol layout — modern modpost on Linux 7.0 doesn't recognize that section format, so cross-module symbol references fail at link time with "undefined!" even though the symbol is defined and "exported" inside the blob. Mirror the existing cv200/av100 vou_init.c pattern: re-declare each blob-exported symbol via DECLARE_BLOB_FUNC + EXPORT_SYMBOL in the corresponding init wrapper. modpost regenerates the modern ksymtab metadata from the wrapper's EXPORT_SYMBOL directives, and the legacy __ksymtab section in the blob is ignored. 92 symbols across 9 init wrappers — mmz (18), vou (35), hidmac (14), rc (10), sio + sensor_i2c (5 each), hiuser (3), isp + hifb (1 each). Plus hi_sched_clock added to osal_v1_shim (no blob defines it, it's a vendor wrapper for the kernel's sched_clock()). * osal_v1_shim: shim __const_udelay / dma_*_coherent / no_llseek / rtc_*_to_*_tm / schedule_work Final batch of cv100_neo modpost unresolved kernel symbols: * __const_udelay — companion to __udelay, dispatches via arm_delay_ops.const_udelay. * dma_alloc_coherent / dma_free_coherent — became macros over dma_alloc_attrs / dma_free_attrs with attrs=0. * no_llseek — removed in 6.12. Provide as a function returning -ESPIPE (what the legacy kernel-internal helper did). * rtc_time_to_tm / rtc_tm_to_time — renamed *_time64_* in 5.6; wrap with the cast from unsigned long → time64_t. * schedule_work — static inline wrapper over queue_work_on; provide as a real exported function symbol. Adds two more Kbuild -D renames (schedule_work, no_llseek) to suppress the static-inline definitions in their headers. --------- Co-authored-by: Vasiliy Yakovlev <vixand@openipc.org>
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.
Summary
Three small follow-ups to the V2/V2A shim work in #170 that surface only when you actually exercise the firmware-tarball build path (the openhisilicon repo CI uses
actions/checkoutso it never hits these):Ship stub
.o.cmdfiles for cv200/av100 blobs (47 files). GitHub-archive tarballs strip dotfile siblings; Linux 7.0 modpost callsread_text_file()andexit(1)s on missing.cmd. cv500 already had them committed (31 files); cv200/av100 didn't.Gate
open_pm.kobuild behindDISABLE_PMinkernel/hi3516av100.kbuild. Vendor blob referencesdev_pm_opp_init_cpufreq_table/cpufreq_table_validate_and_show/cpufreq_generic_attr— all removed in mainline. opensdk already passesDISABLE_PM=1by default; matching the gate makes the directive effective. Lite (4.9) unaffected —DISABLE_PM=1already applied there and the previous unconditionalobj-mmade it a no-op anyway.Add
dev_warnto both shims (cv200_shim.c, av100_shim.c). Same shape as thedev_errshim from kernel/osal_v2_shim,osal_v2a_shim: cover the full V2/V2A blob-symbol set #170 — macro in<linux/dev_printk.h>that funnels through_dev_err; variadic wrapper viadev_vprintk_emitatKERN_WARNING.Verification
End-to-end QEMU smoke on both neo targets (login + DHCP + ping):
hi3516cv200_neohi3516av100_neohi3516cv200_litehi3516av100_liteTest plan
Related
Builds on #170 (full V2/V2A blob-symbol set). Companion to a forthcoming
openipc/linuxPR adding hi3516av100 mainline DT + CRG + higmac driver, and anOpenIPC/firmwarePR with the cv200_neo + av100_neo board files + opensdk hash bump.🤖 Generated with Claude Code