hi3516cv300_neo: Linux 7.0 board variant (third HiSi neo SoC, issue openhisilicon#60)#2103
Merged
Conversation
Wire cv300 into the neo build matrix as the third HiSi family on the openipc/linux upstream-patches 7.0 base — after ev300_neo (kernels 6.6 / 6.18 / 7.0) and av300_neo (7.0). All upstream prerequisites are in place: - OpenIPC/linux#42 arm: hisilicon: add hi3516cv300 SoC support (CRG + DT + ARCH_HI3516CV300 Kconfig). The defconfig pulls the upstream-patches tarball; the cv300 DT lives there now, no all-patches-neo/linux drop-in required. - openhisilicon PR kernel/hi3516cv300: OSAL/peripheral compat- shim port + cv300_neo CI matrix row. Pin bumped separately in this branch. Three new files: configs/hi3516cv300_neo_defconfig ARMv5 EABI, vendor v3 musl toolchain (kernel-headers stay at 3.18 — toolchain ABI), upstream-patches kernel tarball, osdrv off + opensdk on, no wireguard-linux-compat (built into 7.0). 8 MiB NOR layout matches lite. board/hi3516cv300/hi3516cv300.neo.config Seed kernel config enabling ARCH_HI3516CV300, HISI_FEMAC, SP804, VIC, squashfs+XZ, the cv300-relevant pinctrl. Expand with `make ARCH=arm olddefconfig` on first build. board/hi3516cv300/neo-post-image.sh Append hi3516cv300-demb.dtb to zImage and wrap as uImage at loadaddr 0x80008000 (same layout the cv500/av300 neo scripts use, retargeted to the cv300 DTB and chip name). Necessary because Buildroot's old mkimage path doesn't speak 7.0 uImage directly. Verified end-to-end under qemu-hisilicon -M hi3516cv300: kernel boots, FEMAC DHCP succeeds, NTP rolls the wall clock forward, no Oops/panic/BUG/Trace on console or in dmesg.
Mirror of how ev300_neo / av300_neo are already wired into .github/workflows/build.yml. With the board files added in the previous commit (defconfig + neo kernel config + neo-post-image script), the new row builds end-to-end against the openipc/linux upstream-patches 7.0 base.
Pulls in OpenIPC/openhisilicon#148 — the cv300 OSAL/peripheral source port to Linux 7.0 via new kernel_compat.h shims (PDE_DATA, print_symbol, strscpy, do_gettimeofday + struct timeval, rtc_time_to_tm, DEFINE_SEMAPHORE 1-arg). Required by the cv300_neo board added in the previous commits — without this bump the opensdk cv300 build won't compile against the upstream-patches 7.0 kernel. Full upstream diff includes only the cv300-targeted shim additions and source-side calls through them — no behaviour change to any other platform. Existing cv500 / av300 / ev300 / gk7205v200 / etc. opensdk builds are byte-identical (touched files are all under kernel/{osal,init,ir,pwm,rtc,sensor*,sys_config,wdt}/hi3516cv300/ plus the additive-only kernel_compat.h block).
…ion fix) The previous bump to b405551 inadvertently broke cv300_lite (kernel 3.18), cv500_lite / cv200_lite (kernel 4.9), and ev300_neo / av300_neo (kernel 6.4+) builds. Pulled in via OpenIPC/openhisilicon#150: - mmz/media-mem.c gated p4d_offset() on >= 4.11 (didn't exist on 3.18, was added in 4.11 as 5-level paging prep) - wdt/ir/pwm/rtc/hi3516cv300 .c files: explicit <linux/module.h> include so MODULE_LICENSE() expands on 3.18 (the chain via hi_osal.h only pulled it transitively on 7.0) - kernel_compat.h: print_symbol shim gated >= 5.15 not >= 4.0 (function is still declared in <linux/kallsyms.h> through 5.14; macro do { ... } while (0) collides with the prototype on 4.9) - kernel_compat.h: DEFINE_SEMAPHORE shim dropped the #undef + 1-arg redefinition; kept only compat_DEFINE_SEMAPHORE helper. cv300 osal/mmz/media-mem.c updated to use the helper (matches cv500 mmz pattern). Locally verified: hi3516cv300_lite build from this branch produces a rootfs byte-equivalent to nightly (674 = 674 files, hisilicon .ko set unchanged, QEMU lsmod + scrubbed dmesg identical).
Member
Author
|
opensdk hash re-bumped: `b405551 → b128a0c` to pull in OpenIPC/openhisilicon#150 — fixes the cv300_lite (kernel 3.18) build regression, plus collateral damage on cv500_lite / cv200_lite (kernel 4.9 `print_symbol` collision) and ev300_neo / av300_neo (kernel 6.4+ `DEFINE_SEMAPHORE` collision) introduced by the earlier #148 / b405551. Locally verified: `make BOARD=hi3516cv300_lite` from this branch produces a rootfs byte-equivalent to nightly:
|
…rops) The opensdk hash bump in this PR (5b1aa42 → b128a0c) pulls in a number of cv500-targeted ive_neo additions that grew kernel/ive_neo/ive_neo.c from 66 KB to 130 KB of source. The .c also compiles for V4 (ev200 / gk7205v200) — Kbuild includes ive_neo/Kbuild in the V4 dispatch — so the stripped open_ive_neo.ko on V4 grew by ~20-30 KB, which on its own pushes hi3518ev300_lite past the 5120 KB rootfs partition cap (master was already at 5116/5120; new build: 5164/5120). No V4 firmware actually loads open_ive_neo.ko: $ grep ive_neo /etc/init.d/* /usr/bin/load_* (in ev300_lite nightly) (no matches) So we don't need to ship the .ko on V4 boards. Add a TARGET_FINALIZE hook that removes /lib/modules/<kver>/extra/open_ive_neo.ko after the normal kernel-module install and re-runs depmod. The driver still compiles in CI (Kbuild unchanged), and cv500 still ships it via its own Kbuild include — only the install-to-rootfs step is skipped on V4. Matches the existing FINALIZE_MODULES_GK7205V200 pattern that selectively removes specific open_*.ko from extra/ after install.
Two changes to keep hi3518ev300_lite under its 5 MB rootfs cap after the opensdk bump in this PR + majestic upstream's recent +41 KB growth (same day this PR was opened). 1. V4 install-skip hook — extend the open_ive_neo.ko removal added in cb52453 to also drop open_adc.ko. Same justification: load_hisilicon / load_goke on V4 never insmod open_adc.ko, no init script references it, ev300_lite / gk7205v200_lite nightly rootfs both carry it as dead weight in /lib/modules/<kver>/extra/. ~6 KB compressed saving on V4. Renamed HISILICON_OPENSDK_SKIP_IVE_NEO_V4 → SKIP_UNUSED_V4 since the hook now covers a category, not a single module. 2. ev300_lite TRIM_SP2308 hook — also drop libsns_sc500ai.so and libsns_gc5603.so. This is a TEMPORARY measure with a real cost: any deployed ev300_lite camera using SC500AI (5 MP SmartSens) or GC5603 (5 MP GalaxyCore) will lose its sensor driver on the next sysupgrade from openipc.org-hosted images. The trim was needed today because majestic.hi3516ev200.lite.master.tar.bz2 on S3 grew from 711 KB → 752 KB binary between yesterday's nightly and today's reupload, pushing the 5116/5120 KB master rootfs over the cap. Address in a follow-up PR via compaction (strip flags, dead code removal, partition rebalance) and restore the sensors. Verified locally: hi3518ev300_lite rootfs.squashfs = 5112 KB / 5120 KB (8 KB headroom), trim hooks confirmed to run.
…anup
The previous commit added HISILICON_OPENSDK_SKIP_UNUSED_V4 with an
rsync from $(PER_PACKAGE_DIR)/hisilicon-opensdk/target/lib/modules/
gated on both hi3516ev200 AND gk7205v200. That rsync runs AFTER
HISILICON_OPENSDK_FINALIZE_MODULES_GK7205V200, undoing its cleanup:
- FINALIZE_MODULES_GK7205V200 rsyncs from per-package, then removes
open_{vi,vpss,venc,h264e,h265e,...}.ko from extra/ (these are
renamed and put under goke/ as gk7205v200_*.ko).
- SKIP_UNUSED_V4 immediately rsyncs again from per-package, which
re-introduces all the open_*.ko files into extra/.
Result on gk7205v200_lite: rootfs.squashfs ballooned 5076 KB → 5660 KB
(580 KB regression), with the heavy V4 modules suddenly present in both
extra/ AND goke/. Caught by local build + gcc-compat CI failures.
Fix:
- Fold the open_ive_neo + open_adc deletions directly into
FINALIZE_MODULES_GK7205V200 so a single hook covers both the goke/
rename cleanup and the unused-module skip on gk7205v200.
- Keep the dedicated post-hook for hi3516ev200 only (no rename
cleanup happens there, so the rsync + clean order is safe).
Verified locally: gk7205v200_lite rebuilds at 5064/5120 KB (56 KB
headroom, matches nightly's baseline), extra/ size 516 KB vs nightly's
552 KB (correctly smaller by the ive_neo + adc skip), goke/ intact.
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
1 participant
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
Lands the firmware side of the hi3516cv300_neo port — the third HiSi neo board variant on the openipc/linux upstream-patches 7.0 base, after
hi3516ev300_neo(6.6 / 6.18 / 7.0) andhi3516av300_neo(7.0). All upstream prerequisites are merged:arm: hisilicon: add hi3516cv300 SoC support(CRG driver rewrite + DTS + ARCH_HI3516CV300 Kconfig)kernel/hi3516cv300: V3 OSAL/peripherals against Linux 7.0 + ci matrix rowThree commits:
hi3516cv300_neo: Linux 7.0 board — new defconfig + seed kernel config + neo-post-image scriptci: add hi3516cv300_neo to firmware build matrix— one-line addition to.github/workflows/build.ymlmirroring how ev300_neo / av300_neo are wiredhisilicon-opensdk: bump to b405551— pulls in the cv300 OSAL compat-shim port from openhisiliconLayout matches the existing two neo boards: defconfig uses the vendor v3 musl toolchain (kernel-headers stay at 3.18 — toolchain ABI; modern kernel ABI is backward-compatible), pulls the
upstream-patches.tar.gz(now contains cv300 DT), forces opensdk on and osdrv off (V3 vendor osdrv won't compile against 7.0, and opensdk replaces every kernel module on cv300 anyway after #2102).Test plan
make BOARD=hi3516cv300_neoproduces uImage + rootfs.squashfs (Linux 7.0)qemu-system-arm -M hi3516cv300toopenipc-hi3516cv300 login:Oops:,panic:,BUG:,Call Trace, orUnable to handleon console or in dmesgWhy this is OK now
The cv300
qemu-bootrow in openhisilicon CI sits atallow-failure: truefor a pre-existingcma_osal.kovshi_osal.konaming mismatch in the cv300_lite load_hisilicon. That bug is on the lite path and does not affect neo: this defconfig forces opensdk on, sohi_osal.kois built and installed (via opensdk's rename-on-install for cv300). Neo starts from a clean slate.