hisilicon-opensdk: bump for the SC2235 DVP pad fix - #2494
Conversation
openhisilicon c1f9eb5c, one commit on from the current pin, is OpenIPC/openhisilicon#233: the SC2235 driver repeats its DVP pad setup once the sensor is streaming, with stronger drive on 0x3641, and starts streaming again afterwards. The stock sequence sets those enables before stream start, and on the Imou Cue 2 that leaves the SoC with no pixel clock -- the sensor answers on the bus and no VI interrupts ever arrive. Nothing else moved between the two commits, and the fix is a delta appended to the existing table rather than a replacement, so no other register changes for any sensor. It reaches gk7205v200 as well as hi3516ev200, because libraries/sensor/ gk7205v200 is a symlink to hi3516ev200 and both families build the one file. No board regresses on either: no defconfig here and no builder device selects SC2235 today. The library ships only because it is in the shared V4 sensor name list, and the two builder devices that mention it do so in exclusion lists, to prune it. This is what firmware#2446's sensor ini and IQ profile were waiting for, and what OpenIPC/builder#160 needs before a Cue 2 image is worth publishing.
PR Summary by QodoBump hisilicon-opensdk for the SC2235 DVP pad fix
AI Description
Diagram
High-Level Assessment
Files changed (1)
|
Code Review by Qodo
1. Affected cameras remain unverified
|
Problem
hi3516ev200boards with an SC2235 on the DVP pads get no video. The sensor is on thebus and answers, but no VI interrupts arrive — the stock driver sets the DVP pad enables
before the sensor starts streaming, and on at least one camera that leaves the SoC with
no pixel clock.
Change
Bumps
HISILICON_OPENSDK_VERSIONfrom6eb7736atoc1f9eb5c— exactly one commit,OpenIPC/openhisilicon#233, which
repeats the pad setup once the sensor is streaming (stronger drive on
0x3641) and startsstreaming again afterwards. It is a delta appended after the existing register table, not
a replacement, so no other register changes for any sensor.
This is what #2446's SC2235 sensor ini and IQ profile were waiting for, and what
OpenIPC/builder#160 needs before an Imou
Cue 2 image is worth publishing.
Blast radius
The bump reaches
hi3516ev200andgk7205v200—libraries/sensor/gk7205v200is asymlink to
hi3516ev200, so both families build the one file.Nothing regresses on either, because nothing uses that sensor yet:
br-ext-chip-hisiliconorbr-ext-chip-gokedefconfig namessc2235fw_setenv sensor sc2235libsns_sc2235.soto save flashlibsns_sc2235.sois built and shipped for these families only becausesc2235is in theshared V4 sensor name list. The Cue 2 will be the first OpenIPC device to select it.
Hardware tested on
Not by me, and I want to be plain about that. There is no
hi3516ev200board in the lab —the recorded DNS name still resolves but the address does not answer — so I have no camera
to exercise this on.
The evidence is the contributor's, on #2446 and #233: one Imou Cue 2 (IPC-C22EN), flashed
clean with kernel and rootfs and the overlay erased, giving video, both streams in Frigate,
night mode and two-way audio. The reason I am comfortable bumping on one camera's evidence
is the blast-radius section above — with no other board selecting this sensor, there is no
working configuration for the change to break.
Evidence
The bump is one line and the diff between the two openhisilicon commits is the linked PR:
Selector and self-test:
45 boards build this package, so the matrix covers the families that actually consume it.
Scope
general/package/all-patches/linux/LD_PRELOAD, and no binaries that cannot be rebuilt from source