From 45705b3833327889e59a66903cfe6d4be7964705 Mon Sep 17 00:00:00 2001
From: Dmitry Ilyin <6576495+widgetii@users.noreply.github.com>
Date: Sat, 17 Oct 2026 12:41:48 +0300
Subject: [PATCH] hisi: do not probe for a sensor on the Hi3536DV100

The Hi3536DV100 is a decoder with no video input, so a report never
finds a sensor there, but the probe was expensive. On a lab NVR i2c0 has
nothing on the wire and the kernel's hibvt-i2c driver runs it in polling
mode (the device tree gives it no interrupt), so each transfer spins
udelay(50) 0x10000 times -- 3.3 s on the SoC's single core -- before
"wait idle timeout". The table is 38 transfers: the sensors section took
126 s, and the board stopped answering ping for most of it.

Leave possible_i2c_addrs empty on this chip. The whole report now takes
1.0 s on that NVR and issues no i2c transfer.
---
 src/hal/hisi/hal_hisi.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/src/hal/hisi/hal_hisi.c b/src/hal/hisi/hal_hisi.c
index 70fe8701..ac27a384 100644
--- a/src/hal/hisi/hal_hisi.c
+++ b/src/hal/hisi/hal_hisi.c
@@ -678,6 +678,17 @@ void setup_hal_hisi() {
         break;
     }
     possible_i2c_addrs = my_possible_i2c_addrs;
+#ifdef IPCHW_HISI_3536D
+    /* The Hi3536DV100 is a decoder with no video input, so there is no sensor
+     * to find, and probing for one is not free. Its i2c0 has nothing on the
+     * wire on an NVR, and the kernel's hibvt-i2c runs it in polling mode (the
+     * device tree gives it no interrupt): every transfer spins udelay(50)
+     * 0x10000 times, 3.3 s on the SoC's only core, before "wait idle timeout".
+     * The full table is 38 such transfers -- the board stopped answering ping
+     * for two minutes on every report. */
+    if (chip_generation == HISI_3536D)
+        possible_i2c_addrs = NULL;
+#endif
     hal_temperature = hisi_get_temp;
 #ifndef STANDALONE_LIBRARY
     hal_detect_ethernet = hisi_ethdetect;