Skip to contentSkip to conversation

i2c issues in 6.18 on R4, but not r4pro (both mt7988 with i2c-mux on i2c2)

DISCUSSION Frank Wunderlich (linux) 3 participants 10 messages Oct 30, 2025 – Nov 5, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi

sorry for html-mail, my main email-provider (GMX) sent it as html (have enabled text-mail as default in settings) when using web-mailer.

i've noticed i2c-issues (i2c2) on mt7988 BPI-R4 with 6.18. On most bootups the i2c-mux is not detected.
root@bpi-r4-v11:~

# i2cdetect -y  2
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- -- 
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
70: -- -- -- -- -- -- -- --
 
i2c-probe is completed successful (no errors and debugs shown before final "return 0;")
 
sometimes the mux came up with same kernel-binary, and i see also other devices on same i2c bus, but mostly all devices are not shown when i use i2cdetect on the bus.
 
root@bpi-r4-v11:~
# i2cdetect -y  2
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- -- 
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
50: -- UU -- -- -- -- -- UU -- -- -- -- -- -- -- -- 
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
70: UU -- -- -- -- -- -- --                         
root@bpi-r4-v11:~
# [   58.594846] sfp sfp1: failed to read EEPROM: -ENXIO
 
6.17 seems ok (have not seen the issue there yet).
 
i looked through many commits, reverted some for clock (my own, lauras clk_gate,...), pinctrl and the i2c-mux (some reset-related) itself but nothing helped.
 
One strange thing is that i have noticed no issues on bpi-r4pro which uses same SoC, same config and should use same drivers and configuration via dt.
Only diference was the reset for the mux itself in dts as far as i see....so tried to disable it to have it nearly same as on r4pro without any effect.
 
Maybe you have any idea how i can nail it down?
 
root@bpi-r4-v11:~
# dmesg | grep -i 'sfp\|i2c'
[    0.000000] Machine model: Banana Pi BPI-R4 (2x SFP+)
[    1.600387] i2c_dev: i2c /dev entries driver
[    1.605291] /soc/i2c@11003000/rt5190a@64: Fixed dependency cycle(s) with /soc/i2c@11003000/rt5190a@64/regulators/buck1 #i2c0 seems not affected
[   12.685157] platform sfp1: deferred probe pending: (reason unknown)
[   12.691440] platform sfp2: deferred probe pending: (reason unknown)
 
# dmesg | grep -i 'err\|fail\|clk'
... #nothing related to i2c or clk
[    1.623805] pca954x 2-0070: probe failed
....
 
wondered why i2c-clocks are always disabled when i look into the clk_summary (also when i2c was working)

root@bpi-r4-v11:~

# cat /sys/kernel/debug/clk/clk_summary | grep -i -B2 INFRA_I2C_BCK
             infra_usb_sys           0       0        0        125000000   0          0     50000      N               deviceless                
          i2c_sel                    0       1        0        125000000   0          0     50000      N            deviceless                   
             infra_i2c_bck           0       3        0        125000000   0          0     50000      N               11005000.i2c              
root@bpi-r4-v11:~

but looking at the driver the clocks are disabled on probe which looks strange to me

https://elixir.bootlin.com/linux/v6.18-rc1/source/drivers/i2c/busses/i2c-mt65xx.c#L1477

i guess it is some kind of timing issue where clocks are still disabled but not reenabled in mtk_i2c_transfer, but my debug in this function shows that is called and ret=0 after the bulk_enable (and of course flooding console :p ). maybe some speed-calculation-issue?

what i had tried to revert (also my own non-mainline like "convert invalid GATE to MUX"), so they seem not the rootcause:

d51e7cfca3fe 2025-09-25 i2c: mt65xx: convert set_speed function to void Wolfram Sang
b49218365280 2025-09-06 i2c: mediatek: fix potential incorrect use of I2C_MASTER_WRRD Leilk.Liu 614b1c3cbfb0 2025-06-12 i2c: use inclusive callbacks in struct i2c_algorithm Wolfram Sang

c90fa5493f7a 2025-07-31 i2c: mux: pca9541: Use I2C adapter timeout value for arbitration timeout Manikanta Guntupalli 94c296776403 2025-06-03 i2c: muxes: pca954x: Reset if (de)select fails Wojciech Siudy
690de2902dca 2025-06-03 i2c: muxes: pca954x: Use reset controller only Wojciech Siudy

e504d3bdb3d0 2025-09-15 clk: mediatek: clk-gate: Add ops for gates with HW voter Laura Nao 8ceff24a754a 2025-09-15 clk: mediatek: clk-gate: Refactor mtk_clk_register_gate to use mtk_gate struct Laura Nao
bd6f4a91401f 2025-09-02 pinctrl: mediatek: moore: replace struct function_desc with struct pinfunction Bartosz Golaszewski 7a24f1f5b214 2025-09-02 pinctrl: mediatek: mt7988: use PINCTRL_PIN_FUNCTION() Bartosz Golaszewski

regards Frank

#1 lore raw

Hi,

it seems that is caused by these 2 commits:

94c296776403 2025-06-03 i2c: muxes: pca954x: Reset if (de)select fails Wojciech Siudy 690de2902dca 2025-06-03 i2c: muxes: pca954x: Use reset controller only Wojciech Siudy

I'm sure that i had tested with them already reverted, but maybe powercycle was not complete. Git bisect pointed also to "i2c: muxes: pca954x: Use reset controller only".

Added Author in To. Maybe he can tell where the gpio-reset should happen in i2c core (have not found any recent commit which takes the reset-gpios) like stated in the commit

commit 690de2902dca98aec96de004428c020ca902f047

Author: Wojciech Siudy <[email protected]>
Date:   Tue Jun 3 14:47:20 2025 +0200

i2c: muxes: pca954x: Use reset controller only

Fallback to legacy reset-gpios is no more needed, because the
framework
can use reset-gpios properity now as well.

my dts-config:

	pca9545: i2c-mux@70 {
		compatible = "nxp,pca9545";
		reg = <0x70>;
		reset-gpios = <&pio 5 GPIO_ACTIVE_LOW>;

i'm unsure what devices are shown in the i2cdetect output, as there should not be any on same bus, only behind the mux (rtc, eeprom and sfp).

[    1.637174] pca954x 2-0070: supply vdd not found, using dummy regulator
[    1.675701] i2c i2c-2: Added multiplexed i2c bus 3
[    1.680597] i2c i2c-2: Added multiplexed i2c bus 4
[    1.685472] i2c i2c-2: Added multiplexed i2c bus 5
[    1.690344] i2c i2c-2: Added multiplexed i2c bus 6
[    1.695137] pca954x 2-0070: registered 4 multiplexed busses for I2C switch pca9545

looks like devices from channel 0 (rtc@51,eeprom@57) are shown on the main bus too.

root@bpi-r4-v11:~

# i2cdetect -y 2
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- UU -- -- -- -- -- UU -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~
# i2cdetect -y 3
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- UU -- -- -- -- -- UU -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~
# i2cdetect -y 4
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: 50 51 -- -- -- -- -- -- -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~
# i2cdetect -y 5
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~
# i2cdetect -y 6
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~

regards Frank

Am 2025-10-30 11:38, schrieb Frank Wunderlich (linux):

Quoted 92 lines

Hi

sorry for html-mail, my main email-provider (GMX) sent it as html (have enabled text-mail as default in settings) when using web-mailer.

i've noticed i2c-issues (i2c2) on mt7988 BPI-R4 with 6.18. On most bootups the i2c-mux is not detected.
root@bpi-r4-v11:~

# i2cdetect -y  2
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- -- 
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
70: -- -- -- -- -- -- -- --
 
i2c-probe is completed successful (no errors and debugs shown before final "return 0;")
 
sometimes the mux came up with same kernel-binary, and i see also other devices on same i2c bus, but mostly all devices are not shown when i use i2cdetect on the bus.
 
root@bpi-r4-v11:~
# i2cdetect -y  2
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- -- 
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
50: -- UU -- -- -- -- -- UU -- -- -- -- -- -- -- -- 
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 
70: UU -- -- -- -- -- -- --                         
root@bpi-r4-v11:~
# [   58.594846] sfp sfp1: failed to read EEPROM: -ENXIO
 
6.17 seems ok (have not seen the issue there yet).
 
i looked through many commits, reverted some for clock (my own, lauras clk_gate,...), pinctrl and the i2c-mux (some reset-related) itself but nothing helped.
 
One strange thing is that i have noticed no issues on bpi-r4pro which uses same SoC, same config and should use same drivers and configuration via dt.
Only diference was the reset for the mux itself in dts as far as i see....so tried to disable it to have it nearly same as on r4pro without any effect.
 
Maybe you have any idea how i can nail it down?
 
root@bpi-r4-v11:~
# dmesg | grep -i 'sfp\|i2c'
[    0.000000] Machine model: Banana Pi BPI-R4 (2x SFP+)
[    1.600387] i2c_dev: i2c /dev entries driver
[    1.605291] /soc/i2c@11003000/rt5190a@64: Fixed dependency cycle(s) with /soc/i2c@11003000/rt5190a@64/regulators/buck1 #i2c0 seems not affected
[   12.685157] platform sfp1: deferred probe pending: (reason unknown)
[   12.691440] platform sfp2: deferred probe pending: (reason unknown)
 
# dmesg | grep -i 'err\|fail\|clk'
... #nothing related to i2c or clk
[    1.623805] pca954x 2-0070: probe failed
....
 
wondered why i2c-clocks are always disabled when i look into the clk_summary (also when i2c was working)

root@bpi-r4-v11:~

# cat /sys/kernel/debug/clk/clk_summary | grep -i -B2 INFRA_I2C_BCK
             infra_usb_sys           0       0        0        125000000   0          0     50000      N               deviceless                
          i2c_sel                    0       1        0        125000000   0          0     50000      N            deviceless                   
             infra_i2c_bck           0       3        0        125000000   0          0     50000      N               11005000.i2c              
root@bpi-r4-v11:~

but looking at the driver the clocks are disabled on probe which looks strange to me

https://elixir.bootlin.com/linux/v6.18-rc1/source/drivers/i2c/busses/i2c-mt65xx.c#L1477

i guess it is some kind of timing issue where clocks are still disabled but not reenabled in mtk_i2c_transfer, but my debug in this function shows that is called and ret=0 after the bulk_enable (and of course flooding console :p ). maybe some speed-calculation-issue?

what i had tried to revert (also my own non-mainline like "convert invalid GATE to MUX"), so they seem not the rootcause:

d51e7cfca3fe 2025-09-25 i2c: mt65xx: convert set_speed function to void Wolfram Sang
b49218365280 2025-09-06 i2c: mediatek: fix potential incorrect use of I2C_MASTER_WRRD Leilk.Liu 614b1c3cbfb0 2025-06-12 i2c: use inclusive callbacks in struct i2c_algorithm Wolfram Sang

c90fa5493f7a 2025-07-31 i2c: mux: pca9541: Use I2C adapter timeout value for arbitration timeout Manikanta Guntupalli 94c296776403 2025-06-03 i2c: muxes: pca954x: Reset if (de)select fails Wojciech Siudy
690de2902dca 2025-06-03 i2c: muxes: pca954x: Use reset controller only Wojciech Siudy

e504d3bdb3d0 2025-09-15 clk: mediatek: clk-gate: Add ops for gates with HW voter Laura Nao 8ceff24a754a 2025-09-15 clk: mediatek: clk-gate: Refactor mtk_clk_register_gate to use mtk_gate struct Laura Nao
bd6f4a91401f 2025-09-02 pinctrl: mediatek: moore: replace struct function_desc with struct pinfunction Bartosz Golaszewski 7a24f1f5b214 2025-09-02 pinctrl: mediatek: mt7988: use PINCTRL_PIN_FUNCTION() Bartosz Golaszewski

regards Frank

#2 lore raw
Quoted 5 lines

I'm sure that i had tested with them already reverted, but maybe powercycle was not complete.
Git bisect pointed also to "i2c: muxes: pca954x: Use reset controller only".

Can you try enabling CONFIG_RESET_GPIO please? It is probably that this driver is missing. The core relies on that driver for the fallback the above patch makes use of. Sadly, I couldn't convince the reset maintainers of a way to automatically select the driver for existing users :(

#3 lore raw

Am 2025-11-02 13:45, schrieb Wolfram Sang:

Quoted 5 lines

Can you try enabling CONFIG_RESET_GPIO please? It is probably that this driver is missing. The core relies on that driver for the fallback the above patch makes use of. Sadly, I couldn't convince the reset maintainers of a way to automatically select the driver for existing users :(

Thanks for pointing to this and indeed i had not enabled this. This is my fault and not from reset maintainers as i have my own defconfig. Tested and my mux works now again.

Maybe this option should be enabled by default or when the i2c-mux is selected?

When moving from specific to a generic aproach (which makes sense) imho it should be mentioned in commit description which option needed instead :)

btw. the Authors Email seems not working, got

<[email protected]>: host
nokia-com.mail.protection.outlook.com[52.101.73.24] said: 550 5.4.1 Recipient address rejected: Access denied.

or maybe its inbox has a whitelist from which domains he can receive mails and my domain is not added to it.

regards Frank

#4 lore raw
Quoted 3 lines

from reset maintainers as i have my own defconfig. Tested and my mux works now again.

Glad it works again!

Quoted 2 lines

Maybe this option should be enabled by default or when the i2c-mux is selected?

That's what we were discussing. At least a warning will now get printed to the logs when booting suggesting the above solution.

Quoted 3 lines

When moving from specific to a generic aproach (which makes sense) imho it should be
mentioned in commit description which option needed instead :)

We didn't realize this back then. Only when the first bug report came in :/

Quoted 1 line

btw. the Authors Email seems not working, got

Yes, he quit Nokia meanwhile.

#5 lore raw
Quoted 6 lines
Quoted 3 lines

Maybe this option should be enabled by default or when the i2c-mux is selected?

That's what we were discussing. At least a warning will now get printed to the logs when booting suggesting the above solution.

FYI, all the commits leading to this breakage are reverted now. Sorry for your troubles!

#6 lore raw

Il 05/11/25 12:51, Wolfram Sang ha scritto:

Quoted 10 lines
Quoted 6 lines
Quoted 3 lines

Maybe this option should be enabled by default or when the i2c-mux is selected?

That's what we were discussing. At least a warning will now get printed to the logs when booting suggesting the above solution.

FYI, all the commits leading to this breakage are reverted now. Sorry for your troubles!

Wow I replied to a very old email in the topic; Good (but not great) to know that the issue was not in MediaTek specific drivers and nice to see that it's all been resolved.

Cheers,
Angelo

#7 lore raw
In reply to Wolfram Sang

Am 5. November 2025 12:51:23 MEZ schrieb Wolfram Sang <[email protected]>:

Quoted 9 lines
Quoted 6 lines
Quoted 3 lines

Maybe this option should be enabled by default or when the i2c-mux is selected?

That's what we were discussing. At least a warning will now get printed to the logs when booting suggesting the above solution.

FYI, all the commits leading to this breakage are reverted now. Sorry for your troubles!

Why? My issue was solved by adding GPIO_RESET... So it was fault on my side,sorry for bothering.

@angelo
yes it is a final board BPI-R4 v1.1.

Why sometimes it works i guess it is because uboot may previously initialize the device (and without reset working it may be stuck in wrong state).
I'm not sure if my current flashed version had the i2c and i2cmux support,but i got it working on that board because of the connected eeprom (read mac address via tlv_eeprom as macs do not have fixed macs via efuse afaik).

Thanks for help so far.

regards Frank

#8 lore raw

Il 05/11/25 13:08, Frank Wunderlich ha scritto:

Quoted 11 lines

Am 5. November 2025 12:51:23 MEZ schrieb Wolfram Sang <[email protected]>:

Quoted 9 lines
Quoted 6 lines
Quoted 3 lines

Maybe this option should be enabled by default or when the i2c-mux is selected?

That's what we were discussing. At least a warning will now get printed to the logs when booting suggesting the above solution.

FYI, all the commits leading to this breakage are reverted now. Sorry for your troubles!

Quoted 6 lines

Why? My issue was solved by adding GPIO_RESET... So it was fault on my side,sorry for bothering.

@angelo
yes it is a final board BPI-R4 v1.1.

Why sometimes it works i guess it is because uboot may previously initialize the device (and without reset working it may be stuck in wrong state).

Heh, this is something that happens frequently.

Quoted 4 lines

I'm not sure if my current flashed version had the i2c and i2cmux support,but i got it working on that board because of the connected eeprom (read mac address via tlv_eeprom as macs do not have fixed macs via efuse afaik).

Thanks for help so far.

No worries, my pleasure (but I didn't do all that much anyway :P :P :P)

Cheers,
Angelo

Quoted 1 line

regards Frank

#9 lore raw
In reply to Frank Wunderlich (linux)

Il 30/10/25 11:38, Frank Wunderlich (linux) ha scritto:

Quoted 36 lines

Hi

sorry for html-mail, my main email-provider (GMX) sent it as html (have enabled text-mail as default in settings) when using web-mailer.

i've noticed i2c-issues (i2c2) on mt7988 BPI-R4 with 6.18. On most bootups the i2c- mux is not detected.

root@bpi-r4-v11:~

# i2cdetect -y  2
      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: -- -- -- -- -- -- -- --

i2c-probe is completed successful (no errors and debugs shown before final "return 0;")

sometimes the mux came up with same kernel-binary, and i see also other devices on same i2c bus, but mostly all devices are not shown when i use i2cdetect on the bus.

root@bpi-r4-v11:~

# i2cdetect -y  2
      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f
00:                         -- -- -- -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
50: -- UU -- -- -- -- -- UU -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: UU -- -- -- -- -- -- --
root@bpi-r4-v11:~
# [   58.594846] sfp sfp1: failed to read EEPROM: -ENXIO
Quoted 9 lines

6.17 seems ok (have not seen the issue there yet).

i looked through many commits, reverted some for clock (my own, lauras clk_gate,...), pinctrl and the i2c-mux (some reset-related) itself but nothing helped.

One strange thing is that i have noticed no issues on bpi-r4pro which uses same SoC, same config and should use same drivers and configuration via dt. Only diference was the reset for the mux itself in dts as far as i see....so tried to disable it to have it nearly same as on r4pro without any effect.

Maybe you have any idea how i can nail it down?

Did you check if there's any clock, or power rail, or power domain, that you are missing in some node perhaps?

Could be that something is getting powered off temporarily and gifting you with this "amazing" issue...

Quoted 8 lines

root@bpi-r4-v11:~

# dmesg | grep -i 'sfp\|i2c'
[    0.000000] Machine model: Banana Pi BPI-R4 (2x SFP+)
[    1.600387] i2c_dev: i2c /dev entries driver
[    1.605291] /soc/i2c@11003000/rt5190a@64: Fixed dependency cycle(s) with /soc/ i2c@11003000/rt5190a@64/regulators/buck1 #i2c0 seems not affected
[   12.685157] platform sfp1: deferred probe pending: (reason unknown)
[   12.691440] platform sfp2: deferred probe pending: (reason unknown)
Quoted 7 lines
# dmesg | grep -i 'err\|fail\|clk'
... #nothing related to i2c or clk
[    1.623805] pca954x 2-0070: probe failed
....

wondered why i2c-clocks are always disabled when i look into the clk_summary (also when i2c was working)

That is something for which I can give you a clear answer....

Quoted 10 lines

root@bpi-r4-v11:~

# cat /sys/kernel/debug/clk/clk_summary | grep -i -B2 INFRA_I2C_BCK
              infra_usb_sys           0       0        0        125000000   0       
    0     50000      N               deviceless
           i2c_sel                    0       1        0        125000000   0       
    0     50000      N            deviceless
              infra_i2c_bck           0       3        0        125000000   0       
    0     50000      N               11005000.i2c
root@bpi-r4-v11:~
Quoted 3 lines

but looking at the driver the clocks are disabled on probe which looks strange to me

https://elixir.bootlin.com/linux/v6.18-rc1/source/drivers/i2c/busses/i2c- mt65xx.c#L1477

...Not strange at all...

Quoted 3 lines

i guess it is some kind of timing issue where clocks are still disabled but not reenabled in mtk_i2c_transfer, but my debug in this function shows that is called and ret=0 after the bulk_enable (and of course flooding console :p ).

...because on MediaTek, the I2C controllers are "kinda standalone" - meaning that they work as long as they're powered up, with their main clock being either some internal XO or the clk26m or a clock derived from that (it's bogus, as that is an IP implementation detail that nobody has, if not MediaTek's HW engineering dept, and it's not relevant anyway so we don't even need to really know if not for pure curiosity).

The basic "unclocked" functionality of the I2C controller(s) in MediaTek SoCs is XFER detection (not sure if they can do buffering as well while the AP-driven clocks are off).

As a small side note: I think MediaTek are not the only ones doing that, but this is an entirely different story and it's just for your own curiosity, in case.

The I2C clocks therefore are only enabling the CPU to perform read/write(/modify) operations on the I2C Controller registers, which also include DMA transfers.

Speaking of DMA, is there really no support for apdma_sync in MT7988a? ;-)

Quoted 2 lines

maybe some speed-calculation-issue?

Could also be, but if it was that, and directly related to I2C, you'd get errors during communication even when your PCA954X device actually gets probed, because whatever calculation is ran, it's always going to be the same across reboots.

All that I'm saying here, though, ignores hardware issues. So, if you're using a board that is not final, or a final board that has HW issues (not only as in your own unit, but as a design mistake of some sort) it's a different story.

I am bringing up the HW question just because "better safe than sorry", but if using something not final.. I'm mostly sure that you would've said that already and tried with.. well, final hardware, before seeking for help here anyway :-)

Cheers!
Angelo

Quoted 17 lines

what i had tried to revert (also my own non-mainline like "convert invalid GATE to MUX"), so they seem not the rootcause:

d51e7cfca3fe 2025-09-25 i2c: mt65xx: convert set_speed function to void Wolfram Sang b49218365280 2025-09-06 i2c: mediatek: fix potential incorrect use of I2C_MASTER_WRRD Leilk.Liu 614b1c3cbfb0 2025-06-12 i2c: use inclusive callbacks in struct i2c_algorithm Wolfram Sang

c90fa5493f7a 2025-07-31 i2c: mux: pca9541: Use I2C adapter timeout value for arbitration timeout Manikanta Guntupalli 94c296776403 2025-06-03 i2c: muxes: pca954x: Reset if (de)select fails Wojciech Siudy
690de2902dca 2025-06-03 i2c: muxes: pca954x: Use reset controller only Wojciech Siudy

e504d3bdb3d0 2025-09-15 clk: mediatek: clk-gate: Add ops for gates with HW voter Laura Nao 8ceff24a754a 2025-09-15 clk: mediatek: clk-gate: Refactor mtk_clk_register_gate to use mtk_gate struct Laura Nao

bd6f4a91401f 2025-09-02 pinctrl: mediatek: moore: replace struct function_desc with struct pinfunction Bartosz Golaszewski 7a24f1f5b214 2025-09-02 pinctrl: mediatek: mt7988: use PINCTRL_PIN_FUNCTION() Bartosz Golaszewski

regards Frank