[RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[Lucas Tanure]

Hi,

While testing mainline Linux on the Khadas VIM4 (Amlogic A311D2 / T7), I 
am hitting intermittent system instability, storage loss, and memory 
corruption during boot, unless I append clk_ignore_unused to the kernel 
command line.

Because the failure is intermittent, I am running 100-reboot test 
campaigns to bisect the remaining clocks and isolate which clocks are 
critical and cannot be safely disabled.

As testing takes significant time over serial console, I wanted to check 
if Amlogic or the maintainers could shed light on this:

Are there specific T7 clocks or hardware domains that must remain 
critical/protected even without an explicit in-kernel consumer?

Current clocks being disabled that don't kill the board on a test run:

[    0.380313] clk: Disabling unused clocks
[    0.380413] clk: Disabled unused clock: rtc_dualdiv
[    0.380718] clk: Disabled unused clock: rtc_duandiv_in
[    0.381377] clk: Unprepared unused clock: a73_div16
[    0.381972] clk: Unprepared unused clock: cpu_div16
[    0.382605] clk: Unprepared unused clock: f50m
[    0.383174] clk: Unprepared unused clock: fixed_pll_dco
[    0.383786] clk: Unprepared unused clock: hdmi_pll_osc
[    0.384416] clk: Unprepared unused clock: sys1_pll_osc
[    0.385056] clk: Unprepared unused clock: earc_osc
[    0.385652] clk: Unprepared unused clock: pcie_refclk_osc
[    0.386323] clk: Unprepared unused clock: eth_pll_osc
[    0.386961] clk: Unprepared unused clock: pcie_osc
[    0.387548] clk: Unprepared unused clock: mclk_pll_osc
[    0.388187] clk: Unprepared unused clock: usb_pll1_osc
[    0.388826] clk: Unprepared unused clock: usb_pll0_osc
[    0.389465] clk: Unprepared unused clock: tcon_pll_osc
[    0.390105] clk: Unprepared unused clock: top_pll_osc
[    0.390738] clk: Unprepared unused clock: aud_pll_osc
[    0.391361] clk: Unprepared unused clock: ddr_pll_osc

I am not forcing any clock to be disabled; I am forcing all clocks to 
stay on and letting them be disabled if unused one by one.

Any guidance or hints on required platform clocks would be greatly 
appreciated.

Thanks,
Lucas Tanure

Re: [RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[Brian Masney]

Hi Lucas,

On Sun, Aug 23, 2026 at 04:40:42PM +0100, Lucas Tanure wrote:
> While testing mainline Linux on the Khadas VIM4 (Amlogic A311D2 / T7), I am
> hitting intermittent system instability, storage loss, and memory corruption
> during boot, unless I append clk_ignore_unused to the kernel command line.
> 
> Because the failure is intermittent, I am running 100-reboot test campaigns
> to bisect the remaining clocks and isolate which clocks are critical and
> cannot be safely disabled.
> 
> As testing takes significant time over serial console, I wanted to check if
> Amlogic or the maintainers could shed light on this:
> 
> Are there specific T7 clocks or hardware domains that must remain
> critical/protected even without an explicit in-kernel consumer?
> 
> Current clocks being disabled that don't kill the board on a test run:
> 
> [    0.380313] clk: Disabling unused clocks
> [    0.380413] clk: Disabled unused clock: rtc_dualdiv
> [    0.380718] clk: Disabled unused clock: rtc_duandiv_in
> [    0.381377] clk: Unprepared unused clock: a73_div16
> [    0.381972] clk: Unprepared unused clock: cpu_div16
> [    0.382605] clk: Unprepared unused clock: f50m
> [    0.383174] clk: Unprepared unused clock: fixed_pll_dco
> [    0.383786] clk: Unprepared unused clock: hdmi_pll_osc
> [    0.384416] clk: Unprepared unused clock: sys1_pll_osc
> [    0.385056] clk: Unprepared unused clock: earc_osc
> [    0.385652] clk: Unprepared unused clock: pcie_refclk_osc
> [    0.386323] clk: Unprepared unused clock: eth_pll_osc
> [    0.386961] clk: Unprepared unused clock: pcie_osc
> [    0.387548] clk: Unprepared unused clock: mclk_pll_osc
> [    0.388187] clk: Unprepared unused clock: usb_pll1_osc
> [    0.388826] clk: Unprepared unused clock: usb_pll0_osc
> [    0.389465] clk: Unprepared unused clock: tcon_pll_osc
> [    0.390105] clk: Unprepared unused clock: top_pll_osc
> [    0.390738] clk: Unprepared unused clock: aud_pll_osc
> [    0.391361] clk: Unprepared unused clock: ddr_pll_osc
> 
> I am not forcing any clock to be disabled; I am forcing all clocks to stay
> on and letting them be disabled if unused one by one.
> 
> Any guidance or hints on required platform clocks would be greatly
> appreciated.

I have an outstanding series that disables unused clocks via the
sync_state callback:

https://lore.kernel.org/linux-clk/20260626-clk-sync-state-v1-0-4156d8196dc8@redhat.com/

Be sure to not have clk_ignore_unused in your kernel command line.

I suggest comparing the list of which clocks are disabled with that
series against your list of clocks above.

Brian

Re: [RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[Neil Armstrong]

On 8/23/26 17:40, Lucas Tanure wrote:
> Hi,
> 
> While testing mainline Linux on the Khadas VIM4 (Amlogic A311D2 / T7), I am hitting intermittent system instability, storage loss, and memory corruption during boot, unless I append clk_ignore_unused to the kernel command line.
> 
> Because the failure is intermittent, I am running 100-reboot test campaigns to bisect the remaining clocks and isolate which clocks are critical and cannot be safely disabled.
> 
> As testing takes significant time over serial console, I wanted to check if Amlogic or the maintainers could shed light on this:
> 
> Are there specific T7 clocks or hardware domains that must remain critical/protected even without an explicit in-kernel consumer?
> 
> Current clocks being disabled that don't kill the board on a test run:
> 
> [    0.380313] clk: Disabling unused clocks
> [    0.380413] clk: Disabled unused clock: rtc_dualdiv
> [    0.380718] clk: Disabled unused clock: rtc_duandiv_in

Those could be used by low power stuff aswell, not only RTC

> [    0.381377] clk: Unprepared unused clock: a73_div16
> [    0.381972] clk: Unprepared unused clock: cpu_div16

Those should probably be left enabled if they're used by the DVFS code

> [    0.382605] clk: Unprepared unused clock: f50m
> [    0.383174] clk: Unprepared unused clock: fixed_pll_dco
> [    0.383786] clk: Unprepared unused clock: hdmi_pll_osc
> [    0.384416] clk: Unprepared unused clock: sys1_pll_osc
> [    0.385056] clk: Unprepared unused clock: earc_osc
> [    0.385652] clk: Unprepared unused clock: pcie_refclk_osc
> [    0.386323] clk: Unprepared unused clock: eth_pll_osc
> [    0.386961] clk: Unprepared unused clock: pcie_osc
> [    0.387548] clk: Unprepared unused clock: mclk_pll_osc
> [    0.388187] clk: Unprepared unused clock: usb_pll1_osc
> [    0.388826] clk: Unprepared unused clock: usb_pll0_osc
> [    0.389465] clk: Unprepared unused clock: tcon_pll_osc
> [    0.390105] clk: Unprepared unused clock: top_pll_osc
> [    0.390738] clk: Unprepared unused clock: aud_pll_osc
> [    0.391361] clk: Unprepared unused clock: ddr_pll_osc
> 
> I am not forcing any clock to be disabled; I am forcing all clocks to stay on and letting them be disabled if unused one by one.
> 
> Any guidance or hints on required platform clocks would be greatly appreciated.

It's hard to say, this is dependent on each family, Amlogic should be able to help here.

You list some clocks, but I don't understand you say that it's stable only if you have clk_ignore_unused, so how did you get this clock list ?

Neil

> 
> Thanks,
> Lucas Tanure

Re: [RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[Lucas Tanure]

On 24/08/2026 16:29, Brian Masney wrote:
> Hi Lucas,
> 
> On Sun, Aug 23, 2026 at 04:40:42PM +0100, Lucas Tanure wrote:
>> While testing mainline Linux on the Khadas VIM4 (Amlogic A311D2 / T7), I am
>> hitting intermittent system instability, storage loss, and memory corruption
>> during boot, unless I append clk_ignore_unused to the kernel command line.
>>
>> Because the failure is intermittent, I am running 100-reboot test campaigns
>> to bisect the remaining clocks and isolate which clocks are critical and
>> cannot be safely disabled.
>>
>> As testing takes significant time over serial console, I wanted to check if
>> Amlogic or the maintainers could shed light on this:
>>
>> Are there specific T7 clocks or hardware domains that must remain
>> critical/protected even without an explicit in-kernel consumer?
>>
>> Current clocks being disabled that don't kill the board on a test run:
>>
>> [    0.380313] clk: Disabling unused clocks
>> [    0.380413] clk: Disabled unused clock: rtc_dualdiv
>> [    0.380718] clk: Disabled unused clock: rtc_duandiv_in
>> [    0.381377] clk: Unprepared unused clock: a73_div16
>> [    0.381972] clk: Unprepared unused clock: cpu_div16
>> [    0.382605] clk: Unprepared unused clock: f50m
>> [    0.383174] clk: Unprepared unused clock: fixed_pll_dco
>> [    0.383786] clk: Unprepared unused clock: hdmi_pll_osc
>> [    0.384416] clk: Unprepared unused clock: sys1_pll_osc
>> [    0.385056] clk: Unprepared unused clock: earc_osc
>> [    0.385652] clk: Unprepared unused clock: pcie_refclk_osc
>> [    0.386323] clk: Unprepared unused clock: eth_pll_osc
>> [    0.386961] clk: Unprepared unused clock: pcie_osc
>> [    0.387548] clk: Unprepared unused clock: mclk_pll_osc
>> [    0.388187] clk: Unprepared unused clock: usb_pll1_osc
>> [    0.388826] clk: Unprepared unused clock: usb_pll0_osc
>> [    0.389465] clk: Unprepared unused clock: tcon_pll_osc
>> [    0.390105] clk: Unprepared unused clock: top_pll_osc
>> [    0.390738] clk: Unprepared unused clock: aud_pll_osc
>> [    0.391361] clk: Unprepared unused clock: ddr_pll_osc
>>
>> I am not forcing any clock to be disabled; I am forcing all clocks to stay
>> on and letting them be disabled if unused one by one.
>>
>> Any guidance or hints on required platform clocks would be greatly
>> appreciated.
> 
> I have an outstanding series that disables unused clocks via the
> sync_state callback:
> 
> https://lore.kernel.org/linux-clk/20260626-clk-sync-state-v1-0-4156d8196dc8@redhat.com/
> 
> Be sure to not have clk_ignore_unused in your kernel command line.
> 
> I suggest comparing the list of which clocks are disabled with that
> series against your list of clocks above.
Thanks for the input, it did not solve the issue. I am going to produce 
the list with a diff in disabled clocks with and without our change, but
it did not solve the issue.
> 
> Brian
> 

Re: [RFC] clk: meson: t7: Intermittent boot instability and memory corruption on VIM4

[Lucas Tanure]

On 26/08/2026 10:02, Neil Armstrong wrote:
> On 8/23/26 17:40, Lucas Tanure wrote:
>> Hi,
>>
>> While testing mainline Linux on the Khadas VIM4 (Amlogic A311D2 / T7), 
>> I am hitting intermittent system instability, storage loss, and memory 
>> corruption during boot, unless I append clk_ignore_unused to the 
>> kernel command line.
>>
>> Because the failure is intermittent, I am running 100-reboot test 
>> campaigns to bisect the remaining clocks and isolate which clocks are 
>> critical and cannot be safely disabled.
>>
>> As testing takes significant time over serial console, I wanted to 
>> check if Amlogic or the maintainers could shed light on this:
>>
>> Are there specific T7 clocks or hardware domains that must remain 
>> critical/protected even without an explicit in-kernel consumer?
>>
>> Current clocks being disabled that don't kill the board on a test run:
>>
>> [    0.380313] clk: Disabling unused clocks
>> [    0.380413] clk: Disabled unused clock: rtc_dualdiv
>> [    0.380718] clk: Disabled unused clock: rtc_duandiv_in
> 
> Those could be used by low power stuff aswell, not only RTC
> 
>> [    0.381377] clk: Unprepared unused clock: a73_div16
>> [    0.381972] clk: Unprepared unused clock: cpu_div16
> 
> Those should probably be left enabled if they're used by the DVFS code
> 
>> [    0.382605] clk: Unprepared unused clock: f50m
>> [    0.383174] clk: Unprepared unused clock: fixed_pll_dco
>> [    0.383786] clk: Unprepared unused clock: hdmi_pll_osc
>> [    0.384416] clk: Unprepared unused clock: sys1_pll_osc
>> [    0.385056] clk: Unprepared unused clock: earc_osc
>> [    0.385652] clk: Unprepared unused clock: pcie_refclk_osc
>> [    0.386323] clk: Unprepared unused clock: eth_pll_osc
>> [    0.386961] clk: Unprepared unused clock: pcie_osc
>> [    0.387548] clk: Unprepared unused clock: mclk_pll_osc
>> [    0.388187] clk: Unprepared unused clock: usb_pll1_osc
>> [    0.388826] clk: Unprepared unused clock: usb_pll0_osc
>> [    0.389465] clk: Unprepared unused clock: tcon_pll_osc
>> [    0.390105] clk: Unprepared unused clock: top_pll_osc
>> [    0.390738] clk: Unprepared unused clock: aud_pll_osc
>> [    0.391361] clk: Unprepared unused clock: ddr_pll_osc
>>
>> I am not forcing any clock to be disabled; I am forcing all clocks to 
>> stay on and letting them be disabled if unused one by one.
>>
>> Any guidance or hints on required platform clocks would be greatly 
>> appreciated.
> 
> It's hard to say, this is dependent on each family, Amlogic should be 
> able to help here.
Yes, Amlogic, please help.>
> You list some clocks, but I don't understand you say that it's stable 
> only if you have clk_ignore_unused, so how did you get this clock list ?
I got this list when I add logs to the clock framework and run without 
clk_ignore_unused.
> 
> Neil
> 
>>
>> Thanks,
>> Lucas Tanure
>