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

[Lucas Tanure]

On 26/08/2026 18:17, Brian Masney wrote:
> Hi Lucas,
>
> On Wed, Aug 26, 2026 at 05:42:01PM +0100, Lucas Tanure wrote:
> > With Brian's patches, VIM4 does not turn off 4 clocks:
> >
> > t7_sys_sd_emmc_b
> > t7_sys_sd_emmc_c
> > sd_emmc_b
> > sd_emmc_c
> >
> > This is expected since I'm trying to run from an SD card. However, even with
> > these four clocks kept enabled, the board still hangs.
> >
> > Complete list with all clocks being disabled after Brians patches:
> > [    1.002247] clk: Disabled unused clock: t7_sys_pwm_ao_gh
> > [    1.002898] clk: Disabled unused clock: t7_sys_pwm_ao_ef
> > [    1.003560] clk: Disabled unused clock: t7_sys_pwm_ao_cd
> > [    1.004220] clk: Disabled unused clock: t7_sys_pwm_ao_ab
> > [    1.004972] clk: Disabled unused clock: t7_sys_pwm_ef
> > [    1.005509] clk: Disabled unused clock: t7_sys_pwm_cd
> > [    1.006211] clk: Disabled unused clock: t7_sys_pwm_ab
> > [    1.006766] clk: Disabled unused clock: t7_sys_ts_hevc
> > [    1.007434] clk: Disabled unused clock: t7_sys_ts_vpu
> > [    1.008033] clk: Disabled unused clock: t7_sys_ts_nna
> > [    1.008741] clk: Disabled unused clock: t7_sys_ts_gpu
> > [    1.009294] clk: Disabled unused clock: t7_sys_sar_adc
> > [    1.009929] clk: Disabled unused clock: t7_sys_vpu_intr
> > [    1.010609] clk: Disabled unused clock: t7_sys_dspb
> > [    1.011186] clk: Disabled unused clock: t7_sys_dspa
> > [    1.011792] clk: Disabled unused clock: t7_sys_a73pclk_apb
> > [    1.012554] clk: Disabled unused clock: t7_sys_pclk_sys_apb
> > [    1.013213] clk: Disabled unused clock: t7_sys_rsa
> > [    1.013811] clk: Disabled unused clock: t7_sys_mipi_isp_pclk
> > [    1.014529] clk: Disabled unused clock: t7_sys_mmc_apb
> > [    1.015107] clk: Disabled unused clock: t7_sys_hdmirx_pclk
> > [    1.015819] clk: Disabled unused clock: t7_sys_i2c_s_a
> > [    1.016429] clk: Disabled unused clock: t7_sys_hdmitx_apb
> > [    1.017109] clk: Disabled unused clock: t7_sys_i2c_m_f
> > [    1.017740] clk: Disabled unused clock: t7_sys_i2c_m_e
> > [    1.018379] clk: Disabled unused clock: t7_sys_i2c_m_d
> > [    1.019018] clk: Disabled unused clock: t7_sys_i2c_m_c
> > [    1.019657] clk: Disabled unused clock: t7_sys_i2c_m_b
> > [    1.020296] clk: Disabled unused clock: t7_sys_i2c_m_a
> > [    1.020940] clk: Disabled unused clock: t7_sys_i2c_ao_b
> > [    1.021585] clk: Disabled unused clock: t7_sys_i2c_ao_a
> > [    1.022236] clk: Disabled unused clock: t7_sys_pcie_phy
> > [    1.022885] clk: Disabled unused clock: t7_sys_usb
> > [    1.023482] clk: Disabled unused clock: t7_sys_pcie
> > [    1.024088] clk: Disabled unused clock: t7_sys_spicc1
> > [    1.024723] clk: Disabled unused clock: t7_sys_spicc0
> > [    1.025345] clk: Disabled unused clock: t7_sys_g2d
> > [    1.025940] clk: Disabled unused clock: t7_sys_spicc5
> > [    1.026569] clk: Disabled unused clock: t7_sys_ts_a53
> > [    1.027197] clk: Disabled unused clock: t7_sys_ts_a73
> > [    1.027825] clk: Disabled unused clock: t7_sys_spicc4
> > [    1.028454] clk: Disabled unused clock: t7_sys_spicc3
> > [    1.029085] clk: Disabled unused clock: t7_sys_spicc2
> > [    1.029710] clk: Disabled unused clock: t7_sys_aififo
> > [    1.030339] clk: Disabled unused clock: t7_sys_uart_f
> > [    1.030967] clk: Disabled unused clock: t7_sys_uart_e
> > [    1.031595] clk: Disabled unused clock: t7_sys_uart_d
> > [    1.032224] clk: Disabled unused clock: t7_sys_uart_c
> > [    1.032857] clk: Disabled unused clock: t7_sys_uart_b
> > [    1.033482] clk: Disabled unused clock: t7_sys_eth
> > [    1.034077] clk: Disabled unused clock: t7_sys_audio
> > [    1.034694] clk: Disabled unused clock: t7_sys_ir_ctrl
> > [    1.035333] clk: Disabled unused clock: t7_sys_msr_clk
> > [    1.035972] clk: Disabled unused clock: t7_sys_spifc
> > [    1.036596] clk: Disabled unused clock: t7_sys_acodec
> > [    1.037218] clk: Disabled unused clock: t7_sys_smartcard
> > [    1.037880] clk: Disabled unused clock: t7_sys_am2axi2
> > [    1.038518] clk: Disabled unused clock: t7_sys_am2axi1
> > [    1.039157] clk: Disabled unused clock: t7_sys_am2axi0
> > [    1.039796] clk: Disabled unused clock: t7_sys_ampipe_eth
> > [    1.040468] clk: Disabled unused clock: t7_sys_ampipe_nand
> > [    1.041155] clk: Disabled unused clock: t7_sys_deswarp
> > [    1.041790] clk: Disabled unused clock: t7_sys_gdc
> > [    1.042385] clk: Disabled unused clock: t7_sys_cec
> > [    1.042981] clk: Disabled unused clock: t7_sys_aucpu
> > [    1.043600] clk: Disabled unused clock: t7_sys_aocpu
> > [    1.044217] clk: Disabled unused clock: t7_sys_mali
> > [    1.044830] clk: Disabled unused clock: t7_sys_ethphy
> > [    1.045451] clk: Disabled unused clock: t7_sys_mipi_dsi_b
> > [    1.046123] clk: Disabled unused clock: t7_sys_mipi_dsi_a
> > [    1.046795] clk: Disabled unused clock: t7_sys_dos
> > [    1.047390] clk: Disabled unused clock: t7_sys_ddr
> > [    1.047991] clk: Disabled unused clock: pwm_ao_g
> > [    1.048565] clk: Disabled unused clock: pwm_ao_f
> > [    1.049135] clk: Disabled unused clock: pwm_ao_e
> > [    1.049709] clk: Disabled unused clock: pwm_ao_d
> > [    1.050283] clk: Disabled unused clock: pwm_ao_b
> > [    1.050857] clk: Disabled unused clock: pwm_ao_a
> > [    1.051454] clk: Disabled unused clock: rtc_dualdiv
> > [    1.052038] clk: Disabled unused clock: rtc_duandiv_in
> >
> > Also with the patches the disable happens 0.5 second later, from 0.5 seconds
> > after the boot to 1 second after the boot.
> >
> > I am still investigating what is missing a consumer or what should be
> > labeled critical.
>
> I've seen cases where a clock is modeled in Linux, but isn't actually
> used by Linux, so it gets caught in the disable unused clocks sweep.
>
> Does the board immediately crash after the disable unused clock sweep?
yes> If so, you could add a log statement, then 1 second sleep in
> clk_disable_unused_subtree() and wait for the board to crash.
good idea , thanks
> If you can't get timely help from the vendor, then you can add
> CLK_IGNORE_UNUSED to all of those clocks, and slowly start taking them
> off in batches and wait for the crash.
That's what I've been doing for the past few weeks, but I'm probably 
hunting down a set of clock issues, as support for this board is still 
in its early stages. It's a combination of clocks, which makes it way 
more time-consuming.

I went with a more localized approach:

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 2b6fae54702a..3358f4dc871f 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -1489,14 +1489,95 @@ static void clk_unprepare_unused_subtree(struct 
clk_core *core,
                else if (core->ops->unprepare)
                        core->ops->unprepare(core->hw);
                trace_clk_unprepare_complete(core);
+               pr_info("clk: V Unprepared unused clock: %s\n", core->name);
        }
 }

+static const char * const disable_clocks[] = {
+       "t7_sys_pwm_ao_gh",
+       "t7_sys_pwm_ao_ef",
+       "t7_sys_pwm_ef",
+       "t7_sys_pwm_cd",
+       "t7_sys_pwm_ab",
+       "t7_sys_ts_hevc",
+       "t7_sys_ts_vpu",
+       "t7_sys_ts_nna",
+       "t7_sys_ts_gpu",
+       "t7_sys_sar_adc",
+       "t7_sys_vpu_intr",
+       "t7_sys_dspb",
+       "t7_sys_dspa",
+       "t7_sys_a73pclk_apb",
+       "t7_sys_pclk_sys_apb",
+       "t7_sys_rsa",
+       "t7_sys_mipi_isp_pclk",
+       "t7_sys_mmc_apb",
+       "t7_sys_hdmirx_pclk",
+       "t7_sys_i2c_s_a",
+       "t7_sys_hdmitx_apb",
+       "t7_sys_i2c_m_f",
+       "t7_sys_i2c_m_e",
+       "t7_sys_i2c_m_d",
+       "t7_sys_i2c_m_c",
+       "t7_sys_i2c_m_b",
+       "t7_sys_i2c_m_a",
+       "t7_sys_i2c_ao_b",
+       "t7_sys_i2c_ao_a",
+       "t7_sys_pcie_phy",
+       "t7_sys_usb",
+       "t7_sys_pcie",
+       "t7_sys_spicc1",
+       "t7_sys_spicc0",
+       "t7_sys_g2d",
+       "t7_sys_spicc5",
+       "t7_sys_ts_a53",
+       "t7_sys_ts_a73",
+       "t7_sys_spicc4",
+       "t7_sys_spicc3",
+       "t7_sys_spicc2",
+       "t7_sys_aififo",
+       "t7_sys_uart_f",
+       "t7_sys_uart_e",
+       "t7_sys_uart_d",
+       "t7_sys_uart_c",
+       "t7_sys_uart_b",
+       "t7_sys_eth",
+       "t7_sys_audio",
+       "t7_sys_ir_ctrl",
+       "t7_sys_msr_clk",
+       "t7_sys_spifc",
+       "t7_sys_acodec",
+       "t7_sys_smartcard",
+       "t7_sys_am2axi2",
+       "t7_sys_am2axi1",
+       "t7_sys_am2axi0",
+       "t7_sys_ampipe_eth",
+       "t7_sys_ampipe_nand",
+       "t7_sys_deswarp",
+       "t7_sys_gdc",
+       "t7_sys_cec",
+       "t7_sys_aucpu",
+       "t7_sys_aocpu",
+       "t7_sys_mali",
+       "t7_sys_ethphy",
+       "t7_sys_mipi_dsi_b",
+       "t7_sys_mipi_dsi_a",
+       "t7_sys_dos",
+       "t7_sys_ddr",
+       "pwm_ao_g",
+       "pwm_ao_e",
+       "pwm_ao_a",
+       "rtc_dualdiv",
+       "rtc_duandiv_in",
+};
+
+
 static void clk_disable_unused_subtree(struct clk_core *core,
                                       struct device *dev)
 {
        struct clk_core *child;
        unsigned long flags;
+       int i;

        lockdep_assert_held(&prepare_lock);

@@ -1517,6 +1598,11 @@ static void clk_disable_unused_subtree(struct 
clk_core *core,
        if (core->flags & CLK_IGNORE_UNUSED)
                goto unlock_out;

+       for (i = 0; i < ARRAY_SIZE(disable_clocks); i++) {
+               if (strcmp(core->name, disable_clocks[i]) == 0)
+                       goto unlock_out;
+       }
+
        /*
         * some gate clocks have special needs during the disable-unused
         * sequence.  call .disable_unused if available, otherwise fall
@@ -1529,6 +1615,7 @@ static void clk_disable_unused_subtree(struct 
clk_core *core,
                else if (core->ops->disable)
                        core->ops->disable(core->hw);
                trace_clk_disable_complete(core);
+               pr_info("clk: V Disabled unused clock: %s\n", core->name);
        }

This makes it easier to let the sweep disable clocks for testing.

> Brian
Thanks
Lucas