sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

[Jonathan Neuschäfer]

[-- Attachment #1: Type: text/plain, Size: 1052 bytes --]

Hi,

I'm currently porting mainline Linux to an i.MX507-based board
(preliminary DT here[1]). Recently I've enabled esdhc1 in the
devicetree.

However, as soon as the mmc driver (sdhci-esdhc-imx.c) disables clk_per
in sdhci_esdhc_runtime_suspend[2], the whole system locks up in some way.
Booting with CONFIG_PM=n "fixes" the problem; commenting out the
offending line in sdhci_esdhc_runtime_suspend lets the system survive,
but leaves the MMC controller in an unusable state (of course it's not a
proper solution anyway).

I tried to compare the CCM's clocks between i.MX50 and i.MX53, but
unfortunately, the i.MX50 reference manual doesn't have the table called
"Output clocks from CCM".

Does anyone have any ideas or patches that I could try in order to
narrow the problem further down?


Thanks,
Jonathan Neuschäfer

[1]: https://gist.github.com/neuschaefer/27cacf0acf5411219be9372d728d1163
[2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/mmc/host/sdhci-esdhc-imx.c?h=v5.0-rc5#n1412

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

[Fabio Estevam]

Hi Jonathan,

On Thu, Feb 7, 2019 at 8:52 PM Jonathan Neuschäfer
<j.neuschaefer@gmx.net> wrote:
>
> Hi,
>
> I'm currently porting mainline Linux to an i.MX507-based board
> (preliminary DT here[1]). Recently I've enabled esdhc1 in the
> devicetree.
>
> However, as soon as the mmc driver (sdhci-esdhc-imx.c) disables clk_per
> in sdhci_esdhc_runtime_suspend[2], the whole system locks up in some way.
> Booting with CONFIG_PM=n "fixes" the problem; commenting out the
> offending line in sdhci_esdhc_runtime_suspend lets the system survive,
> but leaves the MMC controller in an unusable state (of course it's not a
> proper solution anyway).
>
> I tried to compare the CCM's clocks between i.MX50 and i.MX53, but
> unfortunately, the i.MX50 reference manual doesn't have the table called
> "Output clocks from CCM".

Please check Table 5-10. CCM_CCGR3 Gated Clock Mapping to Target
Module from the MX50 Referene Manual.

>
> Does anyone have any ideas or patches that I could try in order to
> narrow the problem further down?

Does the change below help?

--- a/arch/arm/boot/dts/imx50.dtsi
+++ b/arch/arm/boot/dts/imx50.dtsi
@@ -102,7 +102,7 @@
                                        reg = <0x50004000 0x4000>;
                                        interrupts = <1>;
                                        clocks = <&clks
IMX5_CLK_ESDHC1_IPG_GATE>,
-                                                <&clks IMX5_CLK_DUMMY>,
+                                                <&clks
IMX5_CLK_ESDHC1_IPG_GATE>,
                                                 <&clks
IMX5_CLK_ESDHC1_PER_GATE>;
                                        clock-names = "ipg", "ahb", "per";
                                        bus-width = <4>;
@@ -114,7 +114,7 @@
                                        reg = <0x50008000 0x4000>;
                                        interrupts = <2>;
                                        clocks = <&clks
IMX5_CLK_ESDHC2_IPG_GATE>,
-                                                <&clks IMX5_CLK_DUMMY>,
+                                                <&clks
IMX5_CLK_ESDHC2_IPG_GATE>,
                                                 <&clks
IMX5_CLK_ESDHC2_PER_GATE>;
                                        clock-names = "ipg", "ahb", "per";
                                        bus-width = <4>;

Re: sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

[Jonathan Neuschäfer]

[-- Attachment #1: Type: text/plain, Size: 1827 bytes --]

On Thu, Feb 07, 2019 at 10:50:28PM -0200, Fabio Estevam wrote:
> On Thu, Feb 7, 2019 at 8:52 PM Jonathan Neuschäfer
[...]
> > I tried to compare the CCM's clocks between i.MX50 and i.MX53, but
> > unfortunately, the i.MX50 reference manual doesn't have the table called
> > "Output clocks from CCM".
> 
> Please check Table 5-10. CCM_CCGR3 Gated Clock Mapping to Target
> Module from the MX50 Referene Manual.

Ok, the tables show:

For i.MX50:
[1:0] 0 ipg_clk_root    eSDHCv2_1
[3:2] 1 esdhc1_clk_root eSDHCv2_1

For i.MX53:
1–0 CG0 esdhc1_ipg_hclk: affects ipg_clk and hclk inputs of ESDEHC-1 (esdhc1_clk_enable)
3–2 CG1 esdhc1_perclk:   affects ipg_clk_perclk   input  of ESDEHC-1 (esdhc1_serial_clk_enable)

Table 18-3 (Output clocks from CCM) in the iMX53RM shows that
ESDHCv2-1's ipg_clk_perclk is esdhc1_clk_root, so the clock structure
does seem to be the same here, between i.MX50 and i.MX53…

(The reason why I looked at i.MX53 is that there are several i.MX53
boards in tree, so it appears to be well tested under mainline Linux)


> Does the change below help?
> 
> --- a/arch/arm/boot/dts/imx50.dtsi
> +++ b/arch/arm/boot/dts/imx50.dtsi
> @@ -102,7 +102,7 @@
>                                         reg = <0x50004000 0x4000>;
>                                         interrupts = <1>;
>                                         clocks = <&clks IMX5_CLK_ESDHC1_IPG_GATE>,
> -                                                <&clks IMX5_CLK_DUMMY>,
> +                                                <&clks IMX5_CLK_ESDHC1_IPG_GATE>,
>                                                  <&clks IMX5_CLK_ESDHC1_PER_GATE>;
>                                         clock-names = "ipg", "ahb", "per";

Unfortunately, this doesn't help.


Thanks,
Jonathan Neuschäfer

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

[Fabio Estevam]

On Sat, Feb 9, 2019 at 2:25 PM Jonathan Neuschäfer
<j.neuschaefer@gmx.net> wrote:
>
> On Thu, Feb 07, 2019 at 10:50:28PM -0200, Fabio Estevam wrote:
> > On Thu, Feb 7, 2019 at 8:52 PM Jonathan Neuschäfer
> [...]
> > > I tried to compare the CCM's clocks between i.MX50 and i.MX53, but
> > > unfortunately, the i.MX50 reference manual doesn't have the table called
> > > "Output clocks from CCM".
> >
> > Please check Table 5-10. CCM_CCGR3 Gated Clock Mapping to Target
> > Module from the MX50 Referene Manual.
>
> Ok, the tables show:
>
> For i.MX50:
> [1:0] 0 ipg_clk_root    eSDHCv2_1

and it also shows that ahb_clk_root is controlled by these same bits.

That's why I added the ahb entry in my suggestion.

> [3:2] 1 esdhc1_clk_root eSDHCv2_1
>
> For i.MX53:
> 1–0 CG0 esdhc1_ipg_hclk: affects ipg_clk and hclk inputs of ESDEHC-1 (esdhc1_clk_enable)
> 3–2 CG1 esdhc1_perclk:   affects ipg_clk_perclk   input  of ESDEHC-1 (esdhc1_serial_clk_enable)
>
> Table 18-3 (Output clocks from CCM) in the iMX53RM shows that
> ESDHCv2-1's ipg_clk_perclk is esdhc1_clk_root, so the clock structure
> does seem to be the same here, between i.MX50 and i.MX53…

That's correct. The esdhc clocks seem to differ.

Re: sdhci-esdhc-imx/v5.0-rc5: i.MX50 system hangs when "per" clock is disabled

[Jonathan Neuschäfer]

[-- Attachment #1: Type: text/plain, Size: 803 bytes --]

On Thu, Feb 07, 2019 at 11:52:11PM +0100, Jonathan Neuschäfer wrote:
> However, as soon as the mmc driver (sdhci-esdhc-imx.c) disables clk_per
> in sdhci_esdhc_runtime_suspend[2], the whole system locks up in some way.

I've traced clk prepare/unprepare/enable/disable calls and got this
right before it hangs:

[    1.280981] sdhci_esdhc_runtime_suspend: unprepare clk_per
[    1.283356] clk_core_unprepare esdhc1_per_gate
[    1.285191] clk_core_unprepare esdhc_a_podf
[    1.286881] clk_core_unprepare esdhc_a_pred
[    1.288568] clk_core_unprepare esdhc_a_sel
[    1.290220] clk_core_unprepare pll2_sw

I suspect that there is a dependency on pll2_sw that is not represented
in the imx5 clock driver. I'm comparing it to the Reference Manual now.


Thanks,
Jonathan Neuschäfer

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]