Re: [PATCH] pinctrl: aspeed-g5: Delay acquisition of regmaps

[Linus Walleij <linus.walleij@xxxxxxxxxx>]

On Wed, Jul 24, 2019 at 10:02 AM Andrew Jeffery <andrew@xxxxxxxx> wrote:

> While sorting out some devicetree issues I found that the pinctrl driver
> was failing to acquire its GFX regmap even though the phandle was
> present in the devicetree:
>
>     [    0.124190] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: No GFX phandle found, some mux configurations may fail
>
> Without access to the GFX regmap we fail to configure the mux for the
> VPO function:
>
>     [    1.548866] pinctrl core: add 1 pinctrl maps
>     [    1.549826] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: found group selector 164 for VPO
>     [    1.550638] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: request pin 144 (V20) for 1e6e6000.display
>     [    1.551346] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: request pin 145 (U19) for 1e6e6000.display
>     ...
>     [    1.562057] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: request pin 218 (T22) for 1e6e6000.display
>     [    1.562541] aspeed-g5-pinctrl 1e6e2000.syscon:pinctrl: request pin 219 (R20) for 1e6e6000.display
>     [    1.563113] Muxing pin 144 for VPO
>     [    1.563456] Want SCU8C[0x00000001]=0x1, got 0x0 from 0x00000000
>     [    1.564624] aspeed_gfx 1e6e6000.display: Error applying setting, reverse things back
>
> This turned out to be a simple problem of timing: The ASPEED pinctrl
> driver is probed during arch_initcall(), while GFX is processed much
> later. As such the GFX syscon is not yet registered during the pinctrl
> probe() and we get an -EPROBE_DEFER when we try to look it up, however
> we must not defer probing the pinctrl driver for the inability to mux
> some GFX-related functions.
>
> Switch to lazily grabbing the regmaps when they're first required by the
> mux configuration. This generates a bit of noise in the patch as we have
> to drop the `const` qualifier on arguments for several function
> prototypes, but has the benefit of working.
>
> I've smoke tested this for the ast2500-evb under qemu with a dummy
> graphics device. We now succeed in our attempts to configure the SoC's
> VPO pinmux function.
>
> Fixes: 7d29ed88acbb ("pinctrl: aspeed: Read and write bits in LPC and GFX controllers")
> Signed-off-by: Andrew Jeffery <andrew@xxxxxxxx>

Patch applied for fixes. Good rootcausing!

Yours,
Linus Walleij