On 6/30/2020 9:37 PM, Mark Tomlinson wrote: > On Tue, 2020-06-30 at 20:14 -0700, Florian Fainelli wrote: >> Sorry, it looks like I made a mistake in my testing (or I was lucky), >>> and this patch doesn't fix the issue. What is happening is: >>> 1) nsp-pinmux driver is registered (arch_initcall). >>> 2) nsp-gpio-a driver is registered (arch_initcall_sync). >>> 3) of_platform_default_populate_init() is called (also at level >>> arch_initcall_sync), which scans the device tree, adds the nsp-gpio-a >>> device, runs its probe, and this returns -EPROBE_DEFER with the error >>> message. >>> 4) Only now nsp-pinmux device is probed. >>> >>> Changing the 'arch_initcall_sync' to 'device_initcall' in nsp-gpio-a >>> ensures that the pinmux is probed first since >>> of_platform_default_populate_init() will be called between the two >>> register calls, and the error goes away. Is this change acceptable as a >>> solution? >> >> If probe deferral did not work, certainly but it sounds like this is >> being done just for the sake of eliminating a round of probe deferral, >> is there a functional problem this is fixing? > > No, I'm just trying to prevent an "error" message appearing in syslog. > >>> The actual error message in syslog is: >>> >>> kern.err kernel: gpiochip_add_data_with_key: GPIOs 480..511 >>> (18000020.gpio) failed to register, -517 >>> >>> So an end user sees "err" and "failed", and doesn't know what "-517" >>> means. >> >> How about this instead: >> >> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c >> index 4fa075d49fbc..10d9d0c17c9e 100644 >> --- a/drivers/gpio/gpiolib.c >> +++ b/drivers/gpio/gpiolib.c >> @@ -1818,9 +1818,10 @@ int gpiochip_add_data_with_key(struct gpio_chip >> *gc, void *data, >> ida_simple_remove(&gpio_ida, gdev->id); >> err_free_gdev: >> /* failures here can mean systems won't boot... */ >> - pr_err("%s: GPIOs %d..%d (%s) failed to register, %d\n", __func__, >> - gdev->base, gdev->base + gdev->ngpio - 1, >> - gc->label ? : "generic", ret); >> + if (ret != -EPROBE_DEFER) >> + pr_err("%s: GPIOs %d..%d (%s) failed to register, %d\n", >> + __func__, gdev->base, gdev->base + gdev->ngpio - 1, >> + gc->label ? : "generic", ret); >> kfree(gdev); >> return ret; >> } >> > That was one of my thoughts too. I found someone had tried that > earlier, but it was rejected: > > > https://patchwork.ozlabs.org/project/linux-gpio/patch/1516566774-1786-1-git-send-email-david@xxxxxxxxxxxxxx/ clk or reset APIs do not complain loudly on EPROBE_DEFER, it seems to me that GPIO should follow here. Also, it does look like Linus was in agreement in the end, not sure why it was not applied though. -- Florian