Re: [PATCH] kernel: x86: early-quirks: Replace mdelay with usleep_range in apple_airport_reset

From: Lukas Wunner
Date: Wed Jan 24 2018 - 00:42:02 EST


On Wed, Jan 24, 2018 at 10:38:22AM +0800, Jia-Ju Bai wrote:
> The function apple_airport_reset is not called in atomic context.
> Thus mdelay can be replaced with usleep_range, to avoid busy wait.
>
> This is reported by a static analysis tool named DCNS written by myself.

No, usleep_range() is built on hrtimers and at this point in the
boot sequence we haven't called hrtimers_init() yet.

Look at init/main.c:start_kernel() and note that setup_arch()
(which calls early_quirks()) is called way before hrtimers_init().

Please amend your static checker to consider every call to mdelay()
that occurs before hrtimers_init() a false positive, or manually
verify each patch for this constraint before submission.

I'll test your patch later today on my MacBook Pro but I suspect
it'll cause a boot crash or hang. In any case the choice of mdelay()
here was deliberate.

Thanks,

Lukas

>
> Signed-off-by: Jia-Ju Bai <baijiaju1990@xxxxxxxxx>
> ---
> arch/x86/kernel/early-quirks.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
> index 1e82f78..559e81a 100644
> --- a/arch/x86/kernel/early-quirks.c
> +++ b/arch/x86/kernel/early-quirks.c
> @@ -604,7 +604,7 @@ static void __init apple_airport_reset(int bus, int slot, int func)
> if ((pmcsr & PCI_PM_CTRL_STATE_MASK) != PCI_D0) {
> pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
> write_pci_config_16(bus, slot, func, BCM4331_PM_CAP + PCI_PM_CTRL, pmcsr);
> - mdelay(10);
> + usleep_range(10000, 11000);
>
> pmcsr = read_pci_config_16(bus, slot, func, BCM4331_PM_CAP + PCI_PM_CTRL);
> if ((pmcsr & PCI_PM_CTRL_STATE_MASK) != PCI_D0) {
> --
> 1.7.9.5
>