From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hans de Goede Subject: Re: [PATCH] platform/x86: intel_int0002_vgpio: Implement irq_set_wake Date: Sat, 2 Feb 2019 11:50:23 +0100 Message-ID: <5a56d2e8-c309-77f4-6ea6-a4e9d2bb0f1b@redhat.com> References: <20180924143719.22191-1-hdegoede@redhat.com> <35cfd1f2-43eb-fc49-4777-a21413bb2d41@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org To: Andy Shevchenko Cc: Maxim Mikityanskiy , Andy Shevchenko , Darren Hart , Platform Driver , Linux Kernel Mailing List List-Id: platform-driver-x86.vger.kernel.org Hi, On 2/2/19 12:25 AM, Andy Shevchenko wrote: > On Sat, Feb 2, 2019 at 12:52 AM Hans de Goede wrote: >> On 1/31/19 8:47 PM, Maxim Mikityanskiy wrote: >>> On Mon, Sep 24, 2018 at 5:37 PM Hans de Goede wrote: >>>> >>>> We were relying on the interrupt being shared with the ACPI SCI and the >>>> ACPI core calling irq_set_wake. But that does not always happen on >>>> Bay Trail devices, so we should do it ourselves. >>>> >>>> This fixes wake from USB not working on various Bay Trail devices. >>> >>> This patch breaks suspend on ASUS E202SA (bisecting pointed me to this >>> patch, and if I revert it and build 4.20.5 without this patch, >>> everything works flawlessly). >> >> Thank you for the bug report, can you please test 4.20.5 with the attached >> patch on top? That should fix it. Once I've confirmation that this fixes >> things I will submit the patch upstream. > > Thanks for fast reply, Hans! > > I looked at the patch and would propose one modification, i.e. use > INTEL_CPU_FAM6() macro instead of ICPU(). I just tried and I like the idea, but this does not work. The problem is that macro assumes that the driver_data being passed is always a struct and the & to take the address of the struct is part of the macro instead of passed with the argument, making it impossible to pass a simple integer value or a bunch of flags (in a bitmask). IMHO this is a shortcoming of the macro and should be fixed, moving the & to the callers of the macro. Given that this patch needs to go to stable to fix the regression I think it is best to just keep the patch as is. And then for -next, once the macro is fixed we can convert the int0002_vgpio driver to use it. Regards, Hans