From mboxrd@z Thu Jan 1 00:00:00 1970 From: sudeep.holla@arm.com (Sudeep Holla) Date: Wed, 26 Aug 2015 09:52:16 +0100 Subject: [PATCH v9 1/1] irqchip: imx-gpcv2: IMX GPCv2 driver for wakeup sources In-Reply-To: References: <1440443055-7291-1-git-send-email-shenwei.wang@freescale.com> <55DC3452.3070205@arm.com> <55DC738D.8000302@arm.com> <55DC7FA4.7040902@arm.com> <55DC96BA.2020905@arm.com> Message-ID: <55DD7E40.1040009@arm.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi Thomas, On 25/08/15 20:26, Thomas Gleixner wrote: > On Tue, 25 Aug 2015, Sudeep Holla wrote: >> On 25/08/15 15:54, Shenwei Wang wrote: >>> You don't really understand what happens after a driver calls >>> enable_irq_wake. In suspend state, even the interrupt >>> controller itself is powered off. How can you get the system up >>> again by just using a SKIP_SET_WAKE. >> >> Sorry for that, let me try to understand aloud. So you have >> irq_{un,}mask function that are called when interrupts are enabled and >> disabled. So suppose you have 3 irqs that are enabled and only one of >> then is set as wakeup source. >> >> Now you call enable_irq_wake, you save that in wakeup_sources, fine. >> Later when you enter suspend, you save all the 3 active irqs in >> saved_irq_mask and over-write cpu2wakeup with wakeup_sources, right? >> >> All fine, what I am saying is let irq-core know that you want to mask >> the 2 non-wakeup irqs you have using MASK_ON_SUSPEND. So when >> suspend_device_irqs is called in suspend path, that's done for you >> automatically and the cpu2wakeup will have just 1 wakeup enabled which >> is what you are doing in suspend callback, right ? > > I missed that when I reviewed the patch. You are right, it can be > simplified. > >> Now that it's already done for you, you need not do anything extra and >> hence just set SKIP_SET_WAKE to do nothing. Thanks for confirming this. > > He still needs the set_wake function to capture the wake enabled > interrupts as they are handed over to the low level asm code via > imx_gpcv2_get_wakeup_source(). Though they could be read back from the > hw registers as well. > Correct, I have no objection on how it's saved/restored but was against having 2 different masks and with the approach taken in this patch the non-wakeup interrupts are enabled until syscore_suspend which is wrong. There's whole lot of drivers blindly copy pasting this as theme which I am going through and trying to clean up. I wanted to ensure no more such additions happen meanwhile, so I am watching such patches closely. Regards, Sudeep