From mboxrd@z Thu Jan 1 00:00:00 1970 From: Heiko =?ISO-8859-1?Q?St=FCbner?= Subject: Re: [PATCH] pinctrl: rockchip: Only mask interrupts; never disable Date: Mon, 26 Jan 2015 17:56:19 +0100 Message-ID: <1844800.Wx1PProku3@diego> References: <1422289444-15142-1-git-send-email-dianders@chromium.org> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7Bit Return-path: In-Reply-To: <1422289444-15142-1-git-send-email-dianders@chromium.org> Sender: linux-kernel-owner@vger.kernel.org To: Doug Anderson Cc: Linus Walleij , Jeffy Chen , linux-arm-kernel@lists.infradead.org, linux-rockchip@lists.infradead.org, linux-gpio@vger.kernel.org, linux-kernel@vger.kernel.org List-Id: linux-gpio@vger.kernel.org Am Montag, 26. Januar 2015, 08:24:03 schrieb Doug Anderson: > The Rockchip GPIO interrupt controller totally throws away all status > about an interrupt when you "disable" the interrupt. That has > unfortunate consequences in the following situation: > > 1. An edge-triggered interrupt is enabled and should wake the system. > 2. System suspend happens: interrupt is disabled and marked for wake. > 3. rockchip_irq_suspend() reenables the interrupt so we can wake. > 4. Interrupt happens when asleep. > 5. rockchip_irq_resume() redisables the interrupt. > 6. Disabling the interrupt throws away all status about it. > 7. Normal system resume happens and we enable the interrupt again, > since we threw away status about the interrupt we don't know it > fired while suspended. Even worse: if we need both edges of the > interrupt the logic to swap edges never runs. > > Note: even if we somehow can post the status about wakeup interrupts > in rockchip_irq_resume() we would still have a window of losing any > edges that came in while interrupts were disabled. > > If we use mask only then we don't need to worry. The GPIO Interrupt > controller keeps track of pending interrupts that are enabled and just > masked. > > There was no real strong reason to support the enable/disable > functionality (other than that it seemed right), so let's go back to > just supporting mask/unmask but actually map it to the real > mask/unmask. This ends up with slightly different (and more correct) > behavior than before (f2dd028 pinctrl: rockchip: Fix > enable/disable/mask/unmask). > > Signed-off-by: Doug Anderson I talked with Doug about it and this looks sane, so Reviewed-by: Heiko Stuebner Heiko From mboxrd@z Thu Jan 1 00:00:00 1970 From: heiko@sntech.de (Heiko =?ISO-8859-1?Q?St=FCbner?=) Date: Mon, 26 Jan 2015 17:56:19 +0100 Subject: [PATCH] pinctrl: rockchip: Only mask interrupts; never disable In-Reply-To: <1422289444-15142-1-git-send-email-dianders@chromium.org> References: <1422289444-15142-1-git-send-email-dianders@chromium.org> Message-ID: <1844800.Wx1PProku3@diego> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Am Montag, 26. Januar 2015, 08:24:03 schrieb Doug Anderson: > The Rockchip GPIO interrupt controller totally throws away all status > about an interrupt when you "disable" the interrupt. That has > unfortunate consequences in the following situation: > > 1. An edge-triggered interrupt is enabled and should wake the system. > 2. System suspend happens: interrupt is disabled and marked for wake. > 3. rockchip_irq_suspend() reenables the interrupt so we can wake. > 4. Interrupt happens when asleep. > 5. rockchip_irq_resume() redisables the interrupt. > 6. Disabling the interrupt throws away all status about it. > 7. Normal system resume happens and we enable the interrupt again, > since we threw away status about the interrupt we don't know it > fired while suspended. Even worse: if we need both edges of the > interrupt the logic to swap edges never runs. > > Note: even if we somehow can post the status about wakeup interrupts > in rockchip_irq_resume() we would still have a window of losing any > edges that came in while interrupts were disabled. > > If we use mask only then we don't need to worry. The GPIO Interrupt > controller keeps track of pending interrupts that are enabled and just > masked. > > There was no real strong reason to support the enable/disable > functionality (other than that it seemed right), so let's go back to > just supporting mask/unmask but actually map it to the real > mask/unmask. This ends up with slightly different (and more correct) > behavior than before (f2dd028 pinctrl: rockchip: Fix > enable/disable/mask/unmask). > > Signed-off-by: Doug Anderson I talked with Doug about it and this looks sane, so Reviewed-by: Heiko Stuebner Heiko