| Message ID | 1422289444-15142-1-git-send-email-dianders@chromium.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
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 <dianders@chromium.org> I talked with Doug about it and this looks sane, so Reviewed-by: Heiko Stuebner <heiko@sntech.de> Heiko
On Mon, Jan 26, 2015 at 5:24 PM, Doug Anderson <dianders@chromium.org> wrote: > 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 <dianders@chromium.org> Patch applied with Heiko's review tag, but not as a fix at this point, it will be for v3.20. Yours, Linus Walleij
diff --git a/drivers/pinctrl/pinctrl-rockchip.c b/drivers/pinctrl/pinctrl-rockchip.c index d144330..dee7d5f 100644 --- a/drivers/pinctrl/pinctrl-rockchip.c +++ b/drivers/pinctrl/pinctrl-rockchip.c @@ -89,7 +89,7 @@ struct rockchip_iomux { * @reg_pull: optional separate register for additional pull settings * @clk: clock of the gpio bank * @irq: interrupt of the gpio bank - * @saved_enables: Saved content of GPIO_INTEN at suspend time. + * @saved_masks: Saved content of GPIO_INTEN at suspend time. * @pin_base: first pin number * @nr_pins: number of pins in this bank * @name: name of the bank @@ -108,7 +108,7 @@ struct rockchip_pin_bank { struct regmap *regmap_pull; struct clk *clk; int irq; - u32 saved_enables; + u32 saved_masks; u32 pin_base; u8 nr_pins; char *name; @@ -1545,8 +1545,8 @@ static void rockchip_irq_suspend(struct irq_data *d) struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); struct rockchip_pin_bank *bank = gc->private; - bank->saved_enables = irq_reg_readl(gc, GPIO_INTEN); - irq_reg_writel(gc, gc->wake_active, GPIO_INTEN); + bank->saved_masks = irq_reg_readl(gc, GPIO_INTMASK); + irq_reg_writel(gc, ~gc->wake_active, GPIO_INTMASK); } static void rockchip_irq_resume(struct irq_data *d) @@ -1554,35 +1554,7 @@ static void rockchip_irq_resume(struct irq_data *d) struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); struct rockchip_pin_bank *bank = gc->private; - irq_reg_writel(gc, bank->saved_enables, GPIO_INTEN); -} - -static void rockchip_irq_disable(struct irq_data *d) -{ - struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); - u32 val; - - irq_gc_lock(gc); - - val = irq_reg_readl(gc, GPIO_INTEN); - val &= ~d->mask; - irq_reg_writel(gc, val, GPIO_INTEN); - - irq_gc_unlock(gc); -} - -static void rockchip_irq_enable(struct irq_data *d) -{ - struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); - u32 val; - - irq_gc_lock(gc); - - val = irq_reg_readl(gc, GPIO_INTEN); - val |= d->mask; - irq_reg_writel(gc, val, GPIO_INTEN); - - irq_gc_unlock(gc); + irq_reg_writel(gc, bank->saved_masks, GPIO_INTMASK); } static int rockchip_interrupts_register(struct platform_device *pdev, @@ -1620,6 +1592,14 @@ static int rockchip_interrupts_register(struct platform_device *pdev, continue; } + /* + * Linux assumes that all interrupts start out disabled/masked. + * Our driver only uses the concept of masked and always keeps + * things enabled, so for us that's all masked and all enabled. + */ + writel_relaxed(0xffffffff, bank->reg_base + GPIO_INTMASK); + writel_relaxed(0xffffffff, bank->reg_base + GPIO_INTEN); + gc = irq_get_domain_generic_chip(bank->domain, 0); gc->reg_base = bank->reg_base; gc->private = bank; @@ -1628,8 +1608,6 @@ static int rockchip_interrupts_register(struct platform_device *pdev, gc->chip_types[0].chip.irq_ack = irq_gc_ack_set_bit; gc->chip_types[0].chip.irq_mask = irq_gc_mask_set_bit; gc->chip_types[0].chip.irq_unmask = irq_gc_mask_clr_bit; - gc->chip_types[0].chip.irq_enable = rockchip_irq_enable; - gc->chip_types[0].chip.irq_disable = rockchip_irq_disable; gc->chip_types[0].chip.irq_set_wake = irq_gc_set_wake; gc->chip_types[0].chip.irq_suspend = rockchip_irq_suspend; gc->chip_types[0].chip.irq_resume = rockchip_irq_resume;
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 <dianders@chromium.org> --- drivers/pinctrl/pinctrl-rockchip.c | 48 +++++++++++--------------------------- 1 file changed, 13 insertions(+), 35 deletions(-)