Dovetail fixes to v5.3

Dovetail fixes to v5.3

[Philippe Gerum]

This series goes on top of v5.13-dovetail-rebase. Backport to
v5.10.y-dovetail-rebase is on its way.

[PATCH 1/6] genirq: irq_pipeline: restore oob stall bit on chained IRQs

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

handle_oob_irq() is called recursively when handling chained
interrupts, at which point the oob stage is already stalled:

generic_handle_irq /* oob stage unstalled */
   -> <parent IRQ flow handler>
          -> handle_oob_irq  /* oob stage stalled */
	        -> <parent IRQ demultiplexer>
		       -> generic_handle_irq
			      -> <chained IRQ flow handler>
			             -> handle_oob_irq
				     /* oob stage must stay stalled */
				 ...
		          ...
	           ...

		  /*
		   * oob stage is unstalled upon leaving the outer call
		   * to handle_oob_irq
		   */

Make sure to restore the oob stall bit to its original value on entry
to this routine before leaving it, instead of clearing such bit
unconditionally.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 kernel/irq/pipeline.c | 38 +++++++++++++++++++++++---------------
 1 file changed, 23 insertions(+), 15 deletions(-)

diff --git a/kernel/irq/pipeline.c b/kernel/irq/pipeline.c
index a220d9467a56ce2..090fcc8bbb6a449 100644
--- a/kernel/irq/pipeline.c
+++ b/kernel/irq/pipeline.c
@@ -931,7 +931,7 @@ static inline bool is_active_edge_event(struct irq_desc *desc)
 bool handle_oob_irq(struct irq_desc *desc) /* hardirqs off */
 {
 	struct irq_stage_data *oobd = this_oob_staged();
-	unsigned int irq = irq_desc_get_irq(desc);
+	unsigned int irq = irq_desc_get_irq(desc), s;
 
 	/*
 	 * Flow handlers of chained interrupts have no business
@@ -959,19 +959,8 @@ bool handle_oob_irq(struct irq_desc *desc) /* hardirqs off */
 
 	if (WARN_ON_ONCE(irq_pipeline_debug() && running_inband()))
 		return false;
-	/*
-	 * Running with the oob stage stalled implies hardirqs off, so
-	 * if the oob stage is stalled on pipeline entry, something is
-	 * badly broken in our interrupt state. Pretend the event has
-	 * been handled, which may end up with the device hammering us
-	 * with more interrupts, but there is no safe option at this
-	 * point.
-	 */
-	if (WARN_ON_ONCE(irq_pipeline_debug() &&
-			on_pipeline_entry() && test_oob_stall()))
-		return true;
 
-	stall_oob();
+	s = test_and_stall_oob();
 
 	if (unlikely(desc->istate & IRQS_EDGE)) {
 		do {
@@ -985,7 +974,13 @@ bool handle_oob_irq(struct irq_desc *desc) /* hardirqs off */
 		do_oob_irq(desc);
 	}
 
-	unstall_oob();
+	/*
+	 * Cascaded interrupts enter handle_oob_irq() with the
+	 * out-of-band stage stalled during the parent
+	 * invocation. Make sure to restore accordingly.
+	 */
+	if (likely(!s))
+		unstall_oob();
 
 	/*
 	 * CPU migration and/or stage switching over the handler are
@@ -1060,7 +1055,7 @@ void restore_stage_on_irq(struct irq_stage_data *prevd)
  *	@regs:	Register file coming from the low-level handling code
  *
  *	Inject an IRQ into the pipeline from a CPU interrupt or trap
- *	context.  A flow handler runs for this IRQ.
+ *	context.  A flow handler runs next for this IRQ.
  *
  *      Hard irqs must be off on entry.
  */
@@ -1081,6 +1076,19 @@ int generic_pipeline_irq(unsigned int irq, struct pt_regs *regs)
 			pr_err("IRQ pipeline: interrupts enabled on entry (IRQ%u)\n",
 			       irq);
 		}
+		/*
+		 * Running with the oob stage stalled implies hardirqs
+		 * off.  For this reason, if the oob stage is stalled
+		 * on pipeline entry but we still receive an interrupt
+		 * from the hardware, something is badly broken in our
+		 * interrupt state. Try fixing up, but without great
+		 * hopes.
+		 */
+		if (on_pipeline_entry() && test_oob_stall()) {
+			pr_err("IRQ pipeline: out-of-band stage stalled on IRQ entry\n");
+			unstall_oob();
+		}
+
 		if (unlikely(desc == NULL)) {
 			pr_err("IRQ pipeline: received unhandled IRQ%u\n",
 			       irq);
-- 
2.31.1

[PATCH 2/6] gpio: omap: irq_pipeline: enable out-of-band interrupt handling

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

We must decode the chained interrupts on the out-of-band stage to
enable end-to-end out-of-band delivery. Set IRQF_OOB for the parent
handler.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 drivers/gpio/gpio-omap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c
index 8c02307ab2d0aa7..9e0eb9a7bfa671b 100644
--- a/drivers/gpio/gpio-omap.c
+++ b/drivers/gpio/gpio-omap.c
@@ -1055,7 +1055,7 @@ static int omap_gpio_chip_init(struct gpio_bank *bank, struct irq_chip *irqc)
 
 	ret = devm_request_irq(bank->chip.parent, bank->irq,
 			       omap_gpio_irq_handler,
-			       0, dev_name(bank->chip.parent), bank);
+			       IRQF_OOB, dev_name(bank->chip.parent), bank);
 	if (ret)
 		gpiochip_remove(&bank->chip);
 
-- 
2.31.1

[PATCH 3/6] drm/msm/mdp5: irq_pipeline: enable out-of-band interrupt handling

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

We must decode the chained interrupts on the out-of-band stage to
enable end-to-end out-of-band delivery. Set IRQF_OOB for the parent
handler.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 drivers/gpu/drm/msm/disp/mdp5/mdp5_mdss.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/msm/disp/mdp5/mdp5_mdss.c b/drivers/gpu/drm/msm/disp/mdp5/mdp5_mdss.c
index 8e0f2be155d8625..781d701dfc75457 100644
--- a/drivers/gpu/drm/msm/disp/mdp5/mdp5_mdss.c
+++ b/drivers/gpu/drm/msm/disp/mdp5/mdp5_mdss.c
@@ -255,7 +255,7 @@ int mdp5_mdss_init(struct drm_device *dev)
 	}
 
 	ret = devm_request_irq(dev->dev, platform_get_irq(pdev, 0),
-			       mdss_irq, 0, "mdss_isr", mdp5_mdss);
+			       mdss_irq, IRQF_OOB, "mdss_isr", mdp5_mdss);
 	if (ret) {
 		DRM_DEV_ERROR(dev->dev, "failed to init irq: %d\n", ret);
 		goto fail_irq;
-- 
2.31.1

[PATCH 4/6] pinctrl: samsung: irq_pipeline: enable out-of-band interrupt handling

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

We must decode the chained interrupts on the out-of-band stage to
enable end-to-end out-of-band delivery. Set IRQF_OOB for the parent
handler.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 drivers/pinctrl/samsung/pinctrl-exynos.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pinctrl/samsung/pinctrl-exynos.c b/drivers/pinctrl/samsung/pinctrl-exynos.c
index e1c86adb8f9421e..7da7f80e654380a 100644
--- a/drivers/pinctrl/samsung/pinctrl-exynos.c
+++ b/drivers/pinctrl/samsung/pinctrl-exynos.c
@@ -288,7 +288,7 @@ __init int exynos_eint_gpio_init(struct samsung_pinctrl_drv_data *d)
 	}
 
 	ret = devm_request_irq(dev, d->irq, exynos_eint_gpio_irq,
-					0, dev_name(dev), d);
+					IRQF_OOB, dev_name(dev), d);
 	if (ret) {
 		dev_err(dev, "irq request failed\n");
 		return -ENXIO;
-- 
2.31.1

[PATCH 5/6] genirq: irq_pipeline: export oob status to /proc/interrupts

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 kernel/irq/proc.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c
index 7c5cd42df3b93e7..1428d83d7da8658 100644
--- a/kernel/irq/proc.c
+++ b/kernel/irq/proc.c
@@ -518,6 +518,9 @@ int show_interrupts(struct seq_file *p, void *v)
 		seq_printf(p, " %*s", prec, "");
 #ifdef CONFIG_GENERIC_IRQ_SHOW_LEVEL
 	seq_printf(p, " %-8s", irqd_is_level_type(&desc->irq_data) ? "Level" : "Edge");
+#endif
+#ifdef CONFIG_IRQ_PIPELINE
+	seq_printf(p, " %-3s", irq_settings_is_oob(desc) ? "oob" : "");
 #endif
 	if (desc->name)
 		seq_printf(p, "-%-8s", desc->name);
-- 
2.31.1

[PATCH 6/6] clocksource/exynos_mct: irq_pipeline: enable out-of-band clock events

[Philippe Gerum]

From: Philippe Gerum <rpm@xenomai.org>

Ensure end-of-end out-of-band event delivery by receiving the clock
interrupts directly on the out-of-band stage.

Signed-off-by: Philippe Gerum <rpm@xenomai.org>
---
 drivers/clocksource/exynos_mct.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/clocksource/exynos_mct.c b/drivers/clocksource/exynos_mct.c
index a60cecf62c1cd6c..8d8ad16c13c3712 100644
--- a/drivers/clocksource/exynos_mct.c
+++ b/drivers/clocksource/exynos_mct.c
@@ -340,7 +340,7 @@ static int exynos4_clockevent_init(void)
 	clockevents_config_and_register(&mct_comp_device, clk_rate,
 					0xf, 0xffffffff);
 	if (request_irq(mct_irqs[MCT_G0_IRQ], exynos4_mct_comp_isr,
-			IRQF_TIMER | IRQF_IRQPOLL, "mct_comp_irq",
+			IRQF_TIMER | IRQF_IRQPOLL | IRQF_OOB, "mct_comp_irq",
 			&mct_comp_device))
 		pr_err("%s: request_irq() failed\n", "mct_comp_irq");
 
-- 
2.31.1

Re: [PATCH 5/6] genirq: irq_pipeline: export oob status to /proc/interrupts

[Jan Kiszka]

On 22.07.21 11:48, Philippe Gerum via Xenomai wrote:
> From: Philippe Gerum <rpm@xenomai.org>
> 
> Signed-off-by: Philippe Gerum <rpm@xenomai.org>
> ---
>  kernel/irq/proc.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c
> index 7c5cd42df3b93e7..1428d83d7da8658 100644
> --- a/kernel/irq/proc.c
> +++ b/kernel/irq/proc.c
> @@ -518,6 +518,9 @@ int show_interrupts(struct seq_file *p, void *v)
>  		seq_printf(p, " %*s", prec, "");
>  #ifdef CONFIG_GENERIC_IRQ_SHOW_LEVEL
>  	seq_printf(p, " %-8s", irqd_is_level_type(&desc->irq_data) ? "Level" : "Edge");
> +#endif
> +#ifdef CONFIG_IRQ_PIPELINE
> +	seq_printf(p, " %-3s", irq_settings_is_oob(desc) ? "oob" : "");
>  #endif
>  	if (desc->name)
>  		seq_printf(p, "-%-8s", desc->name);
> 

Small but valuable - already received a question in this direction.

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux

Re: [PATCH 5/6] genirq: irq_pipeline: export oob status to /proc/interrupts

[Philippe Gerum]

Jan Kiszka <jan.kiszka@siemens.com> writes:

> On 22.07.21 11:48, Philippe Gerum via Xenomai wrote:
>> From: Philippe Gerum <rpm@xenomai.org>
>> 
>> Signed-off-by: Philippe Gerum <rpm@xenomai.org>
>> ---
>>  kernel/irq/proc.c | 3 +++
>>  1 file changed, 3 insertions(+)
>> 
>> diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c
>> index 7c5cd42df3b93e7..1428d83d7da8658 100644
>> --- a/kernel/irq/proc.c
>> +++ b/kernel/irq/proc.c
>> @@ -518,6 +518,9 @@ int show_interrupts(struct seq_file *p, void *v)
>>  		seq_printf(p, " %*s", prec, "");
>>  #ifdef CONFIG_GENERIC_IRQ_SHOW_LEVEL
>>  	seq_printf(p, " %-8s", irqd_is_level_type(&desc->irq_data) ? "Level" : "Edge");
>> +#endif
>> +#ifdef CONFIG_IRQ_PIPELINE
>> +	seq_printf(p, " %-3s", irq_settings_is_oob(desc) ? "oob" : "");
>>  #endif
>>  	if (desc->name)
>>  		seq_printf(p, "-%-8s", desc->name);
>> 
>
> Small but valuable - already received a question in this direction.
>

This was deeply buried in my todo list until Byron fortunately mentioned
it.

I'm not 100% happy with the format since this new field might affect
some parsers fed by /proc/interrupts, but I could not find any better
way which would be generic enough, not depending on the specific
printout which may be done by the irqchip driver.

-- 
Philippe.