* Mark Brown <broonie@xxxxxxxxxx> [170327 10:52]:
> On Wed, Mar 22, 2017 at 10:10:49AM -0700, Tony Lindgren wrote:
> > At least Motorola CPCAP PMIC needs it's device interrupts re-read
> > until there are no more interrupts. Otherwise the PMIC interrupt to
> > the SoC will eventually stop toggling. This seems to be a bug in the
> > CPCAP PMIC where it can stop driving the GPIO interrupt to the SoC
> > with pending CPCAP interrupts.
>
> > Let's allow handling issues like this by introducing a flag for
> > handle_reread and by splitting regmap_irq_thread() into two separate
> > functions for regmap_read_irq_status() and regmap_irq_handle_pending().
>
> So, I see your use case but the fact is that as Charles observed this is
> exactly the code used for emulating level triggered IRQs with edge
> triggered interrupt controllers. This means someone is doubtless going
> to end up using it for precisely that. This makes me uncomfortable, we
> do have this open coded into various drivers already but this is more of
> a core thing and it feels like this should be in genirq rather than
> here... that said, looking at the code:
Yes.. But then again we might avoid piling up yet more driver
specific hacks. I don't know what the genirq solution would look
like, handle until we get IRQ_NONE? :)
> > + do {
> > + ret = regmap_read_irq_status(data);
> > + if (ret)
> > + goto out_runtime_put;
> > +
> > + ret = regmap_irq_handle_pending(data);
> > + if (ret < 0)
> > + goto out_runtime_put;
> > +
> > + if (!ret)
> > + break;
> > +
> > + handled += ret;
> > + } while (chip->handle_reread);
>
> There's no protection against screaming interrupts here, I'd really like
> to see that. Also some tracing of the number of times we spin.
Good idea. How about something like below where handle_reread checks
for the total time spent in the thread loop with a large enough
timeout? Or do you have some better ideas in mind?
I tested I can hit that warning with timeout set to much lower
10 ms with retries being 1 or 2 at that point.
Regards,
Tony
8< -------------------------------
>From tony Mon Sep 17 00:00:00 2001
From: Tony Lindgren <tony@xxxxxxxxxxx>
Date: Mon, 27 Mar 2017 08:09:36 -0700
Subject: [PATCH] regmap: irq: Fix lost interrupts by introducing
handle_reread
At least Motorola CPCAP PMIC needs it's device interrupts re-read
until there are no more interrupts. Otherwise the PMIC interrupt to
the SoC will eventually stop toggling. This seems to be a bug in the
CPCAP PMIC where it can stop driving the GPIO interrupt to the SoC
with pending CPCAP interrupts.
Let's allow handling issues like this by introducing a flag for
handle_reread and by splitting regmap_irq_thread() into two separate
functions for regmap_read_irq_status() and regmap_irq_handle_pending().
Cc: Charles Keepax <ckeepax@xxxxxxxxxxxxxxxxxxxxxxxxxxx>
Cc: Lee Jones <lee.jones@xxxxxxxxxx>
Cc: Marcel Partap <mpartap@xxxxxxx>
Cc: Michael Scott <michael.scott@xxxxxxxxxx>
Signed-off-by: Tony Lindgren <tony@xxxxxxxxxxx>
---
drivers/base/regmap/regmap-irq.c | 97 ++++++++++++++++++++++++++++++----------
include/linux/regmap.h | 2 +
2 files changed, 75 insertions(+), 24 deletions(-)
diff --git a/drivers/base/regmap/regmap-irq.c b/drivers/base/regmap/regmap-irq.c
--- a/drivers/base/regmap/regmap-irq.c
+++ b/drivers/base/regmap/regmap-irq.c
@@ -21,6 +21,8 @@
#include "internal.h"
+#define REGMAP_IRQ_THREAD_MAX_MSECS 3000
+
struct regmap_irq_chip_data {
struct mutex lock;
struct irq_chip irq_chip;
@@ -259,27 +261,11 @@ static const struct irq_chip regmap_irq_chip = {
.irq_set_wake = regmap_irq_set_wake,
};
-static irqreturn_t regmap_irq_thread(int irq, void *d)
+static int regmap_read_irq_status(struct regmap_irq_chip_data *data)
{
- struct regmap_irq_chip_data *data = d;
const struct regmap_irq_chip *chip = data->chip;
struct regmap *map = data->map;
int ret, i;
- bool handled = false;
- u32 reg;
-
- if (chip->handle_pre_irq)
- chip->handle_pre_irq(chip->irq_drv_data);
-
- if (chip->runtime_pm) {
- ret = pm_runtime_get_sync(map->dev);
- if (ret < 0) {
- dev_err(map->dev, "IRQ thread failed to resume: %d\n",
- ret);
- pm_runtime_put(map->dev);
- goto exit;
- }
- }
/*
* Read in the statuses, using a single bulk read if possible
@@ -299,7 +285,7 @@ static irqreturn_t regmap_irq_thread(int irq, void *d)
if (ret != 0) {
dev_err(map->dev, "Failed to read IRQ status: %d\n",
ret);
- goto exit;
+ return ret;
}
for (i = 0; i < data->chip->num_regs; i++) {
@@ -315,7 +301,7 @@ static irqreturn_t regmap_irq_thread(int irq, void *d)
break;
default:
BUG();
- goto exit;
+ return ret;
}
}
@@ -330,13 +316,21 @@ static irqreturn_t regmap_irq_thread(int irq, void *d)
dev_err(map->dev,
"Failed to read IRQ status: %d\n",
ret);
- if (chip->runtime_pm)
- pm_runtime_put(map->dev);
- goto exit;
+ return ret;
}
}
}
+ return 0;
+}
+
+static int regmap_irq_handle_pending(struct regmap_irq_chip_data *data)
+{
+ const struct regmap_irq_chip *chip = data->chip;
+ struct regmap *map = data->map;
+ int ret, i, handled = 0;
+ u32 reg;
+
/*
* Ignore masked IRQs and ack if we need to; we ack early so
* there is no race between handling and acknowleding the
@@ -361,14 +355,69 @@ static irqreturn_t regmap_irq_thread(int irq, void *d)
if (data->status_buf[chip->irqs[i].reg_offset /
map->reg_stride] & chip->irqs[i].mask) {
handle_nested_irq(irq_find_mapping(data->domain, i));
- handled = true;
+ handled++;
}
}
+ return handled;
+}
+
+static irqreturn_t regmap_irq_thread(int irq, void *d)
+{
+ struct regmap_irq_chip_data *data = d;
+ const struct regmap_irq_chip *chip = data->chip;
+ struct regmap *map = data->map;
+ ktime_t calltime = 0, retrytime;
+ int ret, handled = 0, retries = 0;
+
+ if (chip->handle_reread)
+ calltime = ktime_get();
+
+ if (chip->handle_pre_irq)
+ chip->handle_pre_irq(chip->irq_drv_data);
+
+ if (chip->runtime_pm) {
+ ret = pm_runtime_get_sync(map->dev);
+ if (ret < 0) {
+ dev_err(map->dev, "IRQ thread failed to resume: %d\n",
+ ret);
+ goto out_runtime_put;
+ }
+ }
+
+ do {
+ ret = regmap_read_irq_status(data);
+ if (ret)
+ goto out_runtime_put;
+
+ ret = regmap_irq_handle_pending(data);
+ if (ret < 0)
+ goto out_runtime_put;
+
+ if (!ret)
+ break;
+
+ handled += ret;
+ retries++;
+
+ if (chip->handle_reread) {
+ s64 msecs;
+
+ retrytime = ktime_get();
+ msecs = ktime_to_ms(ktime_sub(retrytime, calltime));
+ if (msecs > REGMAP_IRQ_THREAD_MAX_MSECS) {
+ dev_err(map->dev,
+ "IRQ thread timeout after %i retries\n",
+ retries);
+ goto out_runtime_put;
+ }
+ }
+ } while (chip->handle_reread);
+
+out_runtime_put:
if (chip->runtime_pm)
pm_runtime_put(map->dev);
-exit:
if (chip->handle_post_irq)
chip->handle_post_irq(chip->irq_drv_data);
diff --git a/include/linux/regmap.h b/include/linux/regmap.h
--- a/include/linux/regmap.h
+++ b/include/linux/regmap.h
@@ -897,6 +897,7 @@ struct regmap_irq {
* @ack_invert: Inverted ack register: cleared bits for ack.
* @wake_invert: Inverted wake register: cleared bits are wake enabled.
* @type_invert: Invert the type flags.
+ * @handle_reread: Read interrupt status until no more interrupts are seen.
* @runtime_pm: Hold a runtime PM lock on the device when accessing it.
*
* @num_regs: Number of registers in each control bank.
@@ -934,6 +935,7 @@ struct regmap_irq_chip {
bool wake_invert:1;
bool runtime_pm:1;
bool type_invert:1;
+ bool handle_reread:1;
int num_regs;
--
2.12.1
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
