Skip to contentSkip to conversation

input: touch: eeti: move ISR code to own function

PATCH In review Daniel Mack 2 participants 11 messages Apr 22, 2019 – Apr 29, 2019 Input drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 input: touch: eeti: move ISR code to own function

Move the ISR handling code to its own function and change the logic to bail immediately in case of .running is false or if the attn_gpio is available but unasserted.

This allows us to call the function at any time to check for the state of attn_gpio.

---
 drivers/input/touchscreen/eeti_ts.c | 30 +++++++++++++++++++----------
 1 file changed, 20 insertions(+), 10 deletions(-)
drivers/input/touchscreen/eeti_ts.c +20 -10
index 7fe41965c5d1..f5724aaa815b 100644
@@ -75,14 +75,19 @@ static void eeti_ts_report_event(struct eeti_ts *eeti, u8 *buf)
 	input_sync(eeti->input);
 }
 
-static irqreturn_t eeti_ts_isr(int irq, void *dev_id)
+static void eeti_ts_read(struct eeti_ts *eeti)
 {
-	struct eeti_ts *eeti = dev_id;
-	int len;
-	int error;
+	int len, error;
 	char buf[6];
 
-	do {
+	for (;;) {
+		if (!eeti->running)
+			break;
+
+		if (eeti->attn_gpio &&
+		    gpiod_get_value_cansleep(eeti->attn_gpio) == 0)
+			break;
+
 		len = i2c_master_recv(eeti->client, buf, sizeof(buf));
 		if (len != sizeof(buf)) {
 			error = len < 0 ? len : -EIO;
@@ -92,12 +97,17 @@ static irqreturn_t eeti_ts_isr(int irq, void *dev_id)
 			break;
 		}
 
-		if (buf[0] & 0x80) {
-			/* Motion packet */
+		/* Motion packet */
+		if (buf[0] & 0x80)
 			eeti_ts_report_event(eeti, buf);
-		}
-	} while (eeti->running &&
-		 eeti->attn_gpio && gpiod_get_value_cansleep(eeti->attn_gpio));
+	}
+}
+
+static irqreturn_t eeti_ts_isr(int irq, void *dev_id)
+{
+	struct eeti_ts *eeti = dev_id;
+
+	eeti_ts_read(eeti);
 
 	return IRQ_HANDLED;
 }
Signature · 1 line
2.20.1
#1 lore raw

PATCH 2/2 input: touch: eeti: read hardware state once after wakeup

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

---
 drivers/input/touchscreen/eeti_ts.c | 1 +
 1 file changed, 1 insertion(+)
drivers/input/touchscreen/eeti_ts.c +1 -0
index f5724aaa815b..674386f910ba 100644
@@ -117,6 +117,7 @@ static void eeti_ts_start(struct eeti_ts *eeti)
 	eeti->running = true;
 	wmb();
 	enable_irq(eeti->client->irq);
+	eeti_ts_read(eeti);
 }
 
 static void eeti_ts_stop(struct eeti_ts *eeti)
Signature · 1 line
2.20.1
maintainer
#2 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Hi Daniel,

Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

Thanks.

Signature · 1 line
Dmitry
#3 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Hi Dmitry,

Quoted 10 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

Thanks,
Daniel

maintainer
#4 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 12 lines · Daniel Mack

Hi Dmitry,

Quoted 9 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Quoted 6 lines
Quoted 1 line

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

Thanks.

Signature · 1 line
Dmitry
#5 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 15 lines · Dmitry Torokhov
Quoted 14 lines · Daniel Mack

Hi Dmitry,

Quoted 10 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

Quoted 7 lines
Quoted 4 lines

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

Thanks,
Daniel

maintainer
#6 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 18 lines · Daniel Mack
Quoted 17 lines · Dmitry Torokhov
Quoted 16 lines · Daniel Mack

Hi Dmitry,

Quoted 10 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

I was about to, but now I wonder if we need a mutex in the isr code now, otherwise there is a chance it will be running concurrently when we are resuming if interrupt does latch.

Thanks.

Signature · 1 line
Dmitry
#7 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 25 lines · Dmitry Torokhov
Quoted 24 lines · Daniel Mack
Quoted 22 lines · Dmitry Torokhov
Quoted 12 lines · Daniel Mack

Hi Dmitry,

Quoted 9 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Quoted 6 lines
Quoted 1 line

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

Quoted 6 lines
Quoted 2 lines

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

I was about to, but now I wonder if we need a mutex in the isr code now, otherwise there is a chance it will be running concurrently when we are resuming if interrupt does latch.

Ah, good point. I'll send a v2.

Thanks,
Daniel

#8 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

In reply to Dmitry Torokhov
Quoted 25 lines · Dmitry Torokhov
Quoted 24 lines · Daniel Mack
Quoted 22 lines · Dmitry Torokhov
Quoted 12 lines · Daniel Mack

Hi Dmitry,

Quoted 9 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Quoted 6 lines
Quoted 1 line

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

Quoted 6 lines
Quoted 2 lines

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

I was about to, but now I wonder if we need a mutex in the isr code now, otherwise there is a chance it will be running concurrently when we are resuming if interrupt does latch.

Actually, to address that, all we need to do is call into eeti_ts_read() before enable_irq(). See my v2.

Thanks,
Daniel

maintainer
#9 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 29 lines · Daniel Mack
Quoted 28 lines · Dmitry Torokhov
Quoted 26 lines · Daniel Mack
Quoted 15 lines · Dmitry Torokhov
Quoted 14 lines · Daniel Mack

Hi Dmitry,

Quoted 10 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

Quoted 7 lines
Quoted 4 lines

The hardware I'm working on doesn't support that unfortunately.

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

Quoted 6 lines
Quoted 3 lines

I was about to, but now I wonder if we need a mutex in the isr code now, otherwise there is a chance it will be running concurrently when we are resuming if interrupt does latch.

Actually, to address that, all we need to do is call into eeti_ts_read() before enable_irq(). See my v2.

This is still a bit racy as interrupt may come after you attempted to read but before enable_irq() and then we need interrupt replaying code to work reliably, which, as far as I understand, is not always the case. I suppose we need at least add a comment that we rely on replays.

What kind of hardware is that? Is there no chance of making interrupt controller handle level interrupts?

Thanks.

Signature · 1 line
Dmitry
#10 lore raw

Re: [PATCH 2/2] input: touch: eeti: read hardware state once after wakeup

Quoted 34 lines · Dmitry Torokhov
Quoted 33 lines · Daniel Mack
Quoted 31 lines · Dmitry Torokhov
Quoted 18 lines · Daniel Mack
Quoted 17 lines · Dmitry Torokhov
Quoted 16 lines · Daniel Mack

Hi Dmitry,

Quoted 10 lines · Dmitry Torokhov
Quoted 7 lines · Daniel Mack

For systems in which the touch IRQ is acting as wakeup source, the interrupt controller might not latch the GPIO IRQ during sleep. In such cases, the interrupt will never occur again after resume, hence the touch screen appears dead.

To fix this, call into eeti_ts_read() once to read the hardware status and to arm the IRQ again.

Can you instead make the interrupt level-triggered?

The hardware I'm working on doesn't support that unfortunately.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

In fact, the whole attn-gpio dance is there because of that, and the GPIO descriptor maps to the same pin that also causes the IRQ in my case.

OK, if the interrupt controller is incapable of dealing with level interrupts then we have to do what you propose.

So you consider these patches for inclusion then? I'm just asking because I can't see them in your tree yet.

I was about to, but now I wonder if we need a mutex in the isr code now, otherwise there is a chance it will be running concurrently when we are resuming if interrupt does latch.

Quoted 7 lines
Quoted 2 lines

Actually, to address that, all we need to do is call into eeti_ts_read() before enable_irq(). See my v2.

This is still a bit racy as interrupt may come after you attempted to read but before enable_irq() and then we need interrupt replaying code to work reliably, which, as far as I understand, is not always the case. I suppose we need at least add a comment that we rely on replays.

Ah, of course. Okay, let's just have a mutex for this then. I'll send a v3.

Quoted 2 lines

What kind of hardware is that? Is there no chance of making interrupt controller handle level interrupts?

It's a PXA3xx platform, and their interrupt controller lacks such a feature.

Thanks,
Daniel