diff mbox

[v3,09/11] i2c: rcar: revoke START request early

Message ID 1447948611-2615-10-git-send-email-wsa@the-dreams.de (mailing list archive)
State Accepted
Delegated to: Geert Uytterhoeven
Headers show

Commit Message

Wolfram Sang Nov. 19, 2015, 3:56 p.m. UTC
From: Wolfram Sang <wsa+renesas@sang-engineering.com>

If we don't clear START generation as soon as possible, it may cause
another message to be generated, e.g. when receiving NACK in address
phase. To keep the race window as small as possible, we clear it right
at the beginning of the interrupt. We don't need any checks since we
always want to stop START and STOP generation on the next occasion after
we started it.

This patch improves the situation but sadly does not completely fix it.
It is still to be researched if we can do better given this HW design.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
---
 drivers/i2c/busses/i2c-rcar.c | 23 +++++++----------------
 1 file changed, 7 insertions(+), 16 deletions(-)

Comments

Sergei Shtylyov March 31, 2016, 9:02 p.m. UTC | #1
Hello.

On 11/19/2015 06:56 PM, Wolfram Sang wrote:

> From: Wolfram Sang <wsa+renesas@sang-engineering.com>
>
> If we don't clear START generation as soon as possible, it may cause
> another message to be generated, e.g. when receiving NACK in address
> phase. To keep the race window as small as possible, we clear it right
> at the beginning of the interrupt. We don't need any checks since we
> always want to stop START and STOP generation on the next occasion after
> we started it.
>
> This patch improves the situation but sadly does not completely fix it.
> It is still to be researched if we can do better given this HW design.
>
> Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

    Thanks for a great work, Wolfram!
    We need this patch in -stable kernels. The R-Car audio just doesn't work 
without it...

> ---
>   drivers/i2c/busses/i2c-rcar.c | 23 +++++++----------------
>   1 file changed, 7 insertions(+), 16 deletions(-)
>
> diff --git a/drivers/i2c/busses/i2c-rcar.c b/drivers/i2c/busses/i2c-rcar.c
> index f237b4fc5b5e55..40979130200935 100644
> --- a/drivers/i2c/busses/i2c-rcar.c
> +++ b/drivers/i2c/busses/i2c-rcar.c
> @@ -83,6 +83,7 @@
>
>   #define RCAR_BUS_PHASE_START	(MDBS | MIE | ESG)
>   #define RCAR_BUS_PHASE_DATA	(MDBS | MIE)
> +#define RCAR_BUS_MASK_DATA	(~(ESG | FSB) & 0xFF)
>   #define RCAR_BUS_PHASE_STOP	(MDBS | MIE | FSB)
>
>   #define RCAR_IRQ_SEND	(MNR | MAL | MST | MAT | MDE)
> @@ -289,13 +290,6 @@ static void rcar_i2c_irq_send(struct rcar_i2c_priv *priv, u32 msr)
>   	if (!(msr & MDE))
>   		return;
>
> -	/*
> -	 * If address transfer phase finished,
> -	 * goto data phase.
> -	 */
> -	if (msr & MAT)
> -		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
> -
>   	if (priv->pos < msg->len) {
>   		/*
>   		 * Prepare next data to ICRXTX register.
> @@ -345,11 +339,7 @@ static void rcar_i2c_irq_recv(struct rcar_i2c_priv *priv, u32 msr)
>   		return;
>
>   	if (msr & MAT) {
> -		/*
> -		 * Address transfer phase finished,
> -		 * but, there is no data at this point.
> -		 * Do nothing.
> -		 */
> +		/* Address transfer phase finished, but no data at this point. */
>   	} else if (priv->pos < msg->len) {
>   		/*
>   		 * get received data
> @@ -365,8 +355,6 @@ static void rcar_i2c_irq_recv(struct rcar_i2c_priv *priv, u32 msr)
>   	 */
>   	if (priv->pos + 1 >= msg->len)
>   		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_STOP);
> -	else
> -		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
>
>   	if (priv->pos == msg->len && !(priv->flags & ID_LAST_MSG))
>   		rcar_i2c_next_msg(priv);
> @@ -432,7 +420,11 @@ static bool rcar_i2c_slave_irq(struct rcar_i2c_priv *priv)
>   static irqreturn_t rcar_i2c_irq(int irq, void *ptr)
>   {
>   	struct rcar_i2c_priv *priv = ptr;
> -	u32 msr;
> +	u32 msr, val;
> +
> +	/* Clear START or STOP as soon as we can */
> +	val = rcar_i2c_read(priv, ICMCR);
> +	rcar_i2c_write(priv, ICMCR, val & RCAR_BUS_MASK_DATA);
>
>   	msr = rcar_i2c_read(priv, ICMSR);
>
> @@ -454,7 +446,6 @@ static irqreturn_t rcar_i2c_irq(int irq, void *ptr)
>   	/* Nack */
>   	if (msr & MNR) {
>   		/* HW automatically sends STOP after received NACK */
> -		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
>   		rcar_i2c_write(priv, ICMIER, RCAR_IRQ_STOP);
>   		rcar_i2c_flags_set(priv, ID_NACK);
>   		goto out;

MBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang March 31, 2016, 10:48 p.m. UTC | #2
On Fri, Apr 01, 2016 at 12:02:56AM +0300, Sergei Shtylyov wrote:
> Hello.
> 
> On 11/19/2015 06:56 PM, Wolfram Sang wrote:
> 
> >From: Wolfram Sang <wsa+renesas@sang-engineering.com>
> >
> >If we don't clear START generation as soon as possible, it may cause
> >another message to be generated, e.g. when receiving NACK in address
> >phase. To keep the race window as small as possible, we clear it right
> >at the beginning of the interrupt. We don't need any checks since we
> >always want to stop START and STOP generation on the next occasion after
> >we started it.
> >
> >This patch improves the situation but sadly does not completely fix it.
> >It is still to be researched if we can do better given this HW design.
> >
> >Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
> 
>    Thanks for a great work, Wolfram!
>    We need this patch in -stable kernels. The R-Car audio just doesn't work
> without it...

Really only this patch? IIRC my tests showed that if you don't remove
the spinlocks (patch 4), the interrupt latency will already be too high
again. In any case, you'd need to do some careful backporting to rip
this out of the whole refactoring series. But maybe you did that already
and have good experiences?
Sergei Shtylyov April 1, 2016, 7:55 p.m. UTC | #3
Hello.

On 04/01/2016 01:48 AM, Wolfram Sang wrote:

>>> From: Wolfram Sang <wsa+renesas@sang-engineering.com>
>>>
>>> If we don't clear START generation as soon as possible, it may cause
>>> another message to be generated, e.g. when receiving NACK in address
>>> phase. To keep the race window as small as possible, we clear it right
>>> at the beginning of the interrupt. We don't need any checks since we
>>> always want to stop START and STOP generation on the next occasion after
>>> we started it.
>>>
>>> This patch improves the situation but sadly does not completely fix it.
>>> It is still to be researched if we can do better given this HW design.
>>>
>>> Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
>>
>>     Thanks for a great work, Wolfram!
>>     We need this patch in -stable kernels. The R-Car audio just doesn't work
>> without it...

> Really only this patch?

    Well, my "reverse" bisection pointed at it. :-)

> IIRC my tests showed that if you don't remove
> the spinlocks (patch 4), the interrupt latency will already be too high
> again.

    Thank you for the valuable info!

> In any case, you'd need to do some careful backporting to rip
> this out of the whole refactoring series.

    Yes, I've already figured that.

> But maybe you did that already
> and have good experiences?

    Not yet, I will report back after more backporting/testing.

MBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang April 1, 2016, 8:14 p.m. UTC | #4
> >IIRC my tests showed that if you don't remove
> >the spinlocks (patch 4), the interrupt latency will already be too high
> >again.
> 
>    Thank you for the valuable info!

Oh, and add patch 6 to it. The context switches are a major problem in
this.
Sergei Shtylyov April 1, 2016, 11:05 p.m. UTC | #5
On 04/01/2016 11:14 PM, Wolfram Sang wrote:
>
>>> IIRC my tests showed that if you don't remove
>>> the spinlocks (patch 4), the interrupt latency will already be too high
>>> again.
>>
>>     Thank you for the valuable info!
>
> Oh, and add patch 6 to it. The context switches are a major problem in
> this.

    Thanks again! I ended up backporting all the patches up to #9. Not sure 
it'd be acceptable for the -stable kernels since there's a lot of refactors 
involved...

MBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang April 2, 2016, 7:27 a.m. UTC | #6
>    Thanks again! I ended up backporting all the patches up to #9. Not sure
> it'd be acceptable for the -stable kernels since there's a lot of refactors
> involved...

That's what I thought, too.
Sergei Shtylyov April 2, 2016, 12:51 p.m. UTC | #7
On 4/2/2016 10:27 AM, Wolfram Sang wrote:

>>     Thanks again! I ended up backporting all the patches up to #9. Not sure
>> it'd be acceptable for the -stable kernels since there's a lot of refactors
>> involved...
>
> That's what I thought, too.

    And I still don't feel good about the patch removing the spinlock -- I'm 
not sure it's SMP-safe...

MBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang April 3, 2016, 8:25 a.m. UTC | #8
>    And I still don't feel good about the patch removing the spinlock -- I'm
> not sure it's SMP-safe...

In what scenario?
Sergei Shtylyov April 3, 2016, 1:35 p.m. UTC | #9
On 04/03/2016 11:25 AM, Wolfram Sang wrote:

>>     And I still don't feel good about the patch removing the spinlock -- I'm
>> not sure it's SMP-safe...
>
> In what scenario?

    I've reviewed the patches once again and it looked like they are safe.

MBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang April 3, 2016, 1:47 p.m. UTC | #10
> >>    And I still don't feel good about the patch removing the spinlock -- I'm
> >>not sure it's SMP-safe...
> >
> >In what scenario?
> 
>    I've reviewed the patches once again and it looked like they are safe.

Thanks.
diff mbox

Patch

diff --git a/drivers/i2c/busses/i2c-rcar.c b/drivers/i2c/busses/i2c-rcar.c
index f237b4fc5b5e55..40979130200935 100644
--- a/drivers/i2c/busses/i2c-rcar.c
+++ b/drivers/i2c/busses/i2c-rcar.c
@@ -83,6 +83,7 @@ 
 
 #define RCAR_BUS_PHASE_START	(MDBS | MIE | ESG)
 #define RCAR_BUS_PHASE_DATA	(MDBS | MIE)
+#define RCAR_BUS_MASK_DATA	(~(ESG | FSB) & 0xFF)
 #define RCAR_BUS_PHASE_STOP	(MDBS | MIE | FSB)
 
 #define RCAR_IRQ_SEND	(MNR | MAL | MST | MAT | MDE)
@@ -289,13 +290,6 @@  static void rcar_i2c_irq_send(struct rcar_i2c_priv *priv, u32 msr)
 	if (!(msr & MDE))
 		return;
 
-	/*
-	 * If address transfer phase finished,
-	 * goto data phase.
-	 */
-	if (msr & MAT)
-		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
-
 	if (priv->pos < msg->len) {
 		/*
 		 * Prepare next data to ICRXTX register.
@@ -345,11 +339,7 @@  static void rcar_i2c_irq_recv(struct rcar_i2c_priv *priv, u32 msr)
 		return;
 
 	if (msr & MAT) {
-		/*
-		 * Address transfer phase finished,
-		 * but, there is no data at this point.
-		 * Do nothing.
-		 */
+		/* Address transfer phase finished, but no data at this point. */
 	} else if (priv->pos < msg->len) {
 		/*
 		 * get received data
@@ -365,8 +355,6 @@  static void rcar_i2c_irq_recv(struct rcar_i2c_priv *priv, u32 msr)
 	 */
 	if (priv->pos + 1 >= msg->len)
 		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_STOP);
-	else
-		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
 
 	if (priv->pos == msg->len && !(priv->flags & ID_LAST_MSG))
 		rcar_i2c_next_msg(priv);
@@ -432,7 +420,11 @@  static bool rcar_i2c_slave_irq(struct rcar_i2c_priv *priv)
 static irqreturn_t rcar_i2c_irq(int irq, void *ptr)
 {
 	struct rcar_i2c_priv *priv = ptr;
-	u32 msr;
+	u32 msr, val;
+
+	/* Clear START or STOP as soon as we can */
+	val = rcar_i2c_read(priv, ICMCR);
+	rcar_i2c_write(priv, ICMCR, val & RCAR_BUS_MASK_DATA);
 
 	msr = rcar_i2c_read(priv, ICMSR);
 
@@ -454,7 +446,6 @@  static irqreturn_t rcar_i2c_irq(int irq, void *ptr)
 	/* Nack */
 	if (msr & MNR) {
 		/* HW automatically sends STOP after received NACK */
-		rcar_i2c_write(priv, ICMCR, RCAR_BUS_PHASE_DATA);
 		rcar_i2c_write(priv, ICMIER, RCAR_IRQ_STOP);
 		rcar_i2c_flags_set(priv, ID_NACK);
 		goto out;