diff mbox series

[v2] i2c: cadence: try reset when master receive arbitration lost

Message ID 20190219012447.5900-1-sxauwsk@163.com
State Rejected
Headers show
Series [v2] i2c: cadence: try reset when master receive arbitration lost | expand

Commit Message

sxauwsk@163.com Feb. 19, 2019, 1:24 a.m. UTC
When the adapter receive arbitration lost error interrupts,
cdns_i2c_master_xfer return to the caller directly instead of resetting
the adapter which resulted in the adapter being out of control.

So when driver detect err_status such as arbitration lost,
then try to repair and fix it.

Reviewed-by: Shubhrajyoti Datta <shubhrajyoti.datta@xilinx.com>
Signed-off-by: Shikai Wang <sxauwsk@163.com>
---
 drivers/i2c/busses/i2c-cadence.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

Comments

Wolfram Sang Feb. 23, 2019, 9:48 a.m. UTC | #1
On Tue, Feb 19, 2019 at 09:24:47AM +0800, Shikai Wang wrote:
> When the adapter receive arbitration lost error interrupts,
> cdns_i2c_master_xfer return to the caller directly instead of resetting
> the adapter which resulted in the adapter being out of control.
> 
> So when driver detect err_status such as arbitration lost,
> then try to repair and fix it.

Just checking because resetting looks like a big hammer to me: there is
no way to gracefully recover form 'arbitration lost' interrupt?
Wolfram Sang March 9, 2019, 9:44 a.m. UTC | #2
On Sat, Feb 23, 2019 at 10:48:43AM +0100, Wolfram Sang wrote:
> On Tue, Feb 19, 2019 at 09:24:47AM +0800, Shikai Wang wrote:
> > When the adapter receive arbitration lost error interrupts,
> > cdns_i2c_master_xfer return to the caller directly instead of resetting
> > the adapter which resulted in the adapter being out of control.
> > 
> > So when driver detect err_status such as arbitration lost,
> > then try to repair and fix it.
> 
> Just checking because resetting looks like a big hammer to me: there is
> no way to gracefully recover form 'arbitration lost' interrupt?

I am still wondering about this but since it is acked by the maintainer:

Applied to for-current, thanks!
Wolfram Sang March 9, 2019, 9:49 a.m. UTC | #3
On Sat, Mar 09, 2019 at 10:44:52AM +0100, Wolfram Sang wrote:
> On Sat, Feb 23, 2019 at 10:48:43AM +0100, Wolfram Sang wrote:
> > On Tue, Feb 19, 2019 at 09:24:47AM +0800, Shikai Wang wrote:
> > > When the adapter receive arbitration lost error interrupts,
> > > cdns_i2c_master_xfer return to the caller directly instead of resetting
> > > the adapter which resulted in the adapter being out of control.
> > > 
> > > So when driver detect err_status such as arbitration lost,
> > > then try to repair and fix it.
> > 
> > Just checking because resetting looks like a big hammer to me: there is
> > no way to gracefully recover form 'arbitration lost' interrupt?
> 
> I am still wondering about this but since it is acked by the maintainer:
> 
> Applied to for-current, thanks!

Oops, I was mixing this up with

[PATCHv4] i2c: cadence: Fix the driver in interrupt flurry case

This one doesn't have an ack from Michal. Dropping it again.
Shubhrajyoti Datta Nov. 5, 2019, 7:42 a.m. UTC | #4
Hi Shikai,

On Tue, Feb 19, 2019 at 8:19 AM Shikai Wang <sxauwsk@163.com> wrote:
>
> When the adapter receive arbitration lost error interrupts,
> cdns_i2c_master_xfer return to the caller directly instead of resetting
> the adapter which resulted in the adapter being out of control.
>
> So when driver detect err_status such as arbitration lost,
> then try to repair and fix it.
>
I am missing the issue that you are facing.
You are having a multimaster scenario and getting arbitration lost.

the current code would attempt a retry did that lead to any issues?

Can you explain the issue that you are facing?
sxauwsk@163.com Nov. 5, 2019, 8:47 a.m. UTC | #5
>Hi Shikai,
>
>On Tue, Feb 19, 2019 at 8:19 AM Shikai Wang <sxauwsk@163.com> wrote:
>>
>> When the adapter receive arbitration lost error interrupts,
>> cdns_i2c_master_xfer return to the caller directly instead of resetting
>> the adapter which resulted in the adapter being out of control.
>>
>> So when driver detect err_status such as arbitration lost,
>> then try to repair and fix it.
>>
>I am missing the issue that you are facing.
>You are having a multimaster scenario and getting arbitration lost.
>
>the current code would attempt a retry did that lead to any issues?
>
>Can you explain the issue that you are facing? 

Of cource,  The following describe my situation.

In my product,  Touchscreen connect to zynq-7000 XC7Z010 by i2c bus( Just connect only one i2c-device of touchscreen),  
when user tap Touchscreen, Touchscreen interrupt send to CPU and notifyed i2c-driver to obtain location data by i2c-bus,  

when Tap the screen frequently,  sometimes CPU get interrupt from touchscreen and try to obtain data,  then detect arbitration lost,
Although i2c-driver try three times,  it's useless.
  
Actually i2c clock-line and data-line keep high, that mean i2c bus free.
Once this situation occur, i2c-control did't work anynay but cpu receive interrputs still.

I am sorry that I have't found a good solution for this issuse;
Shubhrajyoti Datta Nov. 7, 2019, 6:13 a.m. UTC | #6
Hi Shikai,

On Tue, Nov 5, 2019 at 2:18 PM sxauwsk@163.com <sxauwsk@163.com> wrote:
>
> >Hi Shikai,
> >
> >On Tue, Feb 19, 2019 at 8:19 AM Shikai Wang <sxauwsk@163.com> wrote:
> >>
> >> When the adapter receive arbitration lost error interrupts,
> >> cdns_i2c_master_xfer return to the caller directly instead of resetting
> >> the adapter which resulted in the adapter being out of control.
> >>
> >> So when driver detect err_status such as arbitration lost,
> >> then try to repair and fix it.
> >>
> >I am missing the issue that you are facing.
> >You are having a multimaster scenario and getting arbitration lost.
> >
> >the current code would attempt a retry did that lead to any issues?
> >
> >Can you explain the issue that you are facing?
>
> Of cource,  The following describe my situation.
>
> In my product,  Touchscreen connect to zynq-7000 XC7Z010 by i2c bus( Just connect only one i2c-device of touchscreen),
> when user tap Touchscreen, Touchscreen interrupt send to CPU and notifyed i2c-driver to obtain location data by i2c-bus,

So it is single master single slave.
>
> when Tap the screen frequently,  sometimes CPU get interrupt from touchscreen and try to obtain data,  then detect arbitration lost,
the arbitration lost is surprising in non-multimaster scenario.
Is there any other master in the configuration that we may not be triggering.
Or can  you probe the lines?

> Although i2c-driver try three times,  it's useless.

You get bus busy? what is the issue.
>
> Actually i2c clock-line and data-line keep high, that mean i2c bus free.
> Once this situation occur, i2c-control did't work anynay but cpu receive interrputs still.
>
> I am sorry that I have't found a good solution for this issuse;
sxauwsk@163.com Nov. 8, 2019, 12:25 p.m. UTC | #7
>Hi Shikai,
>
>On Tue, Nov 5, 2019 at 2:18 PM sxauwsk@163.com <sxauwsk@163.com> wrote:
>>
>> >Hi Shikai,
>> >
>> >On Tue, Feb 19, 2019 at 8:19 AM Shikai Wang <sxauwsk@163.com> wrote:
>> >>
>> >> When the adapter receive arbitration lost error interrupts,
>> >> cdns_i2c_master_xfer return to the caller directly instead of resetting
>> >> the adapter which resulted in the adapter being out of control.
>> >>
>> >> So when driver detect err_status such as arbitration lost,
>> >> then try to repair and fix it.
>> >>
>> >I am missing the issue that you are facing.
>> >You are having a multimaster scenario and getting arbitration lost.
>> >
>> >the current code would attempt a retry did that lead to any issues?
>> >
>> >Can you explain the issue that you are facing?
>>
>> Of cource,  The following describe my situation.
>>
>> In my product,  Touchscreen connect to zynq-7000 XC7Z010 by i2c bus( Just connect only one i2c-device of touchscreen),
>> when user tap Touchscreen, Touchscreen interrupt send to CPU and notifyed i2c-driver to obtain location data by i2c-bus,
>
>So it is single master single slave.
>>
>> when Tap the screen frequently,  sometimes CPU get interrupt from touchscreen and try to obtain data,  then detect arbitration lost,
>the arbitration lost is surprising in non-multimaster scenario.
>Is there any other master in the configuration that we may not be triggering.
>Or can  you probe the lines?
>
>> Although i2c-driver try three times,  it's useless.
>
>You get bus busy? what is the issue.
>>
>> Actually i2c clock-line and data-line keep high, that mean i2c bus free.
>> Once this situation occur, i2c-control did't work anynay but cpu receive interrputs still.
>>
>> I am sorry that I have't found a good solution for this issuse; 

Yes,  I can't believe that,  Maybe caused by the hardware environment or any other well.
Additionally I have't  probe this issue recently.
diff mbox series

Patch

diff --git a/drivers/i2c/busses/i2c-cadence.c b/drivers/i2c/busses/i2c-cadence.c
index b13605718291..595b0d56ff1a 100644
--- a/drivers/i2c/busses/i2c-cadence.c
+++ b/drivers/i2c/busses/i2c-cadence.c
@@ -548,10 +548,6 @@  static int cdns_i2c_process_msg(struct cdns_i2c *id, struct i2c_msg *msg,
 	cdns_i2c_writereg(CDNS_I2C_IXR_ALL_INTR_MASK,
 			  CDNS_I2C_IDR_OFFSET);
 
-	/* If it is bus arbitration error, try again */
-	if (id->err_status & CDNS_I2C_IXR_ARB_LOST)
-		return -EAGAIN;
-
 	return 0;
 }
 
@@ -624,6 +620,12 @@  static int cdns_i2c_master_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs,
 		if (id->err_status) {
 			cdns_i2c_master_reset(adap);
 
+			/* If it is bus arbitration error, try again */
+			if (id->err_status & CDNS_I2C_IXR_ARB_LOST) {
+				ret = -EAGAIN;
+				goto out;
+			}
+
 			if (id->err_status & CDNS_I2C_IXR_NACK) {
 				ret = -ENXIO;
 				goto out;