Re: [PATCH] i2c-cadence: Do not let signals interrupt I2C transfers

[Mike Looijmans <mike.looijmans-Oq418RWZeHk@public.gmane.org>]

On 03/11/2014 05:49 AM, Suneel Garapati wrote:
> Hi Mike/Soren,
>
>> 

Met vriendelijke groet / kind regards,

Mike Looijmans

TOPIC Embedded Systems
Eindhovenseweg 32-C, NL-5683 KH Best
Postbus 440, NL-5680 AK Best
Telefoon: (+31) (0) 499 33 69 79
Telefax:  (+31) (0) 499 33 69 70
E-mail: mike.looijmans-Oq418RWZeHk@public.gmane.org
Website: www.topic.nl

Please consider the environment before printing this e-mail

-----Original Message-----
>> From: Sören Brinkmann [mailto:soren.brinkmann-gjFFaj9aHVfQT0dZR+AlfA@public.gmane.org]
>> Sent: Tuesday, March 11, 2014 03:30
>> To: Mike Looijmans; Michal Simek
>> Cc: git; wsa-z923LK4zBo2bacvFa/9K2g@public.gmane.org; linux-i2c-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-
>> kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>> Subject: Re: [PATCH] i2c-cadence: Do not let signals interrupt I2C transfers
>>
>> Hi Mike,
>>
>> The cadence driver is not in mainline yet. I think for our vendor tree we can
>> pretty much take it this way.
>> Regarding getting this into mainline, I'll send another iteration of the change
>> set and include these changes.
>>
>> On Mon, 2014-03-10 at 12:12PM +0100, Mike Looijmans wrote:
>>> Pressing CTRL-C while communicating with an I2C device leads to
>>> erratic behaviour. The cause is that the controller will interrupt the
>>> I2C transfer in progress, and leave the client device in an undefined
>>> state. Many drivers do not handle error return codes on I2C transfers.
>>> The calling driver has no way of telling how much of the transfer has
>>> actually completed, so it cannot reliably determine the device's state.
>>>
>>> The best solution here is to not handle signals in the I2C bus driver
>>> at all, but always complete a transaction before returning control.
>>>
>>> See for a related patch and discussion on this topic:
>>> http://lkml.org/lkml/2014/1/9/246
>>
>> Can we get your Signed-off-by, please?

Sure, I forgot the -s, will add it in the next version.

>>
>>> ---
>>>   drivers/i2c/busses/i2c-cadence.c |   12 ++++--------
>>>   1 file changed, 4 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/drivers/i2c/busses/i2c-cadence.c
>>> b/drivers/i2c/busses/i2c-cadence.c
>>> index 86713d6..32ce2ee 100644
>>> --- a/drivers/i2c/busses/i2c-cadence.c
>>> +++ b/drivers/i2c/busses/i2c-cadence.c
>>> @@ -452,16 +452,12 @@ static int cdns_i2c_process_msg(struct cdns_i2c
>> *id, struct i2c_msg *msg,
>>>                      cdns_i2c_msend(id);
>>>
>>>              /* Wait for the signal of completion */
>>> -           ret = wait_for_completion_interruptible_timeout(
>>> -                                                   &id->xfer_done, HZ);
>>> -           if (ret < 1) {
>>> +           ret = wait_for_completion_timeout(&id->xfer_done, HZ);
>>> +           if (ret == 0) {
>>
>> To match the style used throughout the file this should just be
>>        if (!ret) {

Will do.

> Instead of discarding Ctrl+C, Can we wait until the current msg transfer completes
> [to avoid client undefined state or re-init the host to a known state]
>
> I see da-vinci driver handling in a similar way.

My patch for the davinci should have completely ignored the signals too. I 
haven't come across a driver that actually calls the xfer function with more 
than one item anyway. If drivers check the result of the xfer at all, they 
usually don't have a clue how to handle partial transfers. Many drivers just 
ignore the return code completely and don't even report failure.

The gain is a stable I2C system. The loss here is something like <1ms more 
latency to interrupt a process in the case it was just transferring I2C data. 
 From experience I can tell that the gain far outweighs that loss.

Mike.