[PATCH 1/1] soc: ti: knav_qmss_queue: Fix linking RAM setup for queue managers

[santosh.shilimkar at oracle.com]

On 11/20/15 9:04 PM, Michal Morawiec wrote:
> On Fri, Nov 20, 2015 at 03:47:38PM -0800, santosh shilimkar wrote:
>> On 11/20/2015 3:39 PM, Michal Morawiec wrote:
>>> Configure linking RAM for both queue managers also in case
>>> when only linking RAM 0 is specified in device tree.
>>>
>> why ?
> If both queue managers are used then both must be configured
> with valid linking RAM configuration independent of the number
> of linking RAMs used. If the configuration for the QM2 is
> missing there will be a crash when it tries push/pop descriptors
> from its queues. That's what I encountered once I removed
> linking RAM 1 from the device tree since it was not needed.
>
>>> Signed-off-by: Michal Morawiec <michal.1.morawiec.ext at nokia.com>
>>> ---
>>>   drivers/soc/ti/knav_qmss_queue.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/soc/ti/knav_qmss_queue.c b/drivers/soc/ti/knav_qmss_queue.c
>>> index 6d8646d..a809c30 100644
>>> --- a/drivers/soc/ti/knav_qmss_queue.c
>>> +++ b/drivers/soc/ti/knav_qmss_queue.c
>>> @@ -1173,7 +1173,7 @@ static int knav_queue_setup_link_ram(struct knav_device *kdev)
>>>
>>>   		block++;
>>>   		if (!block->size)
>>> -			return 0;
>>> +			continue;
>>>
>> You have to expand this a bit for me. Because you really don't
>> want kernel code to run the configuration on hardware which doesn't
>> exist. I mean device tree suppose to specify the linking RAM for both
>> QMs unless and until there is a reason not too.
> If I understand the current handling in the driver correctly
> the linking RAM(s) is/are used cooperatively by QMs (shared mode).
> So every linking RAM specified in the device tree must be
> configured exactly the same for both QMs (base addres and size).
> If there is only one linking RAM specified then only this one
> must be configured for both QMs.
> For proper operation only one linking RAM is required and in most
> cases this can be internal one as long as it is able to handle
> the number of descriptors used in the system.
> That was the case for me so I moved all regions so that they all
> fit into one internal linking RAM and removed entry for external
> linking RAM. Current driver code however skips configuration of
> second queue manager if second linking RAM is not specified.
>
> Referring to the configuration of the missing hardware you
> mentioned I don't think there is anything missing. It's just
> one resource less used by the HW (QMs).
>
> I hope this explains the intention of my patch.
>
It does. Can you please update the patch change log with above
description and I shall pick it up.

Thanks for explaining the problem.

Regards,
Santosh