lkml.org 
[lkml]   [2023]   [Jul]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v9 06/10] usb: dwc3: qcom: Add support to read IRQ's related to multiport
From


On 7/21/2023 2:05 PM, Johan Hovold wrote:
> On Sun, Jul 16, 2023 at 12:31:05AM +0530, Krishna Kurapati PSSNV wrote:
>> On 7/14/2023 4:10 PM, Krishna Kurapati PSSNV wrote:
>>> On 7/14/2023 2:35 PM, Johan Hovold wrote:
>
>>>> I haven't had time to look at your latest replies yet, but as I already
>>>> said when reviewing v9, it seems you should be using a common helper for
>>>> non-mp and mp.
>
>>>  The gist of my mail was to see if I can defer qcom probe when dwc3
>>> probe fails/or doesn't happen on of_plat_pop (which is logical) so that
>>> we can move setup_irq to after dwc3_register_core so that we know
>>> whether we are MP capable or not. This would help us move all IRQ
>>> reading into one function.
>
>> I see it is difficult to write a common helper. To do so, we need to
>> know whether the device is MP capable or not in advance. And since it is
>> not possible to know it before of_plat_pop is done, I see only few ways
>> to do it:
>>
>> 1. Based on qcom node compatible string, I can read whether the device
>> is MP capable or not and get IRQ's accordingly.
>
> See, it's not impossible. You can also determine whether you have a
> multiport controller from looking at the interrupt names which are
> indexed and distinct for MP.
>
>> 2. Read the port_info in advance but it needs me to go through some DT
>> props and try getting this info. Or read xhci regs like we are doing in
>> core (which is not good). Also since some Dt props can be missing, is it
>> difficult to get the MP capability info before of_plat_pop is done.
>
> That seem unnecessary currently, but long term we probably need to fix
> the design of this driver and defer some setup using callbacks that are
> called when the core driver probes. Perhaps now is the time to add such
> functionality.
>
>> 3. Remove IRQ handling completely. Just because the device has IRQ's
>> present, I don't see a point in adding them to bindings, and because we
>> added them to bindings, we are making a patch to read them (and since
>> this is a little challenging, the whole of multiport series is blocked
>> although I don't need wakeup support on these interrupts right away).
>
> Again, no. The devicetree binding should describe the hardware
> capabilities and that has nothing to do with whether you need this for
> you current project or not.
>
>> Can't we let the rest of the patches go through and let interrupt
>> handling for 2nd, 3rd and 4rth ports be taken care later ? I am asking
>> this because I want the rest of the patches which are in good shape now
>> (after fixing the nits mentioned) to get merged atleast. I will make
>> sure to add interrupt handling later in a different series once this is
>> merged once I send v10.
>
> As I've explained in earlier mails, I don't think that is acceptable as
> you'd be dumping your technical debt on the community which will be left
> to clean up your mess.
>
>> Or if there is a simpler way to do it, I would be happy to take any
>> suggestions and complete this missing part in this series itself.
>

Hi Johan,

Thanks for these comments.

> Using the 'compatible' or 'interrupt-names' properties seems like the
> easiest way to determine whether you have an MP controller or not.
>

Yes, I can make a common helper to get IRQ's based on compatible. I also
provided another implementation (which is more unambiguous and better I
feel) on [1]. I will take one path forward based on your review of that
patch as well.

Thanks a lot again for the reviews !

[1]:
https://lore.kernel.org/all/f6f2456d-0067-6cd6-3282-8cae7c47a2d3@quicinc.com/

Regards,
Krishna,

\
 
 \ /
  Last update: 2023-07-21 10:48    [W:0.951 / U:0.040 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site