lkml.org 
[lkml]   [2019]   [May]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] serial: 8250: Add support for using platform_device resources
On Tue, May 14, 2019 at 02:02:36PM +0200, Esben Haabendal wrote:
> Andy Shevchenko <andy.shevchenko@gmail.com> writes:
> > On Tue, May 14, 2019 at 10:24 AM Esben Haabendal <esben@haabendal.dk> wrote:
> >> Andy Shevchenko <andriy.shevchenko@linux.intel.com> writes:
> >> > On Tue, May 07, 2019 at 02:22:18PM +0200, Esben Haabendal wrote:
> >
> >> We are on repeat here. I don't agree with you here. I have a simple
> >> generic 8250 (16550A) compatible device, and cannot use it in a mfd
> >> driver using the standard mfd-core framework.
> >
> >> The lacking of support for platform_get_resource() in the generic
> >> serial8250 driver is not a feature. It should be supported, just as it
> >> is in several of the specialized 8250 drivers.
> >
> > We are going circles here.
> > What exactly prevents you to use it? Presence of request_mem_region()?
>
> Exactly.

And I completely tired to repeat that this is okay and does not prevent you to
use MFD.

> >> It would still mean that I would have revert to not using convenient and
> >> otherwise fully appropriate API calls like pci_request_regions() and
> >> mfd_add_devices().
> >
> > Yes, here is the issue. 8250 requires the parent not to *request*
> > resources. Because child handles IO access itself.
>
> Ok, clearly we are not discussing the actual IO access. The only issue
> is how to handle the memory resource management.
>
> And yes, serial8250 requires "the parent" to not request the resources.
> But by doing so, it gets in the way of the mfd-core way of splitting a
> properly requested resource.

Which is right thing to do. Requesting resources should be done by their actual
user, no?

Moreover, if you look at /proc/iomem, for example, I bet there will be
a difference with your proposed method and a established one.

> >> The mfd driver in question is for a PCI device. Not being able to
> >> request the PCI regions seems silly.
> >
> > Nope. Otherwise, the parent which *doesn't handle* IO on behalf of
> > child should not request its resources.
>
> If I may, could I get you to discuss this with Lee Jones?
>
> As I read both of your comments in this thread, you are not aligned on
> how mfd drivers should handle resources. And in that case, one of you
> are most likely more right than the other, and if Lee is right, I seem
> to be unable to convince you about that.

MFD has nothing to do with *requesting* resource. It's about *slicing* them.
*Requesting* resource is orthogonal to the *slicing*.

> >> Not being able to register all child devices with the call introduced
> >> for that sole purpose also seems silly.
> >
> >> Please take a look at https://lkml.org/lkml/2019/4/9/576
> >> ("[PATCH v2 2/4] mfd: ioc3: Add driver for SGI IOC3 chip")
> >
> > Thank you for this link.
> > Now, look at this comment:
> >
> > + /*
> > + * Map all IOC3 registers. These are shared between subdevices
> > + * so the main IOC3 module manages them.
> > + */
> >
> > Is it your case? Can we see the code?
>
> That comment seems quite misleading. I am quote certain that the uart
> registers which are part of BAR 0 is not more shared between subdevices
> than the uart registers in BAR 0 in my case.
>
> But BAR 0 as a whole is shared between subdevices. But BAR 0 can be
> (is) split in parts that are exclusive to one subdevice. The only
> difference I see is that I don't have any registers accessed directly by
> the mfd driver.

I stopped here. Let's discuss an actual code.

--
With Best Regards,
Andy Shevchenko


\
 
 \ /
  Last update: 2019-05-14 14:39    [W:0.060 / U:30.040 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site