[PATCH v2 4/7] drm/simpledrm: Add support for system memory framebuffers

[Rob Herring]

On Tue, Oct 18, 2022 at 5:47 AM Thierry Reding <thierry.reding at gmail.com> wrote:
>
> On Mon, Oct 17, 2022 at 01:15:59PM -0500, Rob Herring wrote:
> > On Mon, Oct 17, 2022 at 9:54 AM Thierry Reding <thierry.reding at gmail.com> wrote:
> > >
> > > On Mon, Oct 10, 2022 at 10:12:34AM +0200, Thomas Zimmermann wrote:
> > > > Hi
> > > >
> > > > Am 07.10.22 um 14:49 schrieb Thierry Reding:
> > > > > From: Thierry Reding <treding at nvidia.com>
> > > > >
> > > > > Simple framebuffers can be set up in system memory, which cannot be
> > > > > requested and/or I/O remapped using the I/O resource helpers. Add a
> > > > > separate code path that obtains system memory framebuffers from the
> > > > > reserved memory region referenced in the memory-region property.
> > > > >
> > > > > v2: make screen base a struct iosys_map to avoid sparse warnings
> >
> > [...]
> >
> > > > > +static int simple_framebuffer_init(struct reserved_mem *rmem)
> > > > > +{
> > > > > +   pr_info("framebuffer memory at %pa, size %lu bytes\n", &rmem->base,
> > > > > +           (unsigned long)rmem->size);
> > > > > +
> > > > > +   rmem->ops = &simple_framebuffer_ops;
> > > > > +
> > > > > +   return 0;
> > > > > +}
> > > > > +RESERVEDMEM_OF_DECLARE(simple_framebuffer, "framebuffer", simple_framebuffer_init);
> > > >
> > > > What's the prupose of these code at all?  I looked through the kernel, but
> > > > there aren't many other examples of it.
> > >
> > > This is a fairly standard construct to deal with early memory
> > > reservations. What happens is roughly this: during early kernel boot,
> > > the reserved-memory core code will iterate over all children of the top-
> > > level reserved-memory node and see if they have a compatible string that
> > > matches one of the entries in the table created by these
> > > RESERVEDMEM_OF_DECLARE entries. It will then call the init function for
> > > a matched entry and register a struct reserved_mem for these. The init
> > > function in this case just dumps an informational message to the boot
> > > log to provide some information about the framebuffer region that was
> > > reserved (which can be used for example for troubleshooting purposes)
> > > and sets the device init/release operations (which will be called when a
> > > device is associated with the reserved memory region, i.e. when the
> > > of_reserved_mem_device_init_by_idx() function is called).
> > >
> > > The reason why there aren't many examples of this is because these are
> > > special memory regions that (at least upstream) kernels seldom support.
> > > Perhaps the most common use-cases are the shared DMA pools (such as
> > > CMA).
> >
> > Also, not all regions need to be handled 'early' before slab allocator
> > or drivers are probed. Do you need early handling here? I can't see
> > why other than if fbcon is up early.
>
> No, I don't think this needs early handling. Obviously we want this to
> be available as soon as possible, but since the framebuffer driver is
> built on top of DRM and that all becomes available fairly late, I don't
> think this could ever run *that* early.
>
> So are you saying that in general if we don't need early handling we
> should avoid RESERVEDMEM_OF_DECLARE and instead manually resolve the
> memory regions and inspect them? In other words, RESERVEDMEM_OF_DECLARE
> should only ever be used when this early handling is needed?

Right. Like all the other *_OF_DECLARE() macros, they are only for
when needed before driver probe time. Lots of shared memory mailbox
drivers use reserved-memory regions if you need examples.

Rob