[PATCH] omap2+: add drm device
rob.clark at linaro.org
Wed Mar 14 06:16:33 PDT 2012
On Wed, Mar 14, 2012 at 8:07 AM, Tomi Valkeinen <tomi.valkeinen at ti.com> wrote:
> On Wed, 2012-03-14 at 07:55 -0500, Rob Clark wrote:
>> On Wed, Mar 14, 2012 at 7:38 AM, Tomi Valkeinen <tomi.valkeinen at ti.com> wrote:
>> > Hi,
>> > On Tue, 2012-03-13 at 15:34 -0500, Rob Clark wrote:
>> >> From: Andy Gross <andy.gross at ti.com>
>> >> Register OMAP DRM/KMS platform device, and reserve a CMA region for
>> >> the device to use for buffer allocation. DMM is split into a
>> >> separate device using hwmod.
>> > What's the diff with this and the previous one?
>> Moving drm.c to mach-omap2 (header could not move because
>> omap_reserve() is in plat-omap.. but this seems to be the same as what
>> is done for dspbridge).
>> > I see you still have the platform data there. You didn't comment on
>> > that. Where is it used after the board files are gone when we use DT?
>> I was kind-of thinking that there would be some DT<->data-structure
>> glue somewhere.. not sure if this goes in mach-omap2 or in the driver
>> itself, but presumably it is needed somewhere.
>> It is only really special cases where some board specific device-tree
>> data is needed, so it seems like this is ok to handle later.
> Afaik DT data should only contain information about the hardware. This
> is SW configuration, so I think DT people won't accept things like that.
hmm, then where *does* it go.. it is a bit too much for bootargs..
>> > And how about the size of the contiguous memory area, it was left a bit
>> > unclear to me why it cannot be dynamic.
>> I don't think there is anything preventing adding a bootarg, but I
>> think it is not essential so ok to add later
> Well, maybe not essential to you =). But you are reserving 32MB memory,
> which is quite a big amount. Even if the reserved memory can be used for
> some other purposes, it's still a big chunk of "special" memory being
> reserved even if the user doesn't use or have a display at all.
If you didn't have display, and were tight on memory, wouldn't you
just disable the display device in your kernel config?
> Well, it's not an issue for me either, but I just feel that doing things
> like that without allowing the user to avoid it is a bit bad thing.
> Btw, do you know why the dma_declare_contiguous() takes a dev pointer as
> an argument, if the memory is not private to that device? (at least I
> understood from you that the memory can be used for other purposes).
contiguous use of the memory is private to the device. There is also
a global CMA pool, from which all dma_alloc_xyz() which is not
associated w/ a per-device pool comes from.. but the advantage of a
per-device-pool is it puts an upper limit on how much dma memory that
device can allocate so that it cannot starve other devices.
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
More information about the dri-devel