Re: Fwd: TTM Interface cleanups

[<th...@tu...>]

Kristian Høgsberg wrote:

>Hi Thomas,
>
>I forgot to Cc you on these patches I sent to the dri-devel list.
>What's your take?
>
>cheers,
>Kristian
>
>---------- Forwarded message ----------
>From: Kristian Høgsberg <kr...@bi...>
>Date: Oct 9, 2007 9:26 PM
>Subject: TTM Interface cleanups
>To: dri-devel <dri...@li...>
>
>
>Hi,
>
>These patches may come through one or two times more, depending on how
>well git-send-email works and how this lists handles posts from
>non-subscribers.  Apologies in advance for any spamming I may have
>incurred.
>
>Anyway, these three patches implement a few things we discussed at XDS2007:
>
>  1/3: Eliminate support for fake buffers.  Eric did the mesa side
>work and this patch just removes support for this buffer type from the
>kernel.
>
>  2/3: Take bo type argument out of the ioctl interface.  Now that we
>only have user and dc type buffers, we can remove the buffer type
>argument from the ioctl completely and take buffer_start = 0 to mean
>type_dc and buffer_start != 0 to mean user space buffer at that
>address.  Also, the library wrapper had both "unsigned long start" and
>"void *user_buffer" - I really could not see a use for both so I kept
>only user_buffer.
>
>  3/3: Drop destroy ioctls for fences and buffer objects. This patch
>drop support for the destroy ioctl for both buffers and fences.
>Instead, they're just refcounted so that when the last user drops its
>last reference the object is destroyed.  This is implemented by just
>making the owner/creator to also reference the object through a
>drm_ref_object like all other user of the object.
>
>cheers,
>Kristian
>  
>
>
>  
>
Hi, Christian.
I haven't looked into the implementation detail, but in general this 
looks good to me.
IMHO we should collect all the libdrm changes needed, and then bump 
libdrm major and implement them.

Also on the list is the introduction of the fence error argument, and 
removal of buffer flags at buffer create.

/Thomas