On 07/01/16 16:56, Chris Wilson wrote:
On Thu, Jan 07, 2016 at 12:34:39PM +0000, Dave Gordon wrote:
On 07/01/16 11:58, Chris Wilson wrote:
On Thu, Jan 07, 2016 at 10:20:50AM +0000, Dave Gordon wrote:
There are a number of places where the driver needs a request, but isn't
working on behalf of any specific user or in a specific context. At
present, we associate them with the per-engine default context. A future
patch will abolish those per-engine context pointers; but we can already
eliminate a lot of the references to them, just by making the allocator
allow NULL as a shorthand for "an appropriate context for this ring",
which will mean that the callers don't need to know anything about how
the "appropriate context" is found (e.g. per-ring vs per-device, etc).
So this patch renames the existing i915_gem_request_alloc(), and makes
it local (static inline), and replaces it with a wrapper that provides
a default if the context is NULL, and also has a nicer calling
convention (doesn't require a pointer to an output parameter). Then we
change all callers to use the new convention:
OLD:
err = i915_gem_request_alloc(ring, user_ctx, &req);
if (err) ...
NEW:
req = i915_gem_request_alloc(ring, user_ctx);
if (IS_ERR(req)) ...
OLD:
err = i915_gem_request_alloc(ring, ring->default_context, &req);
if (err) ...
NEW:
req = i915_gem_request_alloc(ring, NULL);
if (IS_ERR(req)) ...
Nak. You haven't fixed i915_gem_request_alloc() at all.
http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?h=breadcrumbs&id=82c72e1a2b4385f0ab07dccee45acef38303e96f
is the patch I have been carrying ever since.
-Chris
I think you'll find that the version of i915_gem_request_alloc()
I've implemented is equivalent to yours, with the *additional* (and
better) semantic of not requiring the caller to specify
(ring->default_param) as the context parameter (which is the main
point, as far as I'm concerned; making the calling convention nicer
was just incidental).
No. Specifying the context is crucial to request allocation. The issue
in the current function call chains are the requests appear out of
nowhere rather than being created with explicit context.
Hiding the context is bad.
-Chris
Why would it be better to name dev_priv->kernel_context at every
callsite rather than once in the called function (especially in my
version where it's very obvious that the new wrapper provides that
default before calling the original underlying allocator).
The context is only important to callers who have one in hand. To the
others, it's an irrelevance; they're just saying, "hey, I want a request
but it's not connected to any context I know of, so gimme a default".
The proof of that is that we can switch them from a per-engine default
to a per-driver default without changing anything else about them. They
really don't need to know how to find the default context, so let the
allocator find it for them.
<aside>
Suppose we some day decide that perhaps there should be *multiple*
/kernel contexts/ (maybe for greater parallelism); my approach will let
us spread non-user requests across all suitable contexts just by
implementing that choice in one place (the new wrapper). By contrast,
naming the default kernel context at every callsite would make this a
far more invasive change.
</aside>
It is a commonplace of modern software practice that encapsulating (and
thus hiding) irrelevant detail is GOOD practice; it is the exposing of
internal implementation detail that is BAD.
.Dave.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
