[Intel-gfx] [PATCH i-g-t v2 07/12] lib/igt_kms: Add support for the OUT_FENCE_PTR property

[Brian Starkey]

On Fri, Dec 16, 2016 at 03:21:45AM -0500, Robert Foss wrote:
>
>
>On 2016-12-14 11:13 AM, Brian Starkey wrote:
>>Hi,
>>
>>On Wed, Dec 14, 2016 at 04:05:04AM -0500, Robert Foss wrote:
>>>From: Gustavo Padovan <gustavo.padovan at collabora.co.uk>
>>>
>>>Add support for the OUT_FENCE_PTR property to enable setting out
>>>fences for
>>>atomic commits.
>>>
>>>Signed-off-by: Gustavo Padovan <gustavo.padovan at collabora.co.uk>
>>>Signed-off-by: Robert Foss <robert.foss at collabora.com>
>>>---
>>>lib/igt_kms.c | 21 ++++++++++++++++++++-
>>>lib/igt_kms.h |  3 +++
>>>2 files changed, 23 insertions(+), 1 deletion(-)
>>>
>>>diff --git a/lib/igt_kms.c b/lib/igt_kms.c
>>>index 8ca49d86..fe1b356d 100644
>>>--- a/lib/igt_kms.c
>>>+++ b/lib/igt_kms.c
>>>@@ -175,7 +175,8 @@ const char
>>>*igt_crtc_prop_names[IGT_NUM_CRTC_PROPS] = {
>>>    "DEGAMMA_LUT",
>>>    "GAMMA_LUT",
>>>    "MODE_ID",
>>>-    "ACTIVE"
>>>+    "ACTIVE",
>>>+    "OUT_FENCE_PTR"
>>>};
>>>
>>>const char *igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
>>>@@ -2103,6 +2104,10 @@ static void
>>>igt_atomic_prepare_crtc_commit(igt_pipe_t *pipe_obj, drmModeAtomicRe
>>>        igt_atomic_populate_crtc_req(req, pipe_obj, IGT_CRTC_ACTIVE,
>>>!!output);
>>>    }
>>>
>>>+    if (pipe_obj->out_fence_ptr)
>>>+        igt_atomic_populate_crtc_req(req, pipe_obj,
>>>IGT_CRTC_OUT_FENCE_PTR,
>>>+            (uint64_t)(uintptr_t) pipe_obj->out_fence_ptr);
>>>+
>>>    /*
>>>     *    TODO: Add all crtc level properties here
>>>     */
>>>@@ -2683,6 +2688,20 @@ igt_pipe_set_gamma_lut(igt_pipe_t *pipe, void
>>>*ptr, size_t length)
>>>}
>>>
>>>/**
>>>+ * igt_pipe_set_out_fence_ptr:
>>>+ * @pipe: pipe pointer to which background color to be set
>>>+ * @fence_ptr: out fence pointer
>>>+ *
>>>+ * Sets the out fence pointer that will be passed to the kernel in
>>>+ * the atomic ioctl. When the kernel returns the out fence pointer
>>>+ * will contain the fd number of the out fence created by KMS.
>>>+ */
>>>+void igt_pipe_set_out_fence_ptr(igt_pipe_t *pipe, int64_t *fence_ptr)
>>>+{
>>>+    pipe->out_fence_ptr = fence_ptr;
>>>+}
>>>+
>>>+/**
>>> * igt_crtc_set_background:
>>> * @pipe: pipe pointer to which background color to be set
>>> * @background: background color value in BGR 16bpc
>>>diff --git a/lib/igt_kms.h b/lib/igt_kms.h
>>>index 9766807c..8eb611c0 100644
>>>--- a/lib/igt_kms.h
>>>+++ b/lib/igt_kms.h
>>>@@ -110,6 +110,7 @@ enum igt_atomic_crtc_properties {
>>>       IGT_CRTC_GAMMA_LUT,
>>>       IGT_CRTC_MODE_ID,
>>>       IGT_CRTC_ACTIVE,
>>>+       IGT_CRTC_OUT_FENCE_PTR,
>>>       IGT_NUM_CRTC_PROPS
>>>};
>>>
>>>@@ -316,6 +317,7 @@ struct igt_pipe {
>>>
>>>    uint64_t mode_blob;
>>>    bool mode_changed;
>>>+    int64_t *out_fence_ptr;
>>
>>I prefer the interface that got suggested before - igt_pipe gets an
>>"int64_t out_fence;" member which tests can query to get the fence
>>value:
>>
>>int igt_pipe_get_last_out_fence(igt_pipe_t *pipe);
>>
>>..and the kernel only ever receives a pointer to pipe->out_fence.
>>
>>The only reason I can see for a test to want to provide its own fence
>>pointer is to test invalid pointer values - and I think it's OK for
>>that test to set the property directly instead of making setting a
>>custom fence pointer the common case for all users.
>>
>>If we don't want to get a fence for every commit then I guess there
>>could be a way for a test to request an out-fence for just the next
>>commit on a pipe (or the inverse - disable fencing for a particular
>>commit):
>>
>>void igt_pipe_request_out_fence(igt_pipe_t *pipe);
>>
>>-Brian
>
>Now I see what you meant in the v1 discussion, I'll amend the 
>implementation in v3 to be the one mentioned above.
>

Partly... my main point on v1 was just to make sure the pointer was to
a 64-bit type.

>The only question I have is about igt_pipe_get_last_out_fence(), is 
>it really necessary? I don't foresee a function like that ever doing 
>more than just returning a struct member. Is it not a bit redundant?
>

I see two reasons for it:
  - It means tests only deal with plain ints, which I think Chris was
    fairly adamant about on v1.
  - The getter can set pipe->out_fence to -1 when its called, making
    the ownership of the fd obvious.

In this case I guess before each commit out_fence needs to be checked,
close()'d if it looks valid, and set to -1 too.

Cheers,
Brian

>
>Rob.
>
>>
>>>};
>>>
>>>typedef struct {
>>>@@ -369,6 +371,7 @@ static inline bool
>>>igt_plane_supports_rotation(igt_plane_t *plane)
>>>void igt_pipe_set_degamma_lut(igt_pipe_t *pipe, void *ptr, size_t
>>>length);
>>>void igt_pipe_set_ctm_matrix(igt_pipe_t *pipe, void *ptr, size_t length);
>>>void igt_pipe_set_gamma_lut(igt_pipe_t *pipe, void *ptr, size_t length);
>>>+void igt_pipe_set_out_fence_ptr(igt_pipe_t *pipe, int64_t *fence_ptr);
>>>
>>>void igt_plane_set_fb(igt_plane_t *plane, struct igt_fb *fb);
>>>void igt_plane_set_fence_fd(igt_plane_t *plane, uint32_t fence_fd);
>>>--
>>>2.11.0
>>>