[Intel-gfx] [PATCH i-g-t 1/2] lib/dummyload: Support returning output fence

[Chris Wilson]

Quoting Tvrtko Ursulin (2017-12-22 13:13:47)
> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> 
> Support creating spin batches which return an output fence using new
> __igt_spin_batch_new_fence / igt_spin_batch_new_fence API.
> 
> This will be used fromthe perf_pmu at interrupts test to ensure user
> interrupt generation from a batch with controlled duration.
> 
> v2: Support out fence with multiple engines as well. (Chris Wilson)
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk> (v1)
> ---
>  lib/igt_dummyload.c | 80 +++++++++++++++++++++++++++++++++++++++++++++++------
>  lib/igt_dummyload.h | 10 +++++++
>  2 files changed, 82 insertions(+), 8 deletions(-)
> 
> diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
> index d19b4e5ea3d2..27eb402bb699 100644
> --- a/lib/igt_dummyload.c
> +++ b/lib/igt_dummyload.c
> @@ -34,6 +34,7 @@
>  #include "intel_chipset.h"
>  #include "intel_reg.h"
>  #include "ioctl_wrappers.h"
> +#include "sw_sync.h"
>  
>  /**
>   * SECTION:igt_dummyload
> @@ -70,9 +71,9 @@ fill_reloc(struct drm_i915_gem_relocation_entry *reloc,
>         reloc->write_domain = write_domains;
>  }
>  
> -static void emit_recursive_batch(igt_spin_t *spin,
> -                                int fd, uint32_t ctx, unsigned engine,
> -                                uint32_t dep)
> +static int emit_recursive_batch(igt_spin_t *spin,
> +                               int fd, uint32_t ctx, unsigned engine,
> +                               uint32_t dep, bool out_fence)
>  {
>  #define SCRATCH 0
>  #define BATCH 1
> @@ -82,6 +83,7 @@ static void emit_recursive_batch(igt_spin_t *spin,
>         struct drm_i915_gem_execbuffer2 execbuf;
>         unsigned int engines[16];
>         unsigned int nengine;
> +       int fence_fd = -1;
>         uint32_t *batch;
>         int i;
>  
> @@ -165,22 +167,44 @@ static void emit_recursive_batch(igt_spin_t *spin,
>         execbuf.buffers_ptr = to_user_pointer(obj + (2 - execbuf.buffer_count));
>         execbuf.rsvd1 = ctx;
>  
> +       if (out_fence)
> +               execbuf.flags |= I915_EXEC_FENCE_OUT;
> +
>         for (i = 0; i < nengine; i++) {
>                 execbuf.flags &= ~ENGINE_MASK;
> -               execbuf.flags = engines[i];
> -               gem_execbuf(fd, &execbuf);
> +               execbuf.flags |= engines[i];
> +               gem_execbuf_wr(fd, &execbuf);
> +               if (out_fence) {
> +                       int _fd = execbuf.rsvd2 >> 32;
> +
> +                       igt_assert(_fd >= 0);
> +                       if (fence_fd == -1) {
> +                               fence_fd = _fd;
> +                       } else {
> +                               int old_fd = fence_fd;
> +
> +                               fence_fd = sync_fence_merge(old_fd, _fd);
> +                               close(old_fd);
> +                               close(_fd);
> +                       }
> +                       igt_assert(fence_fd >= 0);

At some point we will write sync_fence_accumulate that does all of the
above: fence_fd = sync_fence_accumulate(fence_fd, _fd);.

Looks correct, the only thing I would nitpick is spin->out_fence; at
that point it is just the fence representing the spinner, and
spin->fence is descriptive and accurate.

Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
-Chris