[igt-dev] [PATCH i-g-t] tests/kms_cursor_crc: Fix user space read too slow error

[Juha-Pekka Heikkilä]

On Fri, Nov 29, 2019 at 6:04 PM Chris Wilson <chris at chris-wilson.co.uk> wrote:
>
> Quoting Juha-Pekka Heikkila (2019-11-29 15:52:45)
> > Having crc running continuously cause this test sometime
> > fill crc buffer, fix this problem as well as do some generic
> > cleanups.
> >
> > Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila at gmail.com>
> > ---
> >  tests/kms_cursor_crc.c | 106 +++++++++++++++++++++++++------------------------
> >  1 file changed, 54 insertions(+), 52 deletions(-)
> >
> > diff --git a/tests/kms_cursor_crc.c b/tests/kms_cursor_crc.c
> > index 6c4c457..6982437 100644
> > --- a/tests/kms_cursor_crc.c
> > +++ b/tests/kms_cursor_crc.c
> > @@ -52,7 +52,6 @@ typedef struct {
> >         struct igt_fb fb;
> >         igt_output_t *output;
> >         enum pipe pipe;
> > -       igt_crc_t ref_crc;
> >         int left, right, top, bottom;
> >         int screenw, screenh;
> >         int refresh;
> > @@ -60,6 +59,9 @@ typedef struct {
> >         int cursor_max_w, cursor_max_h;
> >         igt_pipe_crc_t *pipe_crc;
> >         unsigned flags;
> > +       igt_plane_t *primary;
> > +       igt_plane_t *cursor;
> > +       cairo_surface_t *surface;
> >  } data_t;
> >
> >  #define TEST_DPMS (1<<0)
> > @@ -89,23 +91,15 @@ static void draw_cursor(cairo_t *cr, int x, int y, int cw, int ch, double a)
> >
> >  static void cursor_enable(data_t *data)
> >  {
> > -       igt_output_t *output = data->output;
> > -       igt_plane_t *cursor =
> > -               igt_output_get_plane_type(output, DRM_PLANE_TYPE_CURSOR);
> > -
> > -       igt_plane_set_fb(cursor, &data->fb);
> > -       igt_plane_set_size(cursor, data->curw, data->curh);
> > -       igt_fb_set_size(&data->fb, cursor, data->curw, data->curh);
> > +       igt_plane_set_fb(data->cursor, &data->fb);
> > +       igt_plane_set_size(data->cursor, data->curw, data->curh);
> > +       igt_fb_set_size(&data->fb, data->cursor, data->curw, data->curh);
> >  }
> >
> >  static void cursor_disable(data_t *data)
> >  {
> > -       igt_output_t *output = data->output;
> > -       igt_plane_t *cursor =
> > -               igt_output_get_plane_type(output, DRM_PLANE_TYPE_CURSOR);
> > -
> > -       igt_plane_set_fb(cursor, NULL);
> > -       igt_plane_set_position(cursor, 0, 0);
> > +       igt_plane_set_fb(data->cursor, NULL);
> > +       igt_plane_set_position(data->cursor, 0, 0);
> >  }
> >
> >  static bool chv_cursor_broken(data_t *data, int x)
> > @@ -144,30 +138,40 @@ static bool cursor_visible(data_t *data, int x, int y)
> >         return true;
> >  }
> >
> > +static void restore_image(data_t *data)
> > +{
> > +       cairo_t *cr;
> > +
> > +       cr = igt_get_cairo_ctx(data->drm_fd, &data->primary_fb);
> > +       cairo_set_operator(cr, CAIRO_OPERATOR_SOURCE);
> > +       cairo_set_source_surface(cr, data->surface, 0, 0);
> > +       cairo_rectangle(cr, 0, 0, data->screenw, data->screenh);
> > +       cairo_fill(cr);
> > +       igt_put_cairo_ctx(data->drm_fd, &data->primary_fb, cr);
> > +       gem_sync(data->drm_fd, data->primary_fb.gem_handle);
>
> This is a generic KMS test, gem_sync() is not supported.

Idea here was to make certain memory copied to FB is really there. I
did ask from Mika for ideas and he suggested changing domain to cause
synchronization, in ioctl_wrappers.c at gem_set_domain() comments
suggested to use gem_sync() instead.

Reason for doing this is I started to get random CRC errors when I
switched away from calling igt_paint_test_pattern() every time and
just use test pattern stored in Cairo surface which seemed much
faster. Adding wait after igt_put_cairo_ctx() removed those CRC errors
so I figured there's my problem.

On my ICL box those random CRC error went away with this pattern but
if gem_sync doesn't do here as I had planned then it can be I just got
lucky with randomness of random CRC errors :) Once CI has ran this I
know more.

/Juha-Pekka