[igt-dev] [PATCH i-g-t v2 2/2] igt/tests/kms_dp_tiled_display: kms test for display port tiled displays
Manasi Navare
manasi.d.navare at intel.com
Wed Aug 28 22:35:14 UTC 2019
On Tue, Aug 27, 2019 at 02:29:49PM -0700, Manasi Navare wrote:
> On Tue, Aug 27, 2019 at 03:49:38AM -0700, Ser, Simon wrote:
> > Thanks for the new version! Here are some more comments. Some of them
> > are nits, some of them are more serious questions. Feel free to let me
> > know if I'm mistaken or if you don't understand one of them.
>
> Thanks for the review comments Simon. I will be working on addressing the following
> review comments and will take over this patch since Madhumitha's internship has ended.
> I really appreciate all your help and feedback on this.
>
> Please find my answers below:
>
> >
> > On Fri, 2019-08-23 at 11:23 -0700, Madhumitha Tolakanahalli Pradeep wrote:
> > > This test validates the tiled DP displays to display a test pattern
> > > seamlessly across the two tiles. It validates the transcoder port
> > > sync feature on i915 to get a tearfree tiled display output.
> > >
> > > Related kernel work patches-
> > > https://patchwork.freedesktop.org/series/59837/#rev4.
> > >
> > > This test can eventually be extended to cover tiled display support
> > > on other connector types.
> > >
> > > v2: Added a check for checking pageflip event timestamps (Simon, Manasi)
> > > Minor style changes (Simon)
> > > Code clean-up and reordering
> > >
> > >
> > > Cc: Manasi Navare <manasi.d.navare at intel.com>
> > > Cc: Petri Latvala <petri.latvala at intel.com>
> > > Cc: Simon Ser <simon.ser at intel.com>
> > >
> > > Cc: <madhumitha.tp at gmail.com>
> > >
> > > Signed-off-by: Madhumitha Tolakanahalli Pradeep <madhumitha.tolakanahalli.pradeep at intel.com>
> > > ---
> > > tests/Makefile.sources | 1 +
> > > tests/kms_dp_tiled_display.c | 346 +++++++++++++++++++++++++++++++++++
> > > tests/meson.build | 1 +
> > > 3 files changed, 348 insertions(+)
> > > create mode 100644 tests/kms_dp_tiled_display.c
> > >
> > > diff --git a/tests/Makefile.sources b/tests/Makefile.sources
> > > index c02e4d94..7561ab9b 100644
> > > --- a/tests/Makefile.sources
> > > +++ b/tests/Makefile.sources
> > > @@ -41,6 +41,7 @@ TESTS_progs = \
> > > kms_cursor_edge_walk \
> > > kms_cursor_legacy \
> > > kms_dp_dsc \
> > > + kms_dp_tiled_display \
> > > kms_draw_crc \
> > > kms_fbcon_fbt \
> > > kms_fence_pin_leak \
> > > diff --git a/tests/kms_dp_tiled_display.c b/tests/kms_dp_tiled_display.c
> > > new file mode 100644
> > > index 00000000..162fbdd9
> > > --- /dev/null
> > > +++ b/tests/kms_dp_tiled_display.c
> > > @@ -0,0 +1,346 @@
> > > +/*
> > > + * Copyright © 2018 Intel Corporation
> > > + *
> > > + * Permission is hereby granted, free of charge, to any person obtaining a
> > > + * copy of this software and associated documentation files (the "Software"),
> > > + * to deal in the Software without restriction, including without limitation
> > > + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> > > + * and/or sell copies of the Software, and to permit persons to whom the
> > > + * Software is furnished to do so, subject to the following conditions:
> > > + *
> > > + * The above copyright notice and this permission notice (including the next
> > > + * paragraph) shall be included in all copies or substantial portions of the
> > > + * Software.
> > > + *
> > > + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> > > + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> > > + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> > > + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> > > + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> > > + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> > > + * IN THE SOFTWARE.
> > > + *
> > > + * Authors:
> > > + * Madhumitha Tolakanahalli Pradeep
> > > + * <madhumitha.tolakanahalli.pradeep at intel.com>
> > > + *
> > > + * Display Port Tiled Display Test
> > > + * This test parses the tile information of the connectors that have TILE
> > > + * property set, sets up the framebuffer with correct offsets corresponding to
> > > + * the tile offsets and does an atomic modeset with two CRTCs for two
> > > + * connectors. Page flip event timestamp from each CRTC is collected and
> > > + * compared to make sure that they occurred in a synchronous manner.
> > > + *
> > > + * This test currently supports only horizontally tiled displays, in line with
> > > + * the displays supported by the kernel at the moment.
> > > + */
> > > +
> > > +#include "igt.h"
> > > +#include "poll.h"
> > > +#include "drm_mode.h"
> > > +#include "drm_fourcc.h"
> > > +
> > > +IGT_TEST_DESCRIPTION("Test for Display Port Tiled Displays");
> > > +
> > > +typedef struct {
> > > + int drm_fd;
> > > + int num_h_tiles;
> > > + igt_display_t *display;
> > > + enum igt_commit_style commit;
> > > +} data_t;
> > > +
> > > +typedef struct {
> > > + igt_output_t *output;
> > > + igt_tile_info_t tile;
> > > + igt_fb_t fb_test_pattern;
> > > + enum pipe pipe;
> > > + enum igt_commit_style commit;
> > > + drmModeConnectorPtr connector;
> > > +} data_connector_t;
> > > +
> > > +static int drm_property_is_tile(drmModePropertyPtr prop)
> > > +{
> > > + return (strcmp(prop->name , "TILE") ? 0 : 1) &&
> > > + drm_property_type_is(prop, DRM_MODE_PROP_BLOB);
> > > +}
> > > +
> > > +static void get_connector_tile_props(data_t *data, drmModeConnectorPtr conn,
> > > + igt_tile_info_t *tile)
> > > +{
> > > + int i = 0;
> > > + drmModePropertyPtr prop;
> > > + drmModePropertyBlobPtr blob;
> > > +
> > > + for (i = 0; i < conn->count_props; i++) {
> > > + prop = drmModeGetProperty(data->drm_fd, conn->props[i]);
> > > +
> > > + igt_assert(prop);
> > > +
> > > + if (!drm_property_is_tile(prop))
> > > + continue;
> >
> > This branch leaks prop.
>
> Freeing prop here before continuing shd fix this, will fix this in the next rev
>
> >
> > > + blob = drmModeGetPropertyBlob(data->drm_fd,
> > > + conn->prop_values[i]);
> > > +
> > > + if (!blob)
> > > + return;
> >
> > This branch leaks prop.
>
> Agree, will add drmModeFreePropertyBlob(blob); before return
>
> >
> > > + igt_parse_connector_tile_blob(blob, tile);
> > > + break;
> > > + }
> > > +
> > > + drmModeFreeProperty(prop);
> > > + drmModeFreePropertyBlob(blob);
> > > +}
> > > +
> > > +static void get_number_of_h_tiles(data_t *data)
> > > +{
> > > + int i;
> > > + drmModeResPtr res;
> > > + drmModeConnectorPtr connector;
> > > + igt_tile_info_t tile = {.num_h_tile = 0};
> > > +
> > > + igt_assert(res = drmModeGetResources(data->drm_fd));
> > > +
> > > + for (i = 0; i < res->count_connectors; i++) {
> > > + connector = drmModeGetConnectorCurrent(data->drm_fd,
> > > + res->connectors[i]);
> > > +
> > > + igt_assert(connector);
> > > +
> > > + if (!(connector->connection == DRM_MODE_CONNECTED &&
> > > + connector->connector_type == DRM_MODE_CONNECTOR_DisplayPort))
> > > + continue;
> >
> > This can be simplified to
> >
> > connection != DRM_MODE_CONNECTED ||
> > connector_type != DRM_MODE_CONNECTOR_DisplayPort
>
> Yes will simplify in the next rev
>
> >
> > Additionally, this branch leaks the connector.
>
> Hmm, so call drmModeFreeConnector(connector); before continuing shd be followed in general throughout
> the code right?
>
> >
> > > + get_connector_tile_props(data, connector, &tile);
> > > + data->num_h_tiles = tile.num_h_tile;
> > > + break;
> > > + }
> > > +
> > > + drmModeFreeResources(res);
> > > + drmModeFreeConnector(connector);
> > > +}
> > > +
> > > +static void get_connector(data_t *data, data_connector_t *conn)
> >
> > Nit: since this fills all connectors, maybe it should be named
> > get_connectors (with an s). Also it's not clear that conn refers to an
> > array of connectors, maybe it should be renamed to conns.
>
> Yes I agree that will make it more readable, I will fix it in next rev
>
> >
> > (Applies to the whole patch)
> >
> > > +{
> > > + int count = 0;
> > > + igt_output_t *output;
> > > +
> > > + for_each_connected_output(data->display, output) {
> > > + conn[count].connector = drmModeGetConnector(data->display->drm_fd,
> > > + output->id);
> > > +
> > > + igt_assert(conn[count].connector);
> > > +
> > > + if (!(conn[count].connector->connector_type ==
> > > + DRM_MODE_CONNECTOR_DisplayPort))
> > > + continue;
> >
> > This can be simplified to
> >
> > if (connector_type != DRM_MODE_CONNECTOR_DisplayPort)
> >
> > Additionally, this branch leaks the connector.
> >
> > > + get_connector_tile_props(data, conn[count].connector,
> > > + &conn[count].tile);
> > > +
> > > + /* Check if the connectors belong to the same tile group */
> > > + if (count > 0)
> > > + igt_assert(conn[count].tile.tile_group_id ==
> > > + conn[count-1].tile.tile_group_id);
> > > +
> > > + count++;
> > > + }
> > > +}
> > > +
> > > +static void
> > > +reset_framebuffer(int drm_fd, igt_output_t *output, igt_fb_t *fb)
> > > +{
> > > + igt_plane_t *primary;
> > > +
> > > + primary = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
> > > + igt_plane_set_fb(primary, NULL);
> > > + igt_remove_fb(drm_fd, fb);
> > > +}
> > > +
> > > +static void reset_output(igt_output_t *output)
> > > +{
> > > + igt_output_set_pipe(output, PIPE_NONE);
> > > +}
> > > +
> > > +static void test_cleanup(data_t *data, data_connector_t *conn)
> > > +{
> > > + int count;
> > > +
> > > + for (count = 0; count < data->num_h_tiles; count++) {
> > > + if (conn[count].output) {
> > > + reset_framebuffer(data->drm_fd, conn[count].output,
> > > + &conn[count].fb_test_pattern);
> > > + reset_output(conn[count].output);
> > > + }
> > > + }
> > > + igt_display_commit2(data->display, data->commit);
> > > +}
> > > +
> > > +static void setup_mode(data_t *data, data_connector_t *conn_data)
> >
> > This function doesn't actually setup the output mode, it only sets up
> > the pipe. Is this an overlook?
>
> yes we just set the pipe/CRTC here and the atomic commit call does a complete
> modeset with check and commit without any output since we didnt setup fb yet.
> But I agree that calling it just setup_pipe robably makes it more intuitive
>
> >
> > > +{
> > > + int count = 0;
> > > + enum pipe pipe;
> > > + igt_output_t *output;
> > > +
> > > + for (count = 0; count < data->num_h_tiles; count++) {
> > > + output = igt_output_from_connector(data->display,
> > > + conn_data[count].connector);
> > > +
> > > + /*
> > > + * The output is set to PIPE_NONE and then assigned a pipe.
> > > + * This is done to ensure a complete modeset occures every
> > > + * time the test is run.
> > > + */
> > > + reset_output(output);
> >
> > Is this necessary?
>
> Yes this is necessary to force a complete modeset for each test since if the pipe
> assignments or the mode has not changed it will not set needs_modeset flag in side kernel
> and will not do a full atomic check and commit.
The igt_display_commit2 call is also needed after setting the pipe to NONE to ensure that
we clear the pipe assignments and then do a modeset.
I just tested that and will add it in next rev as below:
static void reset_mode(data_t *data, data_connector_t *conn_data)
{
int count;
igt_output_t *output;
for (count = 0; count < data->num_h_tiles; count++) {
output = igt_output_from_connector(data->display,
conn_data[count].connector);
igt_output_set_pipe(output, PIPE_NONE);
}
igt_display_commit2(data->display, data->commit);
}
Manasi
>
> >
> > > + for_each_pipe(data->display, pipe) {
> > > + if (count > 0 && pipe == conn_data[count-1].pipe)
> > > + continue;
> > > +
> > > + if (igt_pipe_connector_valid(pipe, output)) {
> > > +
> > > + conn_data[count].pipe = pipe;
> > > + conn_data[count].output = output;
> > > +
> > > + igt_output_set_pipe(conn_data[count].output,
> > > + conn_data[count].pipe);
> > > + break;
> > > + }
> > > + }
> > > + }
> > > + igt_display_commit_atomic(data->display, DRM_MODE_ATOMIC_ALLOW_MODESET,
> > > + NULL);
> > > +}
> > > +
> > > +static void setup_framebuffer(data_t *data, data_connector_t *conn)
> > > +{
> > > + int count;
> > > + igt_plane_t *primary;
> > > +
> > > + for (count = 0; count < data->num_h_tiles; count++) {
> > > +
> > > + igt_create_pattern_fb(data->drm_fd,
> > > + (conn[count].tile.tile_h_size *
> > > + data->num_h_tiles),
> >
> > Do we need to multiply by the number of horizontal tiles here? It seems
> > we only use the first tile_h_size pixels of the buffer. Am I missing
> > something?
> >
>
> Yes we only use the tile_h_size of the buffer at the proper offset of tile_h_loc
> So this below logic shd be modified to only create fb of the size tile_h_size if its in the for loop
> or just create one fb outside the loop of the size tile_h_size * num_tiles and then just use that with dfferent
> offsets for two CRTCs
>
> When I discussed this on dri-devel, people said that real userspace can decide to do either one single fb for all tiles
> or separate ones where each is of the size equal to tile size.
>
> Let me know if there is a preference
>
> > But maybe a test that would better reflect reality would create a
> > single framebuffer for all tiled displays, to display a single image
> > across all of them?
> >
> > > + conn[count].tile.tile_v_size,
> > > + DRM_FORMAT_XBGR8888,
> > > + LOCAL_DRM_FORMAT_MOD_NONE,
> > > + &conn[count].fb_test_pattern);
> > > +
> > > + primary = igt_output_get_plane_type(conn[count].output,
> > > + DRM_PLANE_TYPE_PRIMARY);
> > > +
> > > + igt_plane_set_fb(primary, &conn[count].fb_test_pattern);
> > > +
> > > + igt_fb_set_size(&conn[count].fb_test_pattern, primary,
> > > + conn[count].tile.tile_h_size,
> > > + conn[count].tile.tile_v_size);
> > > +
> > > + igt_fb_set_position(&conn[count].fb_test_pattern, primary,
> > > + (conn[count].tile.tile_h_size *
> > > + conn[count].tile.tile_h_loc),
> > > + (conn[count].tile.tile_v_size *
> > > + conn[count].tile.tile_v_loc));
> > > +
> > > + igt_plane_set_size(primary,
> > > + conn[count].tile.tile_h_size,
> > > + conn[count].tile.tile_v_size);
> > > + }
> > > +}
> > > +
> > > +static void page_flip_handler(int fd, unsigned int seq,
> > > + unsigned int tv_sec, unsigned int tv_usec,
> > > + unsigned int crtc_id, void *data)
> > > +{
> > > + bool expr = false;
> > > + static unsigned int _tv_sec, _tv_usec;
> > > +
> > > + igt_debug("Page Flip Event received from CRTC:%d at %u:%u\n", crtc_id,
> > > + tv_sec, tv_usec);
> >
> > We should also make sure we receive exactly one page-flip per CRTC.
> > Currently we don't use crtc_id (apart from logging purposes).
>
> I see so somehow have a static count of events and make sure per crtc id its the same count or
> some similar logic
>
> >
> > > + /* Skip the following checks for the first page flip event */
> > > + if (_tv_sec == 0 || _tv_usec == 0) {
> > > + _tv_sec = tv_sec;
> > > + _tv_usec = tv_usec;
> > > + return;
> > > + }
> > > +
> > > + /*
> > > + * For seamless tear-free display, the page flip event timestamps
> > > + * from all the tiles should not differ by more than 10us.
> > > + */
> > > + expr = tv_sec == _tv_sec && (abs(tv_usec - _tv_usec) < 10);
> >
> > Nit: this could be renamed to e.g. is_on_time, expr is a pretty vague
> > name.
>
> yes agree will rename to make it more readable
>
> >
> > > + igt_fail_on_f(!expr, "Delayed page flip event from CRTC:%d at %u:%u\n",
> > > + crtc_id, tv_sec, tv_usec);
> > > +
> > > + if (tv_sec < _tv_sec)
> > > + _tv_sec = tv_sec;
> > > + if (tv_usec < _tv_usec)
> > > + _tv_usec = tv_usec;
> >
> > This updates the timestamp on each page-flip. This means that the first
> > and last page-flips could be delayed by more than 10µs if there are
> > intermediate page-flips. For instance, if the first page-flip happens
> > at t=0µs and the second one happens at t=9µs, the last one could happen
> > at t=18µs without making this test fails. Is this something we want to
> > allow?
>
> Nope all of them should happen within the same vblank so the diff between all
> should be <10us, will have to modify the logic to do that.
>
> so basically instead of making <10us condition for the consecutive ones, it should be between the first one
> and all other events, correct?
>
> >
> > > +}
> > > +
> > > +static void wait_for_pageflip(int drm_fd)
> > > +{
> > > + struct pollfd pfd;
> > > + drmEventContext drm_event;
> >
> > Nit: to make sure garbage isn't read from this struct (there are other
> > function pointers), it's probably safer to zero-fill it (= {0}).
>
> Got it
>
> >
> > > + drm_event.version = 3;
> > > + drm_event.page_flip_handler2 = page_flip_handler;
> > > +
> > > + pfd.fd = drm_fd;
> > > + pfd.events = POLLIN;
> > > + pfd.revents = 0;
> > > +
> > > + poll(&pfd, 1, 1000);
> >
> > Maybe we should check poll(2)'s return value?
>
> poll(2)? Or just check return value of the function poll() just for error handling?
>
> >
> > > + if (pfd.revents & POLLIN)
> > > + drmHandleEvent(drm_fd, &drm_event);
> >
> > If we don't get POLLIN (e.g. if we timeout), wait_for_pageflip will
> > silently do nothing, and the test will pass. Instead, we probably want
> > to fail in this case.
>
> Yes thats correct, just add else case with igt_assert?
>
> >
> > > +}
> > > +
> > > +igt_main
> > > +{
> > > + igt_display_t display;
> > > + data_connector_t *conn_data = NULL;
> > > + data_t data = {.drm_fd = 0, .num_h_tiles = 0,
> > > + .display = NULL, .commit = COMMIT_LEGACY};
> >
> > Nit: one can just do assign to {0} to zero-fill the struct.
>
> Got it
>
> >
> > > + igt_fixture {
> > > + data.drm_fd = drm_open_driver_master(DRIVER_ANY);
> > > +
> > > + kmstest_set_vt_graphics_mode();
> > > + igt_display_require(&display, data.drm_fd);
> > > + igt_display_reset(&display);
> > > +
> > > + data.display = &display;
> > > +
> > > + get_number_of_h_tiles(&data);
> > > + igt_debug("Number of Horizontal Tiles: %d\n", data.num_h_tiles);
> > > +
> > > + data.commit = data.display->is_atomic ? COMMIT_ATOMIC : COMMIT_LEGACY;
> >
> > Should we try to run this test at all on drivers that don't support
> > atomic?
> >
> > If the driver doesn't support atomic, user-space will submit two page-
> > flip requests. However it's possible that the hardware executes a page-
> > flip between both requests: user-space calls the first drmModePageFlip,
> > the vblank for this particular CRTC triggers, then user-space calls the
> > second drmModePageFlip. Page-flips will be out-of-sync by one frame.
>
> Hmm, yea should probbaly have igt_require on atomic
>
> >
> > > + if (data.num_h_tiles > 0)
> > > + conn_data = malloc(data.num_h_tiles * sizeof(data_connector_t));
> > > + }
> > > +
> > > + igt_subtest("basic-test-pattern") {
> > > + igt_skip_on(data.num_h_tiles == 0);
> > > + igt_assert(conn_data);
> > > +
> > > + get_connector(&data, conn_data);
> > > + setup_mode(&data, conn_data);
> > > + setup_framebuffer(&data, conn_data);
> > > + igt_display_commit_atomic(data.display, DRM_MODE_ATOMIC_NONBLOCK |
> > > + DRM_MODE_PAGE_FLIP_EVENT, NULL);
> > > + wait_for_pageflip(data.drm_fd);
> >
> > Here we only wait for a single page-flip. This will only stash the
> > timestamp of the frst page-flip without checking the time-stamp of
> > subsequent page-flips. We should wait for as many page-flips as enabled
> > tiled connectors.
>
> That was our initial understanding as well, but with just one wait_for_pageflip call we
> were geting two page flip events for two tiles
> Do we need to call wait for page flip explicitly the number of times = number of tiles?
>
> Regards
> Manasi
>
> >
> > > + test_cleanup(&data, conn_data);
> > > + }
> > > +
> > > + igt_fixture {
> > > + free(conn_data);
> > > + close(data.drm_fd);
> > > + kmstest_restore_vt_mode();
> > > + igt_display_fini(data.display);
> > > + }
> > > +}
> > > diff --git a/tests/meson.build b/tests/meson.build
> > > index a7b2b322..50292df8 100644
> > > --- a/tests/meson.build
> > > +++ b/tests/meson.build
> > > @@ -26,6 +26,7 @@ test_progs = [
> > > 'kms_cursor_edge_walk',
> > > 'kms_cursor_legacy',
> > > 'kms_dp_dsc',
> > > + 'kms_dp_tiled_display',
> > > 'kms_draw_crc',
> > > 'kms_fbcon_fbt',
> > > 'kms_fence_pin_leak',
> _______________________________________________
> igt-dev mailing list
> igt-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/igt-dev
More information about the igt-dev
mailing list