[igt-dev] [PATCH i-g-t] tests/kms_async: Add test to validate asynchronous flips
Paulo Zanoni
paulo.r.zanoni at intel.com
Wed Apr 8 23:37:50 UTC 2020
Em qua, 2020-04-08 às 12:06 +0530, Karthik B S escreveu:
> Thanks a lot for the review.
>
> On 4/3/2020 10:39 PM, Paulo Zanoni wrote:
> > Em sex, 2020-04-03 às 14:47 +0530, Karthik B S escreveu:
> > > Asynchronous flips are issued using the page flip IOCTL.
> > > The test consists of two subtests. The first subtest waits for
> > > the page flip event to be recevied before giving the next flip,
> > > and the second subtest doesn't wait for page flip events.
> > >
> > > The test passes if the IOCTL is succesful.
> > >
> > > Signed-off-by: Karthik B S <karthik.b.s at intel.com>
> > > Suggested-by: Paulo Zanoni <paulo.r.zanoni at intel.com>
> >
> > I feel I deserve a little more credit than just a Suggested-by since
> > this still has a lot of code I wrote :).
> >
>
> Sorry for that. I'll update in rev2. :)
>
> > > ---
> > > tests/Makefile.sources | 1 +
> > > tests/kms_async.c | 206 +++++++++++++++++++++++++++++++++++++++++
> > > tests/meson.build | 1 +
> > > 3 files changed, 208 insertions(+)
> > > create mode 100644 tests/kms_async.c
> > >
> > > diff --git a/tests/Makefile.sources b/tests/Makefile.sources
> > > index 4e44c98c..eedf4fcb 100644
> > > --- a/tests/Makefile.sources
> > > +++ b/tests/Makefile.sources
> > > @@ -29,6 +29,7 @@ TESTS_progs = \
> > > fbdev \
> > > kms_3d \
> > > kms_addfb_basic \
> > > + kms_async \
> > > kms_atomic \
> > > kms_atomic_interruptible \
> > > kms_atomic_transition \
> > > diff --git a/tests/kms_async.c b/tests/kms_async.c
> > > new file mode 100644
> > > index 00000000..0e21f9f6
> > > --- /dev/null
> > > +++ b/tests/kms_async.c
> > > @@ -0,0 +1,206 @@
> > > +/*
> > > + * Copyright © 2020 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: Paulo Zanoni <email>
> > * Karthik B S <email>
> >
>
> Noted.
>
> > > + */
> > > +
> > > +#include "igt.h"
> > > +#include "igt_aux.h"
> > > +#include <poll.h>
> > > +
> > > +#define BUFS 4
> > > +#define WARM_UP_TIME 2
> > > +#define RUN_TIME 10
> >
> > Each subtest takes 12 seconds to run, by definition (sans timeouts).
> > Maybe for IGT we want to make this a little faster? When I wrote this
> > code I needed very precise FPS values, IGT only needs to assert things
> > are working as expected.
> >
> > 1s WARM_UP_TIME and 4s RUN_TIME are probably very fine, and even less
> > than that is also probably fine for IGT. Please play with the tunables.
> >
> > IGT needs to be as fast as it can while still achieving 100%
> > reliability.
> >
>
> Will update this.
>
> > > +
> > > +typedef struct {
> > > + int drm_fd;
> > > + igt_display_t display;
> > > +} data_t;
> > > +
> > > +static drmModeConnectorPtr find_connector_for_modeset(data_t *data)
> > > +{
> > > + igt_output_t *output;
> > > + drmModeConnectorPtr ret = NULL;
> > > +
> > > + for_each_connected_output(&data->display, output) {
> > > + if (output->config.connector->count_modes > 0) {
> > > + ret = output->config.connector;
> > > + break;
> > > + }
> > > + }
> > > +
> > > + igt_assert_f(ret, "Connector NOT found\n");
> > > + return ret;
> > > +}
> > > +
> > > +static void flip_handler(int fd_, unsigned int sequence, unsigned int tv_sec,
> > > + unsigned int tv_usec, void *_data)
> > > +{
> > > + static unsigned int last_ms;
> > > + unsigned int cur_ms;
> > > +
> > > + igt_assert(_data == NULL);
> > > +
> > > + cur_ms = tv_sec * 1000 + tv_usec / 1000;
> > > +
> > > + igt_debug("Flip interval: %dms\n", cur_ms - last_ms);
> > > +
> > > + last_ms = cur_ms;
> > > +
> > > + igt_debug("Flip event received, fd:%d seq:%u tv_sec:%u, "
> > > + "tv_usec:%u data:%p\n",
> > > + fd_, sequence, tv_sec, tv_usec, _data);
> >
> > In my original implementation these messages were under if (0) because
> > they significantly affected FPS.
> >
> > I think we should probably have some kind of igt_assert() here instead
> > of printing stuff. This program is supposed to achieve thousands of
> > frames per second even on slow hardware (I think I got 2500 on a slow
> > GLK?). Any intervals that resemble the monitor refresh rate by a 10x
> > error rate are probably failures.
> >
>
> Does this mean if vblank interval is 16ms, then intervals of
> >1.6ms is a failure? Did I understand this right or Am I missing
> something here.
This test is supposed to flip waaaaay faster than the monitor refresh
rate. So if the refresh rate is 60hz, we should be getting events much
faster than once per 16ms. We need to define a threshold where we can
be confident that issues are detected (no false positives and no false
negatives). I just came up with the 10x thing because I feel it's safe,
but maybe I'm wrong and 10x is not the best value, I only ever tested
one machine. This is the kind of thing we may have to tune after the
test is in production by CI.
>
> > > +}
> > > +
> > > +static void wait_flip_event(data_t *data)
> > > +{
> > > + int ret;
> > > + drmEventContext evctx;
> > > + struct pollfd pfd;
> > > + static int timeouts = 0;
> > > +
> > > + evctx.version = 2;
> > > + evctx.vblank_handler = NULL;
> > > + evctx.page_flip_handler = flip_handler;
> > > +
> > > + pfd.fd = data->drm_fd;
> > > + pfd.events = POLLIN;
> > > + pfd.revents = 0;
> > > +
> > > + ret = poll(&pfd, 1, 10000);
> >
> > 10s timeouts are not igt-friendly. I used a giant number because I
> > wanted to read the logs. IGT doesn't need nor want 10s timeouts.
> >
>
> Will update this.
>
> > > +
> > > + switch (ret) {
> > > + case 0:
> > > + timeouts++;
> > > + igt_debug("Poll timeout %d!\n", timeouts);
> >
> > I don't remember exactly, but I think a timeout was a failure? Maybe
> > igt_assert() here?
> >
> >
> > > + break;
> > > + case 1:
> > > + ret = drmHandleEvent(data->drm_fd, &evctx);
> > > + igt_assert(ret == 0);
> > > + break;
> > > + default:
> > > + /* unexpected */
> > > + igt_assert(0);
> > > + }
> > > +}
> > > +
> > > +static void make_fb(data_t *data, struct igt_fb *fb,
> > > + drmModeConnectorPtr connector, int index)
> > > +{
> > > + uint32_t width, height;
> > > + int rec_width;
> > > +
> > > + width = connector->modes[0].hdisplay;
> > > + height = connector->modes[0].vdisplay;
> > > +
> > > + rec_width = width / (BUFS * 2);
> > > +
> > > + igt_create_fb(data->drm_fd, width, height, DRM_FORMAT_ARGB8888,
> > > + LOCAL_I915_FORMAT_MOD_X_TILED, fb);
> > > + igt_draw_fill_fb(data->drm_fd, fb, 0x88);
> > > + igt_draw_rect_fb(data->drm_fd, NULL, NULL, fb, IGT_DRAW_MMAP_CPU,
> > > + rec_width*2+rec_width*index, height/4,
> > > + rec_width, height/2, rand());
> >
> > Please follow the IGT coding style regarding spaces.
> >
>
> Will update this accordingly.
>
> > > +}
> > > +
> > > +static void test_async_flip(data_t *data, bool wait_for_flips)
> > > +{
> > > + struct igt_fb bufs[BUFS];
> > > + drmModeResPtr res;
> > > + drmModeConnectorPtr connector;
> > > + uint32_t crtc_id;
> > > + int i, ret, frame, warm_end_frame;
> > > + long long int fps;
> > > + struct timeval start, end, diff;
> > > + bool warming_up = true;
> > > +
> >
> > From here...
> >
> > > + res = drmModeGetResources(data->drm_fd);
> > > + igt_assert(res);
> > > +
> > > + kmstest_unset_all_crtcs(data->drm_fd, res);
> > > +
> > > + connector = find_connector_for_modeset(data);
> > > + crtc_id = kmstest_find_crtc_for_connector(data->drm_fd,
> > > + res, connector, 0);
> > > +
> > > + for (i = 0; i < BUFS; i++)
> > > + make_fb(data, &bufs[i], connector, i);
> > > +
> > > + ret = drmModeSetCrtc(data->drm_fd, crtc_id, bufs[0].fb_id, 0, 0,
> > > + &connector->connector_id, 1, &connector->modes[0]);
> > > + igt_assert(ret == 0);
> >
> > ... to here we probably don't want to repeat for every subtest,
> > especially if we add more in the future. Make them part of the fixture
> > and adjust the subtests to deal with it.
> >
>
> Sure. I'll make changes to the subtests accordingly.
>
> > > +
> > > + gettimeofday(&start, NULL);
> > > + frame = 1;
> > > + do {
> > > + int flags = DRM_MODE_PAGE_FLIP_ASYNC;
> > > +
> > > + if (wait_for_flips)
> > > + flags |= DRM_MODE_PAGE_FLIP_EVENT;
> > > +
> > > + ret = drmModePageFlip(data->drm_fd, crtc_id,
> > > + bufs[frame % 4].fb_id,
> > > + flags, NULL);
> > > +
> > > + igt_assert(ret == 0 || ret == -EBUSY);
> > > +
> > > + if (wait_for_flips)
> > > + wait_flip_event(data);
> > > +
> > > + gettimeofday(&end, NULL);
> > > + timersub(&end, &start, &diff);
> > > +
> > > + /* 2s of warm-up time for the freq to stabilize */
> > > + if (warming_up && diff.tv_sec >= WARM_UP_TIME) {
> > > + warming_up = false;
> > > + warm_end_frame = frame;
> > > + start = end;
> > > + }
> > > +
> > > + frame++;
> > > + } while (diff.tv_sec < RUN_TIME);
> > > +
> > > + fps = (frame - warm_end_frame) * 1000 / RUN_TIME;
> > > + igt_info("fps: %lld.%03lld\n", fps / 1000, fps % 1000);
> >
> > We have to igt_assert() that FPS is a value significantly bigger than
> > the mode refresh rate.
> >
>
> Didn't get this?
Printing the value is not enough, humans are not monitoring the IGT
output. We need assertions.
> FPS is expected to be bigger than RR right?
Right, like waaay bigger.
> Could you please help me here, to understand what I'm missing?
Something like:
int threshold = 10; /* TODO: may need tuning */
igt_assert(fps / 1000 > refresh_rate * threshold);
>
> > > +
> > > + for (i = 0; i < BUFS; i++)
> > > + igt_remove_fb(data->drm_fd, &bufs[i]);
> > > +}
> > > +
> > > +igt_main
> > > +{
> > > + data_t data;
> > > +
> > > + igt_fixture {
> > > + data.drm_fd = drm_open_driver_master(DRIVER_INTEL);
> > > + kmstest_set_vt_graphics_mode();
> > > + igt_display_require(&data.display, data.drm_fd);
> > > + igt_display_require_output(&data.display);
> >
> > We have to SKIP in case async flips are not supported by the Kernel.
> >
>
> Could you please let me know how I can do this? Since I can't test with
> a try_commit as we're using the page flip IOCTL in this test.
>
> Any other way with which I can check this?
A quick check on the libdrm headers shows me DRM_CAP_ASYNC_PAGE_FLIP.
Maybe that's enough? Otherwise go with whatever error drmModePageFlip()
returns.
>
> Thanks,
> Karthik
>
> > Thanks,
> > Paulo
> >
> >
> > > + }
> > > +
> > > + igt_subtest("async-flip-with-page-flip-events")
> > > + test_async_flip(&data, true);
> > > + igt_subtest("async-flip-without-page-flip-events")
> > > + test_async_flip(&data, false);
> > > +
> > > + igt_fixture
> > > + igt_display_fini(&data.display);
> > > +}
> > > diff --git a/tests/meson.build b/tests/meson.build
> > > index e882f4dc..0830b1c9 100644
> > > --- a/tests/meson.build
> > > +++ b/tests/meson.build
> > > @@ -13,6 +13,7 @@ test_progs = [
> > > 'fbdev',
> > > 'kms_3d',
> > > 'kms_addfb_basic',
> > > + 'kms_async',
> > > 'kms_atomic',
> > > 'kms_atomic_interruptible',
> > > 'kms_atomic_transition',
More information about the igt-dev
mailing list