From: "Borah, Chaitanya Kumar" <chaitanya.kumar.borah@intel.com>
To: Swati Sharma <swati2.sharma@intel.com>, <igt-dev@lists.freedesktop.org>
Subject: Re: [PATCH i-g-t 1/2] tests/intel: Add kms_hdmi_audio_bw test
Date: Tue, 1 Sep 2026 17:33:54 +0530 [thread overview]
Message-ID: <987b88f2-61a8-499e-91dc-c2561a850528@intel.com> (raw)
In-Reply-To: <6b27f04d-8f8a-4344-82e3-10917f78f0a2@intel.com>
On 8/31/2026 2:25 PM, Borah, Chaitanya Kumar wrote:
>
>
> On 8/19/2026 1:08 PM, Swati Sharma wrote:
>> Add a new IGT test to validate HDMI TMDS audio bandwidth constraints
>> under constrained horizontal blanking intervals.
>>
>> The test injects EDIDs declaring all 7 CEA sample rates (32kHz-192kHz)
>> and observes which rates the driver exposes via ELD (EDID-Like Data)
>> under varying BPC and channel configurations.
>>
>> Subtests:
>> - audio-bw-supported: Baseline with hblank=160 where audio bandwidth
>> is sufficient for all configurations.
>> - audio-bw-pruned: Constrained hblank=80 (CVT RB2) where the driver
>> must prune unsustainable sample rates or disable audio entirely.
>> - suspend-s3/s4-audio-recovery: Verify audio state and sample rates
>> are preserved across system suspend/resume.
>> - runtime-suspend-audio-recovery: Verify audio state is preserved
>> across DPMS off/on cycles.
>>
>> The test uses the HDMI TMDS bandwidth formula from the spec:
>> pkts_avail = FLOOR((CEIL(hblank * bpc/8) - overhead) / 32)
>> pkts_reqd = CEIL(R_AP * T_line)
>> where overhead=74 (HDCP 1.4 rekey always reserved by the driver).
>>
>> Assertions verify:
>> - Audio is active when bandwidth is available (pkts_avail > 0)
>> - Audio is inactive when bandwidth is exhausted (pkts_avail == 0)
>> - No sample rate requiring more packets than available appears in ELD
>>
>> Signed-off-by: Swati Sharma <swati2.sharma@intel.com>
>> Assisted-by: GitHub Copilot:Claude Opus 4.6
>> ---
>> tests/intel/kms_hdmi_audio_bw.c | 626 ++++++++++++++++++++++++++++++++
>> tests/meson.build | 1 +
>> 2 files changed, 627 insertions(+)
>> create mode 100644 tests/intel/kms_hdmi_audio_bw.c
>>
>> diff --git a/tests/intel/kms_hdmi_audio_bw.c b/tests/intel/
>> kms_hdmi_audio_bw.c
>> new file mode 100644
>> index 000000000..4b6f0d001
>> --- /dev/null
>> +++ b/tests/intel/kms_hdmi_audio_bw.c
>> @@ -0,0 +1,626 @@
>> +// SPDX-License-Identifier: MIT
>> +/*
>> + * Copyright © 2026 Intel Corporation
>> + */
>> +
>> +/**
>> + * TEST: kms hdmi audio bw
>> + * Category: Display
>> + * Description: Validate HDMI TMDS audio bandwidth constraints by
>> injecting
>> + * EDIDs with all sample rates declared and observing
>> which rates
>> + * the driver exposes (via ELD) under varying BPC /
>> channel /
>> + * hblank configurations.
>> + * Driver requirement: i915, xe
>> + * Mega feature: Display Audio
>> + */
>> +
>> +#include "config.h"
>> +
>> +#include <math.h>
>> +#include <string.h>
>> +
>> +#include "igt.h"
>> +#include "igt_edid.h"
>> +#include "igt_eld.h"
>> +#include "igt_aux.h"
>> +#include "xe/xe_query.h"
>
> What is this used for?
>
>> +
>> +/**
>> + * SUBTEST: audio-bw-supported
>> + * Description: Baseline test with hblank=160 where audio bandwidth is
>> + * sufficient for all BPC and channel combinations.
>> Verifies
>> + * that no sample rates are pruned.
>> + *
>> + * SUBTEST: audio-bw-pruned
>> + * Description: Constrained test with hblank=80 (CVT RB2) where audio
>> + * bandwidth is limited. Logs which sample rates are pruned
>> + * per BPC and channel combination.
>> + *
>> + * SUBTEST: suspend-%s-audio-recovery
>> + * Description: Validate audio state restoration after %arg[1] with
>> + * constrained hblank=80 and 12bpc.
>> + *
>> + * arg[1]:
>> + *
>> + * @s3: S3 (suspend to RAM)
>> + * @s4: S4 (hibernate)
>> + *
>> + * SUBTEST: runtime-suspend-audio-recovery
>> + * Description: Validate audio state restoration after runtime
>> suspend/resume
>> + * with constrained hblank=80 and 12bpc.
>> + */
>> +
>> +IGT_TEST_DESCRIPTION("Validate HDMI TMDS audio bandwidth constraints. "
>> + "EDIDs declare all sample rates (32k-192k); the test "
>> + "observes which rates survive in the ELD under "
>> + "constrained hblank timings.");
>> +
>> +typedef struct {
>> + int drm_fd;
>> + igt_display_t display;
>> + igt_output_t *output;
>> + igt_crtc_t *crtc;
>> + struct igt_fb fb;
>> +} data_t;
>> +
>> +/* All sample rates declared in the EDID SAD */
>> +#define ALL_SAMPLE_RATES (CEA_SAD_SAMPLING_RATE_32KHZ | \
>> + CEA_SAD_SAMPLING_RATE_44KHZ | \
>> + CEA_SAD_SAMPLING_RATE_48KHZ | \
>> + CEA_SAD_SAMPLING_RATE_88KHZ | \
>> + CEA_SAD_SAMPLING_RATE_96KHZ | \
>> + CEA_SAD_SAMPLING_RATE_176KHZ | \
>> + CEA_SAD_SAMPLING_RATE_192KHZ)
>> +
>> +struct rate_info {
>> + unsigned int flag;
>> + const char *name;
>> + int freq_hz;
>> +};
>> +
>> +static const struct rate_info rate_table[] = {
>> + { CEA_SAD_SAMPLING_RATE_32KHZ, "32k", 32000 },
>> + { CEA_SAD_SAMPLING_RATE_44KHZ, "44.1k", 44100 },
>> + { CEA_SAD_SAMPLING_RATE_48KHZ, "48k", 48000 },
>> + { CEA_SAD_SAMPLING_RATE_88KHZ, "88k", 88200 },
I missed this, lets keep the naming consistent "88.2k"
>> + { CEA_SAD_SAMPLING_RATE_96KHZ, "96k", 96000 },
>> + { CEA_SAD_SAMPLING_RATE_176KHZ, "176k", 176400 },
"176.4k"
Also see
https://lore.kernel.org/igt-dev/20260901113313.627816-1-chaitanya.kumar.borah@intel.com/T/#u
>> + { CEA_SAD_SAMPLING_RATE_192KHZ, "192k", 192000 },
>> +};
>> +
>> +#define ACR_RATE_MAX 1500
>> +#define TOLERANCE_AUDIOCLK_PPM 1000
>> +#define TOLERANCE_PIXELCLK 0.005
>> +#define HBLANK_OVERHEAD_STD 30
>> +#define HBLANK_OVERHEAD_HDCP14 74
>> +#define DI_PACKET_SIZE 32
>> +
>> +static void rates_to_str(unsigned int rates, char *buf, size_t len)
>> +{
>> + int pos = 0;
>> +
>> + buf[0] = '\0';
>> + for (int i = 0; i < ARRAY_SIZE(rate_table); i++) {
>> + if (!(rates & rate_table[i].flag))
>> + continue;
>> + if (pos > 0)
>> + pos += snprintf(buf + pos, len - pos, ",");
>> + pos += snprintf(buf + pos, len - pos, "%s", rate_table[i].name);
>> + }
>> + if (pos == 0)
>> + snprintf(buf, len, "none");
>> +}
>> +
>> +static const int bpc_values[] = { 8, 10, 12 };
>> +static const int channel_values[] = { 2, 8 };
>> +
>> +/*
>> + * 1920x1080@60Hz CVT RB2 — hblank=80 (constrained)
>> + * Available Packets/Line = FLOOR(((BPC/8)*80 - 74) / 32)
>> + * 8bpc=0, 10bpc=0, 12bpc=1
>> + */
>> +static const drmModeModeInfo mode_1080p_hblank80 = {
>> + .clock = 133320,
>> + .hdisplay = 1920,
>> + .hsync_start = 1928,
>> + .hsync_end = 1960,
>> + .htotal = 2000, /* hblank = 80 */
>> + .vdisplay = 1080,
>> + .vsync_start = 1097,
>> + .vsync_end = 1105,
>> + .vtotal = 1111,
>> + .vrefresh = 60,
>> + .flags = DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_NVSYNC,
>> + .type = DRM_MODE_TYPE_DRIVER,
>> + .name = "1920x1080",
>> +};
>> +
>> +/*
>> + * 1920x1080@60Hz with hblank=160 (relaxed baseline)
>> + * Enough hblank for audio at any BPC.
>> + */
>> +static const drmModeModeInfo mode_1080p_hblank160 = {
>> + .clock = 148500,
>> + .hdisplay = 1920,
>> + .hsync_start = 1968,
>> + .hsync_end = 2000,
>> + .htotal = 2080, /* hblank = 160 */
>> + .vdisplay = 1080,
>> + .vsync_start = 1097,
>> + .vsync_end = 1105,
>> + .vtotal = 1111,
>> + .vrefresh = 60,
>> + .flags = DRM_MODE_FLAG_PHSYNC | DRM_MODE_FLAG_NVSYNC,
>> + .type = DRM_MODE_TYPE_DRIVER,
>> + .name = "1920x1080",
>> +};
>> +
>> +static igt_output_t *find_hdmi_output(igt_display_t *display)
>> +{
>> + igt_output_t *output;
>> +
>> + for_each_connected_output(display, output) {
>> + drmModeConnector *c = output->config.connector;
>> +
>> + if (c->connector_type == DRM_MODE_CONNECTOR_HDMIA ||
>> + c->connector_type == DRM_MODE_CONNECTOR_HDMIB)
>> + return output;
>> + }
>> +
>> + return NULL;
>> +}
>> +
>> +static int hblank_of(const drmModeModeInfo *mode)
>> +{
>> + return mode->htotal - mode->hdisplay;
>> +}
>> +
>> +/*
>> + * Driver always reserves HDCP 1.4 rekey overhead (74 clocks) even when
>> + * HDCP is not active: 30 (standard) + 44 (HDCP 1.4 rekey quiet period).
>> + * FIXME: once driver exposes HDCP state, use 30 for no HDCP, 74 for
>> HDCP 1.4.
>> + */
>> +static int avail_pkts_per_line(int bpc, int hblank)
>> +{
>> + int overhead = HBLANK_OVERHEAD_HDCP14;
>> + int tb_blank = (bpc * hblank + 7) / 8; /* CEIL(hblank * bpc/8) */
>> + int avail = (tb_blank - overhead) / DI_PACKET_SIZE;
>> +
>> + return avail > 0 ? avail : 0;
>> +}
>> +
>> +/* Packets required per line for a given audio rate and channel
>> layout */
>> +static int required_pkts_per_line(const drmModeModeInfo *mode, int
>> freq_hz,
>> + int channels)
>> +{
>> + double ap = (channels <= 2) ? 0.25 : 1.0;
>> + double f_pixel_max = mode->clock * 1000.0 * (1 +
>> TOLERANCE_PIXELCLK);
>> + double t_line = mode->htotal / f_pixel_max;
>> + double r_ap = ((freq_hz * ap) + (2 * ACR_RATE_MAX)) *
>> + (1 + TOLERANCE_AUDIOCLK_PPM / 1e6);
>> +
>> + double avg_pkts = r_ap * t_line;
>> +
>> + return (int)avg_pkts + (avg_pkts > (int)avg_pkts ? 1 : 0);
>> +}
>> +
>> +/*
>> + * Build a CEA EDID declaring all 7 sample rates in the SAD.
>> + * Deep-color flags in HDMI VSDB match the requested bpc.
>> + */
>> +static const struct edid *
>> +build_edid(int bpc, int audio_channels)
>> +{
>> + static unsigned char raw_edid[2 * EDID_BLOCK_SIZE];
>> + struct edid *edid;
>> + struct edid_ext *ext;
>> + struct edid_cea *cea;
>> + struct edid_cea_data_block *block;
>> + struct cea_sad sad;
>> + struct hdmi_vsdb hdmi;
>
> nit: Let's call it vsdb.
>
>> + struct cea_speaker_alloc speakers;
>> + size_t offset = 0;
>> +
>> + memset(raw_edid, 0, sizeof(raw_edid));
>> + edid = (struct edid *)raw_edid;
>> + memcpy(edid, igt_kms_get_base_edid(), sizeof(struct edid));
>> + edid->extensions_len = 1;
>> +
>> + ext = &edid->extensions[0];
>> + cea = &ext->data.cea;
>> +
>> + if (audio_channels > 0) {
>> + cea_sad_init_pcm(&sad,
>> + audio_channels,
>> + ALL_SAMPLE_RATES,
>> + CEA_SAD_SAMPLE_SIZE_16 |
>> + CEA_SAD_SAMPLE_SIZE_24);
>> + block = (struct edid_cea_data_block *)&cea->data[offset];
>> + offset += edid_cea_data_block_set_sad(block, &sad, 1);
>> + }
>> +
>> + memset(&hdmi, 0, sizeof(hdmi));
>> + hdmi.src_phy_addr[0] = 0x10;
>> + hdmi.src_phy_addr[1] = 0x00;
>> + hdmi.flags1 = HDMI_VSDB_SUPPORTS_AI;
>> + hdmi.max_tdms_clock = 340000000 / (5 * 1000000);
>
> where do these number come from?
>
>> +
>> + switch (bpc) {
>> + case 12:
>> + hdmi.flags1 |= HDMI_VSDB_DC_36BIT;
>> + /* fall through */
>> + case 10:
>> + hdmi.flags1 |= HDMI_VSDB_DC_30BIT;
>> + /* fall through */
>> + case 8:
>> + break;
>> + }
>> +
>> + block = (struct edid_cea_data_block *)&cea->data[offset];
>> + offset += edid_cea_data_block_set_hdmi_vsdb(block, &hdmi,
>> + sizeof(hdmi));
>> +
>> + memset(&speakers, 0, sizeof(speakers));
>> + speakers.speakers = CEA_SPEAKER_FRONT_LEFT_RIGHT;
>> + if (audio_channels > 2)
>> + speakers.speakers |= CEA_SPEAKER_FRONT_CENTER |
>> + CEA_SPEAKER_LFE |
>> + CEA_SPEAKER_REAR_LEFT_RIGHT;
>
> Not really a problem for this test but we are adding 6 speakers for
> anything with greater than 2 channels.
>
>
>> + block = (struct edid_cea_data_block *)&cea->data[offset];
>> + offset += edid_cea_data_block_set_speaker_alloc(block, &speakers);
>> +
>> + edid_ext_set_cea(ext, offset, 0,
>> + EDID_CEA_BASIC_AUDIO | EDID_CEA_UNDERSCAN |
>> + EDID_CEA_YCBCR444 | EDID_CEA_YCBCR422);
>> + edid_update_checksum(edid);
>> +
>> + return edid;
>> +}
>> +
>> +static void force_edid_and_connector(data_t *data, const struct edid
>> *edid)
>> +{
>> + kmstest_force_edid(data->drm_fd, data->output->config.connector,
>> edid);
>> + igt_skip_on_f(!kmstest_force_connector(data->drm_fd,
>> + data->output->config.connector,
>> + FORCE_CONNECTOR_ON),
>> + "Could not force HDMI connector on\n");
>> +}
>> +
>> +static void cleanup_connector(data_t *data)
>> +{
>> + if (data->output->pending_crtc) {
>> + igt_plane_t *primary;
>> +
>> + primary = igt_output_get_plane_type(data->output,
>> + DRM_PLANE_TYPE_PRIMARY);
>> + igt_plane_set_fb(primary, NULL);
>> + igt_output_set_crtc(data->output, NULL);
>> + igt_display_commit2(&data->display, COMMIT_ATOMIC);
>> + }
>> +
>> + igt_remove_fb(data->drm_fd, &data->fb);
>> +
>> + kmstest_force_connector(data->drm_fd,
>> + data->output->config.connector,
>> + FORCE_CONNECTOR_UNSPECIFIED);
>> + kmstest_force_edid(data->drm_fd,
>> + data->output->config.connector, NULL);
>> +}
>> +
>> +static int try_modeset(data_t *data, const drmModeModeInfo *mode)
>> +{
>> + igt_plane_t *primary;
>> + int ret;
>> +
>> + igt_display_reset(&data->display);
>> +
>> + igt_output_set_crtc(data->output, data->crtc);
>> + igt_output_override_mode(data->output, mode);
>> +
>> + primary = igt_output_get_plane_type(data->output,
>> + DRM_PLANE_TYPE_PRIMARY);
>> +
>> + igt_create_pattern_fb(data->drm_fd,
>> + mode->hdisplay, mode->vdisplay,
>> + DRM_FORMAT_XRGB8888, DRM_FORMAT_MOD_LINEAR,
>> + &data->fb);
>> + igt_plane_set_fb(primary, &data->fb);
>> +
>> + ret = igt_display_try_commit_atomic(&data->display,
>> + DRM_MODE_ATOMIC_ALLOW_MODESET,
>> + NULL);
>> + if (ret) {
>> + igt_plane_set_fb(primary, NULL);
>> + igt_output_set_crtc(data->output, NULL);
>> + igt_remove_fb(data->drm_fd, &data->fb);
>> + }
>> +
>> + return ret;
>> +}
>> +
>> +static bool audio_is_active(void)
>> +{
>> + if (!eld_is_supported())
>> + return false;
>> +
>> + return eld_has_igt();
>> +}
>> +
>> +static unsigned int get_eld_rates(void)
>> +{
>> + struct eld_entry eld;
>> +
>> + if (!eld_get_igt(&eld))
>> + return 0;
>> +
>> + if (eld.sads_len == 0)
>> + return 0;
>> +
>> + return eld.sads[0].rates;
>
> This is heavily dependent on the semantics that currently we only add
> one SAD in build_eld. Let's document this atleast.
>
>> +}
>> +
>> +static void log_eld_rates(unsigned int declared, unsigned int eld_rates)
>> +{
>> + char decl_str[128], eld_str[128], pruned_str[128];
>> + unsigned int pruned = declared & ~eld_rates;
>> +
>> + rates_to_str(declared, decl_str, sizeof(decl_str));
>> + rates_to_str(eld_rates, eld_str, sizeof(eld_str));
>> + rates_to_str(pruned, pruned_str, sizeof(pruned_str));
>> +
>> + igt_info(" SAD declared: %s\n", decl_str);
>> + igt_info(" ELD reports: %s\n", eld_str);
>> + if (pruned)
>> + igt_info(" Pruned: %s\n", pruned_str);
>> +}
>> +
>> +static void assert_per_rate(const drmModeModeInfo *mode, int channels,
>> + int pkts_avail, unsigned int eld_rates)
>> +{
>> + for (int i = 0; i < ARRAY_SIZE(rate_table); i++) {
>> + int req = required_pkts_per_line(mode, rate_table[i].freq_hz,
>> + channels);
>> + bool in_eld = eld_rates & rate_table[i].flag;
>> +
>> + /* A rate that can't fit must not appear in ELD */
>> + igt_assert_f(!(req > pkts_avail && in_eld),
>> + "%s: req=%d > avail=%d but rate present in ELD\n",
>> + rate_table[i].name, req, pkts_avail);
>
> This check does not protect againts cases where the driver over prunes.
> It asserts true all req <= pkts_avail cases, irrespective of in_eld.
>
>> + }
>> +}
>> +
>> +static void log_per_rate_analysis(const drmModeModeInfo *mode,
>> + int bpc, int channels,
>> + int pkts_avail, unsigned int eld_rates)
>> +{
>> + const char *layout = (channels <= 2) ? "L0" : "L1";
>> +
>> + igt_info(" %-6s %-3s pkts: avail=%d\n",
>> + "Rate", layout, pkts_avail);
>> +
>> + for (int i = 0; i < ARRAY_SIZE(rate_table); i++) {
>> + int req = required_pkts_per_line(mode, rate_table[i].freq_hz,
>> + channels);
>> + const char *expect = (req <= pkts_avail && pkts_avail > 0) ?
>> + "fit" : "NO";
>> + const char *eld_has = (eld_rates & rate_table[i].flag) ?
>> + "yes" : "no";
>> +
>> + igt_info(" %5s: req=%d fit=%s (ELD: %s)\n",
>> + rate_table[i].name, req, expect, eld_has);
>> + }
>> +}
>> +
>> +/* Run the BPC × channels matrix for a given mode/hblank. */
>> +static void test_audio_bw_matrix(data_t *data, const drmModeModeInfo
>> *mode)
>> +{
>> + int hblank = hblank_of(mode);
>> +
>> + igt_info("=== Audio BW matrix: %s hblank=%d ===\n",
>> + mode->name, hblank);
>> +
>> + for (int b = 0; b < ARRAY_SIZE(bpc_values); b++) {
>> + int bpc = bpc_values[b];
>> +
>> + for (int c = 0; c < ARRAY_SIZE(channel_values); c++) {
>> + int channels = channel_values[c];
>> + const struct edid *edid;
>> + int pkts, ret;
>> + bool audio;
>> + unsigned int eld_rates;
>> +
>> + edid = build_edid(bpc, channels);
>> + force_edid_and_connector(data, edid);
>> +
>> + igt_output_set_prop_value(data->output,
>> + IGT_CONNECTOR_MAX_BPC, bpc);
>> +
>> + pkts = avail_pkts_per_line(bpc, hblank);
>> +
>> + igt_info("\n %dbpc %dch hblank=%d avail_pkts=%d\n",
>> + bpc, channels, hblank, pkts);
>> +
>> + ret = try_modeset(data, mode);
>> +
>> + if (ret) {
>> + igt_info(" modeset: REJECTED\n");
>> + cleanup_connector(data);
>> + continue;
>> + }
>> +
>> + /* Allow ELD to propagate */
>> + usleep(200 * 1000);
>> +
>> + audio = audio_is_active();
>> + eld_rates = audio ? get_eld_rates() : 0;
>> +
>> + igt_info(" modeset: OK\n");
>> + igt_info(" audio: %s\n", audio ? "active" :
>> "inactive");
>> +
>> + igt_assert_f(!(pkts == 0 && audio),
>> + "Audio active with 0 available packets\n");
>> + igt_assert_f(!(pkts > 0 && !audio),
>> + "Audio inactive with %d available packets\n",
>> + pkts);
>> +
>> + if (audio) {
>> + log_eld_rates(ALL_SAMPLE_RATES, eld_rates);
>> + assert_per_rate(mode, channels, pkts,
>> + eld_rates);
>> + }
>> +
>> + log_per_rate_analysis(mode, bpc, channels,
>> + pkts, eld_rates);
>> +
>> + cleanup_connector(data);
>> + }
>> + }
>> +
>> + igt_info("\n=== End matrix ===\n");
>> +}
>> +
>> +static void test_audio_bw_supported(data_t *data)
>> +{
>> + test_audio_bw_matrix(data, &mode_1080p_hblank160);
>> +}
>> +
>> +static void test_audio_bw_pruned(data_t *data)
>> +{
>> + test_audio_bw_matrix(data, &mode_1080p_hblank80);
>> +}
>> +
>> +static void test_suspend_audio_recovery(data_t *data,
>> + enum igt_suspend_state state)
>> +{
>> + const struct edid *edid;
>> + bool audio_before, audio_after;
>> + unsigned int rates_before, rates_after;
>> + char before_str[128], after_str[128];
>> + int ret;
>> +
>> + edid = build_edid(12, 2);
>> + force_edid_and_connector(data, edid);
>> +
>> + igt_output_set_prop_value(data->output, IGT_CONNECTOR_MAX_BPC, 12);
>> +
>> + ret = try_modeset(data, &mode_1080p_hblank80);
>> + igt_require(ret == 0);
>> +
>> + usleep(200 * 1000);
>
> Is this empirical? I understand it is being already being used
> kms_hdmi_inject but 200ms looks like a lot.
>
>> +
>> + audio_before = audio_is_active();
>> + rates_before = audio_before ? get_eld_rates() : 0;
>> + rates_to_str(rates_before, before_str, sizeof(before_str));
>> + igt_info("Before suspend: audio=%d rates=%s\n",
>> + audio_before, before_str);
>> +
>> + igt_system_suspend_autoresume(state, SUSPEND_TEST_NONE);
>> +
>> + usleep(200 * 1000);
>> +
>> + audio_after = audio_is_active();
>> + rates_after = audio_after ? get_eld_rates() : 0;
>> + rates_to_str(rates_after, after_str, sizeof(after_str));
>> + igt_info("After suspend: audio=%d rates=%s\n",
>> + audio_after, after_str);
>> +
>> + igt_assert_eq(audio_before, audio_after);
>> + if (audio_before)
>> + igt_assert_eq(rates_before, rates_after);
>> +
>> + cleanup_connector(data);
>> +}
>> +
>> +static void test_runtime_suspend_audio(data_t *data)
>> +{
>> + const struct edid *edid;
>> + bool audio_before, audio_after;
>> + unsigned int rates_before, rates_after;
>> + char before_str[128], after_str[128];
>> + int ret;
>> +
>> + edid = build_edid(12, 2);
>> + force_edid_and_connector(data, edid);
>> +
>> + igt_output_set_prop_value(data->output, IGT_CONNECTOR_MAX_BPC, 12);
>> +
>> + ret = try_modeset(data, &mode_1080p_hblank80);
>> + igt_require(ret == 0);
>> +
>> + usleep(200 * 1000);
>> +
>> + audio_before = audio_is_active();
>> + rates_before = audio_before ? get_eld_rates() : 0;
>> + rates_to_str(rates_before, before_str, sizeof(before_str));
>> + igt_info("Before runtime suspend: audio=%d rates=%s\n",
>> + audio_before, before_str);
>> +
>> + kmstest_set_connector_dpms(data->drm_fd,
>> + data->output->config.connector,
>> + DRM_MODE_DPMS_OFF);
>> + usleep(500 * 1000);
>> + kmstest_set_connector_dpms(data->drm_fd,
>> + data->output->config.connector,
>> + DRM_MODE_DPMS_ON);
>> + usleep(500 * 1000);
>> +
>> + audio_after = audio_is_active();
>> + rates_after = audio_after ? get_eld_rates() : 0;
>> + rates_to_str(rates_after, after_str, sizeof(after_str));
>> + igt_info("After runtime suspend: audio=%d rates=%s\n",
>> + audio_after, after_str);
>> +
>> + igt_assert_eq(audio_before, audio_after);
>> + if (audio_before)
>> + igt_assert_eq(rates_before, rates_after);
>> +
>> + cleanup_connector(data);
>> +}
>
> test_suspend_audio_recovery and test_runtime_suspend_audio are almost
> identical. Can we converge them?
>
>> +
>> +int igt_main()
>> +{
>> + data_t data = {};
>> +
>> + igt_fixture() {
>> + data.drm_fd = drm_open_driver_master(DRIVER_INTEL | DRIVER_XE);
>> + igt_require(is_intel_device(data.drm_fd));
>> + kmstest_set_vt_graphics_mode();
>> + igt_display_require(&data.display, data.drm_fd);
>> +
>> + data.output = find_hdmi_output(&data.display);
>> + igt_require_f(data.output, "No HDMI connector found\n");
>> +
>> + data.crtc = igt_first_crtc(&data.display);
>> + igt_require_f(data.crtc, "No usable CRTC found\n");
>> + }
>> +
>> + igt_describe("Baseline: hblank=160, audio should be fully
>> supported "
>> + "for all BPC and channel configurations.");
>> + igt_subtest("audio-bw-supported")
>> + test_audio_bw_supported(&data);
>> +
>> + igt_describe("Constrained: hblank=80 (CVT RB2), audio may be
>> pruned "
>> + "or disabled depending on BPC.");
>> + igt_subtest("audio-bw-pruned")
>> + test_audio_bw_pruned(&data);
We have to re-think the sub test names.
audio-bw-supported seems to prune some SADs.
Opened device: /dev/dri/card0
Starting subtest: audio-bw-supported
=== Audio BW matrix: 1920x1080 hblank=160 ===
...
8bpc 8ch hblank=160 avail_pkts=2
modeset: OK
audio: active
SAD declared: 32k,44.1k,48k,88k,96k,176k,192k
ELD reports: 32k,44.1k,48k,88k,96k
Pruned: 176k,192k
Rate L1 pkts: avail=2
32k: req=1 fit=fit (ELD: yes)
44.1k: req=1 fit=fit (ELD: yes)
48k: req=1 fit=fit (ELD: yes)
88k: req=2 fit=fit (ELD: yes)
96k: req=2 fit=fit (ELD: yes)
176k: req=3 fit=NO (ELD: no)
192k: req=3 fit=NO (ELD: no)
While audio-bw-pruned has a configuration that prunes nothing.
Starting subtest: audio-bw-pruned
=== Audio BW matrix: 1920x1080 hblank=80 ===
...
12bpc 2ch hblank=80 avail_pkts=1
modeset: OK
audio: active
SAD declared: 32k,44.1k,48k,88k,96k,176k,192k
ELD reports: 32k,44.1k,48k,88k,96k,176k,192k
Rate L0 pkts: avail=1
32k: req=1 fit=fit (ELD: yes)
44.1k: req=1 fit=fit (ELD: yes)
48k: req=1 fit=fit (ELD: yes)
88k: req=1 fit=fit (ELD: yes)
96k: req=1 fit=fit (ELD: yes)
176k: req=1 fit=fit (ELD: yes)
192k: req=1 fit=fit (ELD: yes)
Either fold both of them to the same subtest or you can have names like
audio-bw-check-rb1 and audio-bw-check-rb2
==
Chaitanya
>> +
>> + igt_describe("Validate audio recovery after S3 suspend with "
>> + "constrained hblank.");
>> + igt_subtest("suspend-s3-audio-recovery")
>> + test_suspend_audio_recovery(&data, SUSPEND_STATE_MEM);
>
> The documentation for igt_suspend_state says "A memory sleep (non-
> hibernation) target state, respecting the system's mem_sleep default" if
> your intention is to deterministically go to S3 use SUSPEND_STATE_S3
> instead.
>
> May I know why only these two power states were selected in particular?
>> +
>> + igt_describe("Validate audio recovery after S4 hibernate with "
>> + "constrained hblank.");
>> + igt_subtest("suspend-s4-audio-recovery")
>> + test_suspend_audio_recovery(&data, SUSPEND_STATE_DISK);
>> +
>> + igt_describe("Validate audio recovery after runtime suspend with "
>> + "constrained hblank.");
>> + igt_subtest("runtime-suspend-audio-recovery")
>> + test_runtime_suspend_audio(&data);
>> +
>> + igt_fixture() {
>> + igt_display_fini(&data.display);
>> + drm_close_driver(data.drm_fd);
>> + }
>> +}
>> diff --git a/tests/meson.build b/tests/meson.build
>> index a62f447df..facb7ab5d 100644
>> --- a/tests/meson.build
>> +++ b/tests/meson.build
>> @@ -259,6 +259,7 @@ intel_kms_progs = [
>> 'kms_fbc_dirty_rect',
>> 'kms_fbcon_fbt',
>> 'kms_fence_pin_leak',
>> + 'kms_hdmi_audio_bw',
>
> needs to be alphabetical order?
>
>> 'kms_flip_scaled_crc',
>> 'kms_flip_tiling',
>> 'kms_frontbuffer_tracking',
>
next prev parent reply other threads:[~2026-09-01 12:04 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-08-19 7:38 [PATCH i-g-t 1/2] tests/intel: Add kms_hdmi_audio_bw test Swati Sharma
2026-08-19 7:38 ` [PATCH i-g-t 2/2] tests/intel: Add mode-rejected-max-dotclock subtest to kms_cdclk Swati Sharma
2026-08-31 9:32 ` Borah, Chaitanya Kumar
2026-08-19 9:45 ` ✓ Xe.CI.BAT: success for series starting with [i-g-t,1/2] tests/intel: Add kms_hdmi_audio_bw test Patchwork
2026-08-19 10:01 ` ✓ i915.CI.BAT: " Patchwork
2026-08-19 12:28 ` ✓ Xe.CI.FULL: " Patchwork
2026-08-19 15:13 ` ✓ i915.CI.Full: " Patchwork
2026-08-31 8:55 ` [PATCH i-g-t 1/2] " Borah, Chaitanya Kumar
2026-09-01 12:03 ` Borah, Chaitanya Kumar [this message]
-- strict thread matches above, loose matches on Subject: below --
2026-08-14 12:16 Swati Sharma
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=987b88f2-61a8-499e-91dc-c2561a850528@intel.com \
--to=chaitanya.kumar.borah@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=swati2.sharma@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox