From cfa0904a35fd0231f4d05da0190f0a22ed881cce Mon Sep 17 00:00:00 2001 From: Fangzhi Zuo Date: Thu, 18 Sep 2025 16:25:45 -0400 Subject: drm/amd/display: Prevent Gating DTBCLK before It Is Properly Latched [why] 1. With allow_0_dtb_clk enabled, the time required to latch DTBCLK to 600 MHz depends on the SMU. If DTBCLK is not latched to 600 MHz before set_mode completes, gating DTBCLK causes the DP2 sink to lose its clock source. 2. The existing DTBCLK gating sequence ungates DTBCLK based on both pix_clk and ref_dtbclk, but gates DTBCLK when either pix_clk or ref_dtbclk is zero. pix_clk can be zero outside the set_mode sequence before DTBCLK is properly latched, which can lead to DTBCLK being gated by mistake. [how] Consider both pixel_clk and ref_dtbclk when determining when it is safe to gate DTBCLK; this is more accurate. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4701 Fixes: 5949e7c4890c ("drm/amd/display: Enable Dynamic DTBCLK Switch") Reviewed-by: Charlene Liu Reviewed-by: Aurabindo Pillai Signed-off-by: Fangzhi Zuo Signed-off-by: Roman Li Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit d04eb0c402780ca037b62a6aecf23b863545ebca) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/dc/clk_mgr/dcn35/dcn35_clk_mgr.c | 4 +++- drivers/gpu/drm/amd/display/dc/dccg/dcn35/dcn35_dccg.c | 2 +- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/clk_mgr/dcn35/dcn35_clk_mgr.c b/drivers/gpu/drm/amd/display/dc/clk_mgr/dcn35/dcn35_clk_mgr.c index b11383fba35f..1eb04772f5da 100644 --- a/drivers/gpu/drm/amd/display/dc/clk_mgr/dcn35/dcn35_clk_mgr.c +++ b/drivers/gpu/drm/amd/display/dc/clk_mgr/dcn35/dcn35_clk_mgr.c @@ -394,6 +394,8 @@ void dcn35_update_clocks(struct clk_mgr *clk_mgr_base, display_count = dcn35_get_active_display_cnt_wa(dc, context, &all_active_disps); if (new_clocks->dtbclk_en && !new_clocks->ref_dtbclk_khz) new_clocks->ref_dtbclk_khz = 600000; + else if (!new_clocks->dtbclk_en && new_clocks->ref_dtbclk_khz > 590000) + new_clocks->ref_dtbclk_khz = 0; /* * if it is safe to lower, but we are already in the lower state, we don't have to do anything @@ -435,7 +437,7 @@ void dcn35_update_clocks(struct clk_mgr *clk_mgr_base, actual_dtbclk = REG_READ(CLK1_CLK4_CURRENT_CNT); - if (actual_dtbclk) { + if (actual_dtbclk > 590000) { clk_mgr_base->clks.ref_dtbclk_khz = new_clocks->ref_dtbclk_khz; clk_mgr_base->clks.dtbclk_en = new_clocks->dtbclk_en; } diff --git a/drivers/gpu/drm/amd/display/dc/dccg/dcn35/dcn35_dccg.c b/drivers/gpu/drm/amd/display/dc/dccg/dcn35/dcn35_dccg.c index de6d62401362..c899c09ea31b 100644 --- a/drivers/gpu/drm/amd/display/dc/dccg/dcn35/dcn35_dccg.c +++ b/drivers/gpu/drm/amd/display/dc/dccg/dcn35/dcn35_dccg.c @@ -1411,7 +1411,7 @@ static void dccg35_set_dtbclk_dto( __func__, params->otg_inst, params->pixclk_khz, params->ref_dtbclk_khz, req_dtbclk_khz, phase, modulo); - } else { + } else if (!params->ref_dtbclk_khz && !req_dtbclk_khz) { switch (params->otg_inst) { case 0: REG_UPDATE(DCCG_GATE_DISABLE_CNTL5, DTBCLK_P0_GATE_DISABLE, 0); -- cgit v1.2.3 From 71ad9054c1f241be63f9d11df8cbd0aa0352fe16 Mon Sep 17 00:00:00 2001 From: "Mario Limonciello (AMD)" Date: Mon, 3 Nov 2025 11:17:44 -0600 Subject: drm/amd/display: Move sleep into each retry for retrieve_link_cap() [Why] When a monitor is booting it's possible that it isn't ready to retrieve link caps and this can lead to an EDID read failure: ``` [drm:retrieve_link_cap [amdgpu]] *ERROR* retrieve_link_cap: Read receiver caps dpcd data failed. amdgpu 0000:c5:00.0: [drm] *ERROR* No EDID read. ``` [How] Rather than msleep once and try a few times, msleep each time. Should be no changes for existing working monitors, but should correct reading caps on a monitor that is slow to boot. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4672 Reviewed-by: Alex Hung Signed-off-by: Mario Limonciello (AMD) Signed-off-by: Ivan Lipski Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 669dca37b3348a447db04bbdcbb3def94d5997cc) Cc: stable@vger.kernel.org --- .../gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c b/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c index b12c11bd6a14..4d0079b44de1 100644 --- a/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c +++ b/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c @@ -1734,12 +1734,13 @@ static bool retrieve_link_cap(struct dc_link *link) } dpcd_set_source_specific_data(link); - /* Sink may need to configure internals based on vendor, so allow some - * time before proceeding with possibly vendor specific transactions - */ - msleep(post_oui_delay); for (i = 0; i < read_dpcd_retry_cnt; i++) { + /* + * Sink may need to configure internals based on vendor, so allow some + * time before proceeding with possibly vendor specific transactions + */ + msleep(post_oui_delay); status = core_link_read_dpcd( link, DP_DPCD_REV, -- cgit v1.2.3 From 8612badc331bcab2068baefa69e1458085ed89e3 Mon Sep 17 00:00:00 2001 From: "Mario Limonciello (AMD)" Date: Mon, 3 Nov 2025 12:11:31 -0600 Subject: drm/amd/display: Increase DPCD read retries [Why] Empirical measurement of some monitors that fail to read EDID while booting shows that the number of retries with a 30ms delay between tries is as high as 16. [How] Increase number of retries to 20. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4672 Reviewed-by: Alex Hung Signed-off-by: Mario Limonciello (AMD) Signed-off-by: Ivan Lipski Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit ad1c59ad7cf74ec06e32fe2c330ac1e957222288) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c b/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c index 4d0079b44de1..eb262ce42e2d 100644 --- a/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c +++ b/drivers/gpu/drm/amd/display/dc/link/protocols/link_dp_capability.c @@ -1691,7 +1691,7 @@ static bool retrieve_link_cap(struct dc_link *link) union edp_configuration_cap edp_config_cap; union dp_downstream_port_present ds_port = { 0 }; enum dc_status status = DC_ERROR_UNEXPECTED; - uint32_t read_dpcd_retry_cnt = 3; + uint32_t read_dpcd_retry_cnt = 20; int i; struct dp_sink_hw_fw_revision dp_hw_fw_revision; const uint32_t post_oui_delay = 30; // 30ms -- cgit v1.2.3 From c97da4785b3bbc60c24cfd1ffea1d7c8b90ed743 Mon Sep 17 00:00:00 2001 From: Ivan Lipski Date: Thu, 30 Oct 2025 11:25:33 -0400 Subject: drm/amd/display: Add an HPD filter for HDMI MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit [Why] Some monitors perform rapid “autoscan” HPD re‑assertions right after a disconnect or powersaving mode enablement. These appear as a quick disconnect→reconnect with an identical EDID. Since Linux has no HDMI hotplug detection (HPD) filter, these quick reconnects are seen as hotplug events, which can unintentionally wake a system with DPMS off. An example: https://gitlab.freedesktop.org/drm/amd/-/issues/2876 Such 'fake reconnects' are considered when the interval between a disconnect and a connect is within 1500ms (experimentally chosen using several monitors), and the two connections have the same EDID. [How] Implement a time-based debounce mechanism: 1. On HDMI disconnect detection, instead of immediately processing the HPD event, save the current sink and schedule delayed work (default 1500ms) 2. If another HDMI disconnect HPD event arrives during the debounce period, it reschedules the pending work, ensuring only the final state is processed. 3. When the debounce timer expires, re-detect the display and compare the new sink with the cached one using EDID comparison. 4. If sinks match (same EDID), this was a spontaneous HPD toggle: - Update connector state internally - Skip hotplug event to prevent desktop rearrangement If sinks differ, this was a real display change: - Process normally with the hotplug event The debounce delay is configurable via module parameter 'hdmi_hpd_debounce_delay_ms'. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/2876 Reviewed-by: Sun peng (Leo) Li Signed-off-by: Ivan Lipski Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit c918e75e1ed95be76f8e3156a411188f650fe03f) --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 138 ++++++++++++++++++++++ drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h | 6 + 2 files changed, 144 insertions(+) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index 91c0188a29b2..2a7a491a62e0 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -3859,6 +3859,97 @@ void amdgpu_dm_update_connector_after_detect( update_subconnector_property(aconnector); } +static bool are_sinks_equal(const struct dc_sink *sink1, const struct dc_sink *sink2) +{ + if (!sink1 || !sink2) + return false; + if (sink1->sink_signal != sink2->sink_signal) + return false; + + if (sink1->dc_edid.length != sink2->dc_edid.length) + return false; + + if (memcmp(sink1->dc_edid.raw_edid, sink2->dc_edid.raw_edid, + sink1->dc_edid.length) != 0) + return false; + return true; +} + + +/** + * DOC: hdmi_hpd_debounce_work + * + * HDMI HPD debounce delay in milliseconds. When an HDMI display toggles HPD + * (such as during power save transitions), this delay determines how long to + * wait before processing the HPD event. This allows distinguishing between a + * physical unplug (>hdmi_hpd_debounce_delay) + * and a spontaneous RX HPD toggle (base; + struct drm_device *dev = connector->dev; + struct amdgpu_device *adev = drm_to_adev(dev); + struct dc *dc = aconnector->dc_link->ctx->dc; + bool fake_reconnect = false; + bool reallow_idle = false; + bool ret = false; + guard(mutex)(&aconnector->hpd_lock); + + /* Re-detect the display */ + scoped_guard(mutex, &adev->dm.dc_lock) { + if (dc->caps.ips_support && dc->ctx->dmub_srv->idle_allowed) { + dc_allow_idle_optimizations(dc, false); + reallow_idle = true; + } + ret = dc_link_detect(aconnector->dc_link, DETECT_REASON_HPD); + } + + if (ret) { + /* Apply workaround delay for certain panels */ + apply_delay_after_dpcd_poweroff(adev, aconnector->dc_sink); + /* Compare sinks to determine if this was a spontaneous HPD toggle */ + if (are_sinks_equal(aconnector->dc_link->local_sink, aconnector->hdmi_prev_sink)) { + /* + * Sinks match - this was a spontaneous HDMI HPD toggle. + */ + drm_dbg_kms(dev, "HDMI HPD: Sink unchanged after debounce, internal re-enable\n"); + fake_reconnect = true; + } + + /* Update connector state */ + amdgpu_dm_update_connector_after_detect(aconnector); + + drm_modeset_lock_all(dev); + dm_restore_drm_connector_state(dev, connector); + drm_modeset_unlock_all(dev); + + /* Only notify OS if sink actually changed */ + if (!fake_reconnect && aconnector->base.force == DRM_FORCE_UNSPECIFIED) + drm_kms_helper_hotplug_event(dev); + } + + /* Release the cached sink reference */ + if (aconnector->hdmi_prev_sink) { + dc_sink_release(aconnector->hdmi_prev_sink); + aconnector->hdmi_prev_sink = NULL; + } + + scoped_guard(mutex, &adev->dm.dc_lock) { + if (reallow_idle && dc->caps.ips_support) + dc_allow_idle_optimizations(dc, true); + } +} + static void handle_hpd_irq_helper(struct amdgpu_dm_connector *aconnector) { struct drm_connector *connector = &aconnector->base; @@ -3868,6 +3959,7 @@ static void handle_hpd_irq_helper(struct amdgpu_dm_connector *aconnector) struct dm_connector_state *dm_con_state = to_dm_connector_state(connector->state); struct dc *dc = aconnector->dc_link->ctx->dc; bool ret = false; + bool debounce_required = false; if (adev->dm.disable_hpd_irq) return; @@ -3890,6 +3982,14 @@ static void handle_hpd_irq_helper(struct amdgpu_dm_connector *aconnector) if (!dc_link_detect_connection_type(aconnector->dc_link, &new_connection_type)) drm_err(adev_to_drm(adev), "KMS: Failed to detect connector\n"); + /* + * Check for HDMI disconnect with debounce enabled. + */ + debounce_required = (aconnector->hdmi_hpd_debounce_delay_ms > 0 && + dc_is_hdmi_signal(aconnector->dc_link->connector_signal) && + new_connection_type == dc_connection_none && + aconnector->dc_link->local_sink != NULL); + if (aconnector->base.force && new_connection_type == dc_connection_none) { emulated_link_detect(aconnector->dc_link); @@ -3899,7 +3999,34 @@ static void handle_hpd_irq_helper(struct amdgpu_dm_connector *aconnector) if (aconnector->base.force == DRM_FORCE_UNSPECIFIED) drm_kms_helper_connector_hotplug_event(connector); + } else if (debounce_required) { + /* + * HDMI disconnect detected - schedule delayed work instead of + * processing immediately. This allows us to coalesce spurious + * HDMI signals from physical unplugs. + */ + drm_dbg_kms(dev, "HDMI HPD: Disconnect detected, scheduling debounce work (%u ms)\n", + aconnector->hdmi_hpd_debounce_delay_ms); + + /* Cache the current sink for later comparison */ + if (aconnector->hdmi_prev_sink) + dc_sink_release(aconnector->hdmi_prev_sink); + aconnector->hdmi_prev_sink = aconnector->dc_link->local_sink; + if (aconnector->hdmi_prev_sink) + dc_sink_retain(aconnector->hdmi_prev_sink); + + /* Schedule delayed detection. */ + if (mod_delayed_work(system_wq, + &aconnector->hdmi_hpd_debounce_work, + msecs_to_jiffies(aconnector->hdmi_hpd_debounce_delay_ms))) + drm_dbg_kms(dev, "HDMI HPD: Re-scheduled debounce work\n"); + } else { + + /* If the aconnector->hdmi_hpd_debounce_work is scheduled, exit early */ + if (delayed_work_pending(&aconnector->hdmi_hpd_debounce_work)) + return; + scoped_guard(mutex, &adev->dm.dc_lock) { dc_exit_ips_for_hw_access(dc); ret = dc_link_detect(aconnector->dc_link, DETECT_REASON_HPD); @@ -7388,6 +7515,13 @@ static void amdgpu_dm_connector_destroy(struct drm_connector *connector) if (aconnector->mst_mgr.dev) drm_dp_mst_topology_mgr_destroy(&aconnector->mst_mgr); + /* Cancel and flush any pending HDMI HPD debounce work */ + cancel_delayed_work_sync(&aconnector->hdmi_hpd_debounce_work); + if (aconnector->hdmi_prev_sink) { + dc_sink_release(aconnector->hdmi_prev_sink); + aconnector->hdmi_prev_sink = NULL; + } + if (aconnector->bl_idx != -1) { backlight_device_unregister(dm->backlight_dev[aconnector->bl_idx]); dm->backlight_dev[aconnector->bl_idx] = NULL; @@ -8549,6 +8683,10 @@ void amdgpu_dm_connector_init_helper(struct amdgpu_display_manager *dm, mutex_init(&aconnector->hpd_lock); mutex_init(&aconnector->handle_mst_msg_ready); + aconnector->hdmi_hpd_debounce_delay_ms = AMDGPU_DM_HDMI_HPD_DEBOUNCE_MS; + INIT_DELAYED_WORK(&aconnector->hdmi_hpd_debounce_work, hdmi_hpd_debounce_work); + aconnector->hdmi_prev_sink = NULL; + /* * configure support HPD hot plug connector_>polled default value is 0 * which means HPD hot plug not supported diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h index db75e991ac7b..8ca738957598 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h @@ -59,6 +59,7 @@ #define AMDGPU_HDR_MULT_DEFAULT (0x100000000LL) +#define AMDGPU_DM_HDMI_HPD_DEBOUNCE_MS 1500 /* #include "include/amdgpu_dal_power_if.h" #include "amdgpu_dm_irq.h" @@ -819,6 +820,11 @@ struct amdgpu_dm_connector { bool pack_sdp_v1_3; enum adaptive_sync_type as_type; struct amdgpu_hdmi_vsdb_info vsdb_info; + + /* HDMI HPD debounce support */ + unsigned int hdmi_hpd_debounce_delay_ms; + struct delayed_work hdmi_hpd_debounce_work; + struct dc_sink *hdmi_prev_sink; }; static inline void amdgpu_dm_set_mst_status(uint8_t *status, -- cgit v1.2.3 From 5bab4c89390f32b2f491f49a151948cd226dd909 Mon Sep 17 00:00:00 2001 From: Ivan Lipski Date: Wed, 5 Nov 2025 15:27:42 -0500 Subject: drm/amd/display: Clear the CUR_ENABLE register on DCN20 on DPP5 [Why] On DCN20 & DCN30, the 6th DPP's & HUBP's are powered on permanently and cannot be power gated. Thus, when dpp_reset() is invoked for the DPP5, while it's still powered on, the cached cursor_state (dpp_base->pos.cur0_ctl.bits.cur0_enable) and the actual state (CUR0_ENABLE) bit are unsycned. This can cause a double cursor in full screen with non-native scaling. [How] Force disable cursor on DPP5 on plane powerdown for ASICs w/ 6 DPPs/HUBPs. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4673 Reviewed-by: Aric Cyr Signed-off-by: Ivan Lipski Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 79b3c037f972dcb13e325a8eabfb8da835764e15) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/dc/hwss/dcn20/dcn20_hwseq.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn20/dcn20_hwseq.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn20/dcn20_hwseq.c index 9477c9f9e196..59c42db5382e 100644 --- a/drivers/gpu/drm/amd/display/dc/hwss/dcn20/dcn20_hwseq.c +++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn20/dcn20_hwseq.c @@ -614,6 +614,14 @@ void dcn20_dpp_pg_control( * DOMAIN11_PGFSM_PWR_STATUS, pwr_status, * 1, 1000); */ + + /* Force disable cursor on plane powerdown on DPP 5 using dpp_force_disable_cursor */ + if (!power_on) { + struct dpp *dpp5 = hws->ctx->dc->res_pool->dpps[dpp_inst]; + if (dpp5 && dpp5->funcs->dpp_force_disable_cursor) + dpp5->funcs->dpp_force_disable_cursor(dpp5); + } + break; default: BREAK_TO_DEBUGGER(); -- cgit v1.2.3 From 1788ef30725da53face7e311cdf62ad65fababcd Mon Sep 17 00:00:00 2001 From: Fangzhi Zuo Date: Fri, 7 Nov 2025 15:01:30 -0500 Subject: drm/amd/display: Fix pbn to kbps Conversion [Why] Existing routine has two conversion sequence, pbn_to_kbps and kbps_to_pbn with margin. Non of those has without-margin calculation. kbps_to_pbn with margin conversion includes fec overhead which has already been included in pbn_div calculation with 0.994 factor considered. It is a double counted fec overhead factor that causes potential bw loss. [How] Add without-margin calculation. Fix fec overhead double counted issue. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3735 Reviewed-by: Aurabindo Pillai Signed-off-by: Fangzhi Zuo Signed-off-by: Ivan Lipski Tested-by: Dan Wheeler Signed-off-by: Alex Deucher (cherry picked from commit e0dec00f3d05e8c0eceaaebfdca217f8d10d380c) Cc: stable@vger.kernel.org --- .../amd/display/amdgpu_dm/amdgpu_dm_mst_types.c | 59 +++++++++------------- 1 file changed, 23 insertions(+), 36 deletions(-) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c index 5e92eaa67aa3..dbd1da4d85d3 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c @@ -884,26 +884,28 @@ struct dsc_mst_fairness_params { }; #if defined(CONFIG_DRM_AMD_DC_FP) -static uint16_t get_fec_overhead_multiplier(struct dc_link *dc_link) +static uint64_t kbps_to_pbn(int kbps, bool is_peak_pbn) { - u8 link_coding_cap; - uint16_t fec_overhead_multiplier_x1000 = PBN_FEC_OVERHEAD_MULTIPLIER_8B_10B; + uint64_t effective_kbps = (uint64_t)kbps; - link_coding_cap = dc_link_dp_mst_decide_link_encoding_format(dc_link); - if (link_coding_cap == DP_128b_132b_ENCODING) - fec_overhead_multiplier_x1000 = PBN_FEC_OVERHEAD_MULTIPLIER_128B_132B; + if (is_peak_pbn) { // add 0.6% (1006/1000) overhead into effective kbps + effective_kbps *= 1006; + effective_kbps = div_u64(effective_kbps, 1000); + } - return fec_overhead_multiplier_x1000; + return (uint64_t) DIV64_U64_ROUND_UP(effective_kbps * 64, (54 * 8 * 1000)); } -static int kbps_to_peak_pbn(int kbps, uint16_t fec_overhead_multiplier_x1000) +static uint32_t pbn_to_kbps(unsigned int pbn, bool with_margin) { - u64 peak_kbps = kbps; + uint64_t pbn_effective = (uint64_t)pbn; + + if (with_margin) // deduct 0.6% (994/1000) overhead from effective pbn + pbn_effective *= (1000000 / PEAK_FACTOR_X1000); + else + pbn_effective *= 1000; - peak_kbps *= 1006; - peak_kbps *= fec_overhead_multiplier_x1000; - peak_kbps = div_u64(peak_kbps, 1000 * 1000); - return (int) DIV64_U64_ROUND_UP(peak_kbps * 64, (54 * 8 * 1000)); + return DIV_U64_ROUND_UP(pbn_effective * 8 * 54, 64); } static void set_dsc_configs_from_fairness_vars(struct dsc_mst_fairness_params *params, @@ -974,7 +976,7 @@ static int bpp_x16_from_pbn(struct dsc_mst_fairness_params param, int pbn) dc_dsc_get_default_config_option(param.sink->ctx->dc, &dsc_options); dsc_options.max_target_bpp_limit_override_x16 = drm_connector->display_info.max_dsc_bpp * 16; - kbps = div_u64((u64)pbn * 994 * 8 * 54, 64); + kbps = pbn_to_kbps(pbn, false); dc_dsc_compute_config( param.sink->ctx->dc->res_pool->dscs[0], ¶m.sink->dsc_caps.dsc_dec_caps, @@ -1003,12 +1005,11 @@ static int increase_dsc_bpp(struct drm_atomic_state *state, int link_timeslots_used; int fair_pbn_alloc; int ret = 0; - uint16_t fec_overhead_multiplier_x1000 = get_fec_overhead_multiplier(dc_link); for (i = 0; i < count; i++) { if (vars[i + k].dsc_enabled) { initial_slack[i] = - kbps_to_peak_pbn(params[i].bw_range.max_kbps, fec_overhead_multiplier_x1000) - vars[i + k].pbn; + kbps_to_pbn(params[i].bw_range.max_kbps, false) - vars[i + k].pbn; bpp_increased[i] = false; remaining_to_increase += 1; } else { @@ -1104,7 +1105,6 @@ static int try_disable_dsc(struct drm_atomic_state *state, int next_index; int remaining_to_try = 0; int ret; - uint16_t fec_overhead_multiplier_x1000 = get_fec_overhead_multiplier(dc_link); int var_pbn; for (i = 0; i < count; i++) { @@ -1137,7 +1137,7 @@ static int try_disable_dsc(struct drm_atomic_state *state, DRM_DEBUG_DRIVER("MST_DSC index #%d, try no compression\n", next_index); var_pbn = vars[next_index].pbn; - vars[next_index].pbn = kbps_to_peak_pbn(params[next_index].bw_range.stream_kbps, fec_overhead_multiplier_x1000); + vars[next_index].pbn = kbps_to_pbn(params[next_index].bw_range.stream_kbps, true); ret = drm_dp_atomic_find_time_slots(state, params[next_index].port->mgr, params[next_index].port, @@ -1197,7 +1197,6 @@ static int compute_mst_dsc_configs_for_link(struct drm_atomic_state *state, int count = 0; int i, k, ret; bool debugfs_overwrite = false; - uint16_t fec_overhead_multiplier_x1000 = get_fec_overhead_multiplier(dc_link); struct drm_connector_state *new_conn_state; memset(params, 0, sizeof(params)); @@ -1278,7 +1277,7 @@ static int compute_mst_dsc_configs_for_link(struct drm_atomic_state *state, DRM_DEBUG_DRIVER("MST_DSC Try no compression\n"); for (i = 0; i < count; i++) { vars[i + k].aconnector = params[i].aconnector; - vars[i + k].pbn = kbps_to_peak_pbn(params[i].bw_range.stream_kbps, fec_overhead_multiplier_x1000); + vars[i + k].pbn = kbps_to_pbn(params[i].bw_range.stream_kbps, false); vars[i + k].dsc_enabled = false; vars[i + k].bpp_x16 = 0; ret = drm_dp_atomic_find_time_slots(state, params[i].port->mgr, params[i].port, @@ -1300,7 +1299,7 @@ static int compute_mst_dsc_configs_for_link(struct drm_atomic_state *state, DRM_DEBUG_DRIVER("MST_DSC Try max compression\n"); for (i = 0; i < count; i++) { if (params[i].compression_possible && params[i].clock_force_enable != DSC_CLK_FORCE_DISABLE) { - vars[i + k].pbn = kbps_to_peak_pbn(params[i].bw_range.min_kbps, fec_overhead_multiplier_x1000); + vars[i + k].pbn = kbps_to_pbn(params[i].bw_range.min_kbps, false); vars[i + k].dsc_enabled = true; vars[i + k].bpp_x16 = params[i].bw_range.min_target_bpp_x16; ret = drm_dp_atomic_find_time_slots(state, params[i].port->mgr, @@ -1308,7 +1307,7 @@ static int compute_mst_dsc_configs_for_link(struct drm_atomic_state *state, if (ret < 0) return ret; } else { - vars[i + k].pbn = kbps_to_peak_pbn(params[i].bw_range.stream_kbps, fec_overhead_multiplier_x1000); + vars[i + k].pbn = kbps_to_pbn(params[i].bw_range.stream_kbps, false); vars[i + k].dsc_enabled = false; vars[i + k].bpp_x16 = 0; ret = drm_dp_atomic_find_time_slots(state, params[i].port->mgr, @@ -1763,18 +1762,6 @@ clean_exit: return ret; } -static uint32_t kbps_from_pbn(unsigned int pbn) -{ - uint64_t kbps = (uint64_t)pbn; - - kbps *= (1000000 / PEAK_FACTOR_X1000); - kbps *= 8; - kbps *= 54; - kbps /= 64; - - return (uint32_t)kbps; -} - static bool is_dsc_common_config_possible(struct dc_stream_state *stream, struct dc_dsc_bw_range *bw_range) { @@ -1873,7 +1860,7 @@ enum dc_status dm_dp_mst_is_port_support_mode( dc_link_get_highest_encoding_format(stream->link)); cur_link_settings = stream->link->verified_link_cap; root_link_bw_in_kbps = dc_link_bandwidth_kbps(aconnector->dc_link, &cur_link_settings); - virtual_channel_bw_in_kbps = kbps_from_pbn(aconnector->mst_output_port->full_pbn); + virtual_channel_bw_in_kbps = pbn_to_kbps(aconnector->mst_output_port->full_pbn, true); /* pick the end to end bw bottleneck */ end_to_end_bw_in_kbps = min(root_link_bw_in_kbps, virtual_channel_bw_in_kbps); @@ -1926,7 +1913,7 @@ enum dc_status dm_dp_mst_is_port_support_mode( immediate_upstream_port = aconnector->mst_output_port->parent->port_parent; if (immediate_upstream_port) { - virtual_channel_bw_in_kbps = kbps_from_pbn(immediate_upstream_port->full_pbn); + virtual_channel_bw_in_kbps = pbn_to_kbps(immediate_upstream_port->full_pbn, true); virtual_channel_bw_in_kbps = min(root_link_bw_in_kbps, virtual_channel_bw_in_kbps); } else { /* For topology LCT 1 case - only one mstb*/ -- cgit v1.2.3 From 9eb00b5f5697bd56baa3222c7a1426fa15bacfb5 Mon Sep 17 00:00:00 2001 From: Robert McClinton Date: Sun, 16 Nov 2025 12:33:21 -0500 Subject: drm/radeon: delete radeon_fence_process in is_signaled, no deadlock Delete the attempt to progress the queue when checking if fence is signaled. This avoids deadlock. dma-fence_ops::signaled can be called with the fence lock in unknown state. For radeon, the fence lock is also the wait queue lock. This can cause a self deadlock when signaled() tries to make forward progress on the wait queue. But advancing the queue is unneeded because incorrectly returning false from signaled() is perfectly acceptable. Link: https://github.com/brave/brave-browser/issues/49182 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4641 Cc: Alex Deucher Signed-off-by: Robert McClinton Signed-off-by: Alex Deucher (cherry picked from commit 527ba26e50ec2ca2be9c7c82f3ad42998a75d0db) Cc: stable@vger.kernel.org --- drivers/gpu/drm/radeon/radeon_fence.c | 7 ------- 1 file changed, 7 deletions(-) diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c index 5b5b54e876d4..167d6f122b8e 100644 --- a/drivers/gpu/drm/radeon/radeon_fence.c +++ b/drivers/gpu/drm/radeon/radeon_fence.c @@ -360,13 +360,6 @@ static bool radeon_fence_is_signaled(struct dma_fence *f) if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq) return true; - if (down_read_trylock(&rdev->exclusive_lock)) { - radeon_fence_process(rdev, ring); - up_read(&rdev->exclusive_lock); - - if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq) - return true; - } return false; } -- cgit v1.2.3 From 31ab31433c9bd2f255c48dc6cb9a99845c58b1e4 Mon Sep 17 00:00:00 2001 From: Mario Limonciello Date: Tue, 18 Nov 2025 07:18:10 -0600 Subject: drm/amd: Skip power ungate during suspend for VPE During the suspend sequence VPE is already going to be power gated as part of vpe_suspend(). It's unnecessary to call during calls to amdgpu_device_set_pg_state(). It actually can expose a race condition with the firmware if s0i3 sequence starts as well. Drop these calls. Cc: Peyton.Lee@amd.com Reviewed-by: Alex Deucher Signed-off-by: Mario Limonciello Signed-off-by: Alex Deucher (cherry picked from commit 2a6c826cfeedd7714611ac115371a959ead55bda) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 2819aceaab74..076bbc09f30c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -3414,10 +3414,11 @@ int amdgpu_device_set_pg_state(struct amdgpu_device *adev, (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX || adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA)) continue; - /* skip CG for VCE/UVD, it's handled specially */ + /* skip CG for VCE/UVD/VPE, it's handled specially */ if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && + adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VPE && adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && adev->ip_blocks[i].version->funcs->set_powergating_state) { /* enable powergating to save power */ -- cgit v1.2.3 From 80d8a9ad1587b64c545d515ab6cb7ecb9908e1b3 Mon Sep 17 00:00:00 2001 From: Yifan Zha Date: Fri, 14 Nov 2025 17:48:58 +0800 Subject: drm/amdgpu: Skip emit de meta data on gfx11 with rs64 enabled [Why] Accoreding to CP updated to RS64 on gfx11, WRITE_DATA with PREEMPTION_META_MEMORY(dst_sel=8) is illegal for CP FW. That packet is used for MCBP on F32 based system. So it would lead to incorrect GRBM write and FW is not handling that extra case correctly. [How] With gfx11 rs64 enabled, skip emit de meta data. Signed-off-by: Yifan Zha Acked-by: Alex Deucher Signed-off-by: Alex Deucher (cherry picked from commit 8366cd442d226463e673bed5d199df916f4ecbcf) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index d61eb9f187c6..f2be16e700c4 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -5872,9 +5872,9 @@ static void gfx_v11_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, if (flags & AMDGPU_IB_PREEMPTED) control |= INDIRECT_BUFFER_PRE_RESUME(1); - if (vmid) + if (vmid && !ring->adev->gfx.rs64_enable) gfx_v11_0_ring_emit_de_meta(ring, - (!amdgpu_sriov_vf(ring->adev) && flags & AMDGPU_IB_PREEMPTED) ? true : false); + !amdgpu_sriov_vf(ring->adev) && (flags & AMDGPU_IB_PREEMPTED)); } amdgpu_ring_write(ring, header); -- cgit v1.2.3 From a44592339397bc6715917997c6869bdedd1a7256 Mon Sep 17 00:00:00 2001 From: Timur Kristóf Date: Wed, 19 Nov 2025 10:25:42 +0100 Subject: drm/amdgpu/vm: Check PRT uAPI flag instead of PTE flag MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This fixes sparse mappings (aka. partially resident textures). Check the correct flags. Since a recent refactor, the code works with uAPI flags (for mapping buffer objects), and not PTE (page table entry) flags. Fixes: 6716a823d18d ("drm/amdgpu: rework how PTE flags are generated v3") Signed-off-by: Timur Kristóf Reviewed-by: Christian König Signed-off-by: Alex Deucher (cherry picked from commit 8feeab26c80635b802f72b3ed986c693ff8f3212) --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index c1a801203949..b1aaef962ad9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -2078,7 +2078,7 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev, struct amdgpu_bo *bo = before->bo_va->base.bo; amdgpu_vm_it_insert(before, &vm->va); - if (before->flags & AMDGPU_PTE_PRT_FLAG(adev)) + if (before->flags & AMDGPU_VM_PAGE_PRT) amdgpu_vm_prt_get(adev); if (amdgpu_vm_is_bo_always_valid(vm, bo) && @@ -2093,7 +2093,7 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev, struct amdgpu_bo *bo = after->bo_va->base.bo; amdgpu_vm_it_insert(after, &vm->va); - if (after->flags & AMDGPU_PTE_PRT_FLAG(adev)) + if (after->flags & AMDGPU_VM_PAGE_PRT) amdgpu_vm_prt_get(adev); if (amdgpu_vm_is_bo_always_valid(vm, bo) && -- cgit v1.2.3 From 21f46f54769c45ac8ca0dbaa977bc1b436ffdee2 Mon Sep 17 00:00:00 2001 From: Srinivasan Shanmugam Date: Tue, 18 Nov 2025 14:28:33 +0530 Subject: drm/amdgpu/ttm: Fix crash when handling MMIO_REMAP in PDE flags MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The MMIO_REMAP BO is a special 4K IO page that does not have a ttm_tt behind it. However, amdgpu_ttm_tt_pde_flags() was treating it like normal TT/doorbell/preempt memory and unconditionally accessed ttm->caching. For the MMIO_REMAP BO, ttm is NULL, so this leads to a NULL pointer dereference when computing PDE flags. Fix this by checking that ttm is non-NULL before reading ttm->caching. This prevents the crash for MMIO_REMAP and also makes the code more defensive if other BOs ever come through without a ttm_tt. Fixes: fb5a52dbe9fe ("drm/amdgpu: Implement TTM handling for MMIO_REMAP placement") Suggested-by: Jesse Zhang Suggested-by: Christian König Cc: Alex Deucher Signed-off-by: Srinivasan Shanmugam Reviewed-by: Jesse Zhang Tested-by: Jesse Zhang Signed-off-by: Alex Deucher (cherry picked from commit 0db94da5a0a1cacda080b9ec8425fcbe4babc141) --- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index aa9ee5dffa45..9d568c16beb1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1372,7 +1372,7 @@ uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *mem) mem->mem_type == AMDGPU_PL_MMIO_REMAP)) { flags |= AMDGPU_PTE_SYSTEM; - if (ttm->caching == ttm_cached) + if (ttm && ttm->caching == ttm_cached) flags |= AMDGPU_PTE_SNOOPED; } -- cgit v1.2.3 From c156c7f27ecdb7b89dbbeaaa1f40d9fadc3c1680 Mon Sep 17 00:00:00 2001 From: Shikang Fan Date: Wed, 19 Nov 2025 18:05:10 +0800 Subject: drm/amdgpu: Add sriov vf check for VCN per queue reset support. Add SRIOV check when setting VCN ring's supported reset mask. Signed-off-by: Shikang Fan Reviewed-by: Lijo Lazar Signed-off-by: Alex Deucher (cherry picked from commit ee9b603ad43f9870eb75184f9fb0a84f8c3cc852) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c | 2 +- drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c | 4 +++- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c index eacf4e93ba2f..cb7123ec1a5d 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c @@ -141,7 +141,7 @@ static int vcn_v4_0_3_late_init(struct amdgpu_ip_block *ip_block) adev->vcn.supported_reset = amdgpu_get_soft_full_reset_mask(&adev->vcn.inst[0].ring_enc[0]); - if (amdgpu_dpm_reset_vcn_is_supported(adev)) + if (amdgpu_dpm_reset_vcn_is_supported(adev) && !amdgpu_sriov_vf(adev)) adev->vcn.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c index 714350cabf2f..8bd457dea4cf 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_1.c @@ -122,7 +122,9 @@ static int vcn_v5_0_1_late_init(struct amdgpu_ip_block *ip_block) switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { case IP_VERSION(13, 0, 12): - if ((adev->psp.sos.fw_version >= 0x00450025) && amdgpu_dpm_reset_vcn_is_supported(adev)) + if ((adev->psp.sos.fw_version >= 0x00450025) && + amdgpu_dpm_reset_vcn_is_supported(adev) && + !amdgpu_sriov_vf(adev)) adev->vcn.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; break; default: -- cgit v1.2.3