From 9984db63742099ee3f3cff35cf71306d10e64356 Mon Sep 17 00:00:00 2001 From: Roman Li Date: Mon, 14 Apr 2025 12:56:48 -0400 Subject: [PATCH 01/17] drm/amd/display: Fix invalid context error in dml helper [Why] "BUG: sleeping function called from invalid context" error. after: "drm/amd/display: Protect FPU in dml2_validate()/dml21_validate()" The populate_dml_plane_cfg_from_plane_state() uses the GFP_KERNEL flag for memory allocation, which shouldn't be used in atomic contexts. The allocation is needed only for using another helper function get_scaler_data_for_plane(). [How] Modify helpers to pass a pointer to scaler_data within existing context, eliminating the need for dynamic memory allocation/deallocation and copying. Fixes: 366e77cd4923 ("drm/amd/display: Protect FPU in dml2_validate()/dml21_validate()") Reviewed-by: Aurabindo Pillai Signed-off-by: Roman Li Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit bd3e84bc98f81b44f2c43936bdadc3241d654259) Cc: stable@vger.kernel.org --- .../amd/display/dc/dml2/dml2_translation_helper.c | 14 +++++--------- 1 file changed, 5 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c index 2061d43b92e1..ab6baf269801 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c @@ -973,7 +973,9 @@ static void populate_dml_surface_cfg_from_plane_state(enum dml_project_id dml2_p } } -static void get_scaler_data_for_plane(const struct dc_plane_state *in, struct dc_state *context, struct scaler_data *out) +static struct scaler_data *get_scaler_data_for_plane( + const struct dc_plane_state *in, + struct dc_state *context) { int i; struct pipe_ctx *temp_pipe = &context->res_ctx.temp_pipe; @@ -994,7 +996,7 @@ static void get_scaler_data_for_plane(const struct dc_plane_state *in, struct dc } ASSERT(i < MAX_PIPES); - memcpy(out, &temp_pipe->plane_res.scl_data, sizeof(*out)); + return &temp_pipe->plane_res.scl_data; } static void populate_dummy_dml_plane_cfg(struct dml_plane_cfg_st *out, unsigned int location, @@ -1057,11 +1059,7 @@ static void populate_dml_plane_cfg_from_plane_state(struct dml_plane_cfg_st *out const struct dc_plane_state *in, struct dc_state *context, const struct soc_bounding_box_st *soc) { - struct scaler_data *scaler_data = kzalloc(sizeof(*scaler_data), GFP_KERNEL); - if (!scaler_data) - return; - - get_scaler_data_for_plane(in, context, scaler_data); + struct scaler_data *scaler_data = get_scaler_data_for_plane(in, context); out->CursorBPP[location] = dml_cur_32bit; out->CursorWidth[location] = 256; @@ -1126,8 +1124,6 @@ static void populate_dml_plane_cfg_from_plane_state(struct dml_plane_cfg_st *out out->DynamicMetadataTransmittedBytes[location] = 0; out->NumberOfCursors[location] = 1; - - kfree(scaler_data); } static unsigned int map_stream_to_dml_display_cfg(const struct dml2_context *dml2, From f1c6be3999d2be2673a51a9be0caf9348e254e52 Mon Sep 17 00:00:00 2001 From: Aurabindo Pillai Date: Wed, 16 Apr 2025 11:26:54 -0400 Subject: [PATCH 02/17] drm/amd/display: more liberal vmin/vmax update for freesync [Why] FAMS2 expects vmin/vmax to be updated in the case when freesync is off, but supported. But we only update it when freesync is enabled. [How] Change the vsync handler such that dc_stream_adjust_vmin_vmax() its called irrespective of whether freesync is enabled. If freesync is supported, then there is no harm in updating vmin/vmax registers. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3546 Reviewed-by: ChiaHsuan Chung Signed-off-by: Aurabindo Pillai Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit cfb2d41831ee5647a4ae0ea7c24971a92d5dfa0d) Cc: stable@vger.kernel.org --- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) 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 536f73131c2d..93dd9e273c65 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -673,15 +673,21 @@ static void dm_crtc_high_irq(void *interrupt_params) spin_lock_irqsave(&adev_to_drm(adev)->event_lock, flags); if (acrtc->dm_irq_params.stream && - acrtc->dm_irq_params.vrr_params.supported && - acrtc->dm_irq_params.freesync_config.state == - VRR_STATE_ACTIVE_VARIABLE) { + acrtc->dm_irq_params.vrr_params.supported) { + bool replay_en = acrtc->dm_irq_params.stream->link->replay_settings.replay_feature_enabled; + bool psr_en = acrtc->dm_irq_params.stream->link->psr_settings.psr_feature_enabled; + bool fs_active_var_en = acrtc->dm_irq_params.freesync_config.state == VRR_STATE_ACTIVE_VARIABLE; + mod_freesync_handle_v_update(adev->dm.freesync_module, acrtc->dm_irq_params.stream, &acrtc->dm_irq_params.vrr_params); - dc_stream_adjust_vmin_vmax(adev->dm.dc, acrtc->dm_irq_params.stream, - &acrtc->dm_irq_params.vrr_params.adjust); + /* update vmin_vmax only if freesync is enabled, or only if PSR and REPLAY are disabled */ + if (fs_active_var_en || (!fs_active_var_en && !replay_en && !psr_en)) { + dc_stream_adjust_vmin_vmax(adev->dm.dc, + acrtc->dm_irq_params.stream, + &acrtc->dm_irq_params.vrr_params.adjust); + } } /* From 2a24755774ef8db33139e2d9a968cc585c6488da Mon Sep 17 00:00:00 2001 From: Alex Hung Date: Wed, 23 Apr 2025 10:02:11 -0600 Subject: [PATCH 03/17] drm/amd/display: Remove unnecessary DC_FP_START/DC_FP_END [WHY & HOW] Remove the unnecessary DC_FP_START/DC_FP_END pair to reduce time in preempt_disable. It also fixes "BUG: sleeping function called from invalid context" error messages because of calling kzalloc with GFP_KERNEL which can sleep. Reviewed-by: Aurabindo Pillai Signed-off-by: Alex Hung Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 94da0735b67b3ada90a3e2e017d306f070306f1d) --- .../gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c | 6 ------ 1 file changed, 6 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c index 2a59cc61ed8c..944650cb13de 100644 --- a/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c +++ b/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c @@ -2114,8 +2114,6 @@ static bool dcn32_resource_construct( #define REG_STRUCT dccg_regs dccg_regs_init(); - DC_FP_START(); - ctx->dc_bios->regs = &bios_regs; pool->base.res_cap = &res_cap_dcn32; @@ -2501,14 +2499,10 @@ static bool dcn32_resource_construct( if (ASICREV_IS_GC_11_0_3(dc->ctx->asic_id.hw_internal_rev) && (dc->config.sdpif_request_limit_words_per_umc == 0)) dc->config.sdpif_request_limit_words_per_umc = 16; - DC_FP_END(); - return true; create_fail: - DC_FP_END(); - dcn32_resource_destruct(pool); return false; From eba692ca3abca258b3214a6e4126afefad1822f0 Mon Sep 17 00:00:00 2001 From: Austin Zheng Date: Thu, 17 Apr 2025 10:24:29 -0400 Subject: [PATCH 04/17] drm/amd/display: Call FP Protect Before Mode Programming/Mode Support [Why] Memory allocation occurs within dml21_validate() for adding phantom planes. May cause kernel to be tainted due to usage of FP Start. [How] Move FP start from dml21_validate to before mode programming/mode support. Calculations requiring floating point are all done within mode programming or mode support. Reviewed-by: Alvin Lee Signed-off-by: Austin Zheng Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit fe3250f10819b411808ab9ae1d824c5fc9b59170) --- drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c index 5d16f36ec95c..ed6584535e89 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c @@ -234,7 +234,9 @@ static bool dml21_mode_check_and_programming(const struct dc *in_dc, struct dc_s if (!result) return false; + DC_FP_START(); result = dml2_build_mode_programming(mode_programming); + DC_FP_END(); if (!result) return false; @@ -277,7 +279,9 @@ static bool dml21_check_mode_support(const struct dc *in_dc, struct dc_state *co mode_support->dml2_instance = dml_init->dml2_instance; dml21_map_dc_state_into_dml_display_cfg(in_dc, context, dml_ctx); dml_ctx->v21.mode_programming.dml2_instance->scratch.build_mode_programming_locals.mode_programming_params.programming = dml_ctx->v21.mode_programming.programming; + DC_FP_START(); is_supported = dml2_check_mode_supported(mode_support); + DC_FP_END(); if (!is_supported) return false; @@ -288,16 +292,12 @@ bool dml21_validate(const struct dc *in_dc, struct dc_state *context, struct dml { bool out = false; - DC_FP_START(); - /* Use dml_validate_only for fast_validate path */ if (fast_validate) out = dml21_check_mode_support(in_dc, context, dml_ctx); else out = dml21_mode_check_and_programming(in_dc, context, dml_ctx); - DC_FP_END(); - return out; } From 5a3846648c0523fd850b7f0aec78c0139453ab8b Mon Sep 17 00:00:00 2001 From: Wayne Lin Date: Fri, 18 Apr 2025 16:31:59 +0800 Subject: [PATCH 05/17] drm/amd/display: Shift DMUB AUX reply command if necessary [Why] Defined value of dmub AUX reply command field get updated but didn't adjust dm receiving side accordingly. [How] Check the received reply command value to see if it's updated version or not. Adjust it if necessary. Fixes: ead08b95fa50 ("drm/amd/display: Fix race condition in DPIA AUX transfer") Cc: Mario Limonciello Cc: Alex Deucher Reviewed-by: Ray Wu Signed-off-by: Wayne Lin Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit d5c9ade755a9afa210840708a12a8f44c0d532f4) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) 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 93dd9e273c65..d9292da56948 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -12758,8 +12758,11 @@ int amdgpu_dm_process_dmub_aux_transfer_sync( goto out; } + payload->reply[0] = adev->dm.dmub_notify->aux_reply.command & 0xF; + if (adev->dm.dmub_notify->aux_reply.command & 0xF0) + /* The reply is stored in the top nibble of the command. */ + payload->reply[0] = (adev->dm.dmub_notify->aux_reply.command >> 4) & 0xF; - payload->reply[0] = adev->dm.dmub_notify->aux_reply.command; if (!payload->write && p_notify->aux_reply.length && (payload->reply[0] == AUX_TRANSACTION_REPLY_AUX_ACK)) { From bc70e11b550d37fbd9eaed0f113ba560894f1609 Mon Sep 17 00:00:00 2001 From: Wayne Lin Date: Sun, 20 Apr 2025 16:29:07 +0800 Subject: [PATCH 06/17] drm/amd/display: Fix the checking condition in dmub aux handling [Why & How] Fix the checking condition for detecting AUX_RET_ERROR_PROTOCOL_ERROR. It was wrongly checking by "not equals to" Reviewed-by: Ray Wu Signed-off-by: Wayne Lin Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 1db6c9e9b62e1a8912f0a281c941099fca678da3) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) 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 d9292da56948..68bbe7d57773 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -12749,7 +12749,7 @@ int amdgpu_dm_process_dmub_aux_transfer_sync( * Transient states before tunneling is enabled could * lead to this error. We can ignore this for now. */ - if (p_notify->result != AUX_RET_ERROR_PROTOCOL_ERROR) { + if (p_notify->result == AUX_RET_ERROR_PROTOCOL_ERROR) { DRM_WARN("DPIA AUX failed on 0x%x(%d), error %d\n", payload->address, payload->length, p_notify->result); From 396dc51b3b7ea524bf8061f478332d0039e96d5d Mon Sep 17 00:00:00 2001 From: Wayne Lin Date: Sun, 20 Apr 2025 16:56:54 +0800 Subject: [PATCH 07/17] drm/amd/display: Remove incorrect checking in dmub aux handler [Why & How] "Request length != reply length" is expected behavior defined in spec. It's not an invalid reply. Besides, replied data handling logic is not designed to be written in amdgpu_dm_process_dmub_aux_transfer_sync(). Remove the incorrectly handling section. Fixes: ead08b95fa50 ("drm/amd/display: Fix race condition in DPIA AUX transfer") Cc: Mario Limonciello Cc: Alex Deucher Reviewed-by: Ray Wu Signed-off-by: Wayne Lin Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 81b5c6fa62af62fe89ae9576f41aae37830b94cb) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-) 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 68bbe7d57773..d63f219de88e 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -12764,19 +12764,9 @@ int amdgpu_dm_process_dmub_aux_transfer_sync( payload->reply[0] = (adev->dm.dmub_notify->aux_reply.command >> 4) & 0xF; if (!payload->write && p_notify->aux_reply.length && - (payload->reply[0] == AUX_TRANSACTION_REPLY_AUX_ACK)) { - - if (payload->length != p_notify->aux_reply.length) { - DRM_WARN("invalid read length %d from DPIA AUX 0x%x(%d)!\n", - p_notify->aux_reply.length, - payload->address, payload->length); - *operation_result = AUX_RET_ERROR_INVALID_REPLY; - goto out; - } - + (payload->reply[0] == AUX_TRANSACTION_REPLY_AUX_ACK)) memcpy(payload->data, p_notify->aux_reply.data, p_notify->aux_reply.length); - } /* success */ ret = p_notify->aux_reply.length; From 3924f45d4de7250a603fd7b50379237a6a0e5adf Mon Sep 17 00:00:00 2001 From: Wayne Lin Date: Sun, 20 Apr 2025 17:50:03 +0800 Subject: [PATCH 08/17] drm/amd/display: Copy AUX read reply data whenever length > 0 [Why] amdgpu_dm_process_dmub_aux_transfer_sync() should return all exact data reply from the sink side. Don't do the analysis job in it. [How] Remove unnecessary check condition AUX_TRANSACTION_REPLY_AUX_ACK. Fixes: ead08b95fa50 ("drm/amd/display: Fix race condition in DPIA AUX transfer") Cc: Mario Limonciello Cc: Alex Deucher Reviewed-by: Ray Wu Signed-off-by: Wayne Lin Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 9b540e3fe6796fec4fb1344f3be8952fc2f084d4) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) 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 d63f219de88e..64df8ca448b3 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -12763,8 +12763,7 @@ int amdgpu_dm_process_dmub_aux_transfer_sync( /* The reply is stored in the top nibble of the command. */ payload->reply[0] = (adev->dm.dmub_notify->aux_reply.command >> 4) & 0xF; - if (!payload->write && p_notify->aux_reply.length && - (payload->reply[0] == AUX_TRANSACTION_REPLY_AUX_ACK)) + if (!payload->write && p_notify->aux_reply.length) memcpy(payload->data, p_notify->aux_reply.data, p_notify->aux_reply.length); From 65924ec69b29296845c7f628112353438e63ea56 Mon Sep 17 00:00:00 2001 From: Wayne Lin Date: Sun, 20 Apr 2025 19:22:14 +0800 Subject: [PATCH 09/17] drm/amd/display: Fix wrong handling for AUX_DEFER case [Why] We incorrectly ack all bytes get written when the reply actually is defer. When it's defer, means sink is not ready for the request. We should retry the request. [How] Only reply all data get written when receive I2C_ACK|AUX_ACK. Otherwise, reply the number of actual written bytes received from the sink. Add some messages to facilitate debugging as well. Fixes: ad6756b4d773 ("drm/amd/display: Shift dc link aux to aux_payload") Cc: Mario Limonciello Cc: Alex Deucher Reviewed-by: Ray Wu Signed-off-by: Wayne Lin Signed-off-by: Ray Wu Tested-by: Daniel Wheeler Signed-off-by: Alex Deucher (cherry picked from commit 3637e457eb0000bc37d8bbbec95964aad2fb29fd) Cc: stable@vger.kernel.org --- .../display/amdgpu_dm/amdgpu_dm_mst_types.c | 28 ++++++++++++++++--- 1 file changed, 24 insertions(+), 4 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 7ceedf626d23..074b79fd5822 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 @@ -51,6 +51,9 @@ #define PEAK_FACTOR_X1000 1006 +/* + * This function handles both native AUX and I2C-Over-AUX transactions. + */ static ssize_t dm_dp_aux_transfer(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg) { @@ -87,15 +90,25 @@ static ssize_t dm_dp_aux_transfer(struct drm_dp_aux *aux, if (adev->dm.aux_hpd_discon_quirk) { if (msg->address == DP_SIDEBAND_MSG_DOWN_REQ_BASE && operation_result == AUX_RET_ERROR_HPD_DISCON) { - result = 0; + result = msg->size; operation_result = AUX_RET_SUCCESS; } } - if (payload.write && result >= 0) - result = msg->size; + /* + * result equals to 0 includes the cases of AUX_DEFER/I2C_DEFER + */ + if (payload.write && result >= 0) { + if (result) { + /*one byte indicating partially written bytes. Force 0 to retry*/ + drm_info(adev_to_drm(adev), "amdgpu: AUX partially written\n"); + result = 0; + } else if (!payload.reply[0]) + /*I2C_ACK|AUX_ACK*/ + result = msg->size; + } - if (result < 0) + if (result < 0) { switch (operation_result) { case AUX_RET_SUCCESS: break; @@ -114,6 +127,13 @@ static ssize_t dm_dp_aux_transfer(struct drm_dp_aux *aux, break; } + drm_info(adev_to_drm(adev), "amdgpu: DP AUX transfer fail:%d\n", operation_result); + } + + if (payload.reply[0]) + drm_info(adev_to_drm(adev), "amdgpu: AUX reply command not ACK: 0x%02x.", + payload.reply[0]); + return result; } From b7e84fb708392b37e5dbb2a95db9b94a0e3f0aa2 Mon Sep 17 00:00:00 2001 From: Ruijing Dong Date: Fri, 2 May 2025 11:19:26 -0400 Subject: [PATCH 10/17] drm/amdgpu/vcn: using separate VCN1_AON_SOC offset VCN1_AON_SOC_ADDRESS_3_0 offset varies on different VCN generations, the issue in vcn4.0.5 is caused by a different VCN1_AON_SOC_ADDRESS_3_0 offset. This patch does the following: 1. use the same offset for other VCN generations. 2. use the vcn4.0.5 special offset 3. update vcn_4_0 and vcn_5_0 Acked-by: Saleemkhan Jamadar Reviewed-by: Leo Liu Signed-off-by: Ruijing Dong Signed-off-by: Alex Deucher (cherry picked from commit 5c89ceda9984498b28716944633a9a01cbb2c90d) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 1 - drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 1 + drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 1 + drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 1 + drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c | 4 +++- drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c | 1 + drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c | 1 + drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c | 3 ++- 8 files changed, 10 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h index cdcdae7f71ce..83adf81defc7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h @@ -66,7 +66,6 @@ #define VCN_ENC_CMD_REG_WAIT 0x0000000c #define VCN_AON_SOC_ADDRESS_2_0 0x1f800 -#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 #define VCN_VID_IP_ADDRESS_2_0 0x0 #define VCN_AON_IP_ADDRESS_2_0 0x30000 diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c index 8e7a36f26e9c..b8d835c9e17e 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c @@ -39,6 +39,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fa00 #define VCN1_VID_SOC_ADDRESS_3_0 0x48200 +#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x1fd #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x503 diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c index d716510b8dd6..3eec1b8feaee 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c @@ -39,6 +39,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fa00 #define VCN1_VID_SOC_ADDRESS_3_0 0x48200 +#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c index 22ae1939476f..0b19f0ab4480 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c @@ -40,6 +40,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fa00 #define VCN1_VID_SOC_ADDRESS_3_0 0x48200 +#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c index c6f6392c1c20..1f777c125b00 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c @@ -46,6 +46,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fb00 #define VCN1_VID_SOC_ADDRESS_3_0 0x48300 +#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 #define VCN_HARVEST_MMSCH 0 @@ -614,7 +615,8 @@ static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_vcn_inst *vinst, /* VCN global tiling registers */ WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET( - VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect); + VCN, inst_idx, regUVD_GFX10_ADDR_CONFIG), + adev->gfx.config.gb_addr_config, 0, indirect); } /** 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 3e176b4b7c69..012f6ea928ec 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c @@ -45,6 +45,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fb00 #define VCN1_VID_SOC_ADDRESS_3_0 0x48300 +#define VCN1_AON_SOC_ADDRESS_3_0 0x48000 static const struct amdgpu_hwip_reg_entry vcn_reg_list_4_0_3[] = { SOC15_REG_ENTRY_STR(VCN, 0, regUVD_POWER_STATUS), diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c index ba603b2246e2..a1171e6152ed 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c @@ -46,6 +46,7 @@ #define VCN_VID_SOC_ADDRESS_2_0 0x1fb00 #define VCN1_VID_SOC_ADDRESS_3_0 (0x48300 + 0x38000) +#define VCN1_AON_SOC_ADDRESS_3_0 (0x48000 + 0x38000) #define VCN_HARVEST_MMSCH 0 diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c index d99d05f42f1d..b90da3d3e140 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c @@ -533,7 +533,8 @@ static void vcn_v5_0_0_mc_resume_dpg_mode(struct amdgpu_vcn_inst *vinst, /* VCN global tiling registers */ WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( - VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect); + VCN, inst_idx, regUVD_GFX10_ADDR_CONFIG), + adev->gfx.config.gb_addr_config, 0, indirect); return; } From d0ce1aaa8531a4a4707711cab5721374751c51b0 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Thu, 1 May 2025 13:00:16 -0400 Subject: [PATCH 11/17] Revert "drm/amd: Stop evicting resources on APUs in suspend" This reverts commit 3a9626c816db901def438dc2513622e281186d39. This breaks S4 because we end up setting the s3/s0ix flags even when we are entering s4 since prepare is used by both flows. The causes both the S3/s0ix and s4 flags to be set which breaks several checks in the driver which assume they are mutually exclusive. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3634 Cc: Mario Limonciello Reviewed-by: Mario Limonciello Signed-off-by: Alex Deucher (cherry picked from commit ce8f7d95899c2869b47ea6ce0b3e5bf304b2fff4) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 -- drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c | 18 ------------------ drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ++--------- 3 files changed, 2 insertions(+), 29 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index ef6e78224fdf..c3641331d4de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1614,11 +1614,9 @@ static inline void amdgpu_acpi_get_backlight_caps(struct amdgpu_dm_backlight_cap #if defined(CONFIG_ACPI) && defined(CONFIG_SUSPEND) bool amdgpu_acpi_is_s3_active(struct amdgpu_device *adev); bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev); -void amdgpu_choose_low_power_state(struct amdgpu_device *adev); #else static inline bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev) { return false; } static inline bool amdgpu_acpi_is_s3_active(struct amdgpu_device *adev) { return false; } -static inline void amdgpu_choose_low_power_state(struct amdgpu_device *adev) { } #endif void amdgpu_register_gpu_instance(struct amdgpu_device *adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c index b7f8f2ff143d..707e131f89d2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c @@ -1533,22 +1533,4 @@ bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev) #endif /* CONFIG_AMD_PMC */ } -/** - * amdgpu_choose_low_power_state - * - * @adev: amdgpu_device_pointer - * - * Choose the target low power state for the GPU - */ -void amdgpu_choose_low_power_state(struct amdgpu_device *adev) -{ - if (adev->in_runpm) - return; - - if (amdgpu_acpi_is_s0ix_active(adev)) - adev->in_s0ix = true; - else if (amdgpu_acpi_is_s3_active(adev)) - adev->in_s3 = true; -} - #endif /* CONFIG_SUSPEND */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 7f354cd532dc..5ac7bd5942d0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4949,15 +4949,13 @@ int amdgpu_device_prepare(struct drm_device *dev) struct amdgpu_device *adev = drm_to_adev(dev); int i, r; - amdgpu_choose_low_power_state(adev); - if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) return 0; /* Evict the majority of BOs before starting suspend sequence */ r = amdgpu_device_evict_resources(adev); if (r) - goto unprepare; + return r; flush_delayed_work(&adev->gfx.gfx_off_delay_work); @@ -4968,15 +4966,10 @@ int amdgpu_device_prepare(struct drm_device *dev) continue; r = adev->ip_blocks[i].version->funcs->prepare_suspend(&adev->ip_blocks[i]); if (r) - goto unprepare; + return r; } return 0; - -unprepare: - adev->in_s0ix = adev->in_s3 = adev->in_s4 = false; - - return r; } /** From 4aaffc85751da5722e858e4333e8cf0aa4b6c78f Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Thu, 1 May 2025 13:46:46 -0400 Subject: [PATCH 12/17] drm/amdgpu: fix pm notifier handling Set the s3/s0ix and s4 flags in the pm notifier so that we can skip the resource evictions properly in pm prepare based on whether we are suspending or hibernating. Drop the eviction as processes are not frozen at this time, we we can end up getting stuck trying to evict VRAM while applications continue to submit work which causes the buffers to get pulled back into VRAM. v2: Move suspend flags out of pm notifier (Mario) Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4178 Fixes: 2965e6355dcd ("drm/amd: Add Suspend/Hibernate notification callback support") Cc: Mario Limonciello Reviewed-by: Mario Limonciello Signed-off-by: Alex Deucher (cherry picked from commit 06f2dcc241e7e5c681f81fbc46cacdf4bfd7d6d7) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 18 +++++------------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 10 +--------- 2 files changed, 6 insertions(+), 22 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 5ac7bd5942d0..f8b3e04d71ed 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4907,28 +4907,20 @@ static int amdgpu_device_evict_resources(struct amdgpu_device *adev) * @data: data * * This function is called when the system is about to suspend or hibernate. - * It is used to evict resources from the device before the system goes to - * sleep while there is still access to swap. + * It is used to set the appropriate flags so that eviction can be optimized + * in the pm prepare callback. */ static int amdgpu_device_pm_notifier(struct notifier_block *nb, unsigned long mode, void *data) { struct amdgpu_device *adev = container_of(nb, struct amdgpu_device, pm_nb); - int r; switch (mode) { case PM_HIBERNATION_PREPARE: adev->in_s4 = true; - fallthrough; - case PM_SUSPEND_PREPARE: - r = amdgpu_device_evict_resources(adev); - /* - * This is considered non-fatal at this time because - * amdgpu_device_prepare() will also fatally evict resources. - * See https://gitlab.freedesktop.org/drm/amd/-/issues/3781 - */ - if (r) - drm_warn(adev_to_drm(adev), "Failed to evict resources, freeze active processes if problems occur: %d\n", r); + break; + case PM_POST_HIBERNATION: + adev->in_s4 = false; break; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 24ee4710f807..72c807f5822e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -2615,13 +2615,8 @@ static int amdgpu_pmops_freeze(struct device *dev) static int amdgpu_pmops_thaw(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); - struct amdgpu_device *adev = drm_to_adev(drm_dev); - int r; - r = amdgpu_device_resume(drm_dev, true); - adev->in_s4 = false; - - return r; + return amdgpu_device_resume(drm_dev, true); } static int amdgpu_pmops_poweroff(struct device *dev) @@ -2634,9 +2629,6 @@ static int amdgpu_pmops_poweroff(struct device *dev) static int amdgpu_pmops_restore(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); - struct amdgpu_device *adev = drm_to_adev(drm_dev); - - adev->in_s4 = false; return amdgpu_device_resume(drm_dev, true); } From f690e3974755a650259a45d71456decc9c96a282 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 30 Apr 2025 12:45:04 -0400 Subject: [PATCH 13/17] drm/amdgpu/hdp4: use memcfg register to post the write for HDP flush Reading back the remapped HDP flush register seems to cause problems on some platforms. All we need is a read, so read back the memcfg register. Fixes: c9b8dcabb52a ("drm/amdgpu/hdp4.0: do a posting read when flushing HDP") Reported-by: Alexey Klimov Link: https://lists.freedesktop.org/archives/amd-gfx/2025-April/123150.html Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4119 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3908 Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher (cherry picked from commit 5c937b4a6050316af37ef214825b6340b5e9e391) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c b/drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c index f1dc13b3ab38..cbbeadeb53f7 100644 --- a/drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c @@ -41,7 +41,12 @@ static void hdp_v4_0_flush_hdp(struct amdgpu_device *adev, { if (!ring || !ring->funcs->emit_wreg) { WREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); - RREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + /* We just need to read back a register to post the write. + * Reading back the remapped register causes problems on + * some platforms so just read back the memory size register. + */ + if (adev->nbio.funcs->get_memsize) + adev->nbio.funcs->get_memsize(adev); } else { amdgpu_ring_emit_wreg(ring, (adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); } From 0e33e0f339b91eecd9558311449a3d1e728722d4 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 30 Apr 2025 12:46:56 -0400 Subject: [PATCH 14/17] drm/amdgpu/hdp5: use memcfg register to post the write for HDP flush Reading back the remapped HDP flush register seems to cause problems on some platforms. All we need is a read, so read back the memcfg register. Fixes: cf424020e040 ("drm/amdgpu/hdp5.0: do a posting read when flushing HDP") Reported-by: Alexey Klimov Link: https://lists.freedesktop.org/archives/amd-gfx/2025-April/123150.html Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4119 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3908 Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher (cherry picked from commit a5cb344033c7598762e89255e8ff52827abb57a4) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c b/drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c index 43195c079748..086a647308df 100644 --- a/drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c @@ -32,7 +32,12 @@ static void hdp_v5_0_flush_hdp(struct amdgpu_device *adev, { if (!ring || !ring->funcs->emit_wreg) { WREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); - RREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + /* We just need to read back a register to post the write. + * Reading back the remapped register causes problems on + * some platforms so just read back the memory size register. + */ + if (adev->nbio.funcs->get_memsize) + adev->nbio.funcs->get_memsize(adev); } else { amdgpu_ring_emit_wreg(ring, (adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); } From dbc988c689333faeeed44d5561f372ff20395304 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 30 Apr 2025 12:47:37 -0400 Subject: [PATCH 15/17] drm/amdgpu/hdp5.2: use memcfg register to post the write for HDP flush Reading back the remapped HDP flush register seems to cause problems on some platforms. All we need is a read, so read back the memcfg register. Fixes: f756dbac1ce1 ("drm/amdgpu/hdp5.2: do a posting read when flushing HDP") Reported-by: Alexey Klimov Link: https://lists.freedesktop.org/archives/amd-gfx/2025-April/123150.html Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4119 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3908 Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher (cherry picked from commit 4a89b7698e771914b4d5b571600c76e2fdcbe2a9) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/hdp_v5_2.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/hdp_v5_2.c b/drivers/gpu/drm/amd/amdgpu/hdp_v5_2.c index fcb8dd2876bc..40940b4ab400 100644 --- a/drivers/gpu/drm/amd/amdgpu/hdp_v5_2.c +++ b/drivers/gpu/drm/amd/amdgpu/hdp_v5_2.c @@ -33,7 +33,17 @@ static void hdp_v5_2_flush_hdp(struct amdgpu_device *adev, if (!ring || !ring->funcs->emit_wreg) { WREG32_NO_KIQ((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); - RREG32_NO_KIQ((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + if (amdgpu_sriov_vf(adev)) { + /* this is fine because SR_IOV doesn't remap the register */ + RREG32_NO_KIQ((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + } else { + /* We just need to read back a register to post the write. + * Reading back the remapped register causes problems on + * some platforms so just read back the memory size register. + */ + if (adev->nbio.funcs->get_memsize) + adev->nbio.funcs->get_memsize(adev); + } } else { amdgpu_ring_emit_wreg(ring, (adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, From ca28e80abe4219c8f1a2961ae05102d70af6dc87 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 30 Apr 2025 12:48:51 -0400 Subject: [PATCH 16/17] drm/amdgpu/hdp6: use memcfg register to post the write for HDP flush Reading back the remapped HDP flush register seems to cause problems on some platforms. All we need is a read, so read back the memcfg register. Fixes: abe1cbaec6cf ("drm/amdgpu/hdp6.0: do a posting read when flushing HDP") Reported-by: Alexey Klimov Link: https://lists.freedesktop.org/archives/amd-gfx/2025-April/123150.html Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4119 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3908 Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher (cherry picked from commit 84141ff615951359c9a99696fd79a36c465ed847) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/hdp_v6_0.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/hdp_v6_0.c b/drivers/gpu/drm/amd/amdgpu/hdp_v6_0.c index a88d25a06c29..6ccd31c8bc69 100644 --- a/drivers/gpu/drm/amd/amdgpu/hdp_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/hdp_v6_0.c @@ -35,7 +35,12 @@ static void hdp_v6_0_flush_hdp(struct amdgpu_device *adev, { if (!ring || !ring->funcs->emit_wreg) { WREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); - RREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + /* We just need to read back a register to post the write. + * Reading back the remapped register causes problems on + * some platforms so just read back the memory size register. + */ + if (adev->nbio.funcs->get_memsize) + adev->nbio.funcs->get_memsize(adev); } else { amdgpu_ring_emit_wreg(ring, (adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); } From 5a11a2767731139bf87e667331aa2209e33a1d19 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 30 Apr 2025 12:50:02 -0400 Subject: [PATCH 17/17] drm/amdgpu/hdp7: use memcfg register to post the write for HDP flush Reading back the remapped HDP flush register seems to cause problems on some platforms. All we need is a read, so read back the memcfg register. Fixes: 689275140cb8 ("drm/amdgpu/hdp7.0: do a posting read when flushing HDP") Reported-by: Alexey Klimov Link: https://lists.freedesktop.org/archives/amd-gfx/2025-April/123150.html Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4119 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3908 Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher (cherry picked from commit dbc064adfcf9095e7d895bea87b2f75c1ab23236) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/hdp_v7_0.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/hdp_v7_0.c b/drivers/gpu/drm/amd/amdgpu/hdp_v7_0.c index 49f7eb4fbd11..2c9239a22f39 100644 --- a/drivers/gpu/drm/amd/amdgpu/hdp_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/hdp_v7_0.c @@ -32,7 +32,12 @@ static void hdp_v7_0_flush_hdp(struct amdgpu_device *adev, { if (!ring || !ring->funcs->emit_wreg) { WREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); - RREG32((adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2); + /* We just need to read back a register to post the write. + * Reading back the remapped register causes problems on + * some platforms so just read back the memory size register. + */ + if (adev->nbio.funcs->get_memsize) + adev->nbio.funcs->get_memsize(adev); } else { amdgpu_ring_emit_wreg(ring, (adev->rmmio_remap.reg_offset + KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >> 2, 0); }