From 3394069e7de984ea3a4c2c68bf69b65f1c0aa658 Mon Sep 17 00:00:00 2001 From: Victor Skvortsov Date: Wed, 2 Apr 2025 17:35:56 -0400 Subject: drm/amdgpu: Disable ACA on VFs VFs query RAS error counts directly from host with AMDGPU_RAS_VIRT_ERROR_COUNT_QUERY. When ACA is enabled, an unusable aca_sysfs is created rather than amdgpu_ras_sysfs_create() Likewise, VFs depend on host support to query CPERs, rather than ACA component. Signed-off-by: Victor Skvortsov Reviewed-by: Zhigang Luo Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index ebf1f63d0442..5bb7673fd28e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -3794,10 +3794,12 @@ init_ras_enabled_flag: adev->ras_hw_enabled & amdgpu_ras_mask; /* aca is disabled by default except for psp v13_0_6/v13_0_12/v13_0_14 */ - adev->aca.is_enabled = - (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) || - amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 12) || - amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 14)); + if (!amdgpu_sriov_vf(adev)) { + adev->aca.is_enabled = + (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) || + amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 12) || + amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 14)); + } /* bad page feature is not applicable to specific app platform */ if (adev->gmc.is_app_apu && -- cgit v1.2.3 From 5045c6c69872b9dd145e9ecbc9731d81a5954be1 Mon Sep 17 00:00:00 2001 From: Ellen Pan Date: Thu, 10 Apr 2025 22:12:24 -0400 Subject: drm/amdgpu: Direct ret in ras_reset_err_cnt on VF With adding sriov_vf check, we directly return EOPNOTSUPP in ras_reset_error_count as we should not do anything on VF to reset RAS error count. This also fixes the issue that loading guest driver causes register violations. Reviewed-by: Ahmad Rehman Signed-off-by: Ellen Pan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 5bb7673fd28e..17f0911ee7e9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -1498,6 +1498,9 @@ int amdgpu_ras_reset_error_count(struct amdgpu_device *adev, !amdgpu_ras_get_aca_debug_mode(adev)) return -EOPNOTSUPP; + if (amdgpu_sriov_vf(adev)) + return -EOPNOTSUPP; + /* skip ras error reset in gpu reset */ if ((amdgpu_in_reset(adev) || amdgpu_ras_in_recovery(adev)) && ((smu_funcs && smu_funcs->set_debug_mode) || -- cgit v1.2.3 From 0105725e2d985899cef5ee187bb27f040f24f2ab Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Mon, 28 Apr 2025 10:36:35 +0530 Subject: drm/amdgpu: Fix comment style Fix code comment style Signed-off-by: Lijo Lazar Reviewed-by: Asad Kamal Reported-by: kernel test robot Closes: https://lore.kernel.org/oe-kbuild-all/202504271826.xy2fFO28-lkp@intel.com/ Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 17f0911ee7e9..82013b495436 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2165,7 +2165,7 @@ void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) /* Fatal error events are handled on host side */ if (amdgpu_sriov_vf(adev)) return; - /** + /* * If the current interrupt is caused by a non-fatal RAS error, skip * check for fatal error. For fatal errors, FED status of all devices * in XGMI hive gets set when the first device gets fatal error -- cgit v1.2.3 From f5db59067c3130d186ddb619bcbdc6bea0ffb704 Mon Sep 17 00:00:00 2001 From: ganglxie Date: Thu, 24 Apr 2025 17:11:47 +0800 Subject: Refine RAS bad page records counting and parsing in eeprom V3 there is only MCA records in V3, no need to care about PA records. recalculate the value of ras_num_bad_pages when parsing failed and go on with the left records instead of quit. Signed-off-by: ganglxie Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 62 +++++++++++++++++++-------------- 1 file changed, 36 insertions(+), 26 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index f40b35f7f679..babc3c9cad65 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2889,6 +2889,7 @@ static int __amdgpu_ras_convert_rec_from_rom(struct amdgpu_device *adev, if (amdgpu_ras_mca2pa_by_idx(adev, bps, err_data)) return -EINVAL; } + return __amdgpu_ras_restore_bad_pages(adev, err_data->err_addr, adev->umc.retire_unit); } @@ -2903,7 +2904,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, &adev->psp.ras_context.ras->eeprom_control; enum amdgpu_memory_partition nps = AMDGPU_NPS1_PARTITION_MODE; int ret = 0; - uint32_t i; + uint32_t i = 0; if (!con || !con->eh_data || !bps || pages <= 0) return 0; @@ -2924,34 +2925,36 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, mutex_lock(&con->recovery_lock); if (from_rom) { - for (i = 0; i < pages; i++) { - if (control->ras_num_recs - i >= adev->umc.retire_unit) { - if ((bps[i].address == bps[i + 1].address) && - (bps[i].mem_channel == bps[i + 1].mem_channel)) { - //deal with retire_unit records a time - ret = __amdgpu_ras_convert_rec_array_from_rom(adev, - &bps[i], &err_data, nps); - if (ret) - goto free; - i += (adev->umc.retire_unit - 1); + /* there is no pa recs in V3, so skip pa recs processing */ + if (control->tbl_hdr.version < RAS_TABLE_VER_V3) { + for (i = 0; i < pages; i++) { + if (control->ras_num_recs - i >= adev->umc.retire_unit) { + if ((bps[i].address == bps[i + 1].address) && + (bps[i].mem_channel == bps[i + 1].mem_channel)) { + /* deal with retire_unit records a time */ + ret = __amdgpu_ras_convert_rec_array_from_rom(adev, + &bps[i], &err_data, nps); + if (ret) + control->ras_num_bad_pages -= adev->umc.retire_unit; + i += (adev->umc.retire_unit - 1); + } else { + break; + } } else { break; } - } else { - break; } } for (; i < pages; i++) { ret = __amdgpu_ras_convert_rec_from_rom(adev, &bps[i], &err_data, nps); if (ret) - goto free; + control->ras_num_bad_pages -= adev->umc.retire_unit; } } else { ret = __amdgpu_ras_restore_bad_pages(adev, bps, pages); } -free: if (from_rom) kfree(err_data.err_addr); mutex_unlock(&con->recovery_lock); @@ -3040,21 +3043,28 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev) dev_err(adev->dev, "Failed to load EEPROM table records!"); } else { if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) { - for (i = 0; i < control->ras_num_recs; i++) { - if ((control->ras_num_recs - i) >= adev->umc.retire_unit) { - if ((bps[i].address == bps[i + 1].address) && - (bps[i].mem_channel == bps[i + 1].mem_channel)) { - control->ras_num_pa_recs += adev->umc.retire_unit; - i += (adev->umc.retire_unit - 1); + /*In V3, there is no pa recs, and some cases(when address==0) may be parsed + as pa recs, so add verion check to avoid it. + */ + if (control->tbl_hdr.version < RAS_TABLE_VER_V3) { + for (i = 0; i < control->ras_num_recs; i++) { + if ((control->ras_num_recs - i) >= adev->umc.retire_unit) { + if ((bps[i].address == bps[i + 1].address) && + (bps[i].mem_channel == bps[i + 1].mem_channel)) { + control->ras_num_pa_recs += adev->umc.retire_unit; + i += (adev->umc.retire_unit - 1); + } else { + control->ras_num_mca_recs += + (control->ras_num_recs - i); + break; + } } else { - control->ras_num_mca_recs += - (control->ras_num_recs - i); + control->ras_num_mca_recs += (control->ras_num_recs - i); break; } - } else { - control->ras_num_mca_recs += (control->ras_num_recs - i); - break; } + } else { + control->ras_num_mca_recs = control->ras_num_recs; } } -- cgit v1.2.3 From 699bff37a56c814c414f5b7ca9e49ef8aae660b9 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Thu, 10 Apr 2025 19:29:57 +0800 Subject: drm/amdgpu: add get_retire_flip_bits for UMC Add the general interface to get flip bits for RAS bad page retirement. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 4 ++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 15 +++++++++++++++ 2 files changed, 19 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index babc3c9cad65..cf794cf7e262 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -3473,6 +3473,10 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev) if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr) control->ras_num_pa_recs = control->ras_num_recs; + if (adev->umc.ras && + adev->umc.ras->get_retire_flip_bits) + adev->umc.ras->get_retire_flip_bits(adev); + if (control->ras_num_recs) { ret = amdgpu_ras_load_bad_pages(adev); if (ret) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 857693bcd8d4..78a8b8654573 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -78,6 +78,18 @@ #define UMC_NPS_SHIFT 40 #define UMC_NPS_MASK 0xffULL +/* three column bits and one row bit in MCA address flip + * in bad page retirement + */ +#define RETIRE_FLIP_BITS_NUM 4 + +struct amdgpu_umc_flip_bits { + uint32_t flip_bits_in_pa[RETIRE_FLIP_BITS_NUM]; + uint32_t flip_row_bit; + uint32_t r13_in_pa; + uint32_t bit_num; +}; + typedef int (*umc_func)(struct amdgpu_device *adev, uint32_t node_inst, uint32_t umc_inst, uint32_t ch_inst, void *data); @@ -100,6 +112,7 @@ struct amdgpu_umc_ras { bool dump_addr); uint32_t (*get_die_id_from_pa)(struct amdgpu_device *adev, uint64_t mca_addr, uint64_t retired_page); + void (*get_retire_flip_bits)(struct amdgpu_device *adev); }; struct amdgpu_umc_funcs { @@ -130,6 +143,8 @@ struct amdgpu_umc { /* active mask for umc node instance */ unsigned long active_mask; + + struct amdgpu_umc_flip_bits flip_bits; }; int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev); -- cgit v1.2.3 From 937467b7d5f790cd7bdc02ac3b25a7362e8b166e Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Tue, 6 May 2025 16:38:27 +0530 Subject: drm/amdgpu: Log RAS errors during load During driver load, RAS event manager may not be initialized. This will cause any ATHUB event during driver load to be skipped in dmesg log. Log the error in dmesg log for easier diagnosis. Signed-off-by: Lijo Lazar Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index cf794cf7e262..dc07936d2fcb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -4498,8 +4498,11 @@ void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) enum ras_event_type type = RAS_EVENT_TYPE_FATAL; u64 event_id; - if (amdgpu_ras_mark_ras_event(adev, type)) + if (amdgpu_ras_mark_ras_event(adev, type)) { + dev_err(adev->dev, + "uncorrectable hardware error (ERREVENT_ATHUB_INTERRUPT) detected!\n"); return; + } event_id = amdgpu_ras_acquire_event_id(adev, type); -- cgit v1.2.3 From b758667f55a09bc86ac0cb230c4e0e5ca931a3a5 Mon Sep 17 00:00:00 2001 From: Mangesh Gadre Date: Tue, 13 May 2025 13:20:20 +0800 Subject: drm/amdgpu: update ras support check update ras support check for vcn 5.0.1 Signed-off-by: Mangesh Gadre Reviewed-by: Stanley.Yang Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index dc07936d2fcb..8a13a0c7b9cd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -3708,7 +3708,8 @@ static void amdgpu_ras_query_ras_capablity_from_vbios(struct amdgpu_device *adev */ if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(2, 6, 0) || amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 0) || - amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 3)) + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 3) || + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(5, 0, 1)) adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN | 1 << AMDGPU_RAS_BLOCK__JPEG); else -- cgit v1.2.3 From 31e837d242cbb58afed8e0d2a722745bc50ad154 Mon Sep 17 00:00:00 2001 From: ganglxie Date: Thu, 22 May 2025 14:28:52 +0800 Subject: drm/amdgpu: handle old RAS eeprom data in non-nps1 mode Get MCA address from PA in nps1, then convert MCA address to PA in specific nps mode. Signed-off-by: ganglxie Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 16 ++++++++++++++-- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 23 +++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 2 ++ 3 files changed, 39 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 8a13a0c7b9cd..b0b1dbd4ae35 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2886,8 +2886,20 @@ static int __amdgpu_ras_convert_rec_from_rom(struct amdgpu_device *adev, bps->retired_page << AMDGPU_GPU_PAGE_SHIFT)) return -EINVAL; } else { - if (amdgpu_ras_mca2pa_by_idx(adev, bps, err_data)) - return -EINVAL; + if (bps->address) { + if (amdgpu_ras_mca2pa_by_idx(adev, bps, err_data)) + return -EINVAL; + } else { + /* for specific old eeprom data, mca address is not stored, + * calc it from pa + */ + if (amdgpu_umc_pa2mca(adev, bps->retired_page << AMDGPU_GPU_PAGE_SHIFT, + &(bps->address), AMDGPU_NPS1_PARTITION_MODE)) + return -EINVAL; + + if (amdgpu_ras_mca2pa(adev, bps, err_data)) + return -EOPNOTSUPP; + } } return __amdgpu_ras_restore_bad_pages(adev, err_data->err_addr, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 8c6e55b5b967..c92b8794aa73 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -562,3 +562,26 @@ int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev, return 0; } + +int amdgpu_umc_pa2mca(struct amdgpu_device *adev, + uint64_t pa, uint64_t *mca, enum amdgpu_memory_partition nps) +{ + struct ta_ras_query_address_input addr_in; + struct ta_ras_query_address_output addr_out; + int ret; + + /* nps: the pa belongs to */ + addr_in.pa.pa = pa | ((uint64_t)nps << 58); + addr_in.addr_type = TA_RAS_PA_TO_MCA; + ret = psp_ras_query_address(&adev->psp, &addr_in, &addr_out); + if (ret) { + dev_warn(adev->dev, "Failed to query RAS MCA address for 0x%llx", + pa); + + return ret; + } + + *mca = addr_out.ma.err_addr; + + return 0; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 29ce6b1d214a..ec203f9e5ffa 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -189,4 +189,6 @@ int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev, uint64_t err_addr, uint32_t ch, uint32_t umc, uint32_t node, uint32_t socket, struct ta_ras_query_address_output *addr_out, bool dump_addr); +int amdgpu_umc_pa2mca(struct amdgpu_device *adev, + uint64_t pa, uint64_t *mca, enum amdgpu_memory_partition nps); #endif -- cgit v1.2.3 From fce0afca3562ceb5386241159280590beeee2f51 Mon Sep 17 00:00:00 2001 From: ganglxie Date: Thu, 22 May 2025 15:14:28 +0800 Subject: drm/amdgpu: Get mca address for old eeprom records after getting mca address for old eeprom records with 'address==0', it can be correctly parsed under none-nps1, or it will be dropped. Signed-off-by: ganglxie Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index b0b1dbd4ae35..de0944947eaf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2859,6 +2859,15 @@ static int __amdgpu_ras_convert_rec_array_from_rom(struct amdgpu_device *adev, return -EINVAL; } } else { + if (bps[0].address == 0) { + /* for specific old eeprom data, mca address is not stored, + * calc it from pa + */ + if (amdgpu_umc_pa2mca(adev, bps[0].retired_page << AMDGPU_GPU_PAGE_SHIFT, + &(bps[0].address), AMDGPU_NPS1_PARTITION_MODE)) + return -EINVAL; + } + if (amdgpu_ras_mca2pa(adev, &bps[0], err_data)) { if (nps == AMDGPU_NPS1_PARTITION_MODE) memcpy(err_data->err_addr, bps, -- cgit v1.2.3