From af730e08203522dbf5a03853576c5b43c9d1afea Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Fri, 29 Mar 2024 14:42:59 +0800 Subject: drm/amdgpu: Add interface to reserve bad page MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Add interface to reserve bad page. Signed-off-by: YiPeng Chai Reviewed-by: Christian König Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 19 +++++++++++++++++++ 1 file 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 352ce16a0963..fe4fdb079bd1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2792,6 +2792,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } } + mutex_init(&con->page_rsv_lock); mutex_init(&con->page_retirement_lock); init_waitqueue_head(&con->page_retirement_wq); atomic_set(&con->page_retirement_req_cnt, 0); @@ -2842,6 +2843,8 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) atomic_set(&con->page_retirement_req_cnt, 0); + mutex_destroy(&con->page_rsv_lock); + cancel_work_sync(&con->recovery_work); mutex_lock(&con->recovery_lock); @@ -4280,3 +4283,19 @@ void amdgpu_ras_query_boot_status(struct amdgpu_device *adev, u32 num_instances) amdgpu_ras_boot_time_error_reporting(adev, i, boot_error); } } + +int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; + uint64_t start = pfn << AMDGPU_GPU_PAGE_SHIFT; + int ret = 0; + + mutex_lock(&con->page_rsv_lock); + ret = amdgpu_vram_mgr_query_page_status(mgr, start); + if (ret == -ENOENT) + ret = amdgpu_vram_mgr_reserve_range(mgr, start, AMDGPU_GPU_PAGE_SIZE); + mutex_unlock(&con->page_rsv_lock); + + return ret; +} -- cgit v1.2.3 From 98b5bc878d4b522c035309c8f6d3247d54050369 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Tue, 19 Mar 2024 09:57:58 +0800 Subject: drm/amdgpu: add message fifo to handle RAS poison events Add message fifo to handle RAS poison events. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 35 +++++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 18 +++++++++++++++++ 2 files changed, 53 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 fe4fdb079bd1..6641f27cb35c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2703,6 +2703,40 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, } } +int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) +{ + int ret = 0; + struct ras_poison_msg poison_msg; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + memset(&poison_msg, 0, sizeof(poison_msg)); + poison_msg.block = block; + poison_msg.pasid = pasid; + poison_msg.reset = reset; + poison_msg.pasid_fn = pasid_fn; + poison_msg.data = data; + + ret = kfifo_put(&con->poison_fifo, poison_msg); + if (!ret) { + dev_err(adev->dev, "Poison message fifo is full!\n"); + return -ENOSPC; + } + + return 0; +} + +#ifdef PRE_DEFINED_FUNCTION +static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, + struct ras_poison_msg *poison_msg) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + return kfifo_get(&con->poison_fifo, poison_msg); +} +#endif + static int amdgpu_ras_page_retirement_thread(void *param) { struct amdgpu_device *adev = (struct amdgpu_device *)param; @@ -2793,6 +2827,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } mutex_init(&con->page_rsv_lock); + INIT_KFIFO(con->poison_fifo); mutex_init(&con->page_retirement_lock); init_waitqueue_head(&con->page_retirement_wq); atomic_set(&con->page_retirement_req_cnt, 0); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h index ab5bf573378e..2b15996f1ede 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h @@ -26,6 +26,7 @@ #include #include +#include #include "ta_ras_if.h" #include "amdgpu_ras_eeprom.h" #include "amdgpu_smuio.h" @@ -442,6 +443,17 @@ struct ras_query_context { u64 event_id; }; +typedef int (*pasid_notify)(struct amdgpu_device *adev, + uint16_t pasid, void *data); + +struct ras_poison_msg { + enum amdgpu_ras_block block; + uint16_t pasid; + uint32_t reset; + pasid_notify pasid_fn; + void *data; +}; + struct amdgpu_ras { /* ras infrastructure */ /* for ras itself. */ @@ -501,6 +513,8 @@ struct amdgpu_ras { struct mutex page_retirement_lock; atomic_t page_retirement_req_cnt; struct mutex page_rsv_lock; + DECLARE_KFIFO(poison_fifo, struct ras_poison_msg, 128); + /* Fatal error detected flag */ atomic_t fed; @@ -913,4 +927,8 @@ u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn); +int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset); + #endif -- cgit v1.2.3 From f493dd64ee6680dc5bb46d7c800346eadb18049a Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:37:36 +0800 Subject: drm/amdgpu: prepare for logging ecc errors Prepare for logging ecc errors. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 32 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 23 +++++++++++++++++++++++ 2 files changed, 55 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 6641f27cb35c..fb2f88005cf3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2737,6 +2737,35 @@ static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, } #endif +static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) +{ + mutex_init(&ecc_log->lock); + + /* Set any value as siphash key */ + memset(&ecc_log->ecc_key, 0xad, sizeof(ecc_log->ecc_key)); + + INIT_RADIX_TREE(&ecc_log->de_page_tree, GFP_KERNEL); + ecc_log->de_updated = false; +} + +static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) +{ + struct radix_tree_iter iter; + void __rcu **slot; + struct ras_ecc_err *ecc_err; + + mutex_lock(&ecc_log->lock); + radix_tree_for_each_slot(slot, &ecc_log->de_page_tree, &iter, 0) { + ecc_err = radix_tree_deref_slot(slot); + kfree(ecc_err->err_pages.pfn); + kfree(ecc_err); + radix_tree_iter_delete(&ecc_log->de_page_tree, &iter, slot); + } + mutex_unlock(&ecc_log->lock); + + mutex_destroy(&ecc_log->lock); + ecc_log->de_updated = false; +} static int amdgpu_ras_page_retirement_thread(void *param) { struct amdgpu_device *adev = (struct amdgpu_device *)param; @@ -2838,6 +2867,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); } + amdgpu_ras_ecc_log_init(&con->umc_ecc_log); #ifdef CONFIG_X86_MCE_AMD if ((adev->asic_type == CHIP_ALDEBARAN) && (adev->gmc.xgmi.connected_to_cpu)) @@ -2882,6 +2912,8 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) cancel_work_sync(&con->recovery_work); + amdgpu_ras_ecc_log_fini(&con->umc_ecc_log); + mutex_lock(&con->recovery_lock); con->eh_data = NULL; kfree(data->bps); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h index 2b15996f1ede..634654cf2634 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h @@ -27,6 +27,8 @@ #include #include #include +#include +#include #include "ta_ras_if.h" #include "amdgpu_ras_eeprom.h" #include "amdgpu_smuio.h" @@ -454,6 +456,26 @@ struct ras_poison_msg { void *data; }; +struct ras_err_pages { + uint32_t count; + uint64_t *pfn; +}; + +struct ras_ecc_err { + u64 hash_index; + uint64_t status; + uint64_t ipid; + uint64_t addr; + struct ras_err_pages err_pages; +}; + +struct ras_ecc_log_info { + struct mutex lock; + siphash_key_t ecc_key; + struct radix_tree_root de_page_tree; + bool de_updated; +}; + struct amdgpu_ras { /* ras infrastructure */ /* for ras itself. */ @@ -514,6 +536,7 @@ struct amdgpu_ras { atomic_t page_retirement_req_cnt; struct mutex page_rsv_lock; DECLARE_KFIFO(poison_fifo, struct ras_poison_msg, 128); + struct ras_ecc_log_info umc_ecc_log; /* Fatal error detected flag */ atomic_t fed; -- cgit v1.2.3 From a734adfbcdb0e8c382fc41d3fe8b7d194c6535f6 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 11 Mar 2024 17:02:40 +0800 Subject: drm/amdgpu: add poison creation handler Add poison creation handler. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 76 ++++++++++++++++++++++++++++++--- 1 file changed, 69 insertions(+), 7 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 fb2f88005cf3..66186e28ab92 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2076,6 +2076,17 @@ static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj { dev_info(obj->adev->dev, "Poison is created\n"); + + if (amdgpu_ip_version(obj->adev, UMC_HWIP, 0) >= IP_VERSION(12, 0, 0)) { + struct amdgpu_ras *con = amdgpu_ras_get_context(obj->adev); + + amdgpu_ras_put_poison_req(obj->adev, + AMDGPU_RAS_BLOCK__UMC, 0, NULL, NULL, false); + + atomic_inc(&con->page_retirement_req_cnt); + + wake_up(&con->page_retirement_wq); + } } static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, @@ -2727,7 +2738,6 @@ int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, return 0; } -#ifdef PRE_DEFINED_FUNCTION static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, struct ras_poison_msg *poison_msg) { @@ -2735,7 +2745,6 @@ static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, return kfifo_get(&con->poison_fifo, poison_msg); } -#endif static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) { @@ -2766,10 +2775,54 @@ static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) mutex_destroy(&ecc_log->lock); ecc_log->de_updated = false; } + +static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, + enum amdgpu_ras_block ras_block, uint32_t timeout_ms) +{ + int ret = 0; + struct ras_ecc_log_info *ecc_log; + struct ras_query_if info; + uint32_t timeout = timeout_ms; + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + + memset(&info, 0, sizeof(info)); + info.head.block = ras_block; + + ecc_log = &ras->umc_ecc_log; + ecc_log->de_updated = false; + do { + ret = amdgpu_ras_query_error_status(adev, &info); + if (ret) { + dev_err(adev->dev, "Failed to query ras error! ret:%d\n", ret); + return ret; + } + + if (timeout && !ecc_log->de_updated) { + msleep(1); + timeout--; + } + } while (timeout && !ecc_log->de_updated); + + if (timeout_ms && !timeout) { + dev_warn(adev->dev, "Can't find deferred error\n"); + return -ETIMEDOUT; + } + + return 0; +} + +static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, + uint32_t timeout) +{ + amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); +} + static int amdgpu_ras_page_retirement_thread(void *param) { struct amdgpu_device *adev = (struct amdgpu_device *)param; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_poison_msg poison_msg; + enum amdgpu_ras_block ras_block; while (!kthread_should_stop()) { @@ -2780,13 +2833,22 @@ static int amdgpu_ras_page_retirement_thread(void *param) if (kthread_should_stop()) break; - dev_info(adev->dev, "Start processing page retirement. request:%d\n", - atomic_read(&con->page_retirement_req_cnt)); - atomic_dec(&con->page_retirement_req_cnt); - amdgpu_umc_bad_page_polling_timeout(adev, - 0, MAX_UMC_POISON_POLLING_TIME_ASYNC); + if (!amdgpu_ras_get_poison_req(adev, &poison_msg)) + continue; + + ras_block = poison_msg.block; + + dev_info(adev->dev, "Start processing ras block %s(%d)\n", + ras_block_str(ras_block), ras_block); + + if (ras_block == AMDGPU_RAS_BLOCK__UMC) + amdgpu_ras_poison_creation_handler(adev, + MAX_UMC_POISON_POLLING_TIME_ASYNC); + else + amdgpu_umc_bad_page_polling_timeout(adev, + false, MAX_UMC_POISON_POLLING_TIME_ASYNC); } return 0; -- cgit v1.2.3 From 95b4063de4f418135984b33528c44eeb4f9f4baa Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Tue, 19 Mar 2024 10:14:16 +0800 Subject: drm/amdgpu: add interface to update umc v12_0 ecc status Add interface to update umc v12_0 ecc status. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 9 ++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 6 ++++++ drivers/gpu/drm/amd/amdgpu/umc_v12_0.c | 24 ++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/umc_v12_0.h | 3 +++ .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c | 5 +++++ 6 files changed, 49 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 66186e28ab92..63b5723e26ea 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -4215,6 +4215,8 @@ void amdgpu_ras_add_mca_err_addr(struct ras_err_info *err_info, struct ras_err_a { struct ras_err_addr *mca_err_addr; + /* This function will be retired. */ + return; mca_err_addr = kzalloc(sizeof(*mca_err_addr), GFP_KERNEL); if (!mca_err_addr) return; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index f486510fc94c..7006a57277ef 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -437,3 +437,12 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev, return 0; } + +int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, + uint64_t status, uint64_t ipid, uint64_t addr) +{ + if (adev->umc.ras->update_ecc_status) + return adev->umc.ras->update_ecc_status(adev, + status, ipid, addr); + return 0; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 563b0249247e..4f3834fa10a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -66,6 +66,8 @@ struct amdgpu_umc_ras { void *ras_error_status); bool (*check_ecc_err_status)(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, void *ras_error_status); + int (*update_ecc_status)(struct amdgpu_device *adev, + uint64_t status, uint64_t ipid, uint64_t addr); }; struct amdgpu_umc_funcs { @@ -122,4 +124,8 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev, int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, uint32_t reset, uint32_t timeout_ms); + +int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, + uint64_t status, uint64_t ipid, uint64_t addr); + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c index a0122b22eda4..81435533c4a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c @@ -479,6 +479,29 @@ static int umc_v12_0_ras_late_init(struct amdgpu_device *adev, struct ras_common return 0; } +static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev, + uint64_t status, uint64_t ipid, uint64_t addr) +{ + uint16_t hwid, mcatype; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + hwid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, HardwareID); + mcatype = REG_GET_FIELD(ipid, MCMP1_IPIDT0, McaType); + + if ((hwid != MCA_UMC_HWID_V12_0) || (mcatype != MCA_UMC_MCATYPE_V12_0)) + return 0; + + if (!status) + return 0; + + if (!umc_v12_0_is_deferred_error(adev, status)) + return 0; + + con->umc_ecc_log.de_updated = true; + + return 0; +} + struct amdgpu_umc_ras umc_v12_0_ras = { .ras_block = { .hw_ops = &umc_v12_0_ras_hw_ops, @@ -489,5 +512,6 @@ struct amdgpu_umc_ras umc_v12_0_ras = { .ecc_info_query_ras_error_count = umc_v12_0_ecc_info_query_ras_error_count, .ecc_info_query_ras_error_address = umc_v12_0_ecc_info_query_ras_error_address, .check_ecc_err_status = umc_v12_0_check_ecc_err_status, + .update_ecc_status = umc_v12_0_update_ecc_status, }; diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h index 1d5f44dcffdd..5c2d7e127608 100644 --- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h +++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h @@ -62,6 +62,9 @@ /* row bits in SOC physical address */ #define UMC_V12_0_PA_R13_BIT 35 +#define MCA_UMC_HWID_V12_0 0x96 +#define MCA_UMC_MCATYPE_V12_0 0x0 + #define MCA_IPID_LO_2_UMC_CH(_ipid_lo) (((((_ipid_lo) >> 20) & 0x1) * 4) + \ (((_ipid_lo) >> 12) & 0xF)) #define MCA_IPID_LO_2_UMC_INST(_ipid_lo) (((_ipid_lo) >> 21) & 0x7) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c index 7a9f85dc06b5..4d3eca2fc3f1 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c @@ -2716,6 +2716,11 @@ static int mca_umc_mca_get_err_count(const struct mca_ras_info *mca_ras, struct umc_v12_0_is_correctable_error(adev, status0)) *count = (ext_error_code == 0) ? odecc_err_cnt : 1; + amdgpu_umc_update_ecc_status(adev, + entry->regs[MCA_REG_IDX_STATUS], + entry->regs[MCA_REG_IDX_IPID], + entry->regs[MCA_REG_IDX_ADDR]); + return 0; } -- cgit v1.2.3 From 2cf8e50ec381e6a6be3835a421f279d88fcb5ba4 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:38:54 +0800 Subject: drm/amdgpu: Add delay work to retire bad pages Add delay work to retire bad pages. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 36 ++++++++++++++++++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 3 +++ 4 files changed, 40 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 63b5723e26ea..72daa51f8beb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -122,6 +122,8 @@ const char *get_ras_block_str(struct ras_common_if *ras_block) #define MAX_UMC_POISON_POLLING_TIME_ASYNC 100 //ms +#define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100 //ms + enum amdgpu_ras_retire_page_reservation { AMDGPU_RAS_RETIRE_PAGE_RESERVED, AMDGPU_RAS_RETIRE_PAGE_PENDING, @@ -2776,6 +2778,30 @@ static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) ecc_log->de_updated = false; } +static void amdgpu_ras_do_page_retirement(struct work_struct *work) +{ + struct amdgpu_ras *con = container_of(work, struct amdgpu_ras, + page_retirement_dwork.work); + struct amdgpu_device *adev = con->adev; + struct ras_err_data err_data; + + if (amdgpu_in_reset(adev) || atomic_read(&con->in_recovery)) + return; + + amdgpu_ras_error_data_init(&err_data); + + amdgpu_umc_handle_bad_pages(adev, &err_data); + + amdgpu_ras_error_data_fini(&err_data); + + mutex_lock(&con->umc_ecc_log.lock); + if (radix_tree_tagged(&con->umc_ecc_log.de_page_tree, + UMC_ECC_NEW_DETECTED_TAG)) + schedule_delayed_work(&con->page_retirement_dwork, + msecs_to_jiffies(AMDGPU_RAS_RETIRE_PAGE_INTERVAL)); + mutex_unlock(&con->umc_ecc_log.lock); +} + static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, enum amdgpu_ras_block ras_block, uint32_t timeout_ms) { @@ -2814,7 +2840,12 @@ static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, uint32_t timeout) { - amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + int ret; + + ret = amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); + if (!ret) + schedule_delayed_work(&con->page_retirement_dwork, 0); } static int amdgpu_ras_page_retirement_thread(void *param) @@ -2929,6 +2960,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); } + INIT_DELAYED_WORK(&con->page_retirement_dwork, amdgpu_ras_do_page_retirement); amdgpu_ras_ecc_log_init(&con->umc_ecc_log); #ifdef CONFIG_X86_MCE_AMD if ((adev->asic_type == CHIP_ALDEBARAN) && @@ -2974,6 +3006,8 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) cancel_work_sync(&con->recovery_work); + cancel_delayed_work_sync(&con->page_retirement_dwork); + amdgpu_ras_ecc_log_fini(&con->umc_ecc_log); mutex_lock(&con->recovery_lock); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h index 634654cf2634..cb5a0f31d201 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h @@ -537,6 +537,7 @@ struct amdgpu_ras { struct mutex page_rsv_lock; DECLARE_KFIFO(poison_fifo, struct ras_poison_msg, 128); struct ras_ecc_log_info umc_ecc_log; + struct delayed_work page_retirement_dwork; /* Fatal error detected flag */ atomic_t fed; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 0f2d765c4e2d..2bd88218c20e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -89,7 +89,7 @@ out_fini_err_data: return ret; } -static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, +void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, void *ras_error_status) { struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index c83d24097c5c..2d08d076f7c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -133,4 +133,7 @@ int amdgpu_umc_build_pages_hash(struct amdgpu_device *adev, uint64_t *pfns, int len, uint64_t *val); int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev, struct radix_tree_root *ecc_tree, struct ras_ecc_err *ecc_err); + +void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, + void *ras_error_status); #endif -- cgit v1.2.3 From 370fbff4cc6fe02ddeb1aeff43fea3e32b828e6a Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 11 Mar 2024 17:52:37 +0800 Subject: drm/amdgpu: add poison consumption handler Add poison consumption handler. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 43 ++++++++++++++++++++++++++++++--- 1 file changed, 39 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 72daa51f8beb..2b6018e7b696 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2848,12 +2848,35 @@ static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, schedule_delayed_work(&con->page_retirement_dwork, 0); } +static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev, + struct ras_poison_msg *poison_msg) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + uint32_t reset = poison_msg->reset; + uint16_t pasid = poison_msg->pasid; + + kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); + + if (poison_msg->pasid_fn) + poison_msg->pasid_fn(adev, pasid, poison_msg->data); + + if (reset) { + flush_delayed_work(&con->page_retirement_dwork); + + con->gpu_reset_flags |= reset; + amdgpu_ras_reset_gpu(adev); + } + + return 0; +} + static int amdgpu_ras_page_retirement_thread(void *param) { struct amdgpu_device *adev = (struct amdgpu_device *)param; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_poison_msg poison_msg; enum amdgpu_ras_block ras_block; + bool poison_creation_is_handled = false; while (!kthread_should_stop()) { @@ -2874,12 +2897,24 @@ static int amdgpu_ras_page_retirement_thread(void *param) dev_info(adev->dev, "Start processing ras block %s(%d)\n", ras_block_str(ras_block), ras_block); - if (ras_block == AMDGPU_RAS_BLOCK__UMC) + if (ras_block == AMDGPU_RAS_BLOCK__UMC) { amdgpu_ras_poison_creation_handler(adev, MAX_UMC_POISON_POLLING_TIME_ASYNC); - else - amdgpu_umc_bad_page_polling_timeout(adev, - false, MAX_UMC_POISON_POLLING_TIME_ASYNC); + poison_creation_is_handled = true; + } else { + /* poison_creation_is_handled: + * false: no poison creation interrupt, but it has poison + * consumption interrupt. + * true: It has poison creation interrupt at the beginning, + * but it has no poison creation interrupt later. + */ + amdgpu_ras_poison_creation_handler(adev, + poison_creation_is_handled ? + 0 : MAX_UMC_POISON_POLLING_TIME_ASYNC); + + amdgpu_ras_poison_consumption_handler(adev, &poison_msg); + poison_creation_is_handled = false; + } } return 0; -- cgit v1.2.3 From bcc093488503226f0d5519d2f0561c497b15cb39 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Wed, 3 Apr 2024 14:11:41 +0800 Subject: drm/amdgpu: Fix address translation defect retired_page is page frame and should be expanded to the full address when querying status. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou 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 2b6018e7b696..0e6a18ec48bd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2399,7 +2399,7 @@ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, }; status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, - data->bps[i].retired_page); + data->bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT); if (status == -EBUSY) (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; else if (status == -ENOENT) -- cgit v1.2.3 From 48fa90718b2ae1d0f17ba94f84e4f93d9f6068cd Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 8 Apr 2024 14:48:50 +0800 Subject: drm/amdgpu: Use new interface to reserve bad page Use new interface to reserve bad page. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 4 +--- 1 file changed, 1 insertion(+), 3 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 0e6a18ec48bd..8c58fd8a08de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2558,9 +2558,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, goto out; } - amdgpu_vram_mgr_reserve_range(&adev->mman.vram_mgr, - bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT, - AMDGPU_GPU_PAGE_SIZE); + amdgpu_ras_reserve_page(adev, bps[i].retired_page); memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps)); data->count++; -- cgit v1.2.3 From 6f3b69139c3c1f7880ef52cc29571cb74ce8220a Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Wed, 24 Apr 2024 10:21:30 +0800 Subject: drm/amdgpu: Fix ras mode2 reset failure in ras aca mode Fix ras mode2 reset failure in ras aca mode. Signed-off-by: YiPeng Chai Reviewed-by: Yang Wang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 4 ++++ 1 file changed, 4 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 8c58fd8a08de..1adc81a55734 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -1250,6 +1250,10 @@ int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk, { struct ras_manager *obj; + /* in resume phase, no need to create aca fs node */ + if (adev->in_suspend || amdgpu_in_reset(adev)) + return 0; + obj = get_ras_manager(adev, blk); if (!obj) return -EINVAL; -- cgit v1.2.3