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_umc.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') 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; +} -- cgit From f27defca68824e8e97218b8816249f258d3d5d32 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 18 Mar 2024 11:48:07 +0800 Subject: drm/amdgpu: umc v12_0 logs ecc errors 1. umc v12_0 logs ecc errors. 2. Reserve newly detected ecc error pages. 3. Add tag for bad pages, so that they can be retired later. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 67 +++++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 7 +++- drivers/gpu/drm/amd/amdgpu/umc_v12_0.c | 41 +++++++++++++++++++- 3 files changed, 113 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 7006a57277ef..0f2d765c4e2d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -21,10 +21,13 @@ * */ +#include #include "amdgpu.h" #include "umc_v6_7.h" #define MAX_UMC_POISON_POLLING_TIME_SYNC 20 //ms +#define MAX_UMC_HASH_STRING_SIZE 256 + static int amdgpu_umc_convert_error_address(struct amdgpu_device *adev, struct ras_err_data *err_data, uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst) @@ -446,3 +449,67 @@ int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, status, ipid, addr); return 0; } + +static int amdgpu_umc_uint64_cmp(const void *a, const void *b) +{ + uint64_t *addr_a = (uint64_t *)a; + uint64_t *addr_b = (uint64_t *)b; + + if (*addr_a > *addr_b) + return 1; + else if (*addr_a < *addr_b) + return -1; + else + return 0; +} + +/* Use string hash to avoid logging the same bad pages repeatedly */ +int amdgpu_umc_build_pages_hash(struct amdgpu_device *adev, + uint64_t *pfns, int len, uint64_t *val) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + char buf[MAX_UMC_HASH_STRING_SIZE] = {0}; + int offset = 0, i = 0; + uint64_t hash_val; + + if (!pfns || !len) + return -EINVAL; + + sort(pfns, len, sizeof(uint64_t), amdgpu_umc_uint64_cmp, NULL); + + for (i = 0; i < len; i++) + offset += snprintf(&buf[offset], sizeof(buf) - offset, "%llx", pfns[i]); + + hash_val = siphash(buf, offset, &con->umc_ecc_log.ecc_key); + + *val = hash_val; + + return 0; +} + +int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev, + struct radix_tree_root *ecc_tree, struct ras_ecc_err *ecc_err) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_ecc_log_info *ecc_log; + int ret; + + ecc_log = &con->umc_ecc_log; + + mutex_lock(&ecc_log->lock); + ret = radix_tree_insert(ecc_tree, ecc_err->hash_index, ecc_err); + if (!ret) { + struct ras_err_pages *err_pages = &ecc_err->err_pages; + int i; + + /* Reserve memory */ + for (i = 0; i < err_pages->count; i++) + amdgpu_ras_reserve_page(adev, err_pages->pfn[i]); + + radix_tree_tag_set(ecc_tree, + ecc_err->hash_index, UMC_ECC_NEW_DETECTED_TAG); + } + mutex_unlock(&ecc_log->lock); + + return ret; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 4f3834fa10a8..c83d24097c5c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -52,6 +52,8 @@ #define LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) \ LOOP_UMC_NODE_INST((node_inst)) LOOP_UMC_INST_AND_CH((umc_inst), (ch_inst)) +/* Page retirement tag */ +#define UMC_ECC_NEW_DETECTED_TAG 0x1 typedef int (*umc_func)(struct amdgpu_device *adev, uint32_t node_inst, uint32_t umc_inst, uint32_t ch_inst, void *data); @@ -127,5 +129,8 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, uint64_t status, uint64_t ipid, uint64_t addr); - +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); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c index 085dcfe16b5e..6c2b61ef5b57 100644 --- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c @@ -546,8 +546,10 @@ static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev, uint16_t hwid, mcatype; struct ta_ras_query_address_input addr_in; uint64_t page_pfn[UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL]; - uint64_t err_addr; + uint64_t err_addr, hash_val = 0; + struct ras_ecc_err *ecc_err; int count; + int ret; hwid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, HardwareID); mcatype = REG_GET_FIELD(ipid, MCMP1_IPIDT0, McaType); @@ -589,6 +591,43 @@ static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev, return 0; } + ret = amdgpu_umc_build_pages_hash(adev, + page_pfn, count, &hash_val); + if (ret) { + dev_err(adev->dev, "Fail to build error pages hash\n"); + return ret; + } + + ecc_err = kzalloc(sizeof(*ecc_err), GFP_KERNEL); + if (!ecc_err) + return -ENOMEM; + + ecc_err->err_pages.pfn = kcalloc(count, sizeof(*ecc_err->err_pages.pfn), GFP_KERNEL); + if (!ecc_err->err_pages.pfn) { + kfree(ecc_err); + return -ENOMEM; + } + + memcpy(ecc_err->err_pages.pfn, page_pfn, count * sizeof(*ecc_err->err_pages.pfn)); + ecc_err->err_pages.count = count; + + ecc_err->hash_index = hash_val; + ecc_err->status = status; + ecc_err->ipid = ipid; + ecc_err->addr = addr; + + ret = amdgpu_umc_logs_ecc_err(adev, &con->umc_ecc_log.de_page_tree, ecc_err); + if (ret) { + if (ret == -EEXIST) + con->umc_ecc_log.de_updated = true; + else + dev_err(adev->dev, "Fail to log ecc error! ret:%d\n", ret); + + kfree(ecc_err->err_pages.pfn); + kfree(ecc_err); + return ret; + } + con->umc_ecc_log.de_updated = true; return 0; -- cgit 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_umc.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 From e74313be5a71df63e307ad98b6ab202b8a222817 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Fri, 22 Mar 2024 12:06:01 +0800 Subject: drm/amdgpu: add condition check for amdgpu_umc_fill_error_record Add condition check for amdgpu_umc_fill_error_record. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 20 +++++++++++++++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 2 +- 3 files changed, 19 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h index cb5a0f31d201..c8980d5f6540 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h @@ -579,6 +579,7 @@ struct ras_err_data { unsigned long de_count; unsigned long err_addr_cnt; struct eeprom_table_record *err_addr; + unsigned long err_addr_len; u32 err_list_count; struct list_head err_node_list; }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 2bd88218c20e..dcda3d24bee3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -66,6 +66,8 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev, goto out_fini_err_data; } + err_data.err_addr_len = adev->umc.max_ras_err_cnt_per_query; + /* * Translate UMC channel address to Physical address */ @@ -121,6 +123,8 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, if(!err_data->err_addr) dev_warn(adev->dev, "Failed to alloc memory for " "umc error address record!\n"); + else + err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; /* umc query_ras_error_address is also responsible for clearing * error status @@ -146,6 +150,8 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, if(!err_data->err_addr) dev_warn(adev->dev, "Failed to alloc memory for " "umc error address record!\n"); + else + err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; /* umc query_ras_error_address is also responsible for clearing * error status @@ -389,14 +395,20 @@ int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, return 0; } -void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, +int amdgpu_umc_fill_error_record(struct ras_err_data *err_data, uint64_t err_addr, uint64_t retired_page, uint32_t channel_index, uint32_t umc_inst) { - struct eeprom_table_record *err_rec = - &err_data->err_addr[err_data->err_addr_cnt]; + struct eeprom_table_record *err_rec; + + if (!err_data || + !err_data->err_addr || + (err_data->err_addr_cnt >= err_data->err_addr_len)) + return -EINVAL; + + err_rec = &err_data->err_addr[err_data->err_addr_cnt]; err_rec->address = err_addr; /* page frame address is saved */ @@ -408,6 +420,8 @@ void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, err_rec->mcumc_id = umc_inst; err_data->err_addr_cnt++; + + return 0; } int amdgpu_umc_loop_channels(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 2d08d076f7c9..9e77e6d48e3b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -109,7 +109,7 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); -void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, +int amdgpu_umc_fill_error_record(struct ras_err_data *err_data, uint64_t err_addr, uint64_t retired_page, uint32_t channel_index, -- cgit From bfa579b38b865879223f61f6ae295c939d4f7d11 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:40:03 +0800 Subject: drm/amdgpu: prepare to handle pasid poison consumption Prepare to handle pasid poison consumption. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 9 ++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 20 +++++++++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 3 +++ drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 3 ++- 5 files changed, 31 insertions(+), 9 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 3b4591f554f1..7ba05f030dd1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -747,10 +747,17 @@ bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev) return amdgpu_ras_get_fed_status(adev); } +void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) +{ + amdgpu_umc_pasid_poison_handler(adev, block, pasid, pasid_fn, data, reset); +} + void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset) { - amdgpu_umc_poison_handler(adev, block, reset); + amdgpu_umc_pasid_poison_handler(adev, block, 0, NULL, NULL, reset); } int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index c51954c9052e..1de021ebdd46 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -337,6 +337,11 @@ int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev, struct tile_config *config); void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset); + +void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset); + bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev); bool amdgpu_amdkfd_bo_mapped_to_dev(struct amdgpu_device *adev, struct kgd_mem *mem); void amdgpu_amdkfd_block_mmu_notifications(void *p); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index dcda3d24bee3..8ebbca9e2e22 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -252,8 +252,9 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, return 0; } -int amdgpu_umc_poison_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, uint32_t reset) +int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) { int ret = AMDGPU_RAS_SUCCESS; @@ -291,16 +292,14 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, amdgpu_ras_error_data_fini(&err_data); } else { - if (reset) { - amdgpu_umc_bad_page_polling_timeout(adev, - reset, MAX_UMC_POISON_POLLING_TIME_SYNC); - } else { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + amdgpu_ras_put_poison_req(adev, + block, pasid, pasid_fn, data, reset); + atomic_inc(&con->page_retirement_req_cnt); wake_up(&con->page_retirement_wq); - } } } else { if (adev->virt.ops && adev->virt.ops->ras_poison_handler) @@ -313,6 +312,13 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, return ret; } +int amdgpu_umc_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint32_t reset) +{ + return amdgpu_umc_pasid_poison_handler(adev, + block, 0, NULL, NULL, reset); +} + int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 9e77e6d48e3b..5f50c69c3cec 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -106,6 +106,9 @@ int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev); int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); int amdgpu_umc_poison_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset); +int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset); int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c index c3beb872adf8..e1c21d250611 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -190,7 +190,8 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, dev_warn(dev->adev->dev, "poison is consumed by client %d, kick off gpu reset flow\n", client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset); + amdgpu_amdkfd_ras_pasid_poison_consumption_handler(dev->adev, + block, pasid, NULL, NULL, reset); } static bool context_id_expected(struct kfd_dev *dev) -- cgit From 506c245f3f1cd989cb89811a7f06e04ff8813a0d Mon Sep 17 00:00:00 2001 From: Bob Zhou Date: Tue, 23 Apr 2024 13:32:39 +0800 Subject: drm/amdgpu: fix double free err_addr pointer warnings In amdgpu_umc_bad_page_polling_timeout, the amdgpu_umc_handle_bad_pages will be run many times so that double free err_addr in some special case. So set the err_addr to NULL to avoid the warnings. Signed-off-by: Bob Zhou Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 8ebbca9e2e22..540e0f066b26 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -179,6 +179,7 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, } kfree(err_data->err_addr); + err_data->err_addr = NULL; mutex_unlock(&con->page_retirement_lock); } -- cgit