driver inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/IBQK42 ---------------------------------------------------------------------- Driver runs a for-loop when allocating bt pages and mapping them with buffer pages. When a large buffer (e.g. MR over 100GB) is being allocated, it may require a considerable loop count. This will lead to soft lockup: watchdog: BUG: soft lockup - CPU#27 stuck for 22s! ... Call trace: hem_list_alloc_mid_bt+0x124/0x394 [hns_roce_hw_v2] hns_roce_hem_list_request+0xf8/0x160 [hns_roce_hw_v2] hns_roce_mtr_create+0x2e4/0x360 [hns_roce_hw_v2] alloc_mr_pbl+0xd4/0x17c [hns_roce_hw_v2] hns_roce_reg_user_mr+0xf8/0x190 [hns_roce_hw_v2] ib_uverbs_reg_mr+0x118/0x290 watchdog: BUG: soft lockup - CPU#35 stuck for 23s! ... Call trace: hns_roce_hem_list_find_mtt+0x7c/0xb0 [hns_roce_hw_v2] mtr_map_bufs+0xc4/0x204 [hns_roce_hw_v2] hns_roce_mtr_create+0x31c/0x3c4 [hns_roce_hw_v2] alloc_mr_pbl+0xb0/0x160 [hns_roce_hw_v2] hns_roce_reg_user_mr+0x108/0x1c0 [hns_roce_hw_v2] ib_uverbs_reg_mr+0x120/0x2bc Add a cond_resched() to fix soft lockup during these loops. In order not to affect the allocation performance of normal-size buffer, set the loop count of a 100GB MR as the threshold to call cond_resched(). Fixes: 38389eaa4db1 ("RDMA/hns: Add mtr support for mixed multihop addressing") Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com> Signed-off-by: Xinghai Cen <cenxinghai@h-partners.com> --- drivers/infiniband/hw/hns/hns_roce_hem.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c index 6d48ccaed1f0..144b8ed4fbb6 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.c +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c @@ -1420,6 +1420,11 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev, return ret; } +/* This is the bottom bt pages number of a 100G MR on 4K OS, assuming + * the bt page size is not expanded by cal_best_bt_pg_sz() + */ +#define RESCHED_LOOP_CNT_THRESHOLD_ON_4K 12800 + /* construct the base address table and link them by address hop config */ int hns_roce_hem_list_request(struct hns_roce_dev *hr_dev, struct hns_roce_hem_list *hem_list, @@ -1428,6 +1433,7 @@ int hns_roce_hem_list_request(struct hns_roce_dev *hr_dev, { const struct hns_roce_buf_region *r; int ofs, end; + int loop; int unit; int ret; int i; @@ -1445,7 +1451,10 @@ int hns_roce_hem_list_request(struct hns_roce_dev *hr_dev, continue; end = r->offset + r->count; - for (ofs = r->offset; ofs < end; ofs += unit) { + for (ofs = r->offset, loop = 1; ofs < end; ofs += unit, loop++) { + if (!(loop % RESCHED_LOOP_CNT_THRESHOLD_ON_4K)) + cond_resched(); + ret = hem_list_alloc_mid_bt(hr_dev, r, unit, ofs, hem_list->mid_bt[i], &hem_list->btm_bt); @@ -1502,9 +1511,14 @@ void *hns_roce_hem_list_find_mtt(struct hns_roce_dev *hr_dev, struct list_head *head = &hem_list->btm_bt; struct hns_roce_hem_item *hem, *temp_hem; void *cpu_base = NULL; + int loop = 1; int nr = 0; list_for_each_entry_safe(hem, temp_hem, head, sibling) { + if (!(loop % RESCHED_LOOP_CNT_THRESHOLD_ON_4K)) + cond_resched(); + loop++; + if (hem_list_page_is_in_range(hem, offset)) { nr = offset - hem->start; cpu_base = hem->addr + nr * BA_BYTE_LEN; -- 2.33.0