This is a note to let you know that I've just added the patch titled
RDMA/hns: Fix mapping error of zero-hop WQE buffer
to the 6.6-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
rdma-hns-fix-mapping-error-of-zero-hop-wqe-buffer.patch
and it can be found in the queue-6.6 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 8cc9ac82061a71ae3b77e583a7c9c234b2ef8a6a
Author: wenglianfa <wenglianfa@xxxxxxxxxx>
Date: Fri Dec 20 13:52:46 2024 +0800
RDMA/hns: Fix mapping error of zero-hop WQE buffer
[ Upstream commit 8673a6c2d9e483dfeeef83a1f06f59e05636f4d1 ]
Due to HW limitation, the three region of WQE buffer must be mapped
and set to HW in a fixed order: SQ buffer, SGE buffer, and RQ buffer.
Currently when one region is zero-hop while the other two are not,
the zero-hop region will not be mapped. This violate the limitation
above and leads to address error.
Fixes: 38389eaa4db1 ("RDMA/hns: Add mtr support for mixed multihop addressing")
Signed-off-by: wenglianfa <wenglianfa@xxxxxxxxxx>
Signed-off-by: Junxian Huang <huangjunxian6@xxxxxxxxxxxxx>
Link: https://patch.msgid.link/20241220055249.146943-2-huangjunxian6@xxxxxxxxxxxxx
Signed-off-by: Leon Romanovsky <leon@xxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index d3fabf64e390..51ab6041ca91 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -986,6 +986,7 @@ struct hns_roce_hem_item {
size_t count; /* max ba numbers */
int start; /* start buf offset in this hem */
int end; /* end buf offset in this hem */
+ bool exist_bt;
};
/* All HEM items are linked in a tree structure */
@@ -1014,6 +1015,7 @@ hem_list_alloc_item(struct hns_roce_dev *hr_dev, int start, int end, int count,
}
}
+ hem->exist_bt = exist_bt;
hem->count = count;
hem->start = start;
hem->end = end;
@@ -1024,22 +1026,22 @@ hem_list_alloc_item(struct hns_roce_dev *hr_dev, int start, int end, int count,
}
static void hem_list_free_item(struct hns_roce_dev *hr_dev,
- struct hns_roce_hem_item *hem, bool exist_bt)
+ struct hns_roce_hem_item *hem)
{
- if (exist_bt)
+ if (hem->exist_bt)
dma_free_coherent(hr_dev->dev, hem->count * BA_BYTE_LEN,
hem->addr, hem->dma_addr);
kfree(hem);
}
static void hem_list_free_all(struct hns_roce_dev *hr_dev,
- struct list_head *head, bool exist_bt)
+ struct list_head *head)
{
struct hns_roce_hem_item *hem, *temp_hem;
list_for_each_entry_safe(hem, temp_hem, head, list) {
list_del(&hem->list);
- hem_list_free_item(hr_dev, hem, exist_bt);
+ hem_list_free_item(hr_dev, hem);
}
}
@@ -1139,6 +1141,10 @@ int hns_roce_hem_list_calc_root_ba(const struct hns_roce_buf_region *regions,
for (i = 0; i < region_cnt; i++) {
r = (struct hns_roce_buf_region *)®ions[i];
+ /* when r->hopnum = 0, the region should not occupy root_ba. */
+ if (!r->hopnum)
+ continue;
+
if (r->hopnum > 1) {
step = hem_list_calc_ba_range(r->hopnum, 1, unit);
if (step > 0)
@@ -1232,7 +1238,7 @@ static int hem_list_alloc_mid_bt(struct hns_roce_dev *hr_dev,
err_exit:
for (level = 1; level < hopnum; level++)
- hem_list_free_all(hr_dev, &temp_list[level], true);
+ hem_list_free_all(hr_dev, &temp_list[level]);
return ret;
}
@@ -1273,16 +1279,26 @@ static int alloc_fake_root_bt(struct hns_roce_dev *hr_dev, void *cpu_base,
{
struct hns_roce_hem_item *hem;
+ /* This is on the has_mtt branch, if r->hopnum
+ * is 0, there is no root_ba to reuse for the
+ * region's fake hem, so a dma_alloc request is
+ * necessary here.
+ */
hem = hem_list_alloc_item(hr_dev, r->offset, r->offset + r->count - 1,
- r->count, false);
+ r->count, !r->hopnum);
if (!hem)
return -ENOMEM;
- hem_list_assign_bt(hem, cpu_base, phy_base);
+ /* The root_ba can be reused only when r->hopnum > 0. */
+ if (r->hopnum)
+ hem_list_assign_bt(hem, cpu_base, phy_base);
list_add(&hem->list, branch_head);
list_add(&hem->sibling, leaf_head);
- return r->count;
+ /* If r->hopnum == 0, 0 is returned,
+ * so that the root_bt entry is not occupied.
+ */
+ return r->hopnum ? r->count : 0;
}
static int setup_middle_bt(struct hns_roce_dev *hr_dev, void *cpu_base,
@@ -1326,7 +1342,7 @@ setup_root_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem_list *hem_list,
return -ENOMEM;
total = 0;
- for (i = 0; i < region_cnt && total < max_ba_num; i++) {
+ for (i = 0; i < region_cnt && total <= max_ba_num; i++) {
r = ®ions[i];
if (!r->count)
continue;
@@ -1392,9 +1408,9 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
region_cnt);
if (ret) {
for (i = 0; i < region_cnt; i++)
- hem_list_free_all(hr_dev, &head.branch[i], false);
+ hem_list_free_all(hr_dev, &head.branch[i]);
- hem_list_free_all(hr_dev, &head.root, true);
+ hem_list_free_all(hr_dev, &head.root);
}
return ret;
@@ -1457,10 +1473,9 @@ void hns_roce_hem_list_release(struct hns_roce_dev *hr_dev,
for (i = 0; i < HNS_ROCE_MAX_BT_REGION; i++)
for (j = 0; j < HNS_ROCE_MAX_BT_LEVEL; j++)
- hem_list_free_all(hr_dev, &hem_list->mid_bt[i][j],
- j != 0);
+ hem_list_free_all(hr_dev, &hem_list->mid_bt[i][j]);
- hem_list_free_all(hr_dev, &hem_list->root_bt, true);
+ hem_list_free_all(hr_dev, &hem_list->root_bt);
INIT_LIST_HEAD(&hem_list->btm_bt);
hem_list->root_ba = 0;
}
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index acdcd5c9f42f..408ef2a96149 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -767,11 +767,6 @@ int hns_roce_mtr_map(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
for (i = 0, mapped_cnt = 0; i < mtr->hem_cfg.region_count &&
mapped_cnt < page_cnt; i++) {
r = &mtr->hem_cfg.region[i];
- /* if hopnum is 0, no need to map pages in this region */
- if (!r->hopnum) {
- mapped_cnt += r->count;
- continue;
- }
if (r->offset + r->count > page_cnt) {
ret = -EINVAL;
