[PATCH rdma-rc 3/3] RDMA/hns: Fix the chip hanging caused by sending doorbell during reset

From: Wei Hu (Xavier)
Date: Thu Jan 10 2019 - 08:23:33 EST


On hi08 chip, There is a possibility of chip hanging when sending
doorbell during reset. We can fix it by prohibiting doorbell during
reset.

Fixes: 2d40788825ac ("RDMA/hns: Add support for processing send wr and receive wr")
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@xxxxxxxxxx>
---
drivers/infiniband/hw/hns/hns_roce_device.h | 1 +
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 20 +++++++++++++-------
drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 11 +++++++++++
3 files changed, 25 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index cd7c2a6..a755127 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -925,6 +925,7 @@ struct hns_roce_dev {
spinlock_t bt_cmd_lock;
bool active;
bool is_reset;
+ bool dis_db;
unsigned long reset_cnt;
struct hns_roce_ib_iboe iboe;

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 5f476e9..ca89b00 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -587,7 +587,7 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
roce_set_field(sq_db.parameter, V2_DB_PARAMETER_SL_M,
V2_DB_PARAMETER_SL_S, qp->sl);

- hns_roce_write64_k((__le32 *)&sq_db, qp->sq.db_reg_l);
+ hns_roce_write64(hr_dev, (__le32 *)&sq_db, qp->sq.db_reg_l);

qp->sq_next_wqe = ind;
qp->next_sge = sge_ind;
@@ -717,7 +717,7 @@ static int hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev,
unsigned long reset_stage)
{
/* When hardware reset has been completed once or more, we should stop
- * sending mailbox&cmq to hardware. If now in .init_instance()
+ * sending mailbox&cmq&doorbell to hardware. If now in .init_instance()
* function, we should exit with error. If now at HNAE3_INIT_CLIENT
* stage of soft reset process, we should exit with error, and then
* HNAE3_INIT_CLIENT related process can rollback the operation like
@@ -726,6 +726,7 @@ static int hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev,
* reset process once again.
*/
hr_dev->is_reset = true;
+ hr_dev->dis_db = true;

if (reset_stage == HNS_ROCE_STATE_RST_INIT ||
instance_stage == HNS_ROCE_STATE_INIT)
@@ -743,8 +744,8 @@ static int hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev,
const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
unsigned long end;

- /* When hardware reset is detected, we should stop sending mailbox&cmq
- * to hardware, and wait until hardware reset finished. If now
+ /* When hardware reset is detected, we should stop sending mailbox&cmq&
+ * doorbell to hardware, and wait until hardware reset finished. If now
* in .init_instance() function, we should exit with error. If now at
* HNAE3_INIT_CLIENT stage of soft reset process, we should exit with
* error, and then HNAE3_INIT_CLIENT related process can rollback the
@@ -752,6 +753,7 @@ static int hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev,
* related process will exit with error to notify NIC driver to
* reschedule soft reset process once again.
*/
+ hr_dev->dis_db = true;
end = msecs_to_jiffies(HNS_ROCE_V2_HW_RST_TIMEOUT) + jiffies;
while (ops->get_hw_reset_stat(handle) && time_before(jiffies, end))
udelay(1);
@@ -776,9 +778,10 @@ static int hns_roce_v2_cmd_sw_resetting(struct hns_roce_dev *hr_dev)
unsigned long end;

/* When software reset is detected at .init_instance() function, we
- * should stop sending mailbox&cmq to hardware, and
+ * should stop sending mailbox&cmq&doorbell to hardware, and
* wait until hardware reset finished, we should exit with error.
*/
+ hr_dev->dis_db = true;
end = msecs_to_jiffies(HNS_ROCE_V2_HW_RST_TIMEOUT) + jiffies;
while (ops->ae_dev_reset_cnt(handle) == hr_dev->reset_cnt &&
time_before(jiffies, end))
@@ -2388,6 +2391,7 @@ static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
enum ib_cq_notify_flags flags)
{
+ struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
u32 notification_flag;
u32 doorbell[2];
@@ -2413,7 +2417,7 @@ static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
roce_set_bit(doorbell[1], V2_CQ_DB_PARAMETER_NOTIFY_S,
notification_flag);

- hns_roce_write64_k(doorbell, hr_cq->cq_db_l);
+ hns_roce_write64(hr_dev, doorbell, hr_cq->cq_db_l);

return 0;
}
@@ -4570,6 +4574,7 @@ static void hns_roce_v2_init_irq_work(struct hns_roce_dev *hr_dev,

static void set_eq_cons_index_v2(struct hns_roce_eq *eq)
{
+ struct hns_roce_dev *hr_dev = eq->hr_dev;
u32 doorbell[2];

doorbell[0] = 0;
@@ -4596,7 +4601,7 @@ static void set_eq_cons_index_v2(struct hns_roce_eq *eq)
HNS_ROCE_V2_EQ_DB_PARA_S,
(eq->cons_index & HNS_ROCE_V2_CONS_IDX_M));

- hns_roce_write64_k(doorbell, eq->doorbell);
+ hns_roce_write64(hr_dev, doorbell, eq->doorbell);
}

static struct hns_roce_aeqe *get_aeqe_v2(struct hns_roce_eq *eq, u32 entry)
@@ -6137,6 +6142,7 @@ static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
return 0;

hr_dev->active = false;
+ hr_dev->dis_db = true;

event.event = IB_EVENT_DEVICE_FATAL;
event.device = &hr_dev->ib_dev;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 17b3299..5f0f35e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -1740,4 +1740,15 @@ struct hns_roce_wqe_atomic_seg {
__le64 cmp_data;
};

+static inline void hns_roce_write64(struct hns_roce_dev *hr_dev, __le32 val[2],
+ void __iomem *dest)
+{
+ struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+ struct hnae3_handle *handle = priv->handle;
+ const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
+
+ if (!hr_dev->dis_db && !ops->get_hw_reset_stat(handle))
+ hns_roce_write64_k(val, dest);
+}
+
#endif
--
1.9.1