* [PATCH 22.11] net/hns3: fix crash for NEON and SVE
@ 2024-12-03 6:46 Jie Hai
2024-12-03 12:31 ` Luca Boccassi
0 siblings, 1 reply; 2+ messages in thread
From: Jie Hai @ 2024-12-03 6:46 UTC (permalink / raw)
To: stable, luca.boccassi, Dongdong Liu, Yisen Zhuang, Huisong Li,
Chengwen Feng, Wei Hu (Xavier)
Cc: haijie1
From: Huisong Li <lihuisong@huawei.com>
[ upstream commit 01843ab2f2fc8c3137258ec39b2cb6f62ba7b8a2 ]
Driver may fail to allocate bulk mbufs for Neon and SVE when rearm
mbuf. Currently, driver keeps going to handle packets even if there
isn't available descriptors to receive packets at this moment.
As a result, driver probably fills the mbufs with invalid data to
application and accesses to illegal address because of the VLD bit
of the descriptor at the "rx_rearm_start" position still being set.
So driver has to clear VLD bit for this descriptor in this scenario
in case of receiving packets later.
In addition, it is possible that the sum of the "rx_rearm_nb" and
"rx_rearm_start" is greater than total descriptor number of Rx queue
in the above scenario. So the index of rxq->sw_ring[] to set mbuf
pointer to NULL should also be fixed to avoid out-of-bounds memory
access.
Fixes: a3d4f4d291d7 ("net/hns3: support NEON Rx")
Fixes: f81a18f49152 ("net/hns3: fix mbuf leakage when RxQ started after reset")
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
drivers/net/hns3/hns3_rxtx.c | 2 +-
drivers/net/hns3/hns3_rxtx_vec.c | 5 +++++
drivers/net/hns3/hns3_rxtx_vec_sve.c | 5 +++++
3 files changed, 11 insertions(+), 1 deletion(-)
diff --git a/drivers/net/hns3/hns3_rxtx.c b/drivers/net/hns3/hns3_rxtx.c
index 16cb174f4dad..bdc162bc5c8e 100644
--- a/drivers/net/hns3/hns3_rxtx.c
+++ b/drivers/net/hns3/hns3_rxtx.c
@@ -51,7 +51,7 @@ hns3_rx_queue_release_mbufs(struct hns3_rx_queue *rxq)
}
}
for (i = 0; i < rxq->rx_rearm_nb; i++)
- rxq->sw_ring[rxq->rx_rearm_start + i].mbuf = NULL;
+ rxq->sw_ring[(rxq->rx_rearm_start + i) % rxq->nb_rx_desc].mbuf = NULL;
}
for (i = 0; i < rxq->bulk_mbuf_num; i++)
diff --git a/drivers/net/hns3/hns3_rxtx_vec.c b/drivers/net/hns3/hns3_rxtx_vec.c
index cd9264d91bbd..b6aee9af67a8 100644
--- a/drivers/net/hns3/hns3_rxtx_vec.c
+++ b/drivers/net/hns3/hns3_rxtx_vec.c
@@ -66,6 +66,11 @@ hns3_rxq_rearm_mbuf(struct hns3_rx_queue *rxq)
if (unlikely(rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
HNS3_DEFAULT_RXQ_REARM_THRESH) < 0)) {
+ /*
+ * Clear VLD bit for the first descriptor rearmed in case
+ * of going to receive packets later.
+ */
+ rxdp[0].rx.bd_base_info = 0;
rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
return;
}
diff --git a/drivers/net/hns3/hns3_rxtx_vec_sve.c b/drivers/net/hns3/hns3_rxtx_vec_sve.c
index 6f23ba674d09..2ca6a70fed6a 100644
--- a/drivers/net/hns3/hns3_rxtx_vec_sve.c
+++ b/drivers/net/hns3/hns3_rxtx_vec_sve.c
@@ -248,6 +248,11 @@ hns3_rxq_rearm_mbuf_sve(struct hns3_rx_queue *rxq)
if (unlikely(rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
HNS3_DEFAULT_RXQ_REARM_THRESH) < 0)) {
+ /*
+ * Clear VLD bit for the first descriptor rearmed in case
+ * of going to receive packets later.
+ */
+ rxdp[0].rx.bd_base_info = 0;
rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
return;
}
--
2.33.0
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH 22.11] net/hns3: fix crash for NEON and SVE
2024-12-03 6:46 [PATCH 22.11] net/hns3: fix crash for NEON and SVE Jie Hai
@ 2024-12-03 12:31 ` Luca Boccassi
0 siblings, 0 replies; 2+ messages in thread
From: Luca Boccassi @ 2024-12-03 12:31 UTC (permalink / raw)
To: Jie Hai
Cc: stable, Dongdong Liu, Yisen Zhuang, Huisong Li, Chengwen Feng,
Wei Hu (Xavier)
On Tue, 3 Dec 2024 at 06:53, Jie Hai <haijie1@huawei.com> wrote:
>
> From: Huisong Li <lihuisong@huawei.com>
>
> [ upstream commit 01843ab2f2fc8c3137258ec39b2cb6f62ba7b8a2 ]
>
> Driver may fail to allocate bulk mbufs for Neon and SVE when rearm
> mbuf. Currently, driver keeps going to handle packets even if there
> isn't available descriptors to receive packets at this moment.
> As a result, driver probably fills the mbufs with invalid data to
> application and accesses to illegal address because of the VLD bit
> of the descriptor at the "rx_rearm_start" position still being set.
> So driver has to clear VLD bit for this descriptor in this scenario
> in case of receiving packets later.
>
> In addition, it is possible that the sum of the "rx_rearm_nb" and
> "rx_rearm_start" is greater than total descriptor number of Rx queue
> in the above scenario. So the index of rxq->sw_ring[] to set mbuf
> pointer to NULL should also be fixed to avoid out-of-bounds memory
> access.
>
> Fixes: a3d4f4d291d7 ("net/hns3: support NEON Rx")
> Fixes: f81a18f49152 ("net/hns3: fix mbuf leakage when RxQ started after reset")
>
> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> ---
> drivers/net/hns3/hns3_rxtx.c | 2 +-
> drivers/net/hns3/hns3_rxtx_vec.c | 5 +++++
> drivers/net/hns3/hns3_rxtx_vec_sve.c | 5 +++++
> 3 files changed, 11 insertions(+), 1 deletion(-)
Thanks, applied
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-12-03 12:31 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-12-03 6:46 [PATCH 22.11] net/hns3: fix crash for NEON and SVE Jie Hai
2024-12-03 12:31 ` Luca Boccassi
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).