* [dpdk-dev] [PATCH v2 1/3] net/i40e: fix out of order Rx read issue
2016-10-17 18:29 [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue Qi Zhang
@ 2016-10-17 18:29 ` Qi Zhang
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 2/3] net/ixgbe: " Qi Zhang
` (2 subsequent siblings)
3 siblings, 0 replies; 6+ messages in thread
From: Qi Zhang @ 2016-10-17 18:29 UTC (permalink / raw)
To: jingjing.wu, helin.zhang; +Cc: dev, Qi Zhang
In vPMD, when load Rx desc with _mm_loadu_si128,
volatile point will be cast into non-volatile point.
So GCC is allowed to reorder the load instructions,
while Rx read's correctness is reply on these load
instructions to follow a backward sequence strictly,
so we add compile barrier to prevent compiler reorder.
Fixes: 9ed94e5bb04e ("i40e: add vector Rx")
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
---
v2:
- fix check-git-log.sh warning.
- add more detail commit message.
drivers/net/i40e/i40e_rxtx_vec.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/net/i40e/i40e_rxtx_vec.c b/drivers/net/i40e/i40e_rxtx_vec.c
index 0ee0241..ab63501 100644
--- a/drivers/net/i40e/i40e_rxtx_vec.c
+++ b/drivers/net/i40e/i40e_rxtx_vec.c
@@ -305,6 +305,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
/* Read desc statuses backwards to avoid race condition */
/* A.1 load 4 pkts desc */
descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
+ rte_compiler_barrier();
/* B.2 copy 2 mbuf point into rx_pkts */
_mm_storeu_si128((__m128i *)&rx_pkts[pos], mbp1);
@@ -313,8 +314,10 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos+2]);
descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
+ rte_compiler_barrier();
/* B.1 load 2 mbuf point */
descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
+ rte_compiler_barrier();
descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
/* B.2 copy 2 mbuf point into rx_pkts */
--
2.7.4
^ permalink raw reply [flat|nested] 6+ messages in thread* [dpdk-dev] [PATCH v2 2/3] net/ixgbe: fix out of order Rx read issue
2016-10-17 18:29 [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue Qi Zhang
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 1/3] net/i40e: " Qi Zhang
@ 2016-10-17 18:29 ` Qi Zhang
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 3/3] net/fm10k: fix out of ofder " Qi Zhang
2016-10-18 11:33 ` [dpdk-dev] [PATCH v2 0/3] net: fix out of order " Ananyev, Konstantin
3 siblings, 0 replies; 6+ messages in thread
From: Qi Zhang @ 2016-10-17 18:29 UTC (permalink / raw)
To: jingjing.wu, helin.zhang; +Cc: dev, Qi Zhang
In vPMD, when load Rx desc with _mm_loadu_si128,
volatile point will be cast into non-volatile point.
So GCC is allowed to reorder the load instructions,
while Rx read's correctness is reply on these load
instructions to follow a backward sequence strictly,
so we add compile barrier to prevent compiler reorder.
Fixes: c95584dc2b18 ("ixgbe: new vectorized functions for Rx/Tx")
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
---
v2:
- fix check-git-log.sh warning.
- add more detail commit message.
drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
index ad8a9ff..abbf284 100644
--- a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
+++ b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
@@ -343,6 +343,7 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
/* Read desc statuses backwards to avoid race condition */
/* A.1 load 4 pkts desc */
descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
+ rte_compiler_barrier();
/* B.2 copy 2 mbuf point into rx_pkts */
_mm_storeu_si128((__m128i *)&rx_pkts[pos], mbp1);
@@ -351,8 +352,10 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos+2]);
descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
+ rte_compiler_barrier();
/* B.1 load 2 mbuf point */
descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
+ rte_compiler_barrier();
descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
/* B.2 copy 2 mbuf point into rx_pkts */
--
2.7.4
^ permalink raw reply [flat|nested] 6+ messages in thread* [dpdk-dev] [PATCH v2 3/3] net/fm10k: fix out of ofder Rx read issue
2016-10-17 18:29 [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue Qi Zhang
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 1/3] net/i40e: " Qi Zhang
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 2/3] net/ixgbe: " Qi Zhang
@ 2016-10-17 18:29 ` Qi Zhang
2016-10-18 11:33 ` [dpdk-dev] [PATCH v2 0/3] net: fix out of order " Ananyev, Konstantin
3 siblings, 0 replies; 6+ messages in thread
From: Qi Zhang @ 2016-10-17 18:29 UTC (permalink / raw)
To: jingjing.wu, helin.zhang; +Cc: dev, Qi Zhang
In vPMD, when load Rx desc with _mm_loadu_si128,
volatile point will be cast into non-volatile point.
So GCC is allowed to reorder the load instructions,
while Rx read's correctness is reply on these load
instructions to follow a backward sequence strictly,
so we add compile barrier to prevent compiler reorder.
Fixes: 7092be8437bd ("fm10k: add vector Rx")
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
---
v2:
- fix check-git-log.sh warning.
- add more detail commit message.
drivers/net/fm10k/fm10k_rxtx_vec.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/net/fm10k/fm10k_rxtx_vec.c b/drivers/net/fm10k/fm10k_rxtx_vec.c
index b724486..27f3e43 100644
--- a/drivers/net/fm10k/fm10k_rxtx_vec.c
+++ b/drivers/net/fm10k/fm10k_rxtx_vec.c
@@ -478,6 +478,7 @@ fm10k_recv_raw_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
/* Read desc statuses backwards to avoid race condition */
/* A.1 load 4 pkts desc */
descs0[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
+ rte_compiler_barrier();
/* B.2 copy 2 mbuf point into rx_pkts */
_mm_storeu_si128((__m128i *)&rx_pkts[pos], mbp1);
@@ -486,8 +487,10 @@ fm10k_recv_raw_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
mbp2 = _mm_loadu_si128((__m128i *)&mbufp[pos+2]);
descs0[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
+ rte_compiler_barrier();
/* B.1 load 2 mbuf point */
descs0[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
+ rte_compiler_barrier();
descs0[0] = _mm_loadu_si128((__m128i *)(rxdp));
/* B.2 copy 2 mbuf point into rx_pkts */
--
2.7.4
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue
2016-10-17 18:29 [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue Qi Zhang
` (2 preceding siblings ...)
2016-10-17 18:29 ` [dpdk-dev] [PATCH v2 3/3] net/fm10k: fix out of ofder " Qi Zhang
@ 2016-10-18 11:33 ` Ananyev, Konstantin
2016-10-19 9:52 ` Bruce Richardson
3 siblings, 1 reply; 6+ messages in thread
From: Ananyev, Konstantin @ 2016-10-18 11:33 UTC (permalink / raw)
To: Zhang, Qi Z, Wu, Jingjing, Zhang, Helin; +Cc: dev, Zhang, Qi Z
>
> In vPMD, when load Rx desc with _mm_loadu_si128,
> volatile point will be cast into non-volatile point.
> So GCC is allowed to reorder the load instructions,
> while Rx read's correctness is reply on these load
> instructions to follow a backward sequence strictly,
> so we add compile barrier to prevent compiler reorder.
> We already met this issue on i40e with GCC6 and we
> fixed this on ixgbe and fm10k also.
>
> v2:
> - fix check-git-log.sh warning.
> - add more detail commit message.
>
> Qi Zhang (3):
> net/i40e: fix out of order Rx read issue
> net/ixgbe: fix out of order Rx read issue
> net/fm10k: fix out of ofder Rx read issue
>
> drivers/net/fm10k/fm10k_rxtx_vec.c | 3 +++
> drivers/net/i40e/i40e_rxtx_vec.c | 3 +++
> drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c | 3 +++
> 3 files changed, 9 insertions(+)
>
> --
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> 2.7.4
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [dpdk-dev] [PATCH v2 0/3] net: fix out of order Rx read issue
2016-10-18 11:33 ` [dpdk-dev] [PATCH v2 0/3] net: fix out of order " Ananyev, Konstantin
@ 2016-10-19 9:52 ` Bruce Richardson
0 siblings, 0 replies; 6+ messages in thread
From: Bruce Richardson @ 2016-10-19 9:52 UTC (permalink / raw)
To: Ananyev, Konstantin; +Cc: Zhang, Qi Z, Wu, Jingjing, Zhang, Helin, dev
On Tue, Oct 18, 2016 at 11:33:43AM +0000, Ananyev, Konstantin wrote:
>
>
> >
> > In vPMD, when load Rx desc with _mm_loadu_si128,
> > volatile point will be cast into non-volatile point.
> > So GCC is allowed to reorder the load instructions,
> > while Rx read's correctness is reply on these load
> > instructions to follow a backward sequence strictly,
> > so we add compile barrier to prevent compiler reorder.
> > We already met this issue on i40e with GCC6 and we
> > fixed this on ixgbe and fm10k also.
> >
> > v2:
> > - fix check-git-log.sh warning.
> > - add more detail commit message.
> >
> > Qi Zhang (3):
> > net/i40e: fix out of order Rx read issue
> > net/ixgbe: fix out of order Rx read issue
> > net/fm10k: fix out of ofder Rx read issue
> >
> > drivers/net/fm10k/fm10k_rxtx_vec.c | 3 +++
> > drivers/net/i40e/i40e_rxtx_vec.c | 3 +++
> > drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c | 3 +++
> > 3 files changed, 9 insertions(+)
> >
> > --
>
> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
>
Applied to dpdk-next_net/rel_16_11
/Bruce
^ permalink raw reply [flat|nested] 6+ messages in thread