RE: [PATCH 1/3] net/mlx5/hws: support jump FDB Rx
> -Original Message- > From: Bing Zhao > Sent: Wednesday, February 26, 2025 6:29 PM > To: Suanming Mou ; Dariusz Sosnowski > ; Slava Ovsiienko ; Ori > Kam ; Matan Azrad > Cc: dev@dpdk.org; Raslan Darawsheh ; Alex Vesker > > Subject: RE: [PATCH 1/3] net/mlx5/hws: support jump FDB Rx > > Hi Suanming, > > PSB > > > -Original Message- > > From: Suanming Mou > > Sent: Tuesday, February 25, 2025 8:45 AM > > To: Dariusz Sosnowski ; Slava Ovsiienko > > ; Bing Zhao ; Ori Kam > > ; Matan Azrad > > Cc: dev@dpdk.org; Raslan Darawsheh ; Alex Vesker > > > > Subject: [PATCH 1/3] net/mlx5/hws: support jump FDB Rx > > > > Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx > > to Rx is not allowed. JUMP_FDB_RX feature introduces the internal > > loopback for Tx case and allow the REG C0 C1 B be preserved as well. > > > > This commit adds the JUMP_FDB_RX cap bit check and use JUMP_FDB_RX > > instead of FT for dest table FDB Rx case. > > > > Signed-off-by: Suanming Mou > > Signed-off-by: Alex Vesker > > --- > > drivers/common/mlx5/mlx5_prm.h| 9 - > > drivers/net/mlx5/hws/mlx5dr_action.c | 26 > +- > > drivers/net/mlx5/hws/mlx5dr_action.h | 4 > > drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 + > > drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++ > > drivers/net/mlx5/hws/mlx5dr_context.c | 17 + > > drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++ > > 7 files changed, 63 insertions(+), 6 deletions(-) > > [snip] > MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX) > > && > > + action->dest_tbl.type == MLX5DR_TABLE_TYPE_FDB_RX) > > + attr->action_type = > > MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX; > > + else > > + attr->action_type = > MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT; > > Fix the long line warnings, same as other places. > Also mlx5dr_context_cap_stc() seems to contain some calculation inside. If > the context and the CAP will not be changed during the runtime on a specific > port. Can this CAP BIT checking be done only once and saved for later usage? > Since the code is used in the flow insertion.WDYT? Ack for the line. But mlx5dr_context_cap_stc() is planned as get other caps based on the bit as well. And meanwhile that function is called by action creation, the jump action creation is done inside table creation, it is not in the flow insertion routine, so we are OK here. > > > + > > + break; [snip]
[PATCH] app/test-pmd: enable hw checksum offload for txonly
This patch fix txonly forward mode, as it was lacking of MBUF offload flags for IPv4 and UDP when the hardware support it. Signed-off-by: Nicolas PLANEL --- app/test-pmd/txonly.c | 4 1 file changed, 4 insertions(+) diff --git a/app/test-pmd/txonly.c b/app/test-pmd/txonly.c index 59d821a22d6f..bdcf6ea66008 100644 --- a/app/test-pmd/txonly.c +++ b/app/test-pmd/txonly.c @@ -334,6 +334,10 @@ pkt_burst_transmit(struct fwd_stream *fs) ol_flags |= RTE_MBUF_F_TX_QINQ; if (tx_offloads & RTE_ETH_TX_OFFLOAD_MACSEC_INSERT) ol_flags |= RTE_MBUF_F_TX_MACSEC; + if (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) + ol_flags |= RTE_MBUF_F_TX_IP_CKSUM; + if (tx_offloads & RTE_ETH_TX_OFFLOAD_UDP_CKSUM) + ol_flags |= RTE_MBUF_F_TX_UDP_CKSUM; /* * Initialize Ethernet header. -- 2.43.0
RE: [PATCH 3/3] net/mlx5: allow FDB RSS
Hi
> -Original Message-
> From: Suanming Mou
> Sent: Wednesday, February 26, 2025 9:43 PM
> To: Bing Zhao ; Dariusz Sosnowski
> ; Slava Ovsiienko ; Ori Kam
> ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: RE: [PATCH 3/3] net/mlx5: allow FDB RSS
>
>
>
> > -Original Message-
> > From: Bing Zhao
> > Sent: Wednesday, February 26, 2025 6:50 PM
> > To: Suanming Mou ; Dariusz Sosnowski
> > ; Slava Ovsiienko ; Ori
> > Kam ; Matan Azrad
> > Cc: dev@dpdk.org; Raslan Darawsheh
> > Subject: RE: [PATCH 3/3] net/mlx5: allow FDB RSS
> >
> >
> >
> > > -Original Message-
> > > From: Suanming Mou
> > > Sent: Tuesday, February 25, 2025 8:45 AM
> > > To: Dariusz Sosnowski ; Slava Ovsiienko
> > > ; Bing Zhao ; Ori Kam
> > > ; Matan Azrad
> > > Cc: dev@dpdk.org; Raslan Darawsheh
> > > Subject: [PATCH 3/3] net/mlx5: allow FDB RSS
> > >
> > > RSS can be used in FDB Rx rules when JUMP_FDB_RX action to allow
> > > jump from FDB Tx to FDB Rx table.
> > >
> > > Different with NIC RSS, FDB RSS will not do the internal implicit
> > > metadata copy.
> > >
> > > This commit enables the FDB RSS if JUMP_FDB_RX is supported.
> > >
> > > Signed-off-by: Suanming Mou
> > > ---
> > > drivers/net/mlx5/mlx5_flow.c| 4 ++--
> > > drivers/net/mlx5/mlx5_flow_dv.c | 32
> > > +--- drivers/net/mlx5/mlx5_flow_hw.c |
> > > 19 ++-
> > > 3 files changed, 33 insertions(+), 22 deletions(-)
> > >
> > > diff --git a/drivers/net/mlx5/mlx5_flow.c
> > > b/drivers/net/mlx5/mlx5_flow.c index 3fbe89a9d4..9c6a4f39fb 100644
> > > --- a/drivers/net/mlx5/mlx5_flow.c
> > > +++ b/drivers/net/mlx5/mlx5_flow.c
> > > @@ -1977,9 +1977,9 @@ mlx5_flow_validate_action_mark(struct
> > > rte_eth_dev *dev,
> > > RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> > > "mark action not supported for "
> > > "egress");
> > > - if (attr->transfer && mlx5_hws_active(dev))
> > > + if (attr->transfer && !mlx5_hws_active(dev))
> > > return rte_flow_error_set(error, ENOTSUP,
> > > - RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> > > + RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
> > > NULL,
> > > "non-template mark action not
> > > supported
> for
> > transfer");
> > > return 0;
> > > }
> > > diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
> > > b/drivers/net/mlx5/mlx5_flow_dv.c index 633c41e358..61d3101ce8
> > 100644
> > > --- a/drivers/net/mlx5/mlx5_flow_dv.c
> > > +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> > > @@ -8939,21 +8939,23 @@ flow_dv_validate(struct rte_eth_dev *dev,
> > > const struct rte_flow_attr *attr,
> > > RTE_FLOW_ERROR_TYPE_ACTION,
> > > NULL,
> > > "unsupported action MARK");
> > > - if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> > > - return rte_flow_error_set(error, ENOTSUP,
> > > - RTE_FLOW_ERROR_TYPE_ACTION,
> > > - NULL,
> > > - "unsupported action QUEUE");
> > > - if (action_flags & MLX5_FLOW_ACTION_RSS)
> > > - return rte_flow_error_set(error, ENOTSUP,
> > > - RTE_FLOW_ERROR_TYPE_ACTION,
> > > - NULL,
> > > - "unsupported action RSS");
> > > - if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> > > - return rte_flow_error_set(error, EINVAL,
> > > - RTE_FLOW_ERROR_TYPE_ACTION,
> > > - actions,
> > > - "no fate action is found");
> > > + if (!priv->jump_fdb_rx_en) {
> > > + if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> > > + return rte_flow_error_set(error, ENOTSUP,
> > > +
> > > RTE_FLOW_ERROR_TYPE_ACTION,
> > > + NULL,
> > > + "unsupported action
> > > QUEUE");
> > > + if (action_flags & MLX5_FLOW_ACTION_RSS)
> > > + return rte_flow_error_set(error, ENOTSUP,
> > > +
> > > RTE_FLOW_ERROR_TYPE_ACTION,
> > > + NULL,
> > > + "unsupported action
> > > RSS");
> > > + if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> > > + return rte_flow_error_set(error, EINVAL,
> > > +
RE: [PATCH 3/3] net/mlx5: allow FDB RSS
> -Original Message-
> From: Bing Zhao
> Sent: Wednesday, February 26, 2025 6:50 PM
> To: Suanming Mou ; Dariusz Sosnowski
> ; Slava Ovsiienko ; Ori
> Kam ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: RE: [PATCH 3/3] net/mlx5: allow FDB RSS
>
>
>
> > -Original Message-
> > From: Suanming Mou
> > Sent: Tuesday, February 25, 2025 8:45 AM
> > To: Dariusz Sosnowski ; Slava Ovsiienko
> > ; Bing Zhao ; Ori Kam
> > ; Matan Azrad
> > Cc: dev@dpdk.org; Raslan Darawsheh
> > Subject: [PATCH 3/3] net/mlx5: allow FDB RSS
> >
> > RSS can be used in FDB Rx rules when JUMP_FDB_RX action to allow jump
> > from FDB Tx to FDB Rx table.
> >
> > Different with NIC RSS, FDB RSS will not do the internal implicit
> > metadata copy.
> >
> > This commit enables the FDB RSS if JUMP_FDB_RX is supported.
> >
> > Signed-off-by: Suanming Mou
> > ---
> > drivers/net/mlx5/mlx5_flow.c| 4 ++--
> > drivers/net/mlx5/mlx5_flow_dv.c | 32 +---
> > drivers/net/mlx5/mlx5_flow_hw.c | 19 ++-
> > 3 files changed, 33 insertions(+), 22 deletions(-)
> >
> > diff --git a/drivers/net/mlx5/mlx5_flow.c
> > b/drivers/net/mlx5/mlx5_flow.c index 3fbe89a9d4..9c6a4f39fb 100644
> > --- a/drivers/net/mlx5/mlx5_flow.c
> > +++ b/drivers/net/mlx5/mlx5_flow.c
> > @@ -1977,9 +1977,9 @@ mlx5_flow_validate_action_mark(struct
> > rte_eth_dev *dev,
> > RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> > "mark action not supported for "
> > "egress");
> > - if (attr->transfer && mlx5_hws_active(dev))
> > + if (attr->transfer && !mlx5_hws_active(dev))
> > return rte_flow_error_set(error, ENOTSUP,
> > - RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> > + RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
> > NULL,
> > "non-template mark action not
> > supported for
> transfer");
> > return 0;
> > }
> > diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
> > b/drivers/net/mlx5/mlx5_flow_dv.c index 633c41e358..61d3101ce8
> 100644
> > --- a/drivers/net/mlx5/mlx5_flow_dv.c
> > +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> > @@ -8939,21 +8939,23 @@ flow_dv_validate(struct rte_eth_dev *dev,
> > const struct rte_flow_attr *attr,
> > RTE_FLOW_ERROR_TYPE_ACTION,
> > NULL,
> > "unsupported action MARK");
> > - if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> > - return rte_flow_error_set(error, ENOTSUP,
> > - RTE_FLOW_ERROR_TYPE_ACTION,
> > - NULL,
> > - "unsupported action QUEUE");
> > - if (action_flags & MLX5_FLOW_ACTION_RSS)
> > - return rte_flow_error_set(error, ENOTSUP,
> > - RTE_FLOW_ERROR_TYPE_ACTION,
> > - NULL,
> > - "unsupported action RSS");
> > - if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> > - return rte_flow_error_set(error, EINVAL,
> > - RTE_FLOW_ERROR_TYPE_ACTION,
> > - actions,
> > - "no fate action is found");
> > + if (!priv->jump_fdb_rx_en) {
> > + if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> > + return rte_flow_error_set(error, ENOTSUP,
> > +
> > RTE_FLOW_ERROR_TYPE_ACTION,
> > + NULL,
> > + "unsupported action
> > QUEUE");
> > + if (action_flags & MLX5_FLOW_ACTION_RSS)
> > + return rte_flow_error_set(error, ENOTSUP,
> > +
> > RTE_FLOW_ERROR_TYPE_ACTION,
> > + NULL,
> > + "unsupported action
> > RSS");
> > + if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> > + return rte_flow_error_set(error, EINVAL,
> > +
> > RTE_FLOW_ERROR_TYPE_ACTION,
> > + actions,
> > + "no fate action is
> > found");
> > + }
> > } else {
> > if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
> > return rte_flow_error_
RE: [PATCH 2/3] net/mlx5: add jump FDB Rx flag
> -Original Message-
> From: Bing Zhao
> Sent: Wednesday, February 26, 2025 6:38 PM
> To: Suanming Mou ; Dariusz Sosnowski
> ; Slava Ovsiienko ; Ori
> Kam ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: RE: [PATCH 2/3] net/mlx5: add jump FDB Rx flag
>
> Hi
>
> > -Original Message-
> > From: Suanming Mou
> > Sent: Tuesday, February 25, 2025 8:45 AM
> > To: Dariusz Sosnowski ; Slava Ovsiienko
> > ; Bing Zhao ; Ori Kam
> > ; Matan Azrad
> > Cc: dev@dpdk.org; Raslan Darawsheh
> > Subject: [PATCH 2/3] net/mlx5: add jump FDB Rx flag
> >
> > When jump FDB Rx is supported, flow will be able to jump from FDB Tx
> > to FDB Rx, in that case the dest action in FDB Rx table should support
> > FDB Tx as well.
> >
> > Signed-off-by: Suanming Mou
> > ---
> > drivers/common/mlx5/mlx5_devx_cmds.c | 8
> > drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
> > drivers/net/mlx5/linux/mlx5_os.c | 1 +
> > drivers/net/mlx5/mlx5.h | 1 +
> > drivers/net/mlx5/mlx5_flow_hw.c | 8 +++-
> > 5 files changed, 18 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
> > b/drivers/common/mlx5/mlx5_devx_cmds.c
> > index f504b29f31..eb8553e8ad 100644
> > --- a/drivers/common/mlx5/mlx5_devx_cmds.c
> > +++ b/drivers/common/mlx5/mlx5_devx_cmds.c
> > @@ -924,6 +924,7 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
> > uint32_t out[MLX5_ST_SZ_DW(query_hca_cap_out)] = {0};
> > bool hca_cap_2_sup;
> > uint64_t general_obj_types_supported = 0;
> > + uint64_t stc_action_type_127_64;
> > void *hcattr;
> > int rc, i;
> >
> > @@ -1352,6 +1353,13 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
> > attr->fdb_unified_en = MLX5_GET(wqe_based_flow_table_cap,
> > hcattr,
> > fdb_unified_en);
> > + stc_action_type_127_64 =
> MLX5_GET64(wqe_based_flow_table_cap,
> > + hcattr,
> > + stc_action_type_127_64);
> > + if (stc_action_type_127_64 &
> > + (1 <<
> > (MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX -
> > + MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX)))
> > + attr->jump_fdb_rx_en = true;
>
> I assume integer value 1 is enough, since it is a bit field, no need to use
> bool
> type.
Will change it to 1.
>
> > }
> > /* Query HCA attribute for ROCE. */
> > if (attr->roce) {
> > diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h
> > b/drivers/common/mlx5/mlx5_devx_cmds.h
> > index 8de4210fb2..6c726a0d46 100644
> > --- a/drivers/common/mlx5/mlx5_devx_cmds.h
> > +++ b/drivers/common/mlx5/mlx5_devx_cmds.h
> > @@ -326,6 +326,7 @@ struct mlx5_hca_attr {
> > uint32_t lag_rx_port_affinity:1;
> > uint32_t wqe_based_flow_table_sup:1;
> > uint32_t fdb_unified_en:1;
> > + uint32_t jump_fdb_rx_en:1;
> > uint8_t max_header_modify_pattern_length;
> > uint64_t system_image_guid;
> > uint32_t log_max_conn_track_offload:5; diff --git
> > a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
> > index 9410211e3b..4e64026137 100644
> > --- a/drivers/net/mlx5/linux/mlx5_os.c
> > +++ b/drivers/net/mlx5/linux/mlx5_os.c
> > @@ -1718,6 +1718,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> > priv->unified_fdb_en = !!priv->master;
> > DRV_LOG(DEBUG, "port %u: unified FDB %s enabled.",
> > eth_dev->data->port_id, priv->unified_fdb_en ? "is" :
> > "isn't");
> > + priv->jump_fdb_rx_en = sh->cdev-
> > >config.hca_attr.jump_fdb_rx_en;
>
> Can we add some INFO or DEBUG level log to indicate this to the user?
Sure, and I think here we have another improvement as if unified_fdb is not
enabled, jump_fdb_rx_en does not make any sense. Will update.
>
> > if (priv->sh->config.dv_esw_en) {
> > uint32_t usable_bits;
> > uint32_t required_bits;
> > diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h index
> > f73f6e63ff..545ba48b3c 100644
> > --- a/drivers/net/mlx5/mlx5.h
> > +++ b/drivers/net/mlx5/mlx5.h
> > @@ -1987,6 +1987,7 @@ struct mlx5_priv {
> > uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
> > uint32_t tunnel_enabled:1; /* If tunnel offloading is enabled on
> > rxqs. */
> > uint32_t unified_fdb_en:1; /* Unified FDB flag per port. */
> > + uint32_t jump_fdb_rx_en:1; /* Jump from FDB Tx to FDB Rx flag per
> > +port. */
> > uint16_t domain_id; /* Switch domain identifier. */
> > uint16_t vport_id; /* Associated VF vport index (if any). */
> > uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG.
> > */ diff --git a/drivers/net/mlx5/mlx5_flow_hw.c
> > b/drivers/net/mlx5/mlx5_flow_hw.c index f0888dbf0e..83f55ed3e8 100644
> > --- a/drivers/net/mlx5/mlx5_flow_hw.c
> > +++ b/drivers/net/mlx
[PATCH] mempool: micro optimizations
The comparisons lcore_id < RTE_MAX_LCORE and lcore_id != LCORE_ID_ANY are
equivalent, but the latter compiles to fewer bytes of code space.
Similarly for lcore_id >= RTE_MAX_LCORE and lcore_id == LCORE_ID_ANY.
The rte_mempool_get_ops() function is also used in the fast path, so
RTE_VERIFY() was replaced by RTE_ASSERT().
Compilers implicitly consider comparisons of variable == 0 likely, so
unlikely() was added to the check for no mempool cache (mp->cache_size ==
0) in the rte_mempool_default_cache() function.
The rte_mempool_do_generic_put() function for adding objects to a mempool
was refactored as follows:
- The comparison for the request itself being too big, which is considered
unlikely, was moved down and out of the code path where the cache has
sufficient room for the added objects, which is considered the most
likely code path.
- Added __rte_assume() about the cache length, size and threshold, for
compiler optimization when "n" is compile time constant.
- Added __rte_assume() about "ret" being zero, so other functions using
the value returned by this function can be potentially optimized by the
compiler; especially when it merges multiple sequential code paths of
inlined code depending on the return value being either zero or
negative.
- The refactored source code (with comments) made the separate comment
describing the cache flush/add algorithm superfluous, so it was removed.
A few more likely()/unlikely() were added.
A few comments were improved for readability.
Some assertions, RTE_ASSERT(), were added. Most importantly to assert that
the return values of the mempool drivers' enqueue and dequeue operations
are API compliant, i.e. 0 (for success) or negative (for failure), and
never positive.
Signed-off-by: Morten Brørup
---
lib/mempool/rte_mempool.h | 67 ++-
1 file changed, 38 insertions(+), 29 deletions(-)
diff --git a/lib/mempool/rte_mempool.h b/lib/mempool/rte_mempool.h
index c495cc012f..aedc100964 100644
--- a/lib/mempool/rte_mempool.h
+++ b/lib/mempool/rte_mempool.h
@@ -334,7 +334,7 @@ struct __rte_cache_aligned rte_mempool {
#ifdef RTE_LIBRTE_MEMPOOL_STATS
#define RTE_MEMPOOL_STAT_ADD(mp, name, n) do {
\
unsigned int __lcore_id = rte_lcore_id();
\
- if (likely(__lcore_id < RTE_MAX_LCORE))
\
+ if (likely(__lcore_id != LCORE_ID_ANY))
\
(mp)->stats[__lcore_id].name += (n);
\
else
\
rte_atomic_fetch_add_explicit(&((mp)->stats[RTE_MAX_LCORE].name), \
@@ -751,7 +751,7 @@ extern struct rte_mempool_ops_table rte_mempool_ops_table;
static inline struct rte_mempool_ops *
rte_mempool_get_ops(int ops_index)
{
- RTE_VERIFY((ops_index >= 0) && (ops_index < RTE_MEMPOOL_MAX_OPS_IDX));
+ RTE_ASSERT((ops_index >= 0) && (ops_index < RTE_MEMPOOL_MAX_OPS_IDX));
return &rte_mempool_ops_table.ops[ops_index];
}
@@ -791,7 +791,8 @@ rte_mempool_ops_dequeue_bulk(struct rte_mempool *mp,
rte_mempool_trace_ops_dequeue_bulk(mp, obj_table, n);
ops = rte_mempool_get_ops(mp->ops_index);
ret = ops->dequeue(mp, obj_table, n);
- if (ret == 0) {
+ RTE_ASSERT(ret <= 0);
+ if (likely(ret == 0)) {
RTE_MEMPOOL_STAT_ADD(mp, get_common_pool_bulk, 1);
RTE_MEMPOOL_STAT_ADD(mp, get_common_pool_objs, n);
}
@@ -816,11 +817,14 @@ rte_mempool_ops_dequeue_contig_blocks(struct rte_mempool
*mp,
void **first_obj_table, unsigned int n)
{
struct rte_mempool_ops *ops;
+ int ret;
ops = rte_mempool_get_ops(mp->ops_index);
RTE_ASSERT(ops->dequeue_contig_blocks != NULL);
rte_mempool_trace_ops_dequeue_contig_blocks(mp, first_obj_table, n);
- return ops->dequeue_contig_blocks(mp, first_obj_table, n);
+ ret = ops->dequeue_contig_blocks(mp, first_obj_table, n);
+ RTE_ASSERT(ret <= 0);
+ return ret;
}
/**
@@ -848,6 +852,7 @@ rte_mempool_ops_enqueue_bulk(struct rte_mempool *mp, void *
const *obj_table,
rte_mempool_trace_ops_enqueue_bulk(mp, obj_table, n);
ops = rte_mempool_get_ops(mp->ops_index);
ret = ops->enqueue(mp, obj_table, n);
+ RTE_ASSERT(ret <= 0);
#ifdef RTE_LIBRTE_MEMPOOL_DEBUG
if (unlikely(ret < 0))
RTE_MEMPOOL_LOG(CRIT, "cannot enqueue %u objects to mempool %s",
@@ -1333,10 +1338,10 @@ rte_mempool_cache_free(struct rte_mempool_cache *cache);
static __rte_always_inline struct rte_mempool_cache *
rte_mempool_default_cache(struct rte_mempool *mp, unsigned lcore_id)
{
- if (mp->cache_size == 0)
+ if (unlikely(mp->cache_size == 0))
return NULL;
- if (lcore_id >= RTE_MAX_LCORE)
+ if (unlikely(lcore_id ==
Re: [PATCH v3 0/3] net/mlx5: support jump FDB Rx
Hi, From: Dariusz Sosnowski Sent: Wednesday, February 26, 2025 6:04 PM To: Dariusz Sosnowski; Slava Ovsiienko; Bing Zhao; Ori Kam; Suanming Mou; Matan Azrad Cc: dev@dpdk.org; Raslan Darawsheh Subject: [PATCH v3 0/3] net/mlx5: support jump FDB Rx Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback for Tx case and allow the REG C0 C1 B be preserved as well. This series adds the jump from FDB Tx to FDB Rx support and allows RSS action in FDB domain rule when jump FDB Rx is supported. v3: - Rebased on v2 of mlx5 Unified FDB patches (series 34723) Suanming Mou (3): net/mlx5/hws: support jump FDB Rx net/mlx5: add jump FDB Rx flag net/mlx5: allow FDB RSS doc/guides/nics/mlx5.rst | 4 drivers/common/mlx5/mlx5_devx_cmds.c | 8 +++ drivers/common/mlx5/mlx5_devx_cmds.h | 1 + drivers/common/mlx5/mlx5_prm.h| 9 +++- drivers/net/mlx5/hws/mlx5dr_action.c | 26 +- drivers/net/mlx5/hws/mlx5dr_action.h | 4 drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++ drivers/net/mlx5/hws/mlx5dr_context.c | 17 ++ drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++ drivers/net/mlx5/linux/mlx5_os.c | 9 ++-- drivers/net/mlx5/mlx5.h | 1 + drivers/net/mlx5/mlx5_flow.c | 4 ++-- drivers/net/mlx5/mlx5_flow_dv.c | 32 ++- drivers/net/mlx5/mlx5_flow_hw.c | 28 ++- 15 files changed, 125 insertions(+), 31 deletions(-) -- 2.39.5 Series applied to next-net-mlx, Kindest regards, Raslan Darawsheh
Re: [PATCH] mempool: micro optimizations
On Wed, Feb 26, 2025 at 03:59:22PM +, Morten Brørup wrote: > The comparisons lcore_id < RTE_MAX_LCORE and lcore_id != LCORE_ID_ANY are > equivalent, but the latter compiles to fewer bytes of code space. > Similarly for lcore_id >= RTE_MAX_LCORE and lcore_id == LCORE_ID_ANY. > > The rte_mempool_get_ops() function is also used in the fast path, so > RTE_VERIFY() was replaced by RTE_ASSERT(). > > Compilers implicitly consider comparisons of variable == 0 likely, so > unlikely() was added to the check for no mempool cache (mp->cache_size == > 0) in the rte_mempool_default_cache() function. > > The rte_mempool_do_generic_put() function for adding objects to a mempool > was refactored as follows: > - The comparison for the request itself being too big, which is considered > unlikely, was moved down and out of the code path where the cache has > sufficient room for the added objects, which is considered the most > likely code path. > - Added __rte_assume() about the cache length, size and threshold, for > compiler optimization when "n" is compile time constant. > - Added __rte_assume() about "ret" being zero, so other functions using > the value returned by this function can be potentially optimized by the > compiler; especially when it merges multiple sequential code paths of > inlined code depending on the return value being either zero or > negative. > - The refactored source code (with comments) made the separate comment > describing the cache flush/add algorithm superfluous, so it was removed. > > A few more likely()/unlikely() were added. > > A few comments were improved for readability. > > Some assertions, RTE_ASSERT(), were added. Most importantly to assert that > the return values of the mempool drivers' enqueue and dequeue operations > are API compliant, i.e. 0 (for success) or negative (for failure), and > never positive. > > Signed-off-by: Morten Brørup > --- > lib/mempool/rte_mempool.h | 67 ++- > 1 file changed, 38 insertions(+), 29 deletions(-) > Is there any measurable performance change with these modifications? /Bruce
RE: [EXTERNAL] Re: [v5 1/5] vhost: skip crypto op fetch before vring init
Hi Maxime,
> >
> > + if (unlikely(vq == NULL)) {
> > + VC_LOG_ERR("Invalid virtqueue %u", qid);
> > + return 0;
> > + }
> > +
> > + if (unlikely(vq->avail == NULL)) {
> > + VC_LOG_DBG("Virtqueue ring not yet initialized %u", qid);
> > + return 0;
> > + }
> > +
>
> I think the proper fix to:
> 1. acquire the access_lock as reader
> 2. check whether vq is enabled
> 3. acquire the iotlb lock as a reader
> 4. check vq's access_ok
>
Ack. Except for vq->enabled, next version of this patch would carry these
checks.
Current virtio crypto for vhost_user does not request SET_VRING_ENABLE, so this
flag would not be set.
Thanks,
Gowrishankar
> You can have a look at rte_vhost_dequeue_burst for example.
>
> > avail_idx = *((volatile uint16_t *)&vq->avail->idx);
> > start_idx = vq->last_used_idx;
> > count = avail_idx - start_idx;
[v6 4/5] vhost: support asymmetric RSA crypto ops
Support asymmetric RSA crypto operations in vhost-user.
Signed-off-by: Gowrishankar Muthukrishnan
Acked-by: Akhil Goyal
---
doc/guides/rel_notes/release_25_03.rst | 3 +
lib/vhost/vhost_crypto.c | 486 +++--
lib/vhost/virtio_crypto.h | 67
3 files changed, 521 insertions(+), 35 deletions(-)
diff --git a/doc/guides/rel_notes/release_25_03.rst
b/doc/guides/rel_notes/release_25_03.rst
index 8867a4bd74..087a407337 100644
--- a/doc/guides/rel_notes/release_25_03.rst
+++ b/doc/guides/rel_notes/release_25_03.rst
@@ -151,6 +151,9 @@ New Features
See the :doc:`../compressdevs/zsda` guide for more details on the new driver.
+* **Updated vhost library.**
+
+ Updated vhost library to support RSA crypto operations.
Removed Items
-
diff --git a/lib/vhost/vhost_crypto.c b/lib/vhost/vhost_crypto.c
index c014c85797..c55db0108b 100644
--- a/lib/vhost/vhost_crypto.c
+++ b/lib/vhost/vhost_crypto.c
@@ -55,6 +55,14 @@ RTE_LOG_REGISTER_SUFFIX(vhost_crypto_logtype, crypto, INFO);
*/
#define vhost_crypto_desc vring_desc
+struct vhost_crypto_session {
+ union {
+ struct rte_cryptodev_asym_session *asym;
+ struct rte_cryptodev_sym_session *sym;
+ };
+ enum rte_crypto_op_type type;
+};
+
static int
cipher_algo_transform(uint32_t virtio_cipher_algo,
enum rte_crypto_cipher_algorithm *algo)
@@ -207,8 +215,10 @@ struct __rte_cache_aligned vhost_crypto {
uint64_t last_session_id;
- uint64_t cache_session_id;
- struct rte_cryptodev_sym_session *cache_session;
+ uint64_t cache_sym_session_id;
+ struct rte_cryptodev_sym_session *cache_sym_session;
+ uint64_t cache_asym_session_id;
+ struct rte_cryptodev_asym_session *cache_asym_session;
/** socket id for the device */
int socket_id;
@@ -335,10 +345,11 @@ transform_chain_param(struct rte_crypto_sym_xform *xforms,
}
static void
-vhost_crypto_create_sess(struct vhost_crypto *vcrypto,
+vhost_crypto_create_sym_sess(struct vhost_crypto *vcrypto,
VhostUserCryptoSessionParam *sess_param)
{
struct rte_crypto_sym_xform xform1 = {0}, xform2 = {0};
+ struct vhost_crypto_session *vhost_session;
struct rte_cryptodev_sym_session *session;
int ret;
@@ -385,42 +396,277 @@ vhost_crypto_create_sess(struct vhost_crypto *vcrypto,
return;
}
- /* insert hash to map */
- if (rte_hash_add_key_data(vcrypto->session_map,
- &vcrypto->last_session_id, session) < 0) {
+ vhost_session = rte_zmalloc(NULL, sizeof(*vhost_session), 0);
+ if (vhost_session == NULL) {
+ VC_LOG_ERR("Failed to alloc session memory");
+ goto error_exit;
+ }
+
+ vhost_session->type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ vhost_session->sym = session;
+
+ /* insert session to map */
+ if ((rte_hash_add_key_data(vcrypto->session_map,
+ &vcrypto->last_session_id, vhost_session) < 0)) {
VC_LOG_ERR("Failed to insert session to hash table");
+ goto error_exit;
+ }
+
+ VC_LOG_INFO("Session %"PRIu64" created for vdev %i.",
+ vcrypto->last_session_id, vcrypto->dev->vid);
+
+ sess_param->session_id = vcrypto->last_session_id;
+ vcrypto->last_session_id++;
+ return;
+
+error_exit:
+ if (rte_cryptodev_sym_session_free(vcrypto->cid, session) < 0)
+ VC_LOG_ERR("Failed to free session");
+
+ sess_param->session_id = -VIRTIO_CRYPTO_ERR;
+ rte_free(vhost_session);
+}
+
+static int
+tlv_decode(uint8_t *tlv, uint8_t type, uint8_t **data, size_t *data_len)
+{
+ size_t tlen = -EINVAL, len;
+
+ if (tlv[0] != type)
+ return -EINVAL;
- if (rte_cryptodev_sym_session_free(vcrypto->cid, session) < 0)
- VC_LOG_ERR("Failed to free session");
+ if (tlv[1] == 0x82) {
+ len = (tlv[2] << 8) | tlv[3];
+ *data = &tlv[4];
+ tlen = len + 4;
+ } else if (tlv[1] == 0x81) {
+ len = tlv[2];
+ *data = &tlv[3];
+ tlen = len + 3;
+ } else {
+ len = tlv[1];
+ *data = &tlv[2];
+ tlen = len + 2;
+ }
+
+ *data_len = len;
+ return tlen;
+}
+
+static int
+virtio_crypto_asym_rsa_der_to_xform(uint8_t *der, size_t der_len,
+ struct rte_crypto_asym_xform *xform)
+{
+ uint8_t *n = NULL, *e = NULL, *d = NULL, *p = NULL, *q = NULL, *dp =
NULL,
+ *dq = NULL, *qinv = NULL, *v = NULL, *tlv;
+ size_t nlen, elen, dlen, plen, qlen, dplen, dqlen, qinvlen, vlen;
+ int len;
+
+ RTE_SET_USED(der_len);
+
+ if (der[0] != 0x30)
+ return -EINVAL;
+
+ if (der[1] == 0x82)
+ tlv = &der[4];
+
[v6 5/5] examples/vhost_crypto: support asymmetric crypto
Support asymmetric crypto operations.
Signed-off-by: Gowrishankar Muthukrishnan
Acked-by: Akhil Goyal
---
doc/guides/sample_app_ug/vhost_crypto.rst | 5 +++
examples/vhost_crypto/main.c | 50 +--
2 files changed, 43 insertions(+), 12 deletions(-)
diff --git a/doc/guides/sample_app_ug/vhost_crypto.rst
b/doc/guides/sample_app_ug/vhost_crypto.rst
index 7ae7addac4..b00f2bf3ae 100644
--- a/doc/guides/sample_app_ug/vhost_crypto.rst
+++ b/doc/guides/sample_app_ug/vhost_crypto.rst
@@ -33,6 +33,7 @@ Start the vhost_crypto example
--socket-file lcore,PATH
[--zero-copy]
[--guest-polling]
+ [--asymmetric-crypto]
where,
@@ -54,6 +55,10 @@ where,
guest works in polling mode, thus will NOT notify the guest completion of
processing.
+* asymmetric-crypto: the presence of this item means the application
+ can handle the asymmetric crypto requests. When this option is used,
+ symmetric crypto requests can not be handled by the application.
+
The application requires that crypto devices capable of performing
the specified crypto operation are available on application initialization.
This means that HW crypto device/s must be bound to a DPDK driver or
diff --git a/examples/vhost_crypto/main.c b/examples/vhost_crypto/main.c
index b1fe4120b9..8bdfc40c4b 100644
--- a/examples/vhost_crypto/main.c
+++ b/examples/vhost_crypto/main.c
@@ -59,6 +59,7 @@ struct vhost_crypto_options {
uint32_t nb_los;
uint32_t zero_copy;
uint32_t guest_polling;
+ bool asymmetric_crypto;
} options;
enum {
@@ -70,6 +71,8 @@ enum {
OPT_ZERO_COPY_NUM,
#define OPT_POLLING "guest-polling"
OPT_POLLING_NUM,
+#define OPT_ASYM"asymmetric-crypto"
+ OPT_ASYM_NUM,
};
#define NB_SOCKET_FIELDS (2)
@@ -202,9 +205,10 @@ vhost_crypto_usage(const char *prgname)
" --%s ,SOCKET-FILE-PATH\n"
" --%s (lcore,cdev_id,queue_id)[,(lcore,cdev_id,queue_id)]\n"
" --%s: zero copy\n"
- " --%s: guest polling\n",
+ " --%s: guest polling\n"
+ " --%s: asymmetric crypto\n",
prgname, OPT_SOCKET_FILE, OPT_CONFIG,
- OPT_ZERO_COPY, OPT_POLLING);
+ OPT_ZERO_COPY, OPT_POLLING, OPT_ASYM);
}
static int
@@ -223,6 +227,8 @@ vhost_crypto_parse_args(int argc, char **argv)
NULL, OPT_ZERO_COPY_NUM},
{OPT_POLLING, no_argument,
NULL, OPT_POLLING_NUM},
+ {OPT_ASYM, no_argument,
+ NULL, OPT_ASYM_NUM},
{NULL, 0, 0, 0}
};
@@ -262,6 +268,10 @@ vhost_crypto_parse_args(int argc, char **argv)
options.guest_polling = 1;
break;
+ case OPT_ASYM_NUM:
+ options.asymmetric_crypto = true;
+ break;
+
default:
vhost_crypto_usage(prgname);
return -EINVAL;
@@ -376,6 +386,7 @@ vhost_crypto_worker(void *arg)
int callfds[VIRTIO_CRYPTO_MAX_NUM_BURST_VQS];
uint32_t lcore_id = rte_lcore_id();
uint32_t burst_size = MAX_PKT_BURST;
+ enum rte_crypto_op_type cop_type;
uint32_t i, j, k;
uint32_t to_fetch, fetched;
@@ -383,9 +394,13 @@ vhost_crypto_worker(void *arg)
RTE_LOG(INFO, USER1, "Processing on Core %u started\n", lcore_id);
+ cop_type = RTE_CRYPTO_OP_TYPE_SYMMETRIC;
+ if (options.asymmetric_crypto)
+ cop_type = RTE_CRYPTO_OP_TYPE_ASYMMETRIC;
+
for (i = 0; i < NB_VIRTIO_QUEUES; i++) {
if (rte_crypto_op_bulk_alloc(info->cop_pool,
- RTE_CRYPTO_OP_TYPE_SYMMETRIC, ops[i],
+ cop_type, ops[i],
burst_size) < burst_size) {
RTE_LOG(ERR, USER1, "Failed to alloc cops\n");
ret = -1;
@@ -411,12 +426,11 @@ vhost_crypto_worker(void *arg)
fetched);
if (unlikely(rte_crypto_op_bulk_alloc(
info->cop_pool,
- RTE_CRYPTO_OP_TYPE_SYMMETRIC,
+ cop_type,
ops[j], fetched) < fetched)) {
RTE_LOG(ERR, USER1, "Failed realloc\n");
return -1;
}
-
fetched = rte_cryptodev_dequeue_burst(
info->cid, info->qid,
ops_deq[j], RTE_MIN(burst_size,
@@ -477,6 +491,7 @@ m
[v5 4/6] crypto/virtio: add vDPA backend
Add vDPA backend to virtio_user crypto.
Signed-off-by: Gowrishankar Muthukrishnan
---
drivers/crypto/virtio/meson.build | 7 +
drivers/crypto/virtio/virtio_cryptodev.c | 57 +-
drivers/crypto/virtio/virtio_cryptodev.h | 3 +
drivers/crypto/virtio/virtio_logs.h | 6 +-
drivers/crypto/virtio/virtio_pci.h| 7 +
drivers/crypto/virtio/virtio_ring.h | 6 -
drivers/crypto/virtio/virtio_user/vhost.h | 90 +++
.../crypto/virtio/virtio_user/vhost_vdpa.c| 710 +
.../virtio/virtio_user/virtio_user_dev.c | 749 ++
.../virtio/virtio_user/virtio_user_dev.h | 85 ++
drivers/crypto/virtio/virtio_user_cryptodev.c | 575 ++
11 files changed, 2265 insertions(+), 30 deletions(-)
create mode 100644 drivers/crypto/virtio/virtio_user/vhost.h
create mode 100644 drivers/crypto/virtio/virtio_user/vhost_vdpa.c
create mode 100644 drivers/crypto/virtio/virtio_user/virtio_user_dev.c
create mode 100644 drivers/crypto/virtio/virtio_user/virtio_user_dev.h
create mode 100644 drivers/crypto/virtio/virtio_user_cryptodev.c
diff --git a/drivers/crypto/virtio/meson.build
b/drivers/crypto/virtio/meson.build
index d2c3b3ad07..3763e86746 100644
--- a/drivers/crypto/virtio/meson.build
+++ b/drivers/crypto/virtio/meson.build
@@ -16,3 +16,10 @@ sources = files(
'virtio_rxtx.c',
'virtqueue.c',
)
+
+if is_linux
+sources += files('virtio_user_cryptodev.c',
+'virtio_user/vhost_vdpa.c',
+'virtio_user/virtio_user_dev.c')
+deps += ['bus_vdev']
+endif
diff --git a/drivers/crypto/virtio/virtio_cryptodev.c
b/drivers/crypto/virtio/virtio_cryptodev.c
index 92fea557ab..bc737f1e68 100644
--- a/drivers/crypto/virtio/virtio_cryptodev.c
+++ b/drivers/crypto/virtio/virtio_cryptodev.c
@@ -544,24 +544,12 @@ virtio_crypto_init_device(struct rte_cryptodev *cryptodev,
return 0;
}
-/*
- * This function is based on probe() function
- * It returns 0 on success.
- */
-static int
-crypto_virtio_create(const char *name, struct rte_pci_device *pci_dev,
- struct rte_cryptodev_pmd_init_params *init_params)
+int
+crypto_virtio_dev_init(struct rte_cryptodev *cryptodev, uint64_t features,
+ struct rte_pci_device *pci_dev)
{
- struct rte_cryptodev *cryptodev;
struct virtio_crypto_hw *hw;
- PMD_INIT_FUNC_TRACE();
-
- cryptodev = rte_cryptodev_pmd_create(name, &pci_dev->device,
- init_params);
- if (cryptodev == NULL)
- return -ENODEV;
-
cryptodev->driver_id = cryptodev_virtio_driver_id;
cryptodev->dev_ops = &virtio_crypto_dev_ops;
@@ -578,16 +566,41 @@ crypto_virtio_create(const char *name, struct
rte_pci_device *pci_dev,
hw->dev_id = cryptodev->data->dev_id;
hw->virtio_dev_capabilities = virtio_capabilities;
- VIRTIO_CRYPTO_INIT_LOG_DBG("dev %d vendorID=0x%x deviceID=0x%x",
- cryptodev->data->dev_id, pci_dev->id.vendor_id,
- pci_dev->id.device_id);
+ if (pci_dev) {
+ /* pci device init */
+ VIRTIO_CRYPTO_INIT_LOG_DBG("dev %d vendorID=0x%x deviceID=0x%x",
+ cryptodev->data->dev_id, pci_dev->id.vendor_id,
+ pci_dev->id.device_id);
- /* pci device init */
- if (vtpci_cryptodev_init(pci_dev, hw))
+ if (vtpci_cryptodev_init(pci_dev, hw))
+ return -1;
+ }
+
+ if (virtio_crypto_init_device(cryptodev, features) < 0)
return -1;
- if (virtio_crypto_init_device(cryptodev,
- VIRTIO_CRYPTO_PMD_GUEST_FEATURES) < 0)
+ return 0;
+}
+
+/*
+ * This function is based on probe() function
+ * It returns 0 on success.
+ */
+static int
+crypto_virtio_create(const char *name, struct rte_pci_device *pci_dev,
+ struct rte_cryptodev_pmd_init_params *init_params)
+{
+ struct rte_cryptodev *cryptodev;
+
+ PMD_INIT_FUNC_TRACE();
+
+ cryptodev = rte_cryptodev_pmd_create(name, &pci_dev->device,
+ init_params);
+ if (cryptodev == NULL)
+ return -ENODEV;
+
+ if (crypto_virtio_dev_init(cryptodev, VIRTIO_CRYPTO_PMD_GUEST_FEATURES,
+ pci_dev) < 0)
return -1;
rte_cryptodev_pmd_probing_finish(cryptodev);
diff --git a/drivers/crypto/virtio/virtio_cryptodev.h
b/drivers/crypto/virtio/virtio_cryptodev.h
index f8498246e2..fad73d54a8 100644
--- a/drivers/crypto/virtio/virtio_cryptodev.h
+++ b/drivers/crypto/virtio/virtio_cryptodev.h
@@ -76,4 +76,7 @@ uint16_t virtio_crypto_pkt_rx_burst(void *tx_queue,
struct rte_crypto_op **tx_pkts,
uint16_t nb_pkts);
+int crypto_virtio_dev_init(struct rte_cryptodev *cryptodev, uint64_t features,
+ struct rte_pci_device *pci_dev);
+
#endif /* _VI
[v5 6/6] test/crypto: add tests for virtio user PMD
Reuse virtio_crypto tests for testing virtio_crypto_user PMD.
Signed-off-by: Gowrishankar Muthukrishnan
---
app/test/test_cryptodev.c | 7 +++
app/test/test_cryptodev.h | 1 +
app/test/test_cryptodev_asym.c | 15 +++
3 files changed, 23 insertions(+)
diff --git a/app/test/test_cryptodev.c b/app/test/test_cryptodev.c
index 441ecc6ad5..60aacdc155 100644
--- a/app/test/test_cryptodev.c
+++ b/app/test/test_cryptodev.c
@@ -19737,6 +19737,12 @@ test_cryptodev_virtio(void)
return run_cryptodev_testsuite(RTE_STR(CRYPTODEV_NAME_VIRTIO_PMD));
}
+static int
+test_cryptodev_virtio_user(void)
+{
+ return run_cryptodev_testsuite(RTE_STR(CRYPTODEV_NAME_VIRTIO_USER_PMD));
+}
+
static int
test_cryptodev_aesni_mb(void)
{
@@ -20074,6 +20080,7 @@ REGISTER_DRIVER_TEST(cryptodev_dpaa_sec_autotest,
test_cryptodev_dpaa_sec);
REGISTER_DRIVER_TEST(cryptodev_ccp_autotest, test_cryptodev_ccp);
REGISTER_DRIVER_TEST(cryptodev_uadk_autotest, test_cryptodev_uadk);
REGISTER_DRIVER_TEST(cryptodev_virtio_autotest, test_cryptodev_virtio);
+REGISTER_DRIVER_TEST(cryptodev_virtio_user_autotest,
test_cryptodev_virtio_user);
REGISTER_DRIVER_TEST(cryptodev_octeontx_autotest, test_cryptodev_octeontx);
REGISTER_DRIVER_TEST(cryptodev_caam_jr_autotest, test_cryptodev_caam_jr);
REGISTER_DRIVER_TEST(cryptodev_nitrox_autotest, test_cryptodev_nitrox);
diff --git a/app/test/test_cryptodev.h b/app/test/test_cryptodev.h
index bb54a33d62..f6c7478f19 100644
--- a/app/test/test_cryptodev.h
+++ b/app/test/test_cryptodev.h
@@ -64,6 +64,7 @@
#define CRYPTODEV_NAME_MVSAM_PMD crypto_mvsam
#define CRYPTODEV_NAME_CCP_PMD crypto_ccp
#define CRYPTODEV_NAME_VIRTIO_PMD crypto_virtio
+#define CRYPTODEV_NAME_VIRTIO_USER_PMD crypto_virtio_user
#define CRYPTODEV_NAME_OCTEONTX_SYM_PMDcrypto_octeontx
#define CRYPTODEV_NAME_CAAM_JR_PMD crypto_caam_jr
#define CRYPTODEV_NAME_NITROX_PMD crypto_nitrox_sym
diff --git a/app/test/test_cryptodev_asym.c b/app/test/test_cryptodev_asym.c
index ac47be724f..a98e3dc824 100644
--- a/app/test/test_cryptodev_asym.c
+++ b/app/test/test_cryptodev_asym.c
@@ -4118,9 +4118,24 @@ test_cryptodev_virtio_asym(void)
return unit_test_suite_runner(&cryptodev_virtio_asym_testsuite);
}
+static int
+test_cryptodev_virtio_user_asym(void)
+{
+ gbl_driver_id = rte_cryptodev_driver_id_get(
+ RTE_STR(CRYPTODEV_NAME_VIRTIO_USER_PMD));
+ if (gbl_driver_id == -1) {
+ RTE_LOG(ERR, USER1, "virtio user PMD must be loaded.\n");
+ return TEST_FAILED;
+ }
+
+ /* Use test suite registered for crypto_virtio_user PMD */
+ return unit_test_suite_runner(&cryptodev_virtio_asym_testsuite);
+}
+
REGISTER_DRIVER_TEST(cryptodev_openssl_asym_autotest,
test_cryptodev_openssl_asym);
REGISTER_DRIVER_TEST(cryptodev_qat_asym_autotest, test_cryptodev_qat_asym);
REGISTER_DRIVER_TEST(cryptodev_octeontx_asym_autotest,
test_cryptodev_octeontx_asym);
REGISTER_DRIVER_TEST(cryptodev_cn9k_asym_autotest, test_cryptodev_cn9k_asym);
REGISTER_DRIVER_TEST(cryptodev_cn10k_asym_autotest, test_cryptodev_cn10k_asym);
REGISTER_DRIVER_TEST(cryptodev_virtio_asym_autotest,
test_cryptodev_virtio_asym);
+REGISTER_DRIVER_TEST(cryptodev_virtio_user_asym_autotest,
test_cryptodev_virtio_user_asym);
--
2.25.1
[v5 5/6] test/crypto: add asymmetric tests for virtio PMD
Add asymmetric tests for Virtio PMD.
Signed-off-by: Gowrishankar Muthukrishnan
---
app/test/test_cryptodev_asym.c | 28
1 file changed, 28 insertions(+)
diff --git a/app/test/test_cryptodev_asym.c b/app/test/test_cryptodev_asym.c
index 9b5f3c545e..ac47be724f 100644
--- a/app/test/test_cryptodev_asym.c
+++ b/app/test/test_cryptodev_asym.c
@@ -4023,6 +4023,19 @@ static struct unit_test_suite
cryptodev_octeontx_asym_testsuite = {
}
};
+static struct unit_test_suite cryptodev_virtio_asym_testsuite = {
+ .suite_name = "Crypto Device VIRTIO ASYM Unit Test Suite",
+ .setup = testsuite_setup,
+ .teardown = testsuite_teardown,
+ .unit_test_cases = {
+ TEST_CASE_ST(ut_setup_asym, ut_teardown_asym, test_capability),
+ TEST_CASE_ST(ut_setup_asym, ut_teardown_asym,
+ test_rsa_sign_verify_crt),
+ TEST_CASE_ST(ut_setup_asym, ut_teardown_asym,
test_rsa_enc_dec_crt),
+ TEST_CASES_END() /**< NULL terminate unit test array */
+ }
+};
+
static int
test_cryptodev_openssl_asym(void)
{
@@ -4091,8 +4104,23 @@ test_cryptodev_cn10k_asym(void)
return unit_test_suite_runner(&cryptodev_octeontx_asym_testsuite);
}
+static int
+test_cryptodev_virtio_asym(void)
+{
+ gbl_driver_id = rte_cryptodev_driver_id_get(
+ RTE_STR(CRYPTODEV_NAME_VIRTIO_PMD));
+ if (gbl_driver_id == -1) {
+ RTE_LOG(ERR, USER1, "virtio PMD must be loaded.\n");
+ return TEST_FAILED;
+ }
+
+ /* Use test suite registered for crypto_virtio PMD */
+ return unit_test_suite_runner(&cryptodev_virtio_asym_testsuite);
+}
+
REGISTER_DRIVER_TEST(cryptodev_openssl_asym_autotest,
test_cryptodev_openssl_asym);
REGISTER_DRIVER_TEST(cryptodev_qat_asym_autotest, test_cryptodev_qat_asym);
REGISTER_DRIVER_TEST(cryptodev_octeontx_asym_autotest,
test_cryptodev_octeontx_asym);
REGISTER_DRIVER_TEST(cryptodev_cn9k_asym_autotest, test_cryptodev_cn9k_asym);
REGISTER_DRIVER_TEST(cryptodev_cn10k_asym_autotest, test_cryptodev_cn10k_asym);
+REGISTER_DRIVER_TEST(cryptodev_virtio_asym_autotest,
test_cryptodev_virtio_asym);
--
2.25.1
[v5 2/6] crypto/virtio: refactor queue operations
Move existing control queue operations into a common place
that would be shared with other virtio type of devices.
Signed-off-by: Gowrishankar Muthukrishnan
---
drivers/crypto/virtio/meson.build | 1 +
drivers/crypto/virtio/virtio_crypto_algs.h | 2 +-
drivers/crypto/virtio/virtio_cryptodev.c | 563 -
drivers/crypto/virtio/virtio_cvq.c | 129 +
drivers/crypto/virtio/virtio_cvq.h | 33 ++
drivers/crypto/virtio/virtio_pci.h | 6 +-
drivers/crypto/virtio/virtio_ring.h| 12 +-
drivers/crypto/virtio/virtio_rxtx.c| 44 +-
drivers/crypto/virtio/virtio_rxtx.h| 13 +
drivers/crypto/virtio/virtqueue.c | 191 ++-
drivers/crypto/virtio/virtqueue.h | 89 +++-
11 files changed, 692 insertions(+), 391 deletions(-)
create mode 100644 drivers/crypto/virtio/virtio_cvq.c
create mode 100644 drivers/crypto/virtio/virtio_cvq.h
create mode 100644 drivers/crypto/virtio/virtio_rxtx.h
diff --git a/drivers/crypto/virtio/meson.build
b/drivers/crypto/virtio/meson.build
index 45533c9b89..d2c3b3ad07 100644
--- a/drivers/crypto/virtio/meson.build
+++ b/drivers/crypto/virtio/meson.build
@@ -11,6 +11,7 @@ includes += include_directories('../../../lib/vhost')
deps += 'bus_pci'
sources = files(
'virtio_cryptodev.c',
+'virtio_cvq.c',
'virtio_pci.c',
'virtio_rxtx.c',
'virtqueue.c',
diff --git a/drivers/crypto/virtio/virtio_crypto_algs.h
b/drivers/crypto/virtio/virtio_crypto_algs.h
index 4c44af3733..3824017ca5 100644
--- a/drivers/crypto/virtio/virtio_crypto_algs.h
+++ b/drivers/crypto/virtio/virtio_crypto_algs.h
@@ -22,7 +22,7 @@ struct virtio_crypto_session {
phys_addr_t phys_addr;
} aad;
- struct virtio_crypto_op_ctrl_req ctrl;
+ struct virtio_pmd_ctrl ctrl;
};
#endif /* _VIRTIO_CRYPTO_ALGS_H_ */
diff --git a/drivers/crypto/virtio/virtio_cryptodev.c
b/drivers/crypto/virtio/virtio_cryptodev.c
index 6a264bc24a..6bb76ff15e 100644
--- a/drivers/crypto/virtio/virtio_cryptodev.c
+++ b/drivers/crypto/virtio/virtio_cryptodev.c
@@ -64,211 +64,6 @@ static const struct rte_cryptodev_capabilities
virtio_capabilities[] = {
uint8_t cryptodev_virtio_driver_id;
-#define NUM_ENTRY_SYM_CREATE_SESSION 4
-
-static int
-virtio_crypto_send_command(struct virtqueue *vq,
- struct virtio_crypto_op_ctrl_req *ctrl, uint8_t *cipher_key,
- uint8_t *auth_key, struct virtio_crypto_session *session)
-{
- uint8_t idx = 0;
- uint8_t needed = 1;
- uint32_t head = 0;
- uint32_t len_cipher_key = 0;
- uint32_t len_auth_key = 0;
- uint32_t len_ctrl_req = sizeof(struct virtio_crypto_op_ctrl_req);
- uint32_t len_session_input = sizeof(struct virtio_crypto_session_input);
- uint32_t len_total = 0;
- uint32_t input_offset = 0;
- void *virt_addr_started = NULL;
- phys_addr_t phys_addr_started;
- struct vring_desc *desc;
- uint32_t desc_offset;
- struct virtio_crypto_session_input *input;
- int ret;
-
- PMD_INIT_FUNC_TRACE();
-
- if (session == NULL) {
- VIRTIO_CRYPTO_SESSION_LOG_ERR("session is NULL.");
- return -EINVAL;
- }
- /* cipher only is supported, it is available if auth_key is NULL */
- if (!cipher_key) {
- VIRTIO_CRYPTO_SESSION_LOG_ERR("cipher key is NULL.");
- return -EINVAL;
- }
-
- head = vq->vq_desc_head_idx;
- VIRTIO_CRYPTO_INIT_LOG_DBG("vq->vq_desc_head_idx = %d, vq = %p",
- head, vq);
-
- if (vq->vq_free_cnt < needed) {
- VIRTIO_CRYPTO_SESSION_LOG_ERR("Not enough entry");
- return -ENOSPC;
- }
-
- /* calculate the length of cipher key */
- if (session->ctrl.header.algo == VIRTIO_CRYPTO_SERVICE_CIPHER) {
- switch (ctrl->u.sym_create_session.op_type) {
- case VIRTIO_CRYPTO_SYM_OP_CIPHER:
- len_cipher_key =
ctrl->u.sym_create_session.u.cipher.para.keylen;
- break;
- case VIRTIO_CRYPTO_SYM_OP_ALGORITHM_CHAINING:
- len_cipher_key =
-
ctrl->u.sym_create_session.u.chain.para.cipher_param.keylen;
- break;
- default:
- VIRTIO_CRYPTO_SESSION_LOG_ERR("invalid op type");
- return -EINVAL;
- }
- } else if (session->ctrl.header.algo == VIRTIO_CRYPTO_AKCIPHER_RSA) {
- len_cipher_key = ctrl->u.akcipher_create_session.para.keylen;
- } else {
- VIRTIO_CRYPTO_SESSION_LOG_ERR("Invalid crypto service for
cipher key");
- return -EINVAL;
- }
-
- /* calculate the length of auth key */
- if (auth_key) {
- len_auth_key =
- ctrl->u.sym_
[v6 2/5] vhost: update vhost_user crypto session parameters
As per requirements on vhost_user spec, session id should be
located at the end of session parameter.
Update VhostUserCryptoSessionParam structure to support newer QEMU
versions (v9). Due to additional parameters added in QEMU,
received payload from QEMU would be larger than existing payload.
Hence, it would break parsing vhost_user messages.
This patch addresses both of the above problems.
Signed-off-by: Gowrishankar Muthukrishnan
Acked-by: Akhil Goyal
---
lib/vhost/vhost_crypto.c | 12 ++--
lib/vhost/vhost_user.h | 33 +
2 files changed, 35 insertions(+), 10 deletions(-)
diff --git a/lib/vhost/vhost_crypto.c b/lib/vhost/vhost_crypto.c
index d3d13eff07..c014c85797 100644
--- a/lib/vhost/vhost_crypto.c
+++ b/lib/vhost/vhost_crypto.c
@@ -238,7 +238,7 @@ struct vhost_crypto_data_req {
static int
transform_cipher_param(struct rte_crypto_sym_xform *xform,
- VhostUserCryptoSessionParam *param)
+ VhostUserCryptoSymSessionParam *param)
{
int ret;
@@ -274,7 +274,7 @@ transform_cipher_param(struct rte_crypto_sym_xform *xform,
static int
transform_chain_param(struct rte_crypto_sym_xform *xforms,
- VhostUserCryptoSessionParam *param)
+ VhostUserCryptoSymSessionParam *param)
{
struct rte_crypto_sym_xform *xform_cipher, *xform_auth;
int ret;
@@ -342,10 +342,10 @@ vhost_crypto_create_sess(struct vhost_crypto *vcrypto,
struct rte_cryptodev_sym_session *session;
int ret;
- switch (sess_param->op_type) {
+ switch (sess_param->u.sym_sess.op_type) {
case VIRTIO_CRYPTO_SYM_OP_NONE:
case VIRTIO_CRYPTO_SYM_OP_CIPHER:
- ret = transform_cipher_param(&xform1, sess_param);
+ ret = transform_cipher_param(&xform1, &sess_param->u.sym_sess);
if (unlikely(ret)) {
VC_LOG_ERR("Error transform session msg (%i)", ret);
sess_param->session_id = ret;
@@ -353,7 +353,7 @@ vhost_crypto_create_sess(struct vhost_crypto *vcrypto,
}
break;
case VIRTIO_CRYPTO_SYM_OP_ALGORITHM_CHAINING:
- if (unlikely(sess_param->hash_mode !=
+ if (unlikely(sess_param->u.sym_sess.hash_mode !=
VIRTIO_CRYPTO_SYM_HASH_MODE_AUTH)) {
sess_param->session_id = -VIRTIO_CRYPTO_NOTSUPP;
VC_LOG_ERR("Error transform session message (%i)",
@@ -363,7 +363,7 @@ vhost_crypto_create_sess(struct vhost_crypto *vcrypto,
xform1.next = &xform2;
- ret = transform_chain_param(&xform1, sess_param);
+ ret = transform_chain_param(&xform1, &sess_param->u.sym_sess);
if (unlikely(ret)) {
VC_LOG_ERR("Error transform session message (%i)", ret);
sess_param->session_id = ret;
diff --git a/lib/vhost/vhost_user.h b/lib/vhost/vhost_user.h
index 9a905ee5f4..ef486545ba 100644
--- a/lib/vhost/vhost_user.h
+++ b/lib/vhost/vhost_user.h
@@ -99,11 +99,10 @@ typedef struct VhostUserLog {
/* Comply with Cryptodev-Linux */
#define VHOST_USER_CRYPTO_MAX_HMAC_KEY_LENGTH 512
#define VHOST_USER_CRYPTO_MAX_CIPHER_KEY_LENGTH64
+#define VHOST_USER_CRYPTO_MAX_KEY_LENGTH 1024
/* Same structure as vhost-user backend session info */
-typedef struct VhostUserCryptoSessionParam {
- int64_t session_id;
- uint32_t op_code;
+typedef struct VhostUserCryptoSymSessionParam {
uint32_t cipher_algo;
uint32_t cipher_key_len;
uint32_t hash_algo;
@@ -114,10 +113,36 @@ typedef struct VhostUserCryptoSessionParam {
uint8_t dir;
uint8_t hash_mode;
uint8_t chaining_dir;
- uint8_t *ciphe_key;
+ uint8_t *cipher_key;
uint8_t *auth_key;
uint8_t cipher_key_buf[VHOST_USER_CRYPTO_MAX_CIPHER_KEY_LENGTH];
uint8_t auth_key_buf[VHOST_USER_CRYPTO_MAX_HMAC_KEY_LENGTH];
+} VhostUserCryptoSymSessionParam;
+
+
+typedef struct VhostUserCryptoAsymRsaParam {
+ uint32_t padding_algo;
+ uint32_t hash_algo;
+} VhostUserCryptoAsymRsaParam;
+
+typedef struct VhostUserCryptoAsymSessionParam {
+ uint32_t algo;
+ uint32_t key_type;
+ uint32_t key_len;
+ uint8_t *key;
+ union {
+ VhostUserCryptoAsymRsaParam rsa;
+ } u;
+ uint8_t key_buf[VHOST_USER_CRYPTO_MAX_KEY_LENGTH];
+} VhostUserCryptoAsymSessionParam;
+
+typedef struct VhostUserCryptoSessionParam {
+ uint32_t op_code;
+ union {
+ VhostUserCryptoSymSessionParam sym_sess;
+ VhostUserCryptoAsymSessionParam asym_sess;
+ } u;
+ int64_t session_id;
} VhostUserCryptoSessionParam;
typedef struct VhostUserVringArea {
--
2.25.1
[v6 1/5] vhost: skip crypto op fetch before vring init
Until virtio avail ring is initialized (by VHOST_USER_SET_VRING_ADDR),
worker thread should not try to fetch crypto op, which would lead to
memory fault.
Fixes: 939066d96563 ("vhost/crypto: add public function implementation")
Cc: sta...@dpdk.org
Signed-off-by: Gowrishankar Muthukrishnan
Acked-by: Akhil Goyal
---
v6:
- added lock checks.
---
lib/vhost/vhost_crypto.c | 20
1 file changed, 20 insertions(+)
diff --git a/lib/vhost/vhost_crypto.c b/lib/vhost/vhost_crypto.c
index 3dc41a3bd5..d3d13eff07 100644
--- a/lib/vhost/vhost_crypto.c
+++ b/lib/vhost/vhost_crypto.c
@@ -8,6 +8,7 @@
#include
#include
+#include "iotlb.h"
#include "rte_vhost_crypto.h"
#include "vhost.h"
#include "vhost_user.h"
@@ -1580,7 +1581,26 @@ rte_vhost_crypto_fetch_requests(int vid, uint32_t qid,
vq = dev->virtqueue[qid];
+ if (unlikely(vq == NULL)) {
+ VC_LOG_ERR("Invalid virtqueue %u", qid);
+ return 0;
+ }
+
+ if (unlikely(rte_rwlock_read_trylock(&vq->access_lock) != 0))
+ return 0;
+
+ vhost_user_iotlb_rd_lock(vq);
+ if (unlikely(!vq->access_ok)) {
+ VC_LOG_DBG("Virtqueue %u vrings not yet initialized", qid);
+ vhost_user_iotlb_rd_unlock(vq);
+ rte_rwlock_read_unlock(&vq->access_lock);
+ return 0;
+ }
+
avail_idx = *((volatile uint16_t *)&vq->avail->idx);
+ vhost_user_iotlb_rd_unlock(vq);
+ rte_rwlock_read_unlock(&vq->access_lock);
+
start_idx = vq->last_used_idx;
count = avail_idx - start_idx;
count = RTE_MIN(count, VHOST_CRYPTO_MAX_BURST_SIZE);
--
2.25.1
[v6 3/5] examples/vhost_crypto: fix user callbacks
In order to handle new vhost user connection, use new_connection
and destroy_connection callbacks.
Fixes: f5188211c721 ("examples/vhost_crypto: add sample application")
Cc: sta...@dpdk.org
Signed-off-by: Gowrishankar Muthukrishnan
Acked-by: Akhil Goyal
---
examples/vhost_crypto/main.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/examples/vhost_crypto/main.c b/examples/vhost_crypto/main.c
index 558c09a60f..b1fe4120b9 100644
--- a/examples/vhost_crypto/main.c
+++ b/examples/vhost_crypto/main.c
@@ -362,8 +362,8 @@ destroy_device(int vid)
}
static const struct rte_vhost_device_ops virtio_crypto_device_ops = {
- .new_device = new_device,
- .destroy_device = destroy_device,
+ .new_connection = new_device,
+ .destroy_connection = destroy_device,
};
static int
--
2.25.1
[v6 0/5] vhost: add RSA support
This patch series supports asymmetric RSA in vhost crypto library. It also includes changes to improve vhost crypto library: * support newer QEMU versions. * fix broken vhost_crypto example application. * stabilize crypto fastpath operations. Gowrishankar Muthukrishnan (5): vhost: skip crypto op fetch before vring init vhost: update vhost_user crypto session parameters examples/vhost_crypto: fix user callbacks vhost: support asymmetric RSA crypto ops examples/vhost_crypto: support asymmetric crypto doc/guides/rel_notes/release_25_03.rst| 3 + doc/guides/sample_app_ug/vhost_crypto.rst | 5 + examples/vhost_crypto/main.c | 54 ++- lib/vhost/vhost_crypto.c | 518 -- lib/vhost/vhost_user.h| 33 +- lib/vhost/virtio_crypto.h | 67 +++ 6 files changed, 621 insertions(+), 59 deletions(-) -- 2.25.1
[v5 0/6] crypto/virtio: enhancements for RSA and vDPA
This patch series enhances virtio crypto PMD to:
* support RSA
* support packed virtio ring
* support vDPA backend
Depends-on: patch-152109 ("vhost: support asymmetric RSA crypto ops")
Gowrishankar Muthukrishnan (6):
crypto/virtio: add asymmetric RSA support
crypto/virtio: refactor queue operations
crypto/virtio: add packed ring support
crypto/virtio: add vDPA backend
test/crypto: add asymmetric tests for virtio PMD
test/crypto: add tests for virtio user PMD
app/test/test_cryptodev.c |7 +
app/test/test_cryptodev.h |1 +
app/test/test_cryptodev_asym.c| 43 +
doc/guides/cryptodevs/features/virtio.ini |4 +
doc/guides/rel_notes/release_25_03.rst|4 +
drivers/crypto/virtio/meson.build |8 +
drivers/crypto/virtio/virtio_crypto_algs.h|2 +-
.../virtio/virtio_crypto_capabilities.h | 19 +
drivers/crypto/virtio/virtio_cryptodev.c | 1060 +++--
drivers/crypto/virtio/virtio_cryptodev.h | 18 +-
drivers/crypto/virtio/virtio_cvq.c| 228
drivers/crypto/virtio/virtio_cvq.h| 33 +
drivers/crypto/virtio/virtio_logs.h |6 +-
drivers/crypto/virtio/virtio_pci.h| 38 +-
drivers/crypto/virtio/virtio_ring.h | 65 +-
drivers/crypto/virtio/virtio_rxtx.c | 725 ++-
drivers/crypto/virtio/virtio_rxtx.h | 13 +
drivers/crypto/virtio/virtio_user/vhost.h | 90 ++
.../crypto/virtio/virtio_user/vhost_vdpa.c| 710 +++
.../virtio/virtio_user/virtio_user_dev.c | 749
.../virtio/virtio_user/virtio_user_dev.h | 85 ++
drivers/crypto/virtio/virtio_user_cryptodev.c | 575 +
drivers/crypto/virtio/virtqueue.c | 229 +++-
drivers/crypto/virtio/virtqueue.h | 221 +++-
lib/cryptodev/cryptodev_pmd.h |6 +
25 files changed, 4447 insertions(+), 492 deletions(-)
create mode 100644 drivers/crypto/virtio/virtio_cvq.c
create mode 100644 drivers/crypto/virtio/virtio_cvq.h
create mode 100644 drivers/crypto/virtio/virtio_rxtx.h
create mode 100644 drivers/crypto/virtio/virtio_user/vhost.h
create mode 100644 drivers/crypto/virtio/virtio_user/vhost_vdpa.c
create mode 100644 drivers/crypto/virtio/virtio_user/virtio_user_dev.c
create mode 100644 drivers/crypto/virtio/virtio_user/virtio_user_dev.h
create mode 100644 drivers/crypto/virtio/virtio_user_cryptodev.c
--
2.25.1
[v5 1/6] crypto/virtio: add asymmetric RSA support
Asymmetric RSA operations (SIGN, VERIFY, ENCRYPT and DECRYPT) are
supported in virtio PMD.
Signed-off-by: Gowrishankar Muthukrishnan
---
Depends-on: patch-152109 ("vhost: support asymmetric RSA crypto ops")
v5:
- fixed asym enqueue xmit func for correct head idx.
- release notes updated.
---
doc/guides/cryptodevs/features/virtio.ini | 4 +
doc/guides/rel_notes/release_25_03.rst| 4 +
.../virtio/virtio_crypto_capabilities.h | 19 +
drivers/crypto/virtio/virtio_cryptodev.c | 347 +++---
drivers/crypto/virtio/virtio_cryptodev.h | 2 +
drivers/crypto/virtio/virtio_rxtx.c | 247 +++--
lib/cryptodev/cryptodev_pmd.h | 6 +
7 files changed, 551 insertions(+), 78 deletions(-)
diff --git a/doc/guides/cryptodevs/features/virtio.ini
b/doc/guides/cryptodevs/features/virtio.ini
index 8f22b6210b..f720def6b4 100644
--- a/doc/guides/cryptodevs/features/virtio.ini
+++ b/doc/guides/cryptodevs/features/virtio.ini
@@ -4,8 +4,11 @@
;
[Features]
Symmetric crypto = Y
+Asymmetric crypto = Y
Sym operation chaining = Y
OOP LB In LB Out = Y
+RSA PRIV OP KEY EXP= Y
+RSA PRIV OP KEY QT = Y
;
; Supported crypto algorithms of the 'virtio' crypto driver.
@@ -30,6 +33,7 @@ SHA1 HMAC = Y
; Supported Asymmetric algorithms of the 'virtio' crypto driver.
;
[Asymmetric]
+RSA = Y
;
; Supported Operating systems of the 'virtio' crypto driver.
diff --git a/doc/guides/rel_notes/release_25_03.rst
b/doc/guides/rel_notes/release_25_03.rst
index 087a407337..5199cb6f31 100644
--- a/doc/guides/rel_notes/release_25_03.rst
+++ b/doc/guides/rel_notes/release_25_03.rst
@@ -155,6 +155,10 @@ New Features
Updated vhost library to support RSA crypto operations.
+* **Updated virtio crypto driver.**
+
+ Updated virtio crypto driver to support RSA crypto operations.
+
Removed Items
-
diff --git a/drivers/crypto/virtio/virtio_crypto_capabilities.h
b/drivers/crypto/virtio/virtio_crypto_capabilities.h
index 03c30deefd..1b26ff6720 100644
--- a/drivers/crypto/virtio/virtio_crypto_capabilities.h
+++ b/drivers/crypto/virtio/virtio_crypto_capabilities.h
@@ -48,4 +48,23 @@
}, }\
}
+#define VIRTIO_ASYM_CAPABILITIES \
+ { /* RSA */ \
+ .op = RTE_CRYPTO_OP_TYPE_ASYMMETRIC,\
+ {.asym = { \
+ .xform_capa = { \
+ .xform_type = RTE_CRYPTO_ASYM_XFORM_RSA,\
+ .op_types = ((1 << RTE_CRYPTO_ASYM_OP_SIGN) | \
+ (1 << RTE_CRYPTO_ASYM_OP_VERIFY) | \
+ (1 << RTE_CRYPTO_ASYM_OP_ENCRYPT) | \
+ (1 << RTE_CRYPTO_ASYM_OP_DECRYPT)), \
+ {.modlen = {\
+ .min = 1, \
+ .max = 1024,\
+ .increment = 1 \
+ }, }\
+ } \
+}, } \
+ }
+
#endif /* _VIRTIO_CRYPTO_CAPABILITIES_H_ */
diff --git a/drivers/crypto/virtio/virtio_cryptodev.c
b/drivers/crypto/virtio/virtio_cryptodev.c
index 793f50059f..6a264bc24a 100644
--- a/drivers/crypto/virtio/virtio_cryptodev.c
+++ b/drivers/crypto/virtio/virtio_cryptodev.c
@@ -41,6 +41,11 @@ static void virtio_crypto_sym_clear_session(struct
rte_cryptodev *dev,
static int virtio_crypto_sym_configure_session(struct rte_cryptodev *dev,
struct rte_crypto_sym_xform *xform,
struct rte_cryptodev_sym_session *session);
+static void virtio_crypto_asym_clear_session(struct rte_cryptodev *dev,
+ struct rte_cryptodev_asym_session *sess);
+static int virtio_crypto_asym_configure_session(struct rte_cryptodev *dev,
+ struct rte_crypto_asym_xform *xform,
+ struct rte_cryptodev_asym_session *session);
/*
* The set of PCI devices this driver supports
@@ -53,6 +58,7 @@ static const struct rte_pci_id pci_id_virtio_crypto_map[] = {
static const struct rte_cryptodev_capabilities virtio_capabilities[] = {
VIRTIO_SYM_CAPABILITIES,
+ VIRTIO_ASYM_CAPABILITIES,
RTE_CRYPTODEV_END_OF_CAPABILITIES_LIST()
};
@@ -103,22 +109,24 @@ virtio_crypto_send_command(struct virtqueue *vq,
}
/* calculate the length of cipher key */
- if (cipher_key) {
+ if (session->ctrl.header.algo == VIRT
[PATCH v11 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc
tx_bytes is computed after both legs are tested. This might
produce a use after memory free.
The computation is now moved into each leg.
Bugzilla ID: 1440
Fixes: d8a210774e1d ("net/af_xdp: support unaligned umem chunks")
Signed-off-by: Ariel Otilibili
Acked-by: Stephen Hemminger
---
.mailmap| 2 +-
drivers/net/af_xdp/rte_eth_af_xdp.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/.mailmap b/.mailmap
index a03d3cfb591b..ea68d6180ccc 100644
--- a/.mailmap
+++ b/.mailmap
@@ -135,7 +135,7 @@ Anupam Kapoor
Apeksha Gupta
Archana Muniganti
Archit Pandey
-Ariel Otilibili
+Ariel Otilibili
Arkadiusz Kubalewski
Arkadiusz Kusztal
Arnaud Fiorini
diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
b/drivers/net/af_xdp/rte_eth_af_xdp.c
index 814398ba4b44..092bcb73aa0a 100644
--- a/drivers/net/af_xdp/rte_eth_af_xdp.c
+++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
@@ -574,6 +574,7 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t
nb_pkts)
umem->mb_pool->header_size;
offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
desc->addr = addr | offset;
+ tx_bytes += desc->len;
count++;
} else {
struct rte_mbuf *local_mbuf =
@@ -601,11 +602,10 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
uint16_t nb_pkts)
desc->addr = addr | offset;
rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
desc->len);
+ tx_bytes += desc->len;
rte_pktmbuf_free(mbuf);
count++;
}
-
- tx_bytes += mbuf->pkt_len;
}
out:
--
2.30.2
[v5 3/6] crypto/virtio: add packed ring support
Add packed ring support.
Signed-off-by: Gowrishankar Muthukrishnan
---
drivers/crypto/virtio/virtio_cryptodev.c | 125 +++
drivers/crypto/virtio/virtio_cryptodev.h | 13 +-
drivers/crypto/virtio/virtio_cvq.c | 103 +-
drivers/crypto/virtio/virtio_pci.h | 25 ++
drivers/crypto/virtio/virtio_ring.h | 59 ++-
drivers/crypto/virtio/virtio_rxtx.c | 444 ++-
drivers/crypto/virtio/virtqueue.c| 50 ++-
drivers/crypto/virtio/virtqueue.h| 134 ++-
8 files changed, 924 insertions(+), 29 deletions(-)
diff --git a/drivers/crypto/virtio/virtio_cryptodev.c
b/drivers/crypto/virtio/virtio_cryptodev.c
index 6bb76ff15e..92fea557ab 100644
--- a/drivers/crypto/virtio/virtio_cryptodev.c
+++ b/drivers/crypto/virtio/virtio_cryptodev.c
@@ -869,6 +869,125 @@ virtio_crypto_clear_session(
rte_free(ctrl);
}
+static void
+virtio_crypto_clear_session_packed(
+ struct rte_cryptodev *dev,
+ struct virtio_crypto_op_ctrl_req *ctrl)
+{
+ struct virtio_crypto_hw *hw;
+ struct virtqueue *vq;
+ struct vring_packed_desc *desc;
+ uint8_t *status;
+ uint8_t needed = 1;
+ uint32_t head;
+ uint64_t malloc_phys_addr;
+ uint8_t len_inhdr = sizeof(struct virtio_crypto_inhdr);
+ uint32_t len_op_ctrl_req = sizeof(struct virtio_crypto_op_ctrl_req);
+ uint64_t session_id = ctrl->u.destroy_session.session_id;
+ uint16_t flags;
+ uint8_t nb_descs = 0;
+
+ hw = dev->data->dev_private;
+ vq = virtcrypto_cq_to_vq(hw->cvq);
+ head = vq->vq_avail_idx;
+ flags = vq->vq_packed.cached_flags;
+
+ VIRTIO_CRYPTO_SESSION_LOG_INFO("vq->vq_desc_head_idx = %d, "
+ "vq = %p", vq->vq_desc_head_idx, vq);
+
+ if (vq->vq_free_cnt < needed) {
+ VIRTIO_CRYPTO_SESSION_LOG_ERR(
+ "vq->vq_free_cnt = %d is less than %d, "
+ "not enough", vq->vq_free_cnt, needed);
+ return;
+ }
+
+ malloc_phys_addr = rte_malloc_virt2iova(ctrl);
+
+ /* status part */
+ status = &(((struct virtio_crypto_inhdr *)
+ ((uint8_t *)ctrl + len_op_ctrl_req))->status);
+ *status = VIRTIO_CRYPTO_ERR;
+
+ /* indirect desc vring part */
+ desc = vq->vq_packed.ring.desc;
+
+ /* ctrl request part */
+ desc[head].addr = malloc_phys_addr;
+ desc[head].len = len_op_ctrl_req;
+ desc[head].flags = VRING_DESC_F_NEXT | vq->vq_packed.cached_flags;
+ vq->vq_free_cnt--;
+ nb_descs++;
+ if (++vq->vq_avail_idx >= vq->vq_nentries) {
+ vq->vq_avail_idx -= vq->vq_nentries;
+ vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
+ }
+
+ /* status part */
+ desc[vq->vq_avail_idx].addr = malloc_phys_addr + len_op_ctrl_req;
+ desc[vq->vq_avail_idx].len = len_inhdr;
+ desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE;
+ vq->vq_free_cnt--;
+ nb_descs++;
+ if (++vq->vq_avail_idx >= vq->vq_nentries) {
+ vq->vq_avail_idx -= vq->vq_nentries;
+ vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
+ }
+
+ virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
+ vq->hw->weak_barriers);
+
+ virtio_wmb(vq->hw->weak_barriers);
+ virtqueue_notify(vq);
+
+ /* wait for used desc in virtqueue
+* desc_is_used has a load-acquire or rte_io_rmb inside
+*/
+ rte_rmb();
+ while (!desc_is_used(&desc[head], vq)) {
+ rte_rmb();
+ usleep(100);
+ }
+
+ /* now get used descriptors */
+ vq->vq_free_cnt += nb_descs;
+ vq->vq_used_cons_idx += nb_descs;
+ if (vq->vq_used_cons_idx >= vq->vq_nentries) {
+ vq->vq_used_cons_idx -= vq->vq_nentries;
+ vq->vq_packed.used_wrap_counter ^= 1;
+ }
+
+ PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d "
+ "vq->vq_queue_idx=%d "
+ "vq->vq_avail_idx=%d "
+ "vq->vq_used_cons_idx=%d "
+ "vq->vq_packed.cached_flags=0x%x "
+ "vq->vq_packed.used_wrap_counter=%d",
+ vq->vq_free_cnt,
+ vq->vq_queue_index,
+ vq->vq_avail_idx,
+ vq->vq_used_cons_idx,
+ vq->vq_packed.cached_flags,
+ vq->vq_packed.used_wrap_counter);
+
+ if (*status != VIRTIO_CRYPTO_OK) {
+ VIRTIO_CRYPTO_SESSION_LOG_ERR("Close session failed "
+ "status=%"PRIu32", session_id=%"PRIu64"",
+ *status, session_id);
+ rte_free(ctrl);
+ return;
+ }
+
+ VIRTIO_CRYPTO_INIT_LOG_DBG("vq->vq_free_cnt=%d "
+
[PATCH v11 2/2] net/af_xdp: Refactor af_xdp_tx_zc
Both legs of the loop share the same logic: the common parts are about
reserving and filling both address and length into the description.
This is moved into reserve_and_fill().
Bugzilla ID: 1440
Suggested-by: Maryam Tahhan
Signed-off-by: Ariel Otilibili
Acked-by: Maryam Tahhan
---
drivers/net/af_xdp/rte_eth_af_xdp.c | 75 -
1 file changed, 42 insertions(+), 33 deletions(-)
diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
b/drivers/net/af_xdp/rte_eth_af_xdp.c
index 092bcb73aa0a..013f30dfc3bf 100644
--- a/drivers/net/af_xdp/rte_eth_af_xdp.c
+++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
@@ -536,21 +536,49 @@ kick_tx(struct pkt_tx_queue *txq, struct xsk_ring_cons
*cq)
}
}
+static inline struct xdp_desc *
+reserve_and_fill(struct pkt_tx_queue *txq, struct rte_mbuf *mbuf,
+struct xsk_umem_info *umem, void **pkt_ptr)
+{
+ struct xdp_desc *desc = NULL;
+ uint64_t addr, offset;
+ uint32_t idx_tx;
+
+ if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx))
+ goto out;
+
+ desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
+ desc->len = mbuf->pkt_len;
+
+ addr = (uint64_t)mbuf - (uint64_t)umem->buffer
+ - umem->mb_pool->header_size;
+ offset = rte_pktmbuf_mtod(mbuf, uint64_t) - (uint64_t)mbuf
+ + umem->mb_pool->header_size;
+
+ if (pkt_ptr)
+ *pkt_ptr = xsk_umem__get_data(umem->buffer, addr + offset);
+
+ offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
+ desc->addr = addr | offset;
+
+out:
+ return desc;
+}
+
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
static uint16_t
af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
{
struct pkt_tx_queue *txq = queue;
struct xsk_umem_info *umem = txq->umem;
- struct rte_mbuf *mbuf;
+ struct rte_mbuf *mbuf, *local_mbuf = NULL;
unsigned long tx_bytes = 0;
int i;
- uint32_t idx_tx;
uint16_t count = 0;
struct xdp_desc *desc;
- uint64_t addr, offset;
struct xsk_ring_cons *cq = &txq->pair->cq;
uint32_t free_thresh = cq->size >> 1;
+ void *pkt;
if (xsk_cons_nb_avail(cq, free_thresh) >= free_thresh)
pull_umem_cq(umem, XSK_RING_CONS__DEFAULT_NUM_DESCS, cq);
@@ -559,51 +587,32 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
uint16_t nb_pkts)
mbuf = bufs[i];
if (mbuf->pool == umem->mb_pool) {
- if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
+ desc = reserve_and_fill(txq, mbuf, umem, NULL);
+ if (!desc) {
kick_tx(txq, cq);
- if (!xsk_ring_prod__reserve(&txq->tx, 1,
- &idx_tx))
+ desc = reserve_and_fill(txq, mbuf, umem, NULL);
+ if (!desc)
goto out;
}
- desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
- desc->len = mbuf->pkt_len;
- addr = (uint64_t)mbuf - (uint64_t)umem->buffer -
- umem->mb_pool->header_size;
- offset = rte_pktmbuf_mtod(mbuf, uint64_t) -
- (uint64_t)mbuf +
- umem->mb_pool->header_size;
- offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
- desc->addr = addr | offset;
+
tx_bytes += desc->len;
count++;
} else {
- struct rte_mbuf *local_mbuf =
- rte_pktmbuf_alloc(umem->mb_pool);
- void *pkt;
-
- if (local_mbuf == NULL)
+ local_mbuf = rte_pktmbuf_alloc(umem->mb_pool);
+ if (!local_mbuf)
goto out;
- if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
+ desc = reserve_and_fill(txq, local_mbuf, umem, &pkt);
+ if (!desc) {
rte_pktmbuf_free(local_mbuf);
goto out;
}
- desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
desc->len = mbuf->pkt_len;
-
- addr = (uint64_t)local_mbuf - (uint64_t)umem->buffer -
- umem->mb_pool->header_size;
- offset = rte_pktmbuf_mtod(local_mbuf, uint64_t) -
- (uint64_t)local_mbuf +
- umem->mb_pool->header_size;
- pkt = xsk_umem__get_data(umem
Re: [PATCH] common/mlx5: add ConnectX-8 device ID
Hi From: Raslan Darawsheh Sent: Sunday, February 23, 2025 9:16 AM To: NBU-Contact-Thomas Monjalon (EXTERNAL) Cc: dev@dpdk.org Subject: [PATCH] common/mlx5: add ConnectX-8 device ID This adds the ConnectX-8 device id to the list of supported Nvidia devices that run the MLX5 PMDs. Signed-off-by: Raslan Darawsheh Patch applied to next-net-mlx, Kindest regards, Raslan Darawsheh
Re: [PATCH v2 1/1] pcapng: warn if NULL is passed to rte_pcapng_close
Hello Patrick, On Tue, Feb 25, 2025 at 9:42 PM Patrick Robb wrote: > Recheck-request: iol-marvell-Functional > > Putting in a recheck as I believe the fail CI reported is false. > >> >> Thanks for the heads up. This patch did supersede version 2, https://patches.dpdk.org/project/dpdk/patch/20250224164031.587829-2-ariel.otilib...@6wind.com/ Regards, Ariel
[DPDK/other Bug 1665] __rte_trace_mem_get causing out of bounds write
https://bugs.dpdk.org/show_bug.cgi?id=1665
Bug ID: 1665
Summary: __rte_trace_mem_get causing out of bounds write
Product: DPDK
Version: 24.11
Hardware: All
OS: All
Status: UNCONFIRMED
Severity: normal
Priority: Normal
Component: other
Assignee: dev@dpdk.org
Reporter: oleksan...@interfacemasters.com
Target Milestone: ---
When almost out of trace memory, __rte_trace_mem_get can write out of bounds.
It happens in my case if I have trace events of sizes that are not aligned to
__RTE_TRACE_EVENT_HEADER_SZ. like 27,33 etc.
I suspect that the issue is with the incorrect bounds check in
__rte_trace_mem_get.
>uint32_t offset = trace->offset;
>if (unlikely((offset + sz) >= trace->len)) { // assume condition is false,
>and offset is not aligned
>...}
>offset = RTE_ALIGN_CEIL(offset, __RTE_TRACE_EVENT_HEADER_SZ); // after
>offset alignment offset + size might be bigger than trace->len
>void *mem = RTE_PTR_ADD(&trace->mem[0], offset); // returning memory chunk
>that is smaller than requested size
For example:
offset = 21, len = 32, size = 9 -> offset + size is smaller than len
align offset to __RTE_TRACE_EVENT_HEADER_SZ -> offset = 24
offset + size is bigger than len.
--
You are receiving this mail because:
You are the assignee for the bug.
Re: [PATCH] Skip vfio in the scenario of non-privileged mode
On Wed, 22 Jan 2025 16:15:03 +0800
Yang Ming wrote:
> On 2025/1/18 00:47, Stephen Hemminger wrote:
> > Caution: This is an external email. Please be very careful when clicking
> > links or opening attachments. See http://nok.it/nsb for additional
> > information.
> >
> > On Fri, 17 Jan 2025 15:28:47 +0800
> > Yang Ming wrote:
> >
> >> DPDK detect vfio container according the existence of vfio
> >> module. But for container with non-privileged mode, there is
> >> possibility that no VFIO_DIR(/dev/vfio) mapping from host to
> >> container when host have both Intel NIC and Mellanox NIC but
> >> this conntainer only allocate VFs from Mellanox NIC.
> >> In this case, vfio kernel module has already been loaded from
> >> the host.
> >> This scenario will cause the error log occurs in DPDK primary
> >> process as below:
> >> 'EAL: cannot open VFIO container, error 2 (No such file or
> >> directory)'
> >> 'EAL: VFIO support could not be initialized'
> >> Because `rte_vfio_enable()` call `rte_vfio_get_container_fd()`
> >> to execute `vfio_container_fd = open(VFIO_CONTAINER_PATH,
> >> O_RDWR);` but VFIO_CONTAINER_PATH(/dev/vfio/vfio) doesn't exist
> >> in this container.
> >> This scenario will also lead to the delay of DPDK secondary
> >> process because `default_vfio_cfg->vfio_enabled = 0` and
> >> `default_vfio_cfg->vfio_container_fd = -1`, socket error will
> >> be set in DPDK primary process when it sync this info to
> >> the secondary process.
> >> This patch use to skip this kind of useless detection for this
> >> scenario.
> >>
> >> Signed-off-by: Yang Ming
> >> ---
> >> lib/eal/linux/eal_vfio.c | 11 +++
> >> 1 file changed, 11 insertions(+)
> >>
> >> diff --git a/lib/eal/linux/eal_vfio.c b/lib/eal/linux/eal_vfio.c
> >> index 7132e24cba..1679d29263 100644
> >> --- a/lib/eal/linux/eal_vfio.c
> >> +++ b/lib/eal/linux/eal_vfio.c
> >> @@ -7,6 +7,7 @@
> >> #include
> >> #include
> >> #include
> >> +#include
> >>
> >> #include
> >> #include
> >> @@ -1083,6 +1084,7 @@ rte_vfio_enable(const char *modname)
> >>/* initialize group list */
> >>int i, j;
> >>int vfio_available;
> >> + DIR *dir;
> >>const struct internal_config *internal_conf =
> >>eal_get_internal_configuration();
> >>
> >> @@ -1119,6 +1121,15 @@ rte_vfio_enable(const char *modname)
> >>return 0;
> >>}
> >>
> >> + /* return 0 if VFIO directory not exist for container with
> >> non-privileged mode */
> >> + dir = opendir(VFIO_DIR);
> >> + if (dir == NULL) {
> >> + EAL_LOG(DEBUG,
> >> + "VFIO directory not exist, skipping VFIO support...");
> >> + return 0;
> >> + }
> >> + closedir(dir);
> > You need to test the non-container cases.
> > If vfio is loaded /dev/vfio is a character device (not a directory)
> >
> > Also looks suspicious that VFIO_DIR is defined but never used currently.
> >
> Hi Stephen,
> For non-container test, /dev/vfio/vfio will be character device, not
> /dev/vfio.
> Here is the command result on my testing environment with Intel NIC.
>
> [root@computer-1 testuser]# ls -l /dev/vfio
> total 0
> crw-rw-rw-. 1 root root 10, 196 Jan 22 01:50 vfio
> [root@computer-1 testuser]# dpdk-devbind.py -b vfio-pci :04:10.2
> [root@computer-1 testuser]# ls -l /dev/vfio
> total 0
> crw---. 1 root root 239, 0 Jan 22 01:52 59
> crw-rw-rw-. 1 root root 10, 196 Jan 22 01:50 vfio
> [root@computer-1 testuser]# dpdk-devbind.py -b ixgbevf :04:10.2
> [root@computer-1 testuser]# ls -l /dev/vfio
> total 0
> crw-rw-rw-. 1 root root 10, 196 Jan 22 01:50 vfio
>
> Can you confirm your test scenario?
>
>
When vfio-pci is loaded but no device bound:
$ ls -l /dev/vfio
total 0
crw-rw-rw- 1 root root 10, 196 Feb 26 05:39 vfio
After binding device
$ ls -l /dev/vfio
total 0
crw--- 1 root root 511, 0 Feb 26 05:42 15
crw-rw-rw- 1 root root 10, 196 Feb 26 05:39 vfio
So testing for /dev/vfio is good indication that module is loaded.
Not sure what I was thinking earlier.
Re: [PATCH] net/mlx5/hws: change error flow on matcher disconnect
Hi, From: Gregory Etelson Sent: Sunday, February 9, 2025 3:48 PM To: dev@dpdk.org Cc: Gregory Etelson; Maayan Kashani; Raslan Darawsheh; Itamar Gozlan Subject: [PATCH] net/mlx5/hws: change error flow on matcher disconnect From: Itamar Gozlan Currently, when a firmware (FW) failure occurs during matcher error flow, the system attempts to reconnect the matcher, leading to a segmentation fault. This happens because the matcher is freed but remains in the list. Example scenario: Given matchers M1->M2->M3, if a FW failure occurs: 1. System tries to destroy M1 and fails 2. M1 remains in the list but is freed 3. When destroying M2, it attempts to remove itself and create M1->M3 4. This results in a segmentation fault as M1 is already freed Signed-off-by: Itamar Gozlan Patch applied to next-net-mlx, Kindest regards, Raslan Darawsheh
Re: [PATCH] net/mlx5: fix packet hardware ptype translation
Hi,
From: Gregory Etelson
Sent: Sunday, February 9, 2025 4:01 PM
To: dev@dpdk.org
Cc: Gregory Etelson; Maayan Kashani; Raslan Darawsheh; Slava Ovsiienko;
sta...@dpdk.org
Subject: [PATCH] net/mlx5: fix packet hardware ptype translation
From: Viacheslav Ovsiienko
Since ConnectX-8 NIC the hardware recognizes L4 protocol in the
IP fragments if the IP offset is zero and reports the L4 protocol
type in CQE fields. This case was not supported in the ptype translation
table cause not reporting the ptype in the mbufs.
Fixes: ea16068c0064 ("net/mlx5: fix L4 packet type support")
Cc: sta...@dpdk.org
Signed-off-by: Viacheslav Ovsiienko
Patch applied to next-net-mlx,
Kindest regards,
Raslan Darawsheh
Re: [PATCH 1/2] net/mlx5: add DROP destination to mirror clone actions
Hi, From: Gregory Etelson Sent: Thursday, February 13, 2025 10:32 AM To: dev@dpdk.org Cc: Gregory Etelson; Maayan Kashani; Raslan Darawsheh; Tamar Mashiah; sta...@dpdk.org Subject: [PATCH 1/2] net/mlx5: add DROP destination to mirror clone actions From: Tamar Mashiah Add DROP destination to mirror clone actions. cc: sta...@dpdk.org Signed-off-by: Tamar Mashiah Series applied to next-net-mlx, Kindest regards, Raslan Darawsheh
Re: [PATCH] net/mlx5: support ESP in non-template RSS expansion
Hi, From: Gregory Etelson Sent: Monday, February 17, 2025 9:13 AM To: dev@dpdk.org Cc: Gregory Etelson; Maayan Kashani; Raslan Darawsheh Subject: [PATCH] net/mlx5: support ESP in non-template RSS expansion Add support for the ESP protocol in non-template RSS expansion. Signed-off-by: Gregory Etelson Patch applied to next-net-mlx, Kindest regards, Raslan Darawsheh
[PATCH v2 2/3] net/mlx5: add jump FDB Rx flag
When jump FDB Rx is supported, flow will be able to jump
from FDB Tx to FDB Rx, in that case the dest action in FDB
Rx table should support FDB Tx as well.
Signed-off-by: Suanming Mou
---
v2: fix bool and update log info.
---
drivers/common/mlx5/mlx5_devx_cmds.c | 8
drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
drivers/net/mlx5/linux/mlx5_os.c | 9 +++--
drivers/net/mlx5/mlx5.h | 1 +
drivers/net/mlx5/mlx5_flow_hw.c | 8 +++-
5 files changed, 24 insertions(+), 3 deletions(-)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/mlx5_devx_cmds.c
index f504b29f31..440820fd4f 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.c
+++ b/drivers/common/mlx5/mlx5_devx_cmds.c
@@ -924,6 +924,7 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
uint32_t out[MLX5_ST_SZ_DW(query_hca_cap_out)] = {0};
bool hca_cap_2_sup;
uint64_t general_obj_types_supported = 0;
+ uint64_t stc_action_type_127_64;
void *hcattr;
int rc, i;
@@ -1352,6 +1353,13 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
attr->fdb_unified_en = MLX5_GET(wqe_based_flow_table_cap,
hcattr,
fdb_unified_en);
+ stc_action_type_127_64 = MLX5_GET64(wqe_based_flow_table_cap,
+ hcattr,
+ stc_action_type_127_64);
+ if (stc_action_type_127_64 &
+ (1 <<
(MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX -
+ MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX)))
+ attr->jump_fdb_rx_en = 1;
}
/* Query HCA attribute for ROCE. */
if (attr->roce) {
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h
b/drivers/common/mlx5/mlx5_devx_cmds.h
index 8de4210fb2..6c726a0d46 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.h
+++ b/drivers/common/mlx5/mlx5_devx_cmds.h
@@ -326,6 +326,7 @@ struct mlx5_hca_attr {
uint32_t lag_rx_port_affinity:1;
uint32_t wqe_based_flow_table_sup:1;
uint32_t fdb_unified_en:1;
+ uint32_t jump_fdb_rx_en:1;
uint8_t max_header_modify_pattern_length;
uint64_t system_image_guid;
uint32_t log_max_conn_track_offload:5;
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 9410211e3b..77f17f1a14 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1716,8 +1716,13 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
* 4. all representors in HWS
*/
priv->unified_fdb_en = !!priv->master;
- DRV_LOG(DEBUG, "port %u: unified FDB %s enabled.",
- eth_dev->data->port_id, priv->unified_fdb_en ? "is" :
"isn't");
+ /* Jump FDB Rx works only with unified FDB enabled. */
+ if (priv->unified_fdb_en)
+ priv->jump_fdb_rx_en =
sh->cdev->config.hca_attr.jump_fdb_rx_en;
+ DRV_LOG(DEBUG, "port %u: unified FDB %s enabled, jump_fdb_rx %s
enabled.",
+ eth_dev->data->port_id,
+ priv->unified_fdb_en ? "is" : "isn't",
+ priv->jump_fdb_rx_en ? "is" : "isn't");
if (priv->sh->config.dv_esw_en) {
uint32_t usable_bits;
uint32_t required_bits;
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index f73f6e63ff..545ba48b3c 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1987,6 +1987,7 @@ struct mlx5_priv {
uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
uint32_t tunnel_enabled:1; /* If tunnel offloading is enabled on rxqs.
*/
uint32_t unified_fdb_en:1; /* Unified FDB flag per port. */
+ uint32_t jump_fdb_rx_en:1; /* Jump from FDB Tx to FDB Rx flag per port.
*/
uint16_t domain_id; /* Switch domain identifier. */
uint16_t vport_id; /* Associated VF vport index (if any). */
uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index 07d36bfb8f..cdb3dd6420 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -9350,6 +9350,7 @@ flow_hw_grp_create_cb(void *tool_ctx, void *cb_ctx)
struct mlx5_flow_group *grp_data;
struct mlx5dr_table *tbl = NULL;
struct mlx5dr_action *jump;
+ uint32_t hws_flags;
uint32_t idx = 0;
MKSTR(matcher_name, "%s_%s_%u_%u_matcher_list",
attr->transfer ? "FDB" : "NIC", attr->egress ? "egress" :
"ingress",
@@ -9370,10 +9371,15 @@ flow_hw_grp_create_cb(void *tool_ctx, void *cb_ctx)
goto error;
grp_data->tbl = tbl;
if (attr->group) {
+ hws_flags =
[PATCH v2 3/3] net/mlx5: allow FDB RSS
RSS can be used in FDB Rx rules when JUMP_FDB_RX action to allow
jump from FDB Tx to FDB Rx table.
Different with NIC RSS, FDB RSS will not do the internal implicit
metadata copy.
This commit enables the FDB RSS if JUMP_FDB_RX is supported.
Signed-off-by: Suanming Mou
---
v2: update limitation.
---
doc/guides/nics/mlx5.rst| 4
drivers/net/mlx5/mlx5_flow.c| 4 ++--
drivers/net/mlx5/mlx5_flow_dv.c | 32 +---
drivers/net/mlx5/mlx5_flow_hw.c | 20 +++-
4 files changed, 38 insertions(+), 22 deletions(-)
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index d40878e336..0371a397b6 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -968,6 +968,10 @@ Limitations
The flow engine of a process cannot move from active to standby mode
if preceding active application rules are still present and vice versa.
+- Since only when unified FDB is enabled, there will be FDB Rx/Tx, jump FDB Rx
+ is valid only when unfied is enabled. And in unified FDB mode, the tag and
RSS
+ actions are only allowed in FDB Rx domain.
+
Statistics
--
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 69d7621d0b..f8b3e504b3 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1977,9 +1977,9 @@ mlx5_flow_validate_action_mark(struct rte_eth_dev *dev,
RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
"mark action not supported for "
"egress");
- if (attr->transfer && mlx5_hws_active(dev))
+ if (attr->transfer && !mlx5_hws_active(dev))
return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
+ RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
NULL,
"non-template mark action not
supported for transfer");
return 0;
}
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 633c41e358..61d3101ce8 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -8939,21 +8939,23 @@ flow_dv_validate(struct rte_eth_dev *dev, const struct
rte_flow_attr *attr,
RTE_FLOW_ERROR_TYPE_ACTION,
NULL,
"unsupported action MARK");
- if (action_flags & MLX5_FLOW_ACTION_QUEUE)
- return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "unsupported action QUEUE");
- if (action_flags & MLX5_FLOW_ACTION_RSS)
- return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "unsupported action RSS");
- if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
- return rte_flow_error_set(error, EINVAL,
- RTE_FLOW_ERROR_TYPE_ACTION,
- actions,
- "no fate action is found");
+ if (!priv->jump_fdb_rx_en) {
+ if (action_flags & MLX5_FLOW_ACTION_QUEUE)
+ return rte_flow_error_set(error, ENOTSUP,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL,
+ "unsupported action
QUEUE");
+ if (action_flags & MLX5_FLOW_ACTION_RSS)
+ return rte_flow_error_set(error, ENOTSUP,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL,
+ "unsupported action
RSS");
+ if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
+ return rte_flow_error_set(error, EINVAL,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ actions,
+ "no fate action is
found");
+ }
} else {
if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
return rte_flow_error_set(error, EINVAL,
diff --git a/drivers/net/mlx5/
[PATCH v2 1/3] net/mlx5/hws: support jump FDB Rx
Before FW introduced the JUMP_FDB_RX action feature, jump from
FDB Tx to Rx is not allowed. JUMP_FDB_RX feature introduces
the internal loopback for Tx case and allow the REG C0 C1 B be
preserved as well.
This commit adds the JUMP_FDB_RX cap bit check and use
JUMP_FDB_RX instead of FT for dest table FDB Rx case.
Signed-off-by: Suanming Mou
Signed-off-by: Alex Vesker
---
v2: fix line lengh.
---
drivers/common/mlx5/mlx5_prm.h| 9 -
drivers/net/mlx5/hws/mlx5dr_action.c | 26 +-
drivers/net/mlx5/hws/mlx5dr_action.h | 4
drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 +
drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++
drivers/net/mlx5/hws/mlx5dr_context.c | 17 +
drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++
7 files changed, 63 insertions(+), 6 deletions(-)
diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h
index 3fc3b0cd2a..84e3347794 100644
--- a/drivers/common/mlx5/mlx5_prm.h
+++ b/drivers/common/mlx5/mlx5_prm.h
@@ -2466,7 +2466,8 @@ struct mlx5_ifc_wqe_based_flow_table_cap_bits {
u8 reserved_at_60[0x8];
u8 max_header_modify_pattern_length[0x8];
u8 ste_format[0x10];
- u8 stc_action_type[0x80];
+ u8 stc_action_type_63_0[0x40];
+ u8 stc_action_type_127_64[0x40];
u8 header_insert_type[0x10];
u8 header_remove_type[0x10];
u8 trivial_match_definer[0x20];
@@ -3543,6 +3544,11 @@ enum mlx5_ifc_rtc_reparse_mode {
MLX5_IFC_RTC_REPARSE_BY_STC = 0x2,
};
+enum mlx5_ifc_stc_action_type_bit_index {
+ MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX = 64,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX = 71,
+};
+
#define MLX5_IFC_RTC_LINEAR_LOOKUP_TBL_LOG_MAX 16
struct mlx5_ifc_rtc_bits {
@@ -3621,6 +3627,7 @@ enum mlx5_ifc_stc_action_type {
MLX5_IFC_STC_ACTION_TYPE_ALLOW = 0x84,
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_VPORT = 0x85,
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_UPLINK = 0x86,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX = 0x87,
};
enum mlx5_ifc_stc_reparse_mode {
diff --git a/drivers/net/mlx5/hws/mlx5dr_action.c
b/drivers/net/mlx5/hws/mlx5dr_action.c
index b9452a3ebc..e21db5b327 100644
--- a/drivers/net/mlx5/hws/mlx5dr_action.c
+++ b/drivers/net/mlx5/hws/mlx5dr_action.c
@@ -803,6 +803,9 @@ int mlx5dr_action_root_build_attr(struct mlx5dr_rule_action
rule_actions[],
switch (action->type) {
case MLX5DR_ACTION_TYP_TBL:
+ attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
+ attr[i].obj = action->dest_tbl.devx_obj->obj;
+ break;
case MLX5DR_ACTION_TYP_TIR:
attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
attr[i].obj = action->devx_obj;
@@ -1097,6 +1100,17 @@ static void mlx5dr_action_fill_stc_attr(struct
mlx5dr_action *action,
}
break;
case MLX5DR_ACTION_TYP_TBL:
+ attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
+ attr->dest_table_id = obj->id;
+ /* Only for unified FDB Rx case */
+ if (mlx5dr_context_cap_stc(action->ctx,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX)
&&
+ action->dest_tbl.type == MLX5DR_TABLE_TYPE_FDB_RX)
+ attr->action_type =
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX;
+ else
+ attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
+
+ break;
case MLX5DR_ACTION_TYP_DEST_ARRAY:
attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
@@ -1419,17 +1433,19 @@ mlx5dr_action_create_dest_table(struct mlx5dr_context
*ctx,
if (!action)
return NULL;
+ action->dest_tbl.type = tbl->type;
+
if (mlx5dr_action_is_root_flags(flags)) {
if (mlx5dr_context_shared_gvmi_used(ctx))
- action->devx_obj = tbl->local_ft->obj;
+ action->dest_tbl.devx_obj = tbl->local_ft;
else
- action->devx_obj = tbl->ft->obj;
+ action->dest_tbl.devx_obj = tbl->ft;
} else {
+ action->dest_tbl.devx_obj = tbl->ft;
+
ret = mlx5dr_action_create_stcs(action, tbl->ft);
if (ret)
goto free_action;
-
- action->devx_dest.devx_obj = tbl->ft;
}
return action;
@@ -2526,7 +2542,7 @@ mlx5dr_action_create_dest_array(struct mlx5dr_context
*ctx,
case MLX5DR_ACTION_TYP_TBL:
dest_list[i].destination_type =
MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
- dest_list[i].destination_id =
dests[i].
[PATCH v2 0/3] net/mlx5: support jump FDB Rx
Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback for Tx case and allow the REG C0 C1 B be preserved as well. This series adds the jump from FDB Tx to FDB Rx support and allows RSS action in FDB domain rule when jump FDB Rx is supported. Suanming Mou (3): net/mlx5/hws: support jump FDB Rx net/mlx5: add jump FDB Rx flag net/mlx5: allow FDB RSS doc/guides/nics/mlx5.rst | 4 drivers/common/mlx5/mlx5_devx_cmds.c | 8 +++ drivers/common/mlx5/mlx5_devx_cmds.h | 1 + drivers/common/mlx5/mlx5_prm.h| 9 +++- drivers/net/mlx5/hws/mlx5dr_action.c | 26 +- drivers/net/mlx5/hws/mlx5dr_action.h | 4 drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++ drivers/net/mlx5/hws/mlx5dr_context.c | 17 ++ drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++ drivers/net/mlx5/linux/mlx5_os.c | 9 ++-- drivers/net/mlx5/mlx5.h | 1 + drivers/net/mlx5/mlx5_flow.c | 4 ++-- drivers/net/mlx5/mlx5_flow_dv.c | 32 ++- drivers/net/mlx5/mlx5_flow_hw.c | 28 ++- 15 files changed, 125 insertions(+), 31 deletions(-) -- 2.34.1
Re: [PATCH V2 0/4] mlx5 unified fdb
Hi, From: Ophir Munk Sent: Wednesday, February 26, 2025 10:38 AM To: Dariusz Sosnowski; Slava Ovsiienko; Bing Zhao; Ori Kam; Suanming Mou; Matan Azrad Cc: dev@dpdk.org; Raslan Darawsheh Subject: [PATCH V2 0/4] mlx5 unified fdb V1: initial version V2: rebase + determine unified fdb based on FW capability Bing Zhao (1): net/mlx5: remove unneeded FDB flag on representor action Ophir Munk (3): common/mlx5: support FDB unified capability query net/mlx5: support FDB unified domain net/mlx5/hws: allow different types in miss validation drivers/common/mlx5/mlx5_devx_cmds.c | 3 + drivers/common/mlx5/mlx5_devx_cmds.h | 1 + drivers/net/mlx5/hws/mlx5dr_table.c | 18 +- drivers/net/mlx5/linux/mlx5_os.c | 11 + drivers/net/mlx5/mlx5.h | 8 + drivers/net/mlx5/mlx5_flow_hw.c | 521 --- drivers/net/mlx5/mlx5_flow_meter.c | 16 +- drivers/net/mlx5/mlx5_flow_quota.c | 9 +- drivers/net/mlx5/mlx5_hws_cnt.c | 9 +- 9 files changed, 429 insertions(+), 167 deletions(-) -- 2.8.4 rebased squashed second and third patch together, series applied to next-net-mlx, Kindest regards, Raslan Darawsheh
[PATCH v3 0/3] net/mlx5: support jump FDB Rx
Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback for Tx case and allow the REG C0 C1 B be preserved as well. This series adds the jump from FDB Tx to FDB Rx support and allows RSS action in FDB domain rule when jump FDB Rx is supported. v3: - Rebased on v2 of mlx5 Unified FDB patches (series 34723) Suanming Mou (3): net/mlx5/hws: support jump FDB Rx net/mlx5: add jump FDB Rx flag net/mlx5: allow FDB RSS doc/guides/nics/mlx5.rst | 4 drivers/common/mlx5/mlx5_devx_cmds.c | 8 +++ drivers/common/mlx5/mlx5_devx_cmds.h | 1 + drivers/common/mlx5/mlx5_prm.h| 9 +++- drivers/net/mlx5/hws/mlx5dr_action.c | 26 +- drivers/net/mlx5/hws/mlx5dr_action.h | 4 drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++ drivers/net/mlx5/hws/mlx5dr_context.c | 17 ++ drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++ drivers/net/mlx5/linux/mlx5_os.c | 9 ++-- drivers/net/mlx5/mlx5.h | 1 + drivers/net/mlx5/mlx5_flow.c | 4 ++-- drivers/net/mlx5/mlx5_flow_dv.c | 32 ++- drivers/net/mlx5/mlx5_flow_hw.c | 28 ++- 15 files changed, 125 insertions(+), 31 deletions(-) -- 2.39.5
[PATCH v3 3/3] net/mlx5: allow FDB RSS
From: Suanming Mou
RSS can be used in FDB Rx rules when JUMP_FDB_RX action to allow
jump from FDB Tx to FDB Rx table.
Different with NIC RSS, FDB RSS will not do the internal implicit
metadata copy.
This commit enables the FDB RSS if JUMP_FDB_RX is supported.
Signed-off-by: Suanming Mou
Acked-by: Dariusz Sosnowski
---
doc/guides/nics/mlx5.rst| 4
drivers/net/mlx5/mlx5_flow.c| 4 ++--
drivers/net/mlx5/mlx5_flow_dv.c | 32 +---
drivers/net/mlx5/mlx5_flow_hw.c | 20 +++-
4 files changed, 38 insertions(+), 22 deletions(-)
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index def40fc20a..190c58b54d 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -970,6 +970,10 @@ Limitations
The flow engine of a process cannot move from active to standby mode
if preceding active application rules are still present and vice versa.
+- Since only when unified FDB is enabled, there will be FDB Rx/Tx, jump FDB Rx
+ is valid only when unfied is enabled. And in unified FDB mode, the tag and
RSS
+ actions are only allowed in FDB Rx domain.
+
Statistics
--
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 69d7621d0b..f8b3e504b3 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1977,9 +1977,9 @@ mlx5_flow_validate_action_mark(struct rte_eth_dev *dev,
RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
"mark action not supported for "
"egress");
- if (attr->transfer && mlx5_hws_active(dev))
+ if (attr->transfer && !mlx5_hws_active(dev))
return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
+ RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
NULL,
"non-template mark action not
supported for transfer");
return 0;
}
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 633c41e358..61d3101ce8 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -8939,21 +8939,23 @@ flow_dv_validate(struct rte_eth_dev *dev, const struct
rte_flow_attr *attr,
RTE_FLOW_ERROR_TYPE_ACTION,
NULL,
"unsupported action MARK");
- if (action_flags & MLX5_FLOW_ACTION_QUEUE)
- return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "unsupported action QUEUE");
- if (action_flags & MLX5_FLOW_ACTION_RSS)
- return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "unsupported action RSS");
- if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
- return rte_flow_error_set(error, EINVAL,
- RTE_FLOW_ERROR_TYPE_ACTION,
- actions,
- "no fate action is found");
+ if (!priv->jump_fdb_rx_en) {
+ if (action_flags & MLX5_FLOW_ACTION_QUEUE)
+ return rte_flow_error_set(error, ENOTSUP,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL,
+ "unsupported action
QUEUE");
+ if (action_flags & MLX5_FLOW_ACTION_RSS)
+ return rte_flow_error_set(error, ENOTSUP,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL,
+ "unsupported action
RSS");
+ if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
+ return rte_flow_error_set(error, EINVAL,
+
RTE_FLOW_ERROR_TYPE_ACTION,
+ actions,
+ "no fate action is
found");
+ }
} else {
if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
return rte_flow_error_set(error, EINVAL,
diff --gi
[PATCH v3 2/3] net/mlx5: add jump FDB Rx flag
From: Suanming Mou
When jump FDB Rx is supported, flow will be able to jump
from FDB Tx to FDB Rx, in that case the dest action in FDB
Rx table should support FDB Tx as well.
Signed-off-by: Suanming Mou
Acked-by: Dariusz Sosnowski
---
drivers/common/mlx5/mlx5_devx_cmds.c | 8
drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
drivers/net/mlx5/linux/mlx5_os.c | 9 +++--
drivers/net/mlx5/mlx5.h | 1 +
drivers/net/mlx5/mlx5_flow_hw.c | 8 +++-
5 files changed, 24 insertions(+), 3 deletions(-)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/mlx5_devx_cmds.c
index f504b29f31..440820fd4f 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.c
+++ b/drivers/common/mlx5/mlx5_devx_cmds.c
@@ -924,6 +924,7 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
uint32_t out[MLX5_ST_SZ_DW(query_hca_cap_out)] = {0};
bool hca_cap_2_sup;
uint64_t general_obj_types_supported = 0;
+ uint64_t stc_action_type_127_64;
void *hcattr;
int rc, i;
@@ -1352,6 +1353,13 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
attr->fdb_unified_en = MLX5_GET(wqe_based_flow_table_cap,
hcattr,
fdb_unified_en);
+ stc_action_type_127_64 = MLX5_GET64(wqe_based_flow_table_cap,
+ hcattr,
+ stc_action_type_127_64);
+ if (stc_action_type_127_64 &
+ (1 <<
(MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX -
+ MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX)))
+ attr->jump_fdb_rx_en = 1;
}
/* Query HCA attribute for ROCE. */
if (attr->roce) {
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h
b/drivers/common/mlx5/mlx5_devx_cmds.h
index 8de4210fb2..6c726a0d46 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.h
+++ b/drivers/common/mlx5/mlx5_devx_cmds.h
@@ -326,6 +326,7 @@ struct mlx5_hca_attr {
uint32_t lag_rx_port_affinity:1;
uint32_t wqe_based_flow_table_sup:1;
uint32_t fdb_unified_en:1;
+ uint32_t jump_fdb_rx_en:1;
uint8_t max_header_modify_pattern_length;
uint64_t system_image_guid;
uint32_t log_max_conn_track_offload:5;
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 7f13cb2c7d..573e846ed2 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1716,8 +1716,13 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
* 4. all representors in HWS
*/
priv->unified_fdb_en = !!priv->master &&
sh->cdev->config.hca_attr.fdb_unified_en;
- DRV_LOG(DEBUG, "port %u: unified FDB %s enabled.",
- eth_dev->data->port_id, priv->unified_fdb_en ? "is" :
"isn't");
+ /* Jump FDB Rx works only with unified FDB enabled. */
+ if (priv->unified_fdb_en)
+ priv->jump_fdb_rx_en =
sh->cdev->config.hca_attr.jump_fdb_rx_en;
+ DRV_LOG(DEBUG, "port %u: unified FDB %s enabled, jump_fdb_rx %s
enabled.",
+ eth_dev->data->port_id,
+ priv->unified_fdb_en ? "is" : "isn't",
+ priv->jump_fdb_rx_en ? "is" : "isn't");
if (priv->sh->config.dv_esw_en) {
uint32_t usable_bits;
uint32_t required_bits;
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index f73f6e63ff..545ba48b3c 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1987,6 +1987,7 @@ struct mlx5_priv {
uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
uint32_t tunnel_enabled:1; /* If tunnel offloading is enabled on rxqs.
*/
uint32_t unified_fdb_en:1; /* Unified FDB flag per port. */
+ uint32_t jump_fdb_rx_en:1; /* Jump from FDB Tx to FDB Rx flag per port.
*/
uint16_t domain_id; /* Switch domain identifier. */
uint16_t vport_id; /* Associated VF vport index (if any). */
uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index 983af38c96..8909f5b506 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -9350,6 +9350,7 @@ flow_hw_grp_create_cb(void *tool_ctx, void *cb_ctx)
struct mlx5_flow_group *grp_data;
struct mlx5dr_table *tbl = NULL;
struct mlx5dr_action *jump;
+ uint32_t hws_flags;
uint32_t idx = 0;
MKSTR(matcher_name, "%s_%s_%u_%u_matcher_list",
attr->transfer ? "FDB" : "NIC", attr->egress ? "egress" :
"ingress",
@@ -9370,10 +9371,15 @@ flow_hw_grp_create_cb(void *tool_ctx, void *cb_ctx)
goto error;
grp_data->tbl = tbl;
[PATCH v3 1/3] net/mlx5/hws: support jump FDB Rx
From: Suanming Mou
Before FW introduced the JUMP_FDB_RX action feature, jump from
FDB Tx to Rx is not allowed. JUMP_FDB_RX feature introduces
the internal loopback for Tx case and allow the REG C0 C1 B be
preserved as well.
This commit adds the JUMP_FDB_RX cap bit check and use
JUMP_FDB_RX instead of FT for dest table FDB Rx case.
Signed-off-by: Suanming Mou
Signed-off-by: Alex Vesker
Acked-by: Dariusz Sosnowski
---
drivers/common/mlx5/mlx5_prm.h| 9 -
drivers/net/mlx5/hws/mlx5dr_action.c | 26 +-
drivers/net/mlx5/hws/mlx5dr_action.h | 4
drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 +
drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++
drivers/net/mlx5/hws/mlx5dr_context.c | 17 +
drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++
7 files changed, 63 insertions(+), 6 deletions(-)
diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h
index 0885f8757f..742c274a85 100644
--- a/drivers/common/mlx5/mlx5_prm.h
+++ b/drivers/common/mlx5/mlx5_prm.h
@@ -2466,7 +2466,8 @@ struct mlx5_ifc_wqe_based_flow_table_cap_bits {
u8 reserved_at_60[0x8];
u8 max_header_modify_pattern_length[0x8];
u8 ste_format[0x10];
- u8 stc_action_type[0x80];
+ u8 stc_action_type_63_0[0x40];
+ u8 stc_action_type_127_64[0x40];
u8 header_insert_type[0x10];
u8 header_remove_type[0x10];
u8 trivial_match_definer[0x20];
@@ -3543,6 +3544,11 @@ enum mlx5_ifc_rtc_reparse_mode {
MLX5_IFC_RTC_REPARSE_BY_STC = 0x2,
};
+enum mlx5_ifc_stc_action_type_bit_index {
+ MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX = 64,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX = 71,
+};
+
#define MLX5_IFC_RTC_LINEAR_LOOKUP_TBL_LOG_MAX 16
struct mlx5_ifc_rtc_bits {
@@ -3621,6 +3627,7 @@ enum mlx5_ifc_stc_action_type {
MLX5_IFC_STC_ACTION_TYPE_ALLOW = 0x84,
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_VPORT = 0x85,
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_UPLINK = 0x86,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX = 0x87,
};
enum mlx5_ifc_stc_reparse_mode {
diff --git a/drivers/net/mlx5/hws/mlx5dr_action.c
b/drivers/net/mlx5/hws/mlx5dr_action.c
index 200a4995d0..c3e67f8a0b 100644
--- a/drivers/net/mlx5/hws/mlx5dr_action.c
+++ b/drivers/net/mlx5/hws/mlx5dr_action.c
@@ -803,6 +803,9 @@ int mlx5dr_action_root_build_attr(struct mlx5dr_rule_action
rule_actions[],
switch (action->type) {
case MLX5DR_ACTION_TYP_TBL:
+ attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
+ attr[i].obj = action->dest_tbl.devx_obj->obj;
+ break;
case MLX5DR_ACTION_TYP_TIR:
attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
attr[i].obj = action->devx_obj;
@@ -1097,6 +1100,17 @@ static void mlx5dr_action_fill_stc_attr(struct
mlx5dr_action *action,
}
break;
case MLX5DR_ACTION_TYP_TBL:
+ attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
+ attr->dest_table_id = obj->id;
+ /* Only for unified FDB Rx case */
+ if (mlx5dr_context_cap_stc(action->ctx,
+ MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX)
&&
+ action->dest_tbl.type == MLX5DR_TABLE_TYPE_FDB_RX)
+ attr->action_type =
MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX;
+ else
+ attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
+
+ break;
case MLX5DR_ACTION_TYP_DEST_ARRAY:
attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
@@ -1419,17 +1433,19 @@ mlx5dr_action_create_dest_table(struct mlx5dr_context
*ctx,
if (!action)
return NULL;
+ action->dest_tbl.type = tbl->type;
+
if (mlx5dr_action_is_root_flags(flags)) {
if (mlx5dr_context_shared_gvmi_used(ctx))
- action->devx_obj = tbl->local_ft->obj;
+ action->dest_tbl.devx_obj = tbl->local_ft;
else
- action->devx_obj = tbl->ft->obj;
+ action->dest_tbl.devx_obj = tbl->ft;
} else {
+ action->dest_tbl.devx_obj = tbl->ft;
+
ret = mlx5dr_action_create_stcs(action, tbl->ft);
if (ret)
goto free_action;
-
- action->devx_dest.devx_obj = tbl->ft;
}
return action;
@@ -2526,7 +2542,7 @@ mlx5dr_action_create_dest_array(struct mlx5dr_context
*ctx,
case MLX5DR_ACTION_TYP_TBL:
dest_list[i].destination_type =
MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
- dest_list[i].des
RE: release candidate 25.03-rc1
> -Original Message- > From: Thomas Monjalon > Sent: Thursday, February 13, 2025 12:50 AM > To: annou...@dpdk.org > Subject: release candidate 25.03-rc1 > > A new DPDK release candidate is ready for testing: > https://git.dpdk.org/dpdk/tag/?id=v25.03-rc1 > > There are 478 new patches in this snapshot. > > Release notes: > https://doc.dpdk.org/guides/rel_notes/release_25_03.html > > Highlights of 25.03-rc1: > - Staged-Ordered ring (SORING) > - mbuf raw bulk functions > - some Intel drivers merged together > - Yunsilicon xsc networking driver > - ZTE Storage Data Accelerator (ZSDA) driver > - Intel compiler icc support replaced by icx > - more function annotations for analyzers > - more MSVC compatibility > > Please test and report issues on bugs.dpdk.org. > > DPDK 25.03-rc2 is expected in more than two weeks (end of February). > > Thank you everyone > Update the test status for Intel part. dpdk25.03-rc1 all test is done. found six new issues. New issues: 1. [dpdk25.03] nvgre/nvgre_ipv4_checksum_offload: core dumped when quit testpmd -> has fix patch 2. [dpdk-25.03] ice_dcf_acl_filter: the rule that supports both switch and ACL will be prioritized for switch-> Intel dev is under investigating 3. [dpdk-25.03rc1][Ubuntu24.04.1] ipsec_gw_cryptodev_func/qat_3des_cbc_ipv4_transport: buffer overflow detected, Aborted (core dumped)-> has fix patch 4. Bug 1663 - [dpdk-25.03] flexible_rxd/check_IPv4_IPv6_TCP_fields_in_RXD_on_specific_queues: port config rxq/txq failed after attach port-> not fix yet 5. [dpdk-25.03] [NNT] pmd_bonded_8023ad: port start failed after set bonding lacp-> has fix patch 6. [dpdk-25.03]port_control/test_pf_start_stop_reset_close: Failed to manually reset pf-> has fix patch # Basic Intel(R) NIC testing * Build or compile: *Build: cover the build test combination with latest GCC/Clang version and the popular OS revision such as Ubuntu24.10, Ubuntu24.04.1, Fedora41, RHEL10, RHEL9.4, RHEL9.5-RT, FreeBSD14.2, SUSE15.6, OpenAnolis8.9, AzureLinux3.0 etc. - All test passed. *Compile: cover the CFLAGES(O0/O1/O2/O3) with popular OS such as Ubuntu24.04.1 and RHEL9.4. - All test passed with latest dpdk. * PF/VF(i40e, ixgbe): test scenarios including PF/VF-RTE_FLOW/TSO/Jumboframe/checksum offload/VLAN/VXLAN, etc. - All test case is done. found the 5 issue. * PF/VF(ice): test scenarios including Switch features/Package Management/Flow Director/Advanced Tx/Advanced RSS/ACL/DCF/Flexible Descriptor, etc. - Execution rate is done. found the 1,2,4,6 issues. * PF/VF(apfi/cpfi): test scenarios including cpfi_rte_flow/TSO/MTU/Jumboframe/checksum offload, etc. - Execution rate is done. No new issue is found. * Intel NIC single core/NIC performance: test scenarios including PF/VF single core performance test, RFC2544 Zero packet loss performance test, etc. - Execution rate is done. No new issue is found. * Power and IPsec: * Power: test scenarios including bi-direction/Telemetry/Empty Poll Lib/Priority Base Frequency, etc. - Execution rate is done. No new issue is found. * IPsec: test scenarios including ipsec/ipsec-gw/ipsec library basic test - QAT&SW/FIB library, etc. - Execution rate is done. No new issue is found. # Basic cryptodev and virtio testing * Virtio: both function and performance test are covered. Such as PVP/Virtio_loopback/virtio-user loopback/virtio-net VM2VM perf testing/VMAWARE ESXI 8.0U1, etc. - Execution rate is done. No new issue is found. * Cryptodev: *Function test: test scenarios including Cryptodev API testing/CompressDev ISA-L/QAT/ZLIB PMD Testing/FIPS, etc. - Execution rate is done. found the 3 issue. *Performance test: test scenarios including Throughput Performance /Cryptodev Latency, etc. - Execution rate is done. No performance drop. Regards, Xu, Hailin
RE: [PATCH v3 0/3] net/mlx5: support jump FDB Rx
Thanks a lot, Dariusz. > -Original Message- > From: Dariusz Sosnowski > Sent: Thursday, February 27, 2025 12:04 AM > To: Dariusz Sosnowski ; Slava Ovsiienko > ; Bing Zhao ; Ori Kam > ; Suanming Mou ; Matan Azrad > > Cc: dev@dpdk.org; Raslan Darawsheh > Subject: [PATCH v3 0/3] net/mlx5: support jump FDB Rx > > Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to > Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback > for Tx case and allow the REG C0 C1 B be preserved as well. > > This series adds the jump from FDB Tx to FDB Rx support and allows RSS > action in FDB domain rule when jump FDB Rx is supported. > > v3: > - Rebased on v2 of mlx5 Unified FDB patches (series 34723) > > Suanming Mou (3): > net/mlx5/hws: support jump FDB Rx > net/mlx5: add jump FDB Rx flag > net/mlx5: allow FDB RSS > > doc/guides/nics/mlx5.rst | 4 > drivers/common/mlx5/mlx5_devx_cmds.c | 8 +++ > drivers/common/mlx5/mlx5_devx_cmds.h | 1 + > drivers/common/mlx5/mlx5_prm.h| 9 +++- > drivers/net/mlx5/hws/mlx5dr_action.c | 26 +- > drivers/net/mlx5/hws/mlx5dr_action.h | 4 > drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 > drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++ > drivers/net/mlx5/hws/mlx5dr_context.c | 17 ++ > drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++ > drivers/net/mlx5/linux/mlx5_os.c | 9 ++-- > drivers/net/mlx5/mlx5.h | 1 + > drivers/net/mlx5/mlx5_flow.c | 4 ++-- > drivers/net/mlx5/mlx5_flow_dv.c | 32 ++- > drivers/net/mlx5/mlx5_flow_hw.c | 28 ++- > 15 files changed, 125 insertions(+), 31 deletions(-) > > -- > 2.39.5
RE: [PATCH v2 1/3] net/mlx5/hws: support jump FDB Rx
> -Original Message-
> From: Suanming Mou
> Sent: Wednesday, February 26, 2025 10:01 PM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh ; Alex Vesker
>
> Subject: [PATCH v2 1/3] net/mlx5/hws: support jump FDB Rx
>
> Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to
> Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback
> for Tx case and allow the REG C0 C1 B be preserved as well.
>
> This commit adds the JUMP_FDB_RX cap bit check and use JUMP_FDB_RX instead
> of FT for dest table FDB Rx case.
>
> Signed-off-by: Suanming Mou
> Signed-off-by: Alex Vesker
> ---
>
> v2: fix line lengh.
>
> ---
> drivers/common/mlx5/mlx5_prm.h| 9 -
> drivers/net/mlx5/hws/mlx5dr_action.c | 26 +-
> drivers/net/mlx5/hws/mlx5dr_action.h | 4
> drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 +
> drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++
> drivers/net/mlx5/hws/mlx5dr_context.c | 17 +
> drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++
> 7 files changed, 63 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/common/mlx5/mlx5_prm.h
> b/drivers/common/mlx5/mlx5_prm.h index 3fc3b0cd2a..84e3347794 100644
> --- a/drivers/common/mlx5/mlx5_prm.h
> +++ b/drivers/common/mlx5/mlx5_prm.h
> @@ -2466,7 +2466,8 @@ struct mlx5_ifc_wqe_based_flow_table_cap_bits {
> u8 reserved_at_60[0x8];
> u8 max_header_modify_pattern_length[0x8];
> u8 ste_format[0x10];
> - u8 stc_action_type[0x80];
> + u8 stc_action_type_63_0[0x40];
> + u8 stc_action_type_127_64[0x40];
> u8 header_insert_type[0x10];
> u8 header_remove_type[0x10];
> u8 trivial_match_definer[0x20];
> @@ -3543,6 +3544,11 @@ enum mlx5_ifc_rtc_reparse_mode {
> MLX5_IFC_RTC_REPARSE_BY_STC = 0x2,
> };
>
> +enum mlx5_ifc_stc_action_type_bit_index {
> + MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX = 64,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX = 71, };
> +
> #define MLX5_IFC_RTC_LINEAR_LOOKUP_TBL_LOG_MAX 16
>
> struct mlx5_ifc_rtc_bits {
> @@ -3621,6 +3627,7 @@ enum mlx5_ifc_stc_action_type {
> MLX5_IFC_STC_ACTION_TYPE_ALLOW = 0x84,
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_VPORT = 0x85,
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_UPLINK = 0x86,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX = 0x87,
> };
>
> enum mlx5_ifc_stc_reparse_mode {
> diff --git a/drivers/net/mlx5/hws/mlx5dr_action.c
> b/drivers/net/mlx5/hws/mlx5dr_action.c
> index b9452a3ebc..e21db5b327 100644
> --- a/drivers/net/mlx5/hws/mlx5dr_action.c
> +++ b/drivers/net/mlx5/hws/mlx5dr_action.c
> @@ -803,6 +803,9 @@ int mlx5dr_action_root_build_attr(struct
> mlx5dr_rule_action rule_actions[],
>
> switch (action->type) {
> case MLX5DR_ACTION_TYP_TBL:
> + attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
> + attr[i].obj = action->dest_tbl.devx_obj->obj;
> + break;
> case MLX5DR_ACTION_TYP_TIR:
> attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
> attr[i].obj = action->devx_obj;
> @@ -1097,6 +1100,17 @@ static void mlx5dr_action_fill_stc_attr(struct
> mlx5dr_action *action,
> }
> break;
> case MLX5DR_ACTION_TYP_TBL:
> + attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
> + attr->dest_table_id = obj->id;
> + /* Only for unified FDB Rx case */
> + if (mlx5dr_context_cap_stc(action->ctx,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX)
> &&
> + action->dest_tbl.type == MLX5DR_TABLE_TYPE_FDB_RX)
> + attr->action_type =
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX;
> + else
> + attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
> +
> + break;
> case MLX5DR_ACTION_TYP_DEST_ARRAY:
> attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
> attr->action_offset = MLX5DR_ACTION_OFFSET_HIT; @@ -1419,17
> +1433,19 @@ mlx5dr_action_create_dest_table(struct mlx5dr_context *ctx,
> if (!action)
> return NULL;
>
> + action->dest_tbl.type = tbl->type;
> +
> if (mlx5dr_action_is_root_flags(flags)) {
> if (mlx5dr_context_shared_gvmi_used(ctx))
> - action->devx_obj = tbl->local_ft->obj;
> + action->dest_tbl.devx_obj = tbl->local_ft;
> else
> - action->devx_obj = tbl->ft->obj;
> + action->dest_tbl.devx_obj = tbl->ft;
> } else {
> + action->dest_tbl.devx_obj = tbl->ft;
> +
> ret = mlx5dr_action_create_stcs(action, tbl->ft);
> if (ret)
> goto free_action;
> -
> - action->devx_dest.dev
[PATCH v12 2/2] net/af_xdp: Refactor af_xdp_tx_zc
Both legs of the loop share the same logic: the common parts are about
reserving and filling both address and length into the description.
This is moved into reserve_and_fill().
Bugzilla ID: 1440
Suggested-by: Maryam Tahhan
Signed-off-by: Ariel Otilibili
Acked-by: Maryam Tahhan
---
drivers/net/af_xdp/rte_eth_af_xdp.c | 75 -
1 file changed, 42 insertions(+), 33 deletions(-)
diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
b/drivers/net/af_xdp/rte_eth_af_xdp.c
index 092bcb73aa0a..05115150a7b9 100644
--- a/drivers/net/af_xdp/rte_eth_af_xdp.c
+++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
@@ -536,21 +536,49 @@ kick_tx(struct pkt_tx_queue *txq, struct xsk_ring_cons
*cq)
}
}
+static inline struct xdp_desc *
+reserve_and_fill(struct pkt_tx_queue *txq, struct rte_mbuf *mbuf,
+struct xsk_umem_info *umem, void **pkt_ptr)
+{
+ struct xdp_desc *desc = NULL;
+ uint64_t addr, offset;
+ uint32_t idx_tx;
+
+ if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx))
+ goto out;
+
+ desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
+ desc->len = mbuf->pkt_len;
+
+ addr = (uint64_t)mbuf - (uint64_t)umem->buffer
+ - umem->mb_pool->header_size;
+ offset = rte_pktmbuf_mtod(mbuf, uint64_t) - (uint64_t)mbuf
+ + umem->mb_pool->header_size;
+
+ if (pkt_ptr)
+ *pkt_ptr = xsk_umem__get_data(umem->buffer, addr + offset);
+
+ offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
+ desc->addr = addr | offset;
+
+out:
+ return desc;
+}
+
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
static uint16_t
af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
{
struct pkt_tx_queue *txq = queue;
struct xsk_umem_info *umem = txq->umem;
- struct rte_mbuf *mbuf;
+ struct rte_mbuf *mbuf, *local_mbuf = NULL;
unsigned long tx_bytes = 0;
int i;
- uint32_t idx_tx;
uint16_t count = 0;
struct xdp_desc *desc;
- uint64_t addr, offset;
struct xsk_ring_cons *cq = &txq->pair->cq;
uint32_t free_thresh = cq->size >> 1;
+ void *pkt;
if (xsk_cons_nb_avail(cq, free_thresh) >= free_thresh)
pull_umem_cq(umem, XSK_RING_CONS__DEFAULT_NUM_DESCS, cq);
@@ -559,51 +587,32 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
uint16_t nb_pkts)
mbuf = bufs[i];
if (mbuf->pool == umem->mb_pool) {
- if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
+ desc = reserve_and_fill(txq, mbuf, umem, NULL);
+ if (!desc) {
kick_tx(txq, cq);
- if (!xsk_ring_prod__reserve(&txq->tx, 1,
- &idx_tx))
+ desc = reserve_and_fill(txq, mbuf, umem, NULL);
+ if (!desc)
goto out;
}
- desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
- desc->len = mbuf->pkt_len;
- addr = (uint64_t)mbuf - (uint64_t)umem->buffer -
- umem->mb_pool->header_size;
- offset = rte_pktmbuf_mtod(mbuf, uint64_t) -
- (uint64_t)mbuf +
- umem->mb_pool->header_size;
- offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
- desc->addr = addr | offset;
+
tx_bytes += desc->len;
count++;
} else {
- struct rte_mbuf *local_mbuf =
- rte_pktmbuf_alloc(umem->mb_pool);
- void *pkt;
-
- if (local_mbuf == NULL)
+ local_mbuf = rte_pktmbuf_alloc(umem->mb_pool);
+ if (!local_mbuf)
goto out;
- if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
+ desc = reserve_and_fill(txq, local_mbuf, umem, &pkt);
+ if (!desc) {
rte_pktmbuf_free(local_mbuf);
goto out;
}
- desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
desc->len = mbuf->pkt_len;
-
- addr = (uint64_t)local_mbuf - (uint64_t)umem->buffer -
- umem->mb_pool->header_size;
- offset = rte_pktmbuf_mtod(local_mbuf, uint64_t) -
- (uint64_t)local_mbuf +
- umem->mb_pool->header_size;
- pkt = xsk_umem__g
[PATCH v12 0/2] Fix use after free, and refactor af_xdp_tx_zc
Hello, The series addresses Bugzilla ID 1440 in two steps; 1. Fix use after free. 2. Refactor af_xdp_tx_zc(). Thank you, --- v12 * fixed style issues (http://mails.dpdk.org/archives/test-report/2025-February/857083.html) v11 * removed assignments in if statements (Stephen Hemminger) v10 (https://inbox.dpdk.org/dev/20250223215259.448723-1-ariel.otilib...@6wind.com/) * no change v9 (https://inbox.dpdk.org/dev/20250207104552.1663519-1-ariel.otilib...@6wind.com/) * desc->len takes its value from mbuf (Maryam Tahhan) v8 (https://inbox.dpdk.org/dev/20250206204645.1564535-1-ariel.otilib...@6wind.com/) * fixed seg fault (https://inbox.dpdk.org/dev/CAFdtZitaNGhC5Q10ATNa7xXX1JbuWob=yzrcwmq8ltz+qji...@mail.gmail.com/) * addressed feedback from Maryam Tahhan v7 (https://inbox.dpdk.org/dev/20250201100300.2194018-1-ariel.otilib...@6wind.com/) * no change * resent because CI didn't take patch-2 into the series v6 (https://inbox.dpdk.org/dev/20250131231018.2163893-1-ariel.otilib...@6wind.com/) * added missing credits to Maryam Tahhan in v5 v5 (https://inbox.dpdk.org/dev/20250131183439.909831-1-ariel.otilib...@6wind.com/) * fix style issues in the signature of reserve_and_fill() (Stephen Hemminger) v4 (https://inbox.dpdk.org/dev/20250130221853.789366-1-ariel.otilib...@6wind.com/) * redid the refactor (Maryam Tahhan) * marked the fix as acked (Stephen Hemminger) * updated .mailmap, my main e-mail is @6wind.com (https://inbox.dpdk.org/dev/20250115121152.487360-4-otili...@eurecom.fr/) v3 (https://inbox.dpdk.org/dev/20250128231152.249497-1-ariel.otilib...@6wind.com/) * reworded commit message of patch-1 * addressed feedback of Maryam Tahhan v2 (https://inbox.dpdk.org/dev/20250116225151.188214-1-ariel.otilib...@6wind.com/) * reworded commit messages * addressed feedback from Stephen Hemminger v1 (http://inbox.dpdk.org/dev/20250116195640.68885-1-ariel.otilib...@6wind.com/) Ariel Otilibili (2): net/af_xdp: Fix use after free in af_xdp_tx_zc net/af_xdp: Refactor af_xdp_tx_zc .mailmap| 2 +- drivers/net/af_xdp/rte_eth_af_xdp.c | 77 - 2 files changed, 44 insertions(+), 35 deletions(-) -- 2.30.2
[PATCH v11 0/2] Fix use after free, and refactor af_xdp_tx_zc
Hello, The series addresses Bugzilla ID 1440 in two steps; 1. Fix use after free. 2. Refactor af_xdp_tx_zc(). Thank you, --- v11 * removed assignments in if statements (Stephen Hemminger) v10 (https://inbox.dpdk.org/dev/20250223215259.448723-1-ariel.otilib...@6wind.com/) * no change v9 (https://inbox.dpdk.org/dev/20250207104552.1663519-1-ariel.otilib...@6wind.com/) * desc->len takes its value from mbuf (Maryam Tahhan) v8 (https://inbox.dpdk.org/dev/20250206204645.1564535-1-ariel.otilib...@6wind.com/) * fixed seg fault (https://inbox.dpdk.org/dev/CAFdtZitaNGhC5Q10ATNa7xXX1JbuWob=yzrcwmq8ltz+qji...@mail.gmail.com/) * addressed feedback from Maryam Tahhan v7 (https://inbox.dpdk.org/dev/20250201100300.2194018-1-ariel.otilib...@6wind.com/) * no change * resent because CI didn't take patch-2 into the series v6 (https://inbox.dpdk.org/dev/20250131231018.2163893-1-ariel.otilib...@6wind.com/) * added missing credits to Maryam Tahhan in v5 v5 (https://inbox.dpdk.org/dev/20250131183439.909831-1-ariel.otilib...@6wind.com/) * fix style issues in the signature of reserve_and_fill() (Stephen Hemminger) v4 (https://inbox.dpdk.org/dev/20250130221853.789366-1-ariel.otilib...@6wind.com/) * redid the refactor (Maryam Tahhan) * marked the fix as acked (Stephen Hemminger) * updated .mailmap, my main e-mail is @6wind.com (https://inbox.dpdk.org/dev/20250115121152.487360-4-otili...@eurecom.fr/) v3 (https://inbox.dpdk.org/dev/20250128231152.249497-1-ariel.otilib...@6wind.com/) * reworded commit message of patch-1 * addressed feedback of Maryam Tahhan v2 (https://inbox.dpdk.org/dev/20250116225151.188214-1-ariel.otilib...@6wind.com/) * reworded commit messages * addressed feedback from Stephen Hemminger v1 (http://inbox.dpdk.org/dev/20250116195640.68885-1-ariel.otilib...@6wind.com/) Ariel Otilibili (2): net/af_xdp: Fix use after free in af_xdp_tx_zc net/af_xdp: Refactor af_xdp_tx_zc .mailmap| 2 +- drivers/net/af_xdp/rte_eth_af_xdp.c | 77 - 2 files changed, 44 insertions(+), 35 deletions(-) -- 2.30.2
Re: [PATCH v6 2/2] net/af_xdp: Refactor af_xdp_tx_zc
Hello Stephen,
On Mon, Feb 24, 2025 at 8:25 PM Stephen Hemminger <
step...@networkplumber.org> wrote:
>
> Please avoid doing assignment in a conditional statement, can be error
> prone.
> Surprised checkpatch doesn't complain about it.
>
Thanks for the feedback. checkpatch did catch them.
Here is the new series,
https://inbox.dpdk.org/dev/20250226200841.2342632-3-ariel.otilib...@6wind.com/
Regards,
Ariel
>
> Better as:
> desc = reserve_and_fill(txq, mbuf, umem);
> if (!desc) {
> kick_tx(txq, cq);
> desc = reserve_and_fill(txq, mbuf, umem);
> if (!desc)
> goto out;
>
[PATCH v12 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc
tx_bytes is computed after both legs are tested. This might
produce a use after memory free.
The computation is now moved into each leg.
Bugzilla ID: 1440
Fixes: d8a210774e1d ("net/af_xdp: support unaligned umem chunks")
Signed-off-by: Ariel Otilibili
Acked-by: Stephen Hemminger
---
.mailmap| 2 +-
drivers/net/af_xdp/rte_eth_af_xdp.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/.mailmap b/.mailmap
index a03d3cfb591b..ea68d6180ccc 100644
--- a/.mailmap
+++ b/.mailmap
@@ -135,7 +135,7 @@ Anupam Kapoor
Apeksha Gupta
Archana Muniganti
Archit Pandey
-Ariel Otilibili
+Ariel Otilibili
Arkadiusz Kubalewski
Arkadiusz Kusztal
Arnaud Fiorini
diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
b/drivers/net/af_xdp/rte_eth_af_xdp.c
index 814398ba4b44..092bcb73aa0a 100644
--- a/drivers/net/af_xdp/rte_eth_af_xdp.c
+++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
@@ -574,6 +574,7 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t
nb_pkts)
umem->mb_pool->header_size;
offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
desc->addr = addr | offset;
+ tx_bytes += desc->len;
count++;
} else {
struct rte_mbuf *local_mbuf =
@@ -601,11 +602,10 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
uint16_t nb_pkts)
desc->addr = addr | offset;
rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
desc->len);
+ tx_bytes += desc->len;
rte_pktmbuf_free(mbuf);
count++;
}
-
- tx_bytes += mbuf->pkt_len;
}
out:
--
2.30.2
[PATCH v3 3/3] config: create top level variable cc_avx2_flags
Top level variable cc_avx2_flags was created and set according to the
compiler used. It is used in lower level directories like drivers/net
and lib/acl.
Signed-off-by: Andre Muezerie
Acked-by: Bruce Richardson
---
config/x86/meson.build | 6 ++
drivers/net/bnxt/meson.build | 2 +-
drivers/net/enic/meson.build | 2 +-
drivers/net/intel/i40e/meson.build | 2 +-
drivers/net/intel/iavf/meson.build | 2 +-
drivers/net/intel/ice/meson.build | 2 +-
drivers/net/intel/idpf/meson.build | 2 +-
drivers/net/nfp/meson.build| 2 +-
drivers/net/octeon_ep/meson.build | 4 ++--
lib/acl/meson.build| 2 +-
10 files changed, 16 insertions(+), 10 deletions(-)
diff --git a/config/x86/meson.build b/config/x86/meson.build
index 47a5b0c04a..941d4d921f 100644
--- a/config/x86/meson.build
+++ b/config/x86/meson.build
@@ -1,6 +1,12 @@
# SPDX-License-Identifier: BSD-3-Clause
# Copyright(c) 2017-2020 Intel Corporation
+if is_ms_compiler
+cc_avx2_flags = ['/arch:AVX2']
+else
+cc_avx2_flags = ['-mavx2']
+endif
+
# get binutils version for the workaround of Bug 97
binutils_ok = true
if is_linux or cc.get_id() == 'gcc'
diff --git a/drivers/net/bnxt/meson.build b/drivers/net/bnxt/meson.build
index e26cf13a65..fd82d0c409 100644
--- a/drivers/net/bnxt/meson.build
+++ b/drivers/net/bnxt/meson.build
@@ -65,7 +65,7 @@ if arch_subdir == 'x86'
static_rte_bus_pci,
static_rte_kvargs, static_rte_hash],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += bnxt_avx2_lib.extract_objects('bnxt_rxtx_vec_avx2.c')
elif arch_subdir == 'arm' and dpdk_conf.get('RTE_ARCH_64')
sources += files('bnxt_rxtx_vec_neon.c')
diff --git a/drivers/net/enic/meson.build b/drivers/net/enic/meson.build
index 1e26338350..cfe5ec170a 100644
--- a/drivers/net/enic/meson.build
+++ b/drivers/net/enic/meson.build
@@ -38,7 +38,7 @@ if dpdk_conf.has('RTE_ARCH_X86_64')
'enic_rxtx_vec_avx2.c',
dependencies: [static_rte_ethdev, static_rte_bus_pci],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += enic_avx2_lib.extract_objects('enic_rxtx_vec_avx2.c')
endif
diff --git a/drivers/net/intel/i40e/meson.build
b/drivers/net/intel/i40e/meson.build
index ffa40c5d64..2973ed1a01 100644
--- a/drivers/net/intel/i40e/meson.build
+++ b/drivers/net/intel/i40e/meson.build
@@ -49,7 +49,7 @@ if arch_subdir == 'x86'
'i40e_rxtx_vec_avx2.c',
dependencies: [static_rte_ethdev, static_rte_kvargs,
static_rte_hash],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += i40e_avx2_lib.extract_objects('i40e_rxtx_vec_avx2.c')
if cc_has_avx512
diff --git a/drivers/net/intel/iavf/meson.build
b/drivers/net/intel/iavf/meson.build
index 19cd1cfbc8..f7eac7c57a 100644
--- a/drivers/net/intel/iavf/meson.build
+++ b/drivers/net/intel/iavf/meson.build
@@ -37,7 +37,7 @@ if arch_subdir == 'x86'
'iavf_rxtx_vec_avx2.c',
dependencies: [static_rte_ethdev],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += iavf_avx2_lib.extract_objects('iavf_rxtx_vec_avx2.c')
if cc_has_avx512
diff --git a/drivers/net/intel/ice/meson.build
b/drivers/net/intel/ice/meson.build
index a34b7c966a..cbdf38c1c4 100644
--- a/drivers/net/intel/ice/meson.build
+++ b/drivers/net/intel/ice/meson.build
@@ -43,7 +43,7 @@ if arch_subdir == 'x86'
'ice_rxtx_vec_avx2.c',
dependencies: [static_rte_ethdev, static_rte_hash],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += ice_avx2_lib.extract_objects('ice_rxtx_vec_avx2.c')
if cc_has_avx512
diff --git a/drivers/net/intel/idpf/meson.build
b/drivers/net/intel/idpf/meson.build
index 802b13035b..4b272d02b1 100644
--- a/drivers/net/intel/idpf/meson.build
+++ b/drivers/net/intel/idpf/meson.build
@@ -23,7 +23,7 @@ if arch_subdir == 'x86' and dpdk_conf.get('RTE_IOVA_IN_MBUF')
== 1
'idpf_common_rxtx_avx2.c',
dependencies: [static_rte_ethdev, static_rte_hash],
include_directories: includes,
-c_args: [cflags, '-mavx2'])
+c_args: [cflags, cc_avx2_flags])
objs += idpf_avx2_lib.extract_objects('idpf_common_rxtx_avx2.c')
if cc_has_avx512
diff --git a/drivers/net/nfp/meson.build b/drivers/net/nfp/meson.build
index 39762bd45a..0a12b7dce7 100644
--- a/drivers/net/nfp/meson.build
+++ b/drivers/net/nfp/meson.build
@@ -61,7 +61,7 @@ if arch_subdir == 'x86'
avx2_sources,
dependencies: [static_rte_ethdev, static_rte_bus_pci],
include_director
[PATCH v3 0/3] allow faster instruction sets to be used with MSVC
Meson.build files were refactored to add MSVC support for the AVX2 instruction set with MSVC: Top level variable cc_avx2_flags was created and set according to the compiler used. It is used in lower level directories. Andre Muezerie (3): eal: make compatible with instruction set updates for MSVC eal: only use numbers as align parameters for MSVC config: create top level variable cc_avx2_flags config/x86/meson.build | 6 ++ drivers/net/bnxt/meson.build | 2 +- drivers/net/enic/meson.build | 2 +- drivers/net/intel/i40e/meson.build | 2 +- drivers/net/intel/iavf/meson.build | 2 +- drivers/net/intel/ice/meson.build | 2 +- drivers/net/intel/idpf/meson.build | 2 +- drivers/net/nfp/meson.build| 2 +- drivers/net/octeon_ep/meson.build | 4 ++-- lib/acl/meson.build| 2 +- lib/eal/common/rte_random.c| 2 ++ lib/eal/x86/include/rte_vect.h | 11 ++- 12 files changed, 28 insertions(+), 11 deletions(-) -- 2.48.1.vfs.0.0
[PATCH v3 1/3] eal: make compatible with instruction set updates for MSVC
After the instruction set updates for MSVC the error below popped up: ../lib/eal/common/rte_random.c(6): fatal error C1083: Cannot open include file: 'x86intrin.h': No such file or directory The fix is to not include header x86intrin.h with MSVC. Signed-off-by: Andre Muezerie Acked-by: Bruce Richardson --- lib/eal/common/rte_random.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/lib/eal/common/rte_random.c b/lib/eal/common/rte_random.c index 8e62578176..9354358818 100644 --- a/lib/eal/common/rte_random.c +++ b/lib/eal/common/rte_random.c @@ -3,8 +3,10 @@ */ #ifdef __RDSEED__ +#ifndef RTE_TOOLCHAIN_MSVC #include #endif +#endif #include #include -- 2.48.1.vfs.0.0
[PATCH v3 2/3] eal: only use numbers as align parameters for MSVC
After the instruction set updates for MSVC the error below popped up:
..\lib\eal\x86\include\rte_vect.h(82): error C2059: syntax error: '('
The issue is that MSVC does not allow __rte_aligned(RTE_X86_ZMM_SIZE).
It only accepts numbers that are power of 2. So, even though
RTE_X86_ZMM_SIZE represents a number that is a power of two it cannot
be used directly.
https://learn.microsoft.com/en-us/cpp/cpp/align-cpp?view=msvc-170
Signed-off-by: Andre Muezerie
Acked-by: Bruce Richardson
---
lib/eal/x86/include/rte_vect.h | 11 ++-
1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/lib/eal/x86/include/rte_vect.h b/lib/eal/x86/include/rte_vect.h
index 70c78e9b77..9ea158b27e 100644
--- a/lib/eal/x86/include/rte_vect.h
+++ b/lib/eal/x86/include/rte_vect.h
@@ -76,9 +76,18 @@ __extension__ ({\
#ifdef __AVX512F__
-#define RTE_X86_ZMM_SIZE (sizeof(__m512i))
+#define RTE_X86_ZMM_SIZE64
#define RTE_X86_ZMM_MASK (RTE_X86_ZMM_SIZE - 1)
+/*
+ * MSVC does not allow __rte_aligned(sizeof(__m512i)). It only accepts
+ * numbers that are power of 2. So, even though sizeof(__m512i) represents a
+ * number that is a power of two it cannot be used directly.
+ * Ref: https://learn.microsoft.com/en-us/cpp/cpp/align-cpp?view=msvc-170
+ * The static assert below ensures that the hardcoded value defined as
+ * RTE_X86_ZMM_SIZE is equal to sizeof(__m512i).
+ */
+static_assert(RTE_X86_ZMM_SIZE == (sizeof(__m512i)), "Unexpected size of
__m512i");
typedef union __rte_aligned(RTE_X86_ZMM_SIZE) __rte_x86_zmm {
__m512i z;
ymm_ty[RTE_X86_ZMM_SIZE / sizeof(ymm_t)];
--
2.48.1.vfs.0.0
Re: [PATCH] net/mlx5: fix queue counter error check
Hi,
From: Dariusz Sosnowski
Sent: Tuesday, February 25, 2025 11:08 AM
To: Slava Ovsiienko; Bing Zhao; Ori Kam; Suanming Mou; Matan Azrad; Shani Peretz
Cc: dev@dpdk.org; Raslan Darawsheh
Subject: [PATCH] net/mlx5: fix queue counter error check
Whenever queue counter allocation fails, the FW syndrome error
should be checked to determine if maximum number of queue counters
was reached.
Fixes: f0c0731b6d40 ("net/mlx5: add counters for hairpin drop")
Cc: shper...@nvidia.com
Signed-off-by: Dariusz Sosnowski
Acked-by: Bing Zhao
Patch applied to next-net-mlx,
Kindest regards,
Raslan Darawsheh
[PATCH v16 14/29] net/rnp: add Rx burst simple support
add only support simple recv pkts.
Signed-off-by: Wenbo Cao
---
drivers/net/rnp/rnp_ethdev.c | 7 ++
drivers/net/rnp/rnp_rxtx.c | 130 +++
drivers/net/rnp/rnp_rxtx.h | 5 ++
3 files changed, 142 insertions(+)
diff --git a/drivers/net/rnp/rnp_ethdev.c b/drivers/net/rnp/rnp_ethdev.c
index 089e90543c..f1ec2ad73a 100644
--- a/drivers/net/rnp/rnp_ethdev.c
+++ b/drivers/net/rnp/rnp_ethdev.c
@@ -335,6 +335,8 @@ static int rnp_dev_start(struct rte_eth_dev *eth_dev)
goto rxq_start_failed;
/* enable eth rx flow */
RNP_RX_ETH_ENABLE(hw, lane);
+ rnp_rx_func_select(eth_dev);
+ rnp_tx_func_select(eth_dev);
port->port_stopped = 0;
return 0;
@@ -595,6 +597,11 @@ static int rnp_dev_infos_get(struct rte_eth_dev *eth_dev,
/* default port configure */
dev_info->default_rxconf = (struct rte_eth_rxconf) {
.rx_drop_en = 0,
+ .rx_thresh = {
+ .pthresh = RNP_RX_DESC_FETCH_TH,
+ .hthresh = RNP_RX_DESC_FETCH_BURST,
+ },
+ .rx_free_thresh = RNP_DEFAULT_RX_FREE_THRESH,
.offloads = 0,
};
diff --git a/drivers/net/rnp/rnp_rxtx.c b/drivers/net/rnp/rnp_rxtx.c
index e65bc06d36..a742646c7e 100644
--- a/drivers/net/rnp/rnp_rxtx.c
+++ b/drivers/net/rnp/rnp_rxtx.c
@@ -636,3 +636,133 @@ int rnp_rx_queue_start(struct rte_eth_dev *eth_dev,
uint16_t qidx)
return 0;
}
+
+#define RNP_CACHE_FETCH_RX (4)
+static __rte_always_inline int
+rnp_refill_rx_ring(struct rnp_rx_queue *rxq)
+{
+ volatile struct rnp_rx_desc *rxbd;
+ struct rnp_rxsw_entry *rx_swbd;
+ struct rte_eth_dev_data *data;
+ struct rte_mbuf *mb;
+ uint16_t j, i;
+ uint16_t rx_id;
+ int ret;
+
+ rxbd = rxq->rx_bdr + rxq->rxrearm_start;
+ rx_swbd = &rxq->sw_ring[rxq->rxrearm_start];
+ ret = rte_mempool_get_bulk(rxq->mb_pool, (void *)rx_swbd,
+ rxq->rx_free_thresh);
+ data = rte_eth_devices[rxq->attr.port_id].data;
+ if (unlikely(ret != 0)) {
+ if (rxq->rxrearm_nb + rxq->rx_free_thresh >= rxq->attr.nb_desc)
{
+ for (i = 0; i < RNP_CACHE_FETCH_RX; i++) {
+ rx_swbd[i].mbuf = &rxq->fake_mbuf;
+ rxbd[i].d.pkt_addr = 0;
+ rxbd[i].d.cmd = 0;
+ }
+ }
+ data->rx_mbuf_alloc_failed += rxq->rx_free_thresh;
+ return 0;
+ }
+ for (j = 0; j < rxq->rx_free_thresh; ++j) {
+ mb = rx_swbd[j].mbuf;
+ rte_mbuf_refcnt_set(mb, 1);
+ mb->data_off = RTE_PKTMBUF_HEADROOM;
+ mb->port = rxq->attr.port_id;
+
+ rxbd[j].d.pkt_addr = rnp_get_dma_addr(&rxq->attr, mb);
+ rxbd[j].d.cmd = 0;
+ }
+ rxq->rxrearm_start += rxq->rx_free_thresh;
+ if (rxq->rxrearm_start >= rxq->attr.nb_desc - 1)
+ rxq->rxrearm_start = 0;
+ rxq->rxrearm_nb -= rxq->rx_free_thresh;
+
+ rx_id = (uint16_t)((rxq->rxrearm_start == 0) ?
+ (rxq->attr.nb_desc - 1) : (rxq->rxrearm_start - 1));
+ rte_wmb();
+ RNP_REG_WR(rxq->rx_tailreg, 0, rx_id);
+
+ return j;
+}
+
+static __rte_always_inline uint16_t
+rnp_recv_pkts(void *_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+ struct rnp_rx_queue *rxq = (struct rnp_rx_queue *)_rxq;
+ struct rnp_rxsw_entry *rx_swbd;
+ uint32_t state_cmd[RNP_CACHE_FETCH_RX];
+ uint32_t pkt_len[RNP_CACHE_FETCH_RX] = {0};
+ volatile struct rnp_rx_desc *rxbd;
+ struct rte_mbuf *nmb;
+ int nb_dd, nb_rx = 0;
+ int i, j;
+
+ if (unlikely(!rxq->rxq_started || !rxq->rx_link))
+ return 0;
+ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RNP_CACHE_FETCH_RX);
+ rxbd = &rxq->rx_bdr[rxq->rx_tail];
+ rte_prefetch0(rxbd);
+ if (rxq->rxrearm_nb > rxq->rx_free_thresh)
+ rnp_refill_rx_ring(rxq);
+
+ if (!(rxbd->wb.qword1.cmd & RNP_CMD_DD))
+ return 0;
+
+ rx_swbd = &rxq->sw_ring[rxq->rx_tail];
+ for (i = 0; i < nb_pkts;
+ i += RNP_CACHE_FETCH_RX, rxbd += RNP_CACHE_FETCH_RX,
+ rx_swbd += RNP_CACHE_FETCH_RX) {
+ for (j = 0; j < RNP_CACHE_FETCH_RX; j++)
+ state_cmd[j] = rxbd[j].wb.qword1.cmd;
+ rte_atomic_thread_fence(rte_memory_order_acquire);
+
+ for (nb_dd = 0; nb_dd < RNP_CACHE_FETCH_RX &&
+ (state_cmd[nb_dd] &
rte_cpu_to_le_16(RNP_CMD_DD));
+ nb_dd++)
+ ;
+ for (j = 0; j < nb_dd; j++)
+ pkt_len[j] = rxbd[j].wb.qword1.lens;
+
+ for (j = 0; j < nb_dd; ++j) {
+
[PATCH 2/2] net/mlx5: add MPLSoGRE matching in HWS
Remove limitations in validation and MPLS match translation.
For non relaxed mode, assume MPLSoUDP if no protocol is mentioned.
Set UDP dest port to MPLS type only for UDP protocol.
Signed-off-by: Maayan Kashani
Acked-by: Dariusz Sosnowski
---
doc/guides/nics/mlx5.rst | 2 --
drivers/net/mlx5/hws/mlx5dr_definer.c | 28 +--
drivers/net/mlx5/mlx5_flow.c | 7 ---
3 files changed, 13 insertions(+), 24 deletions(-)
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 316c3397d2e..204572c885e 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -356,8 +356,6 @@ Limitations
- L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE
and MPLSoUDP.
-- MPLSoGRE is not supported in HW steering (``dv_flow_en`` = 2).
-
- MPLSoUDP with multiple MPLS headers is only supported in HW steering
(``dv_flow_en`` = 2).
- Match on Geneve header supports the following fields only:
diff --git a/drivers/net/mlx5/hws/mlx5dr_definer.c
b/drivers/net/mlx5/hws/mlx5dr_definer.c
index 837e0c47bda..ba4053c552f 100644
--- a/drivers/net/mlx5/hws/mlx5dr_definer.c
+++ b/drivers/net/mlx5/hws/mlx5dr_definer.c
@@ -1644,14 +1644,9 @@ mlx5dr_definer_conv_item_mpls(struct
mlx5dr_definer_conv_data *cd,
{
const struct rte_flow_item_mpls *m = item->mask;
struct mlx5dr_definer_fc *fc;
- bool inner = cd->tunnel;
-
- if (inner) {
- DR_LOG(ERR, "Inner MPLS item not supported");
- rte_errno = ENOTSUP;
- return rte_errno;
- }
+ bool is_udp;
+ /* If no protocol is set - assume MPLSoUDP */
if (!cd->relaxed) {
/* In order to match on MPLS we must match on ip_protocol and
l4_dport. */
fc = &cd->fc[DR_CALC_FNAME(IP_PROTOCOL, false)];
@@ -1661,14 +1656,17 @@ mlx5dr_definer_conv_item_mpls(struct
mlx5dr_definer_conv_data *cd,
fc->tag_set = &mlx5dr_definer_udp_protocol_set;
DR_CALC_SET(fc, eth_l2, l4_type_bwc, false);
}
-
- /* Currently support only MPLSoUDP */
- fc = &cd->fc[DR_CALC_FNAME(L4_DPORT, false)];
- if (!fc->tag_set) {
- fc->item_idx = item_idx;
- fc->tag_mask_set = &mlx5dr_definer_ones_set;
- fc->tag_set = &mlx5dr_definer_mpls_udp_port_set;
- DR_CALC_SET(fc, eth_l4, destination_port, false);
+ is_udp = (fc->tag_set == &mlx5dr_definer_udp_protocol_set);
+
+ if (is_udp) {
+ /* Set UDP dest port to MPLS. */
+ fc = &cd->fc[DR_CALC_FNAME(L4_DPORT, false)];
+ if (!fc->tag_set) {
+ fc->item_idx = item_idx;
+ fc->tag_mask_set = &mlx5dr_definer_ones_set;
+ fc->tag_set = &mlx5dr_definer_mpls_udp_port_set;
+ DR_CALC_SET(fc, eth_l4, destination_port,
false);
+ }
}
}
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 3fbe89a9d42..69d7621d0bf 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -3812,13 +3812,6 @@ mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev
__rte_unused,
return rte_flow_error_set(error, ENOTSUP,
RTE_FLOW_ERROR_TYPE_ITEM,
item,
"multiple tunnel layers not
supported");
- } else {
- /* Multi-tunnel isn't allowed but MPLS over GRE is an
exception. */
- if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
- !(item_flags & MLX5_FLOW_LAYER_MPLS))
- return rte_flow_error_set(error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ITEM,
item,
- "multiple tunnel layers not
supported");
}
if (!mask)
mask = nic_mask;
--
2.21.0
[PATCH 1/2] doc: fix MPLS limitations under HWS
Update missing limitations.
Fixes: 1e2a51f545b6 ("net/mlx5: support MPLS tunnel with HWS")
Cc: sta...@dpdk.org
Signed-off-by: Maayan Kashani
Acked-by: Dariusz Sosnowski
---
doc/guides/nics/mlx5.rst | 8
1 file changed, 8 insertions(+)
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 690b46a9743..316c3397d2e 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -279,6 +279,14 @@ Limitations
which results in L3 encap.
d. Only in transfer (switchdev) mode.
+ - MPLS:
+
+- RTE_FLOW_ITEM_TYPE_MPLS matching is not supported for mlxcfg
``FLEX_PARSER_PROFILE_ENABLE`` = 1.
+- RTE_FLOW_ITEM_TYPE_MPLS matching is not supported on group 0.
+- Testpmd commands "set mplsoudp_encap" and "set mplsoudp_decap" support
only one MPLS.
+- For non relaxed pattern template, MPLSoUDP match is assumed when trying
to match MPLS w/o former
+ protocol match.
+
- When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
specific VLAN will match for VLAN packets as well:
--
2.21.0
[PATCH] app/test: fix the check of the oop header data
The data of the out-of-place header is never checked. Therefore,
the faulty PMD, which overwrites this data, will not be able to
verify that with tests. New checks to support that were added to
the GCM OOP functions.
Fixes: 51e202f0596f ("test/crypto: rename GCM test code")
Cc: sta...@dpdk.org
Signed-off-by: Arkadiusz Kusztal
---
app/test/test_cryptodev.c | 29 +++--
1 file changed, 27 insertions(+), 2 deletions(-)
diff --git a/app/test/test_cryptodev.c b/app/test/test_cryptodev.c
index 441ecc6ad5..6ef43a784d 100644
--- a/app/test/test_cryptodev.c
+++ b/app/test/test_cryptodev.c
@@ -13769,8 +13769,9 @@ test_authenticated_encryption_oop(const struct
aead_test_data *tdata)
struct crypto_testsuite_params *ts_params = &testsuite_params;
struct crypto_unittest_params *ut_params = &unittest_params;
+ uint32_t i;
int retval;
- uint8_t *ciphertext, *auth_tag;
+ uint8_t *ciphertext, *auth_tag, *buffer_oop;
uint16_t plaintext_pad_len;
struct rte_cryptodev_info dev_info;
@@ -13846,6 +13847,18 @@ test_authenticated_encryption_oop(const struct
aead_test_data *tdata)
ut_params->op->sym->cipher.data.offset);
auth_tag = ciphertext + plaintext_pad_len;
+ /* Check if the data within the offset range is not overwritten in the
OOP */
+ buffer_oop = rte_pktmbuf_mtod(ut_params->obuf, uint8_t *);
+ for (i = 0; i < ut_params->op->sym->cipher.data.offset; i++) {
+ if (buffer_oop[i]) {
+ RTE_LOG(ERR, USER1,
+ "Incorrect value of the output buffer
header\n");
+ debug_hexdump(stdout, "Incorrect value:", buffer_oop,
+ ut_params->op->sym->cipher.data.offset);
+ return TEST_FAILED;
+ }
+ }
+
debug_hexdump(stdout, "ciphertext:", ciphertext, tdata->ciphertext.len);
debug_hexdump(stdout, "auth tag:", auth_tag, tdata->auth_tag.len);
@@ -13878,8 +13891,9 @@ test_authenticated_decryption_oop(const struct
aead_test_data *tdata)
struct crypto_testsuite_params *ts_params = &testsuite_params;
struct crypto_unittest_params *ut_params = &unittest_params;
+ uint32_t i;
int retval;
- uint8_t *plaintext;
+ uint8_t *plaintext, *buffer_oop;
struct rte_cryptodev_info dev_info;
rte_cryptodev_info_get(ts_params->valid_devs[0], &dev_info);
@@ -13957,6 +13971,17 @@ test_authenticated_decryption_oop(const struct
aead_test_data *tdata)
debug_hexdump(stdout, "plaintext:", plaintext, tdata->ciphertext.len);
+ /* Check if the data within the offset range is not overwritten in the
OOP */
+ buffer_oop = rte_pktmbuf_mtod(ut_params->obuf, uint8_t *);
+ for (i = 0; i < ut_params->op->sym->cipher.data.offset; i++) {
+ if (buffer_oop[i]) {
+ RTE_LOG(ERR, USER1,
+ "Incorrect value of the output buffer
header\n");
+ debug_hexdump(stdout, "Incorrect value:", buffer_oop,
+ ut_params->op->sym->cipher.data.offset);
+ return TEST_FAILED;
+ }
+ }
/* Validate obuf */
TEST_ASSERT_BUFFERS_ARE_EQUAL(
plaintext,
--
2.43.0
RE: [PATCH 3/3] net/mlx5: allow FDB RSS
> -Original Message-
> From: Suanming Mou
> Sent: Tuesday, February 25, 2025 8:45 AM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: [PATCH 3/3] net/mlx5: allow FDB RSS
>
> RSS can be used in FDB Rx rules when JUMP_FDB_RX action to allow jump from
> FDB Tx to FDB Rx table.
>
> Different with NIC RSS, FDB RSS will not do the internal implicit metadata
> copy.
>
> This commit enables the FDB RSS if JUMP_FDB_RX is supported.
>
> Signed-off-by: Suanming Mou
> ---
> drivers/net/mlx5/mlx5_flow.c| 4 ++--
> drivers/net/mlx5/mlx5_flow_dv.c | 32 +---
> drivers/net/mlx5/mlx5_flow_hw.c | 19 ++-
> 3 files changed, 33 insertions(+), 22 deletions(-)
>
> diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
> index 3fbe89a9d4..9c6a4f39fb 100644
> --- a/drivers/net/mlx5/mlx5_flow.c
> +++ b/drivers/net/mlx5/mlx5_flow.c
> @@ -1977,9 +1977,9 @@ mlx5_flow_validate_action_mark(struct rte_eth_dev
> *dev,
> RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> "mark action not supported for "
> "egress");
> - if (attr->transfer && mlx5_hws_active(dev))
> + if (attr->transfer && !mlx5_hws_active(dev))
> return rte_flow_error_set(error, ENOTSUP,
> - RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
> + RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
> NULL,
> "non-template mark action not
> supported
> for transfer");
> return 0;
> }
> diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
> b/drivers/net/mlx5/mlx5_flow_dv.c index 633c41e358..61d3101ce8 100644
> --- a/drivers/net/mlx5/mlx5_flow_dv.c
> +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> @@ -8939,21 +8939,23 @@ flow_dv_validate(struct rte_eth_dev *dev, const
> struct rte_flow_attr *attr,
> RTE_FLOW_ERROR_TYPE_ACTION,
> NULL,
> "unsupported action MARK");
> - if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> - return rte_flow_error_set(error, ENOTSUP,
> - RTE_FLOW_ERROR_TYPE_ACTION,
> - NULL,
> - "unsupported action QUEUE");
> - if (action_flags & MLX5_FLOW_ACTION_RSS)
> - return rte_flow_error_set(error, ENOTSUP,
> - RTE_FLOW_ERROR_TYPE_ACTION,
> - NULL,
> - "unsupported action RSS");
> - if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> - return rte_flow_error_set(error, EINVAL,
> - RTE_FLOW_ERROR_TYPE_ACTION,
> - actions,
> - "no fate action is found");
> + if (!priv->jump_fdb_rx_en) {
> + if (action_flags & MLX5_FLOW_ACTION_QUEUE)
> + return rte_flow_error_set(error, ENOTSUP,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + NULL,
> + "unsupported action
> QUEUE");
> + if (action_flags & MLX5_FLOW_ACTION_RSS)
> + return rte_flow_error_set(error, ENOTSUP,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + NULL,
> + "unsupported action
> RSS");
> + if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
> + return rte_flow_error_set(error, EINVAL,
> +
> RTE_FLOW_ERROR_TYPE_ACTION,
> + actions,
> + "no fate action is
> found");
> + }
> } else {
> if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
> return rte_flow_error_set(error, EINVAL, diff --git
> a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c index
> 83f55ed3e8..a063e5ac9c 100644
> --- a/drivers/net/mlx5/mlx5_flow_hw.c
> +++ b/drivers/net/mlx5/mlx5_flow_hw.c
> @@ -7026,6 +7026,7 @@ mlx5_hw_validate_action_queue(struct rte_eth_dev
> *dev,
> uint64_t action_flags,
>
RE: [PATCH 2/3] net/mlx5: add jump FDB Rx flag
Hi
> -Original Message-
> From: Suanming Mou
> Sent: Tuesday, February 25, 2025 8:45 AM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: [PATCH 2/3] net/mlx5: add jump FDB Rx flag
>
> When jump FDB Rx is supported, flow will be able to jump from FDB Tx to
> FDB Rx, in that case the dest action in FDB Rx table should support FDB Tx
> as well.
>
> Signed-off-by: Suanming Mou
> ---
> drivers/common/mlx5/mlx5_devx_cmds.c | 8
> drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
> drivers/net/mlx5/linux/mlx5_os.c | 1 +
> drivers/net/mlx5/mlx5.h | 1 +
> drivers/net/mlx5/mlx5_flow_hw.c | 8 +++-
> 5 files changed, 18 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
> b/drivers/common/mlx5/mlx5_devx_cmds.c
> index f504b29f31..eb8553e8ad 100644
> --- a/drivers/common/mlx5/mlx5_devx_cmds.c
> +++ b/drivers/common/mlx5/mlx5_devx_cmds.c
> @@ -924,6 +924,7 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
> uint32_t out[MLX5_ST_SZ_DW(query_hca_cap_out)] = {0};
> bool hca_cap_2_sup;
> uint64_t general_obj_types_supported = 0;
> + uint64_t stc_action_type_127_64;
> void *hcattr;
> int rc, i;
>
> @@ -1352,6 +1353,13 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
> attr->fdb_unified_en = MLX5_GET(wqe_based_flow_table_cap,
> hcattr,
> fdb_unified_en);
> + stc_action_type_127_64 = MLX5_GET64(wqe_based_flow_table_cap,
> + hcattr,
> + stc_action_type_127_64);
> + if (stc_action_type_127_64 &
> +(1 <<
> (MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX -
> + MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX)))
> + attr->jump_fdb_rx_en = true;
I assume integer value 1 is enough, since it is a bit field, no need to use
bool type.
> }
> /* Query HCA attribute for ROCE. */
> if (attr->roce) {
> diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h
> b/drivers/common/mlx5/mlx5_devx_cmds.h
> index 8de4210fb2..6c726a0d46 100644
> --- a/drivers/common/mlx5/mlx5_devx_cmds.h
> +++ b/drivers/common/mlx5/mlx5_devx_cmds.h
> @@ -326,6 +326,7 @@ struct mlx5_hca_attr {
> uint32_t lag_rx_port_affinity:1;
> uint32_t wqe_based_flow_table_sup:1;
> uint32_t fdb_unified_en:1;
> + uint32_t jump_fdb_rx_en:1;
> uint8_t max_header_modify_pattern_length;
> uint64_t system_image_guid;
> uint32_t log_max_conn_track_offload:5; diff --git
> a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
> index 9410211e3b..4e64026137 100644
> --- a/drivers/net/mlx5/linux/mlx5_os.c
> +++ b/drivers/net/mlx5/linux/mlx5_os.c
> @@ -1718,6 +1718,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> priv->unified_fdb_en = !!priv->master;
> DRV_LOG(DEBUG, "port %u: unified FDB %s enabled.",
> eth_dev->data->port_id, priv->unified_fdb_en ? "is" :
> "isn't");
> + priv->jump_fdb_rx_en = sh->cdev-
> >config.hca_attr.jump_fdb_rx_en;
Can we add some INFO or DEBUG level log to indicate this to the user?
> if (priv->sh->config.dv_esw_en) {
> uint32_t usable_bits;
> uint32_t required_bits;
> diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h index
> f73f6e63ff..545ba48b3c 100644
> --- a/drivers/net/mlx5/mlx5.h
> +++ b/drivers/net/mlx5/mlx5.h
> @@ -1987,6 +1987,7 @@ struct mlx5_priv {
> uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
> uint32_t tunnel_enabled:1; /* If tunnel offloading is enabled on
> rxqs. */
> uint32_t unified_fdb_en:1; /* Unified FDB flag per port. */
> + uint32_t jump_fdb_rx_en:1; /* Jump from FDB Tx to FDB Rx flag per
> +port. */
> uint16_t domain_id; /* Switch domain identifier. */
> uint16_t vport_id; /* Associated VF vport index (if any). */
> uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG.
> */ diff --git a/drivers/net/mlx5/mlx5_flow_hw.c
> b/drivers/net/mlx5/mlx5_flow_hw.c index f0888dbf0e..83f55ed3e8 100644
> --- a/drivers/net/mlx5/mlx5_flow_hw.c
> +++ b/drivers/net/mlx5/mlx5_flow_hw.c
> @@ -9322,6 +9322,7 @@ flow_hw_grp_create_cb(void *tool_ctx, void *cb_ctx)
> struct mlx5_flow_group *grp_data;
> struct mlx5dr_table *tbl = NULL;
> struct mlx5dr_action *jump;
> + uint32_t hws_flags;
> uint32_t idx = 0;
> MKSTR(matcher_name, "%s_%s_%u_%u_matcher_list",
> attr->transfer ? "FDB" : "NIC", attr->egress ? "egress" :
> "ingress", @@ -9342,10 +9343,15 @@ flow_hw_grp_create_cb(void *tool_ctx,
> void *cb_ctx)
> goto error;
> grp_data->tbl = tbl;
> if (attr->group) {
>
Re: [PATCH 1/2] doc: fix MPLS limitations under HWS
Hi,
From: Maayan Kashani
Sent: Wednesday, February 26, 2025 10:02 AM
To: dev@dpdk.org
Cc: Maayan Kashani; Dariusz Sosnowski; Raslan Darawsheh; sta...@dpdk.org; Slava
Ovsiienko; Bing Zhao; Ori Kam; Suanming Mou; Matan Azrad; Michael Baum
Subject: [PATCH 1/2] doc: fix MPLS limitations under HWS
Update missing limitations.
Fixes: 1e2a51f545b6 ("net/mlx5: support MPLS tunnel with HWS")
Cc: sta...@dpdk.org
Signed-off-by: Maayan Kashani
Acked-by: Dariusz Sosnowski
series squashed and applied to next-net-mlx,
Kindest regards,
Raslan Darawsheh
RE: [PATCH V2 4/4] net/mlx5/hws: allow different types in miss validation
> -Original Message-
> From: Ophir Munk
> Sent: Wednesday, February 26, 2025 9:39 AM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Suanming Mou ; Matan Azrad
>
> Cc: dev@dpdk.org; Raslan Darawsheh
> Subject: [PATCH V2 4/4] net/mlx5/hws: allow different types in miss validation
>
> In validating Jump from group A to default miss group B DPDK requests that
> both
> groups types will be identical. In practice jumping from FDB_UNIFIED to FDB_RX
> or FDB_TX group types (or vice versa) is allowed.
> This commit enables this jumping between the different types.
>
> Fixes: b81f95ca770d ("net/mlx5/hws: support default miss table")
>
> Signed-off-by: Ophir Munk
Acked-by: Dariusz Sosnowski
Best regards,
Dariusz Sosnowski
[PATCH] crypto/qat: fix incorrect sm3 state size
This commit fixes incorrectly set state size within
the QAT PMD.
Fixes: 75fd4bbc94ab ("crypto/qat: support SM3 hash algorithm")
Cc: sta...@dpdk.org
Signed-off-by: Arkadiusz Kusztal
---
drivers/crypto/qat/qat_sym_session.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/crypto/qat/qat_sym_session.c
b/drivers/crypto/qat/qat_sym_session.c
index 9b26f8a89b..7f370f03fb 100644
--- a/drivers/crypto/qat/qat_sym_session.c
+++ b/drivers/crypto/qat/qat_sym_session.c
@@ -2473,7 +2473,8 @@ int qat_sym_cd_auth_set(struct qat_sym_session *cdesc,
qat_hash_get_block_size(cdesc->qat_hash_alg) >> 3;
auth_param->u2.inner_prefix_sz =
qat_hash_get_block_size(cdesc->qat_hash_alg);
- auth_param->hash_state_sz = digestsize;
+ auth_param->hash_state_sz = (hash_cd_ctrl->outer_prefix_sz +
+ auth_param->u2.inner_prefix_sz) >> 3;
if (qat_dev_gen == QAT_GEN4 || qat_dev_gen == QAT_GEN5 ||
qat_dev_gen == QAT_VQAT) {
ICP_QAT_FW_HASH_FLAG_MODE2_SET(
--
2.43.0
RE: [PATCH v2 05/14] net/ipn3ke: remove unnecessary dereference
Hi, > -Original Message- > From: Stephen Hemminger > Sent: Thursday, February 20, 2025 10:32 AM > To: dev@dpdk.org > Cc: Stephen Hemminger ; Xu, Rosen > > Subject: [PATCH v2 05/14] net/ipn3ke: remove unnecessary dereference > > Since it is a function pointer, the expression *hw->f_mac_read and > hw->f_mac_read are equivalent in C. > > Remove excession indentation here. > > Signed-off-by: Stephen Hemminger > --- > drivers/net/intel/ipn3ke/ipn3ke_ethdev.c | 27 +--- > 1 file changed, 5 insertions(+), 22 deletions(-) > > diff --git a/drivers/net/intel/ipn3ke/ipn3ke_ethdev.c > b/drivers/net/intel/ipn3ke/ipn3ke_ethdev.c > index baae80d661..c8f06a06cb 100644 > --- a/drivers/net/intel/ipn3ke/ipn3ke_ethdev.c > +++ b/drivers/net/intel/ipn3ke/ipn3ke_ethdev.c > @@ -229,34 +229,17 @@ ipn3ke_mtu_set(struct ipn3ke_hw *hw, uint32_t > mac_num, > uint32_t rx; > uint32_t tmp; > > - if (!(*hw->f_mac_read) || !(*hw->f_mac_write)) > + if (!hw->f_mac_read || !hw->f_mac_write) > return; > > - (*hw->f_mac_read)(hw, > - &tx, > - txaddr, > - mac_num, > - eth_group_sel); > + hw->f_mac_read(hw, &tx, txaddr, mac_num, eth_group_sel); > > - (*hw->f_mac_read)(hw, > - &rx, > - rxaddr, > - mac_num, > - eth_group_sel); > + hw->f_mac_read(hw, &rx, rxaddr, mac_num, eth_group_sel); > > tmp = ipn3ke_mtu_cal(tx, rx); > > - (*hw->f_mac_write)(hw, > - tmp, > - txaddr, > - mac_num, > - eth_group_sel); > - > - (*hw->f_mac_write)(hw, > - tmp, > - rxaddr, > - mac_num, > - eth_group_sel); > + hw->f_mac_write(hw, tmp, txaddr, mac_num, eth_group_sel); > + hw->f_mac_write(hw, tmp, rxaddr, mac_num, eth_group_sel); > } > > static void > -- > 2.47.2 Reviewed-by: Rosen Xu
Re: [PATCH v2 5/5] acl: make compatible with instruction set updates for MSVC
On Tue, Feb 25, 2025 at 05:06:56PM -0800, Andre Muezerie wrote: > Top level 'cc_avx2_flags' was created and holds the correct flags > depending on the compiler used. > > Signed-off-by: Andre Muezerie > --- Acked-by: Bruce Richardson This patch can probably be merged with the patch introducing the new build variable. It and the previous-but-one patch are both very small and are probably better merged.
Re: [PATCH v2 4/5] drivers/net: make compatible with instruction set updates for MSVC
On Tue, Feb 25, 2025 at 05:06:55PM -0800, Andre Muezerie wrote: > Top level 'cc_avx2_flags' was created and holds the correct flags > depending on the compiler used. > > Signed-off-by: Andre Muezerie > --- Acked-by: Bruce Richardson
RE: [PATCH] net/mlx5: support ESP in non-template RSS expansion
> -Original Message- > From: Gregory Etelson > Sent: Monday, February 17, 2025 8:13 AM > To: dev@dpdk.org > Cc: Gregory Etelson ; Maayan Kashani > ; Raslan Darawsheh > Subject: [PATCH] net/mlx5: support ESP in non-template RSS expansion > > External email: Use caution opening links or attachments > > > Add support for the ESP protocol in non-template RSS expansion. > > Signed-off-by: Gregory Etelson Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
[PATCH v16 22/29] net/rnp: add supported packet types
add support parse hw packet types result.
Signed-off-by: Wenbo Cao
---
doc/guides/nics/features/rnp.ini | 1 +
doc/guides/nics/rnp.rst | 2 ++
drivers/net/rnp/base/rnp_bdq_if.h | 4 +++
drivers/net/rnp/rnp_ethdev.c | 31 +
drivers/net/rnp/rnp_rxtx.c| 45 +++
5 files changed, 83 insertions(+)
diff --git a/doc/guides/nics/features/rnp.ini b/doc/guides/nics/features/rnp.ini
index 00b23b0fa2..eb1c27a3d3 100644
--- a/doc/guides/nics/features/rnp.ini
+++ b/doc/guides/nics/features/rnp.ini
@@ -7,6 +7,7 @@
Speed capabilities = Y
Link status = Y
Link status event= Y
+Packet type parsing = Y
Basic stats = Y
Stats per queue = Y
Extended stats = Y
diff --git a/doc/guides/nics/rnp.rst b/doc/guides/nics/rnp.rst
index ea3270a758..58f4f4a3a8 100644
--- a/doc/guides/nics/rnp.rst
+++ b/doc/guides/nics/rnp.rst
@@ -55,6 +55,7 @@ Features
- Jumbo frames
- Scatter-Gather IO support
- Port hardware statistic
+- Packet type parsing
Prerequisites and Pre-conditions
@@ -97,6 +98,7 @@ Listed below are the rte_eth functions supported:
* ``rte_eth_dev_default_mac_addr_set``
* ``rte_eth_dev_mac_addr_add``
* ``rte_eth_dev_mac_addr_remove``
+* ``rte_eth_dev_get_supported_ptypes``
* ``rte_eth_promiscuous_disable``
* ``rte_eth_promiscuous_enable``
* ``rte_eth_allmulticast_enable``
diff --git a/drivers/net/rnp/base/rnp_bdq_if.h
b/drivers/net/rnp/base/rnp_bdq_if.h
index 61a38323e5..a7d27bdebb 100644
--- a/drivers/net/rnp/base/rnp_bdq_if.h
+++ b/drivers/net/rnp/base/rnp_bdq_if.h
@@ -73,6 +73,7 @@ struct rnp_tx_desc {
#define RNP_RX_L3TYPE_IPV4 (0x00UL << RNP_RX_L3TYPE_S)
#define RNP_RX_L3TYPE_IPV6 (0x01UL << RNP_RX_L3TYPE_S)
#define RNP_RX_L4TYPE_S(6)
+#define RNP_RX_L4TYPE_MASK RTE_GENMASK32(7, 6)
#define RNP_RX_L4TYPE_TCP (0x01UL << RNP_RX_L4TYPE_S)
#define RNP_RX_L4TYPE_SCTP (0x02UL << RNP_RX_L4TYPE_S)
#define RNP_RX_L4TYPE_UDP (0x03UL << RNP_RX_L4TYPE_S)
@@ -83,9 +84,12 @@ struct rnp_tx_desc {
#define RNP_RX_IN_L3_ERR RTE_BIT32(11)
#define RNP_RX_IN_L4_ERR RTE_BIT32(12)
#define RNP_RX_TUNNEL_TYPE_S (13)
+#define RNP_RX_TUNNEL_MASK RTE_GENMASK32(14, 13)
#define RNP_RX_PTYPE_VXLAN (0x01UL << RNP_RX_TUNNEL_TYPE_S)
#define RNP_RX_PTYPE_NVGRE (0x02UL << RNP_RX_TUNNEL_TYPE_S)
#define RNP_RX_PTYPE_VLAN RTE_BIT32(15)
+/* mark_data */
+#define RNP_RX_L3TYPE_VALIDRTE_BIT32(31)
/* tx data cmd */
#define RNP_TX_TSO_EN RTE_BIT32(4)
#define RNP_TX_L3TYPE_S(5)
diff --git a/drivers/net/rnp/rnp_ethdev.c b/drivers/net/rnp/rnp_ethdev.c
index 0ce9fa52e9..289d9ed867 100644
--- a/drivers/net/rnp/rnp_ethdev.c
+++ b/drivers/net/rnp/rnp_ethdev.c
@@ -1186,6 +1186,36 @@ rnp_dev_mac_addr_remove(struct rte_eth_dev *dev,
rnp_clear_macaddr(port, index);
}
+static uint32_t *rnp_support_ptypes_get(void)
+{
+ static uint32_t ptypes[] = {
+ RTE_PTYPE_L2_ETHER,
+ RTE_PTYPE_L2_ETHER_TIMESYNC,
+ RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
+ RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
+ RTE_PTYPE_L4_TCP,
+ RTE_PTYPE_L4_UDP,
+ RTE_PTYPE_L4_SCTP,
+ RTE_PTYPE_TUNNEL_VXLAN,
+ RTE_PTYPE_TUNNEL_GRE,
+ RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
+ RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
+ RTE_PTYPE_INNER_L4_TCP,
+ RTE_PTYPE_INNER_L4_UDP,
+ RTE_PTYPE_INNER_L4_SCTP,
+ RTE_PTYPE_UNKNOWN,
+ };
+
+ return ptypes;
+}
+
+static const uint32_t *
+rnp_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused,
+size_t *no_of_elements __rte_unused)
+{
+ return rnp_support_ptypes_get();
+}
+
/* Features supported by this driver */
static const struct eth_dev_ops rnp_eth_dev_ops = {
.dev_configure= rnp_dev_configure,
@@ -1228,6 +1258,7 @@ static const struct eth_dev_ops rnp_eth_dev_ops = {
.mac_addr_set = rnp_dev_mac_addr_set,
.mac_addr_add = rnp_dev_mac_addr_add,
.mac_addr_remove = rnp_dev_mac_addr_remove,
+ .dev_supported_ptypes_get = rnp_dev_supported_ptypes_get,
};
static void
diff --git a/drivers/net/rnp/rnp_rxtx.c b/drivers/net/rnp/rnp_rxtx.c
index 8d67e04990..dd8cde8aff 100644
--- a/drivers/net/rnp/rnp_rxtx.c
+++ b/drivers/net/rnp/rnp_rxtx.c
@@ -639,6 +639,49 @@ int rnp_rx_queue_start(struct rte_eth_dev *eth_dev,
uint16_t qidx)
return 0;
}
+static __rte_always_inline void
+rnp_dev_rx_parse(struct rnp_rx_queue *rxq __rte_unused,
+struct rte_mbuf *m,
+volatile struct rnp_rx_desc rxbd)
+{
+ uint32_t mark_data = rxbd.wb.qword0.mark_data;
+ uint16_t vlan_tci = rxbd.wb.qword1.vlan_tci;
+ uint32_t cmd = rxbd.wb.q
[PATCH v16 17/29] net/rnp: add Rx scatter segment version
add support scatter multi segment received.
Signed-off-by: Wenbo Cao
---
doc/guides/nics/features/rnp.ini | 1 +
doc/guides/nics/rnp.rst | 2 +
drivers/net/rnp/rnp_ethdev.c | 2 +
drivers/net/rnp/rnp_rxtx.c | 130 ++-
drivers/net/rnp/rnp_rxtx.h | 2 +
5 files changed, 135 insertions(+), 2 deletions(-)
diff --git a/doc/guides/nics/features/rnp.ini b/doc/guides/nics/features/rnp.ini
index 6d13370e8f..6453762745 100644
--- a/doc/guides/nics/features/rnp.ini
+++ b/doc/guides/nics/features/rnp.ini
@@ -15,5 +15,6 @@ RSS hash = Y
RSS key update = Y
RSS reta update = Y
Inner RSS= Y
+Scattered Rx = Y
Linux= Y
x86-64 = Y
diff --git a/doc/guides/nics/rnp.rst b/doc/guides/nics/rnp.rst
index 6aa8b00af6..ebf6971140 100644
--- a/doc/guides/nics/rnp.rst
+++ b/doc/guides/nics/rnp.rst
@@ -51,6 +51,8 @@ Features
- Promiscuous mode
- Link state information
- MTU update
+- Jumbo frames
+- Scatter-Gather IO support
Prerequisites and Pre-conditions
diff --git a/drivers/net/rnp/rnp_ethdev.c b/drivers/net/rnp/rnp_ethdev.c
index 4afdccd0ab..3dfc7ef97b 100644
--- a/drivers/net/rnp/rnp_ethdev.c
+++ b/drivers/net/rnp/rnp_ethdev.c
@@ -606,6 +606,8 @@ static int rnp_dev_infos_get(struct rte_eth_dev *eth_dev,
dev_info->reta_size = RNP_RSS_INDIR_SIZE;
/* speed cap info */
dev_info->speed_capa = rnp_get_speed_caps(eth_dev);
+ /* rx support offload cap */
+ dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_SCATTER;
/* default ring configure */
dev_info->default_rxportconf.burst_size = 32;
dev_info->default_txportconf.burst_size = 32;
diff --git a/drivers/net/rnp/rnp_rxtx.c b/drivers/net/rnp/rnp_rxtx.c
index 260c34113a..b2ca9b15fc 100644
--- a/drivers/net/rnp/rnp_rxtx.c
+++ b/drivers/net/rnp/rnp_rxtx.c
@@ -826,7 +826,6 @@ rnp_xmit_simple(void *_txq, struct rte_mbuf **tx_pkts,
uint16_t nb_pkts)
if (txq->tx_next_rs > txq->attr.nb_desc)
txq->tx_next_rs = txq->tx_rs_thresh - 1;
}
-
txq->tx_tail = i;
rte_wmb();
@@ -835,9 +834,136 @@ rnp_xmit_simple(void *_txq, struct rte_mbuf **tx_pkts,
uint16_t nb_pkts)
return start;
}
+static int
+rnp_rxq_bulk_alloc(struct rnp_rx_queue *rxq,
+ volatile struct rnp_rx_desc *rxbd,
+ struct rnp_rxsw_entry *rxe,
+ bool bulk_alloc)
+{
+ struct rte_eth_dev_data *data;
+ struct rte_mbuf *nmb = NULL;
+ uint16_t update_tail;
+
+ if (!bulk_alloc) {
+ data = rte_eth_devices[rxq->attr.port_id].data;
+ nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
+ if (unlikely(!nmb)) {
+ data->rx_mbuf_alloc_failed++;
+ return -ENOMEM;
+ }
+ rxbd->d.pkt_addr = 0;
+ rxbd->d.cmd = 0;
+ rxe->mbuf = NULL;
+ rxe->mbuf = nmb;
+ rxbd->d.pkt_addr = rnp_get_dma_addr(&rxq->attr, nmb);
+ }
+ if (rxq->rxrearm_nb > rxq->rx_free_thresh) {
+ rxq->rxrearm_nb -= rxq->rx_free_thresh;
+ rxq->rxrearm_start += rxq->rx_free_thresh;
+ if (rxq->rxrearm_start >= rxq->attr.nb_desc)
+ rxq->rxrearm_start = 0;
+ update_tail = (uint16_t)((rxq->rxrearm_start == 0) ?
+ (rxq->attr.nb_desc - 1) : (rxq->rxrearm_start -
1));
+ rte_io_wmb();
+ RNP_REG_WR(rxq->rx_tailreg, 0, update_tail);
+ }
+
+ return 0;
+}
+
+static __rte_always_inline uint16_t
+rnp_scattered_rx(void *rx_queue, struct rte_mbuf **rx_pkts,
+uint16_t nb_pkts)
+{
+ struct rnp_rx_queue *rxq = (struct rnp_rx_queue *)rx_queue;
+ volatile struct rnp_rx_desc *bd_ring = rxq->rx_bdr;
+ struct rte_mbuf *first_seg = rxq->pkt_first_seg;
+ struct rte_mbuf *last_seg = rxq->pkt_last_seg;
+ struct rnp_rxsw_entry *sw_ring = rxq->sw_ring;
+ volatile struct rnp_rx_desc *rxbd;
+ volatile struct rnp_rx_desc rxd;
+ struct rnp_rxsw_entry *rxe;
+ struct rte_mbuf *rxm;
+ uint16_t rx_pkt_len;
+ uint16_t nb_rx = 0;
+ uint16_t rx_status;
+ uint16_t rx_id;
+
+ if (unlikely(!rxq->rxq_started || !rxq->rx_link))
+ return 0;
+ rx_id = rxq->rx_tail;
+ while (nb_rx < nb_pkts) {
+ rxbd = &bd_ring[rx_id];
+ rx_status = rxbd->wb.qword1.cmd;
+ if (!(rx_status & rte_cpu_to_le_16(RNP_CMD_DD)))
+ break;
+ rte_atomic_thread_fence(rte_memory_order_acquire);
+ rxd = *rxbd;
+ rxe = &sw_ring[rx_id];
+ rxm = rxe->mbuf;
+ if (rnp_rxq_bulk_alloc(rxq, rxbd, rxe, false))
+ break;
RE: [PATCH 2/2] net/mlx5/hws: add DROP action to multi destination actions
> -Original Message- > From: Gregory Etelson > Sent: Thursday, February 13, 2025 9:33 AM > To: dev@dpdk.org > Cc: Gregory Etelson ; Maayan Kashani > ; Raslan Darawsheh ; Tamar > Mashiah ; sta...@dpdk.org > Subject: [PATCH 2/2] net/mlx5/hws: add DROP action to multi destination > actions > > External email: Use caution opening links or attachments > > > From: Tamar Mashiah > > Add DROP action to multi destination actions, set NOP destination type when > DROP is specified. > > cc: sta...@dpdk.org > > Signed-off-by: Tamar Mashiah Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
RE: [PATCH 1/2] net/mlx5: add DROP destination to mirror clone actions
> -Original Message- > From: Gregory Etelson > Sent: Thursday, February 13, 2025 9:33 AM > To: dev@dpdk.org > Cc: Gregory Etelson ; Maayan Kashani > ; Raslan Darawsheh ; Tamar > Mashiah ; sta...@dpdk.org > Subject: [PATCH 1/2] net/mlx5: add DROP destination to mirror clone actions > > External email: Use caution opening links or attachments > > > From: Tamar Mashiah > > Add DROP destination to mirror clone actions. > > cc: sta...@dpdk.org > > Signed-off-by: Tamar Mashiah Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
RE: [PATCH] net/mlx5/hws: change error flow on matcher disconnect
> -Original Message- > From: Gregory Etelson > Sent: Sunday, February 9, 2025 2:48 PM > To: dev@dpdk.org > Cc: Gregory Etelson ; Maayan Kashani > ; Raslan Darawsheh ; Itamar > Gozlan > Subject: [PATCH] net/mlx5/hws: change error flow on matcher disconnect > > External email: Use caution opening links or attachments > > > From: Itamar Gozlan > > Currently, when a firmware (FW) failure occurs during matcher error flow, the > system attempts to reconnect the matcher, leading to a segmentation fault. > This > happens because the matcher is freed but remains in the list. > > Example scenario: > Given matchers M1->M2->M3, if a FW failure occurs: > 1. System tries to destroy M1 and fails > 2. M1 remains in the list but is freed > 3. When destroying M2, it attempts to remove itself and create M1->M3 4. This > results in a segmentation fault as M1 is already freed > > Signed-off-by: Itamar Gozlan Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
[PATCH V2 0/4] mlx5 unified fdb
V1: initial version V2: rebase + determine unified fdb based on FW capability Bing Zhao (1): net/mlx5: remove unneeded FDB flag on representor action Ophir Munk (3): common/mlx5: support FDB unified capability query net/mlx5: support FDB unified domain net/mlx5/hws: allow different types in miss validation drivers/common/mlx5/mlx5_devx_cmds.c | 3 + drivers/common/mlx5/mlx5_devx_cmds.h | 1 + drivers/net/mlx5/hws/mlx5dr_table.c | 18 +- drivers/net/mlx5/linux/mlx5_os.c | 11 + drivers/net/mlx5/mlx5.h | 8 + drivers/net/mlx5/mlx5_flow_hw.c | 521 --- drivers/net/mlx5/mlx5_flow_meter.c | 16 +- drivers/net/mlx5/mlx5_flow_quota.c | 9 +- drivers/net/mlx5/mlx5_hws_cnt.c | 9 +- 9 files changed, 429 insertions(+), 167 deletions(-) -- 2.8.4
[PATCH V2 1/4] common/mlx5: support FDB unified capability query
This commit queries the FW for the new unified FDB mode and saves it in
mlx5 shared device as fdb_unified_en bit.
Signed-off-by: Ophir Munk
---
drivers/common/mlx5/mlx5_devx_cmds.c | 3 +++
drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
2 files changed, 4 insertions(+)
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c
b/drivers/common/mlx5/mlx5_devx_cmds.c
index bba00a9..f504b29 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.c
+++ b/drivers/common/mlx5/mlx5_devx_cmds.c
@@ -1349,6 +1349,9 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
attr->max_header_modify_pattern_length =
MLX5_GET(wqe_based_flow_table_cap,
hcattr,
max_header_modify_pattern_length);
+ attr->fdb_unified_en = MLX5_GET(wqe_based_flow_table_cap,
+ hcattr,
+ fdb_unified_en);
}
/* Query HCA attribute for ROCE. */
if (attr->roce) {
diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h
b/drivers/common/mlx5/mlx5_devx_cmds.h
index 38548b4..8de4210 100644
--- a/drivers/common/mlx5/mlx5_devx_cmds.h
+++ b/drivers/common/mlx5/mlx5_devx_cmds.h
@@ -325,6 +325,7 @@ struct mlx5_hca_attr {
uint32_t cross_vhca:1;
uint32_t lag_rx_port_affinity:1;
uint32_t wqe_based_flow_table_sup:1;
+ uint32_t fdb_unified_en:1;
uint8_t max_header_modify_pattern_length;
uint64_t system_image_guid;
uint32_t log_max_conn_track_offload:5;
--
2.8.4
[PATCH V2 2/4] net/mlx5: support FDB unified domain
Legacy FDB domain is enhanced with optional three sub-domains: FDB_RX
FDB_TX and FDB_UNIFIED. Based on FW capability bit mlx5 will work in
either legacy or unified mode. To work in fdb unified mode the flow
attribute transfer bit must be set and the fdb sub domains are
determined by the table specialize flags SPECIALIZE_TRANSFER_WIRE_ORIG
and SPECIALIZE_TRANSFER_VPORT_ORIG.
++-+--+
| WIRE_ORIG | VPORT_ORIG | FDB domain |
++-+--+
| 0 | 0 | FDB_UNIFIED |
| 0 | 1 | FDB_TX |
| 1 | 0 | FDB_RX |
++-+--+
There can be no two groups which only differ in their fdb sub domain.
A table must be created in a group which has the same fdb sub domain.
Signed-off-by: Ophir Munk
---
drivers/net/mlx5/linux/mlx5_os.c | 11 +
drivers/net/mlx5/mlx5.h| 8 +
drivers/net/mlx5/mlx5_flow_hw.c| 519 ++---
drivers/net/mlx5/mlx5_flow_meter.c | 16 +-
drivers/net/mlx5/mlx5_flow_quota.c | 9 +-
drivers/net/mlx5/mlx5_hws_cnt.c| 9 +-
6 files changed, 409 insertions(+), 163 deletions(-)
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 091f288..7f13cb2 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1707,6 +1707,17 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
LIST_INIT(&priv->hw_ext_ctrl_flows);
if (priv->sh->config.dv_flow_en == 2) {
#ifdef HAVE_MLX5_HWS_SUPPORT
+ /*
+* Unified FDB flag is only needed for the actions created on
the transfer
+* port. proxy port. It is not needed on the following ports:
+* 1. NIC PF / VF / SF
+* 2. in Verbs or DV/DR mode
+* 3. with unsupported FW
+* 4. all representors in HWS
+*/
+ priv->unified_fdb_en = !!priv->master &&
sh->cdev->config.hca_attr.fdb_unified_en;
+ DRV_LOG(DEBUG, "port %u: unified FDB %s enabled.",
+ eth_dev->data->port_id, priv->unified_fdb_en ? "is" :
"isn't");
if (priv->sh->config.dv_esw_en) {
uint32_t usable_bits;
uint32_t required_bits;
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 2b590c9..f73f6e6 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1986,6 +1986,7 @@ struct mlx5_priv {
uint32_t mark_enabled:1; /* If mark action is enabled on rxqs. */
uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
uint32_t tunnel_enabled:1; /* If tunnel offloading is enabled on rxqs.
*/
+ uint32_t unified_fdb_en:1; /* Unified FDB flag per port. */
uint16_t domain_id; /* Switch domain identifier. */
uint16_t vport_id; /* Associated VF vport index (if any). */
uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
@@ -2226,6 +2227,13 @@ mlx5_is_port_on_mpesw_device(struct mlx5_priv *priv)
return priv->mpesw_port >= 0;
}
+static inline bool
+is_unified_fdb(const struct mlx5_priv *priv)
+{
+ /* Only needed on proxy port in E-Switch mode. */
+ return priv->unified_fdb_en;
+}
+
/* mlx5.c */
int mlx5_getenv_int(const char *);
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index e72b87d..1cc074d 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -282,12 +282,34 @@ mlx5_multi_pattern_reformat_index_to_type(uint32_t ix)
}
static inline enum mlx5dr_table_type
-get_mlx5dr_table_type(const struct rte_flow_attr *attr)
+get_mlx5dr_fdb_table_type(const struct rte_flow_attr *attr,
+ uint32_t specialize, bool fdb_unified_en)
+{
+ if (fdb_unified_en && !!attr->group) {
+ if ((specialize & (RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_WIRE_ORIG
|
+
RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_VPORT_ORIG)) == 0)
+ return MLX5DR_TABLE_TYPE_FDB_UNIFIED;
+ MLX5_ASSERT((specialize &
(RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_WIRE_ORIG |
+
RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_VPORT_ORIG)) !=
+
(RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_WIRE_ORIG |
+
RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_VPORT_ORIG));
+ if (specialize & RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_WIRE_ORIG)
+ return MLX5DR_TABLE_TYPE_FDB_RX;
+ if (specialize & RTE_FLOW_TABLE_SPECIALIZE_TRANSFER_VPORT_ORIG)
+ return MLX5DR_TABLE_TYPE_FDB_TX;
+ }
+
+ return MLX5DR_TABLE_TYPE_FDB;
+}
+
+static inline enum mlx5dr_table_type
+get_mlx5dr_table_type(const struct rte_flow_attr *attr, uint32_t
[PATCH V2 3/4] net/mlx5: remove unneeded FDB flag on representor action
From: Bing Zhao
The "send_to_kernel" action is shared and created during the HWS
configuration stage. By default, the NIC Rx and Tx flags are needed.
There is only one case that the FDB* flags are needed when creating
the action, the same as other shared actions. Only the transfer proxy
port needs the FDB* flags for shared actions.
Removing the unneeded FDB flags will help to save the resources of
memory and STCs.
Fixes: 07836ba156cb ("net/mlx5: fix device checking for send to kernel action")
Cc: jiaw...@nvidia.com
Signed-off-by: Bing Zhao
---
drivers/net/mlx5/mlx5_flow_hw.c | 14 ++
1 file changed, 6 insertions(+), 8 deletions(-)
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index 1cc074d..f0888db 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -9649,21 +9649,18 @@ _create_send_to_kernel_actions(struct mlx5_priv *priv,
int type)
}
static void
-flow_hw_create_send_to_kernel_actions(struct mlx5_priv *priv)
+flow_hw_create_send_to_kernel_actions(struct mlx5_priv *priv, bool is_proxy)
{
#ifdef HAVE_MLX5DV_DR_ACTION_CREATE_DEST_ROOT_TABLE
int i, from, to;
- bool is_vf_sf_dev = priv->sh->dev_cap.vf || priv->sh->dev_cap.sf;
bool unified_fdb = is_unified_fdb(priv);
for (i = MLX5DR_TABLE_TYPE_NIC_RX; i <= MLX5DR_TABLE_TYPE_NIC_TX; i++)
_create_send_to_kernel_actions(priv, i);
- if (priv->sh->config.dv_esw_en && !is_vf_sf_dev) {
- from = unified_fdb ? MLX5DR_TABLE_TYPE_FDB_RX :
- MLX5DR_TABLE_TYPE_FDB;
- to = unified_fdb ? MLX5DR_TABLE_TYPE_FDB_UNIFIED :
- MLX5DR_TABLE_TYPE_FDB;
+ if (is_proxy) {
+ from = unified_fdb ? MLX5DR_TABLE_TYPE_FDB_RX :
MLX5DR_TABLE_TYPE_FDB;
+ to = unified_fdb ? MLX5DR_TABLE_TYPE_FDB_UNIFIED :
MLX5DR_TABLE_TYPE_FDB;
for (i = from; i <= to; i++)
_create_send_to_kernel_actions(priv, i);
}
@@ -9674,6 +9671,7 @@ static void
flow_hw_destroy_send_to_kernel_action(struct mlx5_priv *priv)
{
int i;
+
for (i = MLX5DR_TABLE_TYPE_NIC_RX; i < MLX5DR_TABLE_TYPE_MAX; i++) {
if (priv->hw_send_to_kernel[i]) {
mlx5dr_action_destroy(priv->hw_send_to_kernel[i]);
@@ -12171,7 +12169,7 @@ __flow_hw_configure(struct rte_eth_dev *dev,
}
}
if (!priv->shared_host)
- flow_hw_create_send_to_kernel_actions(priv);
+ flow_hw_create_send_to_kernel_actions(priv, is_proxy);
if (port_attr->nb_conn_tracks || (host_priv && host_priv->hws_ctpool)) {
if (mlx5_flow_ct_init(dev, port_attr->nb_conn_tracks,
nb_q_updated))
goto err;
--
2.8.4
[PATCH V2 4/4] net/mlx5/hws: allow different types in miss validation
In validating Jump from group A to default miss group B DPDK requests
that both groups types will be identical. In practice jumping from
FDB_UNIFIED to FDB_RX or FDB_TX group types (or vice versa) is allowed.
This commit enables this jumping between the different types.
Fixes: b81f95ca770d ("net/mlx5/hws: support default miss table")
Signed-off-by: Ophir Munk
---
drivers/net/mlx5/hws/mlx5dr_table.c | 18 --
1 file changed, 16 insertions(+), 2 deletions(-)
diff --git a/drivers/net/mlx5/hws/mlx5dr_table.c
b/drivers/net/mlx5/hws/mlx5dr_table.c
index b371f42..c1c60b4 100644
--- a/drivers/net/mlx5/hws/mlx5dr_table.c
+++ b/drivers/net/mlx5/hws/mlx5dr_table.c
@@ -626,6 +626,19 @@ int mlx5dr_table_connect_to_miss_table(struct mlx5dr_table
*src_tbl,
return 0;
}
+static bool mlx5dr_table_set_default_miss_valid_types(enum mlx5dr_table_type
from,
+ enum mlx5dr_table_type to)
+{
+ if (from == to ||
+ ((from == MLX5DR_TABLE_TYPE_FDB_UNIFIED &&
+(to == MLX5DR_TABLE_TYPE_FDB_RX || to ==
MLX5DR_TABLE_TYPE_FDB_TX)) ||
+(to == MLX5DR_TABLE_TYPE_FDB_UNIFIED &&
+(from == MLX5DR_TABLE_TYPE_FDB_RX || from ==
MLX5DR_TABLE_TYPE_FDB_TX
+ return true;
+
+ return false;
+}
+
static int mlx5dr_table_set_default_miss_not_valid(struct mlx5dr_table *tbl,
struct mlx5dr_table
*miss_tbl)
{
@@ -637,8 +650,9 @@ static int mlx5dr_table_set_default_miss_not_valid(struct
mlx5dr_table *tbl,
}
if (mlx5dr_table_is_root(tbl) ||
- (miss_tbl && mlx5dr_table_is_root(miss_tbl)) ||
- (miss_tbl && miss_tbl->type != tbl->type)) {
+ (miss_tbl &&
+((mlx5dr_table_is_root(miss_tbl)) ||
+!mlx5dr_table_set_default_miss_valid_types(tbl->type,
miss_tbl->type {
DR_LOG(ERR, "Invalid arguments");
rte_errno = EINVAL;
return -rte_errno;
--
2.8.4
RE: [PATCH] net/mlx5: fix packet hardware ptype translation
> -Original Message-
> From: Gregory Etelson
> Sent: Sunday, February 9, 2025 3:01 PM
> To: dev@dpdk.org
> Cc: Gregory Etelson ; Maayan Kashani
> ; Raslan Darawsheh ; Slava
> Ovsiienko ; sta...@dpdk.org
> Subject: [PATCH] net/mlx5: fix packet hardware ptype translation
>
> External email: Use caution opening links or attachments
>
>
> From: Viacheslav Ovsiienko
>
> Since ConnectX-8 NIC the hardware recognizes L4 protocol in the IP fragments
> if
> the IP offset is zero and reports the L4 protocol type in CQE fields. This
> case was
> not supported in the ptype translation table cause not reporting the ptype in
> the
> mbufs.
>
> Fixes: ea16068c0064 ("net/mlx5: fix L4 packet type support")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Viacheslav Ovsiienko
Acked-by: Dariusz Sosnowski
Best regards,
Dariusz Sosnowski
Re: [PATCH v3] net/mlx5: add eCPRI support
Hi, From: Bing Zhao Sent: Wednesday, February 26, 2025 9:17 AM To: Slava Ovsiienko; dev@dpdk.org; Raslan Darawsheh Cc: Ori Kam; Dariusz Sosnowski; Suanming Mou; Matan Azrad; Hamdan Agbariya Subject: [PATCH v3] net/mlx5: add eCPRI support Support eCPRI item matching over L2 (ETHER / VLAN) in HWS, both for template API and backward compatibility API. Signed-off-by: Bing Zhao Acked-by: Dariusz Sosnowski --- v2: fix the checkpatch warning of unneeded else v3: fix single line if condition code style warning --- Patch applied to next-net-mlx, Kindest regards, Raslan Darawsheh
RE: [PATCH V2 1/4] common/mlx5: support FDB unified capability query
> -Original Message- > From: Ophir Munk > Sent: Wednesday, February 26, 2025 9:39 AM > To: Dariusz Sosnowski ; Slava Ovsiienko > ; Bing Zhao ; Ori Kam > ; Suanming Mou ; Matan Azrad > > Cc: dev@dpdk.org; Raslan Darawsheh > Subject: [PATCH V2 1/4] common/mlx5: support FDB unified capability query > > This commit queries the FW for the new unified FDB mode and saves it in > mlx5 shared device as fdb_unified_en bit. > > Signed-off-by: Ophir Munk Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
RE: [PATCH V2 2/4] net/mlx5: support FDB unified domain
> -Original Message- > From: Ophir Munk > Sent: Wednesday, February 26, 2025 9:39 AM > To: Dariusz Sosnowski ; Slava Ovsiienko > ; Bing Zhao ; Ori Kam > ; Suanming Mou ; Matan Azrad > > Cc: dev@dpdk.org; Raslan Darawsheh > Subject: [PATCH V2 2/4] net/mlx5: support FDB unified domain > > Legacy FDB domain is enhanced with optional three sub-domains: FDB_RX > FDB_TX and FDB_UNIFIED. Based on FW capability bit mlx5 will work in either > legacy or unified mode. To work in fdb unified mode the flow attribute > transfer > bit must be set and the fdb sub domains are determined by the table specialize > flags SPECIALIZE_TRANSFER_WIRE_ORIG and > SPECIALIZE_TRANSFER_VPORT_ORIG. > > ++-+--+ > | WIRE_ORIG | VPORT_ORIG | FDB domain | > ++-+--+ > | 0 | 0 | FDB_UNIFIED | > | 0 | 1 | FDB_TX | > | 1 | 0 | FDB_RX | > ++-+--+ > > There can be no two groups which only differ in their fdb sub domain. > A table must be created in a group which has the same fdb sub domain. > > Signed-off-by: Ophir Munk Acked-by: Dariusz Sosnowski Best regards, Dariusz Sosnowski
RE: [PATCH V2 3/4] net/mlx5: remove unneeded FDB flag on representor action
> -Original Message-
> From: Ophir Munk
> Sent: Wednesday, February 26, 2025 9:39 AM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Suanming Mou ; Matan Azrad
>
> Cc: dev@dpdk.org; Raslan Darawsheh ; Jiawei(Jonny)
> Wang
> Subject: [PATCH V2 3/4] net/mlx5: remove unneeded FDB flag on representor
> action
>
> From: Bing Zhao
>
> The "send_to_kernel" action is shared and created during the HWS configuration
> stage. By default, the NIC Rx and Tx flags are needed.
>
> There is only one case that the FDB* flags are needed when creating the
> action,
> the same as other shared actions. Only the transfer proxy port needs the FDB*
> flags for shared actions.
>
> Removing the unneeded FDB flags will help to save the resources of memory and
> STCs.
>
> Fixes: 07836ba156cb ("net/mlx5: fix device checking for send to kernel
> action")
> Cc: jiaw...@nvidia.com
>
> Signed-off-by: Bing Zhao
Acked-by: Dariusz Sosnowski
Best regards,
Dariusz Sosnowski
Re: [PATCH 3/6] config: allow faster instruction sets to be used with MSVC
On Tue, Feb 25, 2025 at 06:01:38PM -0800, Andre Muezerie wrote: > On Tue, Feb 25, 2025 at 02:28:02PM +, Bruce Richardson wrote: > > On Mon, Feb 24, 2025 at 01:01:16PM -0800, Andre Muezerie wrote: > > > Up to now MSVC has being used with the default mode, which uses SSE2 > > > instructions for scalar floating-point and vector calculations. > > > https://learn.microsoft.com/en-us/cpp/build/reference/arch-x64?view=msvc-170 > > > > > > This patch allows users to specify the CPU for which the generated > > > code should be optimized for in the same way it's done for GCC: by > > > passing the CPU name. > > > When no explicit CPU name is passed, 'native' is assumed (like it > > > happens with GCC) and the code will be optimized for the same CPU > > > type used to compile the code. > > > > > > MSVC does not provide this functionality natively, so logic was > > > added to meson.build to handle these differences, detecting which > > > instruction sets are supported by the CPU(s), passing the best > > > options to MSVC and setting the correct macros (like __AVX512F__) > > > so that the DPDK code can rely on them like it is done with GCC. > > > > > > Signed-off-by: Andre Muezerie > > > --- > > > > Hi Andre, > > > > couple of initial thoughts inline below. > > > > /Bruce > > > > > config/x86/meson.build | 364 - > > > 1 file changed, 325 insertions(+), 39 deletions(-) > > > > > > > There is quite a lot of new code to be added here. Might it be worthwhile > > creating a "config/x86/msvc/" subdirectory with its own meson.build file to > > handle all the complexities of using it. We can have the common material at > > the top of the x86/meson.build file, and then do > > > > if is_ms_compiler > > subdir(msvc) > > subdir_done() > > endif > > > > leaving the rest of the file for the gcc/clang/icx code. > > I think that makes sense, as there's not much common code there that is > common to gcc and msvc. > > > > > I really don't want to have tables like this to maintain in our code if at > > all possible. We used to have something a bit similar in DPDK IIRC, but we > > found it a maintenance nightmare and just switched to using the compiler to > > do all the work. In our existing builds, we just pass the > > cpu_instruction_set parameter straight to the -march flag of the compiler. > > For MSVC support, I believe we should just do the exact same. > > > > Maintaining lists like this will be a problem as new platforms need to be > > constantly added. > > It's great that when using gcc users can just pass the CPU type to it and > that it will set all the macros corresponding to that CPU for them. I > would love to be able to rely on MSVC for that as well, unfortunately MSVC > does not provide that level of granularity, that's why I came up with the > idea of having this table. > > Initially I was also very concerned about the amount of data to be stored > there, and the work required to maintain it. Then I decided to throw away > all the CPU types that do not have SSE4_2. That reduced the table in half. > Adding the entries manually was still a lot of work and error prone. So, > I decided to write some code that uses gcc to build that table for me. > I'll polish that code and add it to the patch. With that it will be almost > zero effort to maintain that table. All that will be required is running > that code on a setup with the latest gcc, providing a file with the CPU > names. Assuming gcc knows about the latest CPUs a new table will be > generated and can be pasted in the meson.build file. > > I assume this does not need to be done too often. In the worst case, if it > happens that DPDK was not updated with the latest CPUs, people can still > pick an earlier CPU with similar characteristics and have similar (if not > same) performance. > Thanks. Having it automated certainly helps, but I'm still not fully convinced that it's a good idea. If everyone else is happy, though, I'm ok to ignore those concerns :-). Let's see what others think. > > > Do we also look to backport them, because if equivalence > > with the linux build is necessary then that will have to be done - as on > > Linux when a new version of GCC comes out, we can then use the new > > instruction set targets on the old releases of DPDK. > > That would be nice, and I'm willing to help with that. It makes it a > better user experience if we can minimize the perceived differences > between the toolsets. I'm not sure if it's a requirement though. If > you're concerned that this would add too much overhead, it could be > decided that no such backport should happen. > > > > > > +if 'RTE_CPUFLAG_AVX512F' in compile_time_cpuflags > > > +machine_args += ['/arch:AVX512'] > > > +elif 'RTE_CPUFLAG_AVX2' in compile_time_cpuflags > > > +machine_args += ['/arch:AVX2'] > > > +elif 'RTE_CPUFLAG_AVX' in compile_time_cpuflags > > > +machine_args += ['/arch:AVX']
Re: [PATCH v2 1/5] eal: make compatible with instruction set updates for MSVC
On Tue, Feb 25, 2025 at 05:06:52PM -0800, Andre Muezerie wrote: > After the instruction set updates for MSVC the error below popped up: > > ../lib/eal/common/rte_random.c(6): fatal error C1083: > Cannot open include file: 'x86intrin.h': No such file or directory > > The fix is to not include header x86intrin.h with MSVC. > > Signed-off-by: Andre Muezerie > --- Acked-by: Bruce Richardson
Re: [PATCH v2 2/5] eal: only use numbers as align parameters for MSVC
On Tue, Feb 25, 2025 at 05:06:53PM -0800, Andre Muezerie wrote:
> After the instruction set updates for MSVC the error below popped up:
>
> ..\lib\eal\x86\include\rte_vect.h(82): error C2059: syntax error: '('
>
> The issue is that MSVC does not allow __rte_aligned(RTE_X86_ZMM_SIZE).
> It only accepts numbers that are power of 2. So, even though
> RTE_X86_ZMM_SIZE represents a number that is a power of two it cannot
> be used directly.
> https://learn.microsoft.com/en-us/cpp/cpp/align-cpp?view=msvc-170
>
> Signed-off-by: Andre Muezerie
> ---
Acked-by: Bruce Richardson
Re: [PATCH v2 3/5] config: create top level variable cc_avx2_flags
On Tue, Feb 25, 2025 at 05:06:54PM -0800, Andre Muezerie wrote: > Top level variable cc_avx2_flags was created and set according to the > compiler used. It is used in lower level directories. > > Signed-off-by: Andre Muezerie > --- > config/x86/meson.build | 6 ++ > 1 file changed, 6 insertions(+) > Acked-by: Bruce Richardson
RE: [PATCH v2 2/5] eal: only use numbers as align parameters for MSVC
> -Original Message-
> From: Andre Muezerie
> Sent: Wednesday, February 26, 2025 1:07 AM
> To: andre...@linux.microsoft.com
> Cc: dev@dpdk.org
> Subject: [PATCH v2 2/5] eal: only use numbers as align parameters for MSVC
>
> After the instruction set updates for MSVC the error below popped up:
>
> ..\lib\eal\x86\include\rte_vect.h(82): error C2059: syntax error: '('
>
> The issue is that MSVC does not allow __rte_aligned(RTE_X86_ZMM_SIZE).
> It only accepts numbers that are power of 2. So, even though
> RTE_X86_ZMM_SIZE represents a number that is a power of two it cannot
> be used directly.
> https://learn.microsoft.com/en-us/cpp/cpp/align-cpp?view=msvc-170
>
> Signed-off-by: Andre Muezerie
> ---
> lib/eal/x86/include/rte_vect.h | 11 ++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/lib/eal/x86/include/rte_vect.h b/lib/eal/x86/include/rte_vect.h
> index 70c78e9b77..0a51c539a4 100644
> --- a/lib/eal/x86/include/rte_vect.h
> +++ b/lib/eal/x86/include/rte_vect.h
> @@ -79,7 +79,16 @@ __extension__ ({\
> #define RTE_X86_ZMM_SIZE (sizeof(__m512i))
> #define RTE_X86_ZMM_MASK (RTE_X86_ZMM_SIZE - 1)
>
> -typedef union __rte_aligned(RTE_X86_ZMM_SIZE) __rte_x86_zmm {
> +/*
> + * MSVC does not allow __rte_aligned(RTE_X86_ZMM_SIZE). It only accepts
> + * numbers that are power of 2. So, even though RTE_X86_ZMM_SIZE represents a
> + * number that is a power of two it cannot be used directly.
> + * Ref: https://learn.microsoft.com/en-us/cpp/cpp/align-cpp?view=msvc-170
> + * The static assert below ensures that RTE_X86_ZMM_SIZE is equal to what is
> + * used in the __rte_aligned() expression.
> + */
> +static_assert(RTE_X86_ZMM_SIZE == 64, "Unexpected size of __m512i");
> +typedef union __rte_aligned(64) __rte_x86_zmm {
Just wonder, would then MSVC understand something like:
#define RTE_X86_ZMM_SIZE64
static_assert(RTE_X86_ZMM_SIZE == sizeof((__m512i), "Unexpected size of
__m512i");
typedef union __rte_aligned(RTE_X86_ZMM_SIZE) __rte_x86_zmm {
?
> __m512i z;
> ymm_ty[RTE_X86_ZMM_SIZE / sizeof(ymm_t)];
> xmm_tx[RTE_X86_ZMM_SIZE / sizeof(xmm_t)];
> --
> 2.48.1.vfs.0.0
RE: [PATCH 1/3] net/mlx5/hws: support jump FDB Rx
Hi Suanming,
PSB
> -Original Message-
> From: Suanming Mou
> Sent: Tuesday, February 25, 2025 8:45 AM
> To: Dariusz Sosnowski ; Slava Ovsiienko
> ; Bing Zhao ; Ori Kam
> ; Matan Azrad
> Cc: dev@dpdk.org; Raslan Darawsheh ; Alex Vesker
>
> Subject: [PATCH 1/3] net/mlx5/hws: support jump FDB Rx
>
> Before FW introduced the JUMP_FDB_RX action feature, jump from FDB Tx to
> Rx is not allowed. JUMP_FDB_RX feature introduces the internal loopback
> for Tx case and allow the REG C0 C1 B be preserved as well.
>
> This commit adds the JUMP_FDB_RX cap bit check and use JUMP_FDB_RX instead
> of FT for dest table FDB Rx case.
>
> Signed-off-by: Suanming Mou
> Signed-off-by: Alex Vesker
> ---
> drivers/common/mlx5/mlx5_prm.h| 9 -
> drivers/net/mlx5/hws/mlx5dr_action.c | 26 +-
> drivers/net/mlx5/hws/mlx5dr_action.h | 4
> drivers/net/mlx5/hws/mlx5dr_cmd.c | 9 +
> drivers/net/mlx5/hws/mlx5dr_cmd.h | 2 ++
> drivers/net/mlx5/hws/mlx5dr_context.c | 17 +
> drivers/net/mlx5/hws/mlx5dr_context.h | 2 ++
> 7 files changed, 63 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/common/mlx5/mlx5_prm.h
> b/drivers/common/mlx5/mlx5_prm.h index 3fc3b0cd2a..84e3347794 100644
> --- a/drivers/common/mlx5/mlx5_prm.h
> +++ b/drivers/common/mlx5/mlx5_prm.h
> @@ -2466,7 +2466,8 @@ struct mlx5_ifc_wqe_based_flow_table_cap_bits {
> u8 reserved_at_60[0x8];
> u8 max_header_modify_pattern_length[0x8];
> u8 ste_format[0x10];
> - u8 stc_action_type[0x80];
> + u8 stc_action_type_63_0[0x40];
> + u8 stc_action_type_127_64[0x40];
> u8 header_insert_type[0x10];
> u8 header_remove_type[0x10];
> u8 trivial_match_definer[0x20];
> @@ -3543,6 +3544,11 @@ enum mlx5_ifc_rtc_reparse_mode {
> MLX5_IFC_RTC_REPARSE_BY_STC = 0x2,
> };
>
> +enum mlx5_ifc_stc_action_type_bit_index {
> + MLX5_IFC_STC_ACTION_TYPE_BIT_64_INDEX = 64,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX = 71, };
> +
> #define MLX5_IFC_RTC_LINEAR_LOOKUP_TBL_LOG_MAX 16
>
> struct mlx5_ifc_rtc_bits {
> @@ -3621,6 +3627,7 @@ enum mlx5_ifc_stc_action_type {
> MLX5_IFC_STC_ACTION_TYPE_ALLOW = 0x84,
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_VPORT = 0x85,
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_UPLINK = 0x86,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX = 0x87,
> };
>
> enum mlx5_ifc_stc_reparse_mode {
> diff --git a/drivers/net/mlx5/hws/mlx5dr_action.c
> b/drivers/net/mlx5/hws/mlx5dr_action.c
> index b9452a3ebc..e21db5b327 100644
> --- a/drivers/net/mlx5/hws/mlx5dr_action.c
> +++ b/drivers/net/mlx5/hws/mlx5dr_action.c
> @@ -803,6 +803,9 @@ int mlx5dr_action_root_build_attr(struct
> mlx5dr_rule_action rule_actions[],
>
> switch (action->type) {
> case MLX5DR_ACTION_TYP_TBL:
> + attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
> + attr[i].obj = action->dest_tbl.devx_obj->obj;
> + break;
> case MLX5DR_ACTION_TYP_TIR:
> attr[i].type = MLX5DV_FLOW_ACTION_DEST_DEVX;
> attr[i].obj = action->devx_obj;
> @@ -1097,6 +1100,17 @@ static void mlx5dr_action_fill_stc_attr(struct
> mlx5dr_action *action,
> }
> break;
> case MLX5DR_ACTION_TYP_TBL:
> + attr->action_offset = MLX5DR_ACTION_OFFSET_HIT;
> + attr->dest_table_id = obj->id;
> + /* Only for unified FDB Rx case */
> + if (mlx5dr_context_cap_stc(action->ctx,
> + MLX5_IFC_STC_ACTION_TYPE_JUMP_FLOW_TABLE_FDB_RX_BIT_INDEX)
> &&
> + action->dest_tbl.type == MLX5DR_TABLE_TYPE_FDB_RX)
> + attr->action_type =
> MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FLOW_TABLE_FDB_RX;
> + else
> + attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
Fix the long line warnings, same as other places.
Also mlx5dr_context_cap_stc() seems to contain some calculation inside. If the
context and the CAP will not be changed during the runtime on a specific port.
Can this CAP BIT checking be done only once and saved for later usage? Since
the code is used in the flow insertion.WDYT?
> +
> + break;
> case MLX5DR_ACTION_TYP_DEST_ARRAY:
> attr->action_type = MLX5_IFC_STC_ACTION_TYPE_JUMP_TO_FT;
> attr->action_offset = MLX5DR_ACTION_OFFSET_HIT; @@ -1419,17
> +1433,19 @@ mlx5dr_action_create_dest_table(struct mlx5dr_context *ctx,
> if (!action)
> return NULL;
>
> + action->dest_tbl.type = tbl->type;
> +
> if (mlx5dr_action_is_root_flags(flags)) {
> if (mlx5dr_context_shared_gvmi_used(ctx))
> - action->devx_obj = tbl->local_ft->obj;
> + action->dest_tbl.devx_obj = tbl->local_ft;
> else
> - action->devx_obj = tbl->ft->obj
RE: [PATCH v5 1/3] cryptodev: add SM4 GCM
> -Original Message- > From: Dooley, Brian > Sent: Monday, February 24, 2025 6:00 PM > To: Akhil Goyal ; Fan Zhang > Cc: dev@dpdk.org; Ji, Kai ; De Lara Guarch, Pablo > ; Dooley, Brian > Subject: [PATCH v5 1/3] cryptodev: add SM4 GCM > > Added support for symmetric algorithm SM4 GCM in cryptodev. > > Signed-off-by: Brian Dooley Acked-by: Pablo de Lara
RE: [PATCH v5 3/3] app/test: add SM4 GCM tests
> -Original Message- > From: Dooley, Brian > Sent: Monday, February 24, 2025 6:00 PM > To: Akhil Goyal ; Fan Zhang > Cc: dev@dpdk.org; Ji, Kai ; De Lara Guarch, Pablo > ; Dooley, Brian > Subject: [PATCH v5 3/3] app/test: add SM4 GCM tests > > Added SM4-GCM tests for the AESNI MB PMD. > > Signed-off-by: Brian Dooley Acked-by: Pablo de Lara
RE: [PATCH v5 2/3] crypto/ipsec_mb: add SM4 GCM support
> -Original Message- > From: Dooley, Brian > Sent: Monday, February 24, 2025 6:00 PM > To: Ji, Kai ; De Lara Guarch, Pablo > > Cc: dev@dpdk.org; gak...@marvell.com; Dooley, Brian > > Subject: [PATCH v5 2/3] crypto/ipsec_mb: add SM4 GCM support > > This patch introduces SM4 GCM algorithm support to the AESNI_MB PMD. > SM4 GCM is available in the v2.0 release of Intel IPsec MB. > > Signed-off-by: Brian Dooley Acked-by: Pablo de Lara
