* Re: [dpdk-dev] [PATCH v2] common/mlx5: add provider query port support to glue library
2021-06-23 13:51 ` David Marchand
@ 2021-06-23 15:39 ` Slava Ovsiienko
2021-06-24 10:10 ` Slava Ovsiienko
` (2 subsequent siblings)
3 siblings, 0 replies; 16+ messages in thread
From: Slava Ovsiienko @ 2021-06-23 15:39 UTC (permalink / raw)
To: David Marchand
Cc: dev, Raslan Darawsheh, Matan Azrad, NBU-Contact-Thomas Monjalon,
dpdk stable
Hi, David
> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> Sent: Wednesday, June 23, 2021 16:52
> To: Slava Ovsiienko <viacheslavo@nvidia.com>
> Cc: dev <dev@dpdk.org>; Raslan Darawsheh <rasland@nvidia.com>; Matan
> Azrad <matan@nvidia.com>; NBU-Contact-Thomas Monjalon
> <thomas@monjalon.net>; dpdk stable <stable@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH v2] common/mlx5: add provider query port
> support to glue library
>
> On Wed, Jun 23, 2021 at 1:27 PM Slava Ovsiienko <viacheslavo@nvidia.com>
> wrote:
> > > > This patch is highly desirable to be provided in DPDK LTS releases
> > > > due to it covers the major compatibility issue.
> > >
> > > This patch is a fix, yet nothing tells this story in the title.
> >
> > This patch is not a fix. Actually it covers the compatibility issue, not a bug.
>
> I still think it counts as a fix in the sense that the mlx5 driver behavior changes
> to an undesired state if rdma-core gets updated.
>
> It's not about preferring "fix" in the title.
> It is more accurate/descriptive to me.
> If you feel strongly against "fix", I won't insist.
I have no strong objections against "fix". The patch definitely can be
categorized as "fix" as well. It would be easier to push the patch to LTS 😊
I just tried to be extremely honest - upstream rdma-core did not provide this API,
now it does, it would be very nice to engage it, allowing full E-Switch support over
upstream rdma-core in some configurations. From other side - you are right,
w/o patch E-Switch might not work in DPDK, with patch - it should work.
Looks like a true magic fix 😊.
>
> Yet "add provider quer port support to glue library" is just black magic to
> most of us.
>
> > The Upstream rdma-core was evolved, its community adopted a slightly
> > different API version than was presented in the vendor version.
> > Our PMD should conform both versions and we provided this patch for
> DPDK.
>
> Let's try differently.
> Place yourself as someone who does not know a thing about the mlx5 driver
> and rdma-core.
> How does such a person understand the impact of this patch?
>
> I would state in the title that the mlx5 driver can now handle correctly rdma-
> core 35.
> Additionally, it could indicate which feature X is now behaving as intended.
> But if feature X is something internal to the mlx5 driver, it is worth skipping.
This rdma-core API mostly reports E-Switch vport assigned indices, the assigning schema
of these ones depends on many factors - kernel/firmware/LAG configs/etc. Formerly,
the vport indices were assigned in direct correspondence with VF index, for these cases
E-Switch is supported fine even w/o API. But the newer kernel drivers with new features supported
changed the vport identification schema and former approach might not work, that's why this
API was introduced.
So, if I understand your comment correctly, we should tell few words the E-Switch
behavior might be affected and the feature malfunction is possible.
With best regards,
Slava
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [PATCH v2] common/mlx5: add provider query port support to glue library
2021-06-23 13:51 ` David Marchand
2021-06-23 15:39 ` Slava Ovsiienko
@ 2021-06-24 10:10 ` Slava Ovsiienko
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 1/2] " Viacheslav Ovsiienko
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 " Viacheslav Ovsiienko
3 siblings, 0 replies; 16+ messages in thread
From: Slava Ovsiienko @ 2021-06-24 10:10 UTC (permalink / raw)
To: David Marchand
Cc: dev, Raslan Darawsheh, Matan Azrad, NBU-Contact-Thomas Monjalon,
dpdk stable
Hi, David
Thank you for the review and comments.
What do you think about the commit message like this?
common/mlx5: fix rdma-core v35 query port API support
The rdma-core mlx5 provider introduced the port attributes query
API since version v35.0 - the mlx5dv_query_port routine. In order
to support this change in the rdma-core the conditional compilation
flag HAVE_MLX5DV_DR_DEVX_PORT_V35 is introduced by the this patch.
In the OFED rdma-core version the new compatible mlx5dv_query_port
routine was introduced as well, replacing the existing proprietary
mlx5dv_query_devx_port routine. The proprietary routine was
controlled in PMD code with HAVE_MLX5DV_DR_DEVX_PORT conditional
flag.
Currently, the OFED rdma-core library contains both versions of
port query API. And this version is a transitional one, there are
the plans to remove the proprietary mlx5dv_query_devx_port routine
and the HAVE_MLX5DV_DR_DEVX_PORT flag in PMD will not work anymore.
We had one more dependency on this flag in the code (for the
mlx5dv_dr_action_create_dest_ib_port routine) and the patch
fixes mentioned dependency also, by introducing the new
dedicated conditional flag - HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT.
The introduced port query API is related to getting kernel
metadata indices assigned to the vports of E-switch. Without
engaging the API the PMD makes the assumptions about the indices,
those might be incorrect for some configurations (for example -
LAG ones) and the E-switch feature might not operate correctly.
This patch is highly desirable to be provided in DPDK LTS releases
due to it covers the major compatibility issue.
(I will add cc: and fixes tag while sending the update)
With best regards,
Slava
> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> Sent: Wednesday, June 23, 2021 16:52
> To: Slava Ovsiienko <viacheslavo@nvidia.com>
> Cc: dev <dev@dpdk.org>; Raslan Darawsheh <rasland@nvidia.com>; Matan
> Azrad <matan@nvidia.com>; NBU-Contact-Thomas Monjalon
> <thomas@monjalon.net>; dpdk stable <stable@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH v2] common/mlx5: add provider query port
> support to glue library
>
> On Wed, Jun 23, 2021 at 1:27 PM Slava Ovsiienko <viacheslavo@nvidia.com>
> wrote:
> > > > This patch is highly desirable to be provided in DPDK LTS releases
> > > > due to it covers the major compatibility issue.
> > >
> > > This patch is a fix, yet nothing tells this story in the title.
> >
> > This patch is not a fix. Actually it covers the compatibility issue, not a bug.
>
> I still think it counts as a fix in the sense that the mlx5 driver behavior
> changes to an undesired state if rdma-core gets updated.
>
> It's not about preferring "fix" in the title.
> It is more accurate/descriptive to me.
> If you feel strongly against "fix", I won't insist.
>
> Yet "add provider quer port support to glue library" is just black magic to
> most of us.
>
>
> > The Upstream rdma-core was evolved, its community adopted a slightly
> > different API version than was presented in the vendor version.
> > Our PMD should conform both versions and we provided this patch for
> DPDK.
>
> Let's try differently.
> Place yourself as someone who does not know a thing about the mlx5 driver
> and rdma-core.
> How does such a person understand the impact of this patch?
>
> I would state in the title that the mlx5 driver can now handle correctly rdma-
> core 35.
> Additionally, it could indicate which feature X is now behaving as intended.
> But if feature X is something internal to the mlx5 driver, it is worth skipping.
>
>
>
> --
> David Marchand
^ permalink raw reply [flat|nested] 16+ messages in thread
* [dpdk-dev] [PATCH v3 1/2] common/mlx5: add provider query port support to glue library
2021-06-23 13:51 ` David Marchand
2021-06-23 15:39 ` Slava Ovsiienko
2021-06-24 10:10 ` Slava Ovsiienko
@ 2021-07-01 7:15 ` Viacheslav Ovsiienko
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
2021-07-06 13:44 ` [dpdk-dev] [PATCH v3 1/2] common/mlx5: add provider query port support to glue library Thomas Monjalon
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 " Viacheslav Ovsiienko
3 siblings, 2 replies; 16+ messages in thread
From: Viacheslav Ovsiienko @ 2021-07-01 7:15 UTC (permalink / raw)
To: dev; +Cc: rasland, matan, thomas, david.marchand, stable
In order to get E-Switch vport identifiers the mlx5 PMD relies
on two approaches:
- use port query API if it is provided by rdma-core library
- otherwise, deduce vport ids from the related VF index
The latter is not reliable and may not work with newer kernel
driver versions and in some configurations (LAG), causing
the E-Switch malfunction. Hence, engaging the port query API
is highly desirable.
The port attributes query API was supported by vendor rdma-core
versions only (in OFED). In Upstream, the rdma-core mlx5 provider
introduced the port attributes query API since version v35.0 - the
mlx5dv_query_port routine, that differs from vendor API. In order
to support this change in the rdma-core the conditional compilation
flag HAVE_MLX5DV_DR_DEVX_PORT_V35 is introduced by the this patch.
In the OFED rdma-core version the new compatible mlx5dv_query_port
routine was introduced as well, replacing the existing proprietary
mlx5dv_query_devx_port routine. The proprietary routine is still
controlled in PMD code with HAVE_MLX5DV_DR_DEVX_PORT conditional
flag.
Currently, the OFED rdma-core library contains both versions of
port query API. And this version is a transitional one, there are
the plans to remove the proprietary mlx5dv_query_devx_port routine
and the HAVE_MLX5DV_DR_DEVX_PORT flag in PMD will not work anymore.
With this patch applied the mlx5 PMD can be compiled and operates
over all rdma-core versions:
- Upstream rdma-core before v35 (deduces vport id from VF name)
(* E-Switch may not operate correctly)
- Upstream rdma-core v35 and above (uses common port query API)
- rdma-core OFED before 5.5 (uses proprietary port query API)
- rdma-core OFED 5.5 and above (uses common port query API)
Despite this patch is not a bugfix (it follows the introduced API
variation in underlying library), it covers the major compatibility
issue and is highly desired to be ported to DPDK LTS.
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
v1: http://patches.dpdk.org/project/dpdk/patch/20210607093726.14546-1-viacheslavo@nvidia.com/
v2: http://patches.dpdk.org/project/dpdk/patch/20210619124830.25297-1-viacheslavo@nvidia.com/#134196
- commit message was clarified
v3: - reword commit message addressing comments
- split patch into small series
---
drivers/common/mlx5/linux/meson.build | 2 +
drivers/common/mlx5/linux/mlx5_glue.c | 55 ++++++++++++++++++++----
drivers/common/mlx5/linux/mlx5_glue.h | 16 ++++++-
drivers/net/mlx5/linux/mlx5_os.c | 60 ++++++++++++---------------
4 files changed, 89 insertions(+), 44 deletions(-)
diff --git a/drivers/common/mlx5/linux/meson.build b/drivers/common/mlx5/linux/meson.build
index 007834a49b..5a0ee51b37 100644
--- a/drivers/common/mlx5/linux/meson.build
+++ b/drivers/common/mlx5/linux/meson.build
@@ -93,6 +93,8 @@ has_sym_args = [
'IBV_WQ_FLAG_RX_END_PADDING' ],
[ 'HAVE_MLX5DV_DR_DEVX_PORT', 'infiniband/mlx5dv.h',
'mlx5dv_query_devx_port' ],
+ [ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
+ 'mlx5dv_query_port' ],
[ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
'mlx5dv_devx_obj_create' ],
[ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
diff --git a/drivers/common/mlx5/linux/mlx5_glue.c b/drivers/common/mlx5/linux/mlx5_glue.c
index d3bd645a5b..38b174aa31 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.c
+++ b/drivers/common/mlx5/linux/mlx5_glue.c
@@ -1087,17 +1087,54 @@ mlx5_glue_devx_wq_query(struct ibv_wq *wq, const void *in, size_t inlen,
static int
mlx5_glue_devx_port_query(struct ibv_context *ctx,
uint32_t port_num,
- struct mlx5dv_devx_port *mlx5_devx_port)
-{
+ struct mlx5_port_info *info)
+{
+ int err = 0;
+
+ info->query_flags = 0;
+#ifdef HAVE_MLX5DV_DR_DEVX_PORT_V35
+ /* The DevX port query API is implemented (rdma-core v35 and above). */
+ struct mlx5_ib_uapi_query_port devx_port;
+
+ memset(&devx_port, 0, sizeof(devx_port));
+ err = mlx5dv_query_port(ctx, port_num, &devx_port);
+ if (err)
+ return err;
+ if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT_REG_C0) {
+ info->vport_meta_tag = devx_port.reg_c0.value;
+ info->vport_meta_mask = devx_port.reg_c0.mask;
+ info->query_flags |= MLX5_PORT_QUERY_REG_C0;
+ }
+ if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT) {
+ info->vport_id = devx_port.vport;
+ info->query_flags |= MLX5_PORT_QUERY_VPORT;
+ }
+#else
#ifdef HAVE_MLX5DV_DR_DEVX_PORT
- return mlx5dv_query_devx_port(ctx, port_num, mlx5_devx_port);
+ /* The legacy DevX port query API is implemented (prior v35). */
+ struct mlx5dv_devx_port devx_port = {
+ .comp_mask = MLX5DV_DEVX_PORT_VPORT |
+ MLX5DV_DEVX_PORT_MATCH_REG_C_0
+ };
+
+ err = mlx5dv_query_devx_port(ctx, port_num, &devx_port);
+ if (err)
+ return err;
+ if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
+ info->vport_meta_tag = devx_port.reg_c_0.value;
+ info->vport_meta_mask = devx_port.reg_c_0.mask;
+ info->query_flags |= MLX5_PORT_QUERY_REG_C0;
+ }
+ if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
+ info->vport_id = devx_port.vport_num;
+ info->query_flags |= MLX5_PORT_QUERY_VPORT;
+ }
#else
- (void)ctx;
- (void)port_num;
- (void)mlx5_devx_port;
- errno = ENOTSUP;
- return errno;
-#endif
+ RTE_SET_USED(ctx);
+ RTE_SET_USED(port_num);
+#endif /* HAVE_MLX5DV_DR_DEVX_PORT */
+#endif /* HAVE_MLX5DV_DR_DEVX_PORT_V35 */
+ return err;
}
static int
diff --git a/drivers/common/mlx5/linux/mlx5_glue.h b/drivers/common/mlx5/linux/mlx5_glue.h
index 97462e9ab8..840d8cf57f 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.h
+++ b/drivers/common/mlx5/linux/mlx5_glue.h
@@ -84,6 +84,20 @@ struct mlx5dv_dr_action;
struct mlx5dv_devx_port;
#endif
+#ifndef HAVE_MLX5DV_DR_DEVX_PORT_V35
+struct mlx5dv_port;
+#endif
+
+#define MLX5_PORT_QUERY_VPORT (1u << 0)
+#define MLX5_PORT_QUERY_REG_C0 (1u << 1)
+
+struct mlx5_port_info {
+ uint16_t query_flags;
+ uint16_t vport_id; /* Associated VF vport index (if any). */
+ uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
+ uint32_t vport_meta_mask; /* Used for vport index field match mask. */
+};
+
#ifndef HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER
struct mlx5dv_dr_flow_meter_attr;
#endif
@@ -311,7 +325,7 @@ struct mlx5_glue {
void *out, size_t outlen);
int (*devx_port_query)(struct ibv_context *ctx,
uint32_t port_num,
- struct mlx5dv_devx_port *mlx5_devx_port);
+ struct mlx5_port_info *info);
int (*dr_dump_domain)(FILE *file, void *domain);
int (*dr_dump_rule)(FILE *file, void *rule);
int (*devx_query_eqn)(struct ibv_context *context, uint32_t cpus,
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 302a881310..92b3009786 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -822,9 +822,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
char name[RTE_ETH_NAME_MAX_LEN];
int own_domain_id = 0;
uint16_t port_id;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
- struct mlx5dv_devx_port devx_port = { .comp_mask = 0 };
-#endif
+ struct mlx5_port_info vport_info = { .query_flags = 0 };
/* Determine if this port representor is supposed to be spawned. */
if (switch_info->representor && dpdk_dev->devargs &&
@@ -1055,29 +1053,27 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
priv->vport_meta_tag = 0;
priv->vport_meta_mask = 0;
priv->pf_bond = spawn->pf_bond;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
/*
- * The DevX port query API is implemented. E-Switch may use
- * either vport or reg_c[0] metadata register to match on
- * vport index. The engaged part of metadata register is
- * defined by mask.
+ * If we have E-Switch we should determine the vport attributes.
+ * E-Switch may use either source vport field or reg_c[0] metadata
+ * register to match on vport index. The engaged part of metadata
+ * register is defined by mask.
*/
if (switch_info->representor || switch_info->master) {
- devx_port.comp_mask = MLX5DV_DEVX_PORT_VPORT |
- MLX5DV_DEVX_PORT_MATCH_REG_C_0;
- err = mlx5_glue->devx_port_query(sh->ctx, spawn->phys_port,
- &devx_port);
+ err = mlx5_glue->devx_port_query(sh->ctx,
+ spawn->phys_port,
+ &vport_info);
if (err) {
DRV_LOG(WARNING,
"can't query devx port %d on device %s",
spawn->phys_port,
mlx5_os_get_dev_device_name(spawn->phys_dev));
- devx_port.comp_mask = 0;
+ vport_info.query_flags = 0;
}
}
- if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
- priv->vport_meta_tag = devx_port.reg_c_0.value;
- priv->vport_meta_mask = devx_port.reg_c_0.mask;
+ if (vport_info.query_flags & MLX5_PORT_QUERY_REG_C0) {
+ priv->vport_meta_tag = vport_info.vport_meta_tag;
+ priv->vport_meta_mask = vport_info.vport_meta_mask;
if (!priv->vport_meta_mask) {
DRV_LOG(ERR, "vport zero mask for port %d"
" on bonding device %s",
@@ -1097,8 +1093,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
goto error;
}
}
- if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
- priv->vport_id = devx_port.vport_num;
+ if (vport_info.query_flags & MLX5_PORT_QUERY_VPORT) {
+ priv->vport_id = vport_info.vport_id;
} else if (spawn->pf_bond >= 0 &&
(switch_info->representor || switch_info->master)) {
DRV_LOG(ERR, "can't deduce vport index for port %d"
@@ -1108,25 +1104,21 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
err = ENOTSUP;
goto error;
} else {
- /* Suppose vport index in compatible way. */
+ /*
+ * Suppose vport index in compatible way. Kernel/rdma_core
+ * support single E-Switch per PF configurations only and
+ * vport_id field contains the vport index for associated VF,
+ * which is deduced from representor port name.
+ * For example, let's have the IB device port 10, it has
+ * attached network device eth0, which has port name attribute
+ * pf0vf2, we can deduce the VF number as 2, and set vport index
+ * as 3 (2+1). This assigning schema should be changed if the
+ * multiple E-Switch instances per PF configurations or/and PCI
+ * subfunctions are added.
+ */
priv->vport_id = switch_info->representor ?
switch_info->port_name + 1 : -1;
}
-#else
- /*
- * Kernel/rdma_core support single E-Switch per PF configurations
- * only and vport_id field contains the vport index for
- * associated VF, which is deduced from representor port name.
- * For example, let's have the IB device port 10, it has
- * attached network device eth0, which has port name attribute
- * pf0vf2, we can deduce the VF number as 2, and set vport index
- * as 3 (2+1). This assigning schema should be changed if the
- * multiple E-Switch instances per PF configurations or/and PCI
- * subfunctions are added.
- */
- priv->vport_id = switch_info->representor ?
- switch_info->port_name + 1 : -1;
-#endif
priv->representor_id = mlx5_representor_id_encode(switch_info,
eth_da->type);
/*
--
2.18.1
^ permalink raw reply [flat|nested] 16+ messages in thread
* [dpdk-dev] [PATCH v3 2/2] common/mlx5: fix compilation flag on port query API
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 1/2] " Viacheslav Ovsiienko
@ 2021-07-01 7:15 ` Viacheslav Ovsiienko
2021-07-06 13:49 ` [dpdk-dev] [dpdk-stable] " Thomas Monjalon
2021-07-06 13:44 ` [dpdk-dev] [PATCH v3 1/2] common/mlx5: add provider query port support to glue library Thomas Monjalon
1 sibling, 1 reply; 16+ messages in thread
From: Viacheslav Ovsiienko @ 2021-07-01 7:15 UTC (permalink / raw)
To: dev; +Cc: rasland, matan, thomas, david.marchand, stable
There is the compilation flag HAVE_MLX5DV_DR_DEVX_PORT
that depends on presence of mlx5dv_query_devx_port routine
in rdma-core library.
The mlx5dv_query_devx_port routine is vendor specific, exists
only in OFED version of rdma-core library and there are the plans
to remove the routine and replace one with Upstream compatible
mlx5dv_query_port version.
As mlx5dv_query_devx_port is being removed we should reconsider
all the dependencies on the HAVE_MLX5DV_DR_DEVX_PORT compilation
flag and introduce the new dedicated compilation flags as needed.
Fixes: 6cfe84fbe7b1 ("net/mlx5: fix port action for LAG")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
drivers/common/mlx5/linux/meson.build | 2 ++
drivers/common/mlx5/linux/mlx5_glue.c | 2 +-
drivers/net/mlx5/mlx5_flow_dv.c | 2 +-
3 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/common/mlx5/linux/meson.build b/drivers/common/mlx5/linux/meson.build
index 5a0ee51b37..5cea1b44d7 100644
--- a/drivers/common/mlx5/linux/meson.build
+++ b/drivers/common/mlx5/linux/meson.build
@@ -95,6 +95,8 @@ has_sym_args = [
'mlx5dv_query_devx_port' ],
[ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
'mlx5dv_query_port' ],
+ [ 'HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT', 'infiniband/mlx5dv.h',
+ 'mlx5dv_dr_action_create_dest_ib_port' ],
[ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
'mlx5dv_devx_obj_create' ],
[ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
diff --git a/drivers/common/mlx5/linux/mlx5_glue.c b/drivers/common/mlx5/linux/mlx5_glue.c
index 38b174aa31..00be8114be 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.c
+++ b/drivers/common/mlx5/linux/mlx5_glue.c
@@ -391,7 +391,7 @@ mlx5_glue_dr_create_flow_action_dest_flow_tbl(void *tbl)
static void *
mlx5_glue_dr_create_flow_action_dest_port(void *domain, uint32_t port)
{
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+#ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
return mlx5dv_dr_action_create_dest_ib_port(domain, port);
#else
#ifdef HAVE_MLX5DV_DR_ESWITCH
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 62edc4fa85..c5d4b01e57 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -10342,7 +10342,7 @@ flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
RTE_FLOW_ERROR_TYPE_ACTION,
NULL,
"No eswitch info was found for port");
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+#ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
/*
* This parameter is transferred to
* mlx5dv_dr_action_create_dest_ib_port().
--
2.18.1
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [dpdk-stable] [PATCH v3 2/2] common/mlx5: fix compilation flag on port query API
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
@ 2021-07-06 13:49 ` Thomas Monjalon
0 siblings, 0 replies; 16+ messages in thread
From: Thomas Monjalon @ 2021-07-06 13:49 UTC (permalink / raw)
To: Viacheslav Ovsiienko; +Cc: dev, stable, rasland, matan, david.marchand, stable
01/07/2021 09:15, Viacheslav Ovsiienko:
> There is the compilation flag HAVE_MLX5DV_DR_DEVX_PORT
> that depends on presence of mlx5dv_query_devx_port routine
> in rdma-core library.
The english syntax is difficult here.
You probably mean:
"The compilation flag HAVE_MLX5DV_DR_DEVX_PORT depends on the presence of
mlx5dv_query_devx_port routine in rdma-core library."
> The mlx5dv_query_devx_port routine is vendor specific, exists
> only in OFED version of rdma-core library and there are the plans
"there are plans"
> to remove the routine and replace one with Upstream compatible
> mlx5dv_query_port version.
Which OFED version will have this change?
> As mlx5dv_query_devx_port is being removed we should reconsider
> all the dependencies on the HAVE_MLX5DV_DR_DEVX_PORT compilation
> flag and introduce the new dedicated compilation flags as needed.
"introduce a new"
It doesn't explain why HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
is a better flag.
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [PATCH v3 1/2] common/mlx5: add provider query port support to glue library
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 1/2] " Viacheslav Ovsiienko
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
@ 2021-07-06 13:44 ` Thomas Monjalon
1 sibling, 0 replies; 16+ messages in thread
From: Thomas Monjalon @ 2021-07-06 13:44 UTC (permalink / raw)
To: Viacheslav Ovsiienko; +Cc: dev, rasland, matan, david.marchand, stable
01/07/2021 09:15, Viacheslav Ovsiienko:
> In order to get E-Switch vport identifiers the mlx5 PMD relies
> on two approaches:
> - use port query API if it is provided by rdma-core library
> - otherwise, deduce vport ids from the related VF index
> The latter is not reliable and may not work with newer kernel
> driver versions and in some configurations (LAG), causing
> the E-Switch malfunction. Hence, engaging the port query API
> is highly desirable.
>
> The port attributes query API was supported by vendor rdma-core
> versions only (in OFED). In Upstream, the rdma-core mlx5 provider
> introduced the port attributes query API since version v35.0 - the
> mlx5dv_query_port routine, that differs from vendor API. In order
> to support this change in the rdma-core the conditional compilation
> flag HAVE_MLX5DV_DR_DEVX_PORT_V35 is introduced by the this patch.
>
> In the OFED rdma-core version the new compatible mlx5dv_query_port
> routine was introduced as well, replacing the existing proprietary
> mlx5dv_query_devx_port routine. The proprietary routine is still
> controlled in PMD code with HAVE_MLX5DV_DR_DEVX_PORT conditional
> flag.
>
> Currently, the OFED rdma-core library contains both versions of
> port query API. And this version is a transitional one, there are
"this version" of what?
> the plans to remove the proprietary mlx5dv_query_devx_port routine
> and the HAVE_MLX5DV_DR_DEVX_PORT flag in PMD will not work anymore.
>
> With this patch applied the mlx5 PMD can be compiled and operates
> over all rdma-core versions:
> - Upstream rdma-core before v35 (deduces vport id from VF name)
> (* E-Switch may not operate correctly)
> - Upstream rdma-core v35 and above (uses common port query API)
> - rdma-core OFED before 5.5 (uses proprietary port query API)
> - rdma-core OFED 5.5 and above (uses common port query API)
This is the kind of explanation which should come first,
with exact function and flag names, before and after.
> Despite this patch is not a bugfix (it follows the introduced API
> variation in underlying library), it covers the major compatibility
> issue and is highly desired to be ported to DPDK LTS.
>
> Cc: stable@dpdk.org
>
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> Acked-by: Matan Azrad <matan@nvidia.com>
[...]
> --- a/drivers/common/mlx5/linux/meson.build
> +++ b/drivers/common/mlx5/linux/meson.build
> @@ -93,6 +93,8 @@ has_sym_args = [
> 'IBV_WQ_FLAG_RX_END_PADDING' ],
> [ 'HAVE_MLX5DV_DR_DEVX_PORT', 'infiniband/mlx5dv.h',
> 'mlx5dv_query_devx_port' ],
> + [ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
> + 'mlx5dv_query_port' ],
It seems there is an indentation issue.
> [ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
> 'mlx5dv_devx_obj_create' ],
> [ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
[...]
> + * pf0vf2, we can deduce the VF number as 2, and set vport index
> + * as 3 (2+1). This assigning schema should be changed if the
> + * multiple E-Switch instances per PF configurations or/and PCI
> + * subfunctions are added.
I don't understand this last sentence.
^ permalink raw reply [flat|nested] 16+ messages in thread
* [dpdk-dev] [PATCH v4 1/2] common/mlx5: add provider query port support to glue library
2021-06-23 13:51 ` David Marchand
` (2 preceding siblings ...)
2021-07-01 7:15 ` [dpdk-dev] [PATCH v3 1/2] " Viacheslav Ovsiienko
@ 2021-07-07 15:54 ` Viacheslav Ovsiienko
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
` (2 more replies)
3 siblings, 3 replies; 16+ messages in thread
From: Viacheslav Ovsiienko @ 2021-07-07 15:54 UTC (permalink / raw)
To: dev; +Cc: rasland, matan, thomas, david.marchand, stable
In order to get E-Switch vport identifiers the mlx5 PMD relies
on two approaches:
[a] use port query API if it is provided by rdma-core library
[b] otherwise, deduce vport ids from the related VF index
The latter is not reliable and may not work with newer kernel
drivers and in some configurations (LAG), causing E-Switch
malfunction. Hence, engaging the port query API is highly
desirable.
Depending on rdma-core version the port query API is:
- very old OFED versions have no query API (approach [b])
- rdma-core OFED < 5.5 provides mlx5dv_query_devx_port,
HAVE_MLX5DV_DR_DEVX_PORT flag is defined (approach [a])
- rdma-core OFED >= 5.5 has mlx5dv_query_port, flag
HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
- future OFED versions might remove mlx5dv_query_devx_port
and HAVE_MLX5DV_DR_DEVX_PORT will not be defined
- Upstream rdma-core < v35 has no port query API (approach [b])
- Upstream rdma-core >= v35 has mlx5dv_query_port, flag
HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
In order to support the new mlx5dv_query_port routine, the
conditional compilation flag HAVE_MLX5DV_DR_DEVX_PORT_V35
is introduced by the this patch. The flag HAVE_MLX5DV_DR_DEVX_PORT
is kept for compatibility with previous rdma-core versions.
Despite this patch is not a bugfix (it follows the introduced API
variation in underlying library), it resolves the compatibility
issue and is highly desired to be ported to DPDK LTS.
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
drivers/common/mlx5/linux/meson.build | 2 +
drivers/common/mlx5/linux/mlx5_glue.c | 55 ++++++++++++++++++++----
drivers/common/mlx5/linux/mlx5_glue.h | 16 ++++++-
drivers/net/mlx5/linux/mlx5_os.c | 60 ++++++++++++---------------
4 files changed, 89 insertions(+), 44 deletions(-)
diff --git a/drivers/common/mlx5/linux/meson.build b/drivers/common/mlx5/linux/meson.build
index 007834a49b..7168edc62c 100644
--- a/drivers/common/mlx5/linux/meson.build
+++ b/drivers/common/mlx5/linux/meson.build
@@ -93,6 +93,8 @@ has_sym_args = [
'IBV_WQ_FLAG_RX_END_PADDING' ],
[ 'HAVE_MLX5DV_DR_DEVX_PORT', 'infiniband/mlx5dv.h',
'mlx5dv_query_devx_port' ],
+ [ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
+ 'mlx5dv_query_port' ],
[ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
'mlx5dv_devx_obj_create' ],
[ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
diff --git a/drivers/common/mlx5/linux/mlx5_glue.c b/drivers/common/mlx5/linux/mlx5_glue.c
index d3bd645a5b..38b174aa31 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.c
+++ b/drivers/common/mlx5/linux/mlx5_glue.c
@@ -1087,17 +1087,54 @@ mlx5_glue_devx_wq_query(struct ibv_wq *wq, const void *in, size_t inlen,
static int
mlx5_glue_devx_port_query(struct ibv_context *ctx,
uint32_t port_num,
- struct mlx5dv_devx_port *mlx5_devx_port)
-{
+ struct mlx5_port_info *info)
+{
+ int err = 0;
+
+ info->query_flags = 0;
+#ifdef HAVE_MLX5DV_DR_DEVX_PORT_V35
+ /* The DevX port query API is implemented (rdma-core v35 and above). */
+ struct mlx5_ib_uapi_query_port devx_port;
+
+ memset(&devx_port, 0, sizeof(devx_port));
+ err = mlx5dv_query_port(ctx, port_num, &devx_port);
+ if (err)
+ return err;
+ if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT_REG_C0) {
+ info->vport_meta_tag = devx_port.reg_c0.value;
+ info->vport_meta_mask = devx_port.reg_c0.mask;
+ info->query_flags |= MLX5_PORT_QUERY_REG_C0;
+ }
+ if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT) {
+ info->vport_id = devx_port.vport;
+ info->query_flags |= MLX5_PORT_QUERY_VPORT;
+ }
+#else
#ifdef HAVE_MLX5DV_DR_DEVX_PORT
- return mlx5dv_query_devx_port(ctx, port_num, mlx5_devx_port);
+ /* The legacy DevX port query API is implemented (prior v35). */
+ struct mlx5dv_devx_port devx_port = {
+ .comp_mask = MLX5DV_DEVX_PORT_VPORT |
+ MLX5DV_DEVX_PORT_MATCH_REG_C_0
+ };
+
+ err = mlx5dv_query_devx_port(ctx, port_num, &devx_port);
+ if (err)
+ return err;
+ if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
+ info->vport_meta_tag = devx_port.reg_c_0.value;
+ info->vport_meta_mask = devx_port.reg_c_0.mask;
+ info->query_flags |= MLX5_PORT_QUERY_REG_C0;
+ }
+ if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
+ info->vport_id = devx_port.vport_num;
+ info->query_flags |= MLX5_PORT_QUERY_VPORT;
+ }
#else
- (void)ctx;
- (void)port_num;
- (void)mlx5_devx_port;
- errno = ENOTSUP;
- return errno;
-#endif
+ RTE_SET_USED(ctx);
+ RTE_SET_USED(port_num);
+#endif /* HAVE_MLX5DV_DR_DEVX_PORT */
+#endif /* HAVE_MLX5DV_DR_DEVX_PORT_V35 */
+ return err;
}
static int
diff --git a/drivers/common/mlx5/linux/mlx5_glue.h b/drivers/common/mlx5/linux/mlx5_glue.h
index 97462e9ab8..840d8cf57f 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.h
+++ b/drivers/common/mlx5/linux/mlx5_glue.h
@@ -84,6 +84,20 @@ struct mlx5dv_dr_action;
struct mlx5dv_devx_port;
#endif
+#ifndef HAVE_MLX5DV_DR_DEVX_PORT_V35
+struct mlx5dv_port;
+#endif
+
+#define MLX5_PORT_QUERY_VPORT (1u << 0)
+#define MLX5_PORT_QUERY_REG_C0 (1u << 1)
+
+struct mlx5_port_info {
+ uint16_t query_flags;
+ uint16_t vport_id; /* Associated VF vport index (if any). */
+ uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
+ uint32_t vport_meta_mask; /* Used for vport index field match mask. */
+};
+
#ifndef HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER
struct mlx5dv_dr_flow_meter_attr;
#endif
@@ -311,7 +325,7 @@ struct mlx5_glue {
void *out, size_t outlen);
int (*devx_port_query)(struct ibv_context *ctx,
uint32_t port_num,
- struct mlx5dv_devx_port *mlx5_devx_port);
+ struct mlx5_port_info *info);
int (*dr_dump_domain)(FILE *file, void *domain);
int (*dr_dump_rule)(FILE *file, void *rule);
int (*devx_query_eqn)(struct ibv_context *context, uint32_t cpus,
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 302a881310..92b3009786 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -822,9 +822,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
char name[RTE_ETH_NAME_MAX_LEN];
int own_domain_id = 0;
uint16_t port_id;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
- struct mlx5dv_devx_port devx_port = { .comp_mask = 0 };
-#endif
+ struct mlx5_port_info vport_info = { .query_flags = 0 };
/* Determine if this port representor is supposed to be spawned. */
if (switch_info->representor && dpdk_dev->devargs &&
@@ -1055,29 +1053,27 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
priv->vport_meta_tag = 0;
priv->vport_meta_mask = 0;
priv->pf_bond = spawn->pf_bond;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
/*
- * The DevX port query API is implemented. E-Switch may use
- * either vport or reg_c[0] metadata register to match on
- * vport index. The engaged part of metadata register is
- * defined by mask.
+ * If we have E-Switch we should determine the vport attributes.
+ * E-Switch may use either source vport field or reg_c[0] metadata
+ * register to match on vport index. The engaged part of metadata
+ * register is defined by mask.
*/
if (switch_info->representor || switch_info->master) {
- devx_port.comp_mask = MLX5DV_DEVX_PORT_VPORT |
- MLX5DV_DEVX_PORT_MATCH_REG_C_0;
- err = mlx5_glue->devx_port_query(sh->ctx, spawn->phys_port,
- &devx_port);
+ err = mlx5_glue->devx_port_query(sh->ctx,
+ spawn->phys_port,
+ &vport_info);
if (err) {
DRV_LOG(WARNING,
"can't query devx port %d on device %s",
spawn->phys_port,
mlx5_os_get_dev_device_name(spawn->phys_dev));
- devx_port.comp_mask = 0;
+ vport_info.query_flags = 0;
}
}
- if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
- priv->vport_meta_tag = devx_port.reg_c_0.value;
- priv->vport_meta_mask = devx_port.reg_c_0.mask;
+ if (vport_info.query_flags & MLX5_PORT_QUERY_REG_C0) {
+ priv->vport_meta_tag = vport_info.vport_meta_tag;
+ priv->vport_meta_mask = vport_info.vport_meta_mask;
if (!priv->vport_meta_mask) {
DRV_LOG(ERR, "vport zero mask for port %d"
" on bonding device %s",
@@ -1097,8 +1093,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
goto error;
}
}
- if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
- priv->vport_id = devx_port.vport_num;
+ if (vport_info.query_flags & MLX5_PORT_QUERY_VPORT) {
+ priv->vport_id = vport_info.vport_id;
} else if (spawn->pf_bond >= 0 &&
(switch_info->representor || switch_info->master)) {
DRV_LOG(ERR, "can't deduce vport index for port %d"
@@ -1108,25 +1104,21 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
err = ENOTSUP;
goto error;
} else {
- /* Suppose vport index in compatible way. */
+ /*
+ * Suppose vport index in compatible way. Kernel/rdma_core
+ * support single E-Switch per PF configurations only and
+ * vport_id field contains the vport index for associated VF,
+ * which is deduced from representor port name.
+ * For example, let's have the IB device port 10, it has
+ * attached network device eth0, which has port name attribute
+ * pf0vf2, we can deduce the VF number as 2, and set vport index
+ * as 3 (2+1). This assigning schema should be changed if the
+ * multiple E-Switch instances per PF configurations or/and PCI
+ * subfunctions are added.
+ */
priv->vport_id = switch_info->representor ?
switch_info->port_name + 1 : -1;
}
-#else
- /*
- * Kernel/rdma_core support single E-Switch per PF configurations
- * only and vport_id field contains the vport index for
- * associated VF, which is deduced from representor port name.
- * For example, let's have the IB device port 10, it has
- * attached network device eth0, which has port name attribute
- * pf0vf2, we can deduce the VF number as 2, and set vport index
- * as 3 (2+1). This assigning schema should be changed if the
- * multiple E-Switch instances per PF configurations or/and PCI
- * subfunctions are added.
- */
- priv->vport_id = switch_info->representor ?
- switch_info->port_name + 1 : -1;
-#endif
priv->representor_id = mlx5_representor_id_encode(switch_info,
eth_da->type);
/*
--
2.18.1
^ permalink raw reply [flat|nested] 16+ messages in thread
* [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 " Viacheslav Ovsiienko
@ 2021-07-07 15:54 ` Viacheslav Ovsiienko
2021-07-07 17:39 ` Thomas Monjalon
2021-07-07 17:12 ` [dpdk-dev] [PATCH v4 1/2] common/mlx5: add provider query port support to glue library Thomas Monjalon
2021-07-08 16:07 ` Raslan Darawsheh
2 siblings, 1 reply; 16+ messages in thread
From: Viacheslav Ovsiienko @ 2021-07-07 15:54 UTC (permalink / raw)
To: dev; +Cc: rasland, matan, thomas, david.marchand, stable
The compilation flag HAVE_MLX5DV_DR_DEVX_PORT depends on presence
of mlx5dv_query_devx_port routine in rdma-core library.
The mlx5dv_query_devx_port routine exists only in OFED versions
of rdma-core library and is being planned to be removed and replaced
with Upstream compatible mlx5dv_query_port.
As mlx5dv_query_devx_port is being removed all the dependencies on
the HAVE_MLX5DV_DR_DEVX_PORT compilation flag are reconsidered.
Fixes: 6cfe84fbe7b1 ("net/mlx5: fix port action for LAG")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
drivers/common/mlx5/linux/meson.build | 2 ++
drivers/common/mlx5/linux/mlx5_glue.c | 2 +-
drivers/net/mlx5/mlx5_flow_dv.c | 2 +-
3 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/common/mlx5/linux/meson.build b/drivers/common/mlx5/linux/meson.build
index 7168edc62c..4ef163e1ef 100644
--- a/drivers/common/mlx5/linux/meson.build
+++ b/drivers/common/mlx5/linux/meson.build
@@ -95,6 +95,8 @@ has_sym_args = [
'mlx5dv_query_devx_port' ],
[ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
'mlx5dv_query_port' ],
+ [ 'HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT', 'infiniband/mlx5dv.h',
+ 'mlx5dv_dr_action_create_dest_ib_port' ],
[ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
'mlx5dv_devx_obj_create' ],
[ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
diff --git a/drivers/common/mlx5/linux/mlx5_glue.c b/drivers/common/mlx5/linux/mlx5_glue.c
index 38b174aa31..00be8114be 100644
--- a/drivers/common/mlx5/linux/mlx5_glue.c
+++ b/drivers/common/mlx5/linux/mlx5_glue.c
@@ -391,7 +391,7 @@ mlx5_glue_dr_create_flow_action_dest_flow_tbl(void *tbl)
static void *
mlx5_glue_dr_create_flow_action_dest_port(void *domain, uint32_t port)
{
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+#ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
return mlx5dv_dr_action_create_dest_ib_port(domain, port);
#else
#ifdef HAVE_MLX5DV_DR_ESWITCH
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 62edc4fa85..c5d4b01e57 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -10342,7 +10342,7 @@ flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
RTE_FLOW_ERROR_TYPE_ACTION,
NULL,
"No eswitch info was found for port");
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+#ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
/*
* This parameter is transferred to
* mlx5dv_dr_action_create_dest_ib_port().
--
2.18.1
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
@ 2021-07-07 17:39 ` Thomas Monjalon
0 siblings, 0 replies; 16+ messages in thread
From: Thomas Monjalon @ 2021-07-07 17:39 UTC (permalink / raw)
To: Viacheslav Ovsiienko; +Cc: dev, rasland, matan, david.marchand, stable
The title could be:
common/mlx5: fix compatibility with OFED port query API
07/07/2021 17:54, Viacheslav Ovsiienko:
> The compilation flag HAVE_MLX5DV_DR_DEVX_PORT depends on presence
> of mlx5dv_query_devx_port routine in rdma-core library.
>
> The mlx5dv_query_devx_port routine exists only in OFED versions
> of rdma-core library and is being planned to be removed and replaced
> with Upstream compatible mlx5dv_query_port.
>
> As mlx5dv_query_devx_port is being removed all the dependencies on
> the HAVE_MLX5DV_DR_DEVX_PORT compilation flag are reconsidered.
May be interesting to record that the new compilation flag
HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT is for OFED only,
and is required for compatibility with a future OFED version.
> Fixes: 6cfe84fbe7b1 ("net/mlx5: fix port action for LAG")
> Cc: stable@dpdk.org
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/2] common/mlx5: add provider query port support to glue library
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 " Viacheslav Ovsiienko
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
@ 2021-07-07 17:12 ` Thomas Monjalon
2021-07-08 16:07 ` Raslan Darawsheh
2 siblings, 0 replies; 16+ messages in thread
From: Thomas Monjalon @ 2021-07-07 17:12 UTC (permalink / raw)
To: Viacheslav Ovsiienko; +Cc: dev, rasland, matan, david.marchand, stable
The title could be:
common/mlx5: use new port query API if available
07/07/2021 17:54, Viacheslav Ovsiienko:
> In order to get E-Switch vport identifiers the mlx5 PMD relies
> on two approaches:
> [a] use port query API if it is provided by rdma-core library
> [b] otherwise, deduce vport ids from the related VF index
> The latter is not reliable and may not work with newer kernel
> drivers and in some configurations (LAG), causing E-Switch
> malfunction. Hence, engaging the port query API is highly
> desirable.
>
> Depending on rdma-core version the port query API is:
> - very old OFED versions have no query API (approach [b])
> - rdma-core OFED < 5.5 provides mlx5dv_query_devx_port,
> HAVE_MLX5DV_DR_DEVX_PORT flag is defined (approach [a])
> - rdma-core OFED >= 5.5 has mlx5dv_query_port, flag
> HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
> - future OFED versions might remove mlx5dv_query_devx_port
> and HAVE_MLX5DV_DR_DEVX_PORT will not be defined
> - Upstream rdma-core < v35 has no port query API (approach [b])
> - Upstream rdma-core >= v35 has mlx5dv_query_port, flag
> HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
>
> In order to support the new mlx5dv_query_port routine, the
> conditional compilation flag HAVE_MLX5DV_DR_DEVX_PORT_V35
> is introduced by the this patch. The flag HAVE_MLX5DV_DR_DEVX_PORT
s/the this/this/
> is kept for compatibility with previous rdma-core versions.
>
> Despite this patch is not a bugfix (it follows the introduced API
> variation in underlying library), it resolves the compatibility
> issue and is highly desired to be ported to DPDK LTS.
>
> Cc: stable@dpdk.org
>
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> Acked-by: Matan Azrad <matan@nvidia.com>
> ---
> drivers/common/mlx5/linux/meson.build | 2 +
> drivers/common/mlx5/linux/mlx5_glue.c | 55 ++++++++++++++++++++----
> drivers/common/mlx5/linux/mlx5_glue.h | 16 ++++++-
> drivers/net/mlx5/linux/mlx5_os.c | 60 ++++++++++++---------------
> 4 files changed, 89 insertions(+), 44 deletions(-)
>
> diff --git a/drivers/common/mlx5/linux/meson.build b/drivers/common/mlx5/linux/meson.build
> index 007834a49b..7168edc62c 100644
> --- a/drivers/common/mlx5/linux/meson.build
> +++ b/drivers/common/mlx5/linux/meson.build
> @@ -93,6 +93,8 @@ has_sym_args = [
> 'IBV_WQ_FLAG_RX_END_PADDING' ],
> [ 'HAVE_MLX5DV_DR_DEVX_PORT', 'infiniband/mlx5dv.h',
> 'mlx5dv_query_devx_port' ],
> + [ 'HAVE_MLX5DV_DR_DEVX_PORT_V35', 'infiniband/mlx5dv.h',
> + 'mlx5dv_query_port' ],
> [ 'HAVE_IBV_DEVX_OBJ', 'infiniband/mlx5dv.h',
> 'mlx5dv_devx_obj_create' ],
> [ 'HAVE_IBV_FLOW_DEVX_COUNTERS', 'infiniband/mlx5dv.h',
> diff --git a/drivers/common/mlx5/linux/mlx5_glue.c b/drivers/common/mlx5/linux/mlx5_glue.c
> index d3bd645a5b..38b174aa31 100644
> --- a/drivers/common/mlx5/linux/mlx5_glue.c
> +++ b/drivers/common/mlx5/linux/mlx5_glue.c
> @@ -1087,17 +1087,54 @@ mlx5_glue_devx_wq_query(struct ibv_wq *wq, const void *in, size_t inlen,
> static int
> mlx5_glue_devx_port_query(struct ibv_context *ctx,
> uint32_t port_num,
> - struct mlx5dv_devx_port *mlx5_devx_port)
> -{
> + struct mlx5_port_info *info)
> +{
> + int err = 0;
> +
> + info->query_flags = 0;
> +#ifdef HAVE_MLX5DV_DR_DEVX_PORT_V35
> + /* The DevX port query API is implemented (rdma-core v35 and above). */
> + struct mlx5_ib_uapi_query_port devx_port;
> +
> + memset(&devx_port, 0, sizeof(devx_port));
> + err = mlx5dv_query_port(ctx, port_num, &devx_port);
> + if (err)
> + return err;
> + if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT_REG_C0) {
> + info->vport_meta_tag = devx_port.reg_c0.value;
> + info->vport_meta_mask = devx_port.reg_c0.mask;
> + info->query_flags |= MLX5_PORT_QUERY_REG_C0;
> + }
> + if (devx_port.flags & MLX5DV_QUERY_PORT_VPORT) {
> + info->vport_id = devx_port.vport;
> + info->query_flags |= MLX5_PORT_QUERY_VPORT;
> + }
> +#else
> #ifdef HAVE_MLX5DV_DR_DEVX_PORT
> - return mlx5dv_query_devx_port(ctx, port_num, mlx5_devx_port);
> + /* The legacy DevX port query API is implemented (prior v35). */
> + struct mlx5dv_devx_port devx_port = {
> + .comp_mask = MLX5DV_DEVX_PORT_VPORT |
> + MLX5DV_DEVX_PORT_MATCH_REG_C_0
> + };
> +
> + err = mlx5dv_query_devx_port(ctx, port_num, &devx_port);
> + if (err)
> + return err;
> + if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
> + info->vport_meta_tag = devx_port.reg_c_0.value;
> + info->vport_meta_mask = devx_port.reg_c_0.mask;
> + info->query_flags |= MLX5_PORT_QUERY_REG_C0;
> + }
> + if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
> + info->vport_id = devx_port.vport_num;
> + info->query_flags |= MLX5_PORT_QUERY_VPORT;
> + }
> #else
> - (void)ctx;
> - (void)port_num;
> - (void)mlx5_devx_port;
> - errno = ENOTSUP;
> - return errno;
> -#endif
> + RTE_SET_USED(ctx);
> + RTE_SET_USED(port_num);
> +#endif /* HAVE_MLX5DV_DR_DEVX_PORT */
> +#endif /* HAVE_MLX5DV_DR_DEVX_PORT_V35 */
> + return err;
> }
>
> static int
> diff --git a/drivers/common/mlx5/linux/mlx5_glue.h b/drivers/common/mlx5/linux/mlx5_glue.h
> index 97462e9ab8..840d8cf57f 100644
> --- a/drivers/common/mlx5/linux/mlx5_glue.h
> +++ b/drivers/common/mlx5/linux/mlx5_glue.h
> @@ -84,6 +84,20 @@ struct mlx5dv_dr_action;
> struct mlx5dv_devx_port;
> #endif
>
> +#ifndef HAVE_MLX5DV_DR_DEVX_PORT_V35
> +struct mlx5dv_port;
> +#endif
> +
> +#define MLX5_PORT_QUERY_VPORT (1u << 0)
> +#define MLX5_PORT_QUERY_REG_C0 (1u << 1)
> +
> +struct mlx5_port_info {
> + uint16_t query_flags;
> + uint16_t vport_id; /* Associated VF vport index (if any). */
> + uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
> + uint32_t vport_meta_mask; /* Used for vport index field match mask. */
> +};
> +
> #ifndef HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER
> struct mlx5dv_dr_flow_meter_attr;
> #endif
> @@ -311,7 +325,7 @@ struct mlx5_glue {
> void *out, size_t outlen);
> int (*devx_port_query)(struct ibv_context *ctx,
> uint32_t port_num,
> - struct mlx5dv_devx_port *mlx5_devx_port);
> + struct mlx5_port_info *info);
> int (*dr_dump_domain)(FILE *file, void *domain);
> int (*dr_dump_rule)(FILE *file, void *rule);
> int (*devx_query_eqn)(struct ibv_context *context, uint32_t cpus,
> diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
> index 302a881310..92b3009786 100644
> --- a/drivers/net/mlx5/linux/mlx5_os.c
> +++ b/drivers/net/mlx5/linux/mlx5_os.c
> @@ -822,9 +822,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> char name[RTE_ETH_NAME_MAX_LEN];
> int own_domain_id = 0;
> uint16_t port_id;
> -#ifdef HAVE_MLX5DV_DR_DEVX_PORT
> - struct mlx5dv_devx_port devx_port = { .comp_mask = 0 };
> -#endif
> + struct mlx5_port_info vport_info = { .query_flags = 0 };
>
> /* Determine if this port representor is supposed to be spawned. */
> if (switch_info->representor && dpdk_dev->devargs &&
> @@ -1055,29 +1053,27 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> priv->vport_meta_tag = 0;
> priv->vport_meta_mask = 0;
> priv->pf_bond = spawn->pf_bond;
> -#ifdef HAVE_MLX5DV_DR_DEVX_PORT
> /*
> - * The DevX port query API is implemented. E-Switch may use
> - * either vport or reg_c[0] metadata register to match on
> - * vport index. The engaged part of metadata register is
> - * defined by mask.
> + * If we have E-Switch we should determine the vport attributes.
> + * E-Switch may use either source vport field or reg_c[0] metadata
> + * register to match on vport index. The engaged part of metadata
> + * register is defined by mask.
> */
> if (switch_info->representor || switch_info->master) {
> - devx_port.comp_mask = MLX5DV_DEVX_PORT_VPORT |
> - MLX5DV_DEVX_PORT_MATCH_REG_C_0;
> - err = mlx5_glue->devx_port_query(sh->ctx, spawn->phys_port,
> - &devx_port);
> + err = mlx5_glue->devx_port_query(sh->ctx,
> + spawn->phys_port,
> + &vport_info);
> if (err) {
> DRV_LOG(WARNING,
> "can't query devx port %d on device %s",
> spawn->phys_port,
> mlx5_os_get_dev_device_name(spawn->phys_dev));
> - devx_port.comp_mask = 0;
> + vport_info.query_flags = 0;
> }
> }
> - if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
> - priv->vport_meta_tag = devx_port.reg_c_0.value;
> - priv->vport_meta_mask = devx_port.reg_c_0.mask;
> + if (vport_info.query_flags & MLX5_PORT_QUERY_REG_C0) {
> + priv->vport_meta_tag = vport_info.vport_meta_tag;
> + priv->vport_meta_mask = vport_info.vport_meta_mask;
> if (!priv->vport_meta_mask) {
> DRV_LOG(ERR, "vport zero mask for port %d"
> " on bonding device %s",
> @@ -1097,8 +1093,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> goto error;
> }
> }
> - if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
> - priv->vport_id = devx_port.vport_num;
> + if (vport_info.query_flags & MLX5_PORT_QUERY_VPORT) {
> + priv->vport_id = vport_info.vport_id;
> } else if (spawn->pf_bond >= 0 &&
> (switch_info->representor || switch_info->master)) {
> DRV_LOG(ERR, "can't deduce vport index for port %d"
> @@ -1108,25 +1104,21 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
> err = ENOTSUP;
> goto error;
> } else {
> - /* Suppose vport index in compatible way. */
> + /*
> + * Suppose vport index in compatible way. Kernel/rdma_core
> + * support single E-Switch per PF configurations only and
> + * vport_id field contains the vport index for associated VF,
> + * which is deduced from representor port name.
> + * For example, let's have the IB device port 10, it has
> + * attached network device eth0, which has port name attribute
> + * pf0vf2, we can deduce the VF number as 2, and set vport index
> + * as 3 (2+1). This assigning schema should be changed if the
> + * multiple E-Switch instances per PF configurations or/and PCI
> + * subfunctions are added.
> + */
> priv->vport_id = switch_info->representor ?
> switch_info->port_name + 1 : -1;
> }
> -#else
> - /*
> - * Kernel/rdma_core support single E-Switch per PF configurations
> - * only and vport_id field contains the vport index for
> - * associated VF, which is deduced from representor port name.
> - * For example, let's have the IB device port 10, it has
> - * attached network device eth0, which has port name attribute
> - * pf0vf2, we can deduce the VF number as 2, and set vport index
> - * as 3 (2+1). This assigning schema should be changed if the
> - * multiple E-Switch instances per PF configurations or/and PCI
> - * subfunctions are added.
> - */
> - priv->vport_id = switch_info->representor ?
> - switch_info->port_name + 1 : -1;
> -#endif
> priv->representor_id = mlx5_representor_id_encode(switch_info,
> eth_da->type);
> /*
>
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [dpdk-dev] [PATCH v4 1/2] common/mlx5: add provider query port support to glue library
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 " Viacheslav Ovsiienko
2021-07-07 15:54 ` [dpdk-dev] [PATCH v4 2/2] common/mlx5: fix compilation flag on port query API Viacheslav Ovsiienko
2021-07-07 17:12 ` [dpdk-dev] [PATCH v4 1/2] common/mlx5: add provider query port support to glue library Thomas Monjalon
@ 2021-07-08 16:07 ` Raslan Darawsheh
2 siblings, 0 replies; 16+ messages in thread
From: Raslan Darawsheh @ 2021-07-08 16:07 UTC (permalink / raw)
To: Slava Ovsiienko, dev
Cc: Matan Azrad, NBU-Contact-Thomas Monjalon, david.marchand, stable
Hi,
> -----Original Message-----
> From: Slava Ovsiienko <viacheslavo@nvidia.com>
> Sent: Wednesday, July 7, 2021 6:54 PM
> To: dev@dpdk.org
> Cc: Raslan Darawsheh <rasland@nvidia.com>; Matan Azrad
> <matan@nvidia.com>; NBU-Contact-Thomas Monjalon
> <thomas@monjalon.net>; david.marchand@redhat.com; stable@dpdk.org
> Subject: [PATCH v4 1/2] common/mlx5: add provider query port support to
> glue library
>
> In order to get E-Switch vport identifiers the mlx5 PMD relies
> on two approaches:
> [a] use port query API if it is provided by rdma-core library
> [b] otherwise, deduce vport ids from the related VF index
> The latter is not reliable and may not work with newer kernel
> drivers and in some configurations (LAG), causing E-Switch
> malfunction. Hence, engaging the port query API is highly
> desirable.
>
> Depending on rdma-core version the port query API is:
> - very old OFED versions have no query API (approach [b])
> - rdma-core OFED < 5.5 provides mlx5dv_query_devx_port,
> HAVE_MLX5DV_DR_DEVX_PORT flag is defined (approach [a])
> - rdma-core OFED >= 5.5 has mlx5dv_query_port, flag
> HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
> - future OFED versions might remove mlx5dv_query_devx_port
> and HAVE_MLX5DV_DR_DEVX_PORT will not be defined
> - Upstream rdma-core < v35 has no port query API (approach [b])
> - Upstream rdma-core >= v35 has mlx5dv_query_port, flag
> HAVE_MLX5DV_DR_DEVX_PORT_V35 is defined (approach [a])
>
> In order to support the new mlx5dv_query_port routine, the
> conditional compilation flag HAVE_MLX5DV_DR_DEVX_PORT_V35
> is introduced by the this patch. The flag HAVE_MLX5DV_DR_DEVX_PORT
> is kept for compatibility with previous rdma-core versions.
>
> Despite this patch is not a bugfix (it follows the introduced API
> variation in underlying library), it resolves the compatibility
> issue and is highly desired to be ported to DPDK LTS.
>
> Cc: stable@dpdk.org
>
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> Acked-by: Matan Azrad <matan@nvidia.com>
Replaced older version and pushed to next-net-mlx,
Kindest regards,
Raslan Darawsheh
^ permalink raw reply [flat|nested] 16+ messages in thread