[PATCH V2 rdma-core 0/6] Enable mlx5 vendor functionality

[PATCH V2 rdma-core 0/6] Enable mlx5 vendor functionality

[Yishai Hadas]

Sending V2 to handle few notes from Jason, details below.

This patch set is the supplementary part of the kernel series that
was accepted few months ago into 4.10.

The below functionality was accepted in mlx5 kernel driver in the vendor
channel path as some driver specific capabilities and functionality, as such it
is exposed to user applications in the same manner by using the direct access
mechanism.

The vendor specific capabilities are reported via the direct mlx5 query
capabilities API.

In addition, an explicit API was added in mlx5 to enable creating a CQ with
some private data. This private data is passed to mlx5 kernel driver in the
vendor channel path.

This follows the requirement to have a more rigid, typed interface than
previous solution that was used on top of ibv_create_cq_ex().

To be able to share the libibverbs code of ibv_create_cq_ex() as of locking,
reference counting a verbs helper API was added, see patch #4 for further details.

Specifically this series:
Report whether mlx5 hardware supports multi packet send WQE offload for RAW QP.
- Multi packet WQE enables sending multiple fix sized packets using a single
  WQE. When the WQE is completed a single CQE will be generated.

Report CQE compressing capabilities and enable creating a CQ with such
capabilities.
- CQE compressing reduces PCI overhead by coalescing and compressing multiple
  CQEs into a single merged CQE. Successful compressing improves message rate
  especially for small packet traffic. 

The series is exposed from 6 patches:
#1:   Pre-patch which aligns mlx5 query device response to 64 bit.
#2-3: Report mlx5 specific hardware capabilities through direct verbs.
#4:   Enable an helper API for post CQ creation for providers.
#5:   Add support in mlx5 provider to create a CQ with compressed CQEs using
      private vendor data.
#6:   Bump package version and update debian relevant files.

Pull request was sent:
https://github.com/linux-rdma/rdma-core/pull/103

Changes from V1:
Patch #4:
- Move to use an helper function from verbs instead of using some provider
  callback.
- Bump the whole IBVERBS_PRIVATE section to 14.
	  
Patch #5:
- Move to use the verbs helper API.
- Naming changes. (replaced 'vendor' and dropped 'ex')
- Update libmlx5.map to use coming version number.(i.e. 14).

An additional patch #6 was added:
- Bump package version and update debian relevant files.
- As a result the relevant libraries symlinks (verb, mlx5) has 1.14 in
  the name as expected. (Tested in both RH & debian).

Bodong Wang (3):
  mlx5: Explicitly align mlx5 query device response to 64 bit
  mlx5: Report multi packet send WQE capability for mlx5 based hardware
  mlx5: Report CQE compression capabilities through mlx5 direct verbs

Yishai Hadas (3):
  verbs: Expose verbs init CQ helper API
  mlx5: Expose direct create_cq with private data
  Bump version and update debian

 CMakeLists.txt                           |  2 +-
 debian/changelog                         |  5 ++-
 debian/ibverbs-providers.symbols         |  2 ++
 debian/libibverbs1.symbols               |  2 +-
 libibverbs/device.c                      | 38 ++++++++++++++---------
 libibverbs/driver.h                      |  3 ++
 libibverbs/ibverbs.h                     |  2 +-
 libibverbs/libibverbs.map                |  3 +-
 libibverbs/verbs.c                       | 26 ++++------------
 providers/mlx5/libmlx5.map               |  5 +++
 providers/mlx5/man/mlx5dv_query_device.3 |  1 +
 providers/mlx5/mlx5-abi.h                | 15 +++++++--
 providers/mlx5/mlx5.c                    | 17 +++++++++--
 providers/mlx5/mlx5.h                    |  6 ++++
 providers/mlx5/mlx5dv.h                  | 33 +++++++++++++++++++-
 providers/mlx5/verbs.c                   | 52 +++++++++++++++++++++++++++++---
 16 files changed, 163 insertions(+), 49 deletions(-)

-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 1/6] mlx5: Explicitly align mlx5 query device response to 64 bit

[Yishai Hadas]

From: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Explicitly align packet pacing response to 64 bit so that further
capabilities that going to be read will match to mlx5 kernel response.

Signed-off-by: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/mlx5-abi.h | 7 ++++++-
 providers/mlx5/verbs.c    | 2 +-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index e6ea428..dde46b9 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -264,6 +264,11 @@ struct mlx5_rss_caps {
 	__u8 reserved[7];
 };
 
+struct mlx5_packet_pacing_caps {
+	struct ibv_packet_pacing_caps caps;
+	__u32  reserved;
+};
+
 struct mlx5_query_device_ex_resp {
 	struct ibv_query_device_resp_ex ibv_resp;
 	__u32				comp_mask;
@@ -271,7 +276,7 @@ struct mlx5_query_device_ex_resp {
 	struct ibv_tso_caps		tso_caps;
 	struct mlx5_rss_caps            rss_caps; /* vendor data channel */
 	__u64				reserved_cqe_comp;
-	struct ibv_packet_pacing_caps	packet_pacing_caps;
+	struct mlx5_packet_pacing_caps	packet_pacing_caps;
 };
 
 #endif /* MLX5_ABI_H */
diff --git a/providers/mlx5/verbs.c b/providers/mlx5/verbs.c
index 67f9748..aba1631 100644
--- a/providers/mlx5/verbs.c
+++ b/providers/mlx5/verbs.c
@@ -1922,7 +1922,7 @@ int mlx5_query_device_ex(struct ibv_context *context,
 	attr->tso_caps = resp.tso_caps;
 	attr->rss_caps.rx_hash_fields_mask = resp.rss_caps.rx_hash_fields_mask;
 	attr->rss_caps.rx_hash_function = resp.rss_caps.rx_hash_function;
-	attr->packet_pacing_caps = resp.packet_pacing_caps;
+	attr->packet_pacing_caps = resp.packet_pacing_caps.caps;
 
 	major     = (raw_fw_ver >> 32) & 0xffff;
 	minor     = (raw_fw_ver >> 16) & 0xffff;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 2/6] mlx5: Report multi packet send WQE capability for mlx5 based hardware

[Yishai Hadas]

From: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Report whether multi packet send WQE is supported or not through mlx5
direct verbs.

Signed-off-by: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/man/mlx5dv_query_device.3 | 1 +
 providers/mlx5/mlx5-abi.h                | 2 ++
 providers/mlx5/mlx5.c                    | 8 +++++++-
 providers/mlx5/mlx5.h                    | 5 +++++
 providers/mlx5/mlx5dv.h                  | 3 ++-
 providers/mlx5/verbs.c                   | 3 +++
 6 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/providers/mlx5/man/mlx5dv_query_device.3 b/providers/mlx5/man/mlx5dv_query_device.3
index 1954714..b33a75b 100644
--- a/providers/mlx5/man/mlx5dv_query_device.3
+++ b/providers/mlx5/man/mlx5dv_query_device.3
@@ -35,6 +35,7 @@ enum mlx5dv_context_flags {
  * This flag indicates if CQE version 0 or 1 is needed.
  */
  MLX5DV_CONTEXT_FLAGS_CQE_V1 = (1 << 0),
+ MLX5DV_CONTEXT_FLAGS_MPW    = (1 << 1), /* Multi packet WQE is supported or not */
 .in -8
 };
 .fi
diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index dde46b9..3574594 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -277,6 +277,8 @@ struct mlx5_query_device_ex_resp {
 	struct mlx5_rss_caps            rss_caps; /* vendor data channel */
 	__u64				reserved_cqe_comp;
 	struct mlx5_packet_pacing_caps	packet_pacing_caps;
+	__u32				support_multi_pkt_send_wqe;
+	__u32				reserved;
 };
 
 #endif /* MLX5_ABI_H */
diff --git a/providers/mlx5/mlx5.c b/providers/mlx5/mlx5.c
index eeaf5ac..7af6a0a 100644
--- a/providers/mlx5/mlx5.c
+++ b/providers/mlx5/mlx5.c
@@ -612,13 +612,18 @@ static int mlx5_map_internal_clock(struct mlx5_device *mdev,
 int mlx5dv_query_device(struct ibv_context *ctx_in,
 			 struct mlx5dv_context *attrs_out)
 {
+	struct mlx5_context *mctx = to_mctx(ctx_in);
+
 	attrs_out->comp_mask = 0;
 	attrs_out->version   = 0;
 	attrs_out->flags     = 0;
 
-	if (to_mctx(ctx_in)->cqe_version == MLX5_CQE_VERSION_V1)
+	if (mctx->cqe_version == MLX5_CQE_VERSION_V1)
 		attrs_out->flags |= MLX5DV_CONTEXT_FLAGS_CQE_V1;
 
+	if (mctx->vendor_cap_flags & MLX5_VENDOR_CAP_FLAGS_MPW)
+		attrs_out->flags |= MLX5DV_CONTEXT_FLAGS_MPW;
+
 	return 0;
 }
 
@@ -827,6 +832,7 @@ static int mlx5_init_context(struct verbs_device *vdev,
 	}
 
 	context->cmds_supp_uhw = resp.cmds_supp_uhw;
+	context->vendor_cap_flags = 0;
 
 	pthread_mutex_init(&context->qp_table_mutex, NULL);
 	pthread_mutex_init(&context->srq_table_mutex, NULL);
diff --git a/providers/mlx5/mlx5.h b/providers/mlx5/mlx5.h
index 57ed1a1..86f2438 100644
--- a/providers/mlx5/mlx5.h
+++ b/providers/mlx5/mlx5.h
@@ -180,6 +180,10 @@ enum {
 	MLX5_USER_CMDS_SUPP_UHW_CREATE_AH    = 1 << 1,
 };
 
+enum mlx5_vendor_cap_flags {
+	MLX5_VENDOR_CAP_FLAGS_MPW		= 1 << 0,
+};
+
 struct mlx5_resource {
 	enum mlx5_rsc_type	type;
 	uint32_t		rsn;
@@ -258,6 +262,7 @@ struct mlx5_context {
 	struct ibv_tso_caps		cached_tso_caps;
 	int				cmds_supp_uhw;
 	uint32_t			uar_size;
+	uint64_t			vendor_cap_flags; /* Use enum mlx5_vendor_cap_flags */
 };
 
 struct mlx5_bitmap {
diff --git a/providers/mlx5/mlx5dv.h b/providers/mlx5/mlx5dv.h
index 2e4995f..c09fad2 100644
--- a/providers/mlx5/mlx5dv.h
+++ b/providers/mlx5/mlx5dv.h
@@ -67,7 +67,8 @@ enum mlx5dv_context_flags {
 	/*
 	 * This flag indicates if CQE version 0 or 1 is needed.
 	 */
-	MLX5DV_CONTEXT_FLAGS_CQE_V1 = (1 << 0),
+	MLX5DV_CONTEXT_FLAGS_CQE_V1	= (1 << 0),
+	MLX5DV_CONTEXT_FLAGS_MPW	= (1 << 1),
 };
 
 /*
diff --git a/providers/mlx5/verbs.c b/providers/mlx5/verbs.c
index aba1631..cea9e46 100644
--- a/providers/mlx5/verbs.c
+++ b/providers/mlx5/verbs.c
@@ -1924,6 +1924,9 @@ int mlx5_query_device_ex(struct ibv_context *context,
 	attr->rss_caps.rx_hash_function = resp.rss_caps.rx_hash_function;
 	attr->packet_pacing_caps = resp.packet_pacing_caps.caps;
 
+	if (resp.support_multi_pkt_send_wqe)
+		mctx->vendor_cap_flags |= MLX5_VENDOR_CAP_FLAGS_MPW;
+
 	major     = (raw_fw_ver >> 32) & 0xffff;
 	minor     = (raw_fw_ver >> 16) & 0xffff;
 	sub_minor = raw_fw_ver & 0xffff;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 3/6] mlx5: Report CQE compression capabilities through mlx5 direct verbs

[Yishai Hadas]

From: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Report max number of compressed and aggregated CQEs in a single
merged CQE and the formats of mini CQE supported for responder.

Signed-off-by: Bodong Wang <bodong-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/mlx5-abi.h |  3 ++-
 providers/mlx5/mlx5.c     |  9 ++++++++-
 providers/mlx5/mlx5.h     |  1 +
 providers/mlx5/mlx5dv.h   | 17 +++++++++++++++++
 providers/mlx5/verbs.c    |  2 ++
 5 files changed, 30 insertions(+), 2 deletions(-)

diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index 3574594..487de3a 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -35,6 +35,7 @@
 
 #include <infiniband/kern-abi.h>
 #include <infiniband/verbs.h>
+#include "mlx5dv.h"
 
 #define MLX5_UVERBS_MIN_ABI_VERSION	1
 #define MLX5_UVERBS_MAX_ABI_VERSION	1
@@ -275,7 +276,7 @@ struct mlx5_query_device_ex_resp {
 	__u32				response_length;
 	struct ibv_tso_caps		tso_caps;
 	struct mlx5_rss_caps            rss_caps; /* vendor data channel */
-	__u64				reserved_cqe_comp;
+	struct mlx5dv_cqe_comp_caps	cqe_comp_caps;
 	struct mlx5_packet_pacing_caps	packet_pacing_caps;
 	__u32				support_multi_pkt_send_wqe;
 	__u32				reserved;
diff --git a/providers/mlx5/mlx5.c b/providers/mlx5/mlx5.c
index 7af6a0a..edb6d35 100644
--- a/providers/mlx5/mlx5.c
+++ b/providers/mlx5/mlx5.c
@@ -613,8 +613,8 @@ int mlx5dv_query_device(struct ibv_context *ctx_in,
 			 struct mlx5dv_context *attrs_out)
 {
 	struct mlx5_context *mctx = to_mctx(ctx_in);
+	uint64_t comp_mask_out = 0;
 
-	attrs_out->comp_mask = 0;
 	attrs_out->version   = 0;
 	attrs_out->flags     = 0;
 
@@ -624,6 +624,13 @@ int mlx5dv_query_device(struct ibv_context *ctx_in,
 	if (mctx->vendor_cap_flags & MLX5_VENDOR_CAP_FLAGS_MPW)
 		attrs_out->flags |= MLX5DV_CONTEXT_FLAGS_MPW;
 
+	if (attrs_out->comp_mask & MLX5DV_CONTEXT_MASK_CQE_COMPRESION) {
+		attrs_out->cqe_comp_caps = mctx->cqe_comp_caps;
+		comp_mask_out |= MLX5DV_CONTEXT_MASK_CQE_COMPRESION;
+	}
+
+	attrs_out->comp_mask = comp_mask_out;
+
 	return 0;
 }
 
diff --git a/providers/mlx5/mlx5.h b/providers/mlx5/mlx5.h
index 86f2438..bb383d4 100644
--- a/providers/mlx5/mlx5.h
+++ b/providers/mlx5/mlx5.h
@@ -263,6 +263,7 @@ struct mlx5_context {
 	int				cmds_supp_uhw;
 	uint32_t			uar_size;
 	uint64_t			vendor_cap_flags; /* Use enum mlx5_vendor_cap_flags */
+	struct mlx5dv_cqe_comp_caps	cqe_comp_caps;
 };
 
 struct mlx5_bitmap {
diff --git a/providers/mlx5/mlx5dv.h b/providers/mlx5/mlx5dv.h
index c09fad2..a2f7acf 100644
--- a/providers/mlx5/mlx5dv.h
+++ b/providers/mlx5/mlx5dv.h
@@ -54,6 +54,16 @@ enum {
 	MLX5_SND_DBR	= 1,
 };
 
+enum mlx5dv_context_comp_mask {
+	MLX5DV_CONTEXT_MASK_CQE_COMPRESION	= 1 << 0,
+	MLX5DV_CONTEXT_MASK_RESERVED		= 1 << 1,
+};
+
+struct mlx5dv_cqe_comp_caps {
+	uint32_t max_num;
+	uint32_t supported_format; /* enum mlx5dv_cqe_comp_res_format */
+};
+
 /*
  * Direct verbs device-specific attributes
  */
@@ -61,6 +71,7 @@ struct mlx5dv_context {
 	uint8_t		version;
 	uint64_t	flags;
 	uint64_t	comp_mask;
+	struct mlx5dv_cqe_comp_caps	cqe_comp_caps;
 };
 
 enum mlx5dv_context_flags {
@@ -277,6 +288,12 @@ struct mlx5_cqe64 {
 	uint8_t		op_own;
 };
 
+enum mlx5dv_cqe_comp_res_format {
+	MLX5DV_CQE_RES_FORMAT_HASH		= 1 << 0,
+	MLX5DV_CQE_RES_FORMAT_CSUM		= 1 << 1,
+	MLX5DV_CQE_RES_FORMAT_RESERVED		= 1 << 2,
+};
+
 static MLX5DV_ALWAYS_INLINE
 uint8_t mlx5dv_get_cqe_owner(struct mlx5_cqe64 *cqe)
 {
diff --git a/providers/mlx5/verbs.c b/providers/mlx5/verbs.c
index cea9e46..a223291 100644
--- a/providers/mlx5/verbs.c
+++ b/providers/mlx5/verbs.c
@@ -1927,6 +1927,8 @@ int mlx5_query_device_ex(struct ibv_context *context,
 	if (resp.support_multi_pkt_send_wqe)
 		mctx->vendor_cap_flags |= MLX5_VENDOR_CAP_FLAGS_MPW;
 
+	mctx->cqe_comp_caps = resp.cqe_comp_caps;
+
 	major     = (raw_fw_ver >> 32) & 0xffff;
 	minor     = (raw_fw_ver >> 16) & 0xffff;
 	sub_minor = raw_fw_ver & 0xffff;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 4/6] verbs: Expose verbs init CQ helper API

[Yishai Hadas]

This patch exposes to providers an helper function to init a CQ
post its creation.

The internal library code was refactored to use this helper
API to have a shared code for all creation flows (i.e. create_cq and
create_cq_ex).

As part of this refactoring reduced the scope of the lock that was taken
as part of creating/destroying a CQ to protect only what it was intended
to do. (i.e. channel->refcnt)

It's the application responsibility not to call concurrently to create a
CQ with a given completion channel and in the same time destroy it from
other thread, no change at that point from the original code.

The helper function going to be used in downstream patch in this
series.

Signed-off-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 libibverbs/device.c       | 38 +++++++++++++++++++++++---------------
 libibverbs/driver.h       |  3 +++
 libibverbs/ibverbs.h      |  2 +-
 libibverbs/libibverbs.map |  3 ++-
 libibverbs/verbs.c        | 26 ++++++--------------------
 5 files changed, 35 insertions(+), 37 deletions(-)

diff --git a/libibverbs/device.c b/libibverbs/device.c
index 6ef670a..ebf4970 100644
--- a/libibverbs/device.c
+++ b/libibverbs/device.c
@@ -132,6 +132,26 @@ __be64 __ibv_get_device_guid(struct ibv_device *device)
 }
 default_symver(__ibv_get_device_guid, ibv_get_device_guid);
 
+void verbs_init_cq(struct ibv_cq *cq, struct ibv_context *context,
+		       struct ibv_comp_channel *channel,
+		       void *cq_context)
+{
+	cq->context		   = context;
+	cq->channel		   = channel;
+
+	if (cq->channel) {
+		pthread_mutex_lock(&context->mutex);
+		++cq->channel->refcnt;
+		pthread_mutex_unlock(&context->mutex);
+	}
+
+	cq->cq_context		   = cq_context;
+	cq->comp_events_completed  = 0;
+	cq->async_events_completed = 0;
+	pthread_mutex_init(&cq->mutex, NULL);
+	pthread_cond_init(&cq->cond, NULL);
+}
+
 static struct ibv_cq_ex *
 __lib_ibv_create_cq_ex(struct ibv_context *context,
 		       struct ibv_cq_init_attr_ex *cq_attr)
@@ -144,23 +164,11 @@ __lib_ibv_create_cq_ex(struct ibv_context *context,
 		return NULL;
 	}
 
-	pthread_mutex_lock(&context->mutex);
-
 	cq = vctx->priv->create_cq_ex(context, cq_attr);
 
-	if (cq) {
-		cq->context		   = context;
-		cq->channel		   = cq_attr->channel;
-		if (cq->channel)
-			++cq->channel->refcnt;
-		cq->cq_context		   = cq_attr->cq_context;
-		cq->comp_events_completed  = 0;
-		cq->async_events_completed = 0;
-		pthread_mutex_init(&cq->mutex, NULL);
-		pthread_cond_init(&cq->cond, NULL);
-	}
-
-	pthread_mutex_unlock(&context->mutex);
+	if (cq)
+		verbs_init_cq(ibv_cq_ex_to_cq(cq), context,
+			        cq_attr->channel, cq_attr->cq_context);
 
 	return cq;
 }
diff --git a/libibverbs/driver.h b/libibverbs/driver.h
index e3ac3f7..ec87afd 100644
--- a/libibverbs/driver.h
+++ b/libibverbs/driver.h
@@ -130,6 +130,9 @@ verbs_get_device(const struct ibv_device *dev)
 typedef struct verbs_device *(*verbs_driver_init_func)(const char *uverbs_sys_path,
 						       int abi_version);
 void verbs_register_driver(const char *name, verbs_driver_init_func init_func);
+void verbs_init_cq(struct ibv_cq *cq, struct ibv_context *context,
+		       struct ibv_comp_channel *channel,
+		       void *cq_context);
 
 int ibv_cmd_get_context(struct ibv_context *context, struct ibv_get_context *cmd,
 			size_t cmd_size, struct ibv_get_context_resp *resp,
diff --git a/libibverbs/ibverbs.h b/libibverbs/ibverbs.h
index 3b08476..eb304ba 100644
--- a/libibverbs/ibverbs.h
+++ b/libibverbs/ibverbs.h
@@ -48,7 +48,7 @@
 #define default_symver(name, api)                                              \
 	asm(".symver " #name "," #api "@@" DEFAULT_ABI)
 #define private_symver(name, api)                                              \
-	asm(".symver " #name "," #api "@@IBVERBS_PRIVATE_13")
+	asm(".symver " #name "," #api "@@IBVERBS_PRIVATE_14")
 
 #define PFX		"libibverbs: "
 
diff --git a/libibverbs/libibverbs.map b/libibverbs/libibverbs.map
index 656c21d..5401241 100644
--- a/libibverbs/libibverbs.map
+++ b/libibverbs/libibverbs.map
@@ -86,7 +86,7 @@ IBVERBS_1.1 {
 
 /* If any symbols in this stanza change ABI then the entire staza gets a new symbol
    version. Also see the private_symver() macro */
-IBVERBS_PRIVATE_13 {
+IBVERBS_PRIVATE_14 {
 	global:
 		/* These historical symbols are now private to libibverbs */
 		ibv_cmd_alloc_mw;
@@ -138,4 +138,5 @@ IBVERBS_PRIVATE_13 {
 		ibv_query_gid_type;
 		ibv_register_driver;
 		verbs_register_driver;
+		verbs_init_cq;
 };
diff --git a/libibverbs/verbs.c b/libibverbs/verbs.c
index e84a2dd..e54fd77 100644
--- a/libibverbs/verbs.c
+++ b/libibverbs/verbs.c
@@ -441,23 +441,10 @@ struct ibv_cq *__ibv_create_cq(struct ibv_context *context, int cqe, void *cq_co
 {
 	struct ibv_cq *cq;
 
-	pthread_mutex_lock(&context->mutex);
-
 	cq = context->ops.create_cq(context, cqe, channel, comp_vector);
 
-	if (cq) {
-		cq->context    	     	   = context;
-		cq->channel		   = channel;
-		if (channel)
-			++channel->refcnt;
-		cq->cq_context 	     	   = cq_context;
-		cq->comp_events_completed  = 0;
-		cq->async_events_completed = 0;
-		pthread_mutex_init(&cq->mutex, NULL);
-		pthread_cond_init(&cq->cond, NULL);
-	}
-
-	pthread_mutex_unlock(&context->mutex);
+	if (cq)
+		verbs_init_cq(cq, context, channel, cq_context);
 
 	return cq;
 }
@@ -477,15 +464,14 @@ int __ibv_destroy_cq(struct ibv_cq *cq)
 	struct ibv_comp_channel *channel = cq->channel;
 	int ret;
 
-	if (channel)
-		pthread_mutex_lock(&channel->context->mutex);
-
 	ret = cq->context->ops.destroy_cq(cq);
 
 	if (channel) {
-		if (!ret)
+		if (!ret) {
+			pthread_mutex_lock(&channel->context->mutex);
 			--channel->refcnt;
-		pthread_mutex_unlock(&channel->context->mutex);
+			pthread_mutex_unlock(&channel->context->mutex);
+		}
 	}
 
 	return ret;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 5/6] mlx5: Expose direct create_cq with private data

[Yishai Hadas]

This patch exposes an API named mlx5dv_create_cq to enable creating a
CQ with some mlx5 private data.

Specifically, it enables creating a CQ in a mode that
few CQEs may be compressed into a single CQE.

To use the shared code of libibverbs when a CQ is created it uses the
verbs_init_cq() helper API that was introduced in previous patch.

Signed-off-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/libmlx5.map |  5 +++++
 providers/mlx5/mlx5-abi.h  |  3 +++
 providers/mlx5/mlx5dv.h    | 13 +++++++++++++
 providers/mlx5/verbs.c     | 45 ++++++++++++++++++++++++++++++++++++++++++---
 4 files changed, 63 insertions(+), 3 deletions(-)

diff --git a/providers/mlx5/libmlx5.map b/providers/mlx5/libmlx5.map
index 7ce3eb1..ffd7b5c 100644
--- a/providers/mlx5/libmlx5.map
+++ b/providers/mlx5/libmlx5.map
@@ -6,3 +6,8 @@ MLX5_1.0 {
 		mlx5dv_init_obj;
 	local: *;
 };
+
+MLX5_1.14 {
+	global:
+		mlx5dv_create_cq;
+} MLX5_1.0;
diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index 487de3a..d05cb40 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -119,6 +119,9 @@ struct mlx5_create_cq {
 	__u64				buf_addr;
 	__u64				db_addr;
 	__u32				cqe_size;
+	__u8                            cqe_comp_en;
+	__u8                            cqe_comp_res_format;
+	__u16                           reserved;
 };
 
 struct mlx5_create_cq_resp {
diff --git a/providers/mlx5/mlx5dv.h b/providers/mlx5/mlx5dv.h
index a2f7acf..a03b1d7 100644
--- a/providers/mlx5/mlx5dv.h
+++ b/providers/mlx5/mlx5dv.h
@@ -82,6 +82,19 @@ enum mlx5dv_context_flags {
 	MLX5DV_CONTEXT_FLAGS_MPW	= (1 << 1),
 };
 
+enum mlx5dv_cq_init_attr_mask {
+	MLX5DV_CQ_INIT_ATTR_MASK_COMPRESSED_CQE	= 1 << 0,
+	MLX5DV_CQ_INIT_ATTR_MASK_RESERVED	= 1 << 1,
+};
+
+struct mlx5dv_cq_init_attr {
+	uint64_t comp_mask; /* Use enum mlx5dv_cq_init_attr_mask */
+	uint8_t cqe_comp_res_format; /* Use enum mlx5dv_cqe_comp_res_format */
+};
+
+struct ibv_cq_ex *mlx5dv_create_cq(struct ibv_context *context,
+				   struct ibv_cq_init_attr_ex *cq_attr,
+				   struct mlx5dv_cq_init_attr *mlx5_cq_attr);
 /*
  * Most device capabilities are exported by ibv_query_device(...),
  * but there is HW device-specific information which is important
diff --git a/providers/mlx5/verbs.c b/providers/mlx5/verbs.c
index a223291..b7a8502 100644
--- a/providers/mlx5/verbs.c
+++ b/providers/mlx5/verbs.c
@@ -341,7 +341,8 @@ enum {
 
 static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 				   const struct ibv_cq_init_attr_ex *cq_attr,
-				   int cq_alloc_flags)
+				   int cq_alloc_flags,
+				   struct mlx5dv_cq_init_attr *mlx5cq_attr)
 {
 	struct mlx5_create_cq		cmd;
 	struct mlx5_create_cq_resp	resp;
@@ -349,6 +350,7 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 	int				cqe_sz;
 	int				ret;
 	int				ncqe;
+	struct mlx5_context *mctx = to_mctx(context);
 	FILE *fp = to_mctx(context)->dbg_fp;
 
 	if (!cq_attr->cqe) {
@@ -428,6 +430,28 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 	cmd.db_addr  = (uintptr_t) cq->dbrec;
 	cmd.cqe_size = cqe_sz;
 
+	if (mlx5cq_attr) {
+		if (mlx5cq_attr->comp_mask & ~(MLX5DV_CQ_INIT_ATTR_MASK_RESERVED - 1)) {
+			mlx5_dbg(fp, MLX5_DBG_CQ,
+				   "Unsupported vendor comp_mask for create_cq\n");
+			errno = EINVAL;
+			goto err_db;
+		}
+
+		if (mlx5cq_attr->comp_mask & MLX5DV_CQ_INIT_ATTR_MASK_COMPRESSED_CQE) {
+			if (mctx->cqe_comp_caps.max_num &&
+			    (mlx5cq_attr->cqe_comp_res_format &
+			     mctx->cqe_comp_caps.supported_format)) {
+				cmd.cqe_comp_en = 1;
+				cmd.cqe_comp_res_format = mlx5cq_attr->cqe_comp_res_format;
+			} else {
+				mlx5_dbg(fp, MLX5_DBG_CQ, "CQE Compression is not supported\n");
+				errno = EINVAL;
+				goto err_db;
+			}
+		}
+	}
+
 	ret = ibv_cmd_create_cq(context, ncqe - 1, cq_attr->channel,
 				cq_attr->comp_vector,
 				ibv_cq_ex_to_cq(&cq->ibv_cq), &cmd.ibv_cmd,
@@ -478,14 +502,29 @@ struct ibv_cq *mlx5_create_cq(struct ibv_context *context, int cqe,
 		return NULL;
 	}
 
-	cq = create_cq(context, &cq_attr, 0);
+	cq = create_cq(context, &cq_attr, 0, NULL);
 	return cq ? ibv_cq_ex_to_cq(cq) : NULL;
 }
 
 struct ibv_cq_ex *mlx5_create_cq_ex(struct ibv_context *context,
 				    struct ibv_cq_init_attr_ex *cq_attr)
 {
-	return create_cq(context, cq_attr, MLX5_CQ_FLAGS_EXTENDED);
+	return create_cq(context, cq_attr, MLX5_CQ_FLAGS_EXTENDED, NULL);
+}
+
+struct ibv_cq_ex *mlx5dv_create_cq(struct ibv_context *context,
+				      struct ibv_cq_init_attr_ex *cq_attr,
+				      struct mlx5dv_cq_init_attr *mlx5_cq_attr)
+{
+	struct ibv_cq_ex *cq;
+
+	cq = create_cq(context, cq_attr, MLX5_CQ_FLAGS_EXTENDED, mlx5_cq_attr);
+	if (!cq)
+		return NULL;
+
+	verbs_init_cq(ibv_cq_ex_to_cq(cq), context,
+		      cq_attr->channel, cq_attr->cq_context);
+	return cq;
 }
 
 int mlx5_resize_cq(struct ibv_cq *ibcq, int cqe)
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH V2 rdma-core 6/6] Bump version and update debian

[Yishai Hadas]

Signed-off-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 CMakeLists.txt                   | 2 +-
 debian/changelog                 | 5 ++++-
 debian/ibverbs-providers.symbols | 2 ++
 debian/libibverbs1.symbols       | 2 +-
 4 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index da80344..24542a7 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -44,7 +44,7 @@ endif()
 set(PACKAGE_NAME "RDMA")
 
 # See Documentation/versioning.md
-set(PACKAGE_VERSION "13")
+set(PACKAGE_VERSION "14")
 
 #-------------------------
 # Basic standard paths
diff --git a/debian/changelog b/debian/changelog
index 3defc05..f47fcec 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,5 +1,8 @@
-rdma-core (13-1) unstable; urgency=low
+rdma-core (14-1) unstable; urgency=low
+
+  * Bump version.
 
+rdma-core (13-1) unstable; urgency=low
   * New version.
   * Adding debian/copyright.
   * Close ITP (Closes: #848971).
diff --git a/debian/ibverbs-providers.symbols b/debian/ibverbs-providers.symbols
index b114deb..9c10852 100644
--- a/debian/ibverbs-providers.symbols
+++ b/debian/ibverbs-providers.symbols
@@ -1,4 +1,6 @@
 libmlx5.so.1 ibverbs-providers #MINVER#
  MLX5_1.0@MLX5_1.0 13-1
+ MLX5_1.14@MLX5_1.14 14-1
  mlx5dv_init_obj@MLX5_1.0 13-1
  mlx5dv_query_device@MLX5_1.0 13-1
+ mlx5dv_create_cq@MLX5_1.14 14-1
diff --git a/debian/libibverbs1.symbols b/debian/libibverbs1.symbols
index d70afd0..b96ef46 100644
--- a/debian/libibverbs1.symbols
+++ b/debian/libibverbs1.symbols
@@ -1,7 +1,7 @@
 libibverbs.so.1 libibverbs1 #MINVER#
  IBVERBS_1.0@IBVERBS_1.0 1.1.6
  IBVERBS_1.1@IBVERBS_1.1 1.1.6
- (symver)IBVERBS_PRIVATE_13 13
+ (symver)IBVERBS_PRIVATE_14 14
  ibv_ack_async_event@IBVERBS_1.0 1.1.6
  ibv_ack_async_event@IBVERBS_1.1 1.1.6
  ibv_ack_cq_events@IBVERBS_1.0 1.1.6
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Jason Gunthorpe]

There are still more places that need updating, see

Documentation/versioning.md

Or ec419980a7890b57ceaf6fefa797a365f2f5fed3

Please follow the commit message pattern from that example..

> index 3defc05..f47fcec 100644
> +++ b/debian/changelog
> @@ -1,5 +1,8 @@
> -rdma-core (13-1) unstable; urgency=low
> +rdma-core (14-1) unstable; urgency=low
> +
> +  * Bump version.

I haven't been adding new change log entires in the past, this is just
a standin file - Debian will maintain their own in their .diff. Same
with rpm.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1043 bytes --]

On Thu, Mar 23, 2017 at 11:19:16AM -0600, Jason Gunthorpe wrote:
> There are still more places that need updating, see
>
> Documentation/versioning.md
>
> Or ec419980a7890b57ceaf6fefa797a365f2f5fed3
>
> Please follow the commit message pattern from that example..
>
> > index 3defc05..f47fcec 100644
> > +++ b/debian/changelog
> > @@ -1,5 +1,8 @@
> > -rdma-core (13-1) unstable; urgency=low
> > +rdma-core (14-1) unstable; urgency=low
> > +
> > +  * Bump version.
>
> I haven't been adding new change log entires in the past, this is just
> a standin file - Debian will maintain their own in their .diff. Same
> with rpm.

It was more convenient for me to bump version in this file too,
so new symbols in debian/*.symbol will have more clear minimal
version (14-1), which is close to rdma-core-14.

Thanks

>
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Jason Gunthorpe]

On Thu, Mar 23, 2017 at 09:11:55PM +0200, Leon Romanovsky wrote:
> > > index 3defc05..f47fcec 100644
> > > +++ b/debian/changelog
> > > @@ -1,5 +1,8 @@
> > > -rdma-core (13-1) unstable; urgency=low
> > > +rdma-core (14-1) unstable; urgency=low
> > > +
> > > +  * Bump version.
> >
> > I haven't been adding new change log entires in the past, this is just
> > a standin file - Debian will maintain their own in their .diff. Same
> > with rpm.
> 
> It was more convenient for me to bump version in this file too,
> so new symbols in debian/*.symbol will have more clear minimal
> version (14-1), which is close to rdma-core-14.

I meant, you still have to bump the version, I was just doing it in
place, because we are not actually keeping a changelog.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 940 bytes --]

On Thu, Mar 23, 2017 at 02:02:14PM -0600, Jason Gunthorpe wrote:
> On Thu, Mar 23, 2017 at 09:11:55PM +0200, Leon Romanovsky wrote:
> > > > index 3defc05..f47fcec 100644
> > > > +++ b/debian/changelog
> > > > @@ -1,5 +1,8 @@
> > > > -rdma-core (13-1) unstable; urgency=low
> > > > +rdma-core (14-1) unstable; urgency=low
> > > > +
> > > > +  * Bump version.
> > >
> > > I haven't been adding new change log entires in the past, this is just
> > > a standin file - Debian will maintain their own in their .diff. Same
> > > with rpm.
> >
> > It was more convenient for me to bump version in this file too,
> > so new symbols in debian/*.symbol will have more clear minimal
> > version (14-1), which is close to rdma-core-14.
>
> I meant, you still have to bump the version, I was just doing it in
> place, because we are not actually keeping a changelog.

And what about 13-1 in all *.symbols files? It will need to be changed
too.
>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Jason Gunthorpe]

On Fri, Mar 24, 2017 at 04:49:05PM +0300, Leon Romanovsky wrote:
> On Thu, Mar 23, 2017 at 02:02:14PM -0600, Jason Gunthorpe wrote:
> > On Thu, Mar 23, 2017 at 09:11:55PM +0200, Leon Romanovsky wrote:
> > > > > index 3defc05..f47fcec 100644
> > > > > +++ b/debian/changelog
> > > > > @@ -1,5 +1,8 @@
> > > > > -rdma-core (13-1) unstable; urgency=low
> > > > > +rdma-core (14-1) unstable; urgency=low
> > > > > +
> > > > > +  * Bump version.
> > > >
> > > > I haven't been adding new change log entires in the past, this is just
> > > > a standin file - Debian will maintain their own in their .diff. Same
> > > > with rpm.
> > >
> > > It was more convenient for me to bump version in this file too,
> > > so new symbols in debian/*.symbol will have more clear minimal
> > > version (14-1), which is close to rdma-core-14.
> >
> > I meant, you still have to bump the version, I was just doing it in
> > place, because we are not actually keeping a changelog.
> 
> And what about 13-1 in all *.symbols files? It will need to be changed
> too.

Do you need a changelog entry to have stuff in the symbols file?

BTW, there are bad looking things in travis with this PR:

parsechangelog/debian: warning:     debian/changelog(l5): found start of entry where expected more change data or trailer
LINE: rdma-core (13-1) unstable; urgency=low
parsechangelog/debian: warning:     debian/changelog(l5): found eof
where expected more change data or trailer

 libmlx5.so.1 ibverbs-providers #MINVER#
  MLX5_1.0@MLX5_1.0 13-1
  MLX5_1.14@MLX5_1.14 14-1
+ MLX5_1.1@MLX5_1.1 14-1
+ mlx5dv_create_cq@MLX5_1.1 14-1

Etc..

https://travis-ci.org/linux-rdma/rdma-core/builds/214329236

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2673 bytes --]

On Fri, Mar 24, 2017 at 12:09:14PM -0600, Jason Gunthorpe wrote:
> On Fri, Mar 24, 2017 at 04:49:05PM +0300, Leon Romanovsky wrote:
> > On Thu, Mar 23, 2017 at 02:02:14PM -0600, Jason Gunthorpe wrote:
> > > On Thu, Mar 23, 2017 at 09:11:55PM +0200, Leon Romanovsky wrote:
> > > > > > index 3defc05..f47fcec 100644
> > > > > > +++ b/debian/changelog
> > > > > > @@ -1,5 +1,8 @@
> > > > > > -rdma-core (13-1) unstable; urgency=low
> > > > > > +rdma-core (14-1) unstable; urgency=low
> > > > > > +
> > > > > > +  * Bump version.
> > > > >
> > > > > I haven't been adding new change log entires in the past, this is just
> > > > > a standin file - Debian will maintain their own in their .diff. Same
> > > > > with rpm.
> > > >
> > > > It was more convenient for me to bump version in this file too,
> > > > so new symbols in debian/*.symbol will have more clear minimal
> > > > version (14-1), which is close to rdma-core-14.
> > >
> > > I meant, you still have to bump the version, I was just doing it in
> > > place, because we are not actually keeping a changelog.
> >
> > And what about 13-1 in all *.symbols files? It will need to be changed
> > too.
>
> Do you need a changelog entry to have stuff in the symbols file?

I think that we need corresponding version in changelog.

If we change version to be 14-1, so all previous declarations in
ibverbs-providers.symbols will need to be updated, because version 13-1
won't exist anymore.

➜  rdma-core git:(master) cat debian/ibverbs-providers.symbols
libmlx5.so.1 ibverbs-providers #MINVER#
 MLX5_1.0@MLX5_1.0 13-1
 mlx5dv_init_obj@MLX5_1.0 13-1
 mlx5dv_query_device@MLX5_1.0 13-1

>
> BTW, there are bad looking things in travis with this PR:
>
> parsechangelog/debian: warning:     debian/changelog(l5): found start of entry where expected more change data or trailer
> LINE: rdma-core (13-1) unstable; urgency=low
> parsechangelog/debian: warning:     debian/changelog(l5): found eof
> where expected more change data or trailer
>
>  libmlx5.so.1 ibverbs-providers #MINVER#
>   MLX5_1.0@MLX5_1.0 13-1
>   MLX5_1.14@MLX5_1.14 14-1
> + MLX5_1.1@MLX5_1.1 14-1
> + mlx5dv_create_cq@MLX5_1.1 14-1
>
> Etc..

It is actual error, strange that travis passed while it was supposed to fail.

The commit https://github.com/linux-rdma/rdma-core/pull/103/commits/4ba27e66294f59db7c969bde1580e7aa8d718cd7
added MLX5_1.1, but in commit https://github.com/linux-rdma/rdma-core/pull/103/commits/9614ecdae2a6a5448cab819a057c703a570b3686
debian was updated with incorrect stanza (MLX5_1.14).

>
> https://travis-ci.org/linux-rdma/rdma-core/builds/214329236

>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2923 bytes --]

On Sat, Mar 25, 2017 at 10:13:26AM +0300, Leon Romanovsky wrote:
> On Fri, Mar 24, 2017 at 12:09:14PM -0600, Jason Gunthorpe wrote:
> > On Fri, Mar 24, 2017 at 04:49:05PM +0300, Leon Romanovsky wrote:
> > > On Thu, Mar 23, 2017 at 02:02:14PM -0600, Jason Gunthorpe wrote:
> > > > On Thu, Mar 23, 2017 at 09:11:55PM +0200, Leon Romanovsky wrote:
> > > > > > > index 3defc05..f47fcec 100644
> > > > > > > +++ b/debian/changelog
> > > > > > > @@ -1,5 +1,8 @@
> > > > > > > -rdma-core (13-1) unstable; urgency=low
> > > > > > > +rdma-core (14-1) unstable; urgency=low
> > > > > > > +
> > > > > > > +  * Bump version.
> > > > > >
> > > > > > I haven't been adding new change log entires in the past, this is just
> > > > > > a standin file - Debian will maintain their own in their .diff. Same
> > > > > > with rpm.
> > > > >
> > > > > It was more convenient for me to bump version in this file too,
> > > > > so new symbols in debian/*.symbol will have more clear minimal
> > > > > version (14-1), which is close to rdma-core-14.
> > > >
> > > > I meant, you still have to bump the version, I was just doing it in
> > > > place, because we are not actually keeping a changelog.
> > >
> > > And what about 13-1 in all *.symbols files? It will need to be changed
> > > too.
> >
> > Do you need a changelog entry to have stuff in the symbols file?
>
> I think that we need corresponding version in changelog.
>
> If we change version to be 14-1, so all previous declarations in
> ibverbs-providers.symbols will need to be updated, because version 13-1
> won't exist anymore.
>
> ➜  rdma-core git:(master) cat debian/ibverbs-providers.symbols
> libmlx5.so.1 ibverbs-providers #MINVER#
>  MLX5_1.0@MLX5_1.0 13-1
>  mlx5dv_init_obj@MLX5_1.0 13-1
>  mlx5dv_query_device@MLX5_1.0 13-1
>
> >
> > BTW, there are bad looking things in travis with this PR:
> >
> > parsechangelog/debian: warning:     debian/changelog(l5): found start of entry where expected more change data or trailer
> > LINE: rdma-core (13-1) unstable; urgency=low
> > parsechangelog/debian: warning:     debian/changelog(l5): found eof
> > where expected more change data or trailer
> >
> >  libmlx5.so.1 ibverbs-providers #MINVER#
> >   MLX5_1.0@MLX5_1.0 13-1
> >   MLX5_1.14@MLX5_1.14 14-1
> > + MLX5_1.1@MLX5_1.1 14-1
> > + mlx5dv_create_cq@MLX5_1.1 14-1
> >
> > Etc..
>
> It is actual error, strange that travis passed while it was supposed to fail.
>
> The commit https://github.com/linux-rdma/rdma-core/pull/103/commits/4ba27e66294f59db7c969bde1580e7aa8d718cd7
> added MLX5_1.1, but in commit https://github.com/linux-rdma/rdma-core/pull/103/commits/9614ecdae2a6a5448cab819a057c703a570b3686
> debian was updated with incorrect stanza (MLX5_1.14).

https://github.com/linux-rdma/rdma-core/pull/107

>
> >
> > https://travis-ci.org/linux-rdma/rdma-core/builds/214329236
>
> >
> > Jason



[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH V2 rdma-core 6/6] Bump version and update debian

[Jason Gunthorpe]

On Sat, Mar 25, 2017 at 10:13:26AM +0300, Leon Romanovsky wrote:
> > Do you need a changelog entry to have stuff in the symbols file?
> 
> I think that we need corresponding version in changelog.

There are lots of counter examples..

libibverbs.so.1 libibverbs1 #MINVER#
 IBVERBS_1.0@IBVERBS_1.0 1.1.6
 IBVERBS_1.1@IBVERBS_1.1 1.1.6

Eg we do not have 1.1.6 in the changelog

> > BTW, there are bad looking things in travis with this PR:
> >
> > parsechangelog/debian: warning:     debian/changelog(l5): found start of entry where expected more change data or trailer
> > LINE: rdma-core (13-1) unstable; urgency=low
> > parsechangelog/debian: warning:     debian/changelog(l5): found eof
> > where expected more change data or trailer
> >
> >  libmlx5.so.1 ibverbs-providers #MINVER#
> >   MLX5_1.0@MLX5_1.0 13-1
> >   MLX5_1.14@MLX5_1.14 14-1
> > + MLX5_1.1@MLX5_1.1 14-1
> > + mlx5dv_create_cq@MLX5_1.1 14-1
> >
> > Etc..
> 
> It is actual error, strange that travis passed while it was supposed to fail.

Maybe we should run dpkg-shlibdeps manually and somehow fail if this
situation happens... Can you look at that?

> The commit
> https://github.com/linux-rdma/rdma-core/pull/103/commits/4ba27e66294f59db7c969bde1580e7aa8d718cd7
> added MLX5_1.1, but in commit
> https://github.com/linux-rdma/rdma-core/pull/103/commits/9614ecdae2a6a5448cab819a057c703a570b3686
> debian was updated with incorrect stanza (MLX5_1.14).

Yes, we went back and forth on this unfortunately, I gave some bad
advice.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html