[PATCH rdma-core 0/5] Clock information

[PATCH rdma-core 0/5] Clock information

[Yishai Hadas]

This series from Feras is the supplementary part of the kernel series that was
sent upstream for 4.16.

The series enables user space applications to query the mlx5 core clock and use
it to translate timestamps from HCA core clock units to nanoseconds.

PR was sent:
https://github.com/linux-rdma/rdma-core/pull/284

Note:
The Travis error in GITHUB is some checkpatch false alarm that relates
to some macro usage.

Yishai

Feras Daoud (5):
  mlx5: Support for user space clock info
  mlx5: Add direct verbs support for query clock info
  mlx5: Add direct verb to translate device timestamp to nanoseconds
  verbs: Add support to read completion's timestamp in wall clock
    nanoseconds
  mlx5: Implement read_completion_wallclock_ns

 debian/ibverbs-providers.symbols           |   2 +
 libibverbs/man/ibv_create_cq_ex.3          |   9 +-
 libibverbs/verbs.h                         |  10 +-
 providers/mlx5/CMakeLists.txt              |   2 +-
 providers/mlx5/cq.c                        | 181 +++++++++++++++++++++++------
 providers/mlx5/libmlx5.map                 |   5 +
 providers/mlx5/man/CMakeLists.txt          |   2 +
 providers/mlx5/man/mlx5dv_get_clock_info.3 |  37 ++++++
 providers/mlx5/man/mlx5dv_query_device.3   |   2 +-
 providers/mlx5/man/mlx5dv_ts_to_ns.3       |  35 ++++++
 providers/mlx5/mlx5-abi.h                  |  24 +++-
 providers/mlx5/mlx5.c                      |  71 +++++++++++
 providers/mlx5/mlx5.h                      |   7 +-
 providers/mlx5/mlx5dv.h                    |  65 ++++++++++-
 providers/mlx5/verbs.c                     |  15 ++-
 15 files changed, 421 insertions(+), 46 deletions(-)
 create mode 100644 providers/mlx5/man/mlx5dv_get_clock_info.3
 create mode 100644 providers/mlx5/man/mlx5dv_ts_to_ns.3

-- 
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 rdma-core 1/5] mlx5: Support for user space clock info

[Yishai Hadas]

From: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

This patch maps the kernel driver's clock info page to userspace.
With this, users will be able to read the clock info directly from
the shared page without any system calls.

Signed-off-by: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Eitan Rabin <rabin-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/mlx5-abi.h | 20 +++++++++++++++++++-
 providers/mlx5/mlx5.c     | 27 +++++++++++++++++++++++++++
 providers/mlx5/mlx5.h     |  2 ++
 3 files changed, 48 insertions(+), 1 deletion(-)

diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index 7b96429..c5d323e 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -81,6 +81,23 @@ enum mlx5_ib_alloc_ucontext_resp_mask {
 	MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_CORE_CLOCK_OFFSET = 1UL << 0,
 };
 
+/* Bit indexes for the mlx5_alloc_ucontext_resp.clock_info_versions bitmap */
+enum {
+	MLX5_IB_CLOCK_INFO_V1	= 0,
+};
+
+struct mlx5_ib_clock_info {
+	__u32 sig;
+	__u32 resv;
+	__u64 nsec;
+	__u64 last_cycles;
+	__u64 frac;
+	__u32 mult;
+	__u32 shift;
+	__u64 mask;
+	__u64 overflow_period;
+};
+
 struct mlx5_alloc_ucontext_resp {
 	struct ibv_get_context_resp	ibv_resp;
 	__u32				qp_tab_size;
@@ -98,7 +115,8 @@ struct mlx5_alloc_ucontext_resp {
 	__u32				response_length;
 	__u8				cqe_version;
 	__u8				cmds_supp_uhw;
-	__u16				reserved2;
+	__u8				reserved2;
+	__u8				clock_info_versions;
 	__u64				hca_core_clock_offset;
 	__u32				log_uar_size;
 	__u32				num_uars_per_page;
diff --git a/providers/mlx5/mlx5.c b/providers/mlx5/mlx5.c
index 52eb4f9..28bb320 100644
--- a/providers/mlx5/mlx5.c
+++ b/providers/mlx5/mlx5.c
@@ -607,6 +607,23 @@ static int mlx5_map_internal_clock(struct mlx5_device *mdev,
 	return 0;
 }
 
+static void mlx5_map_clock_info(struct mlx5_device *mdev,
+				struct ibv_context *ibv_ctx)
+{
+	struct mlx5_context *context = to_mctx(ibv_ctx);
+	void *clock_info_page;
+	off_t offset = 0;
+
+	set_command(MLX5_MMAP_GET_CLOCK_INFO_CMD, &offset);
+	set_index(MLX5_IB_CLOCK_INFO_V1, &offset);
+	clock_info_page = mmap(NULL, mdev->page_size,
+			       PROT_READ, MAP_SHARED, ibv_ctx->cmd_fd,
+			       offset * mdev->page_size);
+
+	if (clock_info_page != MAP_FAILED)
+		context->clock_info_page = clock_info_page;
+}
+
 int mlx5dv_query_device(struct ibv_context *ctx_in,
 			 struct mlx5dv_context *attrs_out)
 {
@@ -1027,6 +1044,14 @@ static int mlx5_init_context(struct verbs_device *vdev,
 		mlx5_map_internal_clock(mdev, ctx);
 	}
 
+	context->clock_info_page = NULL;
+	if (resp.response_length + sizeof(resp.ibv_resp) >=
+	    offsetof(struct mlx5_alloc_ucontext_resp, clock_info_versions) +
+	    sizeof(resp.clock_info_versions) &&
+	    (resp.clock_info_versions & (1 << MLX5_IB_CLOCK_INFO_V1))) {
+		mlx5_map_clock_info(mdev, ctx);
+	}
+
 	mlx5_read_env(&vdev->device, context);
 
 	mlx5_spinlock_init(&context->hugetlb_lock);
@@ -1107,6 +1132,8 @@ static void mlx5_cleanup_context(struct verbs_device *device,
 	if (context->hca_core_clock)
 		munmap(context->hca_core_clock - context->core_clock.offset,
 		       page_size);
+	if (context->clock_info_page)
+		munmap((void *)context->clock_info_page, page_size);
 	close_debug_file(context);
 }
 
diff --git a/providers/mlx5/mlx5.h b/providers/mlx5/mlx5.h
index d8b1858..c0f342d 100644
--- a/providers/mlx5/mlx5.h
+++ b/providers/mlx5/mlx5.h
@@ -60,6 +60,7 @@ enum {
 	MLX5_MMAP_GET_CONTIGUOUS_PAGES_CMD = 1,
 	MLX5_MMAP_GET_CORE_CLOCK_CMD    = 5,
 	MLX5_MMAP_ALLOC_WC		= 6,
+	MLX5_MMAP_GET_CLOCK_INFO_CMD    = 7,
 };
 
 enum {
@@ -287,6 +288,7 @@ struct mlx5_context {
 		uint64_t                mask;
 	} core_clock;
 	void			       *hca_core_clock;
+	const struct mlx5_ib_clock_info *clock_info_page;
 	struct ibv_tso_caps		cached_tso_caps;
 	int				cmds_supp_uhw;
 	uint32_t			uar_size;
-- 
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 rdma-core 2/5] mlx5: Add direct verbs support for query clock info

[Yishai Hadas]

From: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Allow retrieving the current clock info with a direct verb.
This allows DV consumers to directly cache and control the safely copied
version of the clock data. A C11 atomic is used to ensure the copy of
the kernel data is done atomically.

A new mlx5dv attribute "max_clock_info_update_nsec" was added to
indicate the maximum period that the user can correctly use the cached
clock info, it can be retrieved by using the mlx5dv_query_device API.

Signed-off-by: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Eitan Rabin <rabin-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 debian/ibverbs-providers.symbols           |  2 ++
 providers/mlx5/CMakeLists.txt              |  2 +-
 providers/mlx5/libmlx5.map                 |  5 ++++
 providers/mlx5/man/CMakeLists.txt          |  1 +
 providers/mlx5/man/mlx5dv_get_clock_info.3 | 37 +++++++++++++++++++++++++
 providers/mlx5/man/mlx5dv_query_device.3   |  2 +-
 providers/mlx5/mlx5-abi.h                  |  4 +++
 providers/mlx5/mlx5.c                      | 44 ++++++++++++++++++++++++++++++
 providers/mlx5/mlx5dv.h                    | 25 ++++++++++++++++-
 9 files changed, 119 insertions(+), 3 deletions(-)
 create mode 100644 providers/mlx5/man/mlx5dv_get_clock_info.3

diff --git a/debian/ibverbs-providers.symbols b/debian/ibverbs-providers.symbols
index b3adfe6..dc87c07 100644
--- a/debian/ibverbs-providers.symbols
+++ b/debian/ibverbs-providers.symbols
@@ -9,6 +9,7 @@ libmlx5.so.1 ibverbs-providers #MINVER#
  MLX5_1.1@MLX5_1.1 14
  MLX5_1.2@MLX5_1.2 15
  MLX5_1.3@MLX5_1.3 16
+ MLX5_1.4@MLX5_1.4 17
  mlx5dv_init_obj@MLX5_1.0 13
  mlx5dv_init_obj@MLX5_1.2 15
  mlx5dv_query_device@MLX5_1.0 13
@@ -16,3 +17,4 @@ libmlx5.so.1 ibverbs-providers #MINVER#
  mlx5dv_set_context_attr@MLX5_1.2 15
  mlx5dv_create_qp@MLX5_1.3 16
  mlx5dv_create_wq@MLX5_1.3 16
+ mlx5dv_get_clock_info@MLX5_1.4 17
diff --git a/providers/mlx5/CMakeLists.txt b/providers/mlx5/CMakeLists.txt
index 88a406d..a8db43d 100644
--- a/providers/mlx5/CMakeLists.txt
+++ b/providers/mlx5/CMakeLists.txt
@@ -11,7 +11,7 @@ if (MLX5_MW_DEBUG)
 endif()
 
 rdma_shared_provider(mlx5 libmlx5.map
-  1 1.3.${PACKAGE_VERSION}
+  1 1.4.${PACKAGE_VERSION}
   buf.c
   cq.c
   dbrec.c
diff --git a/providers/mlx5/libmlx5.map b/providers/mlx5/libmlx5.map
index f797822..01fb983 100644
--- a/providers/mlx5/libmlx5.map
+++ b/providers/mlx5/libmlx5.map
@@ -23,3 +23,8 @@ MLX5_1.3 {
 		mlx5dv_create_qp;
 		mlx5dv_create_wq;
 } MLX5_1.2;
+
+MLX5_1.4 {
+	global:
+		mlx5dv_get_clock_info;
+} MLX5_1.3;
diff --git a/providers/mlx5/man/CMakeLists.txt b/providers/mlx5/man/CMakeLists.txt
index bb1610b..6ce94a9 100644
--- a/providers/mlx5/man/CMakeLists.txt
+++ b/providers/mlx5/man/CMakeLists.txt
@@ -1,4 +1,5 @@
 rdma_man_pages(
+  mlx5dv_get_clock_info.3
   mlx5dv_init_obj.3
   mlx5dv_query_device.3
   mlx5dv.7
diff --git a/providers/mlx5/man/mlx5dv_get_clock_info.3 b/providers/mlx5/man/mlx5dv_get_clock_info.3
new file mode 100644
index 0000000..194a32e
--- /dev/null
+++ b/providers/mlx5/man/mlx5dv_get_clock_info.3
@@ -0,0 +1,37 @@
+.\" -*- nroff -*-
+.\" Licensed under the OpenIB.org (MIT) - See COPYING.md
+.\"
+.TH MLX5DV_GET_CLOCK_INFO 3 2017-11-08 1.0.0
+.SH "NAME"
+mlx5dv_get_clock_info \- Get device clock information
+.SH "SYNOPSIS"
+.nf
+.B #include <infiniband/mlx5dv.h>
+.sp
+.BI "int mlx5dv_get_clock_info(struct ibv_context *ctx_in,
+.BI "                          struct mlx5dv_clock_info *clock_info);
+.fi
+.SH "DESCRIPTION"
+Get the updated core
+.I clock_info
+from the device driver. This information will be used later to translate the
+completion timestamp from HCA core clock to nanoseconds. The values of the clock are
+updated from the driver's PTP clock, therefore, without a running PTP
+client on the machine, the wall clock conversion will not be accurate.
+.PP
+Pass the latest \fBstruct mlx5dv_clock_info\fR to \fBmlx5dv_ts_to_ns(3)\fR in order to translate
+the completion timestamp from HCA core clock to nanoseconds.
+.PP
+If the clock_info becomes too old then time conversion will return wrong conversion results.
+The user must ensure that  \fBmlx5dv_get_clock_info(3)\fR is called at least once every
+\fBmax_clock_info_update_nsec\fR as returned by the \fBmlx5dv_query_device(3)\fR function.
+.PP
+.fi
+.SH "RETURN VALUE"
+0 on success or the value of errno on failure (which indicates the failure reason).
+.SH "SEE ALSO"
+.BR mlx5dv (7),
+.BR mlx5dv_ts_to_ns (3)
+.SH "AUTHORS"
+.TP
+Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
diff --git a/providers/mlx5/man/mlx5dv_query_device.3 b/providers/mlx5/man/mlx5dv_query_device.3
index 0cffc2b..522b054 100644
--- a/providers/mlx5/man/mlx5dv_query_device.3
+++ b/providers/mlx5/man/mlx5dv_query_device.3
@@ -75,7 +75,7 @@ MLX5DV_CONTEXT_MASK_SWP                 = 1 << 1,
 MLX5DV_CONTEXT_MASK_STRIDING_RQ         = 1 << 2,
 MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS     = 1 << 3,
 MLX5DV_CONTEXT_MASK_DYN_BFREGS          = 1 << 4,
-MLX5DV_CONTEXT_MASK_RESERVED            = 1 << 5,
+MLX5DV_CONTEXT_MASK_CLOCK_INFO_UPDATE   = 1 << 5,
 .in -8
 };
 
diff --git a/providers/mlx5/mlx5-abi.h b/providers/mlx5/mlx5-abi.h
index c5d323e..fb5ea0a 100644
--- a/providers/mlx5/mlx5-abi.h
+++ b/providers/mlx5/mlx5-abi.h
@@ -86,6 +86,10 @@ enum {
 	MLX5_IB_CLOCK_INFO_V1	= 0,
 };
 
+enum {
+	MLX5_IB_CLOCK_INFO_KERNEL_UPDATING = 1,
+};
+
 struct mlx5_ib_clock_info {
 	__u32 sig;
 	__u32 resv;
diff --git a/providers/mlx5/mlx5.c b/providers/mlx5/mlx5.c
index 28bb320..e71c8b0 100644
--- a/providers/mlx5/mlx5.c
+++ b/providers/mlx5/mlx5.c
@@ -673,6 +673,14 @@ int mlx5dv_query_device(struct ibv_context *ctx_in,
 		comp_mask_out |= MLX5DV_CONTEXT_MASK_DYN_BFREGS;
 	}
 
+	if (attrs_out->comp_mask & MLX5DV_CONTEXT_MASK_CLOCK_INFO_UPDATE) {
+		if (mctx->clock_info_page) {
+			attrs_out->max_clock_info_update_nsec =
+					mctx->clock_info_page->overflow_period;
+			comp_mask_out |= MLX5DV_CONTEXT_MASK_CLOCK_INFO_UPDATE;
+		}
+	}
+
 	attrs_out->comp_mask = comp_mask_out;
 
 	return 0;
@@ -868,6 +876,42 @@ int mlx5dv_set_context_attr(struct ibv_context *ibv_ctx,
 	return 0;
 }
 
+typedef _Atomic(uint32_t) atomic_uint32_t;
+
+int mlx5dv_get_clock_info(struct ibv_context *ctx_in,
+			  struct mlx5dv_clock_info *clock_info)
+{
+	struct mlx5_context *ctx = to_mctx(ctx_in);
+	const struct mlx5_ib_clock_info *ci = ctx->clock_info_page;
+	uint32_t retry, tmp_sig;
+	atomic_uint32_t *sig;
+
+	if (!ci)
+		return EINVAL;
+
+	sig = (atomic_uint32_t *)&ci->sig;
+
+	do {
+		retry = 10;
+repeat:
+		tmp_sig = atomic_load(sig);
+		if (unlikely(tmp_sig &
+			     MLX5_IB_CLOCK_INFO_KERNEL_UPDATING)) {
+			if (--retry)
+				goto repeat;
+			return EBUSY;
+		}
+		clock_info->nsec   = ci->nsec;
+		clock_info->last_cycles = ci->last_cycles;
+		clock_info->frac   = ci->frac;
+		clock_info->mult   = ci->mult;
+		clock_info->shift  = ci->shift;
+		clock_info->mask   = ci->mask;
+	} while (unlikely(tmp_sig != atomic_load(sig)));
+
+	return 0;
+}
+
 static void adjust_uar_info(struct mlx5_device *mdev,
 			    struct mlx5_context *context,
 			    struct mlx5_alloc_ucontext_resp resp)
diff --git a/providers/mlx5/mlx5dv.h b/providers/mlx5/mlx5dv.h
index 84c5099..2447517 100644
--- a/providers/mlx5/mlx5dv.h
+++ b/providers/mlx5/mlx5dv.h
@@ -63,7 +63,7 @@ enum mlx5dv_context_comp_mask {
 	MLX5DV_CONTEXT_MASK_STRIDING_RQ		= 1 << 2,
 	MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS	= 1 << 3,
 	MLX5DV_CONTEXT_MASK_DYN_BFREGS		= 1 << 4,
-	MLX5DV_CONTEXT_MASK_RESERVED		= 1 << 5,
+	MLX5DV_CONTEXT_MASK_CLOCK_INFO_UPDATE	= 1 << 5,
 };
 
 struct mlx5dv_cqe_comp_caps {
@@ -102,6 +102,7 @@ struct mlx5dv_context {
 	struct mlx5dv_striding_rq_caps striding_rq_caps;
 	uint32_t	tunnel_offloads_caps;
 	uint32_t	max_dynamic_bfregs;
+	uint64_t	max_clock_info_update_nsec;
 };
 
 enum mlx5dv_context_flags {
@@ -788,4 +789,26 @@ struct mlx5dv_ctx_allocators {
 int mlx5dv_set_context_attr(struct ibv_context *context,
 		enum mlx5dv_set_ctx_attr_type type, void *attr);
 
+struct mlx5dv_clock_info {
+	uint64_t nsec;
+	uint64_t last_cycles;
+	uint64_t frac;
+	uint32_t mult;
+	uint32_t shift;
+	uint64_t mask;
+};
+
+/*
+ * Get mlx5 core clock info
+ *
+ * Output:
+ *      clock_info  - clock info to be filled
+ * Input:
+ *      context     - device context
+ *
+ * Return: 0 on success, or the value of errno on failure
+ */
+int mlx5dv_get_clock_info(struct ibv_context *context,
+			  struct mlx5dv_clock_info *clock_info);
+
 #endif /* _MLX5DV_H_ */
-- 
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 rdma-core 3/5] mlx5: Add direct verb to translate device timestamp to nanoseconds

[Yishai Hadas]

From: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Add a direct verb that translates device timestamps from HCA
core clock units to nanoseconds using the device clock info.

Signed-off-by: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Eitan Rabin <rabin-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/man/CMakeLists.txt    |  1 +
 providers/mlx5/man/mlx5dv_ts_to_ns.3 | 35 +++++++++++++++++++++++++++++++
 providers/mlx5/mlx5dv.h              | 40 ++++++++++++++++++++++++++++++++++++
 3 files changed, 76 insertions(+)
 create mode 100644 providers/mlx5/man/mlx5dv_ts_to_ns.3

diff --git a/providers/mlx5/man/CMakeLists.txt b/providers/mlx5/man/CMakeLists.txt
index 6ce94a9..876b6fc 100644
--- a/providers/mlx5/man/CMakeLists.txt
+++ b/providers/mlx5/man/CMakeLists.txt
@@ -2,5 +2,6 @@ rdma_man_pages(
   mlx5dv_get_clock_info.3
   mlx5dv_init_obj.3
   mlx5dv_query_device.3
+  mlx5dv_ts_to_ns.3
   mlx5dv.7
 )
diff --git a/providers/mlx5/man/mlx5dv_ts_to_ns.3 b/providers/mlx5/man/mlx5dv_ts_to_ns.3
new file mode 100644
index 0000000..197b195
--- /dev/null
+++ b/providers/mlx5/man/mlx5dv_ts_to_ns.3
@@ -0,0 +1,35 @@
+.\" -*- nroff -*-
+.\" Licensed under the OpenIB.org (MIT) - See COPYING.md
+.\"
+.TH MLX5DV_TS_TO_NS 3 2017-11-08 1.0.0
+.SH "NAME"
+mlx5dv_ts_to_ns \- Convert device timestamp from HCA core clock units to
+the corresponding nanosecond counts
+.SH "SYNOPSIS"
+.nf
+.B #include <infiniband/mlx5dv.h>
+.sp
+.BI "uint64_t mlx5dv_ts_to_ns(struct mlx5dv_clock_info *clock_info,
+.BI "			      uint64_t device_timestamp);
+.fi
+.SH "DESCRIPTION"
+.B mlx5dv_ts_to_ns(3)
+Converts a host byte order
+.I device_timestamp
+from HCA core clock units into the corresponding nanosecond wallclock time.
+.PP
+\fBstruct mlx5dv_clock_info\fR can be retrieved using \fBmlx5dv_get_clock_info(3)\fR.
+.PP
+The greater the difference between the device reporting a timestamp and the last
+mlx5dv_clock_info update, the greater the inaccuracy of the clock time conversion.
+
+.fi
+.SH "RETURN VALUE"
+Timestamp in nanoseconds
+.SH "SEE ALSO"
+.BR mlx5dv (7),
+.BR mlx5dv_get_clock_info (3),
+.BR mlx5dv_query_device (3)
+.SH "AUTHORS"
+.TP
+Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
diff --git a/providers/mlx5/mlx5dv.h b/providers/mlx5/mlx5dv.h
index 2447517..2a4a823 100644
--- a/providers/mlx5/mlx5dv.h
+++ b/providers/mlx5/mlx5dv.h
@@ -811,4 +811,44 @@ struct mlx5dv_clock_info {
 int mlx5dv_get_clock_info(struct ibv_context *context,
 			  struct mlx5dv_clock_info *clock_info);
 
+/*
+ * Translate device timestamp to nano-sec
+ *
+ * Input:
+ *      clock_info  - clock info to be filled
+ *      device_timestamp   - timestamp to translate
+ *
+ * Return: nano-sec
+ */
+static inline uint64_t mlx5dv_ts_to_ns(struct mlx5dv_clock_info *clock_info,
+				       uint64_t device_timestamp)
+{
+	uint64_t delta, nsec;
+
+	/*
+	 * device_timestamp & cycles are the free running 'mask' bit counters
+	 * from the hardware hca_core_clock clock.
+	 */
+	delta = (device_timestamp - clock_info->last_cycles) & clock_info->mask;
+	nsec  = clock_info->nsec;
+
+	/*
+	 * Guess if the device_timestamp is more recent than
+	 * clock_info->last_cycles, if not (too far in the future) treat
+	 * it as old time stamp. This will break every max_clock_info_update_nsec.
+	 */
+
+	if (delta > clock_info->mask / 2) {
+		delta = (clock_info->last_cycles - device_timestamp) &
+				clock_info->mask;
+		nsec -= ((delta * clock_info->mult) - clock_info->frac) >>
+				clock_info->shift;
+	} else {
+		nsec += ((delta * clock_info->mult) + clock_info->frac) >>
+				clock_info->shift;
+	}
+
+	return nsec;
+}
+
 #endif /* _MLX5DV_H_ */
-- 
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 rdma-core 4/5] verbs: Add support to read completion's timestamp in wall clock nanoseconds

[Yishai Hadas]

From: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Add support to read completion timestamp and convert it from
HCA clock units to wallclock nanoseconds.

This commit revises the documentation for ibv_wc_read_completion_ts
to clarify the return value time units.

Signed-off-by: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Eitan Rabin <rabin-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 libibverbs/man/ibv_create_cq_ex.3 |  9 +++++++--
 libibverbs/verbs.h                | 10 +++++++++-
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/libibverbs/man/ibv_create_cq_ex.3 b/libibverbs/man/ibv_create_cq_ex.3
index 23f867c..51a6eb5 100644
--- a/libibverbs/man/ibv_create_cq_ex.3
+++ b/libibverbs/man/ibv_create_cq_ex.3
@@ -40,9 +40,10 @@ enum ibv_wc_flags_ex {
         IBV_WC_EX_WITH_SLID                  = 1 << 4,  /* Require slid in WC */
         IBV_WC_EX_WITH_SL                    = 1 << 5,  /* Require sl in WC */
         IBV_WC_EX_WITH_DLID_PATH_BITS        = 1 << 6,  /* Require dlid path bits in WC */
-        IBV_WC_EX_WITH_COMPLETION_TIMESTAMP  = 1 << 7,  /* Require completion timestamp in WC /*
+        IBV_WC_EX_WITH_COMPLETION_TIMESTAMP  = 1 << 7,  /* Require completion device timestamp in WC /*
         IBV_WC_EX_WITH_CVLAN                 = 1 << 8,  /* Require VLAN info in WC */
         IBV_WC_EX_WITH_FLOW_TAG		     = 1 << 9,  /* Require flow tag in WC */
+        IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK  = 1 << 11, /* Require completion wallclock timestamp in WC */
 };
 
 enum ibv_cq_init_attr_mask {
@@ -132,7 +133,11 @@ Below members and functions are used in order to poll the current completion. Th
  Get the dlid_path_bits field from the current completion.
 
 .BI "uint64_t ibv_wc_read_completion_ts(struct ibv_cq_ex " "*cq"); \c
- Get the completion timestamp from the current completion.
+ Get the completion timestamp from the current completion in HCA clock units.
+
+.BI "uint64_t ibv_wc_read_completion_wallclock_ns(struct ibv_cq_ex " *cq ");
+Get the completion timestamp from the current completion and convert it
+from HCA clock units to wall clock nanoseconds.
 
 .BI "uint16_t ibv_wc_read_cvlan(struct ibv_cq_ex " "*cq"); \c
  Get the CVLAN field from the current completion.
diff --git a/libibverbs/verbs.h b/libibverbs/verbs.h
index 997597a..ddda03b 100644
--- a/libibverbs/verbs.h
+++ b/libibverbs/verbs.h
@@ -466,6 +466,7 @@ enum ibv_create_cq_wc_flags {
 	IBV_WC_EX_WITH_CVLAN		= 1 << 8,
 	IBV_WC_EX_WITH_FLOW_TAG		= 1 << 9,
 	IBV_WC_EX_WITH_TM_INFO		= 1 << 10,
+	IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK	= 1 << 11,
 };
 
 enum {
@@ -483,7 +484,8 @@ enum {
 				IBV_WC_EX_WITH_COMPLETION_TIMESTAMP |
 				IBV_WC_EX_WITH_CVLAN |
 				IBV_WC_EX_WITH_FLOW_TAG |
-				IBV_WC_EX_WITH_TM_INFO
+				IBV_WC_EX_WITH_TM_INFO |
+				IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK
 };
 
 enum ibv_wc_flags {
@@ -1205,6 +1207,7 @@ struct ibv_cq_ex {
 	uint32_t (*read_flow_tag)(struct ibv_cq_ex *current);
 	void (*read_tm_info)(struct ibv_cq_ex *current,
 			     struct ibv_wc_tm_info *tm_info);
+	uint64_t (*read_completion_wallclock_ns)(struct ibv_cq_ex *current);
 };
 
 static inline struct ibv_cq *ibv_cq_ex_to_cq(struct ibv_cq_ex *cq)
@@ -1307,6 +1310,11 @@ static inline uint64_t ibv_wc_read_completion_ts(struct ibv_cq_ex *cq)
 	return cq->read_completion_ts(cq);
 }
 
+static inline uint64_t ibv_wc_read_completion_wallclock_ns(struct ibv_cq_ex *cq)
+{
+	return cq->read_completion_wallclock_ns(cq);
+}
+
 static inline uint16_t ibv_wc_read_cvlan(struct ibv_cq_ex *cq)
 {
 	return cq->read_cvlan(cq);
-- 
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 rdma-core 5/5] mlx5: Implement read_completion_wallclock_ns

[Yishai Hadas]

From: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Implement the read_completion_wallclock_ns CQ reader.
This internally fetches the clock page and converts the
device timestamp using the existing DV api.

The commit also revises mlx5_cq_fill_pfns return value.

Signed-off-by: Feras Daoud <ferasda-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Eitan Rabin <rabin-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 providers/mlx5/cq.c    | 181 +++++++++++++++++++++++++++++++++++++++----------
 providers/mlx5/mlx5.h  |   5 +-
 providers/mlx5/verbs.c |  15 ++--
 3 files changed, 162 insertions(+), 39 deletions(-)

diff --git a/providers/mlx5/cq.c b/providers/mlx5/cq.c
index e96418f..25446d0 100644
--- a/providers/mlx5/cq.c
+++ b/providers/mlx5/cq.c
@@ -1004,10 +1004,12 @@ static inline void _mlx5_end_poll(struct ibv_cq_ex *ibcq,
 }
 
 static inline int mlx5_start_poll(struct ibv_cq_ex *ibcq, struct ibv_poll_cq_attr *attr,
-				  int lock, enum polling_mode stall, int cqe_version)
+				  int lock, enum polling_mode stall,
+				  int cqe_version, int clock_update)
 				  ALWAYS_INLINE;
 static inline int mlx5_start_poll(struct ibv_cq_ex *ibcq, struct ibv_poll_cq_attr *attr,
-				  int lock, enum polling_mode stall, int cqe_version)
+				  int lock, enum polling_mode stall,
+				  int cqe_version, int clock_update)
 {
 	struct mlx5_cq *cq = to_mcq(ibv_cq_ex_to_cq(ibcq));
 	struct mlx5_cqe64 *cqe64;
@@ -1066,8 +1068,14 @@ static inline int mlx5_start_poll(struct ibv_cq_ex *ibcq, struct ibv_poll_cq_att
 		}
 
 		cq->flags &= ~(MLX5_CQ_FLAGS_FOUND_CQES);
+
+		goto out;
 	}
 
+	if (clock_update && !err)
+		err = mlx5dv_get_clock_info(ibcq->context, &cq->last_clock_info);
+
+out:
 	return err;
 }
 
@@ -1117,73 +1125,145 @@ static inline int mlx5_next_poll_v1(struct ibv_cq_ex *ibcq)
 static inline int mlx5_start_poll_v0(struct ibv_cq_ex *ibcq,
 				     struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, 0, 0);
+	return mlx5_start_poll(ibcq, attr, 0, 0, 0, 0);
 }
 
 static inline int mlx5_start_poll_v1(struct ibv_cq_ex *ibcq,
 				     struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, 0, 1);
+	return mlx5_start_poll(ibcq, attr, 0, 0, 1, 0);
 }
 
 static inline int mlx5_start_poll_v0_lock(struct ibv_cq_ex *ibcq,
 					  struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, 0, 0);
+	return mlx5_start_poll(ibcq, attr, 1, 0, 0, 0);
 }
 
 static inline int mlx5_start_poll_v1_lock(struct ibv_cq_ex *ibcq,
 					  struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, 0, 1);
+	return mlx5_start_poll(ibcq, attr, 1, 0, 1, 0);
 }
 
 static inline int mlx5_start_poll_adaptive_stall_v0_lock(struct ibv_cq_ex *ibcq,
 							 struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 0);
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 0, 0);
 }
 
 static inline int mlx5_start_poll_stall_v0_lock(struct ibv_cq_ex *ibcq,
 						struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 0);
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 0, 0);
 }
 
 static inline int mlx5_start_poll_adaptive_stall_v1_lock(struct ibv_cq_ex *ibcq,
 							 struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 1);
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 1, 0);
 }
 
 static inline int mlx5_start_poll_stall_v1_lock(struct ibv_cq_ex *ibcq,
 						struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 1);
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 1, 0);
 }
 
 static inline int mlx5_start_poll_stall_v0(struct ibv_cq_ex *ibcq,
 					   struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 0);
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 0, 0);
 }
 
 static inline int mlx5_start_poll_adaptive_stall_v0(struct ibv_cq_ex *ibcq,
 						    struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 0);
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 0, 0);
 }
 
 static inline int mlx5_start_poll_adaptive_stall_v1(struct ibv_cq_ex *ibcq,
 						    struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 1);
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 1, 0);
 }
 
 static inline int mlx5_start_poll_stall_v1(struct ibv_cq_ex *ibcq,
 					   struct ibv_poll_cq_attr *attr)
 {
-	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 1);
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 1, 0);
+}
+
+static inline int mlx5_start_poll_v0_lock_clock_update(struct ibv_cq_ex *ibcq,
+						       struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, 0, 0, 1);
+}
+
+static inline int mlx5_start_poll_v1_lock_clock_update(struct ibv_cq_ex *ibcq,
+						       struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, 0, 1, 1);
+}
+
+static inline int mlx5_start_poll_v1_clock_update(struct ibv_cq_ex *ibcq,
+						  struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, 0, 1, 1);
+}
+
+static inline int mlx5_start_poll_v0_clock_update(struct ibv_cq_ex *ibcq,
+						  struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, 0, 0, 1);
+}
+
+static inline int mlx5_start_poll_stall_v1_lock_clock_update(struct ibv_cq_ex *ibcq,
+							     struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 1, 1);
+}
+
+static inline int mlx5_start_poll_stall_v0_lock_clock_update(struct ibv_cq_ex *ibcq,
+							     struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL, 0, 1);
+}
+
+static inline int mlx5_start_poll_stall_v1_clock_update(struct ibv_cq_ex *ibcq,
+							struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 1, 1);
+}
+
+static inline int mlx5_start_poll_stall_v0_clock_update(struct ibv_cq_ex *ibcq,
+							struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL, 0, 1);
+}
+
+static inline int mlx5_start_poll_adaptive_stall_v0_lock_clock_update(struct ibv_cq_ex *ibcq,
+								      struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 0, 1);
+}
+
+static inline int mlx5_start_poll_adaptive_stall_v1_lock_clock_update(struct ibv_cq_ex *ibcq,
+								      struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 1, POLLING_MODE_STALL_ADAPTIVE, 1, 1);
+}
+
+static inline int mlx5_start_poll_adaptive_stall_v0_clock_update(struct ibv_cq_ex *ibcq,
+								 struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 0, 1);
+}
+
+static inline int mlx5_start_poll_adaptive_stall_v1_clock_update(struct ibv_cq_ex *ibcq,
+								 struct ibv_poll_cq_attr *attr)
+{
+	return mlx5_start_poll(ibcq, attr, 0, POLLING_MODE_STALL_ADAPTIVE, 1, 1);
 }
 
 static inline void mlx5_end_poll_adaptive_stall_lock(struct ibv_cq_ex *ibcq)
@@ -1408,6 +1488,15 @@ static inline uint64_t mlx5_cq_read_wc_completion_ts(struct ibv_cq_ex *ibcq)
 	return be64toh(cq->cqe64->timestamp);
 }
 
+static inline uint64_t
+mlx5_cq_read_wc_completion_wallclock_ns(struct ibv_cq_ex *ibcq)
+{
+	struct mlx5_cq *cq = to_mcq(ibv_cq_ex_to_cq(ibcq));
+
+	return mlx5dv_ts_to_ns(&cq->last_clock_info,
+			       mlx5_cq_read_wc_completion_ts(ibcq));
+}
+
 static inline uint16_t mlx5_cq_read_wc_cvlan(struct ibv_cq_ex *ibcq)
 {
 	struct mlx5_cq *cq = to_mcq(ibv_cq_ex_to_cq(ibcq));
@@ -1437,16 +1526,17 @@ static inline void mlx5_cq_read_wc_tm_info(struct ibv_cq_ex *ibcq,
 #define STALL BIT(1)
 #define V1 BIT(2)
 #define ADAPTIVE BIT(3)
+#define CLOCK_UPDATE BIT(4)
 
-#define mlx5_start_poll_name(cqe_ver, lock, stall, adaptive) \
-	mlx5_start_poll##adaptive##stall##cqe_ver##lock
+#define mlx5_start_poll_name(cqe_ver, lock, stall, adaptive, clock_update) \
+	mlx5_start_poll##adaptive##stall##cqe_ver##lock##clock_update
 #define mlx5_next_poll_name(cqe_ver, adaptive) \
 	mlx5_next_poll##adaptive##cqe_ver
 #define mlx5_end_poll_name(lock, stall, adaptive) \
 	mlx5_end_poll##adaptive##stall##lock
 
-#define POLL_FN_ENTRY(cqe_ver, lock, stall, adaptive) { \
-		.start_poll = &mlx5_start_poll_name(cqe_ver, lock, stall, adaptive), \
+#define POLL_FN_ENTRY(cqe_ver, lock, stall, adaptive, clock_update) { \
+		.start_poll = &mlx5_start_poll_name(cqe_ver, lock, stall, adaptive, clock_update), \
 		.next_poll = &mlx5_next_poll_name(cqe_ver, adaptive), \
 		.end_poll = &mlx5_end_poll_name(lock, stall, adaptive), \
 	}
@@ -1456,29 +1546,44 @@ static const struct op
 	int (*start_poll)(struct ibv_cq_ex *ibcq, struct ibv_poll_cq_attr *attr);
 	int (*next_poll)(struct ibv_cq_ex *ibcq);
 	void (*end_poll)(struct ibv_cq_ex *ibcq);
-} ops[ADAPTIVE + V1 + STALL + SINGLE_THREADED + 1] = {
-	[V1] =  POLL_FN_ENTRY(_v1, _lock, , ),
-	[0] =  POLL_FN_ENTRY(_v0, _lock, , ),
-	[V1 | SINGLE_THREADED] =  POLL_FN_ENTRY(_v1, , , ),
-	[SINGLE_THREADED] =  POLL_FN_ENTRY(_v0, , , ),
-	[V1 | STALL] =  POLL_FN_ENTRY(_v1, _lock, _stall, ),
-	[STALL] =  POLL_FN_ENTRY(_v0, _lock, _stall, ),
-	[V1 | SINGLE_THREADED | STALL] =  POLL_FN_ENTRY(_v1, , _stall, ),
-	[SINGLE_THREADED | STALL] =  POLL_FN_ENTRY(_v0, , _stall, ),
-	[V1 | STALL | ADAPTIVE] =  POLL_FN_ENTRY(_v1, _lock, _stall, _adaptive),
-	[STALL | ADAPTIVE] =  POLL_FN_ENTRY(_v0, _lock, _stall, _adaptive),
-	[V1 | SINGLE_THREADED | STALL | ADAPTIVE] =  POLL_FN_ENTRY(_v1, , _stall, _adaptive),
-	[SINGLE_THREADED | STALL | ADAPTIVE] =  POLL_FN_ENTRY(_v0, , _stall, _adaptive),
+} ops[ADAPTIVE + V1 + STALL + SINGLE_THREADED + CLOCK_UPDATE + 1] = {
+	[V1] =  POLL_FN_ENTRY(_v1, _lock, , ,),
+	[0] =  POLL_FN_ENTRY(_v0, _lock, , ,),
+	[V1 | SINGLE_THREADED] =  POLL_FN_ENTRY(_v1, , , , ),
+	[SINGLE_THREADED] =  POLL_FN_ENTRY(_v0, , , , ),
+	[V1 | STALL] =  POLL_FN_ENTRY(_v1, _lock, _stall, , ),
+	[STALL] =  POLL_FN_ENTRY(_v0, _lock, _stall, , ),
+	[V1 | SINGLE_THREADED | STALL] =  POLL_FN_ENTRY(_v1, , _stall, , ),
+	[SINGLE_THREADED | STALL] =  POLL_FN_ENTRY(_v0, , _stall, , ),
+	[V1 | STALL | ADAPTIVE] = POLL_FN_ENTRY(_v1, _lock, _stall, _adaptive, ),
+	[STALL | ADAPTIVE] =  POLL_FN_ENTRY(_v0, _lock, _stall, _adaptive, ),
+	[V1 | SINGLE_THREADED | STALL | ADAPTIVE] = POLL_FN_ENTRY(_v1, , _stall, _adaptive, ),
+	[SINGLE_THREADED | STALL | ADAPTIVE] = POLL_FN_ENTRY(_v0, , _stall, _adaptive, ),
+	[V1 | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, _lock, , , _clock_update),
+	[0 | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, _lock, , , _clock_update),
+	[V1 | SINGLE_THREADED | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, , , , _clock_update),
+	[SINGLE_THREADED | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, , , , _clock_update),
+	[V1 | STALL | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, _lock, _stall, , _clock_update),
+	[STALL | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, _lock, _stall, , _clock_update),
+	[V1 | SINGLE_THREADED | STALL | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, , _stall, , _clock_update),
+	[SINGLE_THREADED | STALL | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, , _stall, , _clock_update),
+	[V1 | STALL | ADAPTIVE | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, _lock, _stall, _adaptive, _clock_update),
+	[STALL | ADAPTIVE | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, _lock, _stall, _adaptive, _clock_update),
+	[V1 | SINGLE_THREADED | STALL | ADAPTIVE | CLOCK_UPDATE] = POLL_FN_ENTRY(_v1, , _stall, _adaptive, _clock_update),
+	[SINGLE_THREADED | STALL | ADAPTIVE | CLOCK_UPDATE] = POLL_FN_ENTRY(_v0, , _stall, _adaptive, _clock_update),
 };
 
-void mlx5_cq_fill_pfns(struct mlx5_cq *cq, const struct ibv_cq_init_attr_ex *cq_attr)
+int mlx5_cq_fill_pfns(struct mlx5_cq *cq,
+		      const struct ibv_cq_init_attr_ex *cq_attr,
+		      struct mlx5_context *mctx)
 {
-	struct mlx5_context *mctx = to_mctx(ibv_cq_ex_to_cq(&cq->ibv_cq)->context);
 	const struct op *poll_ops = &ops[((cq->stall_enable && cq->stall_adaptive_enable) ? ADAPTIVE : 0) |
 					 (mctx->cqe_version ? V1 : 0) |
 					 (cq->flags & MLX5_CQ_FLAGS_SINGLE_THREADED ?
 						      SINGLE_THREADED : 0) |
-					 (cq->stall_enable ? STALL : 0)];
+					 (cq->stall_enable ? STALL : 0) |
+					 ((cq_attr->wc_flags & IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK) ?
+							CLOCK_UPDATE : 0)];
 
 	cq->ibv_cq.start_poll = poll_ops->start_poll;
 	cq->ibv_cq.next_poll = poll_ops->next_poll;
@@ -1509,6 +1614,14 @@ void mlx5_cq_fill_pfns(struct mlx5_cq *cq, const struct ibv_cq_init_attr_ex *cq_
 		cq->ibv_cq.read_flow_tag = mlx5_cq_read_flow_tag;
 	if (cq_attr->wc_flags & IBV_WC_EX_WITH_TM_INFO)
 		cq->ibv_cq.read_tm_info = mlx5_cq_read_wc_tm_info;
+	if (cq_attr->wc_flags & IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK) {
+		if (!mctx->clock_info_page)
+			return EOPNOTSUPP;
+		cq->ibv_cq.read_completion_wallclock_ns =
+		    mlx5_cq_read_wc_completion_wallclock_ns;
+	}
+
+	return 0;
 }
 
 int mlx5_arm_cq(struct ibv_cq *ibvcq, int solicited)
diff --git a/providers/mlx5/mlx5.h b/providers/mlx5/mlx5.h
index c0f342d..db80846 100644
--- a/providers/mlx5/mlx5.h
+++ b/providers/mlx5/mlx5.h
@@ -387,6 +387,7 @@ struct mlx5_cq {
 	struct mlx5_cqe64		*cqe64;
 	uint32_t			flags;
 	int			umr_opcode;
+	struct mlx5dv_clock_info	last_clock_info;
 };
 
 struct mlx5_tag_entry {
@@ -713,7 +714,9 @@ struct ibv_cq *mlx5_create_cq(struct ibv_context *context, int cqe,
 			       int comp_vector);
 struct ibv_cq_ex *mlx5_create_cq_ex(struct ibv_context *context,
 				    struct ibv_cq_init_attr_ex *cq_attr);
-void mlx5_cq_fill_pfns(struct mlx5_cq *cq, const struct ibv_cq_init_attr_ex *cq_attr);
+int mlx5_cq_fill_pfns(struct mlx5_cq *cq,
+		      const struct ibv_cq_init_attr_ex *cq_attr,
+		      struct mlx5_context *mctx);
 int mlx5_alloc_cq_buf(struct mlx5_context *mctx, struct mlx5_cq *cq,
 		      struct mlx5_buf *buf, int nent, int cqe_sz);
 int mlx5_free_cq_buf(struct mlx5_context *ctx, struct mlx5_buf *buf);
diff --git a/providers/mlx5/verbs.c b/providers/mlx5/verbs.c
index dafed18..1557e4f 100644
--- a/providers/mlx5/verbs.c
+++ b/providers/mlx5/verbs.c
@@ -532,7 +532,8 @@ enum {
 				       IBV_WC_EX_WITH_COMPLETION_TIMESTAMP |
 				       IBV_WC_EX_WITH_CVLAN |
 				       IBV_WC_EX_WITH_FLOW_TAG |
-				       IBV_WC_EX_WITH_TM_INFO
+				       IBV_WC_EX_WITH_TM_INFO |
+				       IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK
 };
 
 enum {
@@ -554,6 +555,7 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 	int				cqe_sz;
 	int				ret;
 	int				ncqe;
+	int				rc;
 	struct mlx5_context *mctx = to_mctx(context);
 	FILE *fp = to_mctx(context)->dbg_fp;
 
@@ -590,6 +592,14 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 		return NULL;
 	}
 
+	if (cq_alloc_flags & MLX5_CQ_FLAGS_EXTENDED) {
+		rc = mlx5_cq_fill_pfns(cq, cq_attr, mctx);
+		if (rc) {
+			errno = rc;
+			goto err;
+		}
+	}
+
 	memset(&cmd, 0, sizeof cmd);
 	cq->cons_index = 0;
 
@@ -696,9 +706,6 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 	cq->stall_adaptive_enable = to_mctx(context)->stall_adaptive_enable;
 	cq->stall_cycles = to_mctx(context)->stall_cycles;
 
-	if (cq_alloc_flags & MLX5_CQ_FLAGS_EXTENDED)
-		mlx5_cq_fill_pfns(cq, cq_attr);
-
 	return &cq->ibv_cq;
 
 err_db:
-- 
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 rdma-core 0/5] Clock information

[Yishai Hadas]

On 1/17/2018 3:51 PM, Yishai Hadas wrote:
> This series from Feras is the supplementary part of the kernel series that was
> sent upstream for 4.16.
> 
> The series enables user space applications to query the mlx5 core clock and use
> it to translate timestamps from HCA core clock units to nanoseconds.
> 
> PR was sent:
> https://github.com/linux-rdma/rdma-core/pull/284
> 
> Note:
> The Travis error in GITHUB is some checkpatch false alarm that relates
> to some macro usage.
> 
> Yishai
> 
> Feras Daoud (5):
>    mlx5: Support for user space clock info
>    mlx5: Add direct verbs support for query clock info
>    mlx5: Add direct verb to translate device timestamp to nanoseconds
>    verbs: Add support to read completion's timestamp in wall clock
>      nanoseconds
>    mlx5: Implement read_completion_wallclock_ns
> 
>   debian/ibverbs-providers.symbols           |   2 +
>   libibverbs/man/ibv_create_cq_ex.3          |   9 +-
>   libibverbs/verbs.h                         |  10 +-
>   providers/mlx5/CMakeLists.txt              |   2 +-
>   providers/mlx5/cq.c                        | 181 +++++++++++++++++++++++------
>   providers/mlx5/libmlx5.map                 |   5 +
>   providers/mlx5/man/CMakeLists.txt          |   2 +
>   providers/mlx5/man/mlx5dv_get_clock_info.3 |  37 ++++++
>   providers/mlx5/man/mlx5dv_query_device.3   |   2 +-
>   providers/mlx5/man/mlx5dv_ts_to_ns.3       |  35 ++++++
>   providers/mlx5/mlx5-abi.h                  |  24 +++-
>   providers/mlx5/mlx5.c                      |  71 +++++++++++
>   providers/mlx5/mlx5.h                      |   7 +-
>   providers/mlx5/mlx5dv.h                    |  65 ++++++++++-
>   providers/mlx5/verbs.c                     |  15 ++-
>   15 files changed, 421 insertions(+), 46 deletions(-)
>   create mode 100644 providers/mlx5/man/mlx5dv_get_clock_info.3
>   create mode 100644 providers/mlx5/man/mlx5dv_ts_to_ns.3
> 

The PR was re-based over the latest code, series was merged.
--
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