From: Feras Daoud <ferasda@xxxxxxxxxxxx>
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@xxxxxxxxxxxx>
Signed-off-by: Eitan Rabin <rabin@xxxxxxxxxxxx>
Reviewed-by: Yishai Hadas <yishaih@xxxxxxxxxxxx>
---
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@xxxxxxxxxxxx>
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@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
