From: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
To: <stable@dpdk.org>
Cc: <bluca@debian.org>, Tim Martin <timothym@nvidia.com>
Subject: [PATCH 22.11] net/mlx5: fix real time counter reading from PCI BAR
Date: Wed, 20 Nov 2024 20:38:55 +0200 [thread overview]
Message-ID: <20241120183855.2218388-1-viacheslavo@nvidia.com> (raw)
From: Tim Martin <timothym@nvidia.com>
[ upstream commit 7271cf3a27fad171880a6085af5bd1cdb5a4a624 ]
There is the mlx5_txpp_read_clock() routine reading
the 64-bit real time counter from the device PCI BAR.
It introduced two issues:
- it checks the PCI BAR mapping into process address
space and tries to map this on demand. This might be
problematic if something goes wrong and mapping fails.
It happens on every read_clock API call, invokes kernel
taking a long time and causing application malfunction.
- the 64-bit counter should be read in single atomic
transaction
Fixes: b94d93ca7380 ("net/mlx5: support reading device clock")
Cc: stable@dpdk.org
Signed-off-by: Tim Martin <timothym@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
.mailmap | 1 +
drivers/net/mlx5/mlx5.c | 4 ++++
drivers/net/mlx5/mlx5_tx.h | 30 ++++++++++++++++++++++++++++++
drivers/net/mlx5/mlx5_txpp.c | 11 ++---------
4 files changed, 37 insertions(+), 9 deletions(-)
diff --git a/.mailmap b/.mailmap
index 025bd2daba..457af9abc2 100644
--- a/.mailmap
+++ b/.mailmap
@@ -1392,6 +1392,7 @@ Timmons C. Player <timmons.player@spirent.com>
Timothy McDaniel <timothy.mcdaniel@intel.com>
Timothy Miskell <timothy.miskell@intel.com>
Timothy Redaelli <tredaelli@redhat.com>
+Tim Martin <timothym@nvidia.com>
Tim Shearer <tim.shearer@overturenetworks.com>
Ting-Kai Ku <ting-kai.ku@intel.com>
Ting Xu <ting.xu@intel.com>
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index d66254740b..1076ad1721 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -1961,6 +1961,7 @@ int
mlx5_proc_priv_init(struct rte_eth_dev *dev)
{
struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_dev_ctx_shared *sh = priv->sh;
struct mlx5_proc_priv *ppriv;
size_t ppriv_size;
@@ -1981,6 +1982,9 @@ mlx5_proc_priv_init(struct rte_eth_dev *dev)
dev->process_private = ppriv;
if (rte_eal_process_type() == RTE_PROC_PRIMARY)
priv->sh->pppriv = ppriv;
+ /* Check and try to map HCA PCI BAR to allow reading real time. */
+ if (sh->dev_cap.rt_timestamp && mlx5_dev_is_pci(dev->device))
+ mlx5_txpp_map_hca_bar(dev);
return 0;
}
diff --git a/drivers/net/mlx5/mlx5_tx.h b/drivers/net/mlx5/mlx5_tx.h
index ff23d87b8a..6b796b97e1 100644
--- a/drivers/net/mlx5/mlx5_tx.h
+++ b/drivers/net/mlx5/mlx5_tx.h
@@ -18,6 +18,7 @@
#include <mlx5_common_mr.h>
#include "mlx5.h"
+#include "mlx5_rx.h"
#include "mlx5_autoconf.h"
/* TX burst subroutines return codes. */
@@ -361,6 +362,35 @@ mlx5_txpp_convert_tx_ts(struct mlx5_dev_ctx_shared *sh, uint64_t mts)
return ci;
}
+/**
+ * Read real time clock counter directly from the device PCI BAR area.
+ * The PCI BAR must be mapped to the process memory space at initialization.
+ *
+ * @param dev
+ * Device to read clock counter from
+ *
+ * @return
+ * 0 - if HCA BAR is not supported or not mapped.
+ * !=0 - read 64-bit value of real-time in UTC formatv (nanoseconds)
+ */
+static __rte_always_inline uint64_t mlx5_read_pcibar_clock(struct rte_eth_dev *dev)
+{
+ struct mlx5_proc_priv *ppriv = dev->process_private;
+
+ if (ppriv && ppriv->hca_bar) {
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_dev_ctx_shared *sh = priv->sh;
+ uint64_t *hca_ptr = (uint64_t *)(ppriv->hca_bar) +
+ __mlx5_64_off(initial_seg, real_time);
+ uint64_t ts = __atomic_load_n(hca_ptr, __ATOMIC_SEQ_CST);
+
+ ts = rte_be_to_cpu_64(ts);
+ ts = mlx5_txpp_convert_rx_ts(sh, ts);
+ return ts;
+ }
+ return 0;
+}
+
/**
* Set Software Parser flags and offsets in Ethernet Segment of WQE.
* Flags must be preliminary initialized to zero.
diff --git a/drivers/net/mlx5/mlx5_txpp.c b/drivers/net/mlx5/mlx5_txpp.c
index 0e1da1d5f5..c43a5a7927 100644
--- a/drivers/net/mlx5/mlx5_txpp.c
+++ b/drivers/net/mlx5/mlx5_txpp.c
@@ -969,7 +969,6 @@ mlx5_txpp_read_clock(struct rte_eth_dev *dev, uint64_t *timestamp)
{
struct mlx5_priv *priv = dev->data->dev_private;
struct mlx5_dev_ctx_shared *sh = priv->sh;
- struct mlx5_proc_priv *ppriv;
uint64_t ts;
int ret;
@@ -995,15 +994,9 @@ mlx5_txpp_read_clock(struct rte_eth_dev *dev, uint64_t *timestamp)
*timestamp = ts;
return 0;
}
- /* Check and try to map HCA PIC BAR to allow reading real time. */
- ppriv = dev->process_private;
- if (ppriv && !ppriv->hca_bar &&
- sh->dev_cap.rt_timestamp && mlx5_dev_is_pci(dev->device))
- mlx5_txpp_map_hca_bar(dev);
/* Check if we can read timestamp directly from hardware. */
- if (ppriv && ppriv->hca_bar) {
- ts = MLX5_GET64(initial_seg, ppriv->hca_bar, real_time);
- ts = mlx5_txpp_convert_rx_ts(sh, ts);
+ ts = mlx5_read_pcibar_clock(dev);
+ if (ts != 0) {
*timestamp = ts;
return 0;
}
--
2.34.1
reply other threads:[~2024-11-20 18:39 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241120183855.2218388-1-viacheslavo@nvidia.com \
--to=viacheslavo@nvidia.com \
--cc=bluca@debian.org \
--cc=stable@dpdk.org \
--cc=timothym@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).