[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b2d98851b62e1cf18bad0996736cd0b6de265f06.camel@gmail.com>
Date: Wed, 28 Aug 2024 09:41:19 -0700
From: Alexander H Duyck <alexander.duyck@...il.com>
To: "Nelson, Shannon" <shannon.nelson@....com>, Mohsin Bashir
<mohsin.bashr@...il.com>, netdev@...r.kernel.org
Cc: alexanderduyck@...com, kuba@...nel.org, andrew@...n.ch,
davem@...emloft.net, edumazet@...gle.com, pabeni@...hat.com,
kernel-team@...a.com, sanmanpradhan@...a.com, sdf@...ichev.me,
jdamato@...tly.com
Subject: Re: [PATCH net-next v2 2/2] eth: fbnic: Add support to fetch group
stats
On Tue, 2024-08-27 at 17:30 -0700, Nelson, Shannon wrote:
> On 8/27/2024 1:59 PM, Mohsin Bashir wrote:
> >
> > Add support for group stats for mac. The fbnic_set_counter helps prevent
> > overriding the default values for counters which are not collected by the device.
> >
> > The 'reset' flag in 'get_eth_mac_stats' allows choosing between
> > resetting the counter to recent most value or fecthing the aggregate
> > values of counters. This is important to cater for cases such as
> > device reset.
> >
> > The 'fbnic_stat_rd64' read 64b stats counters in a consistent fashion using
> > high-low-high approach. This allows to isolate cases where counter is
> > wrapped between the reads.
> >
> > Command: ethtool -S eth0 --groups eth-mac
> > Example Output:
> > eth-mac-FramesTransmittedOK: 421644
> > eth-mac-FramesReceivedOK: 3849708
> > eth-mac-FrameCheckSequenceErrors: 0
> > eth-mac-AlignmentErrors: 0
> > eth-mac-OctetsTransmittedOK: 64799060
> > eth-mac-FramesLostDueToIntMACXmitError: 0
> > eth-mac-OctetsReceivedOK: 5134513531
> > eth-mac-FramesLostDueToIntMACRcvError: 0
> > eth-mac-MulticastFramesXmittedOK: 568
> > eth-mac-BroadcastFramesXmittedOK: 454
> > eth-mac-MulticastFramesReceivedOK: 276106
> > eth-mac-BroadcastFramesReceivedOK: 26119
> > eth-mac-FrameTooLongErrors: 0
> >
> > Signed-off-by: Mohsin Bashir <mohsin.bashr@...il.com>
> > ---
> > v2: Rebase to the latest
> >
> > v1: https://lore.kernel.org/netdev/20240807002445.3833895-1-mohsin.bashr@gmail.com
> > ---
> > drivers/net/ethernet/meta/fbnic/Makefile | 1 +
> > drivers/net/ethernet/meta/fbnic/fbnic.h | 4 ++
> > drivers/net/ethernet/meta/fbnic/fbnic_csr.h | 37 ++++++++++++++
> > .../net/ethernet/meta/fbnic/fbnic_ethtool.c | 49 ++++++++++++++++++
> > .../net/ethernet/meta/fbnic/fbnic_hw_stats.c | 27 ++++++++++
> > .../net/ethernet/meta/fbnic/fbnic_hw_stats.h | 40 +++++++++++++++
> > drivers/net/ethernet/meta/fbnic/fbnic_mac.c | 50 +++++++++++++++++++
> > drivers/net/ethernet/meta/fbnic/fbnic_mac.h | 3 ++
> > 8 files changed, 211 insertions(+)
> > create mode 100644 drivers/net/ethernet/meta/fbnic/fbnic_hw_stats.c
> > create mode 100644 drivers/net/ethernet/meta/fbnic/fbnic_hw_stats.h
> >
> > diff --git a/drivers/net/ethernet/meta/fbnic/Makefile b/drivers/net/ethernet/meta/fbnic/Makefile
> > index 37cfc34a5118..ed4533a73c57 100644
> > --- a/drivers/net/ethernet/meta/fbnic/Makefile
> > +++ b/drivers/net/ethernet/meta/fbnic/Makefile
> > @@ -10,6 +10,7 @@ obj-$(CONFIG_FBNIC) += fbnic.o
> > fbnic-y := fbnic_devlink.o \
> > fbnic_ethtool.o \
> > fbnic_fw.o \
> > + fbnic_hw_stats.o \
> > fbnic_irq.o \
> > fbnic_mac.o \
> > fbnic_netdev.o \
> > diff --git a/drivers/net/ethernet/meta/fbnic/fbnic.h b/drivers/net/ethernet/meta/fbnic/fbnic.h
> > index 28d970f81bfc..0f9e8d79461c 100644
> > --- a/drivers/net/ethernet/meta/fbnic/fbnic.h
> > +++ b/drivers/net/ethernet/meta/fbnic/fbnic.h
> > @@ -11,6 +11,7 @@
> >
> > #include "fbnic_csr.h"
> > #include "fbnic_fw.h"
> > +#include "fbnic_hw_stats.h"
> > #include "fbnic_mac.h"
> > #include "fbnic_rpc.h"
> >
> > @@ -47,6 +48,9 @@ struct fbnic_dev {
> >
> > /* Number of TCQs/RCQs available on hardware */
> > u16 max_num_queues;
> > +
> > + /* Local copy of hardware statistics */
> > + struct fbnic_hw_stats hw_stats;
> > };
> >
> > /* Reserve entry 0 in the MSI-X "others" array until we have filled all
> > diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_csr.h b/drivers/net/ethernet/meta/fbnic/fbnic_csr.h
> > index a64360de0552..21db509acbc1 100644
> > --- a/drivers/net/ethernet/meta/fbnic/fbnic_csr.h
> > +++ b/drivers/net/ethernet/meta/fbnic/fbnic_csr.h
> > @@ -660,6 +660,43 @@ enum {
> > #define FBNIC_SIG_PCS_INTR_MASK 0x11816 /* 0x46058 */
> > #define FBNIC_CSR_END_SIG 0x1184e /* CSR section delimiter */
> >
> > +#define FBNIC_CSR_START_MAC_STAT 0x11a00
> > +#define FBNIC_MAC_STAT_RX_BYTE_COUNT_L 0x11a08 /* 0x46820 */
> > +#define FBNIC_MAC_STAT_RX_BYTE_COUNT_H 0x11a09 /* 0x46824 */
> > +#define FBNIC_MAC_STAT_RX_ALIGN_ERROR_L \
> > + 0x11a0a /* 0x46828 */
> > +#define FBNIC_MAC_STAT_RX_ALIGN_ERROR_H \
> > + 0x11a0b /* 0x4682c */
> > +#define FBNIC_MAC_STAT_RX_TOOLONG_L 0x11a0e /* 0x46838 */
> > +#define FBNIC_MAC_STAT_RX_TOOLONG_H 0x11a0f /* 0x4683c */
> > +#define FBNIC_MAC_STAT_RX_RECEIVED_OK_L \
> > + 0x11a12 /* 0x46848 */
> > +#define FBNIC_MAC_STAT_RX_RECEIVED_OK_H \
> > + 0x11a13 /* 0x4684c */
> > +#define FBNIC_MAC_STAT_RX_PACKET_BAD_FCS_L \
> > + 0x11a14 /* 0x46850 */
> > +#define FBNIC_MAC_STAT_RX_PACKET_BAD_FCS_H \
> > + 0x11a15 /* 0x46854 */
> > +#define FBNIC_MAC_STAT_RX_IFINERRORS_L 0x11a18 /* 0x46860 */
> > +#define FBNIC_MAC_STAT_RX_IFINERRORS_H 0x11a19 /* 0x46864 */
> > +#define FBNIC_MAC_STAT_RX_MULTICAST_L 0x11a1c /* 0x46870 */
> > +#define FBNIC_MAC_STAT_RX_MULTICAST_H 0x11a1d /* 0x46874 */
> > +#define FBNIC_MAC_STAT_RX_BROADCAST_L 0x11a1e /* 0x46878 */
> > +#define FBNIC_MAC_STAT_RX_BROADCAST_H 0x11a1f /* 0x4687c */
> > +#define FBNIC_MAC_STAT_TX_BYTE_COUNT_L 0x11a3e /* 0x468f8 */
> > +#define FBNIC_MAC_STAT_TX_BYTE_COUNT_H 0x11a3f /* 0x468fc */
> > +#define FBNIC_MAC_STAT_TX_TRANSMITTED_OK_L \
> > + 0x11a42 /* 0x46908 */
> > +#define FBNIC_MAC_STAT_TX_TRANSMITTED_OK_H \
> > + 0x11a43 /* 0x4690c */
> > +#define FBNIC_MAC_STAT_TX_IFOUTERRORS_L \
> > + 0x11a46 /* 0x46918 */
> > +#define FBNIC_MAC_STAT_TX_IFOUTERRORS_H \
> > + 0x11a47 /* 0x4691c */
> > +#define FBNIC_MAC_STAT_TX_MULTICAST_L 0x11a4a /* 0x46928 */
> > +#define FBNIC_MAC_STAT_TX_MULTICAST_H 0x11a4b /* 0x4692c */
> > +#define FBNIC_MAC_STAT_TX_BROADCAST_L 0x11a4c /* 0x46930 */
> > +#define FBNIC_MAC_STAT_TX_BROADCAST_H 0x11a4d /* 0x46934 */
>
> These might be more readable if you add another tab between the name and
> the value, then you wouldn't need to do line wraps.
We were trying to keep the format consistent from the top of these
defines to the bottom. If I recall the comment on the byte offset would
end up going past 80 characters if we shifted that over.
> > /* PUL User Registers */
> > #define FBNIC_CSR_START_PUL_USER 0x31000 /* CSR section delimiter */
> > #define FBNIC_PUL_OB_TLP_HDR_AW_CFG 0x3103d /* 0xc40f4 */
> > diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c b/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c
> > index 7064dfc9f5b0..5d980e178941 100644
> > --- a/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c
> > +++ b/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c
> > @@ -16,8 +16,57 @@ fbnic_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *drvinfo)
> > sizeof(drvinfo->fw_version));
> > }
> >
> > +static void fbnic_set_counter(u64 *stat, struct fbnic_stat_counter *counter)
> > +{
> > + if (counter->reported)
> > + *stat = counter->value;
> > +}
> > +
> > +static void
> > +fbnic_get_eth_mac_stats(struct net_device *netdev,
> > + struct ethtool_eth_mac_stats *eth_mac_stats)
> > +{
> > + struct fbnic_net *fbn = netdev_priv(netdev);
> > + struct fbnic_mac_stats *mac_stats;
> > + struct fbnic_dev *fbd = fbn->fbd;
> > + const struct fbnic_mac *mac;
> > +
> > + mac_stats = &fbd->hw_stats.mac;
> > + mac = fbd->mac;
> > +
> > + mac->get_eth_mac_stats(fbd, false, &mac_stats->eth_mac);
> > +
> > + fbnic_set_counter(ð_mac_stats->FramesTransmittedOK,
> > + &mac_stats->eth_mac.FramesTransmittedOK);
> > + fbnic_set_counter(ð_mac_stats->FramesReceivedOK,
> > + &mac_stats->eth_mac.FramesReceivedOK);
> > + fbnic_set_counter(ð_mac_stats->FrameCheckSequenceErrors,
> > + &mac_stats->eth_mac.FrameCheckSequenceErrors);
> > + fbnic_set_counter(ð_mac_stats->AlignmentErrors,
> > + &mac_stats->eth_mac.AlignmentErrors);
> > + fbnic_set_counter(ð_mac_stats->OctetsTransmittedOK,
> > + &mac_stats->eth_mac.OctetsTransmittedOK);
> > + fbnic_set_counter(ð_mac_stats->FramesLostDueToIntMACXmitError,
> > + &mac_stats->eth_mac.FramesLostDueToIntMACXmitError);
> > + fbnic_set_counter(ð_mac_stats->OctetsReceivedOK,
> > + &mac_stats->eth_mac.OctetsReceivedOK);
> > + fbnic_set_counter(ð_mac_stats->FramesLostDueToIntMACRcvError,
> > + &mac_stats->eth_mac.FramesLostDueToIntMACRcvError);
> > + fbnic_set_counter(ð_mac_stats->MulticastFramesXmittedOK,
> > + &mac_stats->eth_mac.MulticastFramesXmittedOK);
> > + fbnic_set_counter(ð_mac_stats->BroadcastFramesXmittedOK,
> > + &mac_stats->eth_mac.BroadcastFramesXmittedOK);
> > + fbnic_set_counter(ð_mac_stats->MulticastFramesReceivedOK,
> > + &mac_stats->eth_mac.MulticastFramesReceivedOK);
> > + fbnic_set_counter(ð_mac_stats->BroadcastFramesReceivedOK,
> > + &mac_stats->eth_mac.BroadcastFramesReceivedOK);
> > + fbnic_set_counter(ð_mac_stats->FrameTooLongErrors,
> > + &mac_stats->eth_mac.FrameTooLongErrors);
> > +}
> > +
> > static const struct ethtool_ops fbnic_ethtool_ops = {
> > .get_drvinfo = fbnic_get_drvinfo,
> > + .get_eth_mac_stats = fbnic_get_eth_mac_stats,
> > };
> >
> > void fbnic_set_ethtool_ops(struct net_device *dev)
> > diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_hw_stats.c b/drivers/net/ethernet/meta/fbnic/fbnic_hw_stats.c
> > new file mode 100644
> > index 000000000000..a0acc7606aa1
> > --- /dev/null
> > +++ b/drivers/net/ethernet/meta/fbnic/fbnic_hw_stats.c
> > @@ -0,0 +1,27 @@
> > +#include "fbnic.h"
> > +
> > +u64 fbnic_stat_rd64(struct fbnic_dev *fbd, u32 reg, u32 offset)
> > +{
> > + u32 prev_upper, upper, lower, diff;
> > +
> > + prev_upper = rd32(fbd, reg + offset);
> > + lower = rd32(fbd, reg);
> > + upper = rd32(fbd, reg + offset);
> > +
> > + diff = upper - prev_upper;
> > + if (!diff)
> > + return ((u64)upper << 32) | lower;
>
> Is there any particular reason you didn't use u64_stats_fetch_begin()
> and u64_stats_fetch_retry() around these to protect the reads?
>
> sln
The thing is there isn't another thread to race against. The checking
here is against hardware so it cannot cooperate with the
stats_fetch_begin/retry.
That said we do have a function that is collecting the 32b register
stats that we probably do need to add this wrapper for as it has to run
in the service task to update the stats.
Powered by blists - more mailing lists