| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <3091c796-acad-4c87-9782-3b67210147c2@linux.dev>
Date: Thu, 18 Sep 2025 11:53:53 +0100
From: Vadim Fedorenko <vadim.fedorenko@...ux.dev>
To: Jakub Kicinski <kuba@...nel.org>
Cc: Andrew Lunn <andrew@...n.ch>, Michael Chan <michael.chan@...adcom.com>,
Pavan Chebbi <pavan.chebbi@...adcom.com>, Tariq Toukan <tariqt@...dia.com>,
Gal Pressman <gal@...dia.com>, intel-wired-lan@...ts.osuosl.org,
Donald Hunter <donald.hunter@...il.com>, Carolina Jubran
<cjubran@...dia.com>, Paolo Abeni <pabeni@...hat.com>,
Simon Horman <horms@...nel.org>, netdev@...r.kernel.org
Subject: Re: [PATCH net-next v3 1/4] ethtool: add FEC bins histogram report
On 18/09/2025 01:41, Jakub Kicinski wrote:
> On Tue, 16 Sep 2025 19:12:54 +0000 Vadim Fedorenko wrote:
>> IEEE 802.3ck-2022 defines counters for FEC bins and 802.3df-2024
>> clarifies it a bit further. Implement reporting interface through as
>> addition to FEC stats available in ethtool.
>> diff --git a/Documentation/netlink/specs/ethtool.yaml b/Documentation/netlink/specs/ethtool.yaml
>> index 7a7594713f1f..de5008266884 100644
>> --- a/Documentation/netlink/specs/ethtool.yaml
>> +++ b/Documentation/netlink/specs/ethtool.yaml
>> @@ -1219,6 +1219,23 @@ attribute-sets:
>> name: udp-ports
>> type: nest
>> nested-attributes: tunnel-udp
>> + -
>> + name: fec-hist
>> + attr-cnt-name: __ethtool-a-fec-hist-cnt
>
> s/__/--/
That will bring strong inconsistency in schema. All other attributes
have counter attribute with __ in the beginning:
name: fec-stat
attr-cnt-name: __ethtool-a-fec-stat-cnt
name: stats-grp
attr-cnt-name: __ethtool-a-stats-grp-cnt
name: stats
attr-cnt-name: __ethtool-a-stats-cnt
>
>> + attributes:
>> + -
>> + name: bin-low
>> + type: u32
>> + -
>> + name: bin-high
>> + type: u32
>
> We should add some doc: strings here so that the important info like
> which one is inclusive is rendered right in the API reference
Yep, will add some doc
>
>> + -
>> + name: bin-val
>> + type: uint
>> + -
>> + name: bin-val-per-lane
>> + type: binary
>
> probably good to doc this too
ack
>
>> + sub-type: u64
>> -
>> name: fec-stat
>> attr-cnt-name: __ethtool-a-fec-stat-cnt
>> @@ -1242,6 +1259,11 @@ attribute-sets:
>> name: corr-bits
>> type: binary
>> sub-type: u64
>> + -
>> + name: hist
>> + type: nest
>> + multi-attr: True
>> + nested-attributes: fec-hist
>> -
>> name: fec
>> attr-cnt-name: __ethtool-a-fec-cnt
>> diff --git a/Documentation/networking/ethtool-netlink.rst b/Documentation/networking/ethtool-netlink.rst
>> index ab20c644af24..b270886c5f5d 100644
>> --- a/Documentation/networking/ethtool-netlink.rst
>> +++ b/Documentation/networking/ethtool-netlink.rst
>> @@ -1541,6 +1541,11 @@ Drivers fill in the statistics in the following structure:
>> .. kernel-doc:: include/linux/ethtool.h
>> :identifiers: ethtool_fec_stats
>>
>> +Statistics may have FEC bins histogram attribute ``ETHTOOL_A_FEC_STAT_HIST``
>> +as defined in IEEE 802.3ck-2022 and 802.3df-2024. Nested attributes will have
>> +the range of FEC errors in the bin (inclusive) and the amount of error events
>> +in the bin.
>
> Does this sound better?
>
> Optional ``ETHTOOL_A_FEC_STAT_HIST`` attributes form a FEC error
> histogram, as defined in IEEE 802.3ck-2022 and 802.3df-2024
> (histogram of number of errors within a single FEC block).
> Each ``ETHTOOL_A_FEC_STAT_HIST`` entry contains error count
> (optionally also broken down by SerDes lane) as well as metadata
> about the bin. Bin range (low, high) is inclusive.
Ack
>
>> static void
>> -nsim_get_fec_stats(struct net_device *dev, struct ethtool_fec_stats *fec_stats)
>> +nsim_get_fec_stats(struct net_device *dev, struct ethtool_fec_stats *fec_stats,
>> + struct ethtool_fec_hist *hist)
>> {
>> + struct ethtool_fec_hist_value *values = hist->values;
>> +
>> + hist->ranges = netdevsim_fec_ranges;
>> +
>> fec_stats->corrected_blocks.total = 123;
>> fec_stats->uncorrectable_blocks.total = 4;
>> +
>> + values[0].bin_value = 445;
>
> Bin 0 had per lane breakdown, can't core add up the lanes for the
> driver?
Like it's done for blocks counter? Should we force drivers to keep 'sum'
value equal to ETHTOOL_STAT_NOT_SET when they provide per-lane values?
>
>> + values[1].bin_value = 12;
>> + values[2].bin_value = 2;
>> + values[0].bin_value_per_lane[0] = 125;
>> + values[0].bin_value_per_lane[1] = 120;
>> + values[0].bin_value_per_lane[2] = 100;
>> + values[0].bin_value_per_lane[3] = 100;
>> }
>
>> +/**
>> + * struct ethtool_fec_hist_range - error bits range for FEC bins histogram
>
> Don't say "FEC bin histogram" I think the word histogram implies that
> the data is bin'ed up.
Ack
>
>> + * statistics
>> + * @low: low bound of the bin (inclusive)
>> + * @high: high bound of the bin (inclusive)
>> + */
>
>> @@ -113,7 +114,11 @@ static int fec_prepare_data(const struct ethnl_req_info *req_base,
>> struct ethtool_fec_stats stats;
>>
>> ethtool_stats_init((u64 *)&stats, sizeof(stats) / 8);
>> - dev->ethtool_ops->get_fec_stats(dev, &stats);
>> + ethtool_stats_init((u64 *)data->fec_stat_hist.values,
>> + ETHTOOL_FEC_HIST_MAX *
>> + sizeof(struct ethtool_fec_hist_value) / 8);
>
> sizeof(data->fec_stat_hist.values) / 8
>
> would save you the multiplication?
yeah...
>
>> + dev->ethtool_ops->get_fec_stats(dev, &stats,
>> + &data->fec_stat_hist);
>>
>> fec_stats_recalc(&data->corr, &stats.corrected_blocks);
>> fec_stats_recalc(&data->uncorr, &stats.uncorrectable_blocks);
>
>> +static int fec_put_hist(struct sk_buff *skb, const struct ethtool_fec_hist *hist)
>
> over 80 chars, please wrap (checkpatch --max-line-length=80)
ouch, that's proper strict check! :)
>
>> +{
>> + const struct ethtool_fec_hist_range *ranges = hist->ranges;
>> + const struct ethtool_fec_hist_value *values = hist->values;
>> + struct nlattr *nest;
>> + int i, j;
>> +
>> + if (!ranges)
>> + return 0;
>> +
>> + for (i = 0; i < ETHTOOL_FEC_HIST_MAX; i++) {
>> + if (i && !ranges[i].low && !ranges[i].high)
>
> low and high should probably be unsigned now
ack
>
>> + break;
>> +
>> + if (WARN_ON_ONCE(values[i].bin_value == ETHTOOL_STAT_NOT_SET))
>> + break;
>> +
>> + nest = nla_nest_start(skb, ETHTOOL_A_FEC_STAT_HIST);
>> + if (!nest)
>> + return -EMSGSIZE;
>> +
>> + if (nla_put_u32(skb, ETHTOOL_A_FEC_HIST_BIN_LOW,
>> + ranges[i].low) ||
>> + nla_put_u32(skb, ETHTOOL_A_FEC_HIST_BIN_HIGH,
>> + ranges[i].high) ||
>> + nla_put_uint(skb, ETHTOOL_A_FEC_HIST_BIN_VAL,
>> + values[i].bin_value))
>> + goto err_cancel_hist;
>> + for (j = 0; j < ETHTOOL_MAX_LANES; j++) {
>> + if (values[i].bin_value_per_lane[j] == ETHTOOL_STAT_NOT_SET)
>
> You know, bin_value could be 'sum', and bin_value_per_lane could be
> simply 'per_lane'.
SG, I'll change it
>
>> + break;
>> + }
>
> {} brackets unnecessary
>
>> + if (j && nla_put_64bit(skb, ETHTOOL_A_FEC_HIST_BIN_VAL_PER_LANE,
>> + sizeof(u64) * j,
>> + values[i].bin_value_per_lane,
>> + ETHTOOL_A_FEC_STAT_PAD))
>> + goto err_cancel_hist;
>> +
>> + nla_nest_end(skb, nest);
>> + }
>> +
>> + return 0;
>> +
>> +err_cancel_hist:
>> + nla_nest_cancel(skb, nest);
>> + return -EMSGSIZE;
>> +}
>
> We need a selftest. Add a case to stats.py and do basic sanity checking
> on what the kernel spits out? Maybe 2 test cases - one for overall and
> one for per-lane, cause mlx5 doesn't have per lane and we'd like the
> test to pass there.
Yeah, sure, I'll add them as another patch to this series
Powered by blists - more mailing lists