Re: [EXT] Re: [v1,ethtool] ethtool: add setting frame preemption of traffic classes

From: Murali Karicheri
Date: Wed Dec 04 2019 - 11:55:59 EST


On 12/04/2019 03:14 AM, Po Liu wrote:
Hi Murali, Ivan,

Thank you for your feedback. Maybe it is better to use "RFC" type for these patches for the discussion.

Yes please.

Murali

Br,
Po Liu

-----Original Message-----
From: Murali Karicheri <m-karicheri2@xxxxxx>
Sent: 2019å12æ4æ 1:42
To: Po Liu <po.liu@xxxxxxx>; rmk+kernel@xxxxxxxxxxxxxxx;
linville@xxxxxxxxxxxxx; netdev-owner@xxxxxxxxxxxxxxx; davem@xxxxxxxxxxxxx;
linux-kernel@xxxxxxxxxxxxxxx; netdev@xxxxxxxxxxxxxxx;
vinicius.gomes@xxxxxxxxx; simon.horman@xxxxxxxxxxxxx; Claudiu Manoil
<claudiu.manoil@xxxxxxx>; Vladimir Oltean <vladimir.oltean@xxxxxxx>;
Xiaoliang Yang <xiaoliang.yang_1@xxxxxxx>; Roy Zang <roy.zang@xxxxxxx>;
Mingkai Hu <mingkai.hu@xxxxxxx>; Jerry Huang <jerry.huang@xxxxxxx>; Leo
Li <leoyang.li@xxxxxxx>
Subject: [EXT] Re: [v1,ethtool] ethtool: add setting frame preemption of traffic
classes

Caution: EXT Email

Hi Po Liu,

Thanks for working on this! Some suggestion below as we are working on adding
this support to Texas Instrument's CPSW driver on AM65x family of SoCs as well.
TRM for that is provided below for your reference.
Relevant section for IET (Frame pre-emption) is

12.2.1.4.6.6.1IET Configuration

https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.ti.c
om%2Flit%2Fug%2Fspruid7d%2Fspruid7d.pdf&amp;data=02%7C01%7Cpo.liu%
40nxp.com%7C15544a9db1ff422e2fdf08d77817647f%7C686ea1d3bc2b4c6fa
92cd99c5c301635%7C0%7C1%7C637109914194485742&amp;sdata=VD5kg%
2FY1SDDWjwMZHjgUNlFrcvXnDvdIPGblWkx4DXs%3D&amp;reserved=0

On 12/03/2019 11:27 AM, Ivan Khoronzhuk wrote:
On Wed, Nov 27, 2019 at 09:58:52AM +0000, Po Liu wrote:

Hi Po Liu,

IEEE Std 802.1Qbu standard defined the frame preemption of port
trffic classes. User can set a value to hardware. The value will be
translated to a binary, each bit represent a traffic class.
Bit "1" means preemptable traffic class. Bit "0" means express
traffic class. MSB represent high number traffic class.

ethtool -k devname

This command would show if the tx-preemption feature is available.
If hareware set preemption feature. The property would be a fixed
value 'on' if hardware support the frame preemption. Feature would
show a fixed value 'off' if hardware don't support the frame preemption.

ethtool devname

This command would show include an item 'preemption'. A following
value '0' means all traffic classes are 'express'. A value none zero
means traffic classes preemption capabilities. The value will be
translated to a binary, each bit represent a traffic class. Bit '1'
means preemptable traffic class. Bit '0' means express traffic class.
MSB represent high number traffic class.

ethtool -s devname preemption N

What about other potential parameters like MAC fragment size, mac hold?
Shouldn't be it considered along with other FP parameters to provide
correct interface later?

Say, preemption, lets name it fp-mask or frame-preemption-mask.
Then other potential setting can be similar and added later:

frame-preemption-mask
frame-preemption-fragsize
frame-preemption-machold
Need additional capabilities as described by Ivan above. Thanks Ivan!

So it would be better to use feature/sub-parameter format so that it can be
extended as needed in the future.

For express/Preemptable mask setting it becomes

ethtool -s devname frame-preemption tc-mask N

For setting min fragment size

ethtool -s devname frame-preemption min-fragsize 64


I thought the fragment size set 64 is enough for Qbv. Anyway, if you prefer more details setting. I think it is better to set a specific serial type. For example, '-p' for set tc-mask N and min-fragsize 64, '-P' for status.

Also the device may be capable of doing Verify process to detect the capability
of neighbor device and show the status. So we should have a way to show this

The verify process maybe disabled as default.

status as well when user type

ethtool devname

We are working currently to add this feature to CPSW driver on AM65x which
will be upstream-ed soon. So want to have this done in an way that we can
extend it later.
Similarly for taprio, there are some parameters that user might want to tune
such as Max SDU size per tc class. I hope we could use ethtool to set the same
on a similar way.

Furthermore, this setting could be extend for a serial setting for mac and traffic class.


Thanks

Murali

....

mac-hold it's rather flag, at least I've used it as priv-flag.
so can or so

frame-preemption-flags


This command would set which traffic classes are frame preemptable.
The value will be translated to a binary, each bit represent a
traffic class. Bit '1' means preemptable traffic class. Bit '0'
means express traffic class. MSB represent high number traffic class.

Signed-off-by: Po Liu <Po.Liu@xxxxxxx>
---
ethtool-copy.h |Â 6 +++++-
ethtool.8.in |Â 8 ++++++++
ethtool.c | 18 ++++++++++++++++++
3 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/ethtool-copy.h b/ethtool-copy.h index 9afd2e6..e04bdf3
100644
--- a/ethtool-copy.h
+++ b/ethtool-copy.h
@@ -1662,6 +1662,9 @@ static __inline__ int
ethtool_validate_duplex(__u8 duplex)
#define AUTONEG_DISABLE 0x00
#define AUTONEG_ENABLE 0x01

+/* Disable preemtion. */
+#define PREEMPTION_DISABLE 0x0
+
/* MDI or MDI-X status/control - if MDI/MDI_X/AUTO is set then
* the driver is required to renegotiate link */ @@ -1878,7 +1881,8
@@ struct ethtool_link_settings {
__s8 link_mode_masks_nwords;
__u8 transceiver;
__u8 reserved1[3];
- __u32 reserved[7];
+ __u32 preemption;
+ __u32 reserved[6];
__u32 link_mode_masks[0];
/* layout of link_mode_masks fields:
* __u32 map_supported[link_mode_masks_nwords];
diff --git a/ethtool.8.in b/ethtool.8.in index 062695a..7d612b2
100644
--- a/ethtool.8.in
+++ b/ethtool.8.in
@@ -236,6 +236,7 @@ ethtool \- query or control network driver and
hardware settings
.B2 autoneg on off
.BN advertise
.BN phyad
+.BN preemption
.B2 xcvr internal external
.RB [ wol \ \*(WO]
.RB [ sopass \ \*(MA]
@@ -703,6 +704,13 @@ lB l lB.
.BI phyad \ N
PHY address.
.TP
+.BI preemption \ N
+Set preemptable traffic classes by bits.
+.B A
+value will be translated to a binary, each bit represent a traffic
class.
+Bit "1" means preemptable traffic class. Bit "0" means express
traffic class.
+MSB represent high number traffic class.
+.TP
.A2 xcvr internal external
Selects transceiver type. Currently only internal and external can be
specified, in the future further types might be added.
diff --git a/ethtool.c b/ethtool.c
index acf183d..d5240f8 100644
--- a/ethtool.c
+++ b/ethtool.c
@@ -928,6 +928,12 @@ dump_link_usettings(const struct
ethtool_link_usettings *link_usettings)
}
}

+ if (link_usettings->base.preemption == PREEMPTION_DISABLE)
+ fprintf(stdout, "ÂÂÂ Preemption: 0x0 (off)\n");
+ else
+ fprintf(stdout, "ÂÂÂ Preemption: 0x%x\n",
+ link_usettings->base.preemption);
+
return 0;
}

@@ -2869,6 +2875,7 @@ static int do_sset(struct cmd_context *ctx)
int port_wanted = -1;
int mdix_wanted = -1;
int autoneg_wanted = -1;
+ int preemption_wanted = -1;
int phyad_wanted = -1;
int xcvr_wanted = -1;
u32 *full_advertising_wanted = NULL; @@ -2957,6 +2964,12 @@
static int do_sset(struct cmd_context *ctx)
} else {
exit_bad_args();
}
+ } else if (!strcmp(argp[i], "preemption")) {
+ gset_changed = 1;
+ i += 1;
+ if (i >= argc)
+ exit_bad_args();
+ preemption_wanted = get_u32(argp[i], 16);
} else if (!strcmp(argp[i], "advertise")) {
gset_changed = 1;
i += 1;
@@ -3094,6 +3107,9 @@ static int do_sset(struct cmd_context *ctx)
}
if (autoneg_wanted != -1)
link_usettings->base.autoneg = autoneg_wanted;
+ if (preemption_wanted != -1)
+ link_usettings->base.preemption
+ = preemption_wanted;
if (phyad_wanted != -1)
link_usettings->base.phy_address = phyad_wanted;
if (xcvr_wanted != -1)
@@ -3186,6 +3202,8 @@ static int do_sset(struct cmd_context *ctx)
fprintf(stderr, "Â not setting transceiver\n");
if (mdix_wanted != -1)
fprintf(stderr, "Â not setting mdix\n");
+ if (preemption_wanted != -1)
+ fprintf(stderr, "Â not setting preemption\n");
}
}

--
2.17.1