Re: [PATCH iproute2-next] tc/htb: Hierarchical QoS hardware offload

On 2021-01-29 18:05, David Ahern wrote:

On 12/15/20 12:42 AM, Maxim Mikityanskiy wrote:

quoted

This commit adds support for configuring HTB in offload mode. HTB
offload eliminates the single qdisc lock in the datapath and offloads
the algorithm to the NIC. The new 'offload' parameter is added to
enable this mode:

     # tc qdisc replace dev eth0 root handle 1: htb offload

Classes are created as usual, but filters should be moved to clsact for
lock-free classification (filters attached to HTB itself are not
supported in the offload mode):

     # tc filter add dev eth0 egress protocol ip flower dst_port 80
     action skbedit priority 1:10

please add the dump in both stdout and json here.

Do you mean to add example output to the commit message?

quoted

Signed-off-by: Maxim Mikityanskiy <redacted>
---
  include/uapi/linux/pkt_sched.h | 1 +
  tc/q_htb.c                     | 9 ++++++++-

missing an update to man/man8/tc-htb.8

OK, I'll add.

quoted

  2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/pkt_sched.h b/include/uapi/linux/pkt_sched.h
index 9e7c2c60..79a699f1 100644
--- a/include/uapi/linux/pkt_sched.h
+++ b/include/uapi/linux/pkt_sched.h

@@ -434,6 +434,7 @@ enum {
  	TCA_HTB_RATE64,
  	TCA_HTB_CEIL64,
  	TCA_HTB_PAD,
+	TCA_HTB_OFFLOAD,
  	__TCA_HTB_MAX,
  };
  

diff --git a/tc/q_htb.c b/tc/q_htb.c
index c609e974..fd11dad6 100644
--- a/tc/q_htb.c
+++ b/tc/q_htb.c

@@ -30,11 +30,12 @@
  static void explain(void)
  {
  	fprintf(stderr, "Usage: ... qdisc add ... htb [default N] [r2q N]\n"
-		"                      [direct_qlen P]\n"
+		"                      [direct_qlen P] [offload]\n"
  		" default  minor id of class to which unclassified packets are sent {0}\n"
  		" r2q      DRR quantums are computed as rate in Bps/r2q {10}\n"
  		" debug    string of 16 numbers each 0-3 {0}\n\n"
  		" direct_qlen  Limit of the direct queue {in packets}\n"
+		" offload  hardware offload\n"

why 'offload hardware offload'? does not make sense to me and

"offload" is a new parameter, and "hardware offload" is the description, 
just like the other parameters above.

you don't
reference hardware below.

Where should I reference it?

quoted

  		"... class add ... htb rate R1 [burst B1] [mpu B] [overhead O]\n"
  		"                      [prio P] [slot S] [pslot PS]\n"
  		"                      [ceil R2] [cburst B2] [mtu MTU] [quantum Q]\n"

@@ -68,6 +69,7 @@ static int htb_parse_opt(struct qdisc_util *qu, int argc,
  	};
  	struct rtattr *tail;
  	unsigned int i; char *p;
+	bool offload = false;
  
  	while (argc > 0) {
  		if (matches(*argv, "r2q") == 0) {

@@ -91,6 +93,8 @@ static int htb_parse_opt(struct qdisc_util *qu, int argc,
  			if (get_u32(&direct_qlen, *argv, 10)) {
  				explain1("direct_qlen"); return -1;
  			}
+		} else if (matches(*argv, "offload") == 0) {
+			offload = true;
  		} else {
  			fprintf(stderr, "What is \"%s\"?\n", *argv);
  			explain();

@@ -103,6 +107,8 @@ static int htb_parse_opt(struct qdisc_util *qu, int argc,
  	if (direct_qlen != ~0U)
  		addattr_l(n, 2024, TCA_HTB_DIRECT_QLEN,
  			  &direct_qlen, sizeof(direct_qlen));
+	if (offload)
+		addattr(n, 2024, TCA_HTB_OFFLOAD);
  	addattr_nest_end(n, tail);
  	return 0;
  }

@@ -344,6 +350,7 @@ static int htb_print_opt(struct qdisc_util *qu, FILE *f, struct rtattr *opt)
  		print_uint(PRINT_ANY, "direct_qlen", " direct_qlen %u",
  			   direct_qlen);
  	}
+	print_uint(PRINT_ANY, "offload", " offload %d", !!tb[TCA_HTB_OFFLOAD]);
  	return 0;
  }
  

On 2/2/21 4:46 AM, Maxim Mikityanskiy wrote:

On 2021-01-29 18:05, David Ahern wrote:

quoted

On 12/15/20 12:42 AM, Maxim Mikityanskiy wrote:

quoted

This commit adds support for configuring HTB in offload mode. HTB
offload eliminates the single qdisc lock in the datapath and offloads
the algorithm to the NIC. The new 'offload' parameter is added to
enable this mode:

     # tc qdisc replace dev eth0 root handle 1: htb offload

Classes are created as usual, but filters should be moved to clsact for
lock-free classification (filters attached to HTB itself are not
supported in the offload mode):

     # tc filter add dev eth0 egress protocol ip flower dst_port 80
     action skbedit priority 1:10

please add the dump in both stdout and json here.

Do you mean to add example output to the commit message?

yes.

quoted

quoted

  diff --git a/tc/q_htb.c b/tc/q_htb.c
index c609e974..fd11dad6 100644

--- a/tc/q_htb.c
+++ b/tc/q_htb.c

@@ -30,11 +30,12 @@

  static void explain(void)
  {
      fprintf(stderr, "Usage: ... qdisc add ... htb [default N] [r2q
N]\n"
-        "                      [direct_qlen P]\n"
+        "                      [direct_qlen P] [offload]\n"
          " default  minor id of class to which unclassified packets
are sent {0}\n"
          " r2q      DRR quantums are computed as rate in Bps/r2q
{10}\n"
          " debug    string of 16 numbers each 0-3 {0}\n\n"
          " direct_qlen  Limit of the direct queue {in packets}\n"
+        " offload  hardware offload\n"

why 'offload hardware offload'? does not make sense to me and

"offload" is a new parameter, and "hardware offload" is the description,
just like the other parameters above.

quoted

you don't
reference hardware below.

Where should I reference it?

I see now. More words are needed to make it clear 'hardware offload' is
the explanation of offload. As it is, the words run together as 'offload
 hardware offload' which is the source of my confusion.