* Re: [PATCH net-next 1/2] mqprio: Add a new hardware offload type in mqprio
From: Yunsheng Lin @ 2017-10-13 2:12 UTC (permalink / raw)
To: Yuval Mintz, davem@davemloft.net
Cc: huangdaode@hisilicon.com, xuwei5@hisilicon.com,
liguozhu@hisilicon.com, Yisen.Zhuang@huawei.com,
gabriele.paoloni@huawei.com, john.garry@huawei.com,
linuxarm@huawei.com, salil.mehta@huawei.com, lipeng321@huawei.com,
netdev@vger.kernel.org, linux-kernel@vger.kernel.org
In-Reply-To: <AM0PR0502MB3683208C0F411F67C70BF6B1BF4B0@AM0PR0502MB3683.eurprd05.prod.outlook.com>
Hi, Yuval
On 2017/10/13 4:10, Yuval Mintz wrote:
>> When a driver supports both dcb and hardware offloaded mqprio, and
>> user is running mqprio and dcb tool concurrently, the configuration
>> set by each tool may be conflicted with each other because the dcb
> (for second 'each') s/each/the
>
Will do, Thanks
>> and mqprio may be using the same hardwere offload component and share
> s/hardwere/hardware
Will do, Thanks
>
>> the tc system in the network stack.
>>
>> This patch adds a new offload type to indicate that the underlying
>> driver offload prio mapping as part of DCB. If the driver would be
> 'should' offload
Will do, Thanks
>
>> incapable of that it would refuse the offload. User would then have
>> to explicitly request that qdisc offload.
>
>
>
^ permalink raw reply
* Re: [PATCH net-next 0/2] Add mqprio hardware offload support in hns3 driver
From: Yunsheng Lin @ 2017-10-13 2:07 UTC (permalink / raw)
To: Yuval Mintz, davem@davemloft.net
Cc: huangdaode@hisilicon.com, xuwei5@hisilicon.com,
liguozhu@hisilicon.com, Yisen.Zhuang@huawei.com,
gabriele.paoloni@huawei.com, john.garry@huawei.com,
linuxarm@huawei.com, salil.mehta@huawei.com, lipeng321@huawei.com,
netdev@vger.kernel.org, linux-kernel@vger.kernel.org
In-Reply-To: <AM0PR0502MB36837A27C3EAA2A08590D709BF4B0@AM0PR0502MB3683.eurprd05.prod.outlook.com>
Hi, Yuval
On 2017/10/13 4:21, Yuval Mintz wrote:
>> This patchset adds a new hardware offload type in mqprio before adding
>> mqprio hardware offload support in hns3 driver.
>
> I think one of the biggest issues in tying this to DCB configuration is the
> non-immediate [and possibly non persistent] configuration.
>
> Scenario #1:
> User is configuring mqprio offloaded with 3 TCs while device is in willing mode.
> Would you expect the driver to immediately respond with a success or instead
> delay the return until the DCBx negotiation is complete and the operational
> num of TCs is actually 3?
Well, when user requsts the mqprio offloaded by a hardware shared by DCB, I expect
the user is not using the dcb tool.
If user is still using dcb tool, then result is undefined.
The scenario you mention maybe can be enforced by setting willing to zero when user
is requesting the mqprio offload, and restore the willing bit when unloaded the mqprio
offload.
But I think the real issue is that dcb and mqprio shares the tc system in the stack,
the problem may be better to be fixed in the stack rather than in the driver, as you
suggested in the DCB patchset. What do you think?
>
> Scenario #2:
> Assume user explicitly offloaded mqprio with 3 TCs, but now DCB configuration
> has changed on the peer side and 4 TCs is the new negotiated operational value.
> Your current driver logic would change the number of TCs underneath the user
> configuration [and it would actually probably work due to mqprio being a crappy
> qdisc]. But was that the user actual intention?
> [I think the likely answer in this scenario is 'yes' since the alternative is no better.
> But I still thought it was worth mentioning]
You are right, the problem also have something to do with mqprio and dcb sharing
the tc in the stack.
Druing testing, when user explicitly offloaded mqprio with 3 TCs, all
queue has a default pfifo mqprio attached, after DCB changes the tc num to 4,
using tc qdisc shows some queue does not have a default pfifo mqprio attached.
Maybe we can add a callback to notify mqprio the configuration has changed.
Thanks
Yunsheng Lin
>
> Cheers,
> Yuval
>
>>
>> Yunsheng Lin (2):
>> mqprio: Add a new hardware offload type in mqprio
>> net: hns3: Add mqprio hardware offload support in hns3 driver
>>
>> drivers/net/ethernet/hisilicon/hns3/hnae3.h | 1 +
>> .../net/ethernet/hisilicon/hns3/hns3pf/hclge_dcb.c | 23 +++++++++++
>> .../net/ethernet/hisilicon/hns3/hns3pf/hns3_enet.c | 46 ++++++++++++++-
>> -------
>> include/uapi/linux/pkt_sched.h | 1 +
>> 4 files changed, 55 insertions(+), 16 deletions(-)
>>
>> --
>> 1.9.1
>
>
>
^ permalink raw reply
* Re: [PATCH net-next v2 1/1] bridge: return error code when deleting Vlan
From: Jamal Hadi Salim @ 2017-10-13 2:03 UTC (permalink / raw)
To: Nikolay Aleksandrov, Roman Mashak, David Ahern
Cc: David Miller, Stephen Hemminger, Linux Kernel Network Developers
In-Reply-To: <e4611694-4248-e796-5cca-24f3d07e0e9e@cumulusnetworks.com>
On 17-10-12 02:12 PM, Nikolay Aleksandrov wrote:
> On 12/10/17 21:07, Roman Mashak wrote:
>> For example, if you attempt to delete a non-existing vlan on a port,
>> the current code succeeds and also sends event :
>>
>> rtnetlink_rcv_msg
>> rtnl_bridge_dellink
>> br_dellink
>> br_afspec
>> br_vlan_info
>>
>> int br_dellink(..)
>> {
>> ...
>> err = br_afspec()
>> if (err == 0)
>> br_ifinfo_notify(RTM_NEWLINK, p);
>> }
>>
>> This is misleading, so a proper errcode has to be produced.
>>
>
> True, but you also change the expected behaviour because now a user can
> clear all vlans with one request (1 - 4094), and after the change that
> will fail with a partial delete if some vlan was missing.
>
The issue is more subtle (per Roman above):
Try to delete a vlan (that doesnt exist).
1) It says "success".
2) Worse: Another process listening (bridge monitor?) gets an _event_
that the vlan has been deleted (when it never existed in the first
place).
> This has been the behaviour forever and some script might depend on it.
> Also IMO, and as David also mentioned, doing a partial delete is not good.
>
I think this is a bug (especially the event part).
cheers,
jamal
^ permalink raw reply
* Re: [PATCH] tracing: bpf: Hide bpf trace events when they are not used
From: Steven Rostedt @ 2017-10-13 1:49 UTC (permalink / raw)
To: Alexei Starovoitov
Cc: LKML, Alexei Starovoitov, Daniel Borkmann, David S. Miller,
netdev
In-Reply-To: <20171013013834.pxh3ancgutrm6m7t@ast-mbp>
On Thu, 12 Oct 2017 18:38:36 -0700
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> actually just noticed that xdp tracepoints are not covered by ifdef.
> They depend on bpf_syscall too. So probably makes sense to wrap
> them together.
> bpf tracepoints are not being actively worked on whereas xdp tracepoints
> keep evolving quickly, so the best is probalby to go via net-next
> if you don't mind.
Hmm, they didn't trigger a warning, with the exception of
trace_xdp_redirect_map. I have code to check if tracepoints are used or
not, and it appears that the xdp can be used without BPF_SYSCALL.
I don't think they should be wrapped together until we know why they
are used. I can still take this patch and just not touch the xdp ones.
Note, my kernel was using trace_xdp_redirect_map_err,
trace_xdp_redirect_err, trace_xdp_redirect and trace_xdp_exception.
As they did appear.
-- Steve
^ permalink raw reply
* [PATCH net-next v2 4/4] net: dsa: remove .set_addr
From: Vivien Didelot @ 2017-10-13 1:41 UTC (permalink / raw)
To: netdev
Cc: linux-kernel, kernel, David S. Miller, Florian Fainelli,
Andrew Lunn, Vivien Didelot
In-Reply-To: <20171013014113.29561-1-vivien.didelot@savoirfairelinux.com>
Now that there is no user for the .set_addr function, remove it from
DSA. If a switch supports this feature (like mv88e6xxx), the
implementation can be done in the driver setup.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
---
include/net/dsa.h | 1 -
net/dsa/dsa2.c | 6 ------
net/dsa/legacy.c | 6 ------
3 files changed, 13 deletions(-)
diff --git a/include/net/dsa.h b/include/net/dsa.h
index ce1d622734d7..2746741f74cf 100644
--- a/include/net/dsa.h
+++ b/include/net/dsa.h
@@ -291,7 +291,6 @@ struct dsa_switch_ops {
enum dsa_tag_protocol (*get_tag_protocol)(struct dsa_switch *ds);
int (*setup)(struct dsa_switch *ds);
- int (*set_addr)(struct dsa_switch *ds, u8 *addr);
u32 (*get_phy_flags)(struct dsa_switch *ds, int port);
/*
diff --git a/net/dsa/dsa2.c b/net/dsa/dsa2.c
index 54ed054777bd..6ac9e11d385c 100644
--- a/net/dsa/dsa2.c
+++ b/net/dsa/dsa2.c
@@ -336,12 +336,6 @@ static int dsa_ds_apply(struct dsa_switch_tree *dst, struct dsa_switch *ds)
if (err)
return err;
- if (ds->ops->set_addr) {
- err = ds->ops->set_addr(ds, dst->cpu_dp->netdev->dev_addr);
- if (err < 0)
- return err;
- }
-
if (!ds->slave_mii_bus && ds->ops->phy_read) {
ds->slave_mii_bus = devm_mdiobus_alloc(ds->dev);
if (!ds->slave_mii_bus)
diff --git a/net/dsa/legacy.c b/net/dsa/legacy.c
index 19ff6e0a21dc..b0fefbffe082 100644
--- a/net/dsa/legacy.c
+++ b/net/dsa/legacy.c
@@ -172,12 +172,6 @@ static int dsa_switch_setup_one(struct dsa_switch *ds,
if (ret)
return ret;
- if (ops->set_addr) {
- ret = ops->set_addr(ds, master->dev_addr);
- if (ret < 0)
- return ret;
- }
-
if (!ds->slave_mii_bus && ops->phy_read) {
ds->slave_mii_bus = devm_mdiobus_alloc(ds->dev);
if (!ds->slave_mii_bus)
--
2.14.2
^ permalink raw reply related
* [PATCH net-next v2 3/4] net: dsa: dsa_loop: remove .set_addr
From: Vivien Didelot @ 2017-10-13 1:41 UTC (permalink / raw)
To: netdev
Cc: linux-kernel, kernel, David S. Miller, Florian Fainelli,
Andrew Lunn, Vivien Didelot
In-Reply-To: <20171013014113.29561-1-vivien.didelot@savoirfairelinux.com>
The .set_addr function does nothing, remove the dsa_loop implementation
before getting rid of it completely in DSA.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
---
drivers/net/dsa/dsa_loop.c | 8 --------
1 file changed, 8 deletions(-)
diff --git a/drivers/net/dsa/dsa_loop.c b/drivers/net/dsa/dsa_loop.c
index d55051abf4ed..3a3f4f7ba364 100644
--- a/drivers/net/dsa/dsa_loop.c
+++ b/drivers/net/dsa/dsa_loop.c
@@ -110,13 +110,6 @@ static void dsa_loop_get_ethtool_stats(struct dsa_switch *ds, int port,
data[i] = ps->ports[port].mib[i].val;
}
-static int dsa_loop_set_addr(struct dsa_switch *ds, u8 *addr)
-{
- dev_dbg(ds->dev, "%s\n", __func__);
-
- return 0;
-}
-
static int dsa_loop_phy_read(struct dsa_switch *ds, int port, int regnum)
{
struct dsa_loop_priv *ps = ds->priv;
@@ -263,7 +256,6 @@ static const struct dsa_switch_ops dsa_loop_driver = {
.get_strings = dsa_loop_get_strings,
.get_ethtool_stats = dsa_loop_get_ethtool_stats,
.get_sset_count = dsa_loop_get_sset_count,
- .set_addr = dsa_loop_set_addr,
.phy_read = dsa_loop_phy_read,
.phy_write = dsa_loop_phy_write,
.port_bridge_join = dsa_loop_port_bridge_join,
--
2.14.2
^ permalink raw reply related
* [PATCH net-next v2 2/4] net: dsa: mv88e6060: setup random mac address
From: Vivien Didelot @ 2017-10-13 1:41 UTC (permalink / raw)
To: netdev
Cc: linux-kernel, kernel, David S. Miller, Florian Fainelli,
Andrew Lunn, Vivien Didelot
In-Reply-To: <20171013014113.29561-1-vivien.didelot@savoirfairelinux.com>
As for mv88e6xxx, setup the switch from within the mv88e6060 driver with
a random MAC address, and remove the .set_addr implementation.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
---
drivers/net/dsa/mv88e6060.c | 30 +++++++++++++++++++-----------
1 file changed, 19 insertions(+), 11 deletions(-)
diff --git a/drivers/net/dsa/mv88e6060.c b/drivers/net/dsa/mv88e6060.c
index 621cdc46ad81..2f9d5e6a0f97 100644
--- a/drivers/net/dsa/mv88e6060.c
+++ b/drivers/net/dsa/mv88e6060.c
@@ -9,6 +9,7 @@
*/
#include <linux/delay.h>
+#include <linux/etherdevice.h>
#include <linux/jiffies.h>
#include <linux/list.h>
#include <linux/module.h>
@@ -188,6 +189,20 @@ static int mv88e6060_setup_port(struct dsa_switch *ds, int p)
return 0;
}
+static int mv88e6060_setup_addr(struct dsa_switch *ds)
+{
+ u8 addr[ETH_ALEN];
+
+ eth_random_addr(addr);
+
+ /* Use the same MAC Address as FD Pause frames for all ports */
+ REG_WRITE(REG_GLOBAL, GLOBAL_MAC_01, (addr[0] << 9) | addr[1]);
+ REG_WRITE(REG_GLOBAL, GLOBAL_MAC_23, (addr[2] << 8) | addr[3]);
+ REG_WRITE(REG_GLOBAL, GLOBAL_MAC_45, (addr[4] << 8) | addr[5]);
+
+ return 0;
+}
+
static int mv88e6060_setup(struct dsa_switch *ds)
{
int ret;
@@ -203,6 +218,10 @@ static int mv88e6060_setup(struct dsa_switch *ds)
if (ret < 0)
return ret;
+ ret = mv88e6060_setup_addr(ds);
+ if (ret < 0)
+ return ret;
+
for (i = 0; i < MV88E6060_PORTS; i++) {
ret = mv88e6060_setup_port(ds, i);
if (ret < 0)
@@ -212,16 +231,6 @@ static int mv88e6060_setup(struct dsa_switch *ds)
return 0;
}
-static int mv88e6060_set_addr(struct dsa_switch *ds, u8 *addr)
-{
- /* Use the same MAC Address as FD Pause frames for all ports */
- REG_WRITE(REG_GLOBAL, GLOBAL_MAC_01, (addr[0] << 9) | addr[1]);
- REG_WRITE(REG_GLOBAL, GLOBAL_MAC_23, (addr[2] << 8) | addr[3]);
- REG_WRITE(REG_GLOBAL, GLOBAL_MAC_45, (addr[4] << 8) | addr[5]);
-
- return 0;
-}
-
static int mv88e6060_port_to_phy_addr(int port)
{
if (port >= 0 && port < MV88E6060_PORTS)
@@ -256,7 +265,6 @@ static const struct dsa_switch_ops mv88e6060_switch_ops = {
.get_tag_protocol = mv88e6060_get_tag_protocol,
.probe = mv88e6060_drv_probe,
.setup = mv88e6060_setup,
- .set_addr = mv88e6060_set_addr,
.phy_read = mv88e6060_phy_read,
.phy_write = mv88e6060_phy_write,
};
--
2.14.2
^ permalink raw reply related
* [PATCH net-next v2 1/4] net: dsa: mv88e6xxx: setup random mac address
From: Vivien Didelot @ 2017-10-13 1:41 UTC (permalink / raw)
To: netdev
Cc: linux-kernel, kernel, David S. Miller, Florian Fainelli,
Andrew Lunn, Vivien Didelot
In-Reply-To: <20171013014113.29561-1-vivien.didelot@savoirfairelinux.com>
An Ethernet switch may support having a MAC address, which can be used
as the switch's source address in transmitted full-duplex Pause frames.
If a DSA switch supports the related .set_addr operation, the DSA core
sets the master's MAC address on the switch. This won't make sense
anymore in a multi-CPU ports system, because there won't be a unique
master device assigned to a switch tree.
Instead, setup the switch from within the Marvell driver with a random
MAC address, and remove the .set_addr implementation.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
---
drivers/net/dsa/mv88e6xxx/chip.c | 33 +++++++++++++++++----------------
1 file changed, 17 insertions(+), 16 deletions(-)
diff --git a/drivers/net/dsa/mv88e6xxx/chip.c b/drivers/net/dsa/mv88e6xxx/chip.c
index d74c7335c512..76cf383dcf90 100644
--- a/drivers/net/dsa/mv88e6xxx/chip.c
+++ b/drivers/net/dsa/mv88e6xxx/chip.c
@@ -932,6 +932,19 @@ static int mv88e6xxx_irl_setup(struct mv88e6xxx_chip *chip)
return 0;
}
+static int mv88e6xxx_mac_setup(struct mv88e6xxx_chip *chip)
+{
+ if (chip->info->ops->set_switch_mac) {
+ u8 addr[ETH_ALEN];
+
+ eth_random_addr(addr);
+
+ return chip->info->ops->set_switch_mac(chip, addr);
+ }
+
+ return 0;
+}
+
static int mv88e6xxx_pvt_map(struct mv88e6xxx_chip *chip, int dev, int port)
{
u16 pvlan = 0;
@@ -2013,6 +2026,10 @@ static int mv88e6xxx_setup(struct dsa_switch *ds)
if (err)
goto unlock;
+ err = mv88e6xxx_mac_setup(chip);
+ if (err)
+ goto unlock;
+
err = mv88e6xxx_phy_setup(chip);
if (err)
goto unlock;
@@ -2043,21 +2060,6 @@ static int mv88e6xxx_setup(struct dsa_switch *ds)
return err;
}
-static int mv88e6xxx_set_addr(struct dsa_switch *ds, u8 *addr)
-{
- struct mv88e6xxx_chip *chip = ds->priv;
- int err;
-
- if (!chip->info->ops->set_switch_mac)
- return -EOPNOTSUPP;
-
- mutex_lock(&chip->reg_lock);
- err = chip->info->ops->set_switch_mac(chip, addr);
- mutex_unlock(&chip->reg_lock);
-
- return err;
-}
-
static int mv88e6xxx_mdio_read(struct mii_bus *bus, int phy, int reg)
{
struct mv88e6xxx_mdio_bus *mdio_bus = bus->priv;
@@ -3785,7 +3787,6 @@ static const struct dsa_switch_ops mv88e6xxx_switch_ops = {
.probe = mv88e6xxx_drv_probe,
.get_tag_protocol = mv88e6xxx_get_tag_protocol,
.setup = mv88e6xxx_setup,
- .set_addr = mv88e6xxx_set_addr,
.adjust_link = mv88e6xxx_adjust_link,
.get_strings = mv88e6xxx_get_strings,
.get_ethtool_stats = mv88e6xxx_get_ethtool_stats,
--
2.14.2
^ permalink raw reply related
* [PATCH net-next v2 0/4] net: dsa: remove .set_addr
From: Vivien Didelot @ 2017-10-13 1:41 UTC (permalink / raw)
To: netdev
Cc: linux-kernel, kernel, David S. Miller, Florian Fainelli,
Andrew Lunn, Vivien Didelot
An Ethernet switch may support having a MAC address, which can be used
as the switch's source address in transmitted full-duplex Pause frames.
If a DSA switch supports the related .set_addr operation, the DSA core
sets the master's MAC address on the switch.
This won't make sense anymore in a multi-CPU ports system, because there
won't be a unique master device assigned to a switch tree.
Moreover this operation is confusing because it makes the user think
that it could be used to program the switch with the MAC address of the
CPU/management port such that MAC address learning can be disabled on
said port, but in fact, that's not how it is currently used.
To fix this, assign a random MAC address at setup time in the mv88e6060
and mv88e6xxx drivers before removing .set_addr completely from DSA.
Changes in v2:
- remove .set_addr implementation from drivers and use a random MAC.
Vivien Didelot (4):
net: dsa: mv88e6xxx: setup random mac address
net: dsa: mv88e6060: setup random mac address
net: dsa: dsa_loop: remove .set_addr
net: dsa: remove .set_addr
drivers/net/dsa/dsa_loop.c | 8 --------
drivers/net/dsa/mv88e6060.c | 30 +++++++++++++++++++-----------
drivers/net/dsa/mv88e6xxx/chip.c | 33 +++++++++++++++++----------------
include/net/dsa.h | 1 -
net/dsa/dsa2.c | 6 ------
net/dsa/legacy.c | 6 ------
6 files changed, 36 insertions(+), 48 deletions(-)
--
2.14.2
^ permalink raw reply
* Re: [PATCH net-next 3/3] sched: act: ife: update parameters via rcu handling
From: Jamal Hadi Salim @ 2017-10-13 1:39 UTC (permalink / raw)
To: Alexander Aring; +Cc: xiyou.wangcong, jiri, netdev, eric.dumazet, bjb
In-Reply-To: <20171011211608.22692-4-aring@mojatatu.com>
On 17-10-11 05:16 PM, Alexander Aring wrote:
> This patch changes the parameter updating via RCU and not protected by a
> spinlock anymore. This reduce the time that the spinlock is being held.
>
> Signed-off-by: Alexander Aring <aring@mojatatu.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
cheers,
jamal
^ permalink raw reply
* Re: [PATCH net-next 2/3] sched: act: ife: migrate to use per-cpu counters
From: Jamal Hadi Salim @ 2017-10-13 1:39 UTC (permalink / raw)
To: Alexander Aring
Cc: xiyou.wangcong, jiri, netdev, eric.dumazet, bjb, Yotam Gigi
In-Reply-To: <20171011211608.22692-3-aring@mojatatu.com>
On 17-10-11 05:16 PM, Alexander Aring wrote:
> This patch migrates the current counter handling which is protected by a
> spinlock to a per-cpu counter handling. This reduce the time where the
> spinlock is being held.
>
> Signed-off-by: Alexander Aring <aring@mojatatu.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
cheers,
jamal
^ permalink raw reply
* Re: [PATCH] tracing: bpf: Hide bpf trace events when they are not used
From: Alexei Starovoitov @ 2017-10-13 1:38 UTC (permalink / raw)
To: Steven Rostedt
Cc: LKML, Alexei Starovoitov, Daniel Borkmann, David S. Miller,
netdev
In-Reply-To: <20171012213501.17fe6f8c@vmware.local.home>
On Thu, Oct 12, 2017 at 09:35:01PM -0400, Steven Rostedt wrote:
> On Thu, 12 Oct 2017 18:14:52 -0700
> Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
>
> > On Thu, Oct 12, 2017 at 06:40:02PM -0400, Steven Rostedt wrote:
> > > From: Steven Rostedt (VMware) <rostedt@goodmis.org>
> > >
> > > All the trace events defined in include/trace/events/bpf.h are only
> > > used when CONFIG_BPF_SYSCALL is defined. But this file gets included by
> > > include/linux/bpf_trace.h which is included by the networking code with
> > > CREATE_TRACE_POINTS defined.
> > >
> > > If a trace event is created but not used it still has data structures
> > > and functions created for its use, even though nothing is using them.
> > > To not waste space, do not define the BPF trace events in bpf.h unless
> > > CONFIG_BPF_SYSCALL is defined.
> > >
> > > Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
> >
> > Looks fine.
> > Acked-by: Alexei Starovoitov <ast@kernel.org>
> >
> > I'm assuming you want to take it through tracing tree along
> > with all other cleanups?
>
> Either way is fine. I have a few other ones. I believe Paul is taking
> the RCU patch. There's no dependency.
>
> I'll take it if it is easier for you. I just need the ack.
actually just noticed that xdp tracepoints are not covered by ifdef.
They depend on bpf_syscall too. So probably makes sense to wrap
them together.
bpf tracepoints are not being actively worked on whereas xdp tracepoints
keep evolving quickly, so the best is probalby to go via net-next
if you don't mind.
^ permalink raw reply
* Re: [PATCH net-next 1/3] sched: act: ife: move encode/decode check to init
From: Jamal Hadi Salim @ 2017-10-13 1:38 UTC (permalink / raw)
To: Alexander Aring
Cc: xiyou.wangcong, jiri, netdev, eric.dumazet, bjb, Yotam Gigi
In-Reply-To: <20171011211608.22692-2-aring@mojatatu.com>
On 17-10-11 05:16 PM, Alexander Aring wrote:
> This patch adds the check of the two possible ife handlings encode
> and decode to the init callback. The decode value is for usability
> aspect and used in userspace code only. The current code offers encode
> else decode only. This patch avoids any other option than this.
>
> Signed-off-by: Alexander Aring <aring@mojatatu.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
cheers,
jamal
^ permalink raw reply
* Re: [PATCH] tracing: bpf: Hide bpf trace events when they are not used
From: Steven Rostedt @ 2017-10-13 1:35 UTC (permalink / raw)
To: Alexei Starovoitov
Cc: LKML, Alexei Starovoitov, Daniel Borkmann, David S. Miller,
netdev
In-Reply-To: <20171013011450.w2uqygdfzwf7ud64@ast-mbp>
On Thu, 12 Oct 2017 18:14:52 -0700
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> On Thu, Oct 12, 2017 at 06:40:02PM -0400, Steven Rostedt wrote:
> > From: Steven Rostedt (VMware) <rostedt@goodmis.org>
> >
> > All the trace events defined in include/trace/events/bpf.h are only
> > used when CONFIG_BPF_SYSCALL is defined. But this file gets included by
> > include/linux/bpf_trace.h which is included by the networking code with
> > CREATE_TRACE_POINTS defined.
> >
> > If a trace event is created but not used it still has data structures
> > and functions created for its use, even though nothing is using them.
> > To not waste space, do not define the BPF trace events in bpf.h unless
> > CONFIG_BPF_SYSCALL is defined.
> >
> > Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
>
> Looks fine.
> Acked-by: Alexei Starovoitov <ast@kernel.org>
>
> I'm assuming you want to take it through tracing tree along
> with all other cleanups?
Either way is fine. I have a few other ones. I believe Paul is taking
the RCU patch. There's no dependency.
I'll take it if it is easier for you. I just need the ack.
-- Steve
^ permalink raw reply
* Re: [PATCH 2/6] ath9k: add a quirk to set use_msi automatically
From: Daniel Drake @ 2017-10-13 1:17 UTC (permalink / raw)
To: AceLan Kao
Cc: Kalle Valo, Christoph Hellwig, QCA ath9k Development,
linux-wireless, netdev, Linux-Kernel@Vger. Kernel. Org
In-Reply-To: <CAFv23Q=dFRfxXQnY3pu_zV-yK9yaw-t3-Kn2q9YFFuyB9jJMog@mail.gmail.com>
On Fri, Oct 13, 2017 at 9:12 AM, AceLan Kao <acelan.kao@canonical.com> wrote:
> Hi Daniel,
>
> After applied the 2 commits you mentioned in the email, ath9k works.
>
> https://marc.info/?l=linux-wireless&m=150631274108016&w=2
> https://github.com/endlessm/linux/commit/739c7a924db8f4434a9617657
Thanks for testing. However the approach was basically rejected in this thread:
[PATCH] PCI MSI: allow alignment restrictions on vector allocation
https://marc.info/?t=150631283200001&r=1&w=2
So we still need an upstream solution.
I am curious what Qualcomm have to say about their hardware corrupting
the MSI Message Data value. Is there any news on them submitting the
MSI support patch?
Separately we have the option of seeing if Intel can help us unblock
the legacy interrupt (assuming it was simply blocked by the BIOS), or
adding an interrupt-polling fallback path to ath9k.
Daniel
^ permalink raw reply
* Re: [PATCH] tracing: bpf: Hide bpf trace events when they are not used
From: Alexei Starovoitov @ 2017-10-13 1:14 UTC (permalink / raw)
To: Steven Rostedt
Cc: LKML, Alexei Starovoitov, Daniel Borkmann, David S. Miller,
netdev
In-Reply-To: <20171012184002.0661a867@gandalf.local.home>
On Thu, Oct 12, 2017 at 06:40:02PM -0400, Steven Rostedt wrote:
> From: Steven Rostedt (VMware) <rostedt@goodmis.org>
>
> All the trace events defined in include/trace/events/bpf.h are only
> used when CONFIG_BPF_SYSCALL is defined. But this file gets included by
> include/linux/bpf_trace.h which is included by the networking code with
> CREATE_TRACE_POINTS defined.
>
> If a trace event is created but not used it still has data structures
> and functions created for its use, even though nothing is using them.
> To not waste space, do not define the BPF trace events in bpf.h unless
> CONFIG_BPF_SYSCALL is defined.
>
> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Looks fine.
Acked-by: Alexei Starovoitov <ast@kernel.org>
I'm assuming you want to take it through tracing tree along
with all other cleanups?
^ permalink raw reply
* Re: [PATCH] Add -target to clang switch while cross compiling.
From: Alexei Starovoitov @ 2017-10-13 1:14 UTC (permalink / raw)
To: Abhijit Ayarekar; +Cc: ast, daniel, netdev, linux-kernel
In-Reply-To: <1507852723-24677-1-git-send-email-abhijit.ayarekar@caviumnetworks.com>
On Thu, Oct 12, 2017 at 04:58:43PM -0700, Abhijit Ayarekar wrote:
> Update to llvm excludes assembly instructions.
> llvm git revision is below
>
> commit 2865ab6996164e7854d55c9e21c065fad7c26569
> Author: Yonghong Song <yhs@fb.com>
> Date: Mon Sep 18 23:29:36 2017 +0000
>
> bpf: add inline-asm support
>
> Signed-off-by: Yonghong Song <yhs@fb.com>
> Acked-by: Alexei Starovoitov <ast@kernel.org>
>
> git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@313593 91177308-0d34-0410-b5e6-96231b3b80d8
correct way to reference a commit is
commit 81eb8447daae ("ipv6: take care of rt6_stats")
Also I think git-svn-id link is broken. I'm not sure why llvm keeps
adding it to commits. Kernel's git history doesn't need them.
So just mention llvm commit in kernel's cannonical way and
mention that it will be part of llvm release 6.0
^ permalink raw reply
* Re: [PATCH 2/6] ath9k: add a quirk to set use_msi automatically
From: AceLan Kao @ 2017-10-13 1:12 UTC (permalink / raw)
To: Daniel Drake
Cc: Kalle Valo, Christoph Hellwig, QCA ath9k Development,
linux-wireless, netdev, Linux-Kernel@Vger. Kernel. Org
In-Reply-To: <CAFv23QkLJ0m112Y8iAGv+6DyFT52Hqnxq5Fy=ryWTWxu8oTXdw@mail.gmail.com>
Hi Daniel,
After applied the 2 commits you mentioned in the email, ath9k works.
https://marc.info/?l=linux-wireless&m=150631274108016&w=2
https://github.com/endlessm/linux/commit/739c7a924db8f4434a9617657
Best regards,
AceLan Kao.
2017-10-05 14:39 GMT+08:00 AceLan Kao <acelan.kao@canonical.com>:
> Hi all,
>
> Please drop my patches, Qualcomm is working internally and will submit
> the MSI patch by themselves.
> Thanks.
>
> Hi Daniel,
>
> I'll try your patches tomorrow.
>
> Best regards,
> AceLan Kao.
>
> 2017-10-02 12:21 GMT+08:00 Daniel Drake <drake@endlessm.com>:
>> Hi AceLan,
>>
>> On Thu, Sep 28, 2017 at 4:28 PM, AceLan Kao <acelan.kao@canonical.com> wrote:
>>> Hi Daniel,
>>>
>>> I've tried your patch, but it doesn't work for me.
>>> Wifi can scan AP, but can't get connected.
>>
>> Can you please clarify which patch(es) you have tried?
>>
>> This is the base patch which adds the infrastructure to request
>> specific MSI IRQ vectors:
>> https://marc.info/?l=linux-wireless&m=150631274108016&w=2
>>
>> This is the ath9k MSI patch which makes use of that:
>> https://github.com/endlessm/linux/commit/739c7a924db8f4434a9617657
>>
>> If you were already able to use ath9k MSI interrupts without specific
>> consideration for which MSI vector numbers were used, these are the
>> possible explanations that spring to mind:
>>
>> 1. You got lucky and it picked a vector number that is 4-aligned. You
>> can check this in the "lspci -vvv" output. You'll see something like:
>> Capabilities: [50] MSI: Enable+ Count=1/4 Maskable+ 64bit+
>> Address: 00000000fee0300c Data: 4142
>> The lower number is the vector number. In my example here 0x42 (66) is
>> not 4-aligned so the failure condition will be hit.
>>
>> 2. You are using interrupt remapping, which I suspect may provide a
>> high likelihood of MSI interrupt vectors being 4-aligned. See if
>> /proc/interrupts shows the IRQ type as IR-PCI-MSI
>> Unfortunately interrupt remapping is not available here,
>> https://lists.linuxfoundation.org/pipermail/iommu/2017-August/023717.html
>>
>> 3. My assumption that all ath9k hardware corrupts the MSI vector
>> number could wrong. However we've seen this on different wifi modules
>> in laptops produced by different OEMs and ODMs, so it seems to be a
>> somewhat widespread problem at least.
>>
>> 4. My assumption that ath9k hardware is corrupting the MSI vector
>> number could be wrong; maybe another component is to blame, could it
>> be a BIOS issue? Admittedly I don't really know how I can debug the
>> layers inbetween seeing the MSI Message Data value disagree with the
>> vector number being handled inside do_IRQ().
>>
>> Daniel
^ permalink raw reply
* Re: [patch net-next 27/34] nfp: bpf: Convert ndo_setup_tc offloads to block callbacks
From: Jakub Kicinski @ 2017-10-13 1:08 UTC (permalink / raw)
To: Jiri Pirko; +Cc: netdev, Daniel Borkmann
In-Reply-To: <20171012171823.1431-28-jiri@resnulli.us>
On Thu, 12 Oct 2017 19:18:16 +0200, Jiri Pirko wrote:
> diff --git a/drivers/net/ethernet/netronome/nfp/bpf/offload.c b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
> index a88bb5b..9e9af88 100644
> --- a/drivers/net/ethernet/netronome/nfp/bpf/offload.c
> +++ b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
> @@ -246,6 +246,10 @@ int nfp_net_bpf_offload(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
> void *code;
> int err;
>
> + if (cls_bpf->common.protocol != htons(ETH_P_ALL) ||
> + cls_bpf->common.chain_index)
> + return -EOPNOTSUPP;
> +
> max_instr = nn_readw(nn, NFP_NET_CFG_BPF_MAX_LEN);
>
> switch (cls_bpf->command) {
It is certainly very ugly but I send a fake struct tc_cls_bpf_offload
here for XDP. Refactoring this mess is pretty high on my priority list
but one way or the other this function will be called from XDP so TC
checks must stay in the TC handler... :(
^ permalink raw reply
* Re: [RFC] Support for UNARP (RFC 1868)
From: Mahesh Bandewar (महेश बंडेवार) @ 2017-10-13 0:53 UTC (permalink / raw)
To: Girish Moodalbail; +Cc: Eric Dumazet, linux-netdev, David Miller, kuznet
In-Reply-To: <c75c04fe-46d6-8ec3-844d-fe379710496b@oracle.com>
On Thu, Oct 12, 2017 at 4:06 PM, Girish Moodalbail
<girish.moodalbail@oracle.com> wrote:
> Hello Eric,
>
> The basic idea is to mark the ARP entry either FAILED or STALE as soon as we
> can so that the subsequent packets that depend on that ARP entry will take
> the slow path (neigh_resolve_output()).
>
> Say, if base_reachable_time is 30 seconds, then an ARP entry will be in
> reachable state somewhere between 15 to 45 seconds. Assuming the worst case,
> the ARP entry will be in REACHABLE state for 45 seconds and the packets
> continue to traverse the network towards the source machine and gets dropped
> their since the VM has moved to destination machine.
>
> Instead, based on the received UNARP packet if we mark the ARP entry
>
> (a) FAILED
> - we move to INCOMPLETE state and start the address resolution by sending
> out ARP packets (up to allowed maximum number) until we get ARP
> response
> back at which point we move the ARP entry state to reachable.
>
> (b) STALE
> - we move to DELAY state and set the next timer to DELAY_PROBE_TIME
> (1 second) and continue to send queued packets in arp_queue.
> - After 1 sec we move to PROBE state and start the address resolution
> like
> in the case(a) above.
>
> I was leaning towards (a).
One could arbitrarily increase the stale timeout (by changing no of
probes). So sender
will continue sending traffic to something that has already gone away.
STALE doesn't
mean bad but here the sender is clearly indicating it's going away so
FAILED seems to
be the only logical option.
> Please see in-line..
>
> <snip>
>
>>
>> Hi Girish
>>
>> Your description (or patch title) is misleading. You apparently
>> implement the receive side of the RFC.
>
>
> You are right, it implements only the receive side of the RFC. If this RFC
> is accepted, then we can change arping(8) to generate UNARP requests. We
> could also add an option to ip-address(8) delete subcommand to generate
> UNARP whenever an address is deleted from the interface.
>
>> And the RFC had Proxy ARP in mind.
>>
>> What about security implications ?
>
>
> Yes, this feature will extend the attack surface for L2 networks. However,
> the attack vectors for this feature should be same as that of the gratuitous
> ARP, right? The same attack mitigation techniques for gratuitous ARPs is
> equally applicable here.
>
>> Will TCP flows be terminated, instead
>> of being smoothly migrated (TCP_REPAIR)
>
>
> The TCP flows will not be terminated. Upon receiving UNARP packet, the ARP
> entry will be marked FAILED. The subsequent TCP packets from the client
> (towards that IP) will be queued (the first 3 packets in arp_queue and then
> other packets get dropped) until the IP address is resolved again through
> the slow path neigh_resolve_output().
>
> The slow path marks the entry as INCOMPLETE and will start sending several
> ARP requests (ucast_solicit + app_solicit + mcast_solicit) to resolve the
> IP. If the resolution is successful, then the TCP packets will be sent out.
> If not, we will invalidate the ARP entry and call arp_error_report() on the
> queued packets (which will end up sending ICMP_HOST_UNREACH error). This
> behavior is same as what will occur if TCP server disappears in the middle
> of a connection.
>
>>
>> What about IPv6 ? Or maybe more abruptly, do we still need to add
>> features to IPv4 in 2017, 22 years after this RFC came ? ;)
>
>
> Legit question :). Well one thing I have seen in Networking is that an old
> idea circles back around later and turns out to be useful in new contexts
> and use cases. Like I enumerated in my initial email there are certain use
> cases in Cloud that might benefit from UNARP.
>
It doesn't make sense to have this implemented only for IPv4. At this time if
equivalent IPv6 feature is missing, I don't see it being useful / acceptable.
> regards,
> ~Girish
>
>>
>> Thanks.
>>
>>
>
^ permalink raw reply
* [next-queue PATCH v7 5/6] net/sched: Add support for HW offloading for CBS
From: Vinicius Costa Gomes @ 2017-10-13 0:40 UTC (permalink / raw)
To: netdev, intel-wired-lan
Cc: Vinicius Costa Gomes, jhs, xiyou.wangcong, jiri, andre.guedes,
ivan.briano, jesus.sanchez-palencia, boon.leong.ong,
richardcochran, henrik, levipearson, rodney.cummings
In-Reply-To: <20171013004005.17416-1-vinicius.gomes@intel.com>
This adds support for offloading the CBS algorithm to the controller,
if supported. Drivers wanting to support CBS offload must implement
the .ndo_setup_tc callback and handle the TC_SETUP_CBS (introduced
here) type.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
---
include/linux/netdevice.h | 1 +
include/net/pkt_sched.h | 9 ++++
net/sched/sch_cbs.c | 104 ++++++++++++++++++++++++++++++++++++++++------
3 files changed, 102 insertions(+), 12 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 31bb3010c69b..1f6c44ef5b21 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -775,6 +775,7 @@ enum tc_setup_type {
TC_SETUP_CLSFLOWER,
TC_SETUP_CLSMATCHALL,
TC_SETUP_CLSBPF,
+ TC_SETUP_CBS,
};
/* These structures hold the attributes of xdp state that are being passed
diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
index 259bc191ba59..7c597b050b36 100644
--- a/include/net/pkt_sched.h
+++ b/include/net/pkt_sched.h
@@ -146,4 +146,13 @@ static inline bool is_classid_clsact_egress(u32 classid)
TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_EGRESS);
}
+struct tc_cbs_qopt_offload {
+ u8 enable;
+ s32 queue;
+ s32 hicredit;
+ s32 locredit;
+ s32 idleslope;
+ s32 sendslope;
+};
+
#endif
diff --git a/net/sched/sch_cbs.c b/net/sched/sch_cbs.c
index 0643587e6dc8..7d2100c5b8aa 100644
--- a/net/sched/sch_cbs.c
+++ b/net/sched/sch_cbs.c
@@ -68,6 +68,8 @@
#define BYTES_PER_KBIT (1000 / 8)
struct cbs_sched_data {
+ bool offload;
+ int queue;
s64 port_rate; /* in bytes/s */
s64 last; /* timestamp in ns */
s64 credits; /* in bytes */
@@ -80,6 +82,11 @@ struct cbs_sched_data {
struct sk_buff *(*dequeue)(struct Qdisc *sch);
};
+static int cbs_enqueue_offload(struct sk_buff *skb, struct Qdisc *sch)
+{
+ return qdisc_enqueue_tail(skb, sch);
+}
+
static int cbs_enqueue_soft(struct sk_buff *skb, struct Qdisc *sch)
{
struct cbs_sched_data *q = qdisc_priv(sch);
@@ -190,6 +197,11 @@ static struct sk_buff *cbs_dequeue_soft(struct Qdisc *sch)
return skb;
}
+static struct sk_buff *cbs_dequeue_offload(struct Qdisc *sch)
+{
+ return qdisc_dequeue_head(sch);
+}
+
static struct sk_buff *cbs_dequeue(struct Qdisc *sch)
{
struct cbs_sched_data *q = qdisc_priv(sch);
@@ -201,14 +213,66 @@ static const struct nla_policy cbs_policy[TCA_CBS_MAX + 1] = {
[TCA_CBS_PARMS] = { .len = sizeof(struct tc_cbs_qopt) },
};
+static void cbs_disable_offload(struct net_device *dev,
+ struct cbs_sched_data *q)
+{
+ struct tc_cbs_qopt_offload cbs = { };
+ const struct net_device_ops *ops;
+ int err;
+
+ if (!q->offload)
+ return;
+
+ q->enqueue = cbs_enqueue_soft;
+ q->dequeue = cbs_dequeue_soft;
+
+ ops = dev->netdev_ops;
+ if (!ops->ndo_setup_tc)
+ return;
+
+ cbs.queue = q->queue;
+ cbs.enable = 0;
+
+ err = ops->ndo_setup_tc(dev, TC_SETUP_CBS, &cbs);
+ if (err < 0)
+ pr_warn("Couldn't disable CBS offload for queue %d\n",
+ cbs.queue);
+}
+
+static int cbs_enable_offload(struct net_device *dev, struct cbs_sched_data *q,
+ const struct tc_cbs_qopt *opt)
+{
+ const struct net_device_ops *ops = dev->netdev_ops;
+ struct tc_cbs_qopt_offload cbs = { };
+ int err;
+
+ if (!ops->ndo_setup_tc)
+ return -EOPNOTSUPP;
+
+ cbs.queue = q->queue;
+
+ cbs.enable = 1;
+ cbs.hicredit = opt->hicredit;
+ cbs.locredit = opt->locredit;
+ cbs.idleslope = opt->idleslope;
+ cbs.sendslope = opt->sendslope;
+
+ err = ops->ndo_setup_tc(dev, TC_SETUP_CBS, &cbs);
+ if (err < 0)
+ return err;
+
+ q->enqueue = cbs_enqueue_offload;
+ q->dequeue = cbs_dequeue_offload;
+
+ return 0;
+}
+
static int cbs_change(struct Qdisc *sch, struct nlattr *opt)
{
struct cbs_sched_data *q = qdisc_priv(sch);
struct net_device *dev = qdisc_dev(sch);
struct nlattr *tb[TCA_CBS_MAX + 1];
- struct ethtool_link_ksettings ecmd;
struct tc_cbs_qopt *qopt;
- s64 link_speed;
int err;
err = nla_parse_nested(tb, TCA_CBS_MAX, opt, cbs_policy, NULL);
@@ -220,23 +284,30 @@ static int cbs_change(struct Qdisc *sch, struct nlattr *opt)
qopt = nla_data(tb[TCA_CBS_PARMS]);
- if (qopt->offload)
- return -EOPNOTSUPP;
+ if (!qopt->offload) {
+ struct ethtool_link_ksettings ecmd;
+ s64 link_speed;
- if (!__ethtool_get_link_ksettings(dev, &ecmd))
- link_speed = ecmd.base.speed;
- else
- link_speed = SPEED_1000;
+ if (!__ethtool_get_link_ksettings(dev, &ecmd))
+ link_speed = ecmd.base.speed;
+ else
+ link_speed = SPEED_1000;
- q->port_rate = link_speed * 1000 * BYTES_PER_KBIT;
+ q->port_rate = link_speed * 1000 * BYTES_PER_KBIT;
- q->enqueue = cbs_enqueue_soft;
- q->dequeue = cbs_dequeue_soft;
+ cbs_disable_offload(dev, q);
+ } else {
+ err = cbs_enable_offload(dev, q, qopt);
+ if (err < 0)
+ return err;
+ }
+ /* Everything went OK, save the parameters used. */
q->hicredit = qopt->hicredit;
q->locredit = qopt->locredit;
q->idleslope = qopt->idleslope;
q->sendslope = qopt->sendslope;
+ q->offload = qopt->offload;
return 0;
}
@@ -244,10 +315,16 @@ static int cbs_change(struct Qdisc *sch, struct nlattr *opt)
static int cbs_init(struct Qdisc *sch, struct nlattr *opt)
{
struct cbs_sched_data *q = qdisc_priv(sch);
+ struct net_device *dev = qdisc_dev(sch);
if (!opt)
return -EINVAL;
+ q->queue = sch->dev_queue - netdev_get_tx_queue(dev, 0);
+
+ q->enqueue = cbs_enqueue_soft;
+ q->dequeue = cbs_dequeue_soft;
+
qdisc_watchdog_init(&q->watchdog, sch);
return cbs_change(sch, opt);
@@ -256,8 +333,11 @@ static int cbs_init(struct Qdisc *sch, struct nlattr *opt)
static void cbs_destroy(struct Qdisc *sch)
{
struct cbs_sched_data *q = qdisc_priv(sch);
+ struct net_device *dev = qdisc_dev(sch);
qdisc_watchdog_cancel(&q->watchdog);
+
+ cbs_disable_offload(dev, q);
}
static int cbs_dump(struct Qdisc *sch, struct sk_buff *skb)
@@ -274,7 +354,7 @@ static int cbs_dump(struct Qdisc *sch, struct sk_buff *skb)
opt.locredit = q->locredit;
opt.sendslope = q->sendslope;
opt.idleslope = q->idleslope;
- opt.offload = 0;
+ opt.offload = q->offload;
if (nla_put(skb, TCA_CBS_PARMS, sizeof(opt), &opt))
goto nla_put_failure;
--
2.14.2
^ permalink raw reply related
* [next-queue PATCH v7 6/6] igb: Add support for CBS offload
From: Vinicius Costa Gomes @ 2017-10-13 0:40 UTC (permalink / raw)
To: netdev, intel-wired-lan
Cc: Andre Guedes, jhs, xiyou.wangcong, jiri, ivan.briano,
jesus.sanchez-palencia, boon.leong.ong, richardcochran, henrik,
levipearson, rodney.cummings
In-Reply-To: <20171013004005.17416-1-vinicius.gomes@intel.com>
From: Andre Guedes <andre.guedes@intel.com>
This patch adds support for Credit-Based Shaper (CBS) qdisc offload
from Traffic Control system. This support enable us to leverage the
Forwarding and Queuing for Time-Sensitive Streams (FQTSS) features
from Intel i210 Ethernet Controller. FQTSS is the former 802.1Qav
standard which was merged into 802.1Q in 2014. It enables traffic
prioritization and bandwidth reservation via the Credit-Based Shaper
which is implemented in hardware by i210 controller.
The patch introduces the igb_setup_tc() function which implements the
support for CBS qdisc hardware offload in the IGB driver. CBS offload
is the only traffic control offload supported by the driver at the
moment.
FQTSS transmission mode from i210 controller is automatically enabled
by the IGB driver when the CBS is enabled for the first hardware
queue. Likewise, FQTSS mode is automatically disabled when CBS is
disabled for the last hardware queue. Changing FQTSS mode requires NIC
reset.
FQTSS feature is supported by i210 controller only.
Signed-off-by: Andre Guedes <andre.guedes@intel.com>
---
drivers/net/ethernet/intel/igb/e1000_defines.h | 23 ++
drivers/net/ethernet/intel/igb/e1000_regs.h | 8 +
drivers/net/ethernet/intel/igb/igb.h | 6 +
drivers/net/ethernet/intel/igb/igb_main.c | 347 +++++++++++++++++++++++++
4 files changed, 384 insertions(+)
diff --git a/drivers/net/ethernet/intel/igb/e1000_defines.h b/drivers/net/ethernet/intel/igb/e1000_defines.h
index 1de82f247312..83cabff1e0ab 100644
--- a/drivers/net/ethernet/intel/igb/e1000_defines.h
+++ b/drivers/net/ethernet/intel/igb/e1000_defines.h
@@ -353,7 +353,18 @@
#define E1000_RXPBS_CFG_TS_EN 0x80000000
#define I210_RXPBSIZE_DEFAULT 0x000000A2 /* RXPBSIZE default */
+#define I210_RXPBSIZE_MASK 0x0000003F
+#define I210_RXPBSIZE_PB_32KB 0x00000020
#define I210_TXPBSIZE_DEFAULT 0x04000014 /* TXPBSIZE default */
+#define I210_TXPBSIZE_MASK 0xC0FFFFFF
+#define I210_TXPBSIZE_PB0_8KB (8 << 0)
+#define I210_TXPBSIZE_PB1_8KB (8 << 6)
+#define I210_TXPBSIZE_PB2_4KB (4 << 12)
+#define I210_TXPBSIZE_PB3_4KB (4 << 18)
+
+#define I210_DTXMXPKTSZ_DEFAULT 0x00000098
+
+#define I210_SR_QUEUES_NUM 2
/* SerDes Control */
#define E1000_SCTL_DISABLE_SERDES_LOOPBACK 0x0400
@@ -1051,4 +1062,16 @@
#define E1000_VLAPQF_P_VALID(_n) (0x1 << (3 + (_n) * 4))
#define E1000_VLAPQF_QUEUE_MASK 0x03
+/* TX Qav Control fields */
+#define E1000_TQAVCTRL_XMIT_MODE BIT(0)
+#define E1000_TQAVCTRL_DATAFETCHARB BIT(4)
+#define E1000_TQAVCTRL_DATATRANARB BIT(8)
+
+/* TX Qav Credit Control fields */
+#define E1000_TQAVCC_IDLESLOPE_MASK 0xFFFF
+#define E1000_TQAVCC_QUEUEMODE BIT(31)
+
+/* Transmit Descriptor Control fields */
+#define E1000_TXDCTL_PRIORITY BIT(27)
+
#endif
diff --git a/drivers/net/ethernet/intel/igb/e1000_regs.h b/drivers/net/ethernet/intel/igb/e1000_regs.h
index 58adbf234e07..8eee081d395f 100644
--- a/drivers/net/ethernet/intel/igb/e1000_regs.h
+++ b/drivers/net/ethernet/intel/igb/e1000_regs.h
@@ -421,6 +421,14 @@ do { \
#define E1000_I210_FLA 0x1201C
+#define E1000_I210_DTXMXPKTSZ 0x355C
+
+#define E1000_I210_TXDCTL(_n) (0x0E028 + ((_n) * 0x40))
+
+#define E1000_I210_TQAVCTRL 0x3570
+#define E1000_I210_TQAVCC(_n) (0x3004 + ((_n) * 0x40))
+#define E1000_I210_TQAVHC(_n) (0x300C + ((_n) * 0x40))
+
#define E1000_INVM_DATA_REG(_n) (0x12120 + 4*(_n))
#define E1000_INVM_SIZE 64 /* Number of INVM Data Registers */
diff --git a/drivers/net/ethernet/intel/igb/igb.h b/drivers/net/ethernet/intel/igb/igb.h
index 06ffb2bc713e..92845692087a 100644
--- a/drivers/net/ethernet/intel/igb/igb.h
+++ b/drivers/net/ethernet/intel/igb/igb.h
@@ -281,6 +281,11 @@ struct igb_ring {
u16 count; /* number of desc. in the ring */
u8 queue_index; /* logical index of the ring*/
u8 reg_idx; /* physical index of the ring */
+ bool cbs_enable; /* indicates if CBS is enabled */
+ s32 idleslope; /* idleSlope in kbps */
+ s32 sendslope; /* sendSlope in kbps */
+ s32 hicredit; /* hiCredit in bytes */
+ s32 locredit; /* loCredit in bytes */
/* everything past this point are written often */
u16 next_to_clean;
@@ -621,6 +626,7 @@ struct igb_adapter {
#define IGB_FLAG_EEE BIT(14)
#define IGB_FLAG_VLAN_PROMISC BIT(15)
#define IGB_FLAG_RX_LEGACY BIT(16)
+#define IGB_FLAG_FQTSS BIT(17)
/* Media Auto Sense */
#define IGB_MAS_ENABLE_0 0X0001
diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c
index 837d9b46a390..be2cf263efa9 100644
--- a/drivers/net/ethernet/intel/igb/igb_main.c
+++ b/drivers/net/ethernet/intel/igb/igb_main.c
@@ -34,6 +34,7 @@
#include <linux/slab.h>
#include <net/checksum.h>
#include <net/ip6_checksum.h>
+#include <net/pkt_sched.h>
#include <linux/net_tstamp.h>
#include <linux/mii.h>
#include <linux/ethtool.h>
@@ -62,6 +63,17 @@
#define BUILD 0
#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
__stringify(BUILD) "-k"
+
+enum queue_mode {
+ QUEUE_MODE_STRICT_PRIORITY,
+ QUEUE_MODE_STREAM_RESERVATION,
+};
+
+enum tx_queue_prio {
+ TX_QUEUE_PRIO_HIGH,
+ TX_QUEUE_PRIO_LOW,
+};
+
char igb_driver_name[] = "igb";
char igb_driver_version[] = DRV_VERSION;
static const char igb_driver_string[] =
@@ -1271,6 +1283,12 @@ static int igb_alloc_q_vector(struct igb_adapter *adapter,
ring->count = adapter->tx_ring_count;
ring->queue_index = txr_idx;
+ ring->cbs_enable = false;
+ ring->idleslope = 0;
+ ring->sendslope = 0;
+ ring->hicredit = 0;
+ ring->locredit = 0;
+
u64_stats_init(&ring->tx_syncp);
u64_stats_init(&ring->tx_syncp2);
@@ -1598,6 +1616,284 @@ static void igb_get_hw_control(struct igb_adapter *adapter)
ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
}
+static void enable_fqtss(struct igb_adapter *adapter, bool enable)
+{
+ struct net_device *netdev = adapter->netdev;
+ struct e1000_hw *hw = &adapter->hw;
+
+ WARN_ON(hw->mac.type != e1000_i210);
+
+ if (enable)
+ adapter->flags |= IGB_FLAG_FQTSS;
+ else
+ adapter->flags &= ~IGB_FLAG_FQTSS;
+
+ if (netif_running(netdev))
+ schedule_work(&adapter->reset_task);
+}
+
+static bool is_fqtss_enabled(struct igb_adapter *adapter)
+{
+ return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
+}
+
+static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
+ enum tx_queue_prio prio)
+{
+ u32 val;
+
+ WARN_ON(hw->mac.type != e1000_i210);
+ WARN_ON(queue < 0 || queue > 4);
+
+ val = rd32(E1000_I210_TXDCTL(queue));
+
+ if (prio == TX_QUEUE_PRIO_HIGH)
+ val |= E1000_TXDCTL_PRIORITY;
+ else
+ val &= ~E1000_TXDCTL_PRIORITY;
+
+ wr32(E1000_I210_TXDCTL(queue), val);
+}
+
+static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
+{
+ u32 val;
+
+ WARN_ON(hw->mac.type != e1000_i210);
+ WARN_ON(queue < 0 || queue > 1);
+
+ val = rd32(E1000_I210_TQAVCC(queue));
+
+ if (mode == QUEUE_MODE_STREAM_RESERVATION)
+ val |= E1000_TQAVCC_QUEUEMODE;
+ else
+ val &= ~E1000_TQAVCC_QUEUEMODE;
+
+ wr32(E1000_I210_TQAVCC(queue), val);
+}
+
+/**
+ * igb_configure_cbs - Configure Credit-Based Shaper (CBS)
+ * @adapter: pointer to adapter struct
+ * @queue: queue number
+ * @enable: true = enable CBS, false = disable CBS
+ * @idleslope: idleSlope in kbps
+ * @sendslope: sendSlope in kbps
+ * @hicredit: hiCredit in bytes
+ * @locredit: loCredit in bytes
+ *
+ * Configure CBS for a given hardware queue. When disabling, idleslope,
+ * sendslope, hicredit, locredit arguments are ignored. Returns 0 if
+ * success. Negative otherwise.
+ **/
+static void igb_configure_cbs(struct igb_adapter *adapter, int queue,
+ bool enable, int idleslope, int sendslope,
+ int hicredit, int locredit)
+{
+ struct net_device *netdev = adapter->netdev;
+ struct e1000_hw *hw = &adapter->hw;
+ u32 tqavcc;
+ u16 value;
+
+ WARN_ON(hw->mac.type != e1000_i210);
+ WARN_ON(queue < 0 || queue > 1);
+
+ if (enable) {
+ set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
+ set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
+
+ /* According to i210 datasheet section 7.2.7.7, we should set
+ * the 'idleSlope' field from TQAVCC register following the
+ * equation:
+ *
+ * For 100 Mbps link speed:
+ *
+ * value = BW * 0x7735 * 0.2 (E1)
+ *
+ * For 1000Mbps link speed:
+ *
+ * value = BW * 0x7735 * 2 (E2)
+ *
+ * E1 and E2 can be merged into one equation as shown below.
+ * Note that 'link-speed' is in Mbps.
+ *
+ * value = BW * 0x7735 * 2 * link-speed
+ * -------------- (E3)
+ * 1000
+ *
+ * 'BW' is the percentage bandwidth out of full link speed
+ * which can be found with the following equation. Note that
+ * idleSlope here is the parameter from this function which
+ * is in kbps.
+ *
+ * BW = idleSlope
+ * ----------------- (E4)
+ * link-speed * 1000
+ *
+ * That said, we can come up with a generic equation to
+ * calculate the value we should set it TQAVCC register by
+ * replacing 'BW' in E3 by E4. The resulting equation is:
+ *
+ * value = idleSlope * 0x7735 * 2 * link-speed
+ * ----------------- -------------- (E5)
+ * link-speed * 1000 1000
+ *
+ * 'link-speed' is present in both sides of the fraction so
+ * it is canceled out. The final equation is the following:
+ *
+ * value = idleSlope * 61034
+ * ----------------- (E6)
+ * 1000000
+ */
+ value = DIV_ROUND_UP_ULL(idleslope * 61034ULL, 1000000);
+
+ tqavcc = rd32(E1000_I210_TQAVCC(queue));
+ tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
+ tqavcc |= value;
+ wr32(E1000_I210_TQAVCC(queue), tqavcc);
+
+ wr32(E1000_I210_TQAVHC(queue), 0x80000000 + hicredit * 0x7735);
+ } else {
+ set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
+ set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
+
+ /* Set idleSlope to zero. */
+ tqavcc = rd32(E1000_I210_TQAVCC(queue));
+ tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
+ wr32(E1000_I210_TQAVCC(queue), tqavcc);
+
+ /* Set hiCredit to zero. */
+ wr32(E1000_I210_TQAVHC(queue), 0);
+ }
+
+ /* XXX: In i210 controller the sendSlope and loCredit parameters from
+ * CBS are not configurable by software so we don't do any 'controller
+ * configuration' in respect to these parameters.
+ */
+
+ netdev_dbg(netdev, "CBS %s: queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
+ (enable) ? "enabled" : "disabled", queue,
+ idleslope, sendslope, hicredit, locredit);
+}
+
+static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
+ bool enable, int idleslope, int sendslope,
+ int hicredit, int locredit)
+{
+ struct igb_ring *ring;
+
+ if (queue < 0 || queue > adapter->num_tx_queues)
+ return -EINVAL;
+
+ ring = adapter->tx_ring[queue];
+
+ ring->cbs_enable = enable;
+ ring->idleslope = idleslope;
+ ring->sendslope = sendslope;
+ ring->hicredit = hicredit;
+ ring->locredit = locredit;
+
+ return 0;
+}
+
+static bool is_any_cbs_enabled(struct igb_adapter *adapter)
+{
+ struct igb_ring *ring;
+ int i;
+
+ for (i = 0; i < adapter->num_tx_queues; i++) {
+ ring = adapter->tx_ring[i];
+
+ if (ring->cbs_enable)
+ return true;
+ }
+
+ return false;
+}
+
+static void igb_setup_tx_mode(struct igb_adapter *adapter)
+{
+ struct net_device *netdev = adapter->netdev;
+ struct e1000_hw *hw = &adapter->hw;
+ u32 val;
+
+ /* Only i210 controller supports changing the transmission mode. */
+ if (hw->mac.type != e1000_i210)
+ return;
+
+ if (is_fqtss_enabled(adapter)) {
+ int i, max_queue;
+
+ /* Configure TQAVCTRL register: set transmit mode to 'Qav',
+ * set data fetch arbitration to 'round robin' and set data
+ * transfer arbitration to 'credit shaper algorithm.
+ */
+ val = rd32(E1000_I210_TQAVCTRL);
+ val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_DATATRANARB;
+ val &= ~E1000_TQAVCTRL_DATAFETCHARB;
+ wr32(E1000_I210_TQAVCTRL, val);
+
+ /* Configure Tx and Rx packet buffers sizes as described in
+ * i210 datasheet section 7.2.7.7.
+ */
+ val = rd32(E1000_TXPBS);
+ val &= ~I210_TXPBSIZE_MASK;
+ val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB |
+ I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB;
+ wr32(E1000_TXPBS, val);
+
+ val = rd32(E1000_RXPBS);
+ val &= ~I210_RXPBSIZE_MASK;
+ val |= I210_RXPBSIZE_PB_32KB;
+ wr32(E1000_RXPBS, val);
+
+ /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
+ * register should not exceed the buffer size programmed in
+ * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
+ * so according to the datasheet we should set MAX_TPKT_SIZE to
+ * 4kB / 64.
+ *
+ * However, when we do so, no frame from queue 2 and 3 are
+ * transmitted. It seems the MAX_TPKT_SIZE should not be great
+ * or _equal_ to the buffer size programmed in TXPBS. For this
+ * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
+ */
+ val = (4096 - 1) / 64;
+ wr32(E1000_I210_DTXMXPKTSZ, val);
+
+ /* Since FQTSS mode is enabled, apply any CBS configuration
+ * previously set. If no previous CBS configuration has been
+ * done, then the initial configuration is applied, which means
+ * CBS is disabled.
+ */
+ max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
+ adapter->num_tx_queues : I210_SR_QUEUES_NUM;
+
+ for (i = 0; i < max_queue; i++) {
+ struct igb_ring *ring = adapter->tx_ring[i];
+
+ igb_configure_cbs(adapter, i, ring->cbs_enable,
+ ring->idleslope, ring->sendslope,
+ ring->hicredit, ring->locredit);
+ }
+ } else {
+ wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
+ wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
+ wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
+
+ val = rd32(E1000_I210_TQAVCTRL);
+ /* According to Section 8.12.21, the other flags we've set when
+ * enabling FQTSS are not relevant when disabling FQTSS so we
+ * don't set they here.
+ */
+ val &= ~E1000_TQAVCTRL_XMIT_MODE;
+ wr32(E1000_I210_TQAVCTRL, val);
+ }
+
+ netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
+ "enabled" : "disabled");
+}
+
/**
* igb_configure - configure the hardware for RX and TX
* @adapter: private board structure
@@ -1609,6 +1905,7 @@ static void igb_configure(struct igb_adapter *adapter)
igb_get_hw_control(adapter);
igb_set_rx_mode(netdev);
+ igb_setup_tx_mode(adapter);
igb_restore_vlan(adapter);
@@ -2150,6 +2447,55 @@ igb_features_check(struct sk_buff *skb, struct net_device *dev,
return features;
}
+static int igb_offload_cbs(struct igb_adapter *adapter,
+ struct tc_cbs_qopt_offload *qopt)
+{
+ struct e1000_hw *hw = &adapter->hw;
+ int err;
+
+ /* CBS offloading is only supported by i210 controller. */
+ if (hw->mac.type != e1000_i210)
+ return -EOPNOTSUPP;
+
+ /* CBS offloading is only supported by queue 0 and queue 1. */
+ if (qopt->queue < 0 || qopt->queue > 1)
+ return -EINVAL;
+
+ err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
+ qopt->idleslope, qopt->sendslope,
+ qopt->hicredit, qopt->locredit);
+ if (err)
+ return err;
+
+ if (is_fqtss_enabled(adapter)) {
+ igb_configure_cbs(adapter, qopt->queue, qopt->enable,
+ qopt->idleslope, qopt->sendslope,
+ qopt->hicredit, qopt->locredit);
+
+ if (!is_any_cbs_enabled(adapter))
+ enable_fqtss(adapter, false);
+
+ } else {
+ enable_fqtss(adapter, true);
+ }
+
+ return 0;
+}
+
+static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
+ void *type_data)
+{
+ struct igb_adapter *adapter = netdev_priv(dev);
+
+ switch (type) {
+ case TC_SETUP_CBS:
+ return igb_offload_cbs(adapter, type_data);
+
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
static const struct net_device_ops igb_netdev_ops = {
.ndo_open = igb_open,
.ndo_stop = igb_close,
@@ -2175,6 +2521,7 @@ static const struct net_device_ops igb_netdev_ops = {
.ndo_set_features = igb_set_features,
.ndo_fdb_add = igb_ndo_fdb_add,
.ndo_features_check = igb_features_check,
+ .ndo_setup_tc = igb_setup_tc,
};
/**
--
2.14.2
^ permalink raw reply related
* [next-queue PATCH v7 3/6] net/sched: Add select_queue() class_ops for mqprio
From: Vinicius Costa Gomes @ 2017-10-13 0:40 UTC (permalink / raw)
To: netdev, intel-wired-lan
Cc: Jesus Sanchez-Palencia, jhs, xiyou.wangcong, jiri, andre.guedes,
ivan.briano, boon.leong.ong, richardcochran, henrik, levipearson,
rodney.cummings
In-Reply-To: <20171013004005.17416-1-vinicius.gomes@intel.com>
From: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
When replacing a child qdisc from mqprio, tc_modify_qdisc() must fetch
the netdev_queue pointer that the current child qdisc is associated
with before creating the new qdisc.
Currently, when using mqprio as root qdisc, the kernel will end up
getting the queue #0 pointer from the mqprio (root qdisc), which leaves
any new child qdisc with a possibly wrong netdev_queue pointer.
Implementing the Qdisc_class_ops select_queue() on mqprio fixes this
issue and avoid an inconsistent state when child qdiscs are replaced.
Signed-off-by: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
---
net/sched/sch_mqprio.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c
index 6bcdfe6e7b63..8c042ae323e3 100644
--- a/net/sched/sch_mqprio.c
+++ b/net/sched/sch_mqprio.c
@@ -396,6 +396,12 @@ static void mqprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
}
}
+static struct netdev_queue *mqprio_select_queue(struct Qdisc *sch,
+ struct tcmsg *tcm)
+{
+ return mqprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
+}
+
static const struct Qdisc_class_ops mqprio_class_ops = {
.graft = mqprio_graft,
.leaf = mqprio_leaf,
@@ -403,6 +409,7 @@ static const struct Qdisc_class_ops mqprio_class_ops = {
.walk = mqprio_walk,
.dump = mqprio_dump_class,
.dump_stats = mqprio_dump_class_stats,
+ .select_queue = mqprio_select_queue,
};
static struct Qdisc_ops mqprio_qdisc_ops __read_mostly = {
--
2.14.2
^ permalink raw reply related
* [next-queue PATCH v7 4/6] net/sched: Introduce Credit Based Shaper (CBS) qdisc
From: Vinicius Costa Gomes @ 2017-10-13 0:40 UTC (permalink / raw)
To: netdev, intel-wired-lan
Cc: Vinicius Costa Gomes, jhs, xiyou.wangcong, jiri, andre.guedes,
ivan.briano, jesus.sanchez-palencia, boon.leong.ong,
richardcochran, henrik, levipearson, rodney.cummings
In-Reply-To: <20171013004005.17416-1-vinicius.gomes@intel.com>
This queueing discipline implements the shaper algorithm defined by
the 802.1Q-2014 Section 8.6.8.2 and detailed in Annex L.
It's primary usage is to apply some bandwidth reservation to user
defined traffic classes, which are mapped to different queues via the
mqprio qdisc.
Only a simple software implementation is added for now.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
---
include/uapi/linux/pkt_sched.h | 18 +++
net/sched/Kconfig | 11 ++
net/sched/Makefile | 1 +
net/sched/sch_cbs.c | 314 +++++++++++++++++++++++++++++++++++++++++
4 files changed, 344 insertions(+)
create mode 100644 net/sched/sch_cbs.c
diff --git a/include/uapi/linux/pkt_sched.h b/include/uapi/linux/pkt_sched.h
index 099bf5528fed..41e349df4bf4 100644
--- a/include/uapi/linux/pkt_sched.h
+++ b/include/uapi/linux/pkt_sched.h
@@ -871,4 +871,22 @@ struct tc_pie_xstats {
__u32 maxq; /* maximum queue size */
__u32 ecn_mark; /* packets marked with ecn*/
};
+
+/* CBS */
+struct tc_cbs_qopt {
+ __u8 offload;
+ __s32 hicredit;
+ __s32 locredit;
+ __s32 idleslope;
+ __s32 sendslope;
+};
+
+enum {
+ TCA_CBS_UNSPEC,
+ TCA_CBS_PARMS,
+ __TCA_CBS_MAX,
+};
+
+#define TCA_CBS_MAX (__TCA_CBS_MAX - 1)
+
#endif
diff --git a/net/sched/Kconfig b/net/sched/Kconfig
index e70ed26485a2..c03d86a7775e 100644
--- a/net/sched/Kconfig
+++ b/net/sched/Kconfig
@@ -172,6 +172,17 @@ config NET_SCH_TBF
To compile this code as a module, choose M here: the
module will be called sch_tbf.
+config NET_SCH_CBS
+ tristate "Credit Based Shaper (CBS)"
+ ---help---
+ Say Y here if you want to use the Credit Based Shaper (CBS) packet
+ scheduling algorithm.
+
+ See the top of <file:net/sched/sch_cbs.c> for more details.
+
+ To compile this code as a module, choose M here: the
+ module will be called sch_cbs.
+
config NET_SCH_GRED
tristate "Generic Random Early Detection (GRED)"
---help---
diff --git a/net/sched/Makefile b/net/sched/Makefile
index 7b915d226de7..80c8f92d162d 100644
--- a/net/sched/Makefile
+++ b/net/sched/Makefile
@@ -52,6 +52,7 @@ obj-$(CONFIG_NET_SCH_FQ_CODEL) += sch_fq_codel.o
obj-$(CONFIG_NET_SCH_FQ) += sch_fq.o
obj-$(CONFIG_NET_SCH_HHF) += sch_hhf.o
obj-$(CONFIG_NET_SCH_PIE) += sch_pie.o
+obj-$(CONFIG_NET_SCH_CBS) += sch_cbs.o
obj-$(CONFIG_NET_CLS_U32) += cls_u32.o
obj-$(CONFIG_NET_CLS_ROUTE4) += cls_route.o
diff --git a/net/sched/sch_cbs.c b/net/sched/sch_cbs.c
new file mode 100644
index 000000000000..0643587e6dc8
--- /dev/null
+++ b/net/sched/sch_cbs.c
@@ -0,0 +1,314 @@
+/*
+ * net/sched/sch_cbs.c Credit Based Shaper
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ *
+ * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
+ *
+ */
+
+/* Credit Based Shaper (CBS)
+ * =========================
+ *
+ * This is a simple rate-limiting shaper aimed at TSN applications on
+ * systems with known traffic workloads.
+ *
+ * Its algorithm is defined by the IEEE 802.1Q-2014 Specification,
+ * Section 8.6.8.2, and explained in more detail in the Annex L of the
+ * same specification.
+ *
+ * There are four tunables to be considered:
+ *
+ * 'idleslope': Idleslope is the rate of credits that is
+ * accumulated (in kilobits per second) when there is at least
+ * one packet waiting for transmission. Packets are transmitted
+ * when the current value of credits is equal or greater than
+ * zero. When there is no packet to be transmitted the amount of
+ * credits is set to zero. This is the main tunable of the CBS
+ * algorithm.
+ *
+ * 'sendslope':
+ * Sendslope is the rate of credits that is depleted (it should be a
+ * negative number of kilobits per second) when a transmission is
+ * ocurring. It can be calculated as follows, (IEEE 802.1Q-2014 Section
+ * 8.6.8.2 item g):
+ *
+ * sendslope = idleslope - port_transmit_rate
+ *
+ * 'hicredit': Hicredit defines the maximum amount of credits (in
+ * bytes) that can be accumulated. Hicredit depends on the
+ * characteristics of interfering traffic,
+ * 'max_interference_size' is the maximum size of any burst of
+ * traffic that can delay the transmission of a frame that is
+ * available for transmission for this traffic class, (IEEE
+ * 802.1Q-2014 Annex L, Equation L-3):
+ *
+ * hicredit = max_interference_size * (idleslope / port_transmit_rate)
+ *
+ * 'locredit': Locredit is the minimum amount of credits that can
+ * be reached. It is a function of the traffic flowing through
+ * this qdisc (IEEE 802.1Q-2014 Annex L, Equation L-2):
+ *
+ * locredit = max_frame_size * (sendslope / port_transmit_rate)
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/skbuff.h>
+#include <net/netlink.h>
+#include <net/sch_generic.h>
+#include <net/pkt_sched.h>
+
+#define BYTES_PER_KBIT (1000 / 8)
+
+struct cbs_sched_data {
+ s64 port_rate; /* in bytes/s */
+ s64 last; /* timestamp in ns */
+ s64 credits; /* in bytes */
+ s32 locredit; /* in bytes */
+ s32 hicredit; /* in bytes */
+ s32 sendslope; /* in kbits/s */
+ s32 idleslope; /* in kbits/s */
+ struct qdisc_watchdog watchdog;
+ int (*enqueue)(struct sk_buff *skb, struct Qdisc *sch);
+ struct sk_buff *(*dequeue)(struct Qdisc *sch);
+};
+
+static int cbs_enqueue_soft(struct sk_buff *skb, struct Qdisc *sch)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+
+ if (sch->q.qlen == 0 && q->credits > 0) {
+ /* We need to stop accumulating credits when there's
+ * no packet enqueued packets and q->credits is
+ * positive.
+ */
+ q->credits = 0;
+ q->last = ktime_get_ns();
+ }
+
+ return qdisc_enqueue_tail(skb, sch);
+}
+
+static int cbs_enqueue(struct sk_buff *skb, struct Qdisc *sch,
+ struct sk_buff **to_free)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+
+ return q->enqueue(skb, sch);
+}
+
+/* timediff is in ns, slope is in kbps */
+static s64 timediff_to_credits(s64 timediff, s32 slope)
+{
+ s64 credits = timediff * slope * BYTES_PER_KBIT;
+
+ do_div(credits, NSEC_PER_SEC);
+
+ return credits;
+}
+
+static s64 delay_from_credits(s64 credits, s32 slope)
+{
+ s64 rate = slope * BYTES_PER_KBIT;
+ s64 delay;
+
+ if (unlikely(rate == 0))
+ return S64_MAX;
+
+ delay = -credits * NSEC_PER_SEC;
+ do_div(delay, rate);
+
+ return delay;
+}
+
+static s64 credits_from_len(unsigned int len, s32 slope, s64 port_rate)
+{
+ /* As do_div() only works on unsigned quantities, convert
+ * slope to a positive number here, and credits to a negative
+ * number before returning.
+ */
+ s64 rate = -slope * BYTES_PER_KBIT;
+ s64 credits;
+
+ if (unlikely(port_rate == 0))
+ return S64_MAX;
+
+ credits = len * rate;
+ do_div(credits, port_rate);
+
+ return -credits;
+}
+
+static struct sk_buff *cbs_dequeue_soft(struct Qdisc *sch)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+ s64 now = ktime_get_ns();
+ struct sk_buff *skb;
+ s64 credits;
+ int len;
+
+ if (q->credits < 0) {
+ credits = timediff_to_credits(now - q->last, q->idleslope);
+
+ credits = q->credits + credits;
+ q->credits = min_t(s64, credits, q->hicredit);
+
+ if (q->credits < 0) {
+ s64 delay;
+
+ delay = delay_from_credits(q->credits, q->idleslope);
+ qdisc_watchdog_schedule_ns(&q->watchdog, now + delay);
+
+ q->last = now;
+
+ return NULL;
+ }
+ }
+
+ skb = qdisc_dequeue_head(sch);
+ if (!skb)
+ return NULL;
+
+ len = qdisc_pkt_len(skb);
+
+ /* As sendslope is a negative number, this will decrease the
+ * amount of q->credits.
+ */
+ credits = credits_from_len(len, q->sendslope, q->port_rate);
+ credits += q->credits;
+
+ q->credits = max_t(s64, credits, q->locredit);
+ q->last = now;
+
+ return skb;
+}
+
+static struct sk_buff *cbs_dequeue(struct Qdisc *sch)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+
+ return q->dequeue(sch);
+}
+
+static const struct nla_policy cbs_policy[TCA_CBS_MAX + 1] = {
+ [TCA_CBS_PARMS] = { .len = sizeof(struct tc_cbs_qopt) },
+};
+
+static int cbs_change(struct Qdisc *sch, struct nlattr *opt)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+ struct net_device *dev = qdisc_dev(sch);
+ struct nlattr *tb[TCA_CBS_MAX + 1];
+ struct ethtool_link_ksettings ecmd;
+ struct tc_cbs_qopt *qopt;
+ s64 link_speed;
+ int err;
+
+ err = nla_parse_nested(tb, TCA_CBS_MAX, opt, cbs_policy, NULL);
+ if (err < 0)
+ return err;
+
+ if (!tb[TCA_CBS_PARMS])
+ return -EINVAL;
+
+ qopt = nla_data(tb[TCA_CBS_PARMS]);
+
+ if (qopt->offload)
+ return -EOPNOTSUPP;
+
+ if (!__ethtool_get_link_ksettings(dev, &ecmd))
+ link_speed = ecmd.base.speed;
+ else
+ link_speed = SPEED_1000;
+
+ q->port_rate = link_speed * 1000 * BYTES_PER_KBIT;
+
+ q->enqueue = cbs_enqueue_soft;
+ q->dequeue = cbs_dequeue_soft;
+
+ q->hicredit = qopt->hicredit;
+ q->locredit = qopt->locredit;
+ q->idleslope = qopt->idleslope;
+ q->sendslope = qopt->sendslope;
+
+ return 0;
+}
+
+static int cbs_init(struct Qdisc *sch, struct nlattr *opt)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+
+ if (!opt)
+ return -EINVAL;
+
+ qdisc_watchdog_init(&q->watchdog, sch);
+
+ return cbs_change(sch, opt);
+}
+
+static void cbs_destroy(struct Qdisc *sch)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+
+ qdisc_watchdog_cancel(&q->watchdog);
+}
+
+static int cbs_dump(struct Qdisc *sch, struct sk_buff *skb)
+{
+ struct cbs_sched_data *q = qdisc_priv(sch);
+ struct tc_cbs_qopt opt = { };
+ struct nlattr *nest;
+
+ nest = nla_nest_start(skb, TCA_OPTIONS);
+ if (!nest)
+ goto nla_put_failure;
+
+ opt.hicredit = q->hicredit;
+ opt.locredit = q->locredit;
+ opt.sendslope = q->sendslope;
+ opt.idleslope = q->idleslope;
+ opt.offload = 0;
+
+ if (nla_put(skb, TCA_CBS_PARMS, sizeof(opt), &opt))
+ goto nla_put_failure;
+
+ return nla_nest_end(skb, nest);
+
+nla_put_failure:
+ nla_nest_cancel(skb, nest);
+ return -1;
+}
+
+static struct Qdisc_ops cbs_qdisc_ops __read_mostly = {
+ .id = "cbs",
+ .priv_size = sizeof(struct cbs_sched_data),
+ .enqueue = cbs_enqueue,
+ .dequeue = cbs_dequeue,
+ .peek = qdisc_peek_dequeued,
+ .init = cbs_init,
+ .reset = qdisc_reset_queue,
+ .destroy = cbs_destroy,
+ .change = cbs_change,
+ .dump = cbs_dump,
+ .owner = THIS_MODULE,
+};
+
+static int __init cbs_module_init(void)
+{
+ return register_qdisc(&cbs_qdisc_ops);
+}
+
+static void __exit cbs_module_exit(void)
+{
+ unregister_qdisc(&cbs_qdisc_ops);
+}
+module_init(cbs_module_init)
+module_exit(cbs_module_exit)
+MODULE_LICENSE("GPL");
--
2.14.2
^ permalink raw reply related
* [next-queue PATCH v7 1/6] net/sched: Check for null dev_queue on create flow
From: Vinicius Costa Gomes @ 2017-10-13 0:40 UTC (permalink / raw)
To: netdev, intel-wired-lan
Cc: Jesus Sanchez-Palencia, jhs, xiyou.wangcong, jiri, andre.guedes,
ivan.briano, boon.leong.ong, richardcochran, henrik, levipearson,
rodney.cummings
In-Reply-To: <20171013004005.17416-1-vinicius.gomes@intel.com>
From: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
In qdisc_alloc() the dev_queue pointer was used without any checks
being performed. If qdisc_create() gets a null dev_queue pointer, it
just passes it along to qdisc_alloc(), leading to a crash. That
happens if a root qdisc implements select_queue() and returns a null
dev_queue pointer for an "invalid handle", for example, or if the
dev_queue associated with the parent qdisc is null.
This patch is in preparation for the next in this series, where
select_queue() is being added to mqprio and as it may return a null
dev_queue.
Signed-off-by: Jesus Sanchez-Palencia <jesus.sanchez-palencia@intel.com>
---
net/sched/sch_generic.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index a0a198768aad..de2408f1ccd3 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -603,8 +603,14 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
struct Qdisc *sch;
unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
int err = -ENOBUFS;
- struct net_device *dev = dev_queue->dev;
+ struct net_device *dev;
+
+ if (!dev_queue) {
+ err = -EINVAL;
+ goto errout;
+ }
+ dev = dev_queue->dev;
p = kzalloc_node(size, GFP_KERNEL,
netdev_queue_numa_node_read(dev_queue));
--
2.14.2
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox