linux-2.6-microblaze.git
3 years agoMerge branch 'add-an-assert-in-napi_consume_skb'
Jakub Kicinski [Wed, 25 Nov 2020 23:08:39 +0000 (15:08 -0800)]
Merge branch 'add-an-assert-in-napi_consume_skb'

Yunsheng Lin says:

====================
Add an assert in napi_consume_skb()

This patch introduces a lockdep_assert_in_softirq() interface and
uses it to assert the case when napi_consume_skb() is not called in
the softirq context.
====================

Link: https://lore.kernel.org/r/1606214969-97849-1-git-send-email-linyunsheng@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: Use lockdep_assert_in_softirq() in napi_consume_skb()
Yunsheng Lin [Tue, 24 Nov 2020 10:49:29 +0000 (18:49 +0800)]
net: Use lockdep_assert_in_softirq() in napi_consume_skb()

Use napi_consume_skb() to assert the case when it is not called
in a atomic softirq context.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agolockdep: Introduce in_softirq lockdep assert
Yunsheng Lin [Tue, 24 Nov 2020 10:49:28 +0000 (18:49 +0800)]
lockdep: Introduce in_softirq lockdep assert

The current semantic for napi_consume_skb() is that caller need
to provide non-zero budget when calling from NAPI context, and
breaking this semantic will cause hard to debug problem, because
_kfree_skb_defer() need to run in atomic context in order to push
the skb to the particular cpu' napi_alloc_cache atomically.

So add the lockdep_assert_in_softirq() to assert when the running
context is not in_softirq, in_softirq means softirq is serving or
BH is disabled, which has a ambiguous semantics due to the BH
disabled confusion, so add a comment to emphasize that.

And the softirq context can be interrupted by hard IRQ or NMI
context, lockdep_assert_in_softirq() need to assert about hard
IRQ or NMI context too.

Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agosoc: qcom: ipa: Constify static qmi structs
Rikard Falkeborn [Sun, 22 Nov 2020 23:40:30 +0000 (00:40 +0100)]
soc: qcom: ipa: Constify static qmi structs

These are only used as input arguments to qmi_handle_init() which
accepts const pointers to both qmi_ops and qmi_msg_handler. Make them
const to allow the compiler to put them in read-only memory.

Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Acked-by: Alex Elder <elder@linaro.org>
Link: https://lore.kernel.org/r/20201122234031.33432-2-rikard.falkeborn@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomptcp: be careful on MPTCP-level ack.
Paolo Abeni [Tue, 24 Nov 2020 21:51:24 +0000 (22:51 +0100)]
mptcp: be careful on MPTCP-level ack.

We can enter the main mptcp_recvmsg() loop even when
no subflows are connected. As note by Eric, that would
result in a divide by zero oops on ack generation.

Address the issue by checking the subflow status before
sending the ack.

Additionally protect mptcp_recvmsg() against invocation
with weird socket states.

v1 -> v2:
 - removed unneeded inline keyword - Jakub

Reported-and-suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Fixes: ea4ca586b16f ("mptcp: refine MPTCP-level ack scheduling")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/5370c0ae03449239e3d1674ddcfb090cf6f20abe.1606253206.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agobridge: mrp: Implement LC mode for MRP
Horatiu Vultur [Tue, 24 Nov 2020 08:25:25 +0000 (09:25 +0100)]
bridge: mrp: Implement LC mode for MRP

Extend MRP to support LC mode(link check) for the interconnect port.
This applies only to the interconnect ring.

Opposite to RC mode(ring check) the LC mode is using CFM frames to
detect when the link goes up or down and based on that the userspace
will need to react.
One advantage of the LC mode over RC mode is that there will be fewer
frames in the normal rings. Because RC mode generates InTest on all
ports while LC mode sends CFM frame only on the interconnect port.

All 4 nodes part of the interconnect ring needs to have the same mode.
And it is not possible to have running LC and RC mode at the same time
on a node.

Whenever the MIM starts it needs to detect the status of the other 3
nodes in the interconnect ring so it would send a frame called
InLinkStatus, on which the clients needs to reply with their link
status.

This patch adds InLinkStatus frame type and extends existing rules on
how to forward this frame.

Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Link: https://lore.kernel.org/r/20201124082525.273820-1-horatiu.vultur@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: sched: alias action flags with TCA_ACT_ prefix
Vlad Buslov [Tue, 24 Nov 2020 16:40:54 +0000 (18:40 +0200)]
net: sched: alias action flags with TCA_ACT_ prefix

Currently both filter and action flags use same "TCA_" prefix which makes
them hard to distinguish to code and confusing for users. Create aliases
for existing action flags constants with "TCA_ACT_" prefix.

Signed-off-by: Vlad Buslov <vlad@buslov.dev>
Link: https://lore.kernel.org/r/20201124164054.893168-1-vlad@buslov.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomptcp: put reference in mptcp timeout timer
Florian Westphal [Tue, 24 Nov 2020 16:24:46 +0000 (17:24 +0100)]
mptcp: put reference in mptcp timeout timer

On close this timer might be scheduled. mptcp uses sk_reset_timer for
this, so the a reference on the mptcp socket is taken.

This causes a refcount leak which can for example be reproduced
with 'mp_join_server_v4.pkt' from the mptcp-packetdrill repo.

The leak has nothing to do with join requests, v1_mp_capable_bind_no_cs.pkt
works too when replacing the last ack mpcapable to v1 instead of v0.

unreferenced object 0xffff888109bba040 (size 2744):
  comm "packetdrill", [..]
  backtrace:
    [..] sk_prot_alloc.isra.0+0x2b/0xc0
    [..] sk_clone_lock+0x2f/0x740
    [..] mptcp_sk_clone+0x33/0x1a0
    [..] subflow_syn_recv_sock+0x2b1/0x690 [..]

Fixes: e16163b6e2b7 ("mptcp: refactor shutdown and close")
Cc: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20201124162446.11448-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: realtek: read actual speed on rtl8211f to detect downshift
Antonio Borneo [Tue, 24 Nov 2020 23:07:56 +0000 (00:07 +0100)]
net: phy: realtek: read actual speed on rtl8211f to detect downshift

The rtl8211f supports downshift and before commit 5502b218e001
("net: phy: use phy_resolve_aneg_linkmode in genphy_read_status")
the read-back of register MII_CTRL1000 was used to detect the
negotiated link speed.
The code added in commit d445dff2df60 ("net: phy: realtek: read
actual speed to detect downshift") is working fine also for this
phy and it's trivial re-using it to restore the downshift
detection on rtl8211f.

Add the phy specific read_status() pointing to the existing
function rtlgen_read_status().

Signed-off-by: Antonio Borneo <antonio.borneo@st.com>
Link: https://lore.kernel.org/r/478f871a-583d-01f1-9cc5-2eea56d8c2a7@huawei.com
Tested-by: Yonglong Liu <liuyonglong@huawei.com>
Link: https://lore.kernel.org/r/20201124230756.887925-1-antonio.borneo@st.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-ptp-use-common-defines-for-ptp-message-types-in-further-drivers'
Jakub Kicinski [Wed, 25 Nov 2020 20:23:26 +0000 (12:23 -0800)]
Merge branch 'net-ptp-use-common-defines-for-ptp-message-types-in-further-drivers'

Christian Eggers says:

====================
net: ptp: use common defines for PTP message types in further drivers

This series replaces further driver internal enumeration / uses of magic
numbers with the newly introduced PTP_MSGTYPE_* defines.
====================

Link: https://lore.kernel.org/r/20201124074418.2609-1-ceggers@arri.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: mscc: use new PTP_MSGTYPE_* defines
Christian Eggers [Tue, 24 Nov 2020 07:44:18 +0000 (08:44 +0100)]
net: phy: mscc: use new PTP_MSGTYPE_* defines

Use recently introduced PTP_MSGTYPE_SYNC and PTP_MSGTYPE_DELAY_REQ
defines instead of a driver internal enumeration.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Reviewed-by: Antoine Tenart <atenart@kernel.org>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_ptp: use PTP wide message type definitions
Christian Eggers [Tue, 24 Nov 2020 07:44:17 +0000 (08:44 +0100)]
mlxsw: spectrum_ptp: use PTP wide message type definitions

Use recently introduced PTP wide defines instead of a driver internal
enumeration.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Cc: Petr Machata <petrm@mellanox.com>
Cc: Jiri Pirko <jiri@nvidia.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: dp83640: use new PTP_MSGTYPE_SYNC define
Christian Eggers [Tue, 24 Nov 2020 07:44:16 +0000 (08:44 +0100)]
net: phy: dp83640: use new PTP_MSGTYPE_SYNC define

Replace use of magic number with recently introduced define.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-phy-add-support-for-shared-interrupts-part-3'
Jakub Kicinski [Wed, 25 Nov 2020 19:18:41 +0000 (11:18 -0800)]
Merge branch 'net-phy-add-support-for-shared-interrupts-part-3'

Ioana Ciornei says:

====================
net: phy: add support for shared interrupts (part 3)

This patch set aims to actually add support for shared interrupts in
phylib and not only for multi-PHY devices. While we are at it,
streamline the interrupt handling in phylib.

For a bit of context, at the moment, there are multiple phy_driver ops
that deal with this subject:

- .config_intr() - Enable/disable the interrupt line.

- .ack_interrupt() - Should quiesce any interrupts that may have been
  fired.  It's also used by phylib in conjunction with .config_intr() to
  clear any pending interrupts after the line was disabled, and before
  it is going to be enabled.

- .did_interrupt() - Intended for multi-PHY devices with a shared IRQ
  line and used by phylib to discern which PHY from the package was the
  one that actually fired the interrupt.

- .handle_interrupt() - Completely overrides the default interrupt
  handling logic from phylib. The PHY driver is responsible for checking
  if any interrupt was fired by the respective PHY and choose
  accordingly if it's the one that should trigger the link state machine.

From my point of view, the interrupt handling in phylib has become
somewhat confusing with all these callbacks that actually read the same
PHY register - the interrupt status.  A more streamlined approach would
be to just move the responsibility to write an interrupt handler to the
driver (as any other device driver does) and make .handle_interrupt()
the only way to deal with interrupts.

Another advantage with this approach would be that phylib would gain
support for shared IRQs between different PHY (not just multi-PHY
devices), something which at the moment would require extending every
PHY driver anyway in order to implement their .did_interrupt() callback
and duplicate the same logic as in .ack_interrupt(). The disadvantage
of making .did_interrupt() mandatory would be that we are slightly
changing the semantics of the phylib API and that would increase
confusion instead of reducing it.

What I am proposing is the following:

- As a first step, make the .ack_interrupt() callback optional so that
  we do not break any PHY driver amid the transition.

- Every PHY driver gains a .handle_interrupt() implementation that, for
  the most part, would look like below:

irq_status = phy_read(phydev, INTR_STATUS);
if (irq_status < 0) {
phy_error(phydev);
return IRQ_NONE;
}

if (!(irq_status & irq_mask))
return IRQ_NONE;

phy_trigger_machine(phydev);

return IRQ_HANDLED;

- Remove each PHY driver's implementation of the .ack_interrupt() by
  actually taking care of quiescing any pending interrupts before
  enabling/after disabling the interrupt line.

- Finally, after all drivers have been ported, remove the
  .ack_interrupt() and .did_interrupt() callbacks from phy_driver.

This patch set is part 3 (and final) of the entire change set and it
addresses the remaining PHY drivers that have not been migrated
previosly. Also, it finally removed the .did_interrupt() and
.ack_interrupt() callbacks since they are of no use anymore.

I do not have access to most of these PHY's, therefore I Cc-ed the
latest contributors to the individual PHY drivers in order to have
access, hopefully, to more regression testing.
====================

Link: https://lore.kernel.org/r/20201123153817.1616814-1-ciorneiioana@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: remove the .did_interrupt() and .ack_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:17 +0000 (17:38 +0200)]
net: phy: remove the .did_interrupt() and .ack_interrupt() callback

Now that all the PHY drivers have been migrated to directly implement
the generic .handle_interrupt() callback for a seamless support of
shared IRQs and all the .config_inter() implementations clear any
pending interrupts, we can safely remove the two callbacks.

With this patch, phylib has a proper support for shared IRQs (and not
just for multi-PHY devices. A PHY driver must implement both the
.handle_interrupt() and .config_intr() callbacks for the IRQs to be
actually used.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: qsemi: remove the use of .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:16 +0000 (17:38 +0200)]
net: phy: qsemi: remove the use of .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Also, add a comment describing the multiple step interrupt
acknoledgement process.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: qsemi: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:15 +0000 (17:38 +0200)]
net: phy: qsemi: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: ti: remove the use of .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:14 +0000 (17:38 +0200)]
net: phy: ti: remove the use of .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Cc: Dan Murphy <dmurphy@ti.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: ti: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:13 +0000 (17:38 +0200)]
net: phy: ti: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Cc: Dan Murphy <dmurphy@ti.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: national: remove the use of the .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:12 +0000 (17:38 +0200)]
net: phy: national: remove the use of the .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: national: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:11 +0000 (17:38 +0200)]
net: phy: national: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: micrel: remove the use of .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:10 +0000 (17:38 +0200)]
net: phy: micrel: remove the use of .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Cc: Divya Koppera <Divya.Koppera@microchip.com>
Cc: Oleksij Rempel <o.rempel@pengutronix.de>
Cc: Philippe Schenker <philippe.schenker@toradex.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Antoine Tenart <atenart@kernel.org>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: micrel: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:09 +0000 (17:38 +0200)]
net: phy: micrel: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Cc: Divya Koppera <Divya.Koppera@microchip.com>
Cc: Oleksij Rempel <o.rempel@pengutronix.de>
Cc: Philippe Schenker <philippe.schenker@toradex.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Antoine Tenart <atenart@kernel.org>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: meson-gxl: remove the use of .ack_callback()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:08 +0000 (17:38 +0200)]
net: phy: meson-gxl: remove the use of .ack_callback()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Cc: Jerome Brunet <jbrunet@baylibre.com>
Cc: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: meson-gxl: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:07 +0000 (17:38 +0200)]
net: phy: meson-gxl: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Cc: Jerome Brunet <jbrunet@baylibre.com>
Cc: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: icplus: remove the use .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:06 +0000 (17:38 +0200)]
net: phy: icplus: remove the use .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Cc: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: icplus: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:05 +0000 (17:38 +0200)]
net: phy: icplus: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Cc: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: intel-xway: remove the use of .ack_interrupt()
Ioana Ciornei [Mon, 23 Nov 2020 15:38:04 +0000 (17:38 +0200)]
net: phy: intel-xway: remove the use of .ack_interrupt()

In preparation of removing the .ack_interrupt() callback, we must replace
its occurrences (aka phy_clear_interrupt), from the 2 places where it is
called from (phy_enable_interrupts and phy_disable_interrupts), with
equivalent functionality.

This means that clearing interrupts now becomes something that the PHY
driver is responsible of doing, before enabling interrupts and after
clearing them. Make this driver follow the new contract.

Cc: Mathias Kresin <dev@kresin.me>
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: intel-xway: implement generic .handle_interrupt() callback
Ioana Ciornei [Mon, 23 Nov 2020 15:38:03 +0000 (17:38 +0200)]
net: phy: intel-xway: implement generic .handle_interrupt() callback

In an attempt to actually support shared IRQs in phylib, we now move the
responsibility of triggering the phylib state machine or just returning
IRQ_NONE, based on the IRQ status register, to the PHY driver. Having
3 different IRQ handling callbacks (.handle_interrupt(),
.did_interrupt() and .ack_interrupt() ) is confusing so let the PHY
driver implement directly an IRQ handler like any other device driver.
Make this driver follow the new convention.

Cc: Mathias Kresin <dev@kresin.me>
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agolan743x: replace polling loop by wait_event_timeout()
Sven Van Asbroeck [Mon, 23 Nov 2020 19:15:29 +0000 (14:15 -0500)]
lan743x: replace polling loop by wait_event_timeout()

The driver's ISR sends a 'software interrupt' event to the probe()
thread using the following method:
- probe(): write 0 to flag, enable s/w interrupt
- probe(): poll on flag, relax using usleep_range()
- ISR    : write 1 to flag

Replace with wake_up() / wait_event_timeout(). Besides being easier
to get right, this abstraction has better timing and memory
consistency properties.

Tested-by: Sven Van Asbroeck <thesven73@gmail.com> # lan7430
Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com>
Link: https://lore.kernel.org/r/20201123191529.14908-2-TheSven73@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agolan743x: clean up software_isr function
Sven Van Asbroeck [Mon, 23 Nov 2020 19:15:28 +0000 (14:15 -0500)]
lan743x: clean up software_isr function

For no apparent reason, this function reads the INT_STS register, and
checks if the software interrupt bit is set. These things have already
been carried out by this function's only caller.

Clean up by removing the redundant code.

Tested-by: Sven Van Asbroeck <thesven73@gmail.com> # lan7430
Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com>
Link: https://lore.kernel.org/r/20201123191529.14908-1-TheSven73@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'mvneta-access-skb_shared_info-only-on-last-frag'
Jakub Kicinski [Tue, 24 Nov 2020 23:09:48 +0000 (15:09 -0800)]
Merge branch 'mvneta-access-skb_shared_info-only-on-last-frag'

Lorenzo Bianconi says:

====================
mvneta: access skb_shared_info only on last frag

Build skb_shared_info on mvneta_rx_swbm stack and sync it to xdp_buff
skb_shared_info area only on the last fragment.
Avoid avoid unnecessary xdp_buff initialization in mvneta_rx_swbm routine.
This a preliminary series to complete xdp multi-buff in mvneta driver.
====================

Link: https://lore.kernel.org/r/cover.1605889258.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: mvneta: alloc skb_shared_info on the mvneta_rx_swbm stack
Lorenzo Bianconi [Fri, 20 Nov 2020 17:05:44 +0000 (18:05 +0100)]
net: mvneta: alloc skb_shared_info on the mvneta_rx_swbm stack

Build skb_shared_info on mvneta_rx_swbm stack and sync it to xdp_buff
skb_shared_info area only on the last fragment. Leftover cache miss in
mvneta_swbm_rx_frame will be addressed introducing mb bit in
xdp_buff/xdp_frame struct

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: mvneta: move skb_shared_info in mvneta_xdp_put_buff caller
Lorenzo Bianconi [Fri, 20 Nov 2020 17:05:43 +0000 (18:05 +0100)]
net: mvneta: move skb_shared_info in mvneta_xdp_put_buff caller

Pass skb_shared_info pointer from mvneta_xdp_put_buff caller. This is a
preliminary patch to reduce accesses to skb_shared_info area and reduce
cache misses.
Remove napi parameter in mvneta_xdp_put_buff signature since it is always
run in NAPI context

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: mvneta: avoid unnecessary xdp_buff initialization
Lorenzo Bianconi [Fri, 20 Nov 2020 17:05:42 +0000 (18:05 +0100)]
net: mvneta: avoid unnecessary xdp_buff initialization

Explicitly initialize mandatory fields defining xdp_buff struct
in mvneta_rx_swbm routine

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: mvpp2: divide fifo for dts-active ports only
Stefan Chulski [Mon, 23 Nov 2020 17:54:33 +0000 (19:54 +0200)]
net: mvpp2: divide fifo for dts-active ports only

Tx/Rx FIFO is a HW resource limited by total size, but shared
by all ports of same CP110 and impacting port-performance.
Do not divide the FIFO for ports which are not enabled in DTS,
so active ports could have more FIFO.
No change in FIFO allocation if all 3 ports on the communication
processor enabled in DTS.

The active port mapping should be done in probe before FIFO-init.

Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Reviewed-by: Russell King <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/1606154073-28267-1-git-send-email-stefanc@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: page_pool: Add page_pool_put_page_bulk() to page_pool.rst
Lorenzo Bianconi [Mon, 23 Nov 2020 15:45:46 +0000 (16:45 +0100)]
net: page_pool: Add page_pool_put_page_bulk() to page_pool.rst

Introduce page_pool_put_page_bulk() entry into the API section of
page_pool.rst

Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/a6a5141b4d7b7b71fa7778b16b48f80095dd3233.1606146163.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: warn if gso_type isn't set for a GSO SKB
Heiner Kallweit [Fri, 20 Nov 2020 23:22:20 +0000 (00:22 +0100)]
net: warn if gso_type isn't set for a GSO SKB

In bug report [0] a warning in r8169 driver was reported that was
caused by an invalid GSO SKB (gso_type was 0). See [1] for a discussion
about this issue. Still the origin of the invalid GSO SKB isn't clear.

It shouldn't be a network drivers task to check for invalid GSO SKB's.
Also, even if issue [0] can be fixed, we can't be sure that a
similar issue doesn't pop up again at another place.
Therefore let gso_features_check() check for such invalid GSO SKB's.

[0] https://bugzilla.kernel.org/show_bug.cgi?id=209423
[1] https://www.spinics.net/lists/netdev/msg690794.html

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://lore.kernel.org/r/97c78d21-7f0b-d843-df17-3589f224d2cf@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: fix spelling mistake "memroy" -> "memory"
Colin Ian King [Mon, 23 Nov 2020 10:34:52 +0000 (10:34 +0000)]
net: hns3: fix spelling mistake "memroy" -> "memory"

There are spelling mistakes in two dev_err messages. Fix them.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Link: https://lore.kernel.org/r/20201123103452.197708-1-colin.king@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'mlxsw-add-support-for-blackhole-nexthops'
Jakub Kicinski [Tue, 24 Nov 2020 20:15:00 +0000 (12:15 -0800)]
Merge branch 'mlxsw-add-support-for-blackhole-nexthops'

Ido Schimmel says:

====================
mlxsw: Add support for blackhole nexthops

This patch set adds support for blackhole nexthops in mlxsw. These
nexthops are exactly the same as other nexthops, but instead of
forwarding packets to an egress router interface (RIF), they are
programmed to silently drop them.

Patches #1-#4 are preparations.

Patch #5 adds support for blackhole nexthops and removes the check that
prevented them from being programmed.

Patch #6 adds a selftests over mlxsw which tests that blackhole nexthops
can be programmed and are marked as offloaded.

Patch #7 extends the existing nexthop forwarding test to also test
blackhole functionality.

Patches #8-#10 add support for a new packet trap ('blackhole_nexthop')
which should be triggered whenever packets are dropped by a blackhole
nexthop. Obviously, by default, the trap action is set to 'drop' so that
dropped packets will not be reported.
====================

Link: https://lore.kernel.org/r/20201123071230.676469-1-idosch@idosch.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: mlxsw: Add blackhole_nexthop trap test
Ido Schimmel [Mon, 23 Nov 2020 07:12:30 +0000 (09:12 +0200)]
selftests: mlxsw: Add blackhole_nexthop trap test

Test that packets hitting a blackhole nexthop are trapped to the CPU
when the trap is enabled. Test that packets are not reported when the
trap is disabled.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_trap: Add blackhole_nexthop trap
Ido Schimmel [Mon, 23 Nov 2020 07:12:29 +0000 (09:12 +0200)]
mlxsw: spectrum_trap: Add blackhole_nexthop trap

Register with devlink the blackhole_nexthop trap so that mlxsw will be
able to report packets dropped due to a blackhole nexthop.

The internal trap identifier is "DISCARD_ROUTER3", which traps packets
dropped in the adjacency table.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agodevlink: Add blackhole_nexthop trap
Ido Schimmel [Mon, 23 Nov 2020 07:12:28 +0000 (09:12 +0200)]
devlink: Add blackhole_nexthop trap

Add a packet trap to report packets that were dropped due to a
blackhole nexthop.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: forwarding: Add blackhole nexthops tests
Ido Schimmel [Mon, 23 Nov 2020 07:12:27 +0000 (09:12 +0200)]
selftests: forwarding: Add blackhole nexthops tests

Test that IPv4 and IPv6 ping fail when the route is using a blackhole
nexthop or a group with a blackhole nexthop. Test that ping passes when
the route starts using a valid nexthop.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: mlxsw: Add blackhole nexthop configuration tests
Ido Schimmel [Mon, 23 Nov 2020 07:12:26 +0000 (09:12 +0200)]
selftests: mlxsw: Add blackhole nexthop configuration tests

Test the mlxsw allows blackhole nexthops to be installed and that the
nexthops are marked as offloaded.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_router: Add support for blackhole nexthops
Ido Schimmel [Mon, 23 Nov 2020 07:12:25 +0000 (09:12 +0200)]
mlxsw: spectrum_router: Add support for blackhole nexthops

Add support for blackhole nexthops by programming them to the adjacency
table with a discard action.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_router: Resolve RIF from nexthop struct instead of neighbour
Ido Schimmel [Mon, 23 Nov 2020 07:12:24 +0000 (09:12 +0200)]
mlxsw: spectrum_router: Resolve RIF from nexthop struct instead of neighbour

The two are the same, but for blackhole nexthops we will not have an
associated neighbour struct, so resolve the RIF from the nexthop struct
itself instead.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_router: Use loopback RIF for unresolved nexthops
Ido Schimmel [Mon, 23 Nov 2020 07:12:23 +0000 (09:12 +0200)]
mlxsw: spectrum_router: Use loopback RIF for unresolved nexthops

Now that the driver creates a loopback RIF during its initialization, it
can be used to program the adjacency entries for unresolved nexthops
instead of other RIFs. The loopback RIF is guaranteed to exist for the
entire life time of the driver, unlike other RIFs that come and go.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_router: Use different trap identifier for unresolved nexthops
Ido Schimmel [Mon, 23 Nov 2020 07:12:22 +0000 (09:12 +0200)]
mlxsw: spectrum_router: Use different trap identifier for unresolved nexthops

Unresolved nexthops are currently written to the adjacency table with a
discard action. Packets hitting such entries are trapped to the CPU via
the 'DISCARD_ROUTER3' trap which can be enabled or disabled on demand,
but is always enabled in order to ensure the kernel can resolve the
unresolved neighbours.

This trap will be needed for blackhole nexthops support. Therefore, move
unresolved nexthops to explicitly program the adjacency entries with a
trap action and a different trap identifier, 'RTR_EGRESS0'.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agomlxsw: spectrum_router: Create loopback RIF during initialization
Ido Schimmel [Mon, 23 Nov 2020 07:12:21 +0000 (09:12 +0200)]
mlxsw: spectrum_router: Create loopback RIF during initialization

Up until now RIFs (router interfaces) were created on demand (e.g.,
when an IP address was added to a netdev). However, sometimes the device
needs to be provided with a RIF when one might not be available.

For example, adjacency entries that drop packets need to be programmed
with an egress RIF despite the RIF not being used to forward packets.

Create such a RIF during initialization so that it could be used later
on to support blackhole nexthops.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge tag 'rxrpc-next-20201123' of git://git.kernel.org/pub/scm/linux/kernel/git...
Jakub Kicinski [Tue, 24 Nov 2020 20:05:58 +0000 (12:05 -0800)]
Merge tag 'rxrpc-next-20201123' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Prelude to gssapi support

Here are some patches that do some reorganisation of the security class
handling in rxrpc to allow implementation of the RxGK security class that
will allow AF_RXRPC to use GSSAPI-negotiated tokens and better crypto.  The
RxGK security class is not included in this patchset.

It does the following things:

 (1) Add a keyrings patch to provide the original key description, as
     provided to add_key(), to the payload preparser so that it can
     interpret the content on that basis.  Unfortunately, the rxrpc_s key
     type wasn't written to interpret its payload as anything other than a
     string of bytes comprising a key, but for RxGK, more information is
     required as multiple Kerberos enctypes are supported.

 (2) Remove the rxk5 security class key parsing.  The rxk5 class never got
     rolled out in OpenAFS and got replaced with rxgk.

 (3) Support the creation of rxrpc keys with multiple tokens of different
     types.  If some types are not supported, the ENOPKG error is
     suppressed if at least one other token's type is supported.

 (4) Punt the handling of server keys (rxrpc_s type) to the appropriate
     security class.

 (5) Organise the security bits in the rxrpc_connection struct into a
     union to make it easier to override for other classes.

 (6) Move some bits from core code into rxkad that won't be appropriate to
     rxgk.

* tag 'rxrpc-next-20201123' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
  rxrpc: Ask the security class how much space to allow in a packet
  rxrpc: rxkad: Don't use pskb_pull() to advance through the response packet
  rxrpc: Organise connection security to use a union
  rxrpc: Don't reserve security header in Tx DATA skbuff
  rxrpc: Merge prime_packet_security into init_connection_security
  rxrpc: Fix example key name in a comment
  rxrpc: Ignore unknown tokens in key payload unless no known tokens
  rxrpc: Make the parsing of xdr payloads more coherent
  rxrpc: Allow security classes to give more info on server keys
  rxrpc: Don't leak the service-side session key to userspace
  rxrpc: Hand server key parsing off to the security class
  rxrpc: Split the server key type (rxrpc_s) into its own file
  rxrpc: Don't retain the server key in the connection
  rxrpc: Support keys with multiple authentication tokens
  rxrpc: List the held token types in the key description in /proc/keys
  rxrpc: Remove the rxk5 security class as it's now defunct
  keys: Provide the original description to the key preparser
====================

Link: https://lore.kernel.org/r/160616220405.830164.2239716599743995145.stgit@warthog.procyon.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: stmmac: Use hrtimer for TX coalescing
Vincent Whitchurch [Fri, 20 Nov 2020 15:02:08 +0000 (16:02 +0100)]
net: stmmac: Use hrtimer for TX coalescing

This driver uses a normal timer for TX coalescing, which means that the
with the default tx-usecs of 1000 microseconds the cleanups actually
happen 10 ms or more later with HZ=100.  This leads to very low
througput with TCP when bridged to a slow link such as a 4G modem.  Fix
this by using an hrtimer instead.

On my ARM platform with HZ=100 and the default TX coalescing settings
(tx-frames 25 tx-usecs 1000), with "tc qdisc add dev eth0 root netem
delay 60ms 40ms rate 50Mbit" run on the server, netperf's TCP_STREAM
improves from ~5.5 Mbps to ~100 Mbps.

Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Link: https://lore.kernel.org/r/20201120150208.6838-1-vincent.whitchurch@axis.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agosctp: Fix some typo
Christophe JAILLET [Sun, 22 Nov 2020 18:07:04 +0000 (19:07 +0100)]
sctp: Fix some typo

s/tranport/transport/

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://lore.kernel.org/r/20201122180704.1366636-1-christophe.jaillet@wanadoo.fr
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: don't include ethtool.h from netdevice.h
Jakub Kicinski [Fri, 20 Nov 2020 22:50:52 +0000 (14:50 -0800)]
net: don't include ethtool.h from netdevice.h

linux/netdevice.h is included in very many places, touching any
of its dependecies causes large incremental builds.

Drop the linux/ethtool.h include, linux/netdevice.h just needs
a forward declaration of struct ethtool_ops.

Fix all the places which made use of this implicit include.

Acked-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Shannon Nelson <snelson@pensando.io>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Link: https://lore.kernel.org/r/20201120225052.1427503-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: pch_gbe: Use 'dma_free_coherent()' to undo 'dma_alloc_coherent()'
Christophe JAILLET [Sat, 21 Nov 2020 09:03:30 +0000 (10:03 +0100)]
net: pch_gbe: Use 'dma_free_coherent()' to undo 'dma_alloc_coherent()'

Memory allocation are done with 'dma_alloc_coherent()'. Be consistent
and use 'dma_free_coherent()' to free the corresponding memory.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://lore.kernel.org/r/20201121090330.1332543-1-christophe.jaillet@wanadoo.fr
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: pch_gbe: Use dma_set_mask_and_coherent to simplify code
Christophe JAILLET [Sat, 21 Nov 2020 09:03:02 +0000 (10:03 +0100)]
net: pch_gbe: Use dma_set_mask_and_coherent to simplify code

'pci_set_dma_mask()' + 'pci_set_consistent_dma_mask()' can be replaced by
an equivalent 'dma_set_mask_and_coherent()' which is much less verbose.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://lore.kernel.org/r/20201121090302.1332491-1-christophe.jaillet@wanadoo.fr
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-dsa-hellcreek-minor-cleanups'
Jakub Kicinski [Tue, 24 Nov 2020 00:57:23 +0000 (16:57 -0800)]
Merge branch 'net-dsa-hellcreek-minor-cleanups'

Kurt Kanzenbach says:

====================
net: dsa: hellcreek: Minor cleanups

fix two minor issues in the hellcreek driver.
====================

Link: https://lore.kernel.org/r/20201121114455.22422-1-kurt@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: dsa: hellcreek: Don't print error message on defer
Kurt Kanzenbach [Sat, 21 Nov 2020 11:44:55 +0000 (12:44 +0100)]
net: dsa: hellcreek: Don't print error message on defer

When DSA is not loaded when the driver is probed an error message is
printed. But, that's not really an error, just a defer. Use dev_err_probe()
instead.

Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: dsa: tag_hellcreek: Cleanup includes
Kurt Kanzenbach [Sat, 21 Nov 2020 11:44:54 +0000 (12:44 +0100)]
net: dsa: tag_hellcreek: Cleanup includes

Remove unused and add needed includes. No functional change.

Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-ptp-introduce-common-defines-for-ptp-message-types'
Jakub Kicinski [Mon, 23 Nov 2020 21:43:48 +0000 (13:43 -0800)]
Merge branch 'net-ptp-introduce-common-defines-for-ptp-message-types'

Christian Eggers says:

====================
net: ptp: introduce common defines for PTP message types

This series introduces commen defines for PTP event messages. Driver
internal defines are removed and some uses of magic numbers are replaced
by the new defines.
====================

Link: https://lore.kernel.org/r/20201120084106.10046-1-ceggers@arri.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoptp: ptp_ines: use new PTP_MSGTYPE_* define(s)
Christian Eggers [Fri, 20 Nov 2020 08:41:06 +0000 (09:41 +0100)]
ptp: ptp_ines: use new PTP_MSGTYPE_* define(s)

Remove driver internal defines for this. Masking msgtype with 0xf is
already done within ptp_get_msgtype().

Signed-off-by: Christian Eggers <ceggers@arri.de>
Cc: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agodpaa2-eth: use new PTP_MSGTYPE_* define(s)
Christian Eggers [Fri, 20 Nov 2020 08:41:05 +0000 (09:41 +0100)]
dpaa2-eth: use new PTP_MSGTYPE_* define(s)

Remove usage of magic numbers.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Cc: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Cc: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ptp: introduce common defines for PTP message types
Christian Eggers [Fri, 20 Nov 2020 08:41:04 +0000 (09:41 +0100)]
net: ptp: introduce common defines for PTP message types

Using PTP wide defines will obsolete different driver internal defines
and uses of magic numbers.

Signed-off-by: Christian Eggers <ceggers@arri.de>
Cc: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agocompat: always include linux/compat.h from net/compat.h
Jakub Kicinski [Sat, 21 Nov 2020 21:48:44 +0000 (13:48 -0800)]
compat: always include linux/compat.h from net/compat.h

We're about to do reshuffling in networking headers and
eliminate some implicit includes. This results in:

In file included from ../net/ipv4/netfilter/arp_tables.c:26:
include/net/compat.h:60:40: error: unknown type name ‘compat_uptr_t’; did you mean ‘compat_ptr_ioctl’?
    struct sockaddr __user **save_addr, compat_uptr_t *ptr,
                                        ^~~~~~~~~~~~~
                                        compat_ptr_ioctl
include/net/compat.h:61:4: error: unknown type name ‘compat_size_t’; did you mean ‘compat_sigset_t’?
    compat_size_t *len);
    ^~~~~~~~~~~~~
    compat_sigset_t

Currently net/compat.h depends on linux/compat.h being included
first. After the upcoming changes this would break the 32bit build.

Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20201121214844.1488283-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agorxrpc: Ask the security class how much space to allow in a packet
David Howells [Wed, 16 Sep 2020 00:34:39 +0000 (01:34 +0100)]
rxrpc: Ask the security class how much space to allow in a packet

Ask the security class how much header and trailer space to allow for when
allocating a packet, given how much data is remaining.

This will allow the rxgk security class to stick both a trailer in as well
as a header as appropriate in the future.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agonet/tun: Call type change netdev notifiers
Martin Schiller [Wed, 18 Nov 2020 06:39:19 +0000 (07:39 +0100)]
net/tun: Call type change netdev notifiers

Call netdev notifiers before and after changing the device type.

Signed-off-by: Martin Schiller <ms@dev.tdt.de>
Link: https://lore.kernel.org/r/20201118063919.29485-1-ms@dev.tdt.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agorxrpc: rxkad: Don't use pskb_pull() to advance through the response packet
David Howells [Tue, 29 Sep 2020 20:48:50 +0000 (21:48 +0100)]
rxrpc: rxkad: Don't use pskb_pull() to advance through the response packet

In the rxkad security class, don't use pskb_pull() to advance through the
contents of the response packet.  There's no point, especially as the next
and last access to the skbuff still has to allow for the wire header in the
offset (which we didn't advance over).

Better to just add the displacement to the next offset.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Organise connection security to use a union
David Howells [Tue, 22 Sep 2020 12:36:17 +0000 (13:36 +0100)]
rxrpc: Organise connection security to use a union

Organise the security information in the rxrpc_connection struct to use a
union to allow for different data for different security classes.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Don't reserve security header in Tx DATA skbuff
David Howells [Thu, 17 Sep 2020 22:23:38 +0000 (23:23 +0100)]
rxrpc: Don't reserve security header in Tx DATA skbuff

Insert the security header into the skbuff representing a DATA packet to be
transmitted rather than using skb_reserve() when the packet is allocated.
This makes it easier to apply crypto that spans the security header and the
data, particularly in the upcoming RxGK class where we have a common
encrypt-and-checksum function that is used in a number of circumstances.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Merge prime_packet_security into init_connection_security
David Howells [Wed, 16 Sep 2020 00:38:15 +0000 (01:38 +0100)]
rxrpc: Merge prime_packet_security into init_connection_security

Merge the ->prime_packet_security() into the ->init_connection_security()
hook as they're always called together.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Fix example key name in a comment
David Howells [Thu, 15 Oct 2020 14:47:45 +0000 (15:47 +0100)]
rxrpc: Fix example key name in a comment

Fix an example of an rxrpc key name in a comment.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Ignore unknown tokens in key payload unless no known tokens
David Howells [Thu, 15 Oct 2020 14:59:36 +0000 (15:59 +0100)]
rxrpc: Ignore unknown tokens in key payload unless no known tokens

When parsing a payload for an rxrpc-type key, ignore any tokens that are
not of a known type and don't give an error for them - unless there are no
tokens of a known type.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Make the parsing of xdr payloads more coherent
David Howells [Sun, 27 Sep 2020 10:17:03 +0000 (11:17 +0100)]
rxrpc: Make the parsing of xdr payloads more coherent

Make the parsing of xdr-encoded payloads, as passed to add_key, more
coherent.  Shuttling back and forth between various variables was a bit
hard to follow.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Allow security classes to give more info on server keys
David Howells [Sun, 27 Sep 2020 10:13:04 +0000 (11:13 +0100)]
rxrpc: Allow security classes to give more info on server keys

Allow a security class to give more information on an rxrpc_s-type key when
it is viewed in /proc/keys.  This will allow the upcoming RxGK security
class to show the enctype name here.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Don't leak the service-side session key to userspace
David Howells [Sun, 27 Sep 2020 10:07:21 +0000 (11:07 +0100)]
rxrpc: Don't leak the service-side session key to userspace

Don't let someone reading a service-side rxrpc-type key get access to the
session key that was exchanged with the client.  The server application
will, at some point, need to be able to read the information in the ticket,
but this probably shouldn't include the key material.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Hand server key parsing off to the security class
David Howells [Wed, 16 Sep 2020 07:37:29 +0000 (08:37 +0100)]
rxrpc: Hand server key parsing off to the security class

Hand responsibility for parsing a server key off to the security class.  We
can determine which class from the description.  This is necessary as rxgk
server keys have different lookup requirements and different content
requirements (dependent on crypto type) to those of rxkad server keys.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Split the server key type (rxrpc_s) into its own file
David Howells [Wed, 16 Sep 2020 07:25:08 +0000 (08:25 +0100)]
rxrpc: Split the server key type (rxrpc_s) into its own file

Split the server private key type (rxrpc_s) out into its own file rather
than mingling it with the authentication/client key type (rxrpc) since they
don't really bear any relation.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Don't retain the server key in the connection
David Howells [Wed, 16 Sep 2020 07:00:44 +0000 (08:00 +0100)]
rxrpc: Don't retain the server key in the connection

Don't retain a pointer to the server key in the connection, but rather get
it on demand when the server has to deal with a response packet.

This is necessary to implement RxGK (GSSAPI-mediated transport class),
where we can't know which key we'll need until we've challenged the client
and got back the response.

This also means that we don't need to do a key search in the accept path in
softirq mode.

Also, whilst we're at it, allow the security class to ask for a kvno and
encoding-type variant of a server key as RxGK needs different keys for
different encoding types.  Keys of this type have an extra bit in the
description:

"<service-id>:<security-index>:<kvno>:<enctype>"

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Support keys with multiple authentication tokens
David Howells [Wed, 16 Sep 2020 07:19:12 +0000 (08:19 +0100)]
rxrpc: Support keys with multiple authentication tokens

rxrpc-type keys can have multiple tokens attached for different security
classes.  Currently, rxrpc always picks the first one, whether or not the
security class it indicates is supported.

Add preliminary support for choosing which security class will be used
(this will need to be directed from a higher layer) and go through the
tokens to find one that's supported.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: List the held token types in the key description in /proc/keys
David Howells [Tue, 8 Sep 2020 21:30:52 +0000 (22:30 +0100)]
rxrpc: List the held token types in the key description in /proc/keys

When viewing an rxrpc-type key through /proc/keys, display a list of held
token types.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agorxrpc: Remove the rxk5 security class as it's now defunct
David Howells [Thu, 3 Sep 2020 07:21:58 +0000 (08:21 +0100)]
rxrpc: Remove the rxk5 security class as it's now defunct

Remove the rxrpc rxk5 security class as it's now defunct and nothing uses
it anymore.

Signed-off-by: David Howells <dhowells@redhat.com>
3 years agokeys: Provide the original description to the key preparser
David Howells [Wed, 16 Sep 2020 10:12:03 +0000 (11:12 +0100)]
keys: Provide the original description to the key preparser

Provide the proposed description (add key) or the original description
(update/instantiate key) when preparsing a key so that the key type can
validate it against the data.

This is important for rxrpc server keys as we need to check that they have
the right amount of key material present - and it's better to do that when
the key is loaded rather than deep in trying to process a response packet.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
cc: keyrings@vger.kernel.org

3 years agoocteontx2-af: Add support for RSS hashing based on Transport protocol field
George Cherian [Fri, 20 Nov 2020 09:39:06 +0000 (15:09 +0530)]
octeontx2-af: Add support for RSS hashing based on Transport protocol field

Add support to choose RSS flow key algorithm with IPv4 transport protocol
field included in hashing input data. This will be enabled by default.
There-by enabling 3/5 tuple hash

Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: George Cherian <george.cherian@marvell.com>
Link: https://lore.kernel.org/r/20201120093906.2873616-1-george.cherian@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge tag 'linux-can-next-for-5.11-20201120' of git://git.kernel.org/pub/scm/linux...
Jakub Kicinski [Sat, 21 Nov 2020 22:58:11 +0000 (14:58 -0800)]
Merge tag 'linux-can-next-for-5.11-20201120' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2020-11-20

The first patch is by Yegor Yefremov and he improves the j1939 documentaton by
adding tables for the CAN identifier and its fields.

Then there are 8 patches by Oliver Hartkopp targeting the CAN driver
infrastructure and drivers. These add support for optional DLC element to the
Classical CAN frame structure. See patch ea7800565a12 ("can: add optional DLC
element to Classical CAN frame structure") for details. Oliver's last patch
adds len8_dlc support to several drivers. Stefan Mätje provides a patch to add
len8_dlc support to the esd_usb2 driver.

The next patch is by Oliver Hartkopp, too and adds support for modification of
Classical CAN DLCs to CAN GW sockets.

The next 3 patches target the nxp,flexcan DT bindings. One patch by my adds the
missing uint32 reference to the clock-frequency property. Joakim Zhang's
patches fix the fsl,clk-source property and add the IMX_SC_R_CAN() macro to the
imx firmware header file, which will be used in the flexcan driver later.

Another patch by Joakim Zhang prepares the flexcan driver for SCU based
stop-mode, by giving the existing, GPR based stop-mode, a _GPR postfix.

The next 5 patches are by me, target the flexcan driver, and clean up the
.ndo_open and .ndo_stop callbacks. These patches try to fix a sporadically
hanging flexcan_close() during simultanious ifdown, sending of CAN messages and
probably open CAN bus. I was never able to reproduce, but these seem to fix the
problem at the reporting user. As these changes are rather big, I'd like to
mainline them via net-next/master.

The next patches are by Jimmy Assarsson and Christer Beskow, they add support
for new USB devices to the existing kvaser_usb driver.

The last patch is by Kaixu Xia and simplifies the return in the
mcp251xfd_chip_softreset() function in the mcp251xfd driver.

* tag 'linux-can-next-for-5.11-20201120' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next: (25 commits)
  can: mcp251xfd: remove useless code in mcp251xfd_chip_softreset
  can: kvaser_usb: Add new Kvaser hydra devices
  can: kvaser_usb: kvaser_usb_hydra: Add support for new device variant
  can: kvaser_usb: Add new Kvaser Leaf v2 devices
  can: kvaser_usb: Add USB_{LEAF,HYDRA}_PRODUCT_ID_END defines
  can: flexcan: flexcan_close(): change order if commands to properly shut down the controller
  can: flexcan: flexcan_open(): completely initialize controller before requesting IRQ
  can: flexcan: flexcan_rx_offload_setup(): factor out mailbox and rx-offload setup into separate function
  can: flexcan: move enabling/disabling of interrupts from flexcan_chip_{start,stop}() to callers
  can: flexcan: factor out enabling and disabling of interrupts into separate function
  can: flexcan: rename macro FLEXCAN_QUIRK_SETUP_STOP_MODE -> FLEXCAN_QUIRK_SETUP_STOP_MODE_GPR
  dt-bindings: firmware: add IMX_SC_R_CAN(x) macro for CAN
  dt-bindings: can: fsl,flexcan: fix fsl,clk-source property
  dt-bindings: can: fsl,flexcan: add uint32 reference to clock-frequency property
  can: gw: support modification of Classical CAN DLCs
  can: drivers: add len8_dlc support for esd_usb2 CAN adapter
  can: drivers: add len8_dlc support for various CAN adapters
  can: drivers: introduce helpers to access Classical CAN DLC values
  can: update documentation for DLC usage in Classical CAN
  can: rename CAN FD related can_len2dlc and can_dlc2len helpers
  ...
====================

Link: https://lore.kernel.org/r/20201120133318.3428231-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: bridge: switch to net core statistics counters handling
Heiner Kallweit [Fri, 20 Nov 2020 11:22:23 +0000 (12:22 +0100)]
net: bridge: switch to net core statistics counters handling

Use netdev->tstats instead of a member of net_bridge for storing
a pointer to the per-cpu counters. This allows us to use core
functionality for statistics handling.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://lore.kernel.org/r/9bad2be2-fd84-7c6e-912f-cee433787018@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-hns3-misc-updates-for-next'
Jakub Kicinski [Sat, 21 Nov 2020 22:33:49 +0000 (14:33 -0800)]
Merge branch 'net-hns3-misc-updates-for-next'

Huazhong Tan says:

====================
net: hns3: misc updates for -next

This series includes some misc updates for the HNS3 ethernet driver.

 #1 adds support for 1280 queues
 #2 adds mapping for BAR45 which is needed by RoCE client.
 #3 extend the interrupt resources.
 #4&#5 add support to query firmware's calculated shaping parameters.
====================

Link: https://lore.kernel.org/r/1605863783-36995-1-git-send-email-tanhuazhong@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: adds debugfs to dump more info of shaping parameters
Yonglong Liu [Fri, 20 Nov 2020 09:16:23 +0000 (17:16 +0800)]
net: hns3: adds debugfs to dump more info of shaping parameters

Adds debugfs to dump new shaping parameters: rate and flag.

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: add support to utilize the firmware calculated shaping parameters
Yonglong Liu [Fri, 20 Nov 2020 09:16:22 +0000 (17:16 +0800)]
net: hns3: add support to utilize the firmware calculated shaping parameters

Since the calculation of the driver is fixed, if the number of
queue or clock changed, the calculated result may be inaccurate.

So for compatible and maintainable, add a new flag to tell the
firmware to calculate the shaping parameters with the specified
rate.

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: add support for pf querying new interrupt resources
Yufeng Mo [Fri, 20 Nov 2020 09:16:21 +0000 (17:16 +0800)]
net: hns3: add support for pf querying new interrupt resources

For HNAE3_DEVICE_VERSION_V3, a maximum of 1281 interrupt
resources are supported. To utilize these new resources,
extend the corresponding field or variable to 16bit type,
and remove the restriction of NIC client that only use a
maximum of 65 interrupt vectors. In addition, the I/O address
of the extended interrupt resources are different, so an extra
handler is needed.

Currently, the total number of interrupts is the sum of RoCE's
number and RoCE's offset (RoCE is in front of NIC), since
the number of both NIC and RoCE are same. For readability,
rewrite the corresponding field of the command, rename the
RoCE's offset field as the number of NIC interrupts, then
the total number of interrupts is sum of the number of RoCE
and NIC, and replace vport->back with hdev in
hclge_init_roce_base_info() for simplifying the code.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: add support for mapping device memory
Huazhong Tan [Fri, 20 Nov 2020 09:16:20 +0000 (17:16 +0800)]
net: hns3: add support for mapping device memory

For device who has device memory accessed through the PCI BAR4,
IO descriptor push of NIC and direct WQE(Work Queue Element) of
RoCE will use this device memory, so add support for mapping
this device memory, and add this info to the RoCE client whose
new feature needs.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: add support for 1280 queues
Yonglong Liu [Fri, 20 Nov 2020 09:16:19 +0000 (17:16 +0800)]
net: hns3: add support for 1280 queues

For DEVICE_VERSION_V1/2, there are total 1024 queues and
queue sets. For DEVICE_VERSION_V3, it increases to 1280,
and can be assigned to one pf, so remove the limitation
of 1024.

To keep compatible with DEVICE_VERSION_V1/2 and old driver
version, the queue number is split into two part:
tqp_num(range 0~1023) and ext_tqp_num(range 1024~1279).

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'ibmvnic-performance-improvements-and-other-updates'
Jakub Kicinski [Sat, 21 Nov 2020 03:51:44 +0000 (19:51 -0800)]
Merge branch 'ibmvnic-performance-improvements-and-other-updates'

Thomas Falcon says:

====================
ibmvnic: Performance improvements and other updates

The first three patches utilize a hypervisor call allowing multiple
TX and RX buffer replenishment descriptors to be sent in one operation,
which significantly reduces hypervisor call overhead. The xmit_more
and Byte Queue Limit API's are leveraged to provide this support
for TX descriptors.

The subsequent two patches remove superfluous code and members in
TX completion handling function and TX buffer structure, respectively,
and remove unused routines.

Finally, four patches which ensure that device queue memory is
cache-line aligned, resolving slowdowns observed in PCI traces,
as well as optimize the driver's NAPI polling function and
to RX buffer replenishment are provided by Dwip Banerjee.

This series provides significant performance improvements, allowing
the driver to fully utilize 100Gb NIC's.
====================

Link: https://lore.kernel.org/r/1605748345-32062-1-git-send-email-tlfalcon@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Do not replenish RX buffers after every polling loop
Dwip N. Banerjee [Thu, 19 Nov 2020 01:12:25 +0000 (19:12 -0600)]
ibmvnic: Do not replenish RX buffers after every polling loop

Reduce the amount of time spent replenishing RX buffers by only doing
so once available buffers has fallen under a certain threshold, in this
case half of the total number of buffers, or if the polling loop exits
before the packets processed is less than its budget. Non-exhaustion of
NAPI budget implies lower incoming packet pressure, allowing the leeway
to refill the buffers in preparation for any impending burst.

Signed-off-by: Dwip N. Banerjee <dnbanerg@us.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Use netdev_alloc_skb instead of alloc_skb to replenish RX buffers
Dwip N. Banerjee [Thu, 19 Nov 2020 01:12:24 +0000 (19:12 -0600)]
ibmvnic: Use netdev_alloc_skb instead of alloc_skb to replenish RX buffers

Take advantage of the additional optimizations in netdev_alloc_skb when
allocating socket buffers to be used for packet reception.

Signed-off-by: Dwip N. Banerjee <dnbanerg@us.ibm.com>
Acked-by: Lijun Pan <ljp@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Correctly re-enable interrupts in NAPI polling routine
Dwip N. Banerjee [Thu, 19 Nov 2020 01:12:23 +0000 (19:12 -0600)]
ibmvnic: Correctly re-enable interrupts in NAPI polling routine

If the current NAPI polling loop exits without completing it's
budget, only re-enable interrupts if there are no entries remaining
in the queue and napi_complete_done is successful. If there are entries
remaining on the queue that were missed, restart the polling loop.

Signed-off-by: Dwip N. Banerjee <dnbanerg@us.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Ensure that device queue memory is cache-line aligned
Dwip N. Banerjee [Thu, 19 Nov 2020 01:12:22 +0000 (19:12 -0600)]
ibmvnic: Ensure that device queue memory is cache-line aligned

PCI bus slowdowns were observed on IBM VNIC devices as a result
of partial cache line writes and non-cache aligned full cache line writes.
Ensure that packet data buffers are cache-line aligned to avoid these
slowdowns.

Signed-off-by: Dwip N. Banerjee <dnbanerg@us.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Remove send_subcrq function
Thomas Falcon [Thu, 19 Nov 2020 01:12:21 +0000 (19:12 -0600)]
ibmvnic: Remove send_subcrq function

It is not longer used, so remove it.

Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Acked-by: Lijun Pan <ljp@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Clean up TX code and TX buffer data structure
Thomas Falcon [Thu, 19 Nov 2020 01:12:20 +0000 (19:12 -0600)]
ibmvnic: Clean up TX code and TX buffer data structure

Remove unused and superfluous code and members in
existing TX implementation and data structures.

Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Introduce xmit_more support using batched subCRQ hcalls
Thomas Falcon [Thu, 19 Nov 2020 01:12:19 +0000 (19:12 -0600)]
ibmvnic: Introduce xmit_more support using batched subCRQ hcalls

Include support for the xmit_more feature utilizing the
H_SEND_SUB_CRQ_INDIRECT hypervisor call which allows the sending
of multiple subordinate Command Response Queue descriptors in one
hypervisor call via a DMA-mapped buffer. This update reduces hypervisor
calls and thus hypervisor call overhead per TX descriptor.

Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: Introduce batched RX buffer descriptor transmission
Thomas Falcon [Thu, 19 Nov 2020 01:12:18 +0000 (19:12 -0600)]
ibmvnic: Introduce batched RX buffer descriptor transmission

Utilize the H_SEND_SUB_CRQ_INDIRECT hypervisor call to send
multiple RX buffer descriptors to the device in one hypervisor
call operation. This change will reduce the number of hypervisor
calls and thus hypervisor call overhead needed to transmit
RX buffer descriptors to the device.

Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>