platform/kernel/linux-rpi.git
4 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 10 Mar 2020 23:20:03 +0000 (16:20 -0700)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2020-03-10

This series contains updates to ice and iavf drivers.

Cleaned up unnecessary parenthesis, which was pointed out by Sergei
Shtylyov.

Mitch updates the iavf and ice drivers to expand the limitation on the
number of queues that the driver can support to account for the newer
800-series capabilities.

Brett cleans up the error messages for both SR-IOV and non SR-IOV use
cases.  Fixed the logic when the ice driver is removed and a bare-metal
VF is passing traffic, which was causing a transmit hang on the VF.
Updated the ice driver to display "Link detected" field via ethtool,
when the driver is in safe mode.  Updated ice driver to properly set
VLAN pruning when transmit anti-spoof is off.

Avinash fixed a corner case in DCB, when switching from IEEE to CEE
mode, the DCBX mode does not get properly updated.

Dave updates the logic when switching from software DCB to firmware DCB
to renegotiate DCBX to ensure the firmware agent has up to date
information about the DCB settings of the link partner.

Lukasz increases the PF's mailbox receive queue size to the maximum to
prevent potential bottleneck or slow down occurring from the PF's
mailbox receive queue being full.

Bruce updates the ice driver to use strscpy() instead of strlcpy().
Cleaned up variable names that were not very descriptive with names that
had more meaning.

Anirudh replaces the use of ENOTSUPP with EOPNOTSUPP in the ice driver.

Jake fixed up a function header comment to properly reflect the variable
size and use.

v2: Dropped patch 5 of the original series, where Tony added tunnel
    offload support.  Based on community feedback, the patch needed
    changes, so giving Tony additional time to work on those changes and
    not hold up the remaining changes in the series.
====================

Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: mt7530: fix macro MIRROR_PORT
DENG Qingfang [Tue, 10 Mar 2020 18:20:50 +0000 (02:20 +0800)]
net: dsa: mt7530: fix macro MIRROR_PORT

The inner pair of parentheses should be around the variable x

Fixes: 37feab6076aa ("net: dsa: mt7530: add support for port mirroring")
Signed-off-by: DENG Qingfang <dqfext@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: microchip: use delayed_work instead of timer + work
George McCollister [Tue, 10 Mar 2020 17:58:59 +0000 (12:58 -0500)]
net: dsa: microchip: use delayed_work instead of timer + work

Simplify ksz_common.c by using delayed_work instead of a combination of
timer and work.

Signed-off-by: George McCollister <george.mccollister@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'flow_offload-follow-ups-to-HW-stats-type-patchset'
David S. Miller [Tue, 10 Mar 2020 23:04:19 +0000 (16:04 -0700)]
Merge branch 'flow_offload-follow-ups-to-HW-stats-type-patchset'

Jiri Pirko says:

====================
flow_offload: follow-ups to HW stats type patchset

This patchset includes couple of patches in reaction to the discussions
to the original HW stats patchset. The first patch is a fix,
the other two patches are basically cosmetics.
====================

Acked-by: Edward Cree <ecree@solarflare.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoflow_offload: restrict driver to pass one allowed bit to flow_action_hw_stats_types_c...
Jiri Pirko [Tue, 10 Mar 2020 15:49:09 +0000 (16:49 +0100)]
flow_offload: restrict driver to pass one allowed bit to flow_action_hw_stats_types_check()

The intention of this helper was to allow driver to specify one type
that it supports, so not only "any" value would pass. So make the API
more strict and allow driver to pass only 1 bit that is going
to be checked.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoflow_offload: turn hw_stats_type into dedicated enum
Jiri Pirko [Tue, 10 Mar 2020 15:49:08 +0000 (16:49 +0100)]
flow_offload: turn hw_stats_type into dedicated enum

Put the values into enum and add an enum to define the bits.

Suggested-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoflow_offload: fix allowed types check
Jiri Pirko [Tue, 10 Mar 2020 15:49:07 +0000 (16:49 +0100)]
flow_offload: fix allowed types check

Change the check to see if the passed allowed type bit is enabled.

Fixes: 319a1d19471e ("flow_offload: check for basic action hw stats type")
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'enetc-Support-extended-BD-rings-at-runtime'
David S. Miller [Tue, 10 Mar 2020 22:48:54 +0000 (15:48 -0700)]
Merge branch 'enetc-Support-extended-BD-rings-at-runtime'

Claudiu Manoil says:

====================
enetc: Support extended BD rings at runtime

First two patches are just misc code cleanup.
The 3rd patch prepares the Rx BD processing code to be extended
to processing both normal and extended BDs.
The last one adds extended Rx BD support for timestamping
without the need of a static config. Finally, the config option
FSL_ENETC_HW_TIMESTAMPING can be dropped.
Care was taken not to impact non-timestamping usecases.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoenetc: Add dynamic allocation of extended Rx BD rings
Claudiu Manoil [Tue, 10 Mar 2020 12:51:24 +0000 (14:51 +0200)]
enetc: Add dynamic allocation of extended Rx BD rings

Hardware timestamping support (PTP) on Rx requires extended
buffer descriptors, double the size of normal Rx descriptors.
On the current controller revision only the timestamping offload
requires extended Rx descriptors.
Since Rx timestamping can be turned on/off at runtime, make Rx ring
allocation configurable at runtime too. As a result, the static
config option FSL_ENETC_HW_TIMESTAMPING can be dropped and the
extended descriptors can be used only when Rx timestamping gets
activated.
The extension has the same size as the base descriptor, making
the descriptor iterators easy to update for the extended case.

Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoenetc: Clean up Rx BD iteration
Claudiu Manoil [Tue, 10 Mar 2020 12:51:23 +0000 (14:51 +0200)]
enetc: Clean up Rx BD iteration

Improve maintainability of the code iterating the Rx buffer
descriptors to prepare it to support iterating extended Rx BD
descriptors as well.
Don't increment by one the h/w descriptor pointers explicitly,
provide an iterator that takes care of the h/w details.

Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoenetc: Clean up of ehtool stats len
Claudiu Manoil [Tue, 10 Mar 2020 12:51:22 +0000 (14:51 +0200)]
enetc: Clean up of ehtool stats len

Refactor the stats len computation code to make it easier
to add new stats counters.

Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoenetc: Drop redundant device node check
Claudiu Manoil [Tue, 10 Mar 2020 12:51:21 +0000 (14:51 +0200)]
enetc: Drop redundant device node check

The existence of the DT port node is the first thing checked
at probe time, and probing won't reach this point if the node
is missing.

Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agopktgen: Allow on loopback device
Lukas Wunner [Tue, 10 Mar 2020 10:49:46 +0000 (11:49 +0100)]
pktgen: Allow on loopback device

When pktgen is used to measure the performance of dev_queue_xmit()
packet handling in the core, it is preferable to not hand down
packets to a low-level Ethernet driver as it would distort the
measurements.

Allow using pktgen on the loopback device, thus constraining
measurements to core code.

Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoflow_offload: use flow_action_for_each in flow_action_mixed_hw_stats_types_check()
Jiri Pirko [Tue, 10 Mar 2020 10:11:57 +0000 (11:11 +0100)]
flow_offload: use flow_action_for_each in flow_action_mixed_hw_stats_types_check()

Instead of manually iterating over entries, use flow_action_for_each
helper. Move the helper and wrap it to fit to 80 cols on the way.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoice: fix incorrect size description of ice_get_nvm_version
Jacob Keller [Thu, 27 Feb 2020 18:15:05 +0000 (10:15 -0800)]
ice: fix incorrect size description of ice_get_nvm_version

The function comment for ice_get_nvm_version indicated that the ver_hi
and ver_lo values were 16 bits. In fact, they are only uint8_t values,
meaning that they have a maximum size of 8 bits. Fix the comment to
match the correct size.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: use variable name more descriptive than type
Bruce Allan [Thu, 27 Feb 2020 18:15:04 +0000 (10:15 -0800)]
ice: use variable name more descriptive than type

The variable name 'type' is not very descriptive. Replace instances of
those with a variable name that is more descriptive or replace it if not
needed.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Use EOPNOTSUPP instead of ENOTSUPP
Anirudh Venkataramanan [Thu, 27 Feb 2020 18:15:03 +0000 (10:15 -0800)]
ice: Use EOPNOTSUPP instead of ENOTSUPP

Using ENOTSUPP almost always results in some bizarre error message to
be printed in userspace. This is likely because ENOTSUPP was defined for
the NFS protocol (as per a comment in include/linux/errno.h). Use
EOPNOTSUPP instead.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Fix format specifier
Tony Nguyen [Thu, 27 Feb 2020 18:15:02 +0000 (10:15 -0800)]
ice: Fix format specifier

Commit ed5a3f664c55 ("ice: Removing hung_queue variable to use txqueue
function parameter") began utilizing the txqueue variable over the
hung_queue variable. hung_queue was an int where txqueue is an unsigned
int. Update the format specifiers to reflect the new type.

Fixes: ed5a3f664c55 ("ice: Removing hung_queue variable to use txqueue function parameter")
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: fix use of deprecated strlcpy()
Bruce Allan [Thu, 27 Feb 2020 18:15:01 +0000 (10:15 -0800)]
ice: fix use of deprecated strlcpy()

checkpatch complains "CHECK:DEPRECATED_API: Deprecated use of 'strlcpy',
prefer 'stracpy or strscpy' instead"; use strscpy.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Increase mailbox receive queue length to maximum
Lukasz Czapnik [Thu, 27 Feb 2020 18:15:00 +0000 (10:15 -0800)]
ice: Increase mailbox receive queue length to maximum

Currently the PF's mailbox receive queue is only 512 entries. This fine,
but considering that all VF's mailbox send queues funnel into the PF's
single mailbox receive queue, let's increase it to the maximum size. This
will help prevent any possible bottleneck/slowdown occurring from the PF's
mailbox receive queue being full.

Signed-off-by: Lukasz Czapnik <lukasz.czapnik@intel.com>
Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Correct setting VLAN pruning
Brett Creeley [Thu, 27 Feb 2020 18:14:59 +0000 (10:14 -0800)]
ice: Correct setting VLAN pruning

VLAN pruning is not always being set correctly due to a previous change
that set Tx antispoof off. ice_vsi_is_vlan_pruning_ena() currently checks
for both Tx antispoof and Rx pruning. The expectation for this function is
to only check Rx pruning so fix the check.

Fixes: cd6d6b83316a ("ice: Fix VF spoofchk")
Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: renegotiate link after FW DCB on
Dave Ertman [Thu, 27 Feb 2020 18:14:58 +0000 (10:14 -0800)]
ice: renegotiate link after FW DCB on

When switching from SW DCB to FW DCB it is necessary
to renegotiate DCBx so that the FW agent can have up
to date information about the DCB settings of the link
partner.

Perform an autoneg restart on the link when activating
FW DCB.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Fix corner case when switching from IEEE to CEE
Avinash JD [Thu, 27 Feb 2020 18:14:57 +0000 (10:14 -0800)]
ice: Fix corner case when switching from IEEE to CEE

While testing DCB for a corner case in which mode is switched from IEEE to
CEE and pfc_ena bitmask unchanged then DCBX mode doesn't get updated.

This is happening because the function ice_dcb_get_mode() is called
in a "no change detected block" instead of "change detected block".

Signed-off-by: Avinash JD <avinash.dayanand@intel.com>
Signed-off-by: Scott Register <scottx.register@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Display Link detected via Ethtool in safe mode
Brett Creeley [Thu, 27 Feb 2020 18:14:56 +0000 (10:14 -0800)]
ice: Display Link detected via Ethtool in safe mode

Currently the "Link detected" field is not shown when the device goes
into safe mode. This is because the safe mode Ethtool ops does not set the
get_link function. Fix this by setting the safe mode Ethtool op get_link
function.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Fix removing driver while bare-metal VFs pass traffic
Brett Creeley [Thu, 27 Feb 2020 18:14:55 +0000 (10:14 -0800)]
ice: Fix removing driver while bare-metal VFs pass traffic

Currently, if there are bare-metal VFs passing traffic and the ice
driver is removed, there is a possibility of VFs triggering a Tx timeout
right before iavf_remove(). This is causing iavf_close() to not be
called because there is a check in the beginning of iavf_remove() that
bails out early if (adapter->state < IAVF_DOWN_PENDING). This makes it
so some resources do not get cleaned up. Specifically, free_irq()
is never called for data interrupts, which results in the following line
of code to trigger:

pci_disable_msix()
free_msi_irqs()
...
BUG_ON(irq_has_action(entry->irq + i));
...

To prevent the Tx timeout from occurring on the VF during driver unload
for ice and the iavf there are a few changes that are needed.

[1] Don't disable all active VF Tx/Rx queues prior to calling
pci_disable_sriov.

[2] Call ice_free_vfs() before disabling the service task.

[3] Disable VF resets when the ice driver is being unloaded by setting
the pf->state flag __ICE_VF_RESETS_DISABLED.

Changing [1] and [2] allow each VF driver's remove flow to successfully
send VIRTCHNL requests, which includes queue disable. This prevents
unexpected Tx timeouts because the PF driver is no longer forcefully
disabling queues.

Due to [1] and [2] there is a possibility that the PF driver will get a
VFLR or reset request over VIRTCHNL from a VF during PF driver unload.
Prevent that by doing [3].

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Improve clarity of prints and variables
Brett Creeley [Thu, 27 Feb 2020 18:14:53 +0000 (10:14 -0800)]
ice: Improve clarity of prints and variables

Currently when the device runs out of MSI-X interrupts a cryptic and
unhelpful message is printed. This will cause confusion when hitting this
case. Fix this by clearing up the error message for both SR-IOV and non
SR-IOV use cases.

Also, make a few minor changes to increase clarity of variables.
1. Change per VF MSI-X and queue pair variables in the PF structure.
2. Use ICE_NONQ_VECS_VF when determining pf->num_msix_per_vf instead of
the magic number "1". This vector is reserved for the OICR.

All of the resource tracking functions were moved to avoid adding
any forward declaration function prototypes.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: allow bigger VFs
Mitch Williams [Thu, 27 Feb 2020 18:14:52 +0000 (10:14 -0800)]
ice: allow bigger VFs

Unlike the XL710 series, 800-series hardware can allocate more than 4
MSI-X vectors per VF. This patch enables that functionality. We
dynamically allocate vectors and queues depending on how many VFs are
enabled. Allocating the maximum number of VFs replicates XL710
behavior with 4 queues and 4 vectors. But allocating a smaller number
of VFs will give you 16 queues and 16 vectors.

Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoiavf: Enable support for up to 16 queues
Mitch Williams [Thu, 27 Feb 2020 18:14:51 +0000 (10:14 -0800)]
iavf: Enable support for up to 16 queues

Previous devices could only allocate 4 MSI-X vectors per VF so there was a
limitation of 4 queues. 800-series hardware can allocate more than 4 MSI-X
vectors, so expand the limitation on the number of queues that the driver
can support to account for these capabilities.

Fix ethtool channel operations to accommodate this change and adjust the
reporting of max number of queues to what is given to us by the PF. Since
we're not requesting queues above this value, just trigger reset to
activate the queues, which we already own.

Finally, fix a test condition that would display an incorrect error
message.

Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Cleanup unneeded parenthesis
Jeff Kirsher [Fri, 21 Feb 2020 22:15:27 +0000 (14:15 -0800)]
ice: Cleanup unneeded parenthesis

Sergei Shtylyov pointed out that two instances of parenthesis are not
needed, so remove them.

Suggested-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
4 years agoMerge branch 'net-Add-support-for-Synopsys-DesignWare-XPCS'
David S. Miller [Tue, 10 Mar 2020 03:13:16 +0000 (20:13 -0700)]
Merge branch 'net-Add-support-for-Synopsys-DesignWare-XPCS'

Jose Abreu says:

====================
net: Add support for Synopsys DesignWare XPCS

This adds support for Synopsys DesignWare XPCS in net subsystem and
integrates it into stmmac.

At 1/8, we start by removing the limitation of stmmac selftests that needed
a PHY to pass all the tests.

Then at 2/8 we use some helpers in stmmac so that some code can be
simplified.

At 3/8, we fallback to dev_fwnode() so that PCI based setups wich may
not have CONFIG_OF can still use FW node.

At 4/8, we adapt stmmac to the new PHYLINK changes as suggested by Russell
King.

We proceed by doing changes in PHYLINK in order to support XPCS: At 5/8 we
add some missing speeds that USXGMII supports and at 6/8 we check if
Autoneg is supported after initial parameters are validated.

Support for XPCS is finally introduced at 7/8, along with the usage of it
in stmmac driver at 8/8.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: Integrate it with DesignWare XPCS
Jose Abreu [Mon, 9 Mar 2020 08:36:27 +0000 (09:36 +0100)]
net: stmmac: Integrate it with DesignWare XPCS

Adds all the necessary logic so that stmmac can be used with Synopsys
DesignWare XPCS.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: Add Synopsys DesignWare XPCS MDIO module
Jose Abreu [Mon, 9 Mar 2020 08:36:26 +0000 (09:36 +0100)]
net: phy: Add Synopsys DesignWare XPCS MDIO module

Synopsys DesignWare XPCS is an MMD that can manage link status,
auto-negotiation, link training, ...

In this commit we add basic support for XPCS using USXGMII interface and
Clause 73 Auto-negotiation.

This is highly tied with PHYLINK and can't be used without it. A given
ethernet driver can use the provided callbacks to add the support for
XPCS.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phylink: Test if MAC/PCS support Autoneg
Jose Abreu [Mon, 9 Mar 2020 08:36:25 +0000 (09:36 +0100)]
net: phylink: Test if MAC/PCS support Autoneg

We may have cases where MAC or PCS do not support Autoneg. Check if it
is supported after validate callback is called.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phylink: Add missing Backplane speeds
Jose Abreu [Mon, 9 Mar 2020 08:36:24 +0000 (09:36 +0100)]
net: phylink: Add missing Backplane speeds

USXGMII also supports these missing backplane speeds.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: Use resolved link config in mac_link_up()
Jose Abreu [Mon, 9 Mar 2020 08:36:23 +0000 (09:36 +0100)]
net: stmmac: Use resolved link config in mac_link_up()

Convert the stmmac ethernet driver to use the finalised link parameters
in mac_link_up() rather than the parameters in mac_config().

Suggested-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: Fallback to dev_fwnode() if needed
Jose Abreu [Mon, 9 Mar 2020 08:36:22 +0000 (09:36 +0100)]
net: stmmac: Fallback to dev_fwnode() if needed

When CONFIG_OF is not enabled, of_fwnode_handle() will return NULL, even
though we can have a FW handle from a given device.

Fallback to dev_fwnode() helper if needed.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: Switch to linkmode_and()/linkmode_andnot()
Jose Abreu [Mon, 9 Mar 2020 08:36:21 +0000 (09:36 +0100)]
net: stmmac: Switch to linkmode_and()/linkmode_andnot()

Use the linkmode_and()/linkmode_andnot() helpers to simplify the code.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: selftests: Do not fail if PHY is not attached
Jose Abreu [Mon, 9 Mar 2020 08:36:20 +0000 (09:36 +0100)]
net: stmmac: selftests: Do not fail if PHY is not attached

If a PHY is not attached, we can still run the tests with MAC loopback
mode. Return -EOPNOTSUPP error code in PHY loopback test so that global
status is not a failure.

Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mlx5-updates-2020-03-09' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Tue, 10 Mar 2020 02:40:42 +0000 (19:40 -0700)]
Merge tag 'mlx5-updates-2020-03-09' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2020-03-09

This series provides updates to mlx5 driver:

1) Use vport metadata matching only when mandatory
2) Introduce root flow table and ethtool steering for uplink representors
3) Expose port speed via FW when link modes are not available
3) Misc cleanups
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hns3-next'
David S. Miller [Tue, 10 Mar 2020 02:36:13 +0000 (19:36 -0700)]
Merge branch 'hns3-next'

Huazhong Tan says:

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

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

[patch 1] fixes some mixed type operation warning.
[patch 2] renames a macro to make it more readable.
[patch 3 & 4]  removes some unnecessary code.
[patch 5] adds check before assert VF reset to prevent some unsuitable
error log.
[patch 6 - 9] some modifications related to printing.

Change log:
V1->V2: fixes a wrong print format in [patch 1] suggested by Jian Shen.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: delete unnecessary logs after kzalloc fails
Yufeng Mo [Sat, 7 Mar 2020 03:42:50 +0000 (11:42 +0800)]
net: hns3: delete unnecessary logs after kzalloc fails

Since kernel already has logs after kzalloc fails,
it's unnecessary to print duplicate logs. So this
patch deletes these logs.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: synchronize some print relating to reset issue
Huazhong Tan [Sat, 7 Mar 2020 03:42:49 +0000 (11:42 +0800)]
net: hns3: synchronize some print relating to reset issue

This patch modifies some printing relating to reset issue.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: print out command code when dump fails in debugfs
Yufeng Mo [Sat, 7 Mar 2020 03:42:48 +0000 (11:42 +0800)]
net: hns3: print out command code when dump fails in debugfs

This patch adds a local variable to save the command code in
some dump cases which need to modify the command code, then
the failing command code can be print out for debugging.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: print out status register when VF receives unknown source interrupt
Huazhong Tan [Sat, 7 Mar 2020 03:42:47 +0000 (11:42 +0800)]
net: hns3: print out status register when VF receives unknown source interrupt

When received an unknown vector 0 interrupt, there is not a
helpful information for user to realize that now. So this patch
prints out the value of the status register for this case, and
uses dev_info() instead of dev_dbg() in hclge_check_event_cause().

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add a check before PF inform VF to reset
Yonglong Liu [Sat, 7 Mar 2020 03:42:46 +0000 (11:42 +0800)]
net: hns3: add a check before PF inform VF to reset

When setting VF's MAC from PF, if the VF driver not loaded, the
firmware will return error to PF.

So PF should check whether VF is alive before sending message to
VF when setting VF's MAC.

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: delete some reduandant code
Guojia Liao [Sat, 7 Mar 2020 03:42:45 +0000 (11:42 +0800)]
net: hns3: delete some reduandant code

In hclge_add_mc_addr_common() and hclge_rm_mc_addr_common(),
variable req had been set as "0" by memset(), so it's unnecessary
to set .entry_type field as "0" with hnae3_set_bit() again.

Signed-off-by: Guojia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: remove an unnecessary resetting check in hclge_handle_hw_ras_error()
Yufeng Mo [Sat, 7 Mar 2020 03:42:44 +0000 (11:42 +0800)]
net: hns3: remove an unnecessary resetting check in hclge_handle_hw_ras_error()

In hclge_handle_hw_ras_error(), it is unnecessary to check
HCLGE_STATE_RST_HANDLING flag, because the reset priority
has been ensured by the process itself.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: rename macro HCLGE_MAX_NCL_CONFIG_LENGTH
Yufeng Mo [Sat, 7 Mar 2020 03:42:43 +0000 (11:42 +0800)]
net: hns3: rename macro HCLGE_MAX_NCL_CONFIG_LENGTH

The name of macro HCLGE_MAX_NCL_CONFIG_LENGTH is inaccurate,
this patch renames it to HCLGE_NCL_CONFIG_LENGTH_IN_EACH_CMD.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: fix some mixed type assignment
Guojia Liao [Sat, 7 Mar 2020 03:42:42 +0000 (11:42 +0800)]
net: hns3: fix some mixed type assignment

This patch cleans up some incorrect type in assignment reported by sparse
and compiler.
The warning as below:
- warning : restricted __le16 degrades to integer
- warning : cast from restricted __le32
- warning : cast from restricted __be32
- warning : cast from restricted __be16
and "mixed operation".

Signed-off-by: Guojia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'ionic-updates'
David S. Miller [Tue, 10 Mar 2020 02:34:04 +0000 (19:34 -0700)]
Merge branch 'ionic-updates'

Shannon Nelson says:

====================
ionic updates

This is a set of small updates for the Pensando ionic driver, some
from internal work, some a result of mailing list discussions.

v4 - don't register mgmt device netdev
v3 - changed __attribute__(packed)) to __packed
v2 - removed print from ionic_init_module()
====================

Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: drop ethtool driver version
Shannon Nelson [Sat, 7 Mar 2020 01:04:08 +0000 (17:04 -0800)]
ionic: drop ethtool driver version

Use the default kernel version in ethtool drv_info output
and drop the module version.

Cc: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: add support for device id 0x1004
Shannon Nelson [Sat, 7 Mar 2020 01:04:07 +0000 (17:04 -0800)]
ionic: add support for device id 0x1004

Add support for the management port device id.  This is to
capture the device and set it up for devlink use, but not set
it up for network operations.  We still use a netdev object
in order to use the napi infrasucture for processing adminq
and notifyq messages, we just don't register the netdev.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: print pci bus lane info
Shannon Nelson [Sat, 7 Mar 2020 01:04:06 +0000 (17:04 -0800)]
ionic: print pci bus lane info

Print the PCI bus lane information so people can keep an
eye out for poor configurations that might not support
the advertised speed.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: support ethtool rxhash disable
Shannon Nelson [Sat, 7 Mar 2020 01:04:05 +0000 (17:04 -0800)]
ionic: support ethtool rxhash disable

We can disable rxhashing by setting rss_types to 0.  The user
can toggle this with "ethtool -K <ethX> rxhash off|on",
which calls into the .ndo_set_features callback with the
NETIF_F_RXHASH feature bit set or cleared.  This patch adds
a check for that bit and updates the FW if necessary.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: clean up bitflag usage
Shannon Nelson [Sat, 7 Mar 2020 01:04:04 +0000 (17:04 -0800)]
ionic: clean up bitflag usage

Remove the unused flags field and and fix the bitflag names
to include the _F_ flag hint.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: improve irq numa locality
Shannon Nelson [Sat, 7 Mar 2020 01:04:03 +0000 (17:04 -0800)]
ionic: improve irq numa locality

Spreading the interrupts across the CPU cores is good for load
balancing, but not necessarily as good when using a CPU/core
that is not part of the NUMA local CPU.  If it can be localized,
the kernel's cpumask_local_spread() service will pick a core
that is on the node close to the PCI device.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: remove pragma packed
Shannon Nelson [Sat, 7 Mar 2020 01:04:02 +0000 (17:04 -0800)]
ionic: remove pragma packed

Replace the misguided "#pragma packed" with tags on each
struct/union definition that actually needs it.  This is safer
and more efficient on the various compilers and architectures.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: keep ionic dev on lif init fail
Shannon Nelson [Sat, 7 Mar 2020 01:04:01 +0000 (17:04 -0800)]
ionic: keep ionic dev on lif init fail

If the basic ionic interface works but the lif creation fails,
don't fail the probe.  This will allow us to use the driver to
help inspect the hw/fw/pci interface for debugging purposes.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mptcp-don-t-auto-adjust-rcvbuf-size-if-locked'
David S. Miller [Tue, 10 Mar 2020 02:30:08 +0000 (19:30 -0700)]
Merge branch 'mptcp-don-t-auto-adjust-rcvbuf-size-if-locked'

Florian Westphal says:

====================
mptcp: don't auto-adjust rcvbuf size if locked

The mptcp receive buffer is auto-sized based on the subflow receive
buffer.  Don't do this if userspace specfied a value via SO_RCVBUF
setsockopt.

Also update selftest program to provide a new option to set a fixed
size.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: don't grow mptcp socket receive buffer when rcvbuf is locked
Florian Westphal [Fri, 6 Mar 2020 20:29:46 +0000 (21:29 +0100)]
mptcp: don't grow mptcp socket receive buffer when rcvbuf is locked

The mptcp rcvbuf size is adjusted according to the subflow rcvbuf size.
This should not be done if userspace did set a fixed value.

Fixes: 600911ff5f72bae ("mptcp: add rmem queue accounting")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: selftests: add rcvbuf set option
Florian Westphal [Fri, 6 Mar 2020 20:29:45 +0000 (21:29 +0100)]
mptcp: selftests: add rcvbuf set option

allows to run the tests with fixed receive buffer by passing
"-R <value>" to mptcp_connect.sh.

While at it, add a default 10 second poll timeout so the "-t"
becomes optional -- this makes mptcp_connect simpler to use
during manual testing.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: mt7530: add support for port mirroring
DENG Qingfang [Fri, 6 Mar 2020 12:35:35 +0000 (20:35 +0800)]
net: dsa: mt7530: add support for port mirroring

Add support for configuring port mirroring through the cls_matchall
classifier. We do a full ingress and/or egress capture towards a
capture port.
MT7530 supports one monitor port and multiple mirrored ports.

Signed-off-by: DENG Qingfang <dqfext@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'batadv-next-for-davem-20200306' of git://git.open-mesh.org/linux-merge
David S. Miller [Tue, 10 Mar 2020 02:14:18 +0000 (19:14 -0700)]
Merge tag 'batadv-next-for-davem-20200306' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
This cleanup patchset includes the following patches:

 - bump version strings, by Simon Wunderlich

 - Avoid RCU list-traversal in spinlock, by Sven Eckelmann

 - Replace zero-length array with flexible-array member,
   by Gustavo A. R. Silva
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'r8169-series-with-improvements-to-rtl_tx'
David S. Miller [Tue, 10 Mar 2020 02:07:42 +0000 (19:07 -0700)]
Merge branch 'r8169-series-with-improvements-to-rtl_tx'

Heiner Kallweit says:

====================
r8169: series with improvements to rtl_tx

This series includes few improvements to rtl_tx().
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: remove now unneeded barrier in rtl_tx
Heiner Kallweit [Fri, 6 Mar 2020 22:58:50 +0000 (23:58 +0100)]
r8169: remove now unneeded barrier in rtl_tx

Until ae84bc187337 ("r8169: don't use bit LastFrag in tx descriptor
after send") we used to access another bit in the descriptor, therefore
it seems the barrier was needed. Since this commit DescOwn is the
only bit we're interested in, so the barrier isn't needed any longer.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: simplify usage of rtl8169_unmap_tx_skb
Heiner Kallweit [Fri, 6 Mar 2020 22:56:38 +0000 (23:56 +0100)]
r8169: simplify usage of rtl8169_unmap_tx_skb

Simplify the parameters taken by rtl8169_unmap_tx_skb, this makes
usage of this function easier to read and understand.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: ensure tx_skb is fully reset after calling rtl8169_unmap_tx_skb
Heiner Kallweit [Fri, 6 Mar 2020 22:55:42 +0000 (23:55 +0100)]
r8169: ensure tx_skb is fully reset after calling rtl8169_unmap_tx_skb

So far tx_skb->skb is the only member of the two structs that is not
reset. Make understanding the code easier by resetting both structs
completely in rtl8169_unmap_tx_skb.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: convert while to for loop in rtl_tx
Heiner Kallweit [Fri, 6 Mar 2020 22:54:47 +0000 (23:54 +0100)]
r8169: convert while to for loop in rtl_tx

Slightly improve the code by converting this while to a for loop.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 's390-qeth-next'
David S. Miller [Tue, 10 Mar 2020 01:16:06 +0000 (18:16 -0700)]
Merge branch 's390-qeth-next'

Julian Wiedmann says:

====================
s390/qeth: updates 2020-03-06

please apply the following patch series for qeth to netdev's net-next
tree.

Just a small update to take care of a regression wrt to IRQ handling in
net-next, reported by Qian Cai. The fix needs some qdio layer changes,
so you will find Vasily's Acked-by in that patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: remove VNICC callback parameter struct
Julian Wiedmann [Fri, 6 Mar 2020 08:13:11 +0000 (09:13 +0100)]
s390/qeth: remove VNICC callback parameter struct

After recent cleanups this is just a complicated wrapper around an u32*.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Alexandra Winter <wintera@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qdio: add tighter controls for IRQ polling
Julian Wiedmann [Fri, 6 Mar 2020 08:13:10 +0000 (09:13 +0100)]
s390/qdio: add tighter controls for IRQ polling

Once the call to qdio_establish() has completed, qdio is free to deliver
data IRQs to the device driver's IRQ poll handler.

For qeth (the only qdio driver that currently uses IRQ polling) this is
problematic, since the IRQs can arrive before its NAPI instance is
even registered. Calling napi_schedule() from qeth_qdio_start_poll()
then crashes in various nasty ways.

Until recently qeth checked for IFF_UP to drop such early interrupts,
but that's fragile as well since it doesn't enforce any ordering.

Fix this properly by bringing up the qdio device in IRQS_DISABLED mode,
and have the driver explicitly opt-in to receive data IRQs.
qeth does so from qeth_open(), which kick-starts a NAPI poll and then
calls qdio_start_irq() from qeth_poll().

Also add a matching qdio_stop_irq() in qeth_stop() to switch the qdio
dataplane back into a disabled state.

Fixes: 3d35dbe6224e ("s390/qeth: don't check for IFF_UP when scheduling napi")
CC: Qian Cai <cai@lca.pw>
Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sched: pie: change tc_pie_xstats->prob
Leslie Monis [Mon, 9 Mar 2020 19:10:33 +0000 (00:40 +0530)]
net: sched: pie: change tc_pie_xstats->prob

Commit 105e808c1da2 ("pie: remove pie_vars->accu_prob_overflows")
changes the scale of probability values in PIE from (2^64 - 1) to
(2^56 - 1). This affects the precision of tc_pie_xstats->prob in
user space.

This patch ensures user space is unaffected.

Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Leslie Monis <lesliemonis@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: add bytes not sent to SCM_TIMESTAMPING_OPT_STATS
Yousuk Seung [Mon, 9 Mar 2020 20:16:40 +0000 (13:16 -0700)]
tcp: add bytes not sent to SCM_TIMESTAMPING_OPT_STATS

Add TCP_NLA_BYTES_NOTSENT to SCM_TIMESTAMPING_OPT_STATS that reports
bytes in the write queue but not sent. This is the same metric as
what is exported with tcp_info.tcpi_notsent_bytes.

Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5e: Show/set Rx network flow classification rules on ul rep
Vlad Buslov [Fri, 10 May 2019 17:42:49 +0000 (20:42 +0300)]
net/mlx5e: Show/set Rx network flow classification rules on ul rep

Reuse infrastructure that already exists for pf in legacy mode to show/set
Rx network flow classification rules for uplink representors.

Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
4 years agonet/mlx5e: Init ethtool steering for representors
Vlad Buslov [Fri, 10 May 2019 17:19:29 +0000 (20:19 +0300)]
net/mlx5e: Init ethtool steering for representors

During transition to uplink representors the code responsible for
initializing ethtool steering functionality wasn't added to representor
init rx routine. This causes NULL pointer dereference during configuration
of network flow classification rule with ethtool (only possible to
reproduce with next commit in this series which registers necessary ethtool
callbacks).

Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
4 years agonet/mlx5e: Show/set Rx flow indir table and RSS hash key on ul rep
Vlad Buslov [Fri, 10 May 2019 12:19:35 +0000 (15:19 +0300)]
net/mlx5e: Show/set Rx flow indir table and RSS hash key on ul rep

Reuse infrastructure that already exists for pf in legacy mode to show/set
Rx flow hash indirection table and RSS hash key for uplink representors.

Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
4 years agonet/mlx5e: Introduce root ft concept for representors netdevs
Saeed Mahameed [Wed, 15 May 2019 09:21:35 +0000 (02:21 -0700)]
net/mlx5e: Introduce root ft concept for representors netdevs

Uplink representor traffic will be redirected to an empty root ft rather
than directly to a direct tir or ttc table, this root ft will be empty and
will be used as a link for auto-chaining with ttc table or ethtool tables
in downstream patches.

On load, fs core will connect uplink rep root_ft with ttc table.  In case
ethtool steering will be used, fs core will auto connect root_ft with
the ethtool bypass tables, which will be connected with the ttc table.

vport_rx_rule[uplink_rep]->root_ft->ethtool->ttc.

For non-uplink representors, for simplicity root_ft will always point at
ttc table, hence the replace vport_rx rule logic is removed.

vport_rx_rule[non_uplink_rep]->root_ft(ttc).

For now ethtool steering support can only be available on uplink rep.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
4 years agonet/mlx5: E-switch, make query inline mode a static function
Parav Pandit [Wed, 18 Dec 2019 09:08:38 +0000 (03:08 -0600)]
net/mlx5: E-switch, make query inline mode a static function

mlx5_eswitch_inline_mode_get() is used only in eswitch_offloads.c.
Hence, make it static and adjacent to its caller function.

Reviewed-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Allocate smaller size tables for ft offload
Paul Blakey [Wed, 8 Jan 2020 12:31:53 +0000 (14:31 +0200)]
net/mlx5: Allocate smaller size tables for ft offload

Instead of giving ft tables one of the largest tables available - 4M,
give it a more reasonable size - 64k. Especially since it will
always be created as a miss hook in the following patch.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix an IS_ERR() vs NULL check
Dan Carpenter [Wed, 4 Mar 2020 14:22:24 +0000 (17:22 +0300)]
net/mlx5e: Fix an IS_ERR() vs NULL check

The esw_vport_tbl_get() function returns error pointers on error.

Fixes: 96e326878fa5 ("net/mlx5e: Eswitch, Use per vport tables for mirroring")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Verify goto chain offload support
Eli Cohen [Tue, 18 Feb 2020 09:59:53 +0000 (11:59 +0200)]
net/mlx5: Verify goto chain offload support

According to PRM, forward to flow table along with either packet
reformat or decap is supported only if reformat_and_fwd_to_table
capability is set for the flow table.

Add dependency on the capability and pack all the conditions for "goto
chain" in a single function.

Fix language in error message in case of not supporting forward to a
lower numbered flow table.

Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: E-Switch, Use vport metadata matching only when mandatory
Majd Dibbiny [Mon, 27 Jan 2020 13:50:29 +0000 (15:50 +0200)]
net/mlx5: E-Switch, Use vport metadata matching only when mandatory

Multi-port RoCE mode requires tagging traffic that passes through the
vport.
This matching can cause performance degradation, therefore disable it
and use the legacy matching on vhca_id and source_port when possible.

Fixes: 92ab1eb392c6 ("net/mlx5: E-Switch, Enable vport metadata matching if firmware supports it")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Tidy up and fix reverse christmas ordring
Mark Bloch [Tue, 25 Feb 2020 19:24:54 +0000 (19:24 +0000)]
net/mlx5: Tidy up and fix reverse christmas ordring

Use reverse chirstmas tree inside mlx5e_ethtool_get_link_ksettings.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Expose port speed when possible
Mark Bloch [Tue, 25 Feb 2020 18:04:40 +0000 (18:04 +0000)]
net/mlx5: Expose port speed when possible

When port speed can't be reported based on ext_eth_proto_capability
or eth_proto_capability instead of reporting speed as unknown check
if the port's speed can be inferred based on the data_rate_oper field.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Saeed Mahameed [Mon, 9 Mar 2020 20:01:26 +0000 (13:01 -0700)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

This series adds some HW bits and definitions for mlx5 driver, to be
used by downstream features in both rdma and netdev branches.

* 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
  net/mlx5: HW bit for goto chain offload support
  net/mlx5: Expose link speed directly
  net/mlx5: Introduce TLS and IPSec objects enums
  net/mlx5: Introduce egress acl forward-to-vport capability
  net/mlx5: Expose raw packet pacing APIs
  net/mlx5e: Replace zero-length array with flexible-array member
  net/mlx5: fix spelling mistake "reserverd" -> "reserved"

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet: sgi: ioc3-eth: Remove phy workaround
Thomas Bogendoerfer [Mon, 9 Mar 2020 12:32:40 +0000 (13:32 +0100)]
net: sgi: ioc3-eth: Remove phy workaround

Commit a8d0f11ee50d ("MIPS: SGI-IP27: Enable ethernet phy on second
Origin 200 module") fixes the root cause of not detected PHYs.
Therefore the workaround can go away now.

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-introduce-Qualcomm-IPA-driver'
David S. Miller [Mon, 9 Mar 2020 05:07:10 +0000 (22:07 -0700)]
Merge branch 'net-introduce-Qualcomm-IPA-driver'

Alex Elder says:

====================
net: introduce Qualcomm IPA driver (UPDATED)

This series presents the driver for the Qualcomm IP Accelerator (IPA).

This is version 2 of this updated series.  It includes the following
small changes since the previous version:
  - Now based on net-next instead of v5.6-rc
  - Config option now named CONFIG_QCOM_IPA
  - Some minor cleanup in the GSI code
  - Small change to replenish logic
  - No longer depends on remoteproc bug fixes
What follows is the basically same explanation as was posted previously.

-Alex

I have posted earlier versions of this code previously, but it has
undergone quite a bit of development since the last time, so rather
than calling it "version 3" I'm just treating it as a new series
(indicating it's been updated in this message).  The fast/data path
is the same as before.  But the driver now (nearly) supports a
second platform, its transaction handling has been generalized
and improved, and modem activities are now handled in a more
unified way.

This series is available (based on net-next in branch "ipa_updated-v2"
in this git repository:
  https://git.linaro.org/people/alex.elder/linux.git

The branch depends on other one other small patch that I sent out
for review earlier.
  https://lore.kernel.org/lkml/20200306042302.17602-1-elder@linaro.org/

I want to address some of the discussion that arose last time.

First, there was the WWAN discussion.  Here's the history:
  - This was last posted nine months ago.
  - Reviewers at that time favored developing a new WWAN subsystem that
    would be used for managing devices like this.  And the suggestion
    was to not accept this driver until that could be developed.
  - Along the way, Apple acquired much of Intel's modem business.
    And as a result, the generic framework became less pressing.
  - I did participate in the WWAN subsystem design however, and
    although it went dormant for a while it's been resurrected:
      https://lore.kernel.org/netdev/20200225100053.16385-1-johannes@sipsolutions.net/
  - Unfortunately the proposed WWAN design was not an easy fit
    with Qualcomm's integrated modem interfaces.  Given that
    rmnet is a supported link type for in the upstream "iproute2"
    package (more on this below), I have opted not to integrate
    with any WWAN subsystem.

So in summary, this driver does not integrate with a generic WWAN
framework.  And I'd like it to be accepted upstream despite that.

Next, Arnd Bergmann had some concerns about flow control.  (Note:
some of my discussions with Arnd about this were offline.) The
overall architecture here also involves the "rmnet" driver:
  drivers/net/ethernet/qualcomm/rmnet

The rmnet driver presents a network device for use.  It connects
with another network device presented, by the IPA driver.  The
rmnet driver wraps (and unwraps) packets transferred to (and from)
the IPA driver with QMAP headers.

   ---------------
   | rmnet_data0 |    <-- "real" netdev
   ---------------
          ||       }- QMAP spoken here
   --------------
   | rmnet_ipa0 |     <-- also netdev, transporting QMAP packets
   --------------
          ||
   --------------
  ( IPA hardware )
   --------------

Arnd's concern was that the rmnet_data0 network device does not
have the benefit of information about the state of the underlying
IPA hardware in order to be effective in controlling TX flow.
The feared result is over-buffering of TX packets (bufferbloat).
I began working on some simple experiments to see whether (or how
much) his concern was warranted.  But it turned out that completing
these experiments was much more work than had been hoped.

The rmnet driver is present in the upstream kernel.  There is also
support for the rmnet link type in the upstream "ip" user space
command in the "iproute2" package.  Changing the layering of rmnet
over IPA likely involves deprecating the rmnet driver and its
support in "iproute2".  I would really rather not go down that
path.

There is precedent for this sort of layering of network devices
(L2TP, VLAN).  And any architecture like this would suffer the
issues Arnd mentioned; the problem is not limited to rmnet and IPA.
I do think this is a problem worth solving, but the prudent thing
to do might be to try to solve it more generally.

So to summarize on this issue, this driver does not attempt to
change the way the rmnet and IPA drivers work together.  And even
though I think Arnd's concerns warrant more investigation, I'd like
this driver to to be accepted upstream without any change to this
architecture.

Finally, a more technical description for the series, and some
acknowledgements to some people who contributed to it.

The IPA is a component present in some Qualcomm SoCs that allows
network functions such as aggregation, filtering, routing, and NAT
to be performed without active involvement of the main application
processor (AP).

In this initial patch series these advanced features are not
implemented.  The IPA driver simply provides a network interface
that makes the modem's LTE network available in Linux.  This initial
series supports only the Qualcomm SDM845 SoC.  The Qualcomm SC7180
SoC is partially supported, and support for other platforms will
follow.

This code is derived from a driver developed by Qualcomm.  A version
of the original source can be seen here:
  https://source.codeaurora.org/quic/la/kernel/msm-4.9/tree
in the "drivers/platform/msm/ipa" directory.  Many were involved in
developing this, but the following individuals deserve explicit
acknowledgement for their substantial contributions:

    Abhishek Choubey
    Ady Abraham
    Chaitanya Pratapa
    David Arinzon
    Ghanim Fodi
    Gidon Studinski
    Ravi Gummadidala
    Shihuan Liu
    Skylar Chang
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoarm64: dts: sdm845: add IPA information
Alex Elder [Fri, 6 Mar 2020 04:28:31 +0000 (22:28 -0600)]
arm64: dts: sdm845: add IPA information

Add IPA-related nodes and definitions to "sdm845.dtsi".

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMAINTAINERS: add entry for the Qualcomm IPA driver
Alex Elder [Fri, 6 Mar 2020 04:28:30 +0000 (22:28 -0600)]
MAINTAINERS: add entry for the Qualcomm IPA driver

Add an entry in the MAINTAINERS file for the Qualcomm IPA driver

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: support build of IPA code
Alex Elder [Fri, 6 Mar 2020 04:28:29 +0000 (22:28 -0600)]
soc: qcom: ipa: support build of IPA code

Add build and Kconfig support for the Qualcomm IPA driver.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: AP/modem communications
Alex Elder [Fri, 6 Mar 2020 04:28:28 +0000 (22:28 -0600)]
soc: qcom: ipa: AP/modem communications

This patch implements two forms of out-of-band communication between
the AP and modem.

  - QMI is a mechanism that allows clients running on the AP
    interact with services running on the modem (and vice-versa).
    The AP IPA driver uses QMI to communicate with the corresponding
    IPA driver resident on the modem, to agree on parameters used
    with the IPA hardware and to ensure both sides are ready before
    entering operational mode.

  - SMP2P is a more primitive mechanism available for the modem and
    AP to communicate with each other.  It provides a means for either
    the AP or modem to interrupt the other, and furthermore, to provide
    32 bits worth of information.  The IPA driver uses SMP2P to tell
    the modem what the state of the IPA clock was in the event of a
    crash.  This allows the modem to safely access the IPA hardware
    (or avoid doing so) when a crash occurs, for example, to access
    information within the IPA hardware.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: modem and microcontroller
Alex Elder [Fri, 6 Mar 2020 04:28:27 +0000 (22:28 -0600)]
soc: qcom: ipa: modem and microcontroller

This patch includes code implementing the modem functionality.
There are several communication paths between the AP and modem,
separate from the main data path provided by IPA.  SMP2P provides
primitive messaging and interrupt capability, and QMI allows more
complex out-of-band messaging to occur between entities on the AP
and modem.  (SMP2P and QMI support are added by the next patch.)
Management of these (plus the network device implementing the data
path) is done by code within "ipa_modem.c".

Sort of unrelated, this patch also includes the code supporting the
microcontroller CPU present on the IPA.  The microcontroller can be
used to implement special handling of packets, but at this time we
don't support that.  Still, it is a component that needs to be
initialized, and in the event of a crash we need to do some
synchronization between the AP and the microcontroller.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: immediate commands
Alex Elder [Fri, 6 Mar 2020 04:28:26 +0000 (22:28 -0600)]
soc: qcom: ipa: immediate commands

One TX endpoint (per EE) is used for issuing immediate commands to
the IPA.  These commands request activites beyond simple data
transfers to be done by the IPA hardware.  For example, the IPA is
able to manage routing packets among endpoints, and immediate commands
are used to configure tables used for that routing.

Immediate commands are built on top of GSI transactions.  They are
different from normal transfers (in that they use a special endpoint,
and their "payload" is interpreted differently), so separate functions
are used to issue immediate command transactions.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: filter and routing tables
Alex Elder [Fri, 6 Mar 2020 04:28:25 +0000 (22:28 -0600)]
soc: qcom: ipa: filter and routing tables

This patch contains code implementing filter and routing tables for
the IPA.  A filter table allows rules to be used for filtering
packets that depart the AP at an endpoint.  A filter table entry
contains the address of a set of rules to apply for each endpoint
that supports filtering.

A routing table allows packets to be routed to an endpoint based
on packet metadata.  It is also a table whose entries each contain
the address of a set of routing rules to apply.

Neither filtering nor routing is supported by the current driver.
All table entries refer to rules that mean "no filtering" and "no
routing."

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: IPA endpoints
Alex Elder [Fri, 6 Mar 2020 04:28:24 +0000 (22:28 -0600)]
soc: qcom: ipa: IPA endpoints

This patch includes the code implementing an IPA endpoint.  This is
the primary abstraction implemented by the IPA.  An endpoint is one
end of a network connection between two entities physically
connected to the IPA.  Specifically, the AP and the modem implement
endpoints, and an (AP endpoint, modem endpoint) pair implements the
transfer of network data in one direction between the AP and modem.

Endpoints are built on top of GSI channels, but IPA endpoints
represent the higher-level functionality that the IPA provides.
Data can be sent through a GSI channel, but it is the IPA endpoint
that represents what is on the "other end" to receive that data.
Other functionality, including aggregation, checksum offload and
(at some future date) IP routing and filtering are all associated
with the IPA endpoint.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: GSI transactions
Alex Elder [Fri, 6 Mar 2020 04:28:23 +0000 (22:28 -0600)]
soc: qcom: ipa: GSI transactions

This patch implements GSI transactions.  A GSI transaction is a
structure that represents a single request (consisting of one or
more TREs) sent to the GSI hardware.  The last TRE in a transaction
includes a flag requesting that the GSI interrupt the AP to notify
that it has completed.

TREs are executed and completed strictly in order.  For this reason,
the completion of a single TRE implies that all previous TREs (in
particular all of those "earlier" in a transaction) have completed.

Whenever there is a need to send a request (a set of TREs) to the
IPA, a GSI transaction is allocated, specifying the number of TREs
that will be required.  Details of the request (e.g. transfer offsets
and length) are represented by in a Linux scatterlist array that is
incorporated in the transaction structure.

Once all commands (TREs) are added to a transaction it is committed.
When the hardware signals that the request has completed, a callback
function allows for cleanup or followup activity to be performed
before the transaction is freed.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: IPA interface to GSI
Alex Elder [Fri, 6 Mar 2020 04:28:22 +0000 (22:28 -0600)]
soc: qcom: ipa: IPA interface to GSI

This patch provides interface functions supplied by the IPA layer
that are called from the GSI layer.  One function is called when a
GSI transaction has completed.  The others allow the GSI layer to
inform the IPA layer when the hardware has been told it has new TREs
to execute, and when the hardware has indicated transactions have
completed.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: the generic software interface
Alex Elder [Fri, 6 Mar 2020 04:28:21 +0000 (22:28 -0600)]
soc: qcom: ipa: the generic software interface

This patch includes "gsi.c", which implements the generic software
interface (GSI) for IPA.  The generic software interface abstracts
channels, which provide a means of transferring data either from the
AP to the IPA, or from the IPA to the AP.  A ring buffer of "transfer
elements" (TREs) is used to describe data transfers to perform.  The
AP writes a doorbell register associated with a channel to let it know
it has added new entries (for an AP->IPA channel) or has finished
processing entries (for an IPA->AP channel).

Each channel also has an event ring buffer, used by the IPA to
communicate information about events related to a channel (for
example, the completion of TREs).  The IPA writes its own doorbell
register, which triggers an interrupt on the AP, to signal that
new event information has arrived.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: GSI headers
Alex Elder [Fri, 6 Mar 2020 04:28:20 +0000 (22:28 -0600)]
soc: qcom: ipa: GSI headers

The Generic Software Interface is a layer of the IPA driver that
abstracts the underlying hardware.  The next patch includes the
main code for GSI (including some additional documentation).  This
patch just includes three GSI header files.

  - "gsi.h" is the top-level GSI header file.  This structure is
    is embedded within the IPA structure.  The main abstraction
    implemented by the GSI code is the channel, and this header
    exposes several operations that can be performed on a GSI channel.

  - "gsi_private.h" exposes some definitions that are intended to be
    private, used only by the main GSI code and the GSI transaction
    code (defined in an upcoming patch).

  - Like "ipa_reg.h", "gsi_reg.h" defines the offsets of the 32-bit
    registers used by the GSI layer, along with masks that define the
    position and width of fields less than 32 bits located within
    these registers.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: qcom: ipa: clocking, interrupts, and memory
Alex Elder [Fri, 6 Mar 2020 04:28:19 +0000 (22:28 -0600)]
soc: qcom: ipa: clocking, interrupts, and memory

This patch incorporates three source files (and their headers).  They're
grouped into one patch mainly for the purpose of making the number and
size of patches in this series somewhat reasonable.

  - "ipa_clock.c" and "ipa_clock.h" implement clocking for the IPA device.
    The IPA has a single core clock managed by the common clock framework.
    In addition, the IPA has three buses whose bandwidth is managed by the
    Linux interconnect framework.  At this time the core clock and all
    three buses are either on or off; we don't yet do any more fine-grained
    management than that.  The core clock and interconnects are enabled
    and disabled as a unit, using a unified clock-like abstraction,
    ipa_clock_get()/ipa_clock_put().

  - "ipa_interrupt.c" and "ipa_interrupt.h" implement IPA interrupts.
    There are two hardware IRQs used by the IPA driver (the other is
    the GSI interrupt, described in a separate patch).  Several types
    of interrupt are handled by the IPA IRQ handler; these are not part
    of data/fast path.

  - The IPA has a region of local memory that is accessible by the AP
    (and modem).  Within that region are areas with certain defined
    purposes.  "ipa_mem.c" and "ipa_mem.h" define those regions, and
    implement their initialization.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>