Ido Schimmel says:

====================
mlxsw: Update main pool computation and pool size limits

Petr says:

In Spectrum ASICs, the shared buffer is an area of memory where packets are
kept until they can be transmitted. There are two resources associated with
shared buffer size: cap_total_buffer_size and cap_guaranteed_shared_buffer.
So far, mlxsw has been using the former as a limit when validating shared
buffer pool size configuration. However, the total size also includes
headrooms and reserved space, which really cannot be used for shared buffer
pools. Patch #1 mends this and has mlxsw use the guaranteed size.

To configure default pool sizes, mlxsw has historically hard-coded one or
two smallish pools, and one "main" pool that took most of the shared buffer
(that would be pool 0 on ingress and pool 4 on egress). During the
development of Spectrum-2, it became clear that the shared buffer size
keeps shrinking as bugs are identified and worked around. In order to
prevent having to tweak the size of pools 0 and 4 to catch up with updates
to values reported by the FW, patch #2 changes the way these pools are set.
Instead of hard-coding a fixed value, the main pool now takes whatever is
left from the guaranteed size after the smaller pool(s) are taken into
account.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Instead of hard-coding the size of the largest pool, calculate it from the
reported guaranteed shared buffer size and sizes of other pools (currently
only the CPU port pool).
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There are two resources associated with shared buffer size:
cap_total_buffer_size, and cap_guaranteed_shared_buffer. So far, mlxsw has
been using the former as a limit to determine how large a pool size is
allowed to be. However, the total size also includes headrooms and reserved
space, which really cannot be used for shared buffer pools.

Therefore convert mlxsw to use the latter resource as a limit. Adjust
hard-coded pool sizes to be the guaranteed size minus 256000 bytes for CPU
port pool. On Spectrum-1 that actually leads to an increase. A follow-up
patch will have this size calculated automatically.
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Setting PCI_EXP_DEVCTL_NOSNOOP_EN for certain chip versions had been
added to the vendor driver more than 10 years ago, and copied from
there to r8169. It has been removed from the vendor driver meanwhile
and I think we can safely remove this too.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tao Ren says:

====================
net: phy: support 1000Base-X auto-negotiation for BCM54616S

This patch series aims at supporting auto negotiation when BCM54616S is
running in 1000Base-X mode: without the patch series, BCM54616S PHY driver
would report incorrect link speed in 1000Base-X mode.

Patch #1 (of 3) modifies assignment to OR when dealing with dev_flags in
phy_attach_direct function, so that dev_flags updated in BCM54616S PHY's
probe callback won't be lost.

Patch #2 (of 3) adds several genphy_c37_* functions to support clause 37
1000Base-X auto-negotiation, and these functions are called in BCM54616S
PHY driver.

Patch #3 (of 3) detects BCM54616S PHY's operation mode and calls according
genphy_c37_* functions to configure auto-negotiation and parse link
attributes (speed, duplex, and etc.) in 1000Base-X mode.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The BCM54616S PHY cannot work properly in RGMII->1000Base-X mode, mainly
because genphy functions are designed for copper links, and 1000Base-X
(clause 37) auto negotiation needs to be handled differently.

This patch enables 1000Base-X support for BCM54616S by customizing 3
driver callbacks, and it's verified to be working on Facebook CMM BMC
platform (RGMII->1000Base-KX):

  - probe: probe callback detects PHY's operation mode based on
    INTERF_SEL[1:0] pins and 1000X/100FX selection bit in SerDES 100-FX
    Control register.

  - config_aneg: calls genphy_c37_config_aneg when the PHY is running in
    1000Base-X mode; otherwise, genphy_config_aneg will be called.

  - read_status: calls genphy_c37_read_status when the PHY is running in
    1000Base-X mode; otherwise, genphy_read_status will be called.

Note: BCM54616S PHY can also be configured in RGMII->100Base-FX mode, and
100Base-FX support is not available as of now.
Signed-off-by: Tao Ren <taoren@fb.com>
Acked-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds support for clause 37 1000Base-X auto-negotiation.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Tao Ren <taoren@fb.com>
Tested-by: René van Dorst <opensource@vdorst.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Modify the assignment to OR when dealing with phydev->dev_flags in
phy_attach_direct function, and this is to make sure dev_flags set in
driver's probe callback won't be lost.
Suggested-by: Andrew Lunn <andrew@lunn.ch>
CC: Heiner Kallweit <hkallweit1@gmail.com>
CC: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Tao Ren <taoren@fb.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since commit 342db221 ("sched: Call skb_get_hash_perturb
in sch_fq_codel") we no longer need anything from this file.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Vivien Didelot says:

====================
The dsa_switch structure represents the physical switch device itself,
and is allocated by the driver. The dsa_switch_tree and dsa_port structures
represent the logical switch fabric (eventually composed of multiple switch
devices) and its ports, and are allocated by the DSA core.

This branch lists the logical ports directly in the fabric which simplifies
the iteration over all ports when assigning the default CPU port or configuring
the D in DSA in drivers like mv88e6xxx.

This also removes the unique dst->cpu_dp pointer and is a first step towards
supporting multiple CPU ports and dropping the DSA_MAX_PORTS limitation.

Because the dsa_port structures are not tied to the dsa_switch structure
anymore, we do not need to provide an helper for the drivers to allocate a
switch structure. Like in many other subsystems, drivers can now embed their
dsa_switch structure as they wish into their private structure. This will
be particularly interesting for the Broadcom drivers which were currently
limited by the dynamically allocated array of DSA ports.

The series implements the list of dsa_port structures, makes use of it,
then drops dst->cpu_dp and the dsa_switch_alloc helper.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Now that ports are dynamically listed in the fabric, there is no need
to provide a special helper to allocate the dsa_switch structure. This
will give more flexibility to drivers to embed this structure as they
wish in their private structure.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Allocate the struct dsa_port the first time it is accessed with
dsa_port_touch, and remove the static dsa_port array from the
dsa_switch structure.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Like the dsa_switch_tree structures, the dsa_port structures will be
allocated on switch registration.

The SJA1105 driver is the only one accessing the dsa_port structure
after the switch allocation and before the switch registration.
For that reason, move switch registration prior to assigning the priv
member of the dsa_port structures.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Instead of digging into the other dsa_switch structures of the fabric
and relying too much on the dsa_to_port helper, use the new list
of switch fabric ports to remap the Port VLAN Map of local bridge
group members or remap the Port VLAN Table entry of external bridge
group members.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Instead of digging into the other dsa_switch structures of the fabric
and relying too much on the dsa_to_port helper, use the new list of
switch fabric ports to define the mask of the local ports allowed to
receive frames from another port of the fabric.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Since mv88e6xxx_pvt_map is a static helper, no need to return
-EOPNOTSUPP if the chip has no PVT, simply silently skip the operation.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of iterating over switches and their
ports when setting up the default CPU port. Unassign it on teardown.

Now that we can iterate over multiple CPU ports, remove dst->cpu_dp.

At the same time, provide a better error message for CPU-less tree.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of iterating over switches and their
ports when looking up the first CPU port in the tree.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Now that we have a potential list of CPU ports, make use of it instead
of only configuring the master device of an unique CPU port.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of iterating over switches and their
ports to find a port from a given node.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of accessing the dsa_switch array
of ports when iterating over DSA ports of a switch to set up the
routing table.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of iterating over switches and their
ports when setting up the switches and their ports.

At the same time, provide setup states and messages for ports and
switches as it is done for the trees.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of iterating over switches and their
ports when looking for a slave device from a given master interface.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use the new ports list instead of accessing the dsa_switch array
of ports in the dsa_to_port helper.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Add a list of switch ports within the switch fabric. This will help the
lookup of a port inside the whole fabric, and it is the first step
towards supporting multiple CPU ports, before deprecating the usage of
the unique dst->cpu_dp pointer.

In preparation for a future allocation of the dsa_port structures,
return -ENOMEM in case no structure is returned, even though this
error cannot be reached yet.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Do not let the drivers access the ds->ports static array directly
while there is a dsa_to_port helper for this purpose.

At the same time, un-const this helper since the SJA1105 driver
assigns the priv member of the returned dsa_port structure.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Heiner Kallweit says:

====================
The attempt to improve performance by changing the PCIe max read request
size was added in the vendor driver more than 10 years back and copied
to r8169 driver. In the vendor driver this has been removed long ago.
Obviously it had no effect, also in my tests I didn't see any
difference. Typically the max payload size is less than 512 bytes
anyway, and the PCI core takes care that the maximum supported value
is set. So let's remove fiddling with PCIe max read request size from
r8169 too. This change allows to simplify the driver in the subsequent
three patches of this series.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

We can remove rtl_hw_start_8168bef() and use rtl_hw_start_8168b()
instead because setting register Config4 is done in
rtl_jumbo_config(), being called from rtl_hw_start().
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

We can remove rtl_hw_start_8168dp() because it's the same as
rtl_hw_start_8168dp() now.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

r8168b_0_hw_jumbo_enable() and r8168b_0_hw_jumbo_disable() both do the
same and just set PCI_EXP_DEVCTL_NOSNOOP_EN. We can simplify the code
by moving this setting for RTL8168B to rtl_hw_start_8168().
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

The attempt to improve performance by changing the PCIe max read request
size was added in the vendor driver more than 10 years back and copied
to r8169 driver. In the vendor driver this has been removed long ago.
Obviously it had no effect, also in my tests I didn't see any
difference. Typically the max payload size is less than 512 bytes
anyway, and the PCI core takes care that the maximum supported value
is set. So let's remove fiddling with PCIe max read request size from
r8169 too.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Karsten Graul says:

====================
More patches to address abnormal termination processing of
sockets and link groups.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

With the introduction of the link group termination worker there is
no longer a need to postpone smc_close_active_abort() to a worker.
To protect socket destruction due to normal and abnormal socket
closing, the socket refcount is increased.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Use a worker for link group termination to guarantee process context.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

If a link group and its connections must be terminated,
* wake up socket waiters
* do not enable buffer reuse

A linkgroup might be terminated while normal connection closing
is running. Avoid buffer reuse and its related LLC DELETE RKEY
call, if linkgroup termination has started. And use the earliest
indication of linkgroup termination possible, namely the removal
from the linkgroup list.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

There are lots of link group termination scenarios. Most of them
still allow to inform the peer of the terminating sockets about aborting.
This patch tries to call smc_close_abort() for terminating sockets.

And the internal TCP socket is reset with tcp_abort().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Usually link groups are freed delayed to enable quick connection
creation for a follow-on SMC socket. Terminated link groups are
freed faster. This patch makes sure, fast schedule of link group
freeing is not rescheduled by a delayed schedule. And it makes sure
link group freeing is not rescheduled, if the real freeing is already
running.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

Locking hierarchy requires that the link group conns_lock can be
taken if the socket lock is held, but not vice versa. Nevertheless
socket termination during abnormal link group termination should
be protected by the socket lock.
This patch reduces the time segments the link group conns_lock is
held to enable usage of lock_sock in smc_lgr_terminate().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

When a link group is to be terminated, it is sufficient to hold
the lgr lock when unlinking the link group from its list.
Move the lock-protected link group unlinking into smc_lgr_terminate().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

The resources for a terminated socket are being cleaned up.
This patch makes sure
* no more data is received for an actively terminated socket
* no more data is sent for an actively or passively terminated socket
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>