linux-2.6-microblaze.git
4 years agoigc: Remove unused registers
Sasha Neftin [Sun, 12 Apr 2020 14:21:58 +0000 (17:21 +0300)]
igc: Remove unused registers

Tx data FIFO Head/Tail, Saved and Packet Count registers
not applicable for i225 LAN controller.
This patch comes to clean up these registers.

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Remove unused IGC_ICS_DRSTA define
Sasha Neftin [Sun, 12 Apr 2020 06:28:29 +0000 (09:28 +0300)]
igc: Remove unused IGC_ICS_DRSTA define

Device reset assert for interrupt cause register not in
use for i225 device.
This patch comes to clean up this define.

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Dump ETQF registers
Andre Guedes [Tue, 7 Apr 2020 21:07:11 +0000 (14:07 -0700)]
igc: Dump ETQF registers

This patch adds the EType Queue Filter (ETQF) registers to the list of
registers dumped by igc_get_regs().

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Refactor ethertype filtering code
Andre Guedes [Tue, 7 Apr 2020 21:07:10 +0000 (14:07 -0700)]
igc: Refactor ethertype filtering code

The whole ethertype filtering code is implemented in igc_ethtool.c and
mixes logic from ethtool and core parts. This patch refactors it so core
logic is moved to igc_main.c, aligning the ethertype filtering code
organization with the rest of the filtering code from the driver (MAC
address and VLAN priority).

Besides moving code to igc_main.c, this patch also does some minor
improvements to the code. Below are some highlights.

In case all filters are already in use and the user tries to add another
filter, we return -ENOSPC instead of -EINVAL so a more meaningful error
code is provided. This also aligns with the behavior implemented in MAC
address filtering code.

With this code refactoring, 'etype_bitmap' array in struct igc_adapter
and 'etype_reg_index' in struct igc_nfc_filter are not needed anymore
and are removed.

Log messages are added to help debugging the ethertype filtering code.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Fix MAX_ETYPE_FILTER value
Andre Guedes [Tue, 7 Apr 2020 21:07:09 +0000 (14:07 -0700)]
igc: Fix MAX_ETYPE_FILTER value

The I225 controller has 8 ethertype filters, not 4. This patch fixes the
MAX_ETYPE_FILTER macro accordingly.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Remove ethertype filter in PTP code
Andre Guedes [Tue, 7 Apr 2020 21:07:08 +0000 (14:07 -0700)]
igc: Remove ethertype filter in PTP code

The driver only supports hardware timestamping for all incoming
traffic (HWTSTAMP_FILTER_ALL) which is enabled via Rx Time Sync
Control (TSYNCRXCTL) register already. Therefore, the ethertype
filter set in in igc_ptp_set_timestamp_mode() is useless so this
patch removes it.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Remove duplicated IGC_RXPBS macro
Andre Guedes [Tue, 7 Apr 2020 21:07:07 +0000 (14:07 -0700)]
igc: Remove duplicated IGC_RXPBS macro

This patch remove the IGC_RXPBS macro defined in line 233 since it is
already defined in line 18 with the exactly same value.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Refactor VLAN priority filtering code
Andre Guedes [Fri, 24 Apr 2020 01:11:19 +0000 (18:11 -0700)]
igc: Refactor VLAN priority filtering code

The whole VLAN priority filtering code is implemented in igc_ethtool.c
and mixes logic from ethtool and core parts. This patch refactors it so
core logic is moved to igc_main.c, aligning the VLAN priority filtering
code organization with the MAC address filtering code.

This patch also takes the opportunity to add some log messages to ease
debugging.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Return -EOPNOTSUPP when VLAN mask doesn't match
Andre Guedes [Fri, 3 Apr 2020 18:17:42 +0000 (11:17 -0700)]
igc: Return -EOPNOTSUPP when VLAN mask doesn't match

The I225 controller supports Rx queue assignment based on VLAN priority
only. Other Tag Control Information (TCI) are valid, but not supported
by the driver. So this patch changes the returning code from igc_add_
ethtool_nfc_entry() to -EOPNOTSUPP in order to provide more meaningful
information on why the function failed.

It also adds a debug messages to give the user a hint about what went
wrong with the NFC setup.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Dump VLANPQF register
Andre Guedes [Fri, 3 Apr 2020 18:17:41 +0000 (11:17 -0700)]
igc: Dump VLANPQF register

This patch adds the VLAN Priority Queue Filter Register (VLANPQF) to the
list of registers dumped by igc_get_regs().

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Rename IGC_VLAPQF macro
Andre Guedes [Fri, 3 Apr 2020 18:17:40 +0000 (11:17 -0700)]
igc: Rename IGC_VLAPQF macro

This patch renames the IGC_VLAPQF macro to IGC_VLANPQF as well as
related macros so they match the register name and fields described in
the data sheet.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Clean up obsolete NVM defines
Sasha Neftin [Wed, 1 Apr 2020 11:38:44 +0000 (14:38 +0300)]
igc: Clean up obsolete NVM defines

Packet buffer allocation, reserved word and pointer guard
not applicable for i225 parts.
This patch comes to clean up these obsolete defines

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: remove IGC_REMOVED function
Vitaly Lifshits [Wed, 1 Apr 2020 09:16:44 +0000 (12:16 +0300)]
igc: remove IGC_REMOVED function

igc driver has leftovers from the previous device that supported
Virtualization. This can be found in the function IGC_REMOVED which
became obsolete, and can be removed.

Signed-off-by: Vitaly Lifshits <vitaly.lifshits@intel.com>
Acked-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Remove PCIe Control register
Sasha Neftin [Wed, 1 Apr 2020 08:43:20 +0000 (11:43 +0300)]
igc: Remove PCIe Control register

GCR (PCIe Control) register not in use and should be removed
This patch clean up this register

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoMerge branch 's390-next'
David S. Miller [Tue, 19 May 2020 19:48:39 +0000 (12:48 -0700)]
Merge branch 's390-next'

Julian Wiedmann says:

====================
s390/net: updates 2020-05-19

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

s390 dropped its support for power management, this removes the relevant
code from the s390 network drivers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/net: remove pm support from iucv drivers
Julian Wiedmann [Tue, 19 May 2020 19:09:04 +0000 (21:09 +0200)]
s390/net: remove pm support from iucv drivers

Commit 394216275c7d ("s390: remove broken hibernate / power management support")
removed support for ARCH_HIBERNATION_POSSIBLE on s390.
So drop the unused pm ops from the iucv drivers.

CC: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/net: remove pm ops from ccwgroup drivers
Julian Wiedmann [Tue, 19 May 2020 19:09:03 +0000 (21:09 +0200)]
s390/net: remove pm ops from ccwgroup drivers

commit 5e1fb45ec8e2 ("s390/ccwgroup: remove pm support") removed power
management support from the ccwgroup bus driver. So remove the
associated callbacks from all ccwgroup drivers.

CC: Vineeth Vijayan <vneethv@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 19 May 2020 19:02:29 +0000 (12:02 -0700)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
1GbE Intel Wired LAN Driver Updates 2020-05-18

This series contains updates to igc driver only.

Sasha adds ECN support for TSO by adding the NETIF_F_TSO_ECN flag, which
aligns with other Intel drivers.  Also cleaned up defines that are not
supported or used in the igc driver.

Andre does most of the changes with updating the log messages for igc
driver.

Vitaly adds support for EEPROM, register and link ethtool
self-tests.

v2: Fixed up the added ethtool self-tests based on feedback from the
    community.  Dropped the four patches that removed '\n' from log
    messages.
v3: Reverted the debug message changes in patch 2 for messages in
    igc_probe, also made reg_test[] static in patch 3 based on community
    feedback
v4: Updated the patch description for patch 2, which referred to changes
    that no longer existed in the patch
v5: Scrubbed patches 4-7 patch description, which also referred to
    changes that no longer existed in the patch
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: work around an irq coalescing related tx timeout
Heiner Kallweit [Mon, 18 May 2020 20:47:16 +0000 (22:47 +0200)]
r8169: work around an irq coalescing related tx timeout

In [0] a user reported reproducible tx timeouts on RTL8168f except
PktCntrDisable is set and irq coalescing is enabled.
Realtek told me that they are not aware of any related hw issue on
this chip version, therefore root cause is still unknown. It's not
clear whether the issue affects one or more chip versions in general,
or whether issue is specific to reporter's system.
Due to this level of uncertainty, and due to the fact that I'm aware
of this one report only, let's apply the workaround on net-next only.
After this change setting irq coalescing via ethtool can reliably
avoid the issue on the affected system.

[0] https://bugzilla.kernel.org/show_bug.cgi?id=207205

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: improve rtl8169_mark_to_asic
Heiner Kallweit [Mon, 18 May 2020 20:22:09 +0000 (22:22 +0200)]
r8169: improve rtl8169_mark_to_asic

Let the compiler decide about inlining, and as confirmed by Eric it's
better to use WRITE_ONCE here to ensure that the descriptor ownership
is transferred to NIC immediately.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: make rtl_rx better readable
Heiner Kallweit [Mon, 18 May 2020 20:14:21 +0000 (22:14 +0200)]
r8169: make rtl_rx better readable

Avoid the goto from the rx error handling branch into the else branch,
and in general avoid having the main rx work in the else branch.
In addition ensure proper reverse xmas tree order of variables in the
for loop.

No functional change intended.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: seeq: Use %pM format specifier for MAC addresses
Andy Shevchenko [Mon, 18 May 2020 19:21:28 +0000 (22:21 +0300)]
net: seeq: Use %pM format specifier for MAC addresses

Convert to %pM instead of using custom code.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: Use %pM format specifier for MAC addresses
Andy Shevchenko [Mon, 18 May 2020 19:18:31 +0000 (22:18 +0300)]
cxgb4: Use %pM format specifier for MAC addresses

Convert to %pM instead of using custom code.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'move-the-SIOCDELRT-and-SIOCADDRT-compat_ioctl-handlers-v3'
David S. Miller [Tue, 19 May 2020 00:35:02 +0000 (17:35 -0700)]
Merge branch 'move-the-SIOCDELRT-and-SIOCADDRT-compat_ioctl-handlers-v3'

Christoph Hellwig says:

====================
move the SIOCDELRT and SIOCADDRT compat_ioctl handlers v3

this series moves the compat_ioctl handlers into the protocol handlers,
avoiding the need to override the address space limited as in the current
handler.

Changes since v3:
 - moar variable reordering

Changes since v1:
 - reorder a bunch of variable declarations
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv4,appletalk: move SIOCADDRT and SIOCDELRT handling into ->compat_ioctl
Christoph Hellwig [Mon, 18 May 2020 06:28:08 +0000 (08:28 +0200)]
ipv4,appletalk: move SIOCADDRT and SIOCDELRT handling into ->compat_ioctl

To prepare removing the global routing_ioctl hack start lifting the code
into the ipv4 and appletalk ->compat_ioctl handlers.  Unlike the existing
handler we don't bother copying in the name - there are no compat issues for
char arrays.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoappletalk: factor out a atrtr_ioctl_addrt helper
Christoph Hellwig [Mon, 18 May 2020 06:28:07 +0000 (08:28 +0200)]
appletalk: factor out a atrtr_ioctl_addrt helper

Add a helper than can be shared with the upcoming compat ioctl handler.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: move SIOCADDRT and SIOCDELRT handling into ->compat_ioctl
Christoph Hellwig [Mon, 18 May 2020 06:28:06 +0000 (08:28 +0200)]
ipv6: move SIOCADDRT and SIOCDELRT handling into ->compat_ioctl

To prepare removing the global routing_ioctl hack start lifting the code
into a newly added ipv6 ->compat_ioctl handler.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: lift copy_from_user out of ipv6_route_ioctl
Christoph Hellwig [Mon, 18 May 2020 06:28:05 +0000 (08:28 +0200)]
ipv6: lift copy_from_user out of ipv6_route_ioctl

Prepare for better compat ioctl handling by moving the user copy out
of ipv6_route_ioctl.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoigc: Remove unneeded register
Sasha Neftin [Sun, 29 Mar 2020 11:35:54 +0000 (14:35 +0300)]
igc: Remove unneeded register

Flow control status register not applicable for i225 parts
so clean up the unneeded define.

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Remove unneeded definition
Sasha Neftin [Sun, 29 Mar 2020 09:01:51 +0000 (12:01 +0300)]
igc: Remove unneeded definition

PHY_FORCE_LIMIT definition not in use and could be removed
i225 parts support auto negotiation mechanism

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Use netdev log helpers in igc_base.c
Andre Guedes [Wed, 25 Mar 2020 00:38:19 +0000 (17:38 -0700)]
igc: Use netdev log helpers in igc_base.c

This patch coverts one pr_debug() call to hw_dbg() in order to keep log
output aligned with the rest of the driver. hw_dbg() is actually a macro
defined in igc_hw.h that expands to netdev_dbg().

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Use netdev log helpers in igc_dump.c
Andre Guedes [Wed, 25 Mar 2020 00:38:18 +0000 (17:38 -0700)]
igc: Use netdev log helpers in igc_dump.c

In igc_dump.c we print log messages using dev_* and pr_* helpers,
generating inconsistent output with the rest of the driver. Since this
is a network device driver, we should preferably use netdev_* helpers
because they append the interface name to the message, helping making
sense out of the logs.

This patch converts all dev_* and pr_* calls to netdev_*.

Quick note about igc_rings_dump(): This function is always called with
valid adapter->netdev so there is not need to check it.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Use netdev log helpers in igc_ptp.c
Andre Guedes [Wed, 25 Mar 2020 00:38:17 +0000 (17:38 -0700)]
igc: Use netdev log helpers in igc_ptp.c

In igc_ptp.c we print log messages using dev_* helpers, generating
inconsistent output with the rest of the driver. Since this is a network
device driver, we should preferably use netdev_* helpers because they
append the interface name to the message, helping making sense out of
the logs.

This patch converts all dev_* calls to netdev_*.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Use netdev log helpers in igc_ethtool.c
Andre Guedes [Wed, 25 Mar 2020 00:38:16 +0000 (17:38 -0700)]
igc: Use netdev log helpers in igc_ethtool.c

In igc_ethtool.c we print log messages using dev_* helpers, generating
inconsistent output with the rest of the driver. Since this is a network
device driver, we should preferably use netdev_* helpers because they
append the interface name to the message, helping making sense the of
the logs.

This patch converts all dev_* calls to netdev_*.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agonet: phy: simplify phy_link_change arguments
Doug Berger [Mon, 18 May 2020 22:23:59 +0000 (15:23 -0700)]
net: phy: simplify phy_link_change arguments

This function was introduced to allow for different handling of
link up and link down events particularly with regard to the
netif_carrier. The third argument do_carrier allowed the flag to
be left unchanged.

Since then the phylink has introduced an implementation that
completely ignores the third parameter since it never wants to
change the flag and the phylib always sets the third parameter
to true so the flag is always changed.

Therefore the third argument (i.e. do_carrier) is no longer
necessary and can be removed. This also means that the phylib
phy_link_down() function no longer needs its second argument.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoigc: add support to eeprom, registers and link self-tests
Vitaly Lifshits [Tue, 5 May 2020 14:06:38 +0000 (17:06 +0300)]
igc: add support to eeprom, registers and link self-tests

Introduced igc_diag.c and igc_diag.h, these files have the
diagnostics functionality of igc driver. For the time being
these files are being used by ethtool self-test callbacks.
Which mean that eeprom, registers and link self-tests for
ethtool were implemented.

Signed-off-by: Vitaly Lifshits <vitaly.lifshits@intel.com>
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Use netdev log helpers in igc_main.c
Andre Guedes [Wed, 25 Mar 2020 00:38:15 +0000 (17:38 -0700)]
igc: Use netdev log helpers in igc_main.c

In igc_main.c we print log messages using both dev_* and netdev_*
helpers, generating inconsistent output. Since this is a network device
driver, we should preferably use netdev_* helpers because they append
the interface name to the message, helping making sense out of the logs.

This patch converts all dev_* calls to netdev_*. There is only two
exceptions:
  1) calls wihtin igc_probe (net_device has not been registered yet)
  2) calls in igc_init_module (module initialization).

It also takes this opportunity to improve some messages.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoigc: Add ECN support for TSO
Sasha Neftin [Sun, 22 Mar 2020 07:58:46 +0000 (09:58 +0200)]
igc: Add ECN support for TSO

Align with other Intel drivers and add ECN support for TSO.

Add NETIF_F_TSO_ECN flag

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agords: convert get_user_pages() --> pin_user_pages()
John Hubbard [Sun, 17 May 2020 01:23:36 +0000 (18:23 -0700)]
rds: convert get_user_pages() --> pin_user_pages()

This code was using get_user_pages_fast(), in a "Case 2" scenario
(DMA/RDMA), using the categorization from [1]. That means that it's
time to convert the get_user_pages_fast() + put_page() calls to
pin_user_pages_fast() + unpin_user_pages() calls.

There is some helpful background in [2]: basically, this is a small
part of fixing a long-standing disconnect between pinning pages, and
file systems' use of those pages.

[1] Documentation/core-api/pin_user_pages.rst

[2] "Explicit pinning of user-space pages":
    https://lwn.net/Articles/807108/

Cc: David S. Miller <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Cc: linux-rdma@vger.kernel.org
Cc: rds-devel@oss.oracle.com
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mptcp-do-not-block-on-subflow-socket'
David S. Miller [Sun, 17 May 2020 19:35:35 +0000 (12:35 -0700)]
Merge branch 'mptcp-do-not-block-on-subflow-socket'

Florian Westphal says:

====================
mptcp: do not block on subflow socket

This series reworks mptcp_sendmsg logic to avoid blocking on the subflow
socket.

It does so by removing the wait loop from mptcp_sendmsg_frag helper.

In order to do that, it moves prerequisites that are currently
handled in mptcp_sendmsg_frag (and cause it to wait until they are
met, e.g. frag cache refill) into the callers.

The worker can just reschedule in case no subflow socket is ready,
since it can't wait -- doing so would block other work items and
doesn't make sense anyway because we should not (re)send data
in case resources are already low.

The sendmsg path can use the existing wait logic until memory
becomes available.

Because large send requests can result in multiple mptcp_sendmsg_frag
calls from mptcp_sendmsg, we may need to restart the socket lookup in
case subflow can't accept more data or memory is low.

Doing so blocks on the mptcp socket, and existing wait handling
releases the msk lock while blocking.

Lastly, no need to use GFP_ATOMIC for extension allocation:
extend __skb_ext_alloc with gfp_t arg instead of hard-coded ATOMIC and
then relax the allocation constraints for mptcp case: those requests
occur in process context.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: allow __skb_ext_alloc to sleep
Florian Westphal [Sat, 16 May 2020 08:46:23 +0000 (10:46 +0200)]
net: allow __skb_ext_alloc to sleep

mptcp calls this from the transmit side, from process context.
Allow a sleeping allocation instead of unconditional GFP_ATOMIC.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: remove inner wait loop from mptcp_sendmsg_frag
Florian Westphal [Sat, 16 May 2020 08:46:22 +0000 (10:46 +0200)]
mptcp: remove inner wait loop from mptcp_sendmsg_frag

previous patches made sure we only call into this function
when these prerequisites are met, so no need to wait on the
subflow socket anymore.

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/7
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: fill skb page frag cache outside of mptcp_sendmsg_frag
Florian Westphal [Sat, 16 May 2020 08:46:21 +0000 (10:46 +0200)]
mptcp: fill skb page frag cache outside of mptcp_sendmsg_frag

The mptcp_sendmsg_frag helper contains a loop that will wait on the
subflow sk.

It seems preferrable to only wait in mptcp_sendmsg() when blocking io is
requested.  mptcp_sendmsg already has such a wait loop that is used when
no subflow socket is available for transmission.

This is another preparation patch that makes sure we call
mptcp_sendmsg_frag only if the page frag cache has been refilled.

Followup patch will remove the wait loop from mptcp_sendmsg_frag().

The retransmit worker doesn't need to do this refill as it won't
transmit new mptcp-level data.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: fill skb extension cache outside of mptcp_sendmsg_frag
Florian Westphal [Sat, 16 May 2020 08:46:20 +0000 (10:46 +0200)]
mptcp: fill skb extension cache outside of mptcp_sendmsg_frag

The mptcp_sendmsg_frag helper contains a loop that will wait on the
subflow sk.

It seems preferrable to only wait in mptcp_sendmsg() when blocking io is
requested.  mptcp_sendmsg already has such a wait loop that is used when
no subflow socket is available for transmission.

This is a preparation patch that makes sure we call
mptcp_sendmsg_frag only if a skb extension has been allocated.

Moreover, such allocation currently uses GFP_ATOMIC while it
could use sleeping allocation instead.

Followup patches will remove the wait loop from mptcp_sendmsg_frag()
and will allow to do a sleeping allocation for the extension.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: avoid blocking in tcp_sendpages
Florian Westphal [Sat, 16 May 2020 08:46:19 +0000 (10:46 +0200)]
mptcp: avoid blocking in tcp_sendpages

The transmit loop continues to xmit new data until an error is returned
or all data was transmitted.

For the blocking i/o case, this means that tcp_sendpages() may block on
the subflow until more space becomes available, i.e. we end up sleeping
with the mptcp socket lock held.

Instead we should check if a different subflow is ready to be used.

This restarts the subflow sk lookup when the tx operation succeeded
and the tcp subflow can't accept more data or if tcp_sendpages
indicates -EAGAIN on a blocking mptcp socket.

In that case we also need to set the NOSPACE bit to make sure we get
notified once memory becomes available.

In case all subflows are busy, the existing logic will wait until a
subflow is ready, releasing the mptcp socket lock while doing so.

The mptcp worker already sets DONTWAIT, so no need to make changes there.

v2:
 * set NOSPACE bit
 * add a comment to clarify that mptcp-sk sndbuf limits need to
   be checked as well.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: break and restart in case mptcp sndbuf is full
Florian Westphal [Sat, 16 May 2020 08:46:18 +0000 (10:46 +0200)]
mptcp: break and restart in case mptcp sndbuf is full

Its not enough to check for available tcp send space.

We also hold on to transmitted data for mptcp-level retransmits.
Right now we will send more and more data if the peer can ack data
at the tcp level fast enough, since that frees up tcp send buffer space.

But we also need to check that data was acked and reclaimed at the mptcp
level.

Therefore add needed check in mptcp_sendmsg, flush tcp data and
wait until more mptcp snd space becomes available if we are over the
limit.  Before we wait for more data, also make sure we start the
retransmit timer if we ran out of sndbuf space.

Otherwise there is a very small chance that we wait forever:

 * receiver is waiting for data
 * sender is blocked because mptcp socket buffer is full
 * at tcp level, all data was acked
 * mptcp-level snd_una was not updated, because last ack
   that acknowledged the last data packet carried an older
   MPTCP-ack.

Restarting the retransmit timer avoids this problem: if TCP
subflow is idle, data is retransmitted from the RTX queue.

New data will make the peer send a new, updated MPTCP-Ack.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: move common nospace-pattern to a helper
Florian Westphal [Sat, 16 May 2020 08:46:17 +0000 (10:46 +0200)]
mptcp: move common nospace-pattern to a helper

Paolo noticed that ssk_check_wmem() has same pattern, so add/use
common helper for both places.

Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: Drop 'pref medium' in route checks
David Ahern [Sun, 17 May 2020 18:00:33 +0000 (12:00 -0600)]
selftests: Drop 'pref medium' in route checks

The 'pref medium' attribute was moved in iproute2 to be near the prefix
which is where it applies versus after the last nexthop. The nexthop
tests were updated to drop the string from route checking, but it crept
in again with the compat tests.

Fixes: 4dddb5be136a ("selftests: net: add new testcases for nexthop API compat mode sysctl")
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ipa-sc7180-suspend-resume'
David S. Miller [Sat, 16 May 2020 23:47:19 +0000 (16:47 -0700)]
Merge branch 'net-ipa-sc7180-suspend-resume'

Alex Elder says:

====================
net: ipa: sc7180 suspend/resume

This series permits suspend/resume to work for the IPA driver
on the Qualcomm SC7180 SoC.  The IPA version on this SoC requires
interrupts to be enabled when the suspend and resume callbacks are
made, and the first patch moves away from using the noirq variants.
The second patch fixes a problem with resume that occurs because
pending interrupts were being cleared before starting a channel.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: do not clear interrupt in gsi_channel_start()
Alex Elder [Fri, 15 May 2020 20:07:31 +0000 (15:07 -0500)]
net: ipa: do not clear interrupt in gsi_channel_start()

In gsi_channel_start() there is harmless-looking comment "Clear the
channel's event ring interrupt in case it's pending".  The intent
was to avoid getting spurious interrupts when first bringing up a
channel.

However we now use channel stop/start to implement suspend and
resume, and an interrupt pending at the time we resume is actually
something we don't want to ignore.

The very first time we bring up the channel we do not expect an
interrupt to be pending, and even if it were, the effect would
simply be to schedule NAPI on that channel, which would find nothing
to do, which is not a problem.

Stop clearing any pending IEOB interrupt in gsi_channel_start().
That leaves one caller of the trivial function gsi_isr_ieob_clear().
Get rid of that function and just open-code it in gsi_isr_ieob()
instead.

This fixes a problem where suspend/resume IPA v4.2 would get stuck
when resuming after a suspend.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: don't use noirq suspend/resume callbacks
Alex Elder [Fri, 15 May 2020 20:07:30 +0000 (15:07 -0500)]
net: ipa: don't use noirq suspend/resume callbacks

Use the suspend and resume callbacks rather than suspend_noirq and
resume_noirq.  With IPA v4.2, we use the CHANNEL_STOP command to
implement a suspend, and without interrupts enabled, that command
won't complete.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Reorganize-trap-data'
David S. Miller [Sat, 16 May 2020 23:42:32 +0000 (16:42 -0700)]
Merge branch 'mlxsw-Reorganize-trap-data'

Ido Schimmel says:

====================
mlxsw: Reorganize trap data

This patch set does not include any functional changes. It merely
reworks the internal storage of traps, trap groups and trap policers in
mlxsw to each use a single array.

These changes allow us to get rid of the multiple arrays we currently
have for traps, which make the trap data easier to validate and extend
with more per-trap information in the future. It will also allow us to
more easily add per-ASIC traps in future submissions.

Last two patches include minor changes to devlink-trap selftests.

Tested with existing devlink-trap selftests.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: mlxsw: Do not hard code trap group name
Ido Schimmel [Sat, 16 May 2020 22:43:10 +0000 (01:43 +0300)]
selftests: mlxsw: Do not hard code trap group name

It can be derived dynamically from the trap's name, so drop it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: devlink_lib: Remove double blank line
Ido Schimmel [Sat, 16 May 2020 22:43:09 +0000 (01:43 +0300)]
selftests: devlink_lib: Remove double blank line

One blank line is enough.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_trap: Store all trap data in one array
Ido Schimmel [Sat, 16 May 2020 22:43:08 +0000 (01:43 +0300)]
mlxsw: spectrum_trap: Store all trap data in one array

Each trap registered with devlink is mapped to one or more Rx listeners.
These listeners allow the switch driver (e.g., mlxsw_spectrum) to
register a function that is called when a packet is received (trapped)
for a specific reason.

Currently, three arrays are used to describe the mapping between the
logical devlink traps and the Rx listeners.

Instead, get rid of these arrays and store all the information in one
array that is easier to validate and extend with more per-trap
information.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_trap: Store all trap group data in one array
Ido Schimmel [Sat, 16 May 2020 22:43:07 +0000 (01:43 +0300)]
mlxsw: spectrum_trap: Store all trap group data in one array

Use one array to store all the information about all the trap groups
instead of hard coding it in code. This will be used in future patches
to disable certain functionality (e.g., policer binding) on a trap group
basis.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_trap: Store all trap policer data in one array
Ido Schimmel [Sat, 16 May 2020 22:43:06 +0000 (01:43 +0300)]
mlxsw: spectrum_trap: Store all trap policer data in one array

Instead of maintaining an array of policers and a linked list, only
maintain an array.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_trap: Move struct definition out of header file
Ido Schimmel [Sat, 16 May 2020 22:43:05 +0000 (01:43 +0300)]
mlxsw: spectrum_trap: Move struct definition out of header file

'struct mlxsw_sp_trap_policer_item' is only used in one file, so move it
there.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: remove remaining call to mdiobus_unregister
Heiner Kallweit [Sat, 16 May 2020 22:05:08 +0000 (00:05 +0200)]
r8169: remove remaining call to mdiobus_unregister

After having switched to devm_mdiobus_register() also this remaining
call to mdiobus_unregister() can be removed.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'ethtool-set_channels-add-a-few-more-checks'
David S. Miller [Sat, 16 May 2020 20:56:30 +0000 (13:56 -0700)]
Merge branch 'ethtool-set_channels-add-a-few-more-checks'

Jakub Kicinski says:

====================
ethtool: set_channels: add a few more checks

There seems to be a few more things we can check in the core before
we call drivers' ethtool_ops->set_channels. Adding the checks to
the core simplifies the drivers. This set only includes changes
to the NFP driver as an example.

There is a small risk in the first patch that someone actually
purposefully accepts a strange configuration without RX or TX
channels, but I couldn't find such a driver in the tree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethtool: don't call set_channels in drivers if config didn't change
Jakub Kicinski [Fri, 15 May 2020 19:49:02 +0000 (12:49 -0700)]
ethtool: don't call set_channels in drivers if config didn't change

Don't call drivers if nothing changed. Netlink code already
contains this logic.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: don't check lack of RX/TX channels
Jakub Kicinski [Fri, 15 May 2020 19:49:01 +0000 (12:49 -0700)]
nfp: don't check lack of RX/TX channels

Core will now perform this check.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethtool: check if there is at least one channel for TX/RX in the core
Jakub Kicinski [Fri, 15 May 2020 19:49:00 +0000 (12:49 -0700)]
ethtool: check if there is at least one channel for TX/RX in the core

Having a channel config with no ability to RX or TX traffic is
clearly wrong. Check for this in the core so the drivers don't
have to.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: Use 32-bit DATA_ACK when possible
Christoph Paasch [Thu, 14 May 2020 15:53:03 +0000 (08:53 -0700)]
mptcp: Use 32-bit DATA_ACK when possible

RFC8684 allows to send 32-bit DATA_ACKs as long as the peer is not
sending 64-bit data-sequence numbers. The 64-bit DSN is only there for
extreme scenarios when a very high throughput subflow is combined with a
long-RTT subflow such that the high-throughput subflow wraps around the
32-bit sequence number space within an RTT of the high-RTT subflow.

It is thus a rare scenario and we should try to use the 32-bit DATA_ACK
instead as long as possible. It allows to reduce the TCP-option overhead
by 4 bytes, thus makes space for an additional SACK-block. It also makes
tcpdumps much easier to read when the DSN and DATA_ACK are both either
32 or 64-bit.

Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetns: enable to inherit devconf from current netns
Nicolas Dichtel [Wed, 13 May 2020 13:58:43 +0000 (15:58 +0200)]
netns: enable to inherit devconf from current netns

The goal is to be able to inherit the initial devconf parameters from the
current netns, ie the netns where this new netns has been created.

This is useful in a containers environment where /proc/sys is read only.
For example, if a pod is created with specifics devconf parameters and has
the capability to create netns, the user expects to get the same parameters
than his 'init_net', which is not the real init_net in this case.

Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodpaa2-eth: add bulking to XDP_TX
Ioana Ciornei [Wed, 13 May 2020 13:55:46 +0000 (16:55 +0300)]
dpaa2-eth: add bulking to XDP_TX

Add driver level bulking to the XDP_TX action.

An array of frame descriptors is held for each Tx frame queue and
populated accordingly when the action returned by the XDP program is
XDP_TX. The frames will be actually enqueued only when the array is
filled. At the end of the NAPI cycle a flush on the queued frames is
performed in order to enqueue the remaining FDs.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: broadcom: fix checkpatch complains about tabs
Kevin Lo [Sat, 16 May 2020 02:09:26 +0000 (10:09 +0800)]
net: phy: broadcom: fix checkpatch complains about tabs

This patch makes checkpatch happy for tabs

Signed-off-by: Kevin Lo <kevlo@kevlo.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mlx5-updates-2020-05-15' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Fri, 15 May 2020 23:36:46 +0000 (16:36 -0700)]
Merge tag 'mlx5-updates-2020-05-15' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2020-05-15

mlx5 core and mlx5e (netdev) updates:

1) Two fixes for release all FW pages support.
2) Improvement in calculating the send queue stop room on tx
3) Flow steering auto-groups creation improvements
4) TC offload fix for Connection tracking with NAT action
5) IPoIB support for self looback to allow communication between ipoib
pkey child interfaces on the same host.
6) DCBNL cleanup to avoid #ifdef DCBNL all over the main mlx5e code
7) Small and trivial code cleanup
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethernet: ti: am65-cpts: Add missing inline qualifier to stub functions
Nathan Chancellor [Fri, 15 May 2020 22:33:18 +0000 (15:33 -0700)]
ethernet: ti: am65-cpts: Add missing inline qualifier to stub functions

When building with Clang:

In file included from drivers/net/ethernet/ti/am65-cpsw-ethtool.c:15:
drivers/net/ethernet/ti/am65-cpts.h:58:12: warning: unused function
'am65_cpts_ns_gettime' [-Wunused-function]
static s64 am65_cpts_ns_gettime(struct am65_cpts *cpts)
           ^
drivers/net/ethernet/ti/am65-cpts.h:63:12: warning: unused function
'am65_cpts_estf_enable' [-Wunused-function]
static int am65_cpts_estf_enable(struct am65_cpts *cpts,
           ^
drivers/net/ethernet/ti/am65-cpts.h:69:13: warning: unused function
'am65_cpts_estf_disable' [-Wunused-function]
static void am65_cpts_estf_disable(struct am65_cpts *cpts, int idx)
            ^
3 warnings generated.

These functions need to be marked as inline, which adds __maybe_unused,
to avoid these warnings, which is the pattern for stub functions.

Fixes: ec008fa2a9e5 ("ethernet: ti: am65-cpts: add routines to support taprio offload")
Link: https://github.com/ClangBuiltLinux/linux/issues/1026
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5e: Take DCBNL-related definitions into dedicated files
Tariq Toukan [Wed, 23 Oct 2019 12:00:52 +0000 (15:00 +0300)]
net/mlx5e: Take DCBNL-related definitions into dedicated files

Take DCBNL-related definitions out of the common en.h header,
Use a dedicated header file for exposing them.
Some need not to be exposed, use them locally in the .c file.
Use stubs to eliminate use of CONFIG_MLX5_CORE_EN_DCB in the
generic control flows.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Calculate SQ stop room in a robust way
Maxim Mikityanskiy [Mon, 30 Mar 2020 13:28:34 +0000 (16:28 +0300)]
net/mlx5e: Calculate SQ stop room in a robust way

Currently, different formulas are used to estimate the space that may be
taken by WQEs in the SQ during a single packet transmit. This space is
called stop room, and it's checked in the end of packet transmit to find
out if the next packet could overflow the SQ. If it could, the driver
tells the kernel to stop sending next packets.

Many factors affect the stop room:

1. Padding with NOPs to avoid WQEs spanning over page boundaries.

2. Enabled and disabled offloads (TLS, upcoming MPWQE).

3. The maximum size of a WQE.

The padding is performed before every WQE if it doesn't fit the current
page.

The current formula assumes that only one padding will be required per
packet, and it doesn't take into account that the WQEs posted during the
transmission of a single packet might exceed the page size in very rare
circumstances. For example, to hit this condition with 4096-byte pages,
TLS offload will have to interrupt an almost-full MPWQE session, be in
the resync flow and try to transmit a near to maximum amount of data.

To avoid SQ overflows in such rare cases after MPWQE is added, this
patch introduces a more robust formula to estimate the stop room. The
new formula uses the fact that a WQE of size X will not require more
than X-1 WQEBBs of padding. More exact estimations are possible, but
they result in much more complex and error-prone code for little gain.

Before this patch, the TLS stop room included space for both INNOVA and
ConnectX TLS offloads that couldn't run at the same time anyway, so this
patch accounts only for the active one.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: IPoIB, Drop multicast packets that this interface sent
Erez Shitrit [Mon, 4 May 2020 08:46:25 +0000 (11:46 +0300)]
net/mlx5e: IPoIB, Drop multicast packets that this interface sent

After enabled loopback packets for IPoIB, we need to drop these packets
that this HCA has replicated and came back to the same interface that
sent them.

Fixes: 4c6c615e3f30 ("net/mlx5e: IPoIB, Add PKEY child interface nic profile")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: IPoIB, Enable loopback packets for IPoIB interfaces
Erez Shitrit [Sun, 3 May 2020 10:01:37 +0000 (13:01 +0300)]
net/mlx5e: IPoIB, Enable loopback packets for IPoIB interfaces

Enable loopback of unicast and multicast traffic for IPoIB enhanced
mode.
This will allow interfaces with the same pkey to communicate between
them e.g cloned interfaces that located in different namespaces.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: CT: Fix offload with CT action after CT NAT action
Roi Dayan [Sun, 12 Apr 2020 12:39:15 +0000 (15:39 +0300)]
net/mlx5e: CT: Fix offload with CT action after CT NAT action

It could be a chain of rules will do action CT again after CT NAT
Before this fix matching will break as we get into the CT table
after NAT changes and not CT NAT.
Fix this by adding pre ct and pre ct nat tables to skip ct/ct_nat
tables and go straight to post_ct table if ct/nat was already done.

Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Move internal timer read function to clock library
Eran Ben Elisha [Thu, 7 May 2020 11:13:33 +0000 (14:13 +0300)]
net/mlx5: Move internal timer read function to clock library

Move mlx5_read_internal_timer() into lib/clock.c file as it is being
used there. As such, make this function a static one.

In addition, rearrange headers include to support function move.

Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: Aya Levin <ayal@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Wait for inactive autogroups
Paul Blakey [Thu, 7 May 2020 09:01:39 +0000 (12:01 +0300)]
net/mlx5: Wait for inactive autogroups

Currently, if one thread tries to add an entry to an autogrouped table
with no free matching group, while another thread is in the process of
creating a new matching autogroup, it doesn't wait for the new group
creation, and creates an unnecessary new autogroup.

Instead of skipping inactive, wait on the write lock of those groups.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Drain wq first during PCI device removal
Parav Pandit [Fri, 1 May 2020 14:42:45 +0000 (09:42 -0500)]
net/mlx5: Drain wq first during PCI device removal

mlx5_unload_one() is done with cleanup = true only once.

So instead of doing health wq drain inside the if(), directly do
during PCI device removal.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Have single error unwinding path
Parav Pandit [Fri, 1 May 2020 15:20:01 +0000 (10:20 -0500)]
net/mlx5: Have single error unwinding path

Having multiple error unwinding path are error prone.
Lets have just one error unwinding path.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix a bug of releasing wrong chunks on > 4K page size systems
Eran Ben Elisha [Sun, 3 May 2020 08:01:39 +0000 (11:01 +0300)]
net/mlx5: Fix a bug of releasing wrong chunks on > 4K page size systems

On systems with page size larger than 4K, a fwp object has few 4K chunks.
Fix a bug in fwp free flow where the chunk address was dropped and
fwp->addr was used instead (first chunk address). This caused a wrong
update of fwp->bitmask which later can cause errors in re-alloc fwp
chunk flow.

In order to fix this it, re-factor the release flow:
- Free 4k: Releases a specific 4k chunk inside the fwp, defined by
  starting address.
- Free fwp: Unconditionally release the whole fwp and its resources.
Free addr will call free fwp if all chunks were released, in order to do
code sharing.

In addition, fix npages to count for all released chunks correctly.

Fixes: c6168161f693 ("net/mlx5: Add support for release all pages event")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Dedicate fw page to the requesting function
Eran Ben Elisha [Sun, 3 May 2020 07:15:58 +0000 (10:15 +0300)]
net/mlx5: Dedicate fw page to the requesting function

The cited patch assumes that all chuncks in a fw page belong to the same
function, thus the driver must dedicate fw page to the requesting
function, which is actually what was intedned in the original fw pages
allocator design, hence the fwp->func_id !

Up until the cited patch everything worked ok, but now "relase all pages"
is broken on systems with page_size > 4k.

Fix this by dedicating fw page to the requesting function id via adding a
func_id parameter to alloc_4k() function.

Fixes: c6168161f693 ("net/mlx5: Add support for release all pages event")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
David S. Miller [Fri, 15 May 2020 20:48:59 +0000 (13:48 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Move the bpf verifier trace check into the new switch statement in
HEAD.

Resolve the overlapping changes in hinic, where bug fixes overlap
the addition of VF support.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Fri, 15 May 2020 20:10:06 +0000 (13:10 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix sk_psock reference count leak on receive, from Xiyu Yang.

 2) CONFIG_HNS should be invisible, from Geert Uytterhoeven.

 3) Don't allow locking route MTUs in ipv6, RFCs actually forbid this,
    from Maciej Żenczykowski.

 4) ipv4 route redirect backoff wasn't actually enforced, from Paolo
    Abeni.

 5) Fix netprio cgroup v2 leak, from Zefan Li.

 6) Fix infinite loop on rmmod in conntrack, from Florian Westphal.

 7) Fix tcp SO_RCVLOWAT hangs, from Eric Dumazet.

 8) Various bpf probe handling fixes, from Daniel Borkmann.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (68 commits)
  selftests: mptcp: pm: rm the right tmp file
  dpaa2-eth: properly handle buffer size restrictions
  bpf: Restrict bpf_trace_printk()'s %s usage and add %pks, %pus specifier
  bpf: Add bpf_probe_read_{user, kernel}_str() to do_refine_retval_range
  bpf: Restrict bpf_probe_read{, str}() only to archs where they work
  MAINTAINERS: Mark networking drivers as Maintained.
  ipmr: Add lockdep expression to ipmr_for_each_table macro
  ipmr: Fix RCU list debugging warning
  drivers: net: hamradio: Fix suspicious RCU usage warning in bpqether.c
  net: phy: broadcom: fix BCM54XX_SHD_SCR3_TRDDAPD value for BCM54810
  tcp: fix error recovery in tcp_zerocopy_receive()
  MAINTAINERS: Add Jakub to networking drivers.
  MAINTAINERS: another add of Karsten Graul for S390 networking
  drivers: ipa: fix typos for ipa_smp2p structure doc
  pppoe: only process PADT targeted at local interfaces
  selftests/bpf: Enforce returning 0 for fentry/fexit programs
  bpf: Enforce returning 0 for fentry/fexit progs
  net: stmmac: fix num_por initialization
  security: Fix the default value of secid_to_secctx hook
  libbpf: Fix register naming in PT_REGS s390 macros
  ...

4 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Linus Torvalds [Fri, 15 May 2020 20:06:56 +0000 (13:06 -0700)]
Merge tag 'for-linus' of git://git./linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "A few minor bug fixes for user visible defects, and one regression:

   - Various bugs from static checkers and syzkaller

   - Add missing error checking in mlx4

   - Prevent RTNL lock recursion in i40iw

   - Fix segfault in cxgb4 in peer abort cases

   - Fix a regression added in 5.7 where the IB_EVENT_DEVICE_FATAL could
     be lost, and wasn't delivered to all the FDs"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/uverbs: Move IB_EVENT_DEVICE_FATAL to destroy_uobj
  RDMA/uverbs: Do not discard the IB_EVENT_DEVICE_FATAL event
  RDMA/iw_cxgb4: Fix incorrect function parameters
  RDMA/core: Fix double put of resource
  IB/core: Fix potential NULL pointer dereference in pkey cache
  IB/hfi1: Fix another case where pq is left on waitlist
  IB/i40iw: Remove bogus call to netdev_master_upper_dev_get()
  IB/mlx4: Test return value of calls to ib_get_cached_pkey
  RDMA/rxe: Always return ERR_PTR from rxe_create_mmap_info()
  i40iw: Fix error handling in i40iw_manage_arp_cache()

4 years agoMerge tag 'linux-kselftest-5.7-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 15 May 2020 19:57:50 +0000 (12:57 -0700)]
Merge tag 'linux-kselftest-5.7-rc6' of git://git./linux/kernel/git/shuah/linux-kselftest

Pull kselftest fixes from Shuah Khan:

 - lkdtm runner fixes to prevent dmesg clearing and shellcheck errors

 - ftrace test handling when test module doesn't exist

 - nsfs test fix to replace zero-length array with flexible-array

 - dmabuf-heaps test fix to return clear error value

* tag 'linux-kselftest-5.7-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests/lkdtm: Use grep -E instead of egrep
  selftests/lkdtm: Don't clear dmesg when running tests
  selftests/ftrace: mark irqsoff_tracer.tc test as unresolved if the test module does not exist
  tools/testing: Replace zero-length array with flexible-array
  kselftests: dmabuf-heaps: Fix confused return value on expected error testing

4 years agoMerge tag 'riscv-for-linus-5.7-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 15 May 2020 19:47:15 +0000 (12:47 -0700)]
Merge tag 'riscv-for-linus-5.7-rc6' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Palmer Dabbelt:
 "A handful of build fixes, all found by Huawei's autobuilder.

  None of these patches should have any functional impact on kernels
  that build, and they're mostly related to various features
  intermingling with !MMU.

  While some of these might be better hoisted to generic code, it seems
  better to have the simple fixes in the meanwhile.

  As far as I know these are the only outstanding patches for 5.7"

* tag 'riscv-for-linus-5.7-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: mmiowb: Fix implicit declaration of function 'smp_processor_id'
  riscv: pgtable: Fix __kernel_map_pages build error if NOMMU
  riscv: Make SYS_SUPPORTS_HUGETLBFS depends on MMU
  riscv: Disable ARCH_HAS_DEBUG_VIRTUAL if NOMMU
  riscv: Add pgprot_writecombine/device and PAGE_SHARED defination if NOMMU
  riscv: stacktrace: Fix undefined reference to `walk_stackframe'
  riscv: Fix unmet direct dependencies built based on SOC_VIRT
  riscv: perf: RISCV_BASE_PMU should be independent
  riscv: perf_event: Make some funciton static

4 years agoMerge branch 'mptcp-fix-MP_JOIN-failure-handling'
David S. Miller [Fri, 15 May 2020 19:30:13 +0000 (12:30 -0700)]
Merge branch 'mptcp-fix-MP_JOIN-failure-handling'

Paolo Abeni says:

====================
mptcp: fix MP_JOIN failure handling

Currently if we hit an MP_JOIN failure on the third ack, the child socket is
closed with reset, but the request socket is not deleted, causing weird
behaviors.

The main problem is that MPTCP's MP_JOIN code needs to plug it's own
'valid 3rd ack' checks and the current TCP callbacks do not allow that.

This series tries to address the above shortcoming introducing a new MPTCP
specific bit in a 'struct tcp_request_sock' hole, and leveraging that to allow
tcp_check_req releasing the request socket when needed.

The above allows cleaning-up a bit current MPTCP hooking in tcp_check_req().

An alternative solution, possibly cleaner but more invasive, would be
changing the 'bool *own_req' syn_recv_sock() argument into 'int *req_status'
and let MPTCP set it to 'REQ_DROP'.

v1 -> v2:
 - be more conservative about drop_req initialization

RFC -> v1:
 - move the drop_req bit inside tcp_request_sock (Eric)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: cope better with MP_JOIN failure
Paolo Abeni [Fri, 15 May 2020 17:22:17 +0000 (19:22 +0200)]
mptcp: cope better with MP_JOIN failure

Currently, on MP_JOIN failure we reset the child
socket, but leave the request socket untouched.

tcp_check_req will deal with it according to the
'tcp_abort_on_overflow' sysctl value - by default the
req socket will stay alive.

The above leads to inconsistent behavior on MP JOIN
failure, and bad listener overflow accounting.

This patch addresses the issue leveraging the infrastructure
just introduced to ask the TCP stack to drop the req on
failure.

The child socket is not freed anymore by subflow_syn_recv_sock(),
instead it's moved to a dead state and will be disposed by the
next sock_put done by the TCP stack, so that listener overflow
accounting is not affected by MP JOIN failure.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoinet_connection_sock: factor out destroy helper.
Paolo Abeni [Fri, 15 May 2020 17:22:16 +0000 (19:22 +0200)]
inet_connection_sock: factor out destroy helper.

Move the steps to prepare an inet_connection_sock for
forced disposal inside a separate helper. No functional
changes inteded, this will just simplify the next patch.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: add new sock flag to deal with join subflows
Paolo Abeni [Fri, 15 May 2020 17:22:15 +0000 (19:22 +0200)]
mptcp: add new sock flag to deal with join subflows

MP_JOIN subflows must not land into the accept queue.
Currently tcp_check_req() calls an mptcp specific helper
to detect such scenario.

Such helper leverages the subflow context to check for
MP_JOIN subflows. We need to deal also with MP JOIN
failures, even when the subflow context is not available
due allocation failure.

A possible solution would be changing the syn_recv_sock()
signature to allow returning a more descriptive action/
error code and deal with that in tcp_check_req().

Since the above need is MPTCP specific, this patch instead
uses a TCP request socket hole to add a MPTCP specific flag.
Such flag is used by the MPTCP syn_recv_sock() to tell
tcp_check_req() how to deal with the request socket.

This change is a no-op for !MPTCP build, and makes the
MPTCP code simpler. It allows also the next patch to deal
correctly with MP JOIN failure.

v1 -> v2:
 - be more conservative on drop_req initialization (Mat)

RFC -> v1:
 - move the drop_req bit inside tcp_request_sock (Eric)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 15 May 2020 18:08:46 +0000 (11:08 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fix from Catalin Marinas:
 "Fix flush_icache_range() second argument in machine_kexec() to be an
  address rather than size"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: fix the flush_icache_range arguments in machine_kexec

4 years agonet: phy: tja11xx: execute cable test on link up
Oleksij Rempel [Thu, 14 May 2020 19:42:18 +0000 (21:42 +0200)]
net: phy: tja11xx: execute cable test on link up

A typical 100Base-T1 link should be always connected. If the link is in
a shot or open state, it is a failure. In most cases, we won't be able
to automatically handle this issue, but we need to log it or notify user
(if possible).

With this patch, the cable will be tested on "ip l s dev .. up" attempt
and send ethnl notification to the user space.

This patch was tested with TJA1102 PHY and "ethtool --monitor" command.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Fri, 15 May 2020 17:57:21 +0000 (10:57 -0700)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Alexei Starovoitov says:

====================
pull-request: bpf 2020-05-15

The following pull-request contains BPF updates for your *net* tree.

We've added 9 non-merge commits during the last 2 day(s) which contain
a total of 14 files changed, 137 insertions(+), 43 deletions(-).

The main changes are:

1) Fix secid_to_secctx LSM hook default value, from Anders.

2) Fix bug in mmap of bpf array, from Andrii.

3) Restrict bpf_probe_read to archs where they work, from Daniel.

4) Enforce returning 0 for fentry/fexit progs, from Yonghong.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: broadcom: add support for BCM54811 PHY
Kevin Lo [Fri, 15 May 2020 17:24:47 +0000 (01:24 +0800)]
net: phy: broadcom: add support for BCM54811 PHY

The BCM54811 PHY shares many similarities with the already supported BCM54810
PHY but additionally requires some semi-unique configuration.

Signed-off-by: Kevin Lo <kevlo@kevlo.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'cxgb4-improve-and-tune-TC-MQPRIO-offload'
David S. Miller [Fri, 15 May 2020 17:54:08 +0000 (10:54 -0700)]
Merge branch 'cxgb4-improve-and-tune-TC-MQPRIO-offload'

Rahul Lakkireddy says:

====================
cxgb4: improve and tune TC-MQPRIO offload

Patch 1 improves the Tx path's credit request and recovery mechanism
when running under heavy load.

Patch 2 adds ability to tune the burst buffer sizes of all traffic
classes to improve performance for <= 1500 MTU, under heavy load.

Patch 3 adds support to track EOTIDs and dump software queue
contexts used by TC-MQPRIO offload.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add EOTID tracking and software context dump
Rahul Lakkireddy [Fri, 15 May 2020 17:11:05 +0000 (22:41 +0530)]
cxgb4: add EOTID tracking and software context dump

Rework and add support for dumping EOTID software context used by
TC-MQPRIO. Also track number of EOTIDs in use.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: tune burst buffer size for TC-MQPRIO offload
Rahul Lakkireddy [Fri, 15 May 2020 17:11:04 +0000 (22:41 +0530)]
cxgb4: tune burst buffer size for TC-MQPRIO offload

For each traffic class, firmware handles up to 4 * MTU amount of data
per burst cycle. Under heavy load, this small buffer size is a
bottleneck when buffering large TSO packets in <= 1500 MTU case.
Increase the burst buffer size to 8 * MTU when supported.

Also, keep the driver's traffic class configuration API similar to
the firmware API counterpart.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: improve credits recovery in TC-MQPRIO Tx path
Rahul Lakkireddy [Fri, 15 May 2020 17:11:03 +0000 (22:41 +0530)]
cxgb4: improve credits recovery in TC-MQPRIO Tx path

Request credit update for every half credits consumed, including
the current request. Also, avoid re-trying to post packets when there
are no credits left. The credit update reply via interrupt will
eventually restore the credits and will invoke the Tx path again.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Fri, 15 May 2020 17:43:52 +0000 (10:43 -0700)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2020-05-15

The following pull-request contains BPF updates for your *net-next* tree.

We've added 37 non-merge commits during the last 1 day(s) which contain
a total of 67 files changed, 741 insertions(+), 252 deletions(-).

The main changes are:

1) bpf_xdp_adjust_tail() now allows to grow the tail as well, from Jesper.

2) bpftool can probe CONFIG_HZ, from Daniel.

3) CAP_BPF is introduced to isolate user processes that use BPF infra and
   to secure BPF networking services by dropping CAP_SYS_ADMIN requirement
   in certain cases, from Alexei.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: mt7530: fix VLAN setup
DENG Qingfang [Fri, 15 May 2020 15:25:55 +0000 (23:25 +0800)]
net: dsa: mt7530: fix VLAN setup

Allow DSA to add VLAN entries even if VLAN filtering is disabled, so
enabling it will not block the traffic of existent ports in the bridge

Signed-off-by: DENG Qingfang <dqfext@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Implement-classifier-action-terse-dump-mode'
David S. Miller [Fri, 15 May 2020 17:23:11 +0000 (10:23 -0700)]
Merge branch 'Implement-classifier-action-terse-dump-mode'

Vlad Buslov says:

====================
Implement classifier-action terse dump mode

Output rate of current upstream kernel TC filter dump implementation if
relatively low (~100k rules/sec depending on configuration). This
constraint impacts performance of software switch implementation that
rely on TC for their datapath implementation and periodically call TC
filter dump to update rules stats. Moreover, TC filter dump output a lot
of static data that don't change during the filter lifecycle (filter
key, specific action details, etc.) which constitutes significant
portion of payload on resulting netlink packets and increases amount of
syscalls necessary to dump all filters on particular Qdisc. In order to
significantly improve filter dump rate this patch sets implement new
mode of TC filter dump operation named "terse dump" mode. In this mode
only parameters necessary to identify the filter (handle, action cookie,
etc.) and data that can change during filter lifecycle (filter flags,
action stats, etc.) are preserved in dump output while everything else
is omitted.

Userspace API is implemented using new TCA_DUMP_FLAGS tlv with only
available flag value TCA_DUMP_FLAGS_TERSE. Internally, new API requires
individual classifier support (new tcf_proto_ops->terse_dump()
callback). Support for action terse dump is implemented in act API and
don't require changing individual action implementations.

The following table provides performance comparison between regular
filter dump and new terse dump mode for two classifier-action profiles:
one minimal config with L2 flower classifier and single gact action and
another heavier config with L2+5tuple flower classifier with
tunnel_key+mirred actions.

 Classifier-action type      |        dump |  terse dump | X improvement
                             | (rules/sec) | (rules/sec) |
-----------------------------+-------------+-------------+---------------
 L2 with gact                |       141.8 |       293.2 |          2.07
 L2+5tuple tunnel_key+mirred |        76.4 |       198.8 |          2.60

Benchmark details: to measure the rate tc filter dump and terse dump
commands are invoked on ingress Qdisc that have one million filters
configured using following commands.

> time sudo tc -s filter show dev ens1f0 ingress >/dev/null

> time sudo tc -s filter show terse dev ens1f0 ingress >/dev/null

Value in results table is calculated by dividing 1000000 total rules by
"real" time reported by time command.

Setup details: 2x Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.40GHz, 32GB memory
====================

Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>