platform/kernel/linux-rpi.git
5 years agonet: phy: add macros for PHYID matching
Heiner Kallweit [Fri, 9 Nov 2018 23:39:14 +0000 (00:39 +0100)]
net: phy: add macros for PHYID matching

Add macros for PHYID matching to be used in PHY driver configs.
By using these macros some boilerplate code can be avoided.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'phylib-simplifications'
David S. Miller [Sun, 11 Nov 2018 17:41:32 +0000 (09:41 -0800)]
Merge branch 'phylib-simplifications'

Heiner Kallweit says:

====================
net: phy: further phylib simplifications after recent changes to the state machine

After the recent changes to the state machine phylib can be further
simplified (w/o having to make any assumptions).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: improve and inline phy_change
Heiner Kallweit [Fri, 9 Nov 2018 17:58:01 +0000 (18:58 +0100)]
net: phy: improve and inline phy_change

Now that phy_mac_interrupt() doesn't call phy_change() any longer it's
called from phy_interrupt() only. Therefore phy_interrupt_is_valid()
returns true always and the check can be removed.
In case of PHY_HALTED phy_interrupt() bails out immediately,
therefore the second check for PHY_HALTED including the call to
phy_disable_interrupts() can be removed.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: simplify phy_mac_interrupt and related functions
Heiner Kallweit [Fri, 9 Nov 2018 17:56:52 +0000 (18:56 +0100)]
net: phy: simplify phy_mac_interrupt and related functions

When using phy_mac_interrupt() the irq number is set to
PHY_IGNORE_INTERRUPT, therefore phy_interrupt_is_valid() returns false.
As a result phy_change() effectively just calls phy_trigger_machine()
when called from phy_mac_interrupt() via phy_change_work(). So we can
call phy_trigger_machine() from phy_mac_interrupt() directly and
remove some now unneeded code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: don't set state PHY_CHANGELINK in phy_change
Heiner Kallweit [Fri, 9 Nov 2018 17:55:50 +0000 (18:55 +0100)]
net: phy: don't set state PHY_CHANGELINK in phy_change

State PHY_CHANGELINK isn't needed here, we can call the state machine
directly. We just have to remove the check for phy_polling_mode() to
make this work also in interrupt mode. Removing this check doesn't
cause any overhead because when not polling the state machine is
called only if required by some event.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'remove-PHY_HAS_INTERRUPT'
David S. Miller [Sun, 11 Nov 2018 17:37:04 +0000 (09:37 -0800)]
Merge branch 'remove-PHY_HAS_INTERRUPT'

Heiner Kallweit says:

====================
net: phy: replace PHY_HAS_INTERRUPT with a check for config_intr and ack_interrupt

Flag PHY_HAS_INTERRUPT is used only here for this small check. I think
using interrupts isn't possible if a driver defines neither
config_intr nor ack_interrupts callback. So we can replace checking
flag PHY_HAS_INTERRUPT with checking for these callbacks.
This allows to remove this flag from all driver configs.

v2:
- add helper for check in patch 1
- remove PHY_HAS_INTERRUPT from all drivers, not only Realtek
- remove flag PHY_HAS_INTERRUPT completely

v3:
- rebase patch 2
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: remove flag PHY_HAS_INTERRUPT from driver configs
Heiner Kallweit [Fri, 9 Nov 2018 17:17:22 +0000 (18:17 +0100)]
net: phy: remove flag PHY_HAS_INTERRUPT from driver configs

Now that flag PHY_HAS_INTERRUPT has been replaced with a check for
callbacks config_intr and ack_interrupt, we can remove setting this
flag from all driver configs.
Last but not least remove flag PHY_HAS_INTERRUPT completely.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: replace PHY_HAS_INTERRUPT with a check for config_intr and ack_interrupt
Heiner Kallweit [Fri, 9 Nov 2018 17:16:28 +0000 (18:16 +0100)]
net: phy: replace PHY_HAS_INTERRUPT with a check for config_intr and ack_interrupt

Flag PHY_HAS_INTERRUPT is used only here for this small check. I think
using interrupts isn't possible if a driver defines neither
config_intr nor ack_interrupts callback. So we can replace checking
flag PHY_HAS_INTERRUPT with checking for these callbacks.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: Fix SKB list traversal in sctp_intl_store_ordered().
David S. Miller [Sun, 11 Nov 2018 03:32:23 +0000 (19:32 -0800)]
sctp: Fix SKB list traversal in sctp_intl_store_ordered().

Same change as made to sctp_intl_store_reasm().

To be fully correct, an iterator has an undefined value when something
like skb_queue_walk() naturally terminates.

This will actually matter when SKB queues are converted over to
list_head.

Formalize what this code ends up doing with the current
implementation.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: Fix SKB list traversal in sctp_intl_store_reasm().
David S. Miller [Sun, 11 Nov 2018 03:28:27 +0000 (19:28 -0800)]
sctp: Fix SKB list traversal in sctp_intl_store_reasm().

To be fully correct, an iterator has an undefined value when something
like skb_queue_walk() naturally terminates.

This will actually matter when SKB queues are converted over to
list_head.

Formalize what this code ends up doing with the current
implementation.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoiucv: Remove SKB list assumptions.
David S. Miller [Thu, 23 Aug 2018 00:01:51 +0000 (17:01 -0700)]
iucv: Remove SKB list assumptions.

Eliminate the assumption that SKBs and SKB list heads can
be cast to eachother in SKB list handling code.

This change also appears to fix a bug since the list->next pointer is
sampled outside of holding the SKB queue lock.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobrcmfmac: Use standard SKB list accessors in brcmf_sdiod_sglist_rw.
David S. Miller [Sun, 12 Aug 2018 04:14:34 +0000 (21:14 -0700)]
brcmfmac: Use standard SKB list accessors in brcmf_sdiod_sglist_rw.

Instead of direct SKB list pointer accesses.

The loops in this function had to be rewritten to accommodate this
more easily.

The first loop iterates now over the target list in the outer loop,
and triggers an mmc data operation when the per-operation limits are
hit.

Then after the loops, if we have any residue, we trigger the last
and final operation.

For the page aligned workaround, where we have to copy the read data
back into the original list of SKBs, we use a two-tiered loop.  The
outer loop stays the same and iterates over pktlist, and then we have
an inner loop which uses skb_peek_next().  The break logic has been
simplified because we know that the aggregate length of the SKBs in
the source and destination lists are the same.

This change also ends up fixing a bug, having to do with the
maintainance of the seg_sz variable and how it drove the outermost
loop.  It begins as:

seg_sz = target_list->qlen;

ie. the number of packets in the target_list queue.  The loop
structure was then:

while (seq_sz) {
...
while (not at end of target_list) {
...
sg_cnt++
...
}
...
seg_sz -= sg_cnt;

The assumption built into that last statement is that sg_cnt counts
how many packets from target_list have been fully processed by the
inner loop.  But this not true.

If we hit one of the limits, such as the max segment size or the max
request size, we will break and copy a partial packet then contine
back up to the top of the outermost loop.

With the new loops we don't have this problem as we don't guard the
loop exit with a packet count, but instead use the progression of the
pkt_next SKB through the list to the end.  The general structure is:

sg_cnt = 0;
skb_queue_walk(target_list, pkt_next) {
pkt_offset = 0;
...
sg_cnt++;
...
while (pkt_offset < pkt_next->len) {
pkt_offset += sg_data_size;
if (queued up max per request)
mmc_submit_one();
}
}
if (sg_cnt)
mmc_submit_one();

The variables that maintain where we are in the MMC command state such
as req_sz, sg_cnt, and sgl are reset when we emit one of these full
sized requests.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoOVS: remove VLAN_TAG_PRESENT - fixup
Michał Mirosław [Sat, 10 Nov 2018 18:55:34 +0000 (19:55 +0100)]
OVS: remove VLAN_TAG_PRESENT - fixup

It turns out I missed one VLAN_TAG_PRESENT in OVS code while rebasing.
This fixes it.

Fixes: 9df46aefafa6 ("OVS: remove use of VLAN_TAG_PRESENT")
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoinfiniband: nes: Fix more direct skb list accesses.
David S. Miller [Sat, 10 Nov 2018 05:18:48 +0000 (21:18 -0800)]
infiniband: nes: Fix more direct skb list accesses.

The following:

skb = skb->next;
...
if (skb == (struct sk_buff *)queue)

is transformed into:

skb = skb_peek_next(skb, queue);
...
if (!skb)

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: leds: Don't make our own link speed names
Kyle Roeschley [Fri, 9 Nov 2018 18:48:03 +0000 (12:48 -0600)]
net: phy: leds: Don't make our own link speed names

The phy core provides a handy phy_speed_to_str() helper, so use that
instead of doing our own formatting of the different known link speeds.
To do this, increase PHY_LED_TRIGGER_SPEED_SUFFIX_SIZE to 11 so we can fit
'Unsupported' if necessary.

Signed-off-by: Kyle Roeschley <kyle.roeschley@ni.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: improve struct phy_device member interrupts handling
Heiner Kallweit [Fri, 9 Nov 2018 17:35:52 +0000 (18:35 +0100)]
net: phy: improve struct phy_device member interrupts handling

As a heritage from the very early days of phylib member interrupts is
defined as u32 even though it's just a flag whether interrupts are
enabled. So we can change it to a bitfield member. In addition change
the code dealing with this member in a way that it's clear we're
dealing with a bool value.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.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>
5 years agoMerge branch 'dpaa2-eth-defer-probe-on-object-allocate'
David S. Miller [Sat, 10 Nov 2018 04:08:59 +0000 (20:08 -0800)]
Merge branch 'dpaa2-eth-defer-probe-on-object-allocate'

Ioana Ciornei says:

====================
dpaa2-eth: defer probe on object allocate

Allocatable objects on the fsl-mc bus may be probed by the fsl_mc_allocator
after the first attempts of other drivers to use them. Defer the probe when
this situation happens.

Changes in v2:
  - proper handling of IS_ERR_OR_NULL
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodpaa2-ptp: defer probe when portal allocation failed
Ioana Ciornei [Fri, 9 Nov 2018 15:26:46 +0000 (15:26 +0000)]
dpaa2-ptp: defer probe when portal allocation failed

The fsl_mc_portal_allocate can fail when the requested MC portals are
not yet probed by the fsl_mc_allocator. In this situation, the driver
should defer the probe.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodpaa2-eth: defer probe on object allocate
Ioana Ciornei [Fri, 9 Nov 2018 15:26:45 +0000 (15:26 +0000)]
dpaa2-eth: defer probe on object allocate

The fsl_mc_object_allocate function can fail because not all allocatable
objects are probed by the fsl_mc_allocator at the call time. Defer the
dpaa2-eth probe when this happens.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoudp6: cleanup stats accounting in recvmsg()
Paolo Abeni [Fri, 9 Nov 2018 14:52:45 +0000 (15:52 +0100)]
udp6: cleanup stats accounting in recvmsg()

In the udp6 code path, we needed multiple tests to select the correct
mib to be updated. Since we touch at least a counter at each iteration,
it's convenient to use the recently introduced __UDPX_MIB() helper once
and remove some code duplication.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: use the new __netdev_tx_sent_queue() BQL optimisation
Jakub Kicinski [Sat, 10 Nov 2018 02:50:00 +0000 (18:50 -0800)]
nfp: use the new __netdev_tx_sent_queue() BQL optimisation

__netdev_tx_sent_queue() was added in commit e59020abf0f
("net: bql: add __netdev_tx_sent_queue()") and allows for
better GSO performance.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ptp-more-accurate-PHC-system-clock-synchronization'
David S. Miller [Sat, 10 Nov 2018 03:43:51 +0000 (19:43 -0800)]
Merge branch 'ptp-more-accurate-PHC-system-clock-synchronization'

Miroslav Lichvar says:

====================
More accurate PHC<->system clock synchronization

RFC->v1:
- added new patches
- separated PHC timestamp from ptp_system_timestamp
- fixed memory leak in PTP_SYS_OFFSET_EXTENDED
- changed PTP_SYS_OFFSET_EXTENDED to work with array of arrays
- fixed PTP_SYS_OFFSET_EXTENDED to break correctly from loop
- fixed timecounter updates in drivers
- split gettimex in igb driver
- fixed ptp_read_* functions to be available without
  CONFIG_PTP_1588_CLOCK

This series enables a more accurate synchronization between PTP hardware
clocks and the system clock.

The first two patches are minor cleanup/bug fixes.

The third patch adds an extended version of the PTP_SYS_OFFSET ioctl,
which returns three timestamps for each measurement. The idea is to
shorten the interval between the system timestamps to contain just the
reading of the lowest register of the PHC in order to reduce the error
in the measured offset and get a smaller upper bound on the maximum
error.

The fourth patch deprecates the original gettime function.

The remaining patches update the gettime function in order to support
the new ioctl in the e1000e, igb, ixgbe, and tg3 drivers.

Tests with few different NICs in different machines show that:
- with an I219 (e1000e) the measured delay was reduced from 2500 to 1300
  ns and the error in the measured offset, when compared to the cross
  timestamping supported by the driver, was reduced by a factor of 5
- with an I210 (igb) the delay was reduced from 5100 to 1700 ns
- with an I350 (igb) the delay was reduced from 2300 to 750 ns
- with an X550 (ixgbe) the delay was reduced from 1950 to 650 ns
- with a BCM5720 (tg3) the delay was reduced from 2400 to 1200 ns
====================

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotg3: extend PTP gettime function to read system clock
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:49 +0000 (11:14 +0100)]
tg3: extend PTP gettime function to read system clock

This adds support for the PTP_SYS_OFFSET_EXTENDED ioctl.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoixgbe: extend PTP gettime function to read system clock
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:48 +0000 (11:14 +0100)]
ixgbe: extend PTP gettime function to read system clock

This adds support for the PTP_SYS_OFFSET_EXTENDED ioctl.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoigb: extend PTP gettime function to read system clock
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:47 +0000 (11:14 +0100)]
igb: extend PTP gettime function to read system clock

This adds support for the PTP_SYS_OFFSET_EXTENDED ioctl.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoe1000e: extend PTP gettime function to read system clock
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:46 +0000 (11:14 +0100)]
e1000e: extend PTP gettime function to read system clock

This adds support for the PTP_SYS_OFFSET_EXTENDED ioctl.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoptp: deprecate gettime64() in favor of gettimex64()
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:45 +0000 (11:14 +0100)]
ptp: deprecate gettime64() in favor of gettimex64()

When a driver provides gettimex64(), use it in the PTP_SYS_OFFSET ioctl
and POSIX clock's gettime() instead of gettime64(). Drivers should
provide only one of the functions.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoptp: add PTP_SYS_OFFSET_EXTENDED ioctl
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:44 +0000 (11:14 +0100)]
ptp: add PTP_SYS_OFFSET_EXTENDED ioctl

The PTP_SYS_OFFSET ioctl, which can be used to measure the offset
between a PHC and the system clock, includes the total time that the
driver needs to read the PHC timestamp.

This typically involves reading of multiple PCI registers (sometimes in
multiple iterations) and the register that contains the lowest bits of
the timestamp is not read in the middle between the two readings of the
system clock. This asymmetry causes the measured offset to have a
significant error.

Introduce a new ioctl, driver function, and helper functions, which
allow the reading of the lowest register to be isolated from the other
readings in order to reduce the asymmetry. The ioctl returns three
timestamps for each measurement:
- system time right before reading the lowest bits of the PHC timestamp
- PHC time
- system time immediately after reading the lowest bits of the PHC
  timestamp

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoptp: check gettime64 return code in PTP_SYS_OFFSET ioctl
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:43 +0000 (11:14 +0100)]
ptp: check gettime64 return code in PTP_SYS_OFFSET ioctl

If a gettime64 call fails, return the error and avoid copying data back
to user.

Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoptp: reorder declarations in ptp_ioctl()
Miroslav Lichvar [Fri, 9 Nov 2018 10:14:42 +0000 (11:14 +0100)]
ptp: reorder declarations in ptp_ioctl()

Reorder declarations of variables as reversed Christmas tree.

Cc: Richard Cochran <richardcochran@gmail.com>
Suggested-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'hns3-add-code-optimization-for-VF-reset-and-some-new-reset-feature'
David S. Miller [Sat, 10 Nov 2018 00:47:35 +0000 (16:47 -0800)]
Merge branch 'hns3-add-code-optimization-for-VF-reset-and-some-new-reset-feature'

Huazhong Tan says:

====================
hns3: add code optimization for VF reset and some new reset feature

Currently hardware supports below reset:
1. VF reset: triggered by sending cmd to IMP(Integrated Management
   Processor). Only reset specific VF function and do not affect
   other PF or VF.
2. PF reset: triggered by sending cmd to IMP. Only reset specific PF
   and it's VF.
3. PF FLR: triggered by PCIe subsystem. Only reset specific PF and
   it's VF.
4. VF FLR: triggered by PCIe subsystem. Only reset specific VF function
   and do not affect other PF or VF.
5. Core reset: triggered by writing to register. Reset most hardware
   unit, such as SSU, which affects all the PF and VF.
6. Global reset: triggered by writing to register. Reset all hardware
   unit, which affects all the PF and VF.
7. IMP reset: triggered by IMU(Intelligent Management Unit) when
   IMP is not longer feeding IMU's watchdog. IMU will reload the IMP
   firmware and IMP will perform global reset after firmware reloading,
   which affects all the PF and VF.

Current driver only support PF/VF reset, incomplete core and global
reset(lacking the vf reset handling). So this patchset adds complete
reset support in hns3 driver.

Also, this patchset contains some optimization related to reset.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add PCIe FLR support for VF
Huazhong Tan [Fri, 9 Nov 2018 14:07:56 +0000 (22:07 +0800)]
net: hns3: add PCIe FLR support for VF

This patch implements the .reset_prepare and .reset_done
ops from pci framework to support the VF FLR.

This patch uses hclgevf_set_def_reset_request() and
hclgevf_reset_event() to handle FLR, so when
hdev->default_reset_request is non zero, it means there is
some reset requseted by hclgevf_set_def_reset_request() need
to be processed. Also get the hdev from the ae_dev because
hclgevf_reset_event is called with handle being NULL.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: do VF's pci re-initialization while PF doing FLR
Huazhong Tan [Fri, 9 Nov 2018 14:07:55 +0000 (22:07 +0800)]
net: hns3: do VF's pci re-initialization while PF doing FLR

While doing PF FLR, VF's PCIe configuration space will be cleared, so
the pci and vector of VF should be re-initialized in the VF's reset
process while PF doing FLR.

Also, this patch fixes some memory not freed problem when pci
re-initialization is done during reset process.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add PCIe FLR support for PF
Huazhong Tan [Fri, 9 Nov 2018 14:07:54 +0000 (22:07 +0800)]
net: hns3: add PCIe FLR support for PF

This patch implements the .reset_prepare and .reset_done
ops from pci framework to support the PF FLR.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: implement the IMP reset processing for PF
Huazhong Tan [Fri, 9 Nov 2018 14:07:53 +0000 (22:07 +0800)]
net: hns3: implement the IMP reset processing for PF

The current code only print the prompt message after receiving
the IMP reset interrupt and does not perform the corresponding driver
reset operation. This patch implements the missing IMP reset handling
in the driver.
1. The driver sets the HCLGE_STATE_CMD_DISABLE to stop sending command
   after receiving the IMP reset interrupt.
2. The driver needs to notify the hardware to reload the IMP firmware.
3. The IMP firmware reloading makes the reset time of hardware longer,
   so it is necessary to extend the driver's waiting time to wait for
   the hardware reset to complete.
4. In hclge_check_event_cause, IMP reset event should have higher
   priority than other events.

Also, after clearing HCLGE_STATE_CMD_DISABLE in the hclge_cmd_init(),
it needs to check whether there is a pending reset, if so, just set
the HCLGE_STATE_CMD_DISABLE back and return.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: stop napi polling when HNS3_NIC_STATE_DOWN is set
Huazhong Tan [Fri, 9 Nov 2018 14:07:52 +0000 (22:07 +0800)]
net: hns3: stop napi polling when HNS3_NIC_STATE_DOWN is set

When calling napi_disable during reset down process, if NAPIF_STATE_MISSED
is set, napi_complete will call __napi_schedule to do the polling again.
So this patch uses HNS3_NIC_STATE_DOWN to ensure the polling is not
scheduled again.

Also, when napi_complete returns true, it means polling is scheduled
again, it is not neccssary to enable the interrupt.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add error handler for hclgevf_reset()
Huazhong Tan [Fri, 9 Nov 2018 14:07:51 +0000 (22:07 +0800)]
net: hns3: add error handler for hclgevf_reset()

Since hclgevf_reset() may fail for some reasons, so it needs an error
handler to deal with it. When VF reset failed, VF can only be restored
by a higher level reset asserted by PF. So, it needs to reinitialize
its command queue, then it can respond to higher level reset.

Also, this patch adds error logging in the hclgevf_notify_client().

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: stop handling command queue while resetting VF
Huazhong Tan [Fri, 9 Nov 2018 14:07:50 +0000 (22:07 +0800)]
net: hns3: stop handling command queue while resetting VF

According to hardware's description, after the reset occurs, the driver
needs to re-initialize the command queue before sending and receiving
any commands. Therefore, the VF's driver needs to identify the command
queue needs to re-initialize with HCLGEVF_STATE_CMD_DISABLE, and does
not allow sending or receiving commands before the re-initialization.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add reset handling for VF when doing Core/Global/IMP reset
Huazhong Tan [Fri, 9 Nov 2018 14:07:49 +0000 (22:07 +0800)]
net: hns3: add reset handling for VF when doing Core/Global/IMP reset

When a Core/Global/IMP reset occurs, the hardware sets the reset status
register of all PF/VF and reports a reset interrupt to all PF/VF and
firmware.

When receiving the reset interrupt:
1. The firmware will wait for 100 ms before resetting the hardware and
   clear the reset status register of all PF when hardware reset is done.
2. The PF/VF driver needs to down the netdev within 100 ms and then wait
   for hardware reset to finish.
3. After firmware clearing the reset status register of all PF, the PF
   driver reinitializes the hardware and clear the reset status register
   of it's VF.
4. After PF driver clearing the reset status register of VF, the VF driver
   reinitializes the hardware.

This patch mainly add handling for the step 4.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add reset handling for VF when doing PF reset
Huazhong Tan [Fri, 9 Nov 2018 14:07:48 +0000 (22:07 +0800)]
net: hns3: add reset handling for VF when doing PF reset

When PF performs a function reset, the hardware will reset both PF
and all the VF belong to this PF. Hence, both PF's driver and VF's
driver need to perform corresponding reset operations.

Before PF driver asserting function reset to hardware, it firstly
set up VF's hardware reset status, and inform the VF driver with
HNAE3_VF_PF_FUNC_RESET, then VF driver sets this reset type to
reset_pending and shechule reset task to stop IO and waits for the
hardware reset status to clear. When PF driver has reinitialized the
hardware and is ready to process mailbox from VF, PF driver clears
VF's hardware reset status for VF to continue its reset process.

Also, this patch uses readl_poll_timeout to simplify the hardware reset
status waitting.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: adjust VF's reset process
Huazhong Tan [Fri, 9 Nov 2018 14:07:47 +0000 (22:07 +0800)]
net: hns3: adjust VF's reset process

Currently when VF need to reset itself, it will send a cmd to PF,
after receiving the VF reset requset, PF sends a cmd to inform
VF to enter the reset process and send a cmd to firmware to do the
actual reset for the VF, it is possible that firmware has resetted
the VF, but VF has not entered the reset process, which may cause
IO not stopped problem when firmware is resetting VF.

This patch fixes it by adjusting the VF reset process, when VF
need to reset itself, it will enter the reset process first, and
it will tell the PF to send cmd to firmware to reset itself.

Add member reset_pending to struct hclgevf_dev, which indicates that
there is reset event need to be processed by the VF's reset task, and
the VF's reset task chooses the highest-level one and clears other
low-level one when it processes reset_pending.

hclge_inform_reset_assert_to_vf function is unused now, but it will
be used to support the PF reset with VF working, so declare it in
the header file.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add reset_hdev to reinit the hdev in VF's reset process
Huazhong Tan [Fri, 9 Nov 2018 14:07:46 +0000 (22:07 +0800)]
net: hns3: add reset_hdev to reinit the hdev in VF's reset process

When doing reset, the reset handling function only need to
reinitialize hardware, it makes sense to add a function to
do that job. Also the error handling of hclgevf_init_hdev is
different when it is used in reset process.

This patch adds reset_hdev to reinitialize hardware when resetting.
Also, this patch removes the hclgevf_dev_ongoing_full_reset because
it is unused now.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4vf: free mac_hlist properly
Arjun Vynipadath [Fri, 9 Nov 2018 09:22:53 +0000 (14:52 +0530)]
cxgb4vf: free mac_hlist properly

The locally maintained list for tracking hash mac table was
not freed during driver remove.

Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4vf: fix memleak in mac_hlist initialization
Arjun Vynipadath [Fri, 9 Nov 2018 09:22:01 +0000 (14:52 +0530)]
cxgb4vf: fix memleak in mac_hlist initialization

mac_hlist was initialized during adapter_up, which will be called
every time a vf device is first brought up, or every time when device
is brought up again after bringing all devices down. This means our
state of previous list is lost, causing a memleak if entries are
present in the list. To fix that, move list init to the condition
that performs initial one time adapter setup.

Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: free mac_hlist properly
Arjun Vynipadath [Fri, 9 Nov 2018 09:20:25 +0000 (14:50 +0530)]
cxgb4: free mac_hlist properly

The locally maintained list for tracking hash mac table was
not freed during driver remove.

Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tcp: remove BUG_ON from tcp_v4_err
Li RongQing [Fri, 9 Nov 2018 09:04:51 +0000 (17:04 +0800)]
net: tcp: remove BUG_ON from tcp_v4_err

if skb is NULL pointer, and the following access of skb's
skb_mstamp_ns will trigger panic, which is same as BUG_ON

Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoxen/netfront: remove unnecessary wmb
Jacob Wen [Fri, 9 Nov 2018 06:53:59 +0000 (14:53 +0800)]
xen/netfront: remove unnecessary wmb

RING_PUSH_REQUESTS_AND_CHECK_NOTIFY is already able to make sure backend sees
requests before req_prod is updated.

Signed-off-by: Jacob Wen <jian.w.wen@oracle.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'nfp-abm-move-code-and-improve-parameter-validation'
David S. Miller [Fri, 9 Nov 2018 04:48:01 +0000 (20:48 -0800)]
Merge branch 'nfp-abm-move-code-and-improve-parameter-validation'

Jakub Kicinski says:

====================
nfp: abm: move code and improve parameter validation

This set starts by separating Qdisc handling code into a new file.
Next two patches allow early access to TLV-based capabilities during
probe, previously the capabilities were parsed just before netdevs
were registered, but its cleaner to do some basic validation earlier
and avoid cleanup work.

Next three patches improve RED's parameter validation.  First we provide
a more precise message about why offload failed (and move the parameter
validation to a helper).  Next we make sure we don't set the top bit
in the 32 bit max RED threshold value.  Because FW is treating the value
as signed it reportedly causes slow downs (unnecessary queuing and
marking) when top bit is set with recent firmwares.  Last (and perhaps
least importantly) we offload the harddrop parameter of the Qdisc.
We don't plan to offload harddrop RED, but it seems prudent to make
sure user didn't set that flag as device behaviour would have differed.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: abm: refuse RED offload with harddrop set
Jakub Kicinski [Fri, 9 Nov 2018 03:50:39 +0000 (19:50 -0800)]
nfp: abm: refuse RED offload with harddrop set

RED Qdisc will now inform the drivers about the state of the harddrop
flag.  Refuse to offload in case harddrop is set.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: red: inform offloads about harddrop setting
Jakub Kicinski [Fri, 9 Nov 2018 03:50:38 +0000 (19:50 -0800)]
net: sched: red: inform offloads about harddrop setting

To mirror software behaviour on offload more precisely inform
the drivers about the state of the harddrop flag.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: abm: don't set negative threshold
Jakub Kicinski [Fri, 9 Nov 2018 03:50:37 +0000 (19:50 -0800)]
nfp: abm: don't set negative threshold

Turns out the threshold value is used in signed compares in the FW,
so we should avoid setting the top bit.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: abm: provide more precise info about offload parameter validation
Jakub Kicinski [Fri, 9 Nov 2018 03:50:36 +0000 (19:50 -0800)]
nfp: abm: provide more precise info about offload parameter validation

Improve log messages printed when RED can't be offloaded because
of Qdisc parameters.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: parse vNIC TLV capabilities at alloc time
Jakub Kicinski [Fri, 9 Nov 2018 03:50:35 +0000 (19:50 -0800)]
nfp: parse vNIC TLV capabilities at alloc time

In certain cases initialization logic which follows allocation of
the vNIC structure may want to validate the capabilities of that vNIC.
This is easy before vNIC is initialized for normal capabilities which
are at fixed offsets in control memory, easy to locate and read, but
poses a challenge if the capabilities are in form of TLVs.  Parse
the TLVs early on so other code can just access parsed info, instead
of having to do the parsing by itself.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: pass ctrl_bar pointer to nfp_net_alloc
Jakub Kicinski [Fri, 9 Nov 2018 03:50:34 +0000 (19:50 -0800)]
nfp: pass ctrl_bar pointer to nfp_net_alloc

Move setting ctrl_bar pointer to the nfp_net_alloc function,
to make sure we can parse capabilities early in the following
patch.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: abm: split qdisc offload code into a separate file
Jakub Kicinski [Fri, 9 Nov 2018 03:50:33 +0000 (19:50 -0800)]
nfp: abm: split qdisc offload code into a separate file

The Qdisc offload code is logically separate, and we will soon
do significant surgery on it to support more Qdiscs, so move
it to a separate file.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotcp_bbr: update comments to reflect pacing_margin_percent
Neal Cardwell [Fri, 9 Nov 2018 02:54:00 +0000 (21:54 -0500)]
tcp_bbr: update comments to reflect pacing_margin_percent

Recently, in commit ab408b6dc744 ("tcp: switch tcp and sch_fq to new
earliest departure time model"), the TCP BBR code switched to a new
approach of using an explicit bbr_pacing_margin_percent for shaving a
pacing rate "haircut", rather than the previous implict
approach. Update an old comment to reflect the new approach.

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-Use-__vlan_hwaccel_-helpers'
David S. Miller [Fri, 9 Nov 2018 04:45:05 +0000 (20:45 -0800)]
Merge branch 'net-Use-__vlan_hwaccel_-helpers'

Michał Mirosław says:

====================
net: Use __vlan_hwaccel_*() helpers

This series removes from networking core and driver code an assumption
about how VLAN tag presence is stored in an skb. This will allow to free
up overloading of VLAN.CFI bit to incidate tag's presence.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosky2: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:06 +0000 (00:18 +0100)]
sky2: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlx4: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:05 +0000 (00:18 +0100)]
mlx4: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobenet: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:04 +0000 (00:18 +0100)]
benet: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4/tunnel: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:04 +0000 (00:18 +0100)]
ipv4/tunnel: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobridge: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:03 +0000 (00:18 +0100)]
bridge: use __vlan_hwaccel helpers

This removes assumption than vlan_tci != 0 when tag is present.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years ago8021q: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:03 +0000 (00:18 +0100)]
8021q: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfnetlink/queue: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:02 +0000 (00:18 +0100)]
nfnetlink/queue: use __vlan_hwaccel helpers

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/core: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:02 +0000 (00:18 +0100)]
net/core: use __vlan_hwaccel helpers

This removes assumptions about VLAN_TAG_PRESENT bit.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: use __vlan_hwaccel helpers
Michał Mirosław [Thu, 8 Nov 2018 23:18:01 +0000 (00:18 +0100)]
cxgb4: use __vlan_hwaccel helpers

Use __vlan_hwaccel_put_tag() to set vlan tag and proto fields.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: move __skb_checksum_complete*() to skbuff.c
Cong Wang [Thu, 8 Nov 2018 22:05:42 +0000 (14:05 -0800)]
net: move __skb_checksum_complete*() to skbuff.c

__skb_checksum_complete_head() and __skb_checksum_complete()
are both declared in skbuff.h, they fit better in skbuff.c
than datagram.c.

Cc: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-ethernet-ti-cpsw-fix-vlan-mcast'
David S. Miller [Fri, 9 Nov 2018 04:30:58 +0000 (20:30 -0800)]
Merge branch 'net-ethernet-ti-cpsw-fix-vlan-mcast'

Ivan Khoronzhuk says:

====================
net: ethernet: ti: cpsw: fix vlan mcast

The cpsw holds separate mcast entires for vlan entries. At this moment
driver adds only not vlan mcast addresses, omitting vlan/mcast entries.
As result mcast for vlans doesn't work. It can be fixed by adding same
mcast entries for every created vlan, but this patchseries uses more
sophisticated way and allows to create mcast entries only for vlans
that really require it. Generic functions from this series can be
reused for fixing vlan and macvlan unicast.

Simple example of ALE table before and after this series, having same
mcast entries as for vlan 100 as for real device (reserved vlan 2),
and one mcast address only for vlan 100 - 01:1b:19:00:00:00.

<---- Before this patchset ---->
vlan , vid = 2, untag_force = 0x5, reg_mcast = 0x5, mem_list = 0x5
mcast, vid = 2, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
ucast, vid = 2, addr = 74:da:ea:47:7d:9d, persistant, port_num = 0x0
vlan , vid = 0, untag_force = 0x7, reg_mcast = 0x0, mem_list = 0x7
mcast, vid = 2, addr = 33:33:00:00:00:01, port_mask = 0x1
mcast, vid = 2, addr = 01:00:5e:00:00:01, port_mask = 0x1
vlan , vid = 1, untag_force = 0x3, reg_mcast = 0x3, mem_list = 0x3
mcast, vid = 1, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
ucast, vid = 1, addr = 74:da:ea:47:7d:9c, persistant, port_num = 0x0
mcast, vid = 1, addr = 33:33:00:00:00:01, port_mask = 0x1
mcast, vid = 1, addr = 01:00:5e:00:00:01, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:00, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:03, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:0e, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:00, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:03, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:0e, port_mask = 0x1
mcast, vid = 2, addr = 33:33:ff:47:7d:9d, port_mask = 0x1
mcast, vid = 2, addr = 33:33:00:00:00:fb, port_mask = 0x1
mcast, vid = 2, addr = 33:33:00:01:00:03, port_mask = 0x1
mcast, vid = 1, addr = 33:33:ff:47:7d:9c, port_mask = 0x1
mcast, vid = 1, addr = 33:33:00:00:00:fb, port_mask = 0x1
mcast, vid = 1, addr = 33:33:00:01:00:03, port_mask = 0x1
mcast, vid = 1, addr = 01:00:5e:00:00:fb, port_mask = 0x1
mcast, vid = 1, addr = 01:00:5e:00:00:fc, port_mask = 0x1
vlan , vid = 100, untag_force = 0x0, reg_mcast = 0x5, mem_list = 0x5
ucast, vid = 100, addr = 74:da:ea:47:7d:9d, persistant, port_num = 0x0
mcast, vid = 100, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
mcast, vid = 2, addr = 01:1b:19:00:00:00, port_mask = 0x1
 ^^^
 Here mcast entry (ptpl2), has to be added only for vlan 100
 but added for reserved vlan 2...that's not enough.

<---- After this patchset ---->
vlan , vid = 2, untag_force = 0x5, reg_mcast = 0x5, mem_list = 0x5
mcast, vid = 2, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
ucast, vid = 2, addr = 74:da:ea:47:7d:9d, persistant, port_num = 0x0
vlan , vid = 0, untag_force = 0x7, reg_mcast = 0x0, mem_list = 0x7
mcast, vid = 2, addr = 33:33:00:00:00:01, port_mask = 0x1
mcast, vid = 2, addr = 01:00:5e:00:00:01, port_mask = 0x1
vlan , vid = 1, untag_force = 0x3, reg_mcast = 0x3, mem_list = 0x3
mcast, vid = 1, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
ucast, vid = 1, addr = 74:da:ea:47:7d:9c, persistant, port_num = 0x0
mcast, vid = 1, addr = 33:33:00:00:00:01, port_mask = 0x1
mcast, vid = 1, addr = 01:00:5e:00:00:01, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:00, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:03, port_mask = 0x1
mcast, vid = 2, addr = 01:80:c2:00:00:0e, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:00, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:03, port_mask = 0x1
mcast, vid = 1, addr = 01:80:c2:00:00:0e, port_mask = 0x1
mcast, vid = 2, addr = 33:33:ff:47:7d:9d, port_mask = 0x1
mcast, vid = 1, addr = 33:33:ff:47:7d:9c, port_mask = 0x1
mcast, vid = 2, addr = 33:33:00:00:00:fb, port_mask = 0x1
mcast, vid = 2, addr = 33:33:00:01:00:03, port_mask = 0x1
mcast, vid = 1, addr = 33:33:00:00:00:fb, port_mask = 0x1
mcast, vid = 1, addr = 33:33:00:01:00:03, port_mask = 0x1
vlan , vid = 100, untag_force = 0x0, reg_mcast = 0x5, mem_list = 0x5
ucast, vid = 100, addr = 74:da:ea:47:7d:9d, persistant, port_num = 0x0
mcast, vid = 100, addr = ff:ff:ff:ff:ff:ff, port_mask = 0x1
mcast, vid = 100, addr = 33:33:00:00:00:01, port_mask = 0x1
mcast, vid = 100, addr = 01:00:5e:00:00:01, port_mask = 0x1
mcast, vid = 100, addr = 33:33:ff:47:7d:9d, port_mask = 0x1
mcast, vid = 100, addr = 01:80:c2:00:00:00, port_mask = 0x1
mcast, vid = 100, addr = 01:80:c2:00:00:03, port_mask = 0x1
mcast, vid = 100, addr = 01:80:c2:00:00:0e, port_mask = 0x1
mcast, vid = 100, addr = 33:33:00:00:00:fb, port_mask = 0x1
mcast, vid = 100, addr = 33:33:00:01:00:03, port_mask = 0x1
mcast, vid = 100, addr = 01:1b:19:00:00:00, port_mask = 0x1
 ^^^
    Here mcast entry (ptpl2), is added only for vlan 100
    as it should be.

Based on net-next/master

v2..v1:
  net: ethernet: ti: cpsw: fix vlan mcast
- removed limit for legacy switch cpsw mode
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: ti: cpsw: fix vlan configuration while down/up
Ivan Khoronzhuk [Thu, 8 Nov 2018 20:27:57 +0000 (22:27 +0200)]
net: ethernet: ti: cpsw: fix vlan configuration while down/up

The vlan configuration is not restored after interface donw/up sequence
(if dual-emac - both interfaces). Tested on am572x EVM.

Steps to check:
~# ip link add link eth1 name eth1.100 type vlan id 100
~# ifconfig eth0 down
~# ifconfig eth1 down

Try to remove vid and observe warning:
~# ip link del eth1.100
[  739.526757] net eth1: removing vlanid 100 from vlan filter
[  739.533322] failed to kill vid 0081/100 for device eth1

This patch fixes it, restoring only vlan ALE entries and all other
unicast/multicast entries are restored by system calling rx_mode ndo.

Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: ti: cpsw: fix vlan mcast
Ivan Khoronzhuk [Thu, 8 Nov 2018 20:27:56 +0000 (22:27 +0200)]
net: ethernet: ti: cpsw: fix vlan mcast

At this moment, mcast addresses are added for real device only
(reserved vlans for dual-emac mode), even if a mcast address was added
for some vlan only, thus ALE doesn't have corresponding vlan mcast
entries after vlan socket joined multicast group. So ALE drops vlan
frames with mcast addresses intended for vlans and potentially can
receive mcast frames for base ndev. That's not correct. So, fix it by
creating only vlan/mcast entries as requested. Patch doesn't use any
additional lists and is based on device mc address list and cpsw ALE
table entries.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: 8021q: vlan_core: allow use list of vlans for real device
Ivan Khoronzhuk [Thu, 8 Nov 2018 20:27:55 +0000 (22:27 +0200)]
net: 8021q: vlan_core: allow use list of vlans for real device

It's redundancy for the drivers to hold the list of vlans when
absolutely the same list exists in vlan core. In most cases it's
needed only to traverse the vlan devices, their vids and sync some
settings with h/w, so add API to simplify this.

At least some of these drivers also can benefit:
grep "for_each.*vid" -r drivers/net/ethernet/

drivers/net/ethernet/hisilicon/hns3/hns3_enet.c:
drivers/net/ethernet/synopsys/dwc-xlgmac-hw.c:
drivers/net/ethernet/qlogic/qlge/qlge_main.c:
drivers/net/ethernet/qlogic/qlcnic/qlcnic_main.c:
drivers/net/ethernet/via/via-rhine.c:
drivers/net/ethernet/via/via-velocity.c:
drivers/net/ethernet/intel/igb/igb_main.c:
drivers/net/ethernet/intel/ice/ice_main.c:
drivers/net/ethernet/intel/e1000/e1000_main.c:
drivers/net/ethernet/intel/i40e/i40e_main.c:
drivers/net/ethernet/intel/e1000e/netdev.c:
drivers/net/ethernet/intel/igbvf/netdev.c:
drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c:
drivers/net/ethernet/intel/ixgb/ixgb_main.c:
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c:
drivers/net/ethernet/amd/xgbe/xgbe-dev.c:
drivers/net/ethernet/emulex/benet/be_main.c:
drivers/net/ethernet/neterion/vxge/vxge-main.c:
drivers/net/ethernet/adaptec/starfire.c:
drivers/net/ethernet/brocade/bna/bnad.c:

Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: core: dev_addr_lists: add auxiliary func to handle reference address updates
Ivan Khoronzhuk [Thu, 8 Nov 2018 20:27:54 +0000 (22:27 +0200)]
net: core: dev_addr_lists: add auxiliary func to handle reference address updates

In order to avoid all table update, and only remove or add new
address, the auxiliary function exists, named __hw_addr_sync_dev().
It allows end driver do nothing when nothing changed and add/rm when
concrete address is firstly added or lastly removed. But it doesn't
include cases when an address of real device or vlan was reused by
other vlans or vlan/macval devices.

For handaling events when address was reused/unreused the patch adds
new auxiliary routine - __hw_addr_ref_sync_dev(). It allows to do
nothing when nothing was changed and do updates only for an address
being added/reused/deleted/unreused. Thus, clone address changes for
vlans can be mirrored in the table. The function is exclusive with
__hw_addr_sync_dev(). It's responsibility of the end driver to
identify address vlan device, if it needs so.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosfc: use the new __netdev_tx_sent_queue BQL optimisation
Edward Cree [Thu, 8 Nov 2018 19:47:19 +0000 (19:47 +0000)]
sfc: use the new __netdev_tx_sent_queue BQL optimisation

As added in 3e59020abf0f ("net: bql: add __netdev_tx_sent_queue()"), which
 see for performance rationale.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-Remove-VLAN_TAG_PRESENT-from-drivers'
David S. Miller [Fri, 9 Nov 2018 03:49:32 +0000 (19:49 -0800)]
Merge branch 'net-Remove-VLAN_TAG_PRESENT-from-drivers'

Michał Mirosław says:

====================
net: Remove VLAN_TAG_PRESENT from drivers

This series removes VLAN_TAG_PRESENT use from network drivers in
preparation to removing its special meaning.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogianfar: remove use of VLAN_TAG_PRESENT
Michał Mirosław [Thu, 8 Nov 2018 17:44:50 +0000 (18:44 +0100)]
gianfar: remove use of VLAN_TAG_PRESENT

Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoOVS: remove use of VLAN_TAG_PRESENT
Michał Mirosław [Thu, 8 Nov 2018 17:44:50 +0000 (18:44 +0100)]
OVS: remove use of VLAN_TAG_PRESENT

This is a minimal change to allow removing of VLAN_TAG_PRESENT.
It leaves OVS unable to use CFI bit, as fixing this would need
a deeper surgery involving userspace interface.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocnic: remove use of VLAN_TAG_PRESENT
Michał Mirosław [Thu, 8 Nov 2018 17:44:50 +0000 (18:44 +0100)]
cnic: remove use of VLAN_TAG_PRESENT

This just removes VLAN_TAG_PRESENT use.  VLAN TCI=0 special meaning is
deeply embedded in the driver code and so is left as is.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoi40iw: remove use of VLAN_TAG_PRESENT
Michał Mirosław [Thu, 8 Nov 2018 17:44:49 +0000 (18:44 +0100)]
i40iw: remove use of VLAN_TAG_PRESENT

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: socionext: refactor netsec_alloc_dring()
Ilias Apalodimas [Thu, 8 Nov 2018 15:19:55 +0000 (17:19 +0200)]
net: socionext: refactor netsec_alloc_dring()

return -ENOMEM directly instead of assigning it in a variable

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: socionext: different approach on DMA
Ilias Apalodimas [Thu, 8 Nov 2018 15:19:54 +0000 (17:19 +0200)]
net: socionext: different approach on DMA

Current driver dynamically allocates an skb and maps it as DMA Rx
buffer. In order to prepare for upcoming XDP changes, let's introduce a
different allocation scheme.
Buffers are allocated dynamically and mapped into hardware.
During the Rx operation the driver uses build_skb() to produce the
necessary buffers for the network stack.
This change increases performance ~15% on 64b packets with smmu disabled
and ~5% with smmu enabled

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: qca_spi: Add available buffer space verification
Stefan Wahren [Thu, 8 Nov 2018 13:38:21 +0000 (14:38 +0100)]
net: qca_spi: Add available buffer space verification

Interferences on the SPI line could distort the response of
available buffer space. So at least we should check that the
response doesn't exceed the maximum available buffer space.
In error case increase a new error counter and retry it later.
This behavior avoids buffer errors in the QCA7000, which
results in an unnecessary chip reset including packet loss.

Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosock: Reset dst when changing sk_mark via setsockopt
David Barmann [Thu, 8 Nov 2018 14:13:35 +0000 (08:13 -0600)]
sock: Reset dst when changing sk_mark via setsockopt

When setting the SO_MARK socket option, if the mark changes, the dst
needs to be reset so that a new route lookup is performed.

This fixes the case where an application wants to change routing by
setting a new sk_mark.  If this is done after some packets have already
been sent, the dst is cached and has no effect.

Signed-off-by: David Barmann <david.barmann@stackpath.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 's390-qeth-next'
David S. Miller [Fri, 9 Nov 2018 01:22:24 +0000 (17:22 -0800)]
Merge branch 's390-qeth-next'

Julian Wiedmann says:

====================
s390/qeth: updates 2018-11-08

please apply the following qeth patches to net-next.

The first patch allows one more device type to query the FW for a MAC address,
the others are all basically just removal of duplicated or unused code.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: don't process hsuid in qeth_l3_setup_netdev()
Julian Wiedmann [Thu, 8 Nov 2018 14:06:22 +0000 (15:06 +0100)]
s390/qeth: don't process hsuid in qeth_l3_setup_netdev()

qeth_l3_setup_netdev() checks if the hsuid attribute is set on the qeth
device, and propagates it to the net_device. In the past this was needed
to pick up any hsuid that was set before allocation of the net_device.

With commit d3d1b205e89f ("s390/qeth: allocate netdevice early") this
is no longer necessary, qeth_l3_dev_hsuid_store() always stores the
hsuid straight into dev->perm_addr.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: remove unused fallback in Layer3's MAC code
Julian Wiedmann [Thu, 8 Nov 2018 14:06:21 +0000 (15:06 +0100)]
s390/qeth: remove unused fallback in Layer3's MAC code

If the CREATE ADDR sent by qeth_l3_iqd_read_initial_mac() fails, its
callback sets a random MAC address on the net_device. The error then
propagates back, and qeth_l3_setup_netdev() bails out without
registering the net_device.

Any subsequent call to qeth_l3_setup_netdev() will then attempt a fresh
CREATE ADDR which either 1) also fails, or 2) sets a proper MAC address
on the net_device. Consequently, the net_device will never be registered
with a random MAC and we can drop the fallback code.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: remove two IPA command helpers
Julian Wiedmann [Thu, 8 Nov 2018 14:06:20 +0000 (15:06 +0100)]
s390/qeth: remove two IPA command helpers

qeth_l3_send_ipa_arp_cmd() is merely a wrapper around
qeth_send_control_data() now. So push the length adjustment into
QETH_SETASS_BASE_LEN, and remove the wrapper. While at it, also remove
some redundant 0-initializations.

qeth_send_setassparms() requires that callers prepare their command
parameters, so that they can be copied into the parameter area in one
go. Skip the indirection, and just let callers set up the command
themselves.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: replace open-coded cmd setup
Julian Wiedmann [Thu, 8 Nov 2018 14:06:19 +0000 (15:06 +0100)]
s390/qeth: replace open-coded cmd setup

Call qeth_prepare_ipa_cmd() during setup of a new IPA cmd buffer, so
that it is used for all commands. Thus ARP and SNMP requests don't have
to do their own initialization.

This will now also set the proper MPC protocol version for SNMP requests
on L2 devices.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: remove card list
Julian Wiedmann [Thu, 8 Nov 2018 14:06:18 +0000 (15:06 +0100)]
s390/qeth: remove card list

Re-implement the card-by-RDEV lookup by using device model concepts, and
remove the now redundant list of all qeth card instances in the system.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: unify transmit code
Julian Wiedmann [Thu, 8 Nov 2018 14:06:17 +0000 (15:06 +0100)]
s390/qeth: unify transmit code

Since commit 82bf5c0867f6 ("s390/qeth: add support for IPv6 TSO"),
qeth_xmit() also knows how to build TSO packets and is practically
identical to qeth_l3_xmit().
Convert qeth_l3_xmit() into a thin wrapper that merely strips the
L2 header off a packet, and calls qeth_xmit() for the actual
TX processing.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: handle af_iucv skbs in qeth_l3_fill_header()
Julian Wiedmann [Thu, 8 Nov 2018 14:06:16 +0000 (15:06 +0100)]
s390/qeth: handle af_iucv skbs in qeth_l3_fill_header()

Filling the HW header from one single function will make it easier to
rip out all the duplicated transmit code in qeth_l3_xmit(). On top, this
saves one conditional branch in the TSO path.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: utilize virtual MAC for Layer2 OSD devices
Julian Wiedmann [Thu, 8 Nov 2018 14:06:15 +0000 (15:06 +0100)]
s390/qeth: utilize virtual MAC for Layer2 OSD devices

By default, READ MAC on a Layer2 OSD device returns the adapter's
burnt-in MAC address. Given the default scenario of many virtual devices
on the same adapter, qeth can't make any use of this address and
therefore skips the READ MAC call for this device type.

But in some configurations, the READ MAC command for a Layer2 OSD device
actually returns a pre-provisioned, virtual MAC address. So enable the
READ MAC code to detect this situation, and let the L2 subdriver
call READ MAC for OSD devices.

This also removes the QETH_LAYER2_MAC_READ flag, which protects L2
devices against calling READ MAC multiple times. Instead protect the
whole call to qeth_l2_request_initial_mac().

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoopenvswitch: remove BUG_ON from get_dpdev
Li RongQing [Thu, 8 Nov 2018 12:40:20 +0000 (20:40 +0800)]
openvswitch: remove BUG_ON from get_dpdev

if local is NULL pointer, and the following access of local's
dev will trigger panic, which is same as BUG_ON

Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ICMP-error-handling-for-UDP-tunnels'
David S. Miller [Fri, 9 Nov 2018 01:13:09 +0000 (17:13 -0800)]
Merge branch 'ICMP-error-handling-for-UDP-tunnels'

Stefano Brivio says:

====================
ICMP error handling for UDP tunnels

This series introduces ICMP error handling for UDP tunnels and
encapsulations and related selftests. We need to handle ICMP errors to
support PMTU discovery and route redirection -- this support is entirely
missing right now:

- patch 1/11 adds a socket lookup for UDP tunnels that use, by design,
  the same destination port on both endpoints -- i.e. VXLAN and GENEVE
- patches 2/11 to 7/11 are specific to VxLAN and GENEVE
- patches 8/11 and 9/11 add infrastructure for lookup of encapsulations
  where sent packets cannot be matched via receiving socket lookup, i.e.
  FoU and GUE
- patches 10/11 and 11/11 are specific to FoU and GUE

v2: changes are listed in the single patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: pmtu: Introduce FoU and GUE PMTU exceptions tests
Stefano Brivio [Thu, 8 Nov 2018 11:19:24 +0000 (12:19 +0100)]
selftests: pmtu: Introduce FoU and GUE PMTU exceptions tests

Introduce eight tests, for FoU and GUE, with IPv4 and IPv6 payload,
on IPv4 and IPv6 transport, that check that PMTU exceptions are created
with the right value when exceeding the MTU on a link of the path.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agofou, fou6: ICMP error handlers for FoU and GUE
Stefano Brivio [Thu, 8 Nov 2018 11:19:23 +0000 (12:19 +0100)]
fou, fou6: ICMP error handlers for FoU and GUE

As the destination port in FoU and GUE receiving sockets doesn't
necessarily match the remote destination port, we can't associate errors
to the encapsulating tunnels with a socket lookup -- we need to blindly
try them instead. This means we don't even know if we are handling errors
for FoU or GUE without digging into the packets.

Hence, implement a single handler for both, one for IPv4 and one for IPv6,
that will check whether the packet that generated the ICMP error used a
direct IP encapsulation or if it had a GUE header, and send the error to
the matching protocol handler, if any.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoudp: Support for error handlers of tunnels with arbitrary destination port
Stefano Brivio [Thu, 8 Nov 2018 11:19:22 +0000 (12:19 +0100)]
udp: Support for error handlers of tunnels with arbitrary destination port

ICMP error handling is currently not possible for UDP tunnels not
employing a receiving socket with local destination port matching the
remote one, because we have no way to look them up.

Add an err_handler tunnel encapsulation operation that can be exported by
tunnels in order to pass the error to the protocol implementing the
encapsulation. We can't easily use a lookup function as we did for VXLAN
and GENEVE, as protocol error handlers, which would be in turn called by
implementations of this new operation, handle the errors themselves,
together with the tunnel lookup.

Without a socket, we can't be sure which encapsulation error handler is
the appropriate one: encapsulation handlers (the ones for FoU and GUE
introduced in the next patch, e.g.) will need to check the new error codes
returned by protocol handlers to figure out if errors match the given
encapsulation, and, in turn, report this error back, so that we can try
all of them in __udp{4,6}_lib_err_encap_no_sk() until we have a match.

v2:
- Name all arguments in err_handler prototypes (David Miller)

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: Convert protocol error handlers from void to int
Stefano Brivio [Thu, 8 Nov 2018 11:19:21 +0000 (12:19 +0100)]
net: Convert protocol error handlers from void to int

We'll need this to handle ICMP errors for tunnels without a sending socket
(i.e. FoU and GUE). There, we might have to look up different types of IP
tunnels, registered as network protocols, before we get a match, so we
want this for the error handlers of IPPROTO_IPIP and IPPROTO_IPV6 in both
inet_protos and inet6_protos. These error codes will be used in the next
patch.

For consistency, return sensible error codes in protocol error handlers
whenever handlers can't handle errors because, even if valid, they don't
match a protocol or any of its states.

This has no effect on existing error handling paths.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: pmtu: Introduce tests for IPv4/IPv6 over GENEVE over IPv4/IPv6
Stefano Brivio [Thu, 8 Nov 2018 11:19:20 +0000 (12:19 +0100)]
selftests: pmtu: Introduce tests for IPv4/IPv6 over GENEVE over IPv4/IPv6

Use a router between endpoints, implemented via namespaces, set a low MTU
between router and destination endpoint, exceed it and check PMTU value in
route exceptions.

v2:
- Introduce IPv4 tests right away, if iproute2 doesn't support the 'df'
  link option they will be skipped (David Ahern)

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogeneve: Allow configuration of DF behaviour
Stefano Brivio [Thu, 8 Nov 2018 11:19:19 +0000 (12:19 +0100)]
geneve: Allow configuration of DF behaviour

draft-ietf-nvo3-geneve-08 says:

   It is strongly RECOMMENDED that Path MTU Discovery ([RFC1191],
   [RFC1981]) be used by setting the DF bit in the IP header when Geneve
   packets are transmitted over IPv4 (this is the default with IPv6).

Now that ICMP error handling is working for GENEVE, we can comply with
this recommendation.

Make this configurable, though, to avoid breaking existing setups. By
default, DF won't be set. It can be set or inherited from inner IPv4
packets. If it's configured to be inherited and we are encapsulating IPv6,
it will be set.

This only applies to non-lwt tunnels: if an external control plane is
used, tunnel key will still control the DF flag.

v2:
- DF behaviour configuration only applies for non-lwt tunnels, apply DF
  setting only if (!geneve->collect_md) in geneve_xmit_skb()
  (Stephen Hemminger)

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogeneve: ICMP error lookup handler
Stefano Brivio [Thu, 8 Nov 2018 11:19:18 +0000 (12:19 +0100)]
geneve: ICMP error lookup handler

Export an encap_err_lookup() operation to match an ICMP error against a
valid VNI.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>