Skip to content

L
linux
Switch branch/tag
  1. 19 Dec, 2018 9 commits
    • David S. Miller's avatar
      Merge branch 'vxlan-Various-fixes' · 59fc137e
      David S. Miller authored 
      Petr Machata says:

====================
vxlan: Various fixes

This patch set contains three fixes for the vxlan driver.

Patch #1 fixes handling of offload mark on replaced VXLAN FDB entries. A
way to trigger this is to replace the FDB entry with one that can not be
offloaded. A future patch set should make it possible to veto such FDB
changes. However the FDB might still fail to be offloaded due to another
issue, and the offload mark should reflect that.

Patch #2 fixes problems in __vxlan_dev_create() when a call to
rtnl_configure_link() fails. These failures would be tricky to hit on a
real system, the most likely vector is through an error in vxlan_open().
However, with the abovementioned vetoing patchset, vetoing the created
entry would trigger the same problems (and be easier to reproduce).

Patch #3 fixes a problem in vxlan_changelink(). In situations where the
default remote configured in the FDB table (if any) does not exactly
match the remote address configured at the VXLAN device, changing the
remote address breaks the default FDB entry. Patch #4 is then a self
test for this issue.

v3:
- Patch #2:
    - Reuse the same errout block for both cleanup paths. Use a bool to
      decide whether the unregister_netdevice() call should be made.

v2:
- Drop former patch #3
- Patch #2:
    - Delete the default entry before calling unregister_netdevice(). That
      takes care of former patch #3, hence tweak the commit message to
      mention that problem as well.
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      59fc137e
    • Petr Machata's avatar
      selftests: net: Add test_vxlan_fdb_changelink.sh · 55cbe079
      Petr Machata authored 
      Add a test to exercise the fix from the previous patch.
Signed-off-by: default avatarPetr Machata <petrm@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      55cbe079
    • Petr Machata's avatar
      vxlan: changelink: Fix handling of default remotes · ce5e098f
      Petr Machata authored 
      Default remotes are stored as FDB entries with an Ethernet address of
00:00:00:00:00:00. When a request is made to change a remote address of
a VXLAN device, vxlan_changelink() first deletes the existing default
remote, and then creates a new FDB entry.

This works well as long as the list of default remotes matches exactly
the configuration of a VXLAN remote address. Thus when the VXLAN device
has a remote of X, there should be exactly one default remote FDB entry
X. If the VXLAN device has no remote address, there should be no such
entry.

Besides using "ip link set", it is possible to manipulate the list of
default remotes by using the "bridge fdb". It is therefore easy to break
the above condition. Under such circumstances, the __vxlan_fdb_delete()
call doesn't delete the FDB entry itself, but just one remote. The
following vxlan_fdb_create() then creates a new FDB entry, leading to a
situation where two entries exist for the address 00:00:00:00:00:00,
each with a different subset of default remotes.

An even more obvious breakage rooted in the same cause can be observed
when a remote address is configured for a VXLAN device that did not have
one before. In that case vxlan_changelink() doesn't remove any remote,
and just creates a new FDB entry for the new address:

$ ip link add name vx up type vxlan id 2000 dstport 4789
$ bridge fdb ap dev vx 00:00:00:00:00:00 dst 192.0.2.20 self permanent
$ bridge fdb ap dev vx 00:00:00:00:00:00 dst 192.0.2.30 self permanent
$ ip link set dev vx type vxlan remote 192.0.2.30
$ bridge fdb sh dev vx | grep 00:00:00:00:00:00
00:00:00:00:00:00 dst 192.0.2.30 self permanent <- new entry, 1 rdst
00:00:00:00:00:00 dst 192.0.2.20 self permanent <- orig. entry, 2 rdsts
00:00:00:00:00:00 dst 192.0.2.30 self permanent

To fix this, instead of calling vxlan_fdb_create() directly, defer to
vxlan_fdb_update(). That has logic to handle the duplicates properly.
Additionally, it also handles notifications, so drop that call from
changelink as well.

Fixes: 0241b836 ("vxlan: fix default fdb entry netlink notify ordering during netdev create")
Signed-off-by: default avatarPetr Machata <petrm@mellanox.com>
Acked-by: default avatarRoopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      ce5e098f
    • Petr Machata's avatar
      vxlan: Fix error path in __vxlan_dev_create() · 6db92468
      Petr Machata authored 
      When a failure occurs in rtnl_configure_link(), the current code
calls unregister_netdevice() to roll back the earlier call to
register_netdevice(), and jumps to errout, which calls
vxlan_fdb_destroy().

However unregister_netdevice() calls transitively ndo_uninit, which is
vxlan_uninit(), and that already takes care of deleting the default FDB
entry by calling vxlan_fdb_delete_default(). Since the entry added
earlier in __vxlan_dev_create() is exactly the default entry, the
cleanup code in the errout block always leads to double free and thus a
panic.

Besides, since vxlan_fdb_delete_default() always destroys the FDB entry
with notification enabled, the deletion of the default entry is notified
even before the addition was notified.

Instead, move the unregister_netdevice() call after the manual destroy,
which solves both problems.

Fixes: 0241b836 ("vxlan: fix default fdb entry netlink notify ordering during netdev create")
Signed-off-by: default avatarPetr Machata <petrm@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      6db92468
    • Petr Machata's avatar
      vxlan: Unmark offloaded bit on replaced FDB entries · 6ad0b5a4
      Petr Machata authored 
      When rdst of an offloaded FDB entry is replaced, it certainly isn't
offloaded anymore. Drivers are notified about such replacements, and can
re-mark the entry as offloaded again if they so wish. However until a
driver does so explicitly, assume a replaced FDB entry is not offloaded.

Note that replaces coming via vxlan_fdb_external_learn_add() are always
immediately followed by an explicit offload marking.

Fixes: 0efe1173 ("vxlan: Support marking RDSTs as offloaded")
Signed-off-by: default avatarPetr Machata <petrm@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      6ad0b5a4
    • David S. Miller's avatar
      Merge branch 'macb-DMA-race-fixes' · a9d6d897
      David S. Miller authored 
      Anssi Hannula says:

====================
net: macb: DMA race condition fixes

Here are a couple of race condition fixes for the macb driver. The first
two are for issues observed at runtime on real HW.

v2:
- added received Tested-bys and Acked-bys to the first two patches
- in patch 3/3, moved the timestamp protection barrier closer to the
  timestamp reads
- in patch 3/3, removed unnecessary move of the addr assignment in
  gem_rx() to keep the patch minimal for maximum clarity
- in patch 3/3, clarified commit message and comments

The 3/3 is the same one I improperly sent last week as a standalone
patch.
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      a9d6d897
    • Anssi Hannula's avatar
      net: macb: add missing barriers when reading descriptors · 6e0af298
      Anssi Hannula authored 
      When reading buffer descriptors on RX or on TX completion, an
RX_USED/TX_USED bit is checked first to ensure that the descriptors have
been populated, i.e. the ownership has been transferred. However, there
are no memory barriers to ensure that the data protected by the
RX_USED/TX_USED bit is up-to-date with respect to that bit.

Specifically:

- TX timestamp descriptors may be loaded before ctrl is loaded for the
  TX_USED check, which is racy as the descriptors may be updated between
  the loads, causing old timestamp descriptor data to be used.

- RX ctrl may be loaded before addr is loaded for the RX_USED check,
  which is racy as a new frame may be written between the loads, causing
  old ctrl descriptor data to be used.
  This issue exists for both macb_rx() and gem_rx() variants.

Fix the races by adding DMA read memory barriers on those paths and
reordering the reads in macb_rx().

I have not observed any actual problems in practice caused by these
being missing, though.

Tested on a ZynqMP based system.

Fixes: 89e5785f ("[PATCH] Atmel MACB ethernet driver")
Signed-off-by: default avatarAnssi Hannula <anssi.hannula@bitwise.fi>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      6e0af298
    • Anssi Hannula's avatar
      net: macb: fix dropped RX frames due to a race · 8159ecab
      Anssi Hannula authored 
      Bit RX_USED set to 0 in the address field allows the controller to write
data to the receive buffer descriptor.

The driver does not ensure the ctrl field is ready (cleared) when the
controller sees the RX_USED=0 written by the driver. The ctrl field might
only be cleared after the controller has already updated it according to
a newly received frame, causing the frame to be discarded in gem_rx() due
to unexpected ctrl field contents.

A message is logged when the above scenario occurs:

  macb ff0b0000.ethernet eth0: not whole frame pointed by descriptor

Fix the issue by ensuring that when the controller sees RX_USED=0 the
ctrl field is already cleared.

This issue was observed on a ZynqMP based system.

Fixes: 4df95131 ("net/macb: change RX path for GEM")
Signed-off-by: default avatarAnssi Hannula <anssi.hannula@bitwise.fi>
Tested-by: default avatarClaudiu Beznea <claudiu.beznea@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      8159ecab
    • Anssi Hannula's avatar
      net: macb: fix random memory corruption on RX with 64-bit DMA · e100a897
      Anssi Hannula authored 
      64-bit DMA addresses are split in upper and lower halves that are
written in separate fields on GEM. For RX, bit 0 of the address is used
as the ownership bit (RX_USED). When the RX_USED bit is unset the
controller is allowed to write data to the buffer.

The driver does not guarantee that the controller already sees the upper
half when the RX_USED bit is cleared, possibly resulting in the
controller writing an incoming frame to an address with an incorrect
upper half and therefore possibly corrupting unrelated system memory.

Fix that by adding the necessary DMA memory barrier between the writes.

This corruption was observed on a ZynqMP based system.

Fixes: fff8019a ("net: macb: Add 64 bit addressing support for GEM")
Signed-off-by: default avatarAnssi Hannula <anssi.hannula@bitwise.fi>
Acked-by: default avatarHarini Katakam <harini.katakam@xilinx.com>
Tested-by: default avatarClaudiu Beznea <claudiu.beznea@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Cc: Michal Simek <michal.simek@xilinx.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      e100a897
  3. 18 Dec, 2018 16 commits
  5. 17 Dec, 2018 1 commit
  7. 16 Dec, 2018 12 commits
    • Marcin Wojtas's avatar
      net: mvneta: fix operation for 64K PAGE_SIZE · e735fd55
      Marcin Wojtas authored 
      Recent changes in the mvneta driver reworked allocation
and handling of the ingress buffers to use entire pages.
Apart from that in SW BM scenario the HW must be informed
via PRXDQS about the biggest possible incoming buffer
that can be propagated by RX descriptors.

The BufferSize field was filled according to the MTU-dependent
pkt_size value. Later change to PAGE_SIZE broke RX operation
when usin 64K pages, as the field is simply too small.

This patch conditionally limits the value passed to the BufferSize
of the PRXDQS register, depending on the PAGE_SIZE used.
On the occasion remove now unused frag_size field of the mvneta_port
structure.

Fixes: 562e2f46 ("net: mvneta: Improve the buffer allocation method for SWBM")
Signed-off-by: default avatarMarcin Wojtas <mw@semihalf.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      e735fd55
    • David S. Miller's avatar
      Merge branch 'hns-fixes' · 369a094d
      David S. Miller authored 
      Peng Li says:

====================
net: hns: Code improvements & fixes for HNS driver

This patchset introduces some code improvements and fixes
for the identified problems in the HNS driver.

Every patch is independent.
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      369a094d
    • Yonglong Liu's avatar
      net: hns: Fix ping failed when use net bridge and send multicast · 6adafc35
      Yonglong Liu authored 
      Create a net bridge, add eth and vnet to the bridge. The vnet is used
by a virtual machine. When ping the virtual machine from the outside
host and the virtual machine send multicast at the same time, the ping
package will lost.

The multicast package send to the eth, eth will send it to the bridge too,
and the bridge learn the mac of eth. When outside host ping the virtual
mechine, it will match the promisc entry of the eth which is not expected,
and the bridge send it to eth not to vnet, cause ping lost.

So this patch change promisc tcam entry position to the END of 512 tcam
entries, which indicate lower priority. And separate one promisc entry to
two: mc & uc, to avoid package match the wrong tcam entry.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      6adafc35
    • Yonglong Liu's avatar
      net: hns: Add mac pcs config when enable|disable mac · 726ae5c9
      Yonglong Liu authored 
      In some case, when mac enable|disable and adjust link, may cause hard to
link(or abnormal) between mac and phy. This patch adds the code for rx PCS
to avoid this bug.

Disable the rx PCS when driver disable the gmac, and enable the rx PCS
when driver enable the mac.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      726ae5c9
    • Yonglong Liu's avatar
      net: hns: Fix ntuple-filters status error. · 7e74a19c
      Yonglong Liu authored 
      The ntuple-filters features is forced on by chip.
But it shows "ntuple-filters: off [fixed]" when use ethtool.
This patch make it correct with "ntuple-filters: on [fixed]".
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      7e74a19c
    • Yonglong Liu's avatar
      net: hns: Avoid net reset caused by pause frames storm · a57275d3
      Yonglong Liu authored 
      There will be a large number of MAC pause frames on the net,
which caused tx timeout of net device. And then the net device
was reset to try to recover it. So that is not useful, and will
cause some other problems.

So need doubled ndev->watchdog_timeo if device watchdog occurred
until watchdog_timeo up to 40s and then try resetting to recover
it.

When collecting dfx information such as hardware registers when tx timeout.
Some registers for count were cleared when read. So need move this task
before update net state which also read the count registers.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      a57275d3
    • Yonglong Liu's avatar
      net: hns: Free irq when exit from abnormal branch · c82bd077
      Yonglong Liu authored 
      1.In "hns_nic_init_irq", if request irq fail at index i,
  the function return directly without releasing irq resources
  that already requested.

2.In "hns_nic_net_up" after "hns_nic_init_irq",
  if exceptional branch occurs, irqs that already requested
  are not release.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      c82bd077
    • Yonglong Liu's avatar
      net: hns: Clean rx fbd when ae stopped. · 31f6b61d
      Yonglong Liu authored 
      If there are packets in hardware when changing the speed or duplex,
it may cause hardware hang up.

This patch adds the code to wait rx fbd clean up when ae stopped.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      31f6b61d
    • Yonglong Liu's avatar
      net: hns: Fixed bug that netdev was opened twice · 5778b13b
      Yonglong Liu authored 
      After resetting dsaf to try to repair chip error such as ecc error,
the net device will be open if net interface is up. But at this time
if there is the users set the net device up with the command ifconfig,
the net device will be opened twice consecutively.

Function napi_enable was called when open device. And Kernel panic will
be occurred if it was called twice consecutively. Such as follow:
static inline void napi_enable(struct napi_struct *n)
{
         BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
         smp_mb__before_clear_bit();
         clear_bit(NAPI_STATE_SCHED, &n->state);
}

[37255.571996] Kernel panic - not syncing: BUG!
[37255.595234] Call trace:
[37255.597694] [<ffff80000008ab48>] dump_backtrace+0x0/0x1a0
[37255.603114] [<ffff80000008ad08>] show_stack+0x20/0x28
[37255.608187] [<ffff8000009c4944>] dump_stack+0x98/0xb8
[37255.613258] [<ffff8000009c149c>] panic+0x10c/0x26c
[37255.618070] [<ffff80000070f134>] hns_nic_net_up+0x30c/0x4e0
[37255.623664] [<ffff80000070f39c>] hns_nic_net_open+0x94/0x12c
[37255.629346] [<ffff80000084be78>] __dev_open+0xf4/0x168
[37255.634504] [<ffff80000084c1ac>] __dev_change_flags+0x98/0x15c
[37255.640359] [<ffff80000084c29c>] dev_change_flags+0x2c/0x68
[37255.769580] [<ffff8000008dc400>] devinet_ioctl+0x650/0x704
[37255.775086] [<ffff8000008ddc38>] inet_ioctl+0x98/0xb4
[37255.780159] [<ffff800000827b7c>] sock_do_ioctl+0x44/0x84
[37255.785490] [<ffff800000828e04>] sock_ioctl+0x248/0x30c
[37255.790737] [<ffff80000026dc6c>] do_vfs_ioctl+0x480/0x618
[37255.796156] [<ffff80000026de94>] SyS_ioctl+0x90/0xa4
[37255.801139] SMP: stopping secondary CPUs
[37255.805079] kbox: catch panic event.
[37255.809586] collected_len = 128928, LOG_BUF_LEN_LOCAL = 131072
[37255.816103] flush cache 0xffff80003f000000  size 0x800000
[37255.822192] flush cache 0xffff80003f000000  size 0x800000
[37255.828289] flush cache 0xffff80003f000000  size 0x800000
[37255.834378] kbox: no notify die func register. no need to notify
[37255.840413] ---[ end Kernel panic - not syncing: BUG!

This patchset fix this bug according to the flag NIC_STATE_DOWN.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      5778b13b
    • Yonglong Liu's avatar
      net: hns: Some registers use wrong address according to the datasheet. · 4ad26f11
      Yonglong Liu authored 
      According to the hip06 datasheet:
1.Six registers use wrong address:
  RCB_COM_SF_CFG_INTMASK_RING
  RCB_COM_SF_CFG_RING_STS
  RCB_COM_SF_CFG_RING
  RCB_COM_SF_CFG_INTMASK_BD
  RCB_COM_SF_CFG_BD_RINT_STS
  DSAF_INODE_VC1_IN_PKT_NUM_0_REG
2.The offset of DSAF_INODE_VC1_IN_PKT_NUM_0_REG should be
  0x103C + 0x80 * all_chn_num
3.The offset to show the value of DSAF_INODE_IN_DATA_STP_DISC_0_REG
  is wrong, so the value of DSAF_INODE_SW_VLAN_TAG_DISC_0_REG will be
  overwrite

These registers are only used in "ethtool -d", so that did not cause ndev
to misfunction.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      4ad26f11
    • Yonglong Liu's avatar
      net: hns: All ports can not work when insmod hns ko after rmmod. · 308c6caf
      Yonglong Liu authored 
      There are two test cases:
1. Remove the 4 modules:hns_enet_drv/hns_dsaf/hnae/hns_mdio,
   and install them again, must use "ifconfig down/ifconfig up"
   command pair to bring port to work.

   This patch calls phy_stop function when init phy to fix this bug.

2. Remove the 2 modules:hns_enet_drv/hns_dsaf, and install them again,
   all ports can not use anymore, because of the phy devices register
   failed(phy devices already exists).

   Phy devices are registered when hns_dsaf installed, this patch
   removes them when hns_dsaf removed.

The two cases are sometimes related, fixing the second case also requires
fixing the first case, so fix them together.
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      308c6caf
    • Yonglong Liu's avatar
      net: hns: Incorrect offset address used for some registers. · 4e1d4be6
      Yonglong Liu authored 
      According to the hip06 Datasheet:
1. The offset of INGRESS_SW_VLAN_TAG_DISC should be 0x1A00+4*all_chn_num
2. The offset of INGRESS_IN_DATA_STP_DISC should be 0x1A50+4*all_chn_num
Signed-off-by: default avatarYonglong Liu <liuyonglong@huawei.com>
Signed-off-by: default avatarPeng Li <lipeng321@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      4e1d4be6
  9. 15 Dec, 2018 2 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7