Recent months, our customer reported several kernel crashes all
preceding with following message:
NETDEV WATCHDOG: eth2 (enic): transmit queue 0 timed out
Error message of one of those crashes:
BUG: unable to handle kernel paging request at ffffffffa007e090
After analyzing severl vmcores, I found that most of crashes are
caused by memory corruption. And all the corrupted memory areas
are overwritten by data of network packets. Moreover, I also found
that the tx queues were enabled over watchdog reset.
After going through the source code, I found that in enic_stop(),
the tx queues stopped by netif_tx_disable() could be woken up over
a small time window between netif_tx_disable() and the
napi_disable() by the following code path:
napi_poll->
enic_poll_msix_wq->
vnic_cq_service->
enic_wq_service->
netif_wake_subqueue(enic->netdev, q_number)->
test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)
In turn, upper netowrk stack could queue skb to ENIC NIC though
enic_hard_start_xmit(). And this might introduce some race condition.
Our customer comfirmed that this kind of kernel crash doesn't occur over
90 days since they applied this patch.
Signed-off-by: Firo Yang <firo.yang@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are two reasons for this.
First, the xmit_more flag conceptually doesn't fit into the skb, as
xmit_more is not a property related to the skb.
Its only a hint to the driver that the stack is about to transmit another
packet immediately.
Second, it was only done this way to not have to pass another argument
to ndo_start_xmit().
We can place xmit_more in the softnet data, next to the device recursion.
The recursion counter is already written to on each transmit. The "more"
indicator is placed right next to it.
Drivers can use the netdev_xmit_more() helper instead of skb->xmit_more
to check the "more packets coming" hint.
skb->xmit_more is retained (but always 0) to not cause build breakage.
This change takes care of the simple s/skb->xmit_more/netdev_xmit_more()/
conversions. Remaining drivers are converted in the next patches.
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The enic driver relies on the CONFIG_CPUMASK_OFFSTACK feature to
dynamically allocate a struct member, but this is normally intended for
local variables.
Building with clang, I get a warning for a few locations that check the
address of the cpumask_var_t:
drivers/net/ethernet/cisco/enic/enic_main.c:122:22: error: address of array 'enic->msix[i].affinity_mask' will always evaluate to 'true' [-Werror,-Wpointer-bool-conversion]
As far as I can tell, the code is still correct, as the truth value of
the pointer is what we need in this configuration. To get rid of
the warning, use cpumask_available() instead of checking the
pointer directly.
Fixes: 322cf7e3a4 ("enic: assign affinity hint to interrupts")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of IPv6 pkts, ipv4_csum_ok is 0. Because of this, driver does
not set skb->ip_summed. So IPv6 rx checksum is not offloaded.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit ab123fe071 ("enic: handle mtu change for vf properly")
ASSERT_RTNL() is added to _enic_change_mtu() to prevent it from being
called without rtnl held. enic_probe() calls enic_change_mtu()
without rtnl held. At this point netdev is not registered yet.
Remove call to enic_change_mtu and assign the mtu to netdev->mtu.
Fixes: ab123fe071 ("enic: handle mtu change for vf properly")
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When driver gets notification for mtu change, driver does not handle it for
all RQs. It handles only RQ[0].
Fix is to use enic_change_mtu() interface to change mtu for vf.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In failure path, we overwrite err to what vnic_rq_disable() returns. In
case it returns 0, enic_open() returns success in case of error.
Reported-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Fixes: e8588e2685 ("enic: enable rq before updating rq descriptors")
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
lockdep spotted that we are using rfs_h.lock in enic_get_rxnfc() without
initializing. rfs_h.lock is initialized in enic_open(). But ethtool_ops
can be called when interface is down.
Move enic_rfs_flw_tbl_init to enic_probe.
INFO: trying to register non-static key.
the code is fine but needs lockdep annotation.
turning off the locking correctness validator.
CPU: 18 PID: 1189 Comm: ethtool Not tainted 4.17.0-rc7-devel+ #27
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
Call Trace:
dump_stack+0x85/0xc0
register_lock_class+0x550/0x560
? __handle_mm_fault+0xa8b/0x1100
__lock_acquire+0x81/0x670
lock_acquire+0xb9/0x1e0
? enic_get_rxnfc+0x139/0x2b0 [enic]
_raw_spin_lock_bh+0x38/0x80
? enic_get_rxnfc+0x139/0x2b0 [enic]
enic_get_rxnfc+0x139/0x2b0 [enic]
ethtool_get_rxnfc+0x8d/0x1c0
dev_ethtool+0x16c8/0x2400
? __mutex_lock+0x64d/0xa00
? dev_load+0x6a/0x150
dev_ioctl+0x253/0x4b0
sock_do_ioctl+0x9a/0x130
sock_ioctl+0x1af/0x350
do_vfs_ioctl+0x8e/0x670
? syscall_trace_enter+0x1e2/0x380
ksys_ioctl+0x60/0x90
__x64_sys_ioctl+0x16/0x20
do_syscall_64+0x5a/0x170
entry_SYSCALL_64_after_hwframe+0x49/0xbe
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 48398b6e70 ("enic: set UDP rss flag") driver needed to set a
single bit to enable UDP rss. This is changed to two bit. One for UDP
IPv4 and other bit for UDP IPv6. The hardware which supports this is not
released yet. When released, driver should set 2 bit to enable UDP rss for
both IPv4 and IPv6.
Also add spinlock around vnic_dev_capable_rss_hash_type().
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 624dbf55a3 ("driver/net: enic: Try DMA 64 first, then
failover to DMA") DMA mask was changed from 40 bits to 64 bits.
Hardware actually supports only 47 bits.
Fixes: 624dbf55a3 ("driver/net: enic: Try DMA 64 first, then failover to DMA")
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit d11790941d ("enic: Add vxlan offload support for IPv6 pkts")
added vxlan offload support for IPv6 pkts. Required change in
enic_udp_tunnel_del was not made. This creates a bug where once user
adds IPv6 tunnel, hw offload for that cannot be deleted.
This patch removes check for IP proto in tunnel delete path. Driver need
not check for IP proto since same UDP port cannot be used to create two
tunnels.
Fixes: d11790941d ("enic: Add vxlan offload support for IPv6 pkts")
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New adapter needs CMD_OPENF_IG_DESCCACHE flag to be set. If this flag is
not set, fw flushes the global IG desc cache. This flag is nop in older
adapter.
Also increment driver version
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rq should be enabled before posting the buffers to rq desc. If not hw sees
stale value and casuses DMAR errors.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New hardware needs UDP flag set to enable UDP L4 rss hash. Add ethtool
get option to display supported rss flow hash.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some adaptors do not support vxlan offload when multi wq is configured.
If hw supports multi wq, BIT(2) is set in a1.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New adaptors supports vxlan offload for inner IPv6 and outer IPv6 vxlan
pkts.
Fw sets BIT(0) & BIT(1) in a1 if hw supports ipv6 inner & outer pkt
offload.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
To compute pseudo IP header csum, we need to check the inner header for
encap pkt, not outer IP header.
Also add pseudo csum for IPv6 inner pkt.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of tx clean up, we set '-1' as budget. This means clean up until
wq is empty or till (1 << 32) pkts are cleaned. Under heavy load this
will run for long time and cause
"watchdog: BUG: soft lockup - CPU#25 stuck for 21s!" warning.
This patch sets wq clean up budget to 256.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ethtool ops to advertise sw timestamping.
Call skb_tx_timestamp() just before ringing the wq doorbell.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use setup_timer function instead of initializing timer with the
function and data fields.
Signed-off-by: Allen Pais <allen.lkml@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With -Wformat-truncation, gcc throws the following warning.
Fix this by increasing the size of devname to accommodate 15 character
netdev interface name and description.
Remove length format precision for %s. We can fit entire name.
Also increment the version.
drivers/net/ethernet/cisco/enic/enic_main.c: In function ‘enic_open’:
drivers/net/ethernet/cisco/enic/enic_main.c:1740:15: warning: ‘%u’ directive output may be truncated writing between 1 and 2 bytes into a region of size between 1 and 12 [-Wformat-truncation=]
"%.11s-rx-%u", netdev->name, i);
^~
drivers/net/ethernet/cisco/enic/enic_main.c:1740:5: note: directive argument in the range [0, 16]
"%.11s-rx-%u", netdev->name, i);
^~~~~~~~~~~~~
drivers/net/ethernet/cisco/enic/enic_main.c:1738:4: note: ‘snprintf’ output between 6 and 18 bytes into a destination of size 16
snprintf(enic->msix[intr].devname,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
sizeof(enic->msix[intr].devname),
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"%.11s-rx-%u", netdev->name, i);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of busy poll, napi_complete_done returns false and does not
dequeue napi. In this case do not unmask the intr. We are guaranteed
napi is called again. This reduces unnecessary iowrites.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define ndo_features_check. Hw supports offload only for ipv4 inner and
ipv4 outer pkt.
Code refactor for setting inner tcp pseudo csum.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Defines enic_udp_tunnel_add/del for configuring vxlan tunnel offload.
enic supports offload of only one ipv4/udp port.
There are two modes that fw supports for vxlan offload.
mode 0: fcoe bit is set for encapsulated packet. fcoe_fc_crc_ok is set
if checksum of csum is ok. This bit is or of ip_csum_ok and
tcp_udp_csum_ok
mode 2: BIT(0) in rss_hash is set if it is encapsulated packet.
BIT(1) is set if outer_ip_csum_ok/
BIT(2) is set if outer_tcp_csum_ok
tcp_udp_csum_ok/ipv4_csum_ok is set if inner csum is OK.
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
napi_complete_done() allows to opt-in for gro_flush_timeout,
added back in linux-3.19, commit 3b47d30396
("net: gro: add a per device gro flush timer")
This allows for more efficient GRO aggregation without
sacrifying latencies.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The network device operation for reading statistics is only called
in one place, and it ignores the return value. Having a structure
return value is potentially confusing because some future driver could
incorrectly assume that the return value was used.
Fix all drivers with ndo_get_stats64 to have a void function.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Driver sets the skb l4/l3 hash based on NIC_CFG_RSS_HASH_TYPE_*,
which is bit mask. This is wrong. Hw actually provides us enum.
Use CQ_ENET_RQ_DESC_RSS_TYPE_* to set l3 and l4 hash type.
Fixes: bf751ba802 ("driver/net: enic: record q_number and rss_hash for skb")
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
i is defined as unsigned.
So print it with %u.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Driver sets vlan_feature to netdev->features as hardware supports all of
them on vlan interface.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
NAPI drivers no longer need to observe a particular protocol
to benefit from busy polling (CONFIG_NET_RX_BUSY_POLL=y)
napi_hash_add() and napi_hash_del() are automatically called
from core networking stack, respectively from
netif_napi_add() and netif_napi_del()
This patch depends on free_netdev() and netif_napi_del() being
called from process context, which seems to be the norm.
Drivers might still prefer to call napi_hash_del() on their
own, since they might combine all the rcu grace periods into
a single one, knowing their NAPI structures lifetime, while
core networking stack has no idea of a possible combining.
Once this patch proves to not bring serious regressions,
we will cleanup drivers to either remove napi_hash_del()
or provide appropriate rcu grace periods combining.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The affinity hint is used by the user space daemon, irqbalancer, to
indicate a preferred CPU mask for irqs. This patch sets the irq affinity
hint to local numa core first, when exausted we try non-local numa cores.
Also set tx xps cpus mask bassed on affinity hint.
v2: remove the global affinity policy.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current code invokes hang reset in case of error interrupt. We should
hang reset only in case of tx timeout. This because of the way hang reset
is implemented in firmware. Hang reset takes more firmware resources than
soft reset. Adaptor does not generate error interrupt in case of tx
timeout.
Hang reset only in case of tx timeout, in .ndo_tx_timeout. Do soft reset
otherwise. Introduce deferred work, enic_tx_hang_reset, to do hang reset.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some of the enic adaptors are know to generate spurious interrupts. When
error interrupt is generated, driver just resets the device. This patch
resets the device only when an error is occurred.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
err_out_vnic_unregister is used regardless of whether
SRIOV is enabled or not.
Reported-by: Jesse Brandeburg <jesse.brangeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
devcmd is an interface for driver to communicate with fw/adaptor. It
involves writing data to hardware registers and waiting for the result.
This mechanism does not scale well. The queuing of "no wait" devcmds is
done in firmware memory rather than on the host. Firmware memory is a
rather more scarce and valuable resource than host memory. A devcmd storm
from one vf can disrupt the service on other pf/vf. The lack of flow
control allows for possible denial of server from one VM to another.
Devcmd2 uses work queue to post the devcmds, just like tx work queue. This
allows better flow control.
Initialize devcmd2, if fails we fall back to devcmd1.
Also change the driver version.
Signed-off-by: N V V Satyanarayana Reddy <nalreddy@cisco.com>
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adaptive interrupt coalescing is available for msix. This patch adds the support
for msi poll. Interface for adaptive interrupt coalescing is already added in
driver. We just did not enable it for legacy intr & msi.
enic_calc_int_moderation() & enic_set_int_moderation() are defined as static
after enic_poll. Since enic_poll needs it, move both of these function
definitions above enic_poll. No change in functionality.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In enic_poll, we clean tx and rx queues, when low latency busy socket polling
is happening, enic_poll will only clean tx queue. After cleaning tx, it should
return total budget for re-poll.
There is a small window between vnic_intr_unmask() and enic_poll_unlock_napi().
In this window if an irq occurs and napi is scheduled on different cpu, it tries
to acquire enic_poll_lock_napi() and fails. Unlock napi_poll before unmasking
the interrupt.
v2:
Do not change tx wonk done behaviour. Consider only rx work done for completing
napi.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We use spinlock to access a single flag. We can avoid spin_locks by using
atomic variable and atomic_cmpxchg(). Use atomic_cmpxchg to set the flag
for idle to poll. And a simple atomic_set to unlock (set idle from poll).
In napi poll, if gro is enabled, we call napi_gro_receive() to deliver the
packets. Before we call napi_complete(), i.e while re-polling, if low
latency busy poll is called, we use netif_receive_skb() to deliver the packets.
At this point if there are some skb's held in GRO, busy poll could deliver the
packets out of order. So we call napi_gro_flush() to flush skbs before we
move the napi poll to idle.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We do not check the return value of enic_dev_stats_dump(). If allocation
fails, we will hit NULL pointer reference.
Return only if memory allocation fails. For other failures, we return the
previously recorded values.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/rocker/rocker.c
The rocker commit was two overlapping changes, one to rename
the ->vport member to ->pport, and another making the bitmask
expression use '1ULL' instead of plain '1'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the built-in function instead of memset.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We should complete notify_check before returning the credits. Once we return the
credits, adaptor may access the notify data.
Signed-off-by: Sujith Sankar <ssujith@cisco.com>
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
this patch fixes following sparse warnings:
enic_main.c:92:28: warning: symbol 'mod_table' was not declared. Should it be static?
enic_main.c:109:28: warning: symbol 'mod_range' was not declared. Should it be static?
enic_main.c:1306:5: warning: symbol 'enic_busy_poll' was not declared. Should it be static?
Signed-off-by: Lad, Prabhakar <prabhakar.csengg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
arch/arm/boot/dts/imx6sx-sdb.dts
net/sched/cls_bpf.c
Two simple sets of overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
With the commit d75b1ade56 ("net: less interrupt masking in NAPI") napi repoll
is done only when work_done == budget. When we are in busy_poll we return 0 in
napi_poll. We should return budget.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
