Changing MTU size of an xgmac network interface while it is active can
cause a panic like
skbuff: skb_over_panic: text:c03bc62c len:1090 put:1090 head:edfb6900 data:edfb6942 tail:0xedfb6d84 end:0xedfb6bc0 dev:eth0
------------[ cut here ]------------
kernel BUG at net/core/skbuff.c:126!
Internal error: Oops - BUG: 0 [#1] SMP ARM
Modules linked in:
CPU: 0 PID: 762 Comm: python Tainted: G W 3.10.0-00015-g3e33cd7 #309
task: edcfe000 ti: ed67e000 task.ti: ed67e000
PC is at skb_panic+0x64/0x70
LR is at wake_up_klogd+0x5c/0x68
This happens because xgmac_change_mtu modifies dev->mtu before the
network interface is quiesced. And thus there still might be buffers
in use which have a buffer size based on the old MTU.
To fix this I moved the change of dev->mtu after the call to
xgmac_stop.
Another modification is required (in xgmac_stop) to ensure that
xgmac_xmit is really not called anymore (xgmac_tx_complete might wake
up the queue again).
I've tested the fix by switching MTU size every second between 600 and
1500 while network traffic was going on. The test box survived a test
of several hours (until I've stopped it) whereas w/o this fix above
panic occurs after several minutes (at most).
Change since v1:
- remove call to netif_stop_queue at beginning of xgmac_stop
- use netif_tx_disable instead of locking+netif_stop_queue
Signed-off-by: Andreas Herrmann <andreas.herrmann@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Highbank and Midway xgmac h/w have different number of MAC address filter
registers with 7 and 31, respectively. Highbank has been wrong, so fix it
and detect the number of filter registers at run-time. Unfortunately,
the version register is the same on both SOCs, so simply test if write to
the last filter register will take a value. It always reads as 0 if not.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even in promiscuous mode, we need to add filter addresses for correct
operation. This fixes silent failures when using a bridge and adding
addresses using the "bridge fdb add" command.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 2ee68f621a (net: calxedaxgmac: fix various errors in
xgmac_set_rx_mode), a fix to clean-up old address entries was added.
However, the loop to zero out the entries failed to increment the register
address resulting in only 1 entry getting cleared. Fix this to correctly
use the loop index. Also, the end of the loop condition was off by 1 and
should have been <= rather than <.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On a DMA mapping error in xgmac_xmit, we should simply free the skb and
return NETDEV_TX_OK rather than -EIO. In the case of errors in mapping
frags, we need to undo everything that has been setup.
Reported-by: Andreas Herrmann <andreas.herrmann@calxeda.com>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the mismatch in the DMA mapping and unmapping sizes for receive. The
unmap size must be equal to the map size and should not be the actual
received frame length. The map size should also be adjusted by the
NET_IP_ALIGN size since the h/w buffer size (dma_buf_sz) includes this
offset.
Also, add a missing dma_mapping_error check in xgmac_rx_refill.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rx_sa_filter_fail and tx_undeflow events are unused and impossible
to occur based on how the h/w is used. We never filter on source MAC
address and TX store and forward mode prevents underflow events.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix xgmac_set_rx_mode to handle several conditions that were not handled
correctly as Lennert Buytenhek describes:
If we have, say, 100 unicast addresses, and 5 multicast addresses, the
unicast address count check will evaluate to true, and set use_hash to
true. The multicast address check will however evaluate to false, and
use_hash won't be set to true again, and XGMAC_FRAME_FILTER_HMC won't
be OR'd into XGMAC_FRAME_FILTER, but since use_hash was still true
from the unicast check, netdev_for_each_mc_addr() will program the
multicast addresses into the hash table instead of using the MAC
address registers, but since the HMC bit wasn't set, the hash table
won't be checked for multicast addresses on receive, and we'll stop
receiving multicast packets entirely.
Also, there is no code that zeroes out MAC address registers reg..31
at the end of this function, meaning that under the right conditions,
unicast/multicast addresses that were previously in the filter but
were then deleted won't be cleared out.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a race condition where the interrupt handler may have called
napi_schedule before napi_enable is called. This would disable interrupts
and never actually schedule napi to run.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the xgmac transmit start and completion work locklessly, it is
possible for xgmac_xmit to stop the tx queue after the xgmac_tx_complete
has run resulting in the tx queue never being woken up. Fix this by
ensuring that ring buffer index updates are visible and recheck the ring
space after stopping the queue. Also fix an off-by-one bug where we need
to stop the queue when the ring buffer space is equal to MAX_SKB_FRAGS.
The implementation used here was copied from
drivers/net/ethernet/broadcom/tg3.c.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Reviewed-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ensure that the descriptor writes are visible before the ring buffer head
is updated. Since writel is a barrier, we can simply update the head after
the writel.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TX completion code may have freed an skb before the entire sg list
was transmitted. The DMA unmap calls for the fragments could also get
skipped. Now set the skb pointer on every entry in the ring, not just
the head of the sg list. We then use the FS (first segment) bit in the
descriptors to determine skb head vs. fragment.
This also fixes similar bug in xgmac_free_tx_skbufs where clean-up of
a sg list that wraps at the end of the ring buffer would not get
unmapped.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is possible for the xgmac_tx_complete to run concurrently with
xgmac_tx_err since there are no locks. Fix this by moving the tx
error handling to a workqueue so we can disable napi while we reset
the transmitter.
Reported-by: Andreas Herrmann <andreas.herrmann@calxeda.com>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
xgmac_desc_get_buf_len appears to have a copy/paste error. flags is the
wrong field to read. We should be reading buf_size field. cpu_to_le32
should also be le32_to_cpu. This never really mattered as this function
is only used for DMA mapping calls which happen to be nops with coherent
DMA.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The xgmac does not actually handle frag lists, so this option should not
be set. It does not appear to have had any impact though.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver core clears the driver data to NULL after device_release
or on probe failure, since commit 0998d06310
(device-core: Ensure drvdata = NULL when no driver is bound).
Thus, it is not needed to manually clear the device driver data to NULL.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Acked-by: Rob Herring <rob.herring@calxeda.com>
Acked-by: Roland Stigge <stigge@antcom.de>
Acked-by: Mugunthan V N <mugunthanvnm@ti.com>
Reviewed-by: H Hartley Sweeten <hsweeten@visionengravers.com>
Tested-by: Roland Stigge <stigge@antcom.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
If NO_DMA=y:
drivers/built-in.o: In function `xgmac_xmit':
drivers/net/ethernet/calxeda/xgmac.c:1102: undefined reference to `dma_mapping_error'
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Rob Herring <rob.herring@calxeda.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
The "changed" variable should be a 64 bit type, otherwise it can't store
all the features. The way the code is now the test for whether
NETIF_F_RXCSUM changed is always false and we return immediately.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
SIMPLE_DEV_PM_OPS macro can handle !CONFIG_PM_SLEEP case nicely, so there is no
need to define PM_OPS for both CONFIG_PM_SLEEP and !CONFIG_PM_SLEEP cases.
Remove the unneeded definitions.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
WOL is broken because the magic packet status bit is getting set rather
than the enable bit. The PMT interrupt is not getting serviced because
the PMT interrupt is also enabled on the global interrupt, but not
cleared by the global interrupt and the global interrupt is higher
priority. This fixes both of these issues to get WOL working.
There's still a problem with receive after resume, but at least now we
can wake-up.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If skb allocation for the rx ring fails repeatedly, we can reach a point
were the ring is empty. In this condition, the driver is out of sync with
the h/w. While this has always been possible, the removal of the skb
recycling seems to have made triggering this problem easier.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bring in the 'net' tree so that we can get some ipv4/ipv6 bug
fixes that some net-next work will build upon.
Signed-off-by: David S. Miller <davem@davemloft.net>
The xgmac driver assumes 1 frame per descriptor. If a frame larger than
the descriptor's buffer size is received, the frame will spill over into
the next descriptor. So check for received frames that span more than one
descriptor and discard them. This prevents a crash if we receive erroneous
large packets.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
NET_ADDR_SET is set in dev_set_mac_address() no need to alter
dev->addr_assign_type value in drivers.
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
On gcc 4.7, we will get alignment traps in the ip stack if we don't align
the ip headers on receive. The h/w can support this, so use ip aligned
allocations.
Cut down the unnecessary padding on the allocation. The buffer can start on
any byte alignment, but the size including the begining offset must be 8
byte aligned. So the h/w buffer size must include the NET_IP_ALIGN offset.
Thanks to Eric Dumazet for the initial patch highlighting the padding issues.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Only generate tx interrupts on every ring size / 4 descriptors. Move the
netif_stop_queue call to the end of the xmit function rather than
checking at the beginning.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The interrupts have already been cleared, so we don't need to clear them
again. Also, we could miss interrupts if they are cleared, but we don't
process the packet.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The standard readl/writel accessors involve a spinlock and cache sync
operation on ARM platforms with an outer cache. Only DMA triggering
accesses need this, so use the raw variants instead in the critical paths.
The relaxed variants would be more appropriate, but don't exist on all
arches.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New received frames will trigger the rx DMA to poll the DMA descriptors,
so there is no need to tell the h/w to poll. We also want to enable
dropping frames from the fifo when there is no buffer.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enable the tx dma to start reading the next frame while sending the current
frame.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Over time, skb recycling infrastructure got litle interest and
many bugs. Generic rx path skb allocation is now using page
fragments for efficient GRO / TCP coalescing, and recyling
a tx skb for rx path is not worth the pain.
Last identified bug is that fat skbs can be recycled
and it can endup using high order pages after few iterations.
With help from Maxime Bizon, who pointed out that commit
87151b8689 (net: allow pskb_expand_head() to get maximum tailroom)
introduced this regression for recycled skbs.
Instead of fixing this bug, lets remove skb recycling.
Drivers wanting really hot skbs should use build_skb() anyway,
to allocate/populate sk_buff right before netif_receive_skb()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Maxime Bizon <mbizon@freebox.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enabling RX cut-thru mode yields better performance as received frames
start getting written to memory before a whole frame is received.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Increase the number of outstanding read and write AXI transactions from 1
to 8 for better performance.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix intermittent hangs in xgmac_rx_refill. If a ring buffer entry already
had an skb allocated, then xgmac_rx_refill would get stuck in a loop. This
can happen on a rx error when we just leave the skb allocated to the entry.
[ 7884.510000] INFO: rcu_preempt detected stall on CPU 0 (t=727315 jiffies)
[ 7884.510000] [<c0010a59>] (unwind_backtrace+0x1/0x98) from [<c006fd93>] (__rcu_pending+0x11b/0x2c4)
[ 7884.510000] [<c006fd93>] (__rcu_pending+0x11b/0x2c4) from [<c0070b95>] (rcu_check_callbacks+0xed/0x1a8)
[ 7884.510000] [<c0070b95>] (rcu_check_callbacks+0xed/0x1a8) from [<c0036abb>] (update_process_times+0x2b/0x48)
[ 7884.510000] [<c0036abb>] (update_process_times+0x2b/0x48) from [<c004e8fd>] (tick_sched_timer+0x51/0x94)
[ 7884.510000] [<c004e8fd>] (tick_sched_timer+0x51/0x94) from [<c0045527>] (__run_hrtimer+0x4f/0x1e8)
[ 7884.510000] [<c0045527>] (__run_hrtimer+0x4f/0x1e8) from [<c0046003>] (hrtimer_interrupt+0xd7/0x1e4)
[ 7884.510000] [<c0046003>] (hrtimer_interrupt+0xd7/0x1e4) from [<c00101d3>] (twd_handler+0x17/0x24)
[ 7884.510000] [<c00101d3>] (twd_handler+0x17/0x24) from [<c006be39>] (handle_percpu_devid_irq+0x59/0x114)
[ 7884.510000] [<c006be39>] (handle_percpu_devid_irq+0x59/0x114) from [<c0069aab>] (generic_handle_irq+0x17/0x2c)
[ 7884.510000] [<c0069aab>] (generic_handle_irq+0x17/0x2c) from [<c000cc8d>] (handle_IRQ+0x35/0x7c)
[ 7884.510000] [<c000cc8d>] (handle_IRQ+0x35/0x7c) from [<c033b153>] (__irq_svc+0x33/0xb8)
[ 7884.510000] [<c033b153>] (__irq_svc+0x33/0xb8) from [<c0244b06>] (xgmac_rx_refill+0x3a/0x140)
[ 7884.510000] [<c0244b06>] (xgmac_rx_refill+0x3a/0x140) from [<c02458ed>] (xgmac_poll+0x265/0x3bc)
[ 7884.510000] [<c02458ed>] (xgmac_poll+0x265/0x3bc) from [<c029fcbf>] (net_rx_action+0xc3/0x200)
[ 7884.510000] [<c029fcbf>] (net_rx_action+0xc3/0x200) from [<c0030cab>] (__do_softirq+0xa3/0x1bc)
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix net tx watchdog timeout recovery. The descriptor ring was reset,
but the DMA engine was not reset to the beginning of the ring.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use eth_hw_addr_random() instead of calling random_ether_addr()
to set addr_assign_type correctly to NET_ADDR_RANDOM.
Reset the state to NET_ADDR_PERM as soon as the MAC get
changed via .ndo_set_mac_address.
v2: adapt to renamed eth_hw_addr_random()
Signed-off-by: Danny Kukawka <danny.kukawka@bisect.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make local function static, make ethtool_ops const.
Compile tested only.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix allyesconfig build on architectures without IOMEM:
drivers/net/ethernet/calxeda/xgmac.c:1800:2:
error: implicit declaration of function 'iounmap' [-Werror=implicit-function-declaration]
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Rob Herring <rob.herring@calxeda.com>
Acked-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for the XGMAC 10Gb ethernet device in the Calxeda Highbank
SOC.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andreas Herrmann