Use round_jiffies_relative to align the ehea workqueue and avoid
extra wakeups.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we enable multiqueue by default the ehea driver is using
quite a lot of memory for its buffer pools. With 4 queues we
consume 64MB in the jumbo packet ring, 16MB in the medium packet
ring and 16MB in the tiny packet ring.
We should only fill the jumbo ring once the MTU is increased but
for now halve it's size so it consumes 32MB. Also reduce the tiny
packet ring, with 4 queues we had 16k entries which is overkill.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
"remote_list" is of type
struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
Probably a copy'n'paste error.
Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
To ease skb->truesize sanitization, its better to be able to localize
all references to skb frags size.
Define accessors : skb_frag_size() to fetch frag size, and
skb_frag_size_{set|add|sub}() to manipulate it.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The tcp_end field is not actually used by the hardware, so there
is no need to set it.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add GRO support to the ehea driver.
v3:
[cascardo] no need to enable GRO, since it's enabled by default
[cascardo] vgrp was removed in the vlan cleanup
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for adding GRO to ehea, remove LRO.
v3:
[cascardo] fixed conflict with vlan cleanup
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Switch to using ndo_get_stats64 to get 64bit statistics.
v3:
[cascardo] use rtnl_link_stats64 as port stats
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The queue macros are many levels deep and it makes it harder to
work your way through them when many of the versions are unused.
Remove the unused versions.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a nonlinear skb fits within the immediate area, use skb_copy_bits
instead of copying the frags by hand.
v3:
[cascardo] fixed conflict with use of skb frag API
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
write_swqe2_TSO and write_swqe2_nonTSO are almost identical.
For TSO we have to set the TSO and mss bits in the wqe and we only
put the header in the immediate area, no data. Collapse both
functions into write_swqe2_immediate.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on a patch from Michael Ellerman, clean up a significant
portion of the transmit path. There was a lot of duplication here.
Even worse, we were always checksumming tx packets and ignoring the
skb->ip_summed field.
Also remove NETIF_F_FRAGLIST from dev->features, I'm not sure why
it was enabled.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ehea adapter has a mode where it will avoid partial cacheline DMA
writes on receive by always padding packets to fall on a cacheline
boundary.
Unfortunately we currently aren't allocating enough space for a full
ethernet MTU packet to be rounded up, so this optimisation doesn't hit.
It's unfortunate that the next largest packet size exposed by the
hypervisor interface is 2kB, meaning our skb allocation comes out of a
4kB SLAB. However the performance increase due to this optimisation is
quite large and my TCP stream numbers increase from 900MB to 1000MB/sec.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We weren't enabling any VLAN features so we missed out on checksum
offload and TSO when using VLANs. Enable them.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems like the ehea xmit routine and an ethtool change of TSO
mode could race, resulting in corrupt packets. Checking gso_size
is enough and we can use the helper function.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The num_tx_qps module option allows a user to configure a different
number of tx and rx queues. Now the networking stack is multiqueue
aware it makes little sense just to enable the tx queues and not the
rx queues so remove the option.
v3:
[cascardo] fixed conflict with get_stats change
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 18604c5485 (ehea: NAPI multi queue TX/RX path for SMP) added
driver specific logic for exiting napi mode. I'm not sure what it was
trying to solve and it should be up to the network stack to decide when
we are done polling so remove it.
v3:
[cascardo] Fixed extra parentheses.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ehea driver had some multiqueue support but was missing the last
few years of networking stack improvements:
- Use skb_record_rx_queue to record which queue an skb came in on.
- Remove the driver specific netif_queue lock and use the networking
stack transmit lock instead.
- Remove the driver specific transmit queue hashing and use
skb_get_queue_mapping instead.
- Use netif_tx_{start|stop|wake}_queue where appropriate. We can also
remove pr->queue_stopped and just check the queue status directly.
- Print all 16 queues in the ethtool stats.
We now enable multiqueue by default since it is a clear win on all my
testing so far.
v3:
[cascardo] fixed use_mcs parameter description
[cascardo] set ehea_ethtool_stats_keys as const
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the deprecated NETIF_F_LLTX feature. Since the network stack
now provides the locking we can remove the driver specific
pr->xmit_lock.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Breno Leitao <leitao@linux.vnet.ibm.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently ehea ndo_get_stats can sleep in two places, in a hcall
and in a GFP_KERNEL alloc, which is not correct.
This patch creates a delayed workqueue that grabs the information each 1
sec from the hardware, and place it into the device structure, so that,
.ndo_get_stats quickly returns the device structure statistics block.
Signed-off-by: Breno Leitao <brenohl@br.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Santiago Leon <santil@linux.vnet.ibm.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on finds for Stephen Rothwell, where current defconfig's
enable a ethernet driver and it is not compiled due to the newly
added NET_VENDOR_* component of Kconfig.
This patch enables all the "new" Kconfig options so that current
defconfig's will continue to compile the expected drivers. In
addition, by enabling all the new Kconfig options does not add
any un-expected options.
CC: Stephen Rothwll <sfc@canb.auug.org.au>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In commit 9aa3283595 (ehea/ibm*: Move the
IBM drivers) the IBM_NEW_EMAC* were renames to IBM_EMAC*
The conversion was incomplete so that even if the driver was added to
the .config it wasn't built, but there were no errors). In this commit
we also update the various defconfigs that use EMAC to use the new
Kconfig symbol, and explicitly add the NET_VENDOR_IBM guard.
We do not explicitly select the Kconfig dependencies, as this would force
EMAC on. Doing it in the defconfig allows more flexibility.
Tested on a canyondlands board.
Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the IBM drivers into drivers/net/ethernet/ibm/ and make the
necessary Kconfig and Makefile changes.
- Renamed ibm_new_emac to emac
- Cleaned up Makefile and Kconfig options which referred to
IBM_NEW_EMAC to IBM_EMAC
- ibmlana driver is a National Semiconductor SONIC driver so
it was not moved
CC: Christoph Raisch <raisch@de.ibm.com>
CC: Santiago Leon <santil@linux.vnet.ibm.com>
CC: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CC: David Gibson <dwg@au1.ibm.com>
CC: Kyle Lucke <klucke@us.ibm.com>
CC: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
