pasemi_mac: Remove SKB copy/recycle logic
It doesn't really buy us much, since copying is about as expensive
as the allocation in the first place. Just remove it for now.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: SKB unmap optimization
Avoid touching skb_shinfo() in the unmap path, since it turns out to
normally cause cache misses and delays. instead, save number of fragments
in the TX_RING_INFO structures since that's all that's needed anyway.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: Software-based LRO support
Implement LRO for pasemi_mac. Pretty straightforward.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: Improve RX interrupt mitigation
Currently the receive side interrupts will go off on the reception of
a packet, NAPI will poll the ring and keep polling as long as there's
a decent amount of packets to receive.
This is less than optimal, especially for LRO where it's better if we
have a more substantial amount of packets to process at once, to get
the real LRO benefits.
So, set the count threshold to a higher value and use the timeout feature
that will give us an interrupt even if not enough packets have come in
to set off the count threshold.
FIXME: It'd be real nice to have ethtool support for users to tune this
at runtime.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: Fix TX cleaning
This is a bit awkward. We don't have a timer-delayed interrupt on TX
complete, but we have a count threshold. So set that reasonably high
(32 packets), and schedule the NAPI poll when it goes off. Also bump a
regular timer that will take care of rotting packets for the last 1..31
ones in case we don't trigger a TX interrupt (and there's no RX activity
that would otherwise trigger the poll).
The longer-term fix is to separate TX from RX NAPI and do two separate
poll loops.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: performance tweaks
* Seems like we do better with a smaller RX ring, probably because chances of
still having the SKB cached are better
* Const-ify variables to get better code generation and fewer reloads
* Move prefetching around a little, and try to prefetch the whole SKB
* Set NETIF_F_HIGHDMA
* Misc other minor tweaks
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: Convert to new dma library
Convert the pasemi_mac driver to the new platform global DMA manaagement
library. This also does a couple of other minor cleanups w.r.t. channel
management.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: Move register definitions to include/asm-powerpc
Move the common register formats and descriptor layouts from
drivers/net/pasemi_mac.h to include/asm-poewrpc/pasemi_dma.h
Previously only the ethernet driver was using them, but other drivers
are coming up that will also use them, so it makes sense to share the
constants.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
pasemi_mac: RX/TX ring management cleanup
Prepare a bit for supporting multiple TX queues by cleaning up some
of the ring management and shuffle things around a bit.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
When using larger MTU's sky2 driver changes from allocating one
data area, to using multiple pages. The threshold for this was based on
a heuristic where the cost of a single allocation is bigger than one
page. Since the allocator has changed, this heuristic is now incorrect;
instead just make the threshold be when the total size of the allocation
is greater than one page.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
The sky2 driver was not aligning the IP header on receive buffers.
This workaround is only needed on hardware with broken FIFO, newer chips
without FIFO can just DMA to unaligned address.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
- whitespaces vs tabs
- use 80 cols
- use if_mii
- use netdev_priv
- remove useless cast to void *
- PCI device id does not need to be globally available
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
drivers/net/r6040.c: In function 'rx_buf_alloc':
drivers/net/r6040.c:262: warning: passing argument 2 of 'pci_map_single' makes pointer from integer without a cast
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Disable PEX errors. The HW generates false positives.
Update RSS hash function to a symmetric algorithm.
Update T3C HW support
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Do not use skb->cb to stash unmap info,
save the info to the descriptor state.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Bump up FW version to 5.0.
Do not downgrade FW within the same major version range.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Subdevice ID is not necessarily set to 1.
Use wild card for PCI device matching
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
dm_write_async and dm_write_reg_async are almost identical.
Move common functionality to dm_write_async_helper (saves ~256b).
Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This patch #if 0's the following unused functions:
- espi.c:t1_espi_set_misc_ctrl()
- sge.c:t1_sched_set_max_avail_bytes()
- sge.c:t1_sched_set_drain_bits_per_us()
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
- Support to add/delete/store/restore 64 and 128 Ethernet addresses for Xframe I and Xframe II respectively.
Signed-off-by: Sreenivasa Honnur <sreenivasa.honnur@neterion.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for configuring secondary unicast addresses. Unicast
addresses take precendece over multicast addresses when filling
the exact address filters to avoid going to promiscous mode.
When more unicast addresses are present than filter slots,
unicast filtering is disabled and all slots can be used for
multicast addresses.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add functions for reg_pattern_test and reg_set_and check
Changed macros to use these functions
Compiled x86, untested
Size decreased ~2K
old:
$ size drivers/net/e1000e/ethtool.o
text data bss dec hex filename
14461 0 0 14461 387d drivers/net/e1000e/ethtool.o
new:
$ size drivers/net/e1000e/ethtool.o
text data bss dec hex filename
12498 0 0 12498 30d2 drivers/net/e1000e/ethtool.o
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Minimal macro to function conversion in e1000_ethtool.c
Adds functions reg_pattern_test and reg_set_and_check
Changes REG_PATTERN_TEST and REG_SET_AND_CHECK macros
to call these functions.
Saves ~2.5KB
Compiled x86, untested (no hardware)
old:
$ size drivers/net/e1000/e1000_ethtool.o
text data bss dec hex filename
16778 0 0 16778 418a drivers/net/e1000/e1000_ethtool.o
new:
$ size drivers/net/e1000/e1000_ethtool.o
text data bss dec hex filename
14128 0 0 14128 3730 drivers/net/e1000/e1000_ethtool.o
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
formerly e1000/e1000e only updated traffic counters once every
2 seconds with the register values of bytes/packets. With newer
code however in the interrupt and polling code we can real-time
fill in these values in the netstats struct for users to see.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
formerly e1000/e1000e only updated traffic counters once every
2 seconds with the register values of bytes/packets. With newer
code however in the interrupt and polling code we can real-time
fill in these values in the netstats struct for users to see.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for the RDC R6040 MAC we can find in the RDC
R-321x System-on-chips.
Signed-off-by: Sten Wang <sten.wang@rdc.com.tw>
Signed-off-by: Daniel Gimpelevich <daniel@gimpelevich.san-francisco.ca.us>
Signed-off-by: Florian Fainelli <florian.fainelli@telecomint.eu>
On the systems that have cache incoherent DMA, including ARM, there
is a race condition between software allocating a new receive buffer
and hardware writing into a buffer. The two race on touching the last
Receive Frame Descriptor (RFD). It has its el-bit set and its next
link equal to 0. When hardware encounters this buffer it attempts to
write data to it and then update Status Word bits and Actual Count in
the RFD. At the same time software may try to clear the el-bit and
set the link address to a new buffer.
Since the entire RFD is once cache-line, the two write operations can
collide. This can lead to the receive unit stalling or interpreting
random memory as its receive area.
The fix is to set the el-bit on and the size to 0 on the next to last
buffer in the chain. When the hardware encounters this buffer it stops
and does not write to it at all. The hardware issues an RNR interrupt
with the receive unit in the No Resources state. Software can write
to the tail of the list because it knows hardware will stop on the
previous descriptor that was marked as the end of list.
Once it has a new next to last buffer prepared, it can clear the el-bit
and set the size on the previous one. The race on this buffer is safe
since the link already points to a valid next buffer and the software
can handle the race setting the size (assuming aligned 16 bit writes
are atomic with respect to the DMA read). If the hardware sees the
el-bit cleared without the size set, it will move on to the next buffer
and skip this one. If it sees the size set but the el-bit still set,
it will complete that buffer and then RNR interrupt and wait.
Signed-off-by: David Acker <dacker@roinet.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Olof Johansson