This patch makes cxgb3 invoke the GRO hooks instead of LRO. As
GRO has a compatible external interface to LRO this is a very
straightforward replacement.
I've kept the ioctl controls for per-queue LRO switches. However,
we should not encourage anyone to use these.
Because of that, I've also kept the skb construction code in
cxgb3. Hopefully we can phase out those per-queue switches
and then kill this too.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The lro manager's frag_align_pad setting was missing,
leading to misaligned access to the skb passed up
to the stack.
Tested-by: Rick Jones <rick.jones2@hp.com>
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I have a system with a Chelsio adapter (driven by cxgb3) whose ports are
part of a Linux bridge. Recently I updated the kernel and discovered
that things stopped working because cxgb3 was doing LRO on packets that
were passed into the bridge code for forwarding. (Incidentally, this
problem manifested itself in a strange way that made debugging a bit
interesting -- for some reason, the skb_warn_if_lro() check in bridge
didn't trigger and these LROed packets were forwarded out a forcedeth
interface, and caused the forcedeth transmit path to get stuck)
This is because cxgb3 has no way of keeping state for the LRO flag until
the interface is brought up, so if the bridging code disables LRO while
the interface is down, then cxgb3_up() will just reenable LRO, and on my
Debian system at least, the init scripts add interfaces to a bridge
before bringing the interfaces up.
Fix this by keeping track of each interface's LRO state in cxgb3 so that
when bridge disables LRO, it stays disabled in cxgb3_up() when the
interface is brought up. I did this by changing the rx_csum_offload
flag into a pair of bit flags; the effect of this on the rx_eth() fast
path is miniscule enough that it should be fine (eg on x86, a cmpb
instruction becomes a testb instruction).
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The accelerated iSCSI traffic could use a private IP address unknown to the OS:
- The IP address is required in both drivers to manage ARP requests and connection set up.
- Added an control call to retrieve the ip address.
- Reply to ARP requests dedicated to the private IP address.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Karen Xie <kxie@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement NIC Tx multiqueue.
Bump up driver version.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function is_pure_response() does "ntohl(var) & const" and then
essentially just tests whether the result is 0 or not; this can be done
more efficiently by computing "var & htonl(const)" instead and doing the
byte swap at compile time instead of run time.
This change slightly shrinks the compiled code; eg on x86-64 we save a
couple of bswapl instructions:
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-8 (-8)
function old new delta
t3_sge_intr_msix_napi 544 536 -8
and this also has the pleasant side effect of fixing a sparse warning:
drivers/net/cxgb3/sge.c:2313:15: warning: restricted degrades to integer
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add consistency in alloc_ring() parameter checking
to avoid potential memory leaks.
alloc_ring() callers are correct fo far.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The generic packet receive code takes care of setting
netdev->last_rx when necessary, for the sake of the
bonding ARP monitor.
Drivers need not do it any more.
Some cases had to be skipped over because the drivers
were making use of the ->last_rx value themselves.
Signed-off-by: David S. Miller <davem@davemloft.net>
when a fatal error occurs, bring ports down, reset the chip,
and bring ports back up.
Factorize code used for both EEH and fatal error recovery.
Fix timer usage when bringing up/resetting sge queue sets.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A SGE queue set timer might access registers while in EEH recovery,
triggering an EEH error loop. Stop all timers early in EEH process.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The generic lro code checks TCP flags/options.
Remove duplicate tests done in the driver.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER
architecture does:
This enables us to cleanly fix the Calgary IOMMU issue that some devices
are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423).
I think that per-device dma_mapping_ops support would be also helpful for
KVM people to support PCI passthrough but Andi thinks that this makes it
difficult to support the PCI passthrough (see the above thread). So I
CC'ed this to KVM camp. Comments are appreciated.
A pointer to dma_mapping_ops to struct dev_archdata is added. If the
pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's
NULL, the system-wide dma_ops pointer is used as before.
If it's useful for KVM people, I plan to implement a mechanism to register
a hook called when a new pci (or dma capable) device is created (it works
with hot plugging). It enables IOMMUs to set up an appropriate
dma_mapping_ops per device.
The major obstacle is that dma_mapping_error doesn't take a pointer to the
device unlike other DMA operations. So x86 can't have dma_mapping_ops per
device. Note all the POWER IOMMUs use the same dma_mapping_error function
so this is not a problem for POWER but x86 IOMMUs use different
dma_mapping_error functions.
The first patch adds the device argument to dma_mapping_error. The patch
is trivial but large since it touches lots of drivers and dma-mapping.h in
all the architecture.
This patch:
dma_mapping_error() doesn't take a pointer to the device unlike other DMA
operations. So we can't have dma_mapping_ops per device.
Note that POWER already has dma_mapping_ops per device but all the POWER
IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device
argument.
[akpm@linux-foundation.org: fix sge]
[akpm@linux-foundation.org: fix svc_rdma]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix bnx2x]
[akpm@linux-foundation.org: fix s2io]
[akpm@linux-foundation.org: fix pasemi_mac]
[akpm@linux-foundation.org: fix sdhci]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix sparc]
[akpm@linux-foundation.org: fix ibmvscsi]
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Muli Ben-Yehuda <muli@il.ibm.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Reset the chip when the PCI link goes down.
Preserve the napi structure when a sge qset's resources are freed.
Replay only HW initialization when the chip comes out of reset.
Signed-off-by: Divy Le ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Using iWARP with a Chelsio T3 NIC generates the following lockdep warning:
=================================
[ INFO: inconsistent lock state ]
2.6.25-rc6 #50
---------------------------------
inconsistent {softirq-on-W} -> {in-softirq-W} usage.
swapper/0 [HC0[0]:SC1[1]:HE0:SE0] takes:
(&adap->sge.reg_lock){-+..}, at: [<ffffffff880e5ee2>] cxgb_offload_ctl+0x3af/0x507 [cxgb3]
The problem is that reg_lock is used with plain spin_lock() in
drivers/net/cxgb3/sge.c but is used with spin_lock_irqsave() in
drivers/net/cxgb3/cxgb3_offload.c. This is technically a false
positive, since the uses in sge.c are only in the initialization and
cleanup paths and cannot overlap with any use in interrupt context.
The best fix is probably just to use spin_lock_irq() with reg_lock in
sge.c. Even though it's not strictly required for correctness, it
avoids triggering lockdep and the extra overhead of disabling
interrupts is not important at all in the initialization and cleanup
slow paths.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
The last change in the Tx queue stop mechanism opens a window
where the Tx queue might be stopped after pending credits
returned.
Tx credits are returned via a control message generated by the HW.
It returns tx credits on demand, triggered by a completion bit
set in selective transmit packet headers.
The current code can lead to the Tx queue stopped
with all pending credits returned, and the current frame
not triggering a credit return. The Tx queue will then never be
awaken.
The driver could alternatively request a completion for packets
that stop the queue. It's however safer at this point to go back
to the pre-existing behaviour.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
1. Add common code for stopping queue.
2. No need to call netif_stop_queue followed by netif_wake_queue (and
infact a netif_start_queue could have been used instead), instead
call stop_queue if required, and remove code under USE_GTS macro.
3. There is no need to check for netif_queue_stopped, as the network
core guarantees that for us (I am sure every driver could remove
that check, eg e1000 - I have tested that path a few billion times
with about a few hundred thousand qstops but the condition never
hit even once).
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
When PCI error recovery was added to cxgb3, a function t3_io_slot_reset()
was added. This function can call back into t3_prep_adapter() at any
time, so t3_prep_adapter() can no longer be marked __devinit.
This patch removes the __devinit annotation from t3_prep_adapter() and
all the functions that it calls, which fixes
WARNING: drivers/net/cxgb3/built-in.o(.text+0x2427): Section mismatch in reference from the function t3_io_slot_reset() to the function .devinit.text:t3_prep_adapter()
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
set_pci_drvdata() stores a pointer to the adapter,
not the net device.
Add missing softirq blocking in t3_mgmt_tx.
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>
Rather than hand-rolling our own prototype, make the code more
future-proof by using the standard irq_handler_t typedef.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Fix warnings from sparse related to shadowed variables and routines
that should be declared static.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Send small TX_DATA work requests as immediate data even when
there are fragments. this avoids doing multiple DMAs for
small fragmented packets.
The driver already implements this optimization for small
contiguous packets.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
cxgb3 used netdev_priv() and dev->priv for different purposes.
In 2.6.23, netdev_priv() == dev->priv, cxgb3 needs a fix.
This patch is a partial backport of Dave Miller's changes in the
net-2.6.24 git branch.
Without this fix, cxgb3 crashes on 2.6.23.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Streamline sge page management.
Fix dma mappings when buffers are recycled.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Fix netpoll handler to work with line interrupt, msi and msi-x.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
eth_type_trans() now sets skb->dev.
References to skb->dev should happen after it is called.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
To clearly state the intent of copying to linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
To clearly state the intent of copying from linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that it is also an offset from skb->head, reduces its size from 8 to 4 bytes
on 64bit architectures, allowing us to combine the 4 bytes hole left by the
layer headers conversion, reducing struct sk_buff size to 256 bytes, i.e. 4
64byte cachelines, and since the sk_buff slab cache is SLAB_HWCACHE_ALIGN...
:-)
Many calculations that previously required that skb->{transport,network,
mac}_header be first converted to a pointer now can be done directly, being
meaningful as offsets or pointers.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the places where we need a pointer to the transport header, it is
still legal to touch skb->h.raw directly if just adding to,
subtracting from or setting it to another layer header.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the quite common 'skb->h.raw - skb->data' sequence.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->h.raw = skb->data' operation, so that we can
later turn skb->h.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple cases:
skb->h.raw = skb->data;
skb->h.raw = {skb_push|[__]skb_pull}()
The next ones will handle the slightly more "complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the quite common 'skb->nh.raw - skb->data' sequence.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->nh.raw = skb->data' operation, so that we can
later turn skb->nh.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->mac.raw = skb->data' operation, so that we can
later turn skb->mac.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One less thing for drivers writers to worry about.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
