This patch adds error reporting to the tg3_phydsp_write() function and
converts a few more locations to use this function over the inlined
equivalent.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
smp_mb() inside tg3_tx_avail() is used twice in the normal
tg3_start_xmit() path (see illustration below). The full memory
barrier is only necessary during race conditions with tx completion.
We can speed up the tx path by replacing smp_mb() in tg3_tx_avail()
with a compiler barrier. The compiler barrier is to force the
compiler to fetch the tx_prod and tx_cons from memory.
In the race condition between tg3_start_xmit() and tg3_tx(),
we have the following situation:
tg3_start_xmit() tg3_tx()
if (!tg3_tx_avail())
BUG();
...
if (!tg3_tx_avail())
netif_tx_stop_queue(); update_tx_index();
smp_mb(); smp_mb();
if (tg3_tx_avail()) if (netif_tx_queue_stopped() &&
netif_tx_wake_queue(); tg3_tx_avail())
With smp_mb() removed from tg3_tx_avail(), we need to add smp_mb() to
tg3_start_xmit() as shown above to properly order netif_tx_stop_queue()
and tg3_tx_avail() to check the ring index. If it is not strictly
ordered, the tx queue can be stopped forever.
This improves performance by about 3% with 2 ports running
bi-directional 64-byte packets.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ASPM workaround setting obtained from NVRAM only works with devices
older than 5717. This patch enforces the restriction.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch changes the code to only manage the PCIe gphy power for
CPMU-less devices only. The CPMU takes over management for newer
chips.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The phy test register location has been repurposed for 5717+ devices.
This patch changes the code to avoid this location for these devices.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch creates a TG3_FLG3_5717_PLUS flag to collectively describe
the set of changes in the ASIC that will apply to all future chip
revisions.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TSS flag needs to be turned off during tg3_close(). If the device
fails to allocate more than one MSI-X vector the next time the device is
brought up, transmits will fail.
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tg3 commit e712699734 entitled
"tg3: Preserve PCIe MPS setting for new devs" attempted to ensure the
PCIe link negotiated Maximum Payload Size (MPS) setting was 128 bytes
for all devices that didn't support higher speeds. The 5784 device was
mistakenly added to this list when it shouldn't have. This patch
removes the 5784 ASIC rev devices from that list.
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
use skb->len for accounting as xt_quota does. Since nf_conntrack works
at the network layer, skb_network_offset should always returns ZERO.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The tuple got from unique_tuple() doesn't need to be really unique, so the
check for the unique tuple isn't necessary, when there isn't any other
choice. Eliminating the unnecessary nf_nat_used_tuple() can save some CPU
cycles too.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The only user of unique_tuple() get_unique_tuple() doesn't care about the
return value of unique_tuple(), so make unique_tuple() return void (nothing).
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This removes duplicate code by providing a default implementation
which is used by 3 of the 4 modules that provide these call.
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
IPVS was merged into the kernel quite a long time ago and
has been seeing wide-spread production use for even longer.
It seems appropriate for it to be no longer tagged as EXPERIMENTAL
Signed-off-as: Simon Horman <horms@verge.net.au>
Signed-off-by: Patrick McHardy <kaber@trash.net>
some users of nf_ct_ext_exist() know ct->ext isn't NULL. For these users, the
check for ct->ext isn't necessary, the function __nf_ct_ext_exist() can be
used instead.
the type of the return value of nf_ct_ext_exist() is changed to bool.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
We currently disable BH for the whole duration of get_counters()
On machines with a lot of cpus and large tables, this might be too long.
We can disable preemption during the whole function, and disable BH only
while fetching counters for the current cpu.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
If user misconfigures ingress and causes a redirection loop, don't
overwhelm the log. This is also a error case so make it unlikely.
Found by inspection, luckily not in real system.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The other calls to kmalloc in the same function use GFP_ATOMIC, and indeed
two locks are held within the body of the function.
The semantic match that finds this problem is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@@ identifier f; @@
*f(...,GFP_ATOMIC,...)
... when != spin_unlock(...)
when != read_unlock(...)
when != write_unlock(...)
when != read_unlock_irq(...)
when != write_unlock_irq(...)
when != read_unlock_irqrestore(...)
when != write_unlock_irqrestore(...)
when != spin_unlock_irq(...)
when != spin_unlock_irqrestore(...)
*f(...,GFP_KERNEL,...)
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
GET_STATS request uses the same memory region as the response.
If a new request for get stats is fired before the response for
the previous get_stats request is received, the response will
corrupt the new request, causing the f/w to misbehave.
Signed-off-by: Somnath K <somnathk@serverengines.com>
Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move be_check_sriov_fn_type to appropriate place to correctly determine
if the be2net driver needs to work as a VF driver or a PF driver.
Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Matt Carlson
David S. Miller
Changli Gao
Simon Horman
Eric Dumazet
Stephen Hemminger
Julia Lawall
Ajit Khaparde