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: NStephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The 8169/8110SC currently announces itself as:
[...]
eth0: RTL8169sc/8110sc at 0x........, ..:..:..:..:..:.., XID 18000000 IRQ ..
                                                             ^^^^^^^^
It uses RTL_GIGA_MAC_VER_05 and this part of the changeset can cut
its performance by a factor of 2~2.5 as reported by Timo.

(the driver includes code just before the hunk to write the ChipCmd
register when mac_version == RTL_GIGA_MAC_VER_0[1-4])
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Timo Jantunen <jeti@welho.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

The 8168 ignores the requests to fetch the Tx descriptors when
the relevant TxPoll bit is already set. It easily kills the
performances of the 8168. David Gundersen has noticed that it
is enough to wait for the completion of the DMA transfer (NPQ
bit is cleared) before writing the TxPoll register again.

The extra IO traffic added by the proposed workaround could be
minimalized but it is not a high-priority task.

Fix for:
http://bugzilla.kernel.org/show_bug.cgi?id=7924
http://bugzilla.kernel.org/show_bug.cgi?id=8688
(http://bugzilla.kernel.org/show_bug.cgi?id=7555 ?)
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: David Gundersen <gundy@iinet.net.au>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

The phys of the 8110SC (RTL_GIGA_MAC_VER_{05/06}) act abnormally in
gigabit mode if they are applied the parameters in rtl8169_hw_phy_config
which actually aim the 8110S/SB.

It is ok to return early from rtl8169_hw_phy_config as it does not
apply to the 8101 and 8168 families.
Signed-off-by: NEdward Hsu <edward_hsu@realtek.com.tw>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>

Theory  : though needless, it should not have hurt.
Practice: it does not play nice with DEBUG_SHIRQ + LOCKDEP + UP
(see https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=242572).

The patch makes sense in itself but I should dig why it has an effect
on #242572 (assuming that NAPI do not change in a near future).
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Fix extracted from Realtek's driver (8.002.00/20070713) for the PHY
attached to 8111/8168b chipsets.

The check against mac_version is just usual paranoia during the bugfix
period of the kernel cycle. -- FR

Tested on Asus M2A-VM motherboard by Roger So.
No regression on my Asrock 945G DVI either (built-in 8168 + 2x8169).
Signed-off-by: NRoger So <roger.so@gmail.com>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

All drivers implement ethtool get_perm_addr the same way -- by calling
the generic function.  So we can inline the generic function into the
caller and avoid going through the drivers.
Signed-off-by: NMatthew Wilcox <matthew@wil.cx>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Mark variables in drivers/* with uninitialized_var() if such a warning
appears, and analysis proves that the var is initialized properly on all
paths it is used.
Signed-off-by: NJeff Garzik <jeff@garzik.org>

It does not really make sense to update the RX config register
before the mac filtering registers.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Merged from Realtek's r8169-6.001 driver.

I have added some locking to protect against the arp monitoring
timer in the bonding driver. Accessing the configuration registers
is otherwise performed under RTNL locking.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

It does not cost much and it will ease the identification of (so far)
unknown devices.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Signed-off-by: NRolf Eike Beer <eike-kernel@sf-tec.de>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Align the IP header when the chipset can DMA at any location (plain 0x8169).
Otherwise (0x8136/0x8168) obey the constraint imposed by the hardware.

This patch complements the previous alignment rework done for copybreak.

Original idea from Philip Craig <philipc@snapgear.com>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Philip Craig <philipc@snapgear.com>
Cc: Mike Isely <isely@pobox.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

No functionnal change:
- trim the old history log
- whitespace/indent/case police
- unsigned int where signedness does not matter
- removal of obsolete assert
- needless cast from void * (dev_instance)
- remove dead code once related to power management
- use netdev_alloc_skb.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Stephen Hemminger <shemminger@linux-foundation.org>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

It has been documented as deprecated:
- in MODULE_PARM_DESC since may 2005 ;
- at the top of the source file and in printk since june 2004.

Good bye.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Extracted from version 1.001.00 of Realtek's r8101.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

The rx copybreak part is straightforward.

The align field in struct rtl_cfg_info is related to the alignment
requirements of the DMA operation. Its value is set at 2 to limit the
scale of possible regression but my old v1.21 8169 datasheet claims a
8 bytes requirements (which never appeared in the driver, of course)
and the 8101/8168 go with a plain 8 bytes alignment.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

This one includes:

- more tweaks to rtl_hw_start_8168

- a work around for a Rx FiFO overflow issue on the 8168Bb
  - rtl8169_{intr_mask/napi_event} are replaced with per-device fields,
    namely tp->{intr/napi}_event
  - rtl_cfg_info is converted to C99 for readability but the values are
    not changed for the 8169/8110 and the 8101

Includes ChipCmd fix from Bernhard Walle <bwalle@suse.de> (2007/02/24).
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

- new identifier for the 8110SCe

- the PCI latency timer is set unconditionally. This part is identical
  in Realtek's r8168 (8.001.00) and r8101 (1.001.00)

- initialization of the cache line size register is for the 8169s only

- more magic in rtl_hw_start_8169

- it is not possible to factor out the setting of the the irq event mask
  with the 8168 and the 8101 any more. Pushed it into the hw_start handler.

- rtl_set_rx_tx_config_registers() and write to the ChipCmd register are
  issued identically for the whole 8169/8110 family: the 8110SCd/8110SCe
  are handled the same way

- work around for AMD platform.

Some registers definitions in Realtek's driver are let aside for later.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

...still a bit yucky though.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Same thing as the previous change for rtl_hw_start_8168.

The 8101 related code in rtl_hw_start_8169 (see RTL_GIGA_MAC_VER_13)
goes away.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

rtl_hw_start_8168 inherits the content of rtl_hw_start_8169 minus
the code which depends on RTL_GIGA_MAC_VER_XY (XY != {11/12}).
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

They aim to limit the amount of moved code when the hw_start
handler gets more specialized.

No functional change.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Rationale: rtl8169_hw_start will not help maintaining an unified
driver for different chipsets but people at Realtek are probably
too polite to say it distinctly.

Let's add the hook and keep hw_start handler unchanged.

As can be seen from the content of rtl8169_pci_tbl, the RTL_CFG_1
entry in rtl_cfg_info was unused. I recycled it for the 0x8168.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

- pci_dma_sync_single_for_cpu is not needed for a single large packet
- remove the function pointer to help gcc optimizing the inline
  pci_dma functions.
Signed-off-by: NStephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

It hasn't "summed" anything in over 7 years, and it's
just a straight mempcy ala skb_copy_to_linear_data()
so just get rid of it.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

The former style suggests a modulo arithmetic misuse but
the expression should never be < 0. Even if it does, the
driver will simply loop longer than expected (not that
the remaining parts of the system will necessarily
appreciate it...).

Let's warn the user when something goes wrong and try
to go over it.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Use netdev_alloc_skb and remove the useless sk_buff * argument of
rtl8169_alloc_rx_skb.
Signed-off-by: NStephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>

Many drivers had code that did kill_vid, but they weren't doing vlan
filtering. With new API the stub is unneeded unless device sets
NETIF_F_HW_VLAN_FILTER.

Bad habit: I couldn't resist fixing a couple of nearby style things
in acenic, and forcedeth.
Signed-off-by: NStephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Signed-off-by: NArnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

One less thing for drivers writers to worry about.
Signed-off-by: NArnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The PM hooks are no-op if the r8169 interface is down (i.e. !IFF_UP).
However, as the chipset is enabled, the device will not work after a
suspend/resume cycle. The patch always issue the required PCI suspend
sequence and removes the module unload/reload workaround.
Signed-off-by: NArnaud Patard <apatard@mandriva.com>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

The irq handler schedules a NAPI poll request unconditionally as soon as
the status register is not clean. It has been there - and wrong - for
ages but a recent timing change made it apparently easier to trigger.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Cc: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Initialize the timer with the rest of the private-struct.
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Added during bf793295

The current code requests a reset but prohibits autoneg, 1000 Mb/s,
100 Mb/s and full duplex. The 8168 does not like it at all.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

This patch splits the vlan_group struct into a multi-allocated struct. On
x86_64, the size of the original struct is a little more than 32KB, causing
a 4-order allocation, which is prune to problems caused by buddy-system
external fragmentation conditions.

I couldn't just use vmalloc() because vfree() cannot be called in the
softirq context of the RCU callback.
Signed-off-by: NDan Aloni <da-x@monatomic.org>
Acked-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

flush_scheduled_work() in net_device->close has a slight tendency
to deadlock with tasks on the workqueue that hold RTNL.

rtl8169_close/down simply need the recovery tasks to not meddle
with the hardware while the device is going down.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Checked in Realtek's driver, this one has no business being there.
The driver still works but there is a noticeable performance drop.
Signed-off-by: NFrancois Romieu <romieu@fr.zoreil.com>