Fix a subtle race condition between tg3_start_xmit() and tg3_tx()
discovered by Herbert Xu <herbert@gondor.apana.org.au>:
CPU0 CPU1
tg3_start_xmit()
if (tx_ring_full) {
tx_lock
tg3_tx()
if (!netif_queue_stopped)
netif_stop_queue()
if (!tx_ring_full)
update_tx_ring
netif_wake_queue()
tx_unlock
}
Even though tx_ring is updated before the if statement in tg3_tx() in
program order, it can be re-ordered by the CPU as shown above. This
scenario can cause the tx queue to be stopped forever if tg3_tx() has
just freed up the entire tx_ring. The possibility of this happening
should be very rare though.
The following changes are made:
1. Add memory barrier to fix the above race condition.
2. Eliminate the private tx_lock altogether and rely solely on
netif_tx_lock. This eliminates one spinlock in tg3_start_xmit()
when the ring is full.
3. Because of 2, use netif_tx_lock in tg3_tx() before calling
netif_wake_queue().
4. Change TX_BUFFS_AVAIL to an inline function with a memory barrier.
Herbert and David suggested using the memory barrier instead of
volatile.
5. Check for the full wake queue condition before getting
netif_tx_lock in tg3_tx(). This reduces the number of unnecessary
spinlocks when the tx ring is full in a steady-state condition.
6. Update version to 3.65.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use GSO to workaround a rare TSO bug on some chips. This hardware
bug may be triggered when the TSO header size is greater than 80
bytes. When this condition is detected in a TSO packet, the driver
will use GSO to segment the packet to workaround the hardware bug.
Thanks to Juergen Kreileder <jk@blackdown.de> for reporting the
problem and collecting traces to help debug the problem.
And thanks to Herbert Xu <herbert@gondor.apana.org.au> for providing
the GSO mechanism that happens to be the perfect workaround for this
problem.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Clear a bit to enable a hardware fix for some ASF related problem.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add workaround to limit the burst size of rx BDs being DMA'ed to the
chip. This works around hardware errata on a number of 5750, 5752,
and 5755 chips.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Herbert Xu pointed out that it is unsafe to call netif_tx_disable()
from LLTX drivers because it uses dev->xmit_lock to synchronize
whereas LLTX drivers use private locks.
Convert tg3 to non-LLTX to fix this issue. tg3 is a lockless driver
where hard_start_xmit and tx completion handling can run concurrently
under normal conditions. A tx_lock is only needed to prevent
netif_stop_queue and netif_wake_queue race condtions when the queue
is full.
So whether we use LLTX or non-LLTX, it makes practically no
difference.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add recovery logic when we suspect that the system is re-ordering
MMIOs. Re-ordered MMIOs to the send mailbox can cause bogus tx
completions and hit BUG_ON() in the tx completion path.
tg3 already has logic to handle re-ordered MMIOs by flushing the MMIOs
that must be strictly ordered (such as the send mailbox). Determining
when to enable the flush is currently a manual process of adding known
chipsets to a list.
The new code replaces the BUG_ON() in the tx completion path with the
call to tg3_tx_recover(). It will set the TG3_FLAG_MBOX_WRITE_REORDER
flag and reset the chip later in the workqueue to recover and start
flushing MMIOs to the mailbox.
A message to report the problem will be printed. We will then decide
whether or not to add the host bridge to the list of chipsets that do
re-ordering.
We may add some additional code later to print the host bridge's ID so
that the user can report it more easily.
The assumption that re-ordering can only happen on x86 systems is also
removed.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Get rid of all the SUN_570X logic and instead:
1) Make sure MEMARB_ENABLE is set when we probe the SRAM
for config information. If that is off we will get
timeouts.
2) Always try to sync with the firmware, if there is no
firmware running do not treat it as an error and instead
just report it the first time we notice this condition.
3) If there is no valid SRAM signature, assume the device
is onboard by setting TG3_FLAG_EEPROM_WRITE_PROT.
Update driver version and release date.
With help from Michael Chan and Fabio Massimo Di Nitto.
Signed-off-by: David S. Miller <davem@davemloft.net>
Add some PHY workaround code to reduce jitter on some PHYs.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Speed up SRAM read and write functions if possible by using MMIO
instead of config. cycles. With this change, the post reset signature
done at the end of D3 power change must now be moved before the D3
power change.
IBM reported a problem on powerpc blades during ethtool self test that
was caused by the memory test taking excessively long. Config. cycles
are very slow on powerpc and the memory test can take more than 10
seconds to complete using config. cycles.
David Miller informed me that an earlier version of the patch caused
problems on sparc64 systems with built-in tg3 chips. This version
fixes the problem by excluding all SUN built-in tg3 chips from doing
MMIO SRAM access.
TG3_FLAG_EEPROM_WRITE_PROT is also set unconditionally when
TG3_FLG2_SUN_570X is set. This should be sane as all SUN chips are
built-in and do not require Vaux switching.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kill the TG3_FLAG_NO_{TX|RX}_PSEUDO_CSUM flags because they are not
very useful. This will free up some bits for new flags.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for new chip 5755 which is very similar to 5787.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support one-shot MSI on 5787.
This one-shot MSI idea is credited to David Miller. In this mode, MSI
disables itself automatically after it is generated, saving the driver
a register access to disable it for NAPI.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support 5787 hardware TSO using a new flag TG3_FLG2_HW_TSO_2.
Since the TSO interface is slightly different and these chips have
finally fixed the 4GB DMA problem and do not have the 40-bit DMA
problem, a new hard_start_xmit is used for these chips. All previous
chips will use the old hard_start_xmit that is now renamed
tg3_start_xmit_dma_bug().
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support additional nvrams and new nvram format for 5787 and 5754.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 40-bit DMA workaround recently implemented for 5714, 5715, and
5780 needs to be expanded because there may be other tg3 devices
behind the EPB Express to PCIX bridge in the 5780 class device.
For example, some 4-port card or mother board designs have 5704 behind
the 5714.
All devices behind the EPB require the 40-bit DMA workaround.
Thanks to Chris Elmquist again for reporting the problem and testing
the patch.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make sure tg3_reset_task() is flushed in the close and suspend paths
as noted by Jeff Garzik.
In the close path, calling flush_scheduled_work() may cause deadlock
if linkwatch_event() is on the workqueue. linkwatch_event() will try
to get the rtnl_lock() which is already held by tg3_close(). So
instead, we set a flag in tg3_reset_task() and tg3_close() polls
the flag until it is cleared.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add nvram lock count so that calls to tg3_nvram_lock()/unlock() can
be nested. Add error checking to all callers of tg3_nvram_lock()
where appropriate. To prevent nvram lock failures after halting the
firmware, it is also necessary to release firmware's nvram lock in
tg3_halt_cpu().
Update version to 3.48.
Based on David Miller's initial patch.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Resubmitting after recommendation to use GET_REG32_1() instead of
GET_REG32_LOOP(..., 1). Retested. Problem remains fixed.
Prevent tg3_get_regs() from reading reserved and undocumented registers
at RX_CPU_BASE and TX_CPU_BASE offsets which caused hostile behavior
on PCIe platforms.
Acked-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the following bugs in tg3_set_power_state():
1. Both WOL and ASF flags require switching to aux power.
2. Add a missing handshake with firmware to enable WOL.
3. Turn off the PHY if both WOL and ASF are disabled.
4. Add nvram arbitration before halting the firmware.
5. Fix tg3_setup_copper_phy() to switch to 100Mbps when
changing to low power state.
Update revision and date.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change the ASF heart beat to 5 seconds for faster detection of system
crash. The driver sends the heartbeat every 2 seconds and the ASF
firmware will timeout and reset the device if no heartbeat is received
after 5 seconds. The old scheme of 2 minutes is ineffective.
tg3_write_mem_fast() is added to speed up the IO to send the heartbeat.
When no workaround is needed, it will use direct MMIO to memory space
to write to memory.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Add complete support for 5714/5715. These chips are very similar to
5780 so the changes are very trivial. A TG3_FLG2_5780_CLASS flag is
added to identify these chips.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Fix 5780 PHY related problems:
1. MAC_RX_MODE reset must be done before setting up the MAC_MODE
register on 5705_PLUS chips or the chip will stop receiving after
a while. The MAC_RX_MODE reset is needed to prevent intermittently
losing the first receive packet on serdes chips.
2. Skip MAC loopback test on 5780 because of hardware errata. Normal
traffic including PHY loopback is not affected by the errata.
3. PHY loopback fails intermittently on 5708S and this is fixed by
putting the PHY in loopback mode first before programming the MAC
mode register. A MAC_RX_MODE reset is also added.
4. Return -EINVAL in tg3_nway_reset() if device is in TBI mode. Allow
nway_reset if 5780S is in parallel detect mode.
5. Add missing PHY IDs in KNOWN_PHY_ID() macro.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the new workaround for 5703 A1/A2 if it is behind
certain ICH bridges. The workaround disables memory and uses config.
cycles only to access all registers. The 5702/03 chips can mistakenly
decode the special cycles from the ICH chipsets as memory write cycles,
causing corruption of register and memory space. Only certain ICH
bridges will drive special cycles with non-zero data during the address
phase which can fall within the 5703's address range. This is not an ICH
bug as the PCI spec allows non-zero address during special cycles.
However, only these ICH bridges are known to drive non-zero addresses
during special cycles.
The indirect_lock is also changed to spin_lock_irqsave from spin_lock_bh
because it is used in irq handler when using the indirect method to
disable interrupts.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the mailbox read method and also adds an inline function
tw32_mailbox_f() for mailbox writes that require read flush.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the basic function pointers to do register accesses in
the fast path. This was suggested by David Miller. The idea is that
various register access methods for different hardware errata can easily
be implemented with these function pointers and performance will not be
degraded on chips that use normal register access methods.
The various register read write macros (e.g. tw32, tr32, tw32_mailbox)
are redefined to call the function pointers.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add 5780S support by adding a new tg3_setup_fiber_mii_phy() function and
a timer function for parallel link detection. 5780S uses standard MII
registers for 1000BaseX and runs in GMII mode as opposed to TBI mode on
older serdes chips.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Consolidate all DMA watermark settings for standard and jumbo frames on
all chips in tg3_init_bufmgr_config() and add new settings for 5780.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new rx_pkt_buf_sz to the tg3 structure to support variable buffer
sizes on the standard ring.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add basic jumbo frames support for 5780. This chip supports jumbo frames
on the standard receive ring without the jumbo ring. The
TG3_FLAG_JUMBO_ENABLE is changed to TG3_FLAG_JUMBO_RING_ENABLE to
indicate using the jumbo ring on 5704 and older chips. A new
TG3_FLG2_JUMBO_CAPABLE flag is added to indicate jumbo frames support
with or without the jumbo ring.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add 5780 PCI IDs, chip IDs, and other basic support.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for ethtool -C with verification of user parameters.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move all driver spinlocks to be taken at sw IRQ
context only.
This fixes the skb_copy() we were doing with hw
IRQs disabled (which is illegal and triggers a
BUG() with HIGHMEM enabled). It also simplifies
the locking all over the driver tremendously.
We accomplish this feat by creating a special
sequence to synchronize with the hw IRQ handler
using a binary state and synchronize_irq().
This idea is from Herbert Xu.
Thanks to Michael Chan for helping to track down
all of the race conditions in initial versions
of this code.
Signed-off-by: David S. Miller <davem@davemloft.net>
Even though we do software interrupt mitigation
via NAPI, it still helps to have some minimal
hw assisted mitigation.
This helps, particularly, on systems where register
I/O overhead is much greater than the CPU horsepower.
For example, it helps on NUMA systems. In such cases
the PIO overhead to disable interrupts for NAPI accounts
for the majority of the packet processing cost. The
CPU is fast enough such that only a single packet is
processed by each NAPI poll call.
Thanks to Michael Chan for reviewing this patch.
Signed-off-by: David S. Miller <davem@davemloft.net>
When supported, use the TAGGED interrupt processing support
the chip provides. In this mode, instead of a "on/off" binary
semaphore, an incrementing tag scheme is used to ACK interrupts.
All MSI supporting chips support TAGGED mode, so the tg3_msi()
interrupt handler uses it unconditionally. This invariant is
verified when MSI support is tested.
Since we can invoke tg3_poll() multiple times per interrupt under
high packet load, we fetch a new copy of the tag value in the
status block right before we actually do the work.
Also, because the tagged status tells the chip exactly which
work we have processed, we can make two optimizations:
1) tg3_restart_ints() need not check tg3_has_work()
2) the tg3_timer() need not poke the chip 10 times per
second to keep from losing interrupt events
Based upon valuable feedback from Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for the NVRAM lock-out feature for TPM in 5752. If lock-out
is enabled, certain NVRAM registers cannot be written to.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add bit definitions for the new GPIO3 in 5752. GPIO3 must be driven as
output when it is unused.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 5752 A0 chip ID is wrong in hardware. The simplest way to workaround
it is to change it to the correct value in tp->pci_chip_rev_id. This
way, it is easier to check for the ASIC_REV_5752 in the rest of the
driver.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some minor 5752 fixes mostly for correctness and add 5752 PHY ID.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace existing ASIC_REV_5752 definition with ASIC_REV_5752_A0,
and add definition for ASIC_REV_5752_A1. Then, add ASIC_REV_5752_A1
to check for setting TG3_FLG2_5750_PLUS in tg3_get_invariants.
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define TG3_FLG2_5750_PLUS flag and set it in tg3_get_invariants for
ASIC_REV_5750 or ASIC_REV_5752.
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ASIC_REV_5752 definition.
Track-down all references to ASIC_REV_5750 and mirror them with
references to the newly defined ASIC_REV_5752.
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Michael Chan