Change details:
- Found a leak in sk_buff unmapping of PCI dma addresses where boundary
conditions are not properly handled in freeing all Tx buffers. Freeing
of all Tx buffers is done considering sk_buffs data and fragments can
be mapped at the boundary.
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_buff unmap_array grows greater than 65536 (x2) with Tx ring of 65536.
Reducing TXQ depth and safe(max) acking of Tx events to 32768 (same as Rx).
Add defines for TX and RX queue depths.
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Check the BNAD_RF_MBOX_IRQ_DISABLED flag after acquiring the bna_lock,
since checking the flag and executing bna_mbox_handler needs to be atomic.
If not, it opens up window where flag is reset when it was checked, but got
set while spinning on the lock by the other thread which is actually
holding the lock
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Async mode of Tx/Rx queue initialization in BNAD from a task queue context
runs into non-unique taskq allocation issues. Get rid of Tx/Rx
initialization from task q context
- In the attach function, wait for IOC enable, then do Tx/Rx queue
initialization. Default BNA attributes are used when IOC enable from attach
fails and values are set to:
1 TxQ, 1 RxQ, 1 Unicast MAC, 1 RIT entry
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Add tx_skb counters and NAPI debug counters to ethtool stats.
- Add rlb stats strings to bnad_net_stats_strings{} array. rlb_stats field
was added to struct bfi_enet_stats {} but the corresponding name structure
array for ethtool was not initialized with right strings, even though the
actual name structure array got expanded. This caused a NULL pointer
violation and a crash when doing ehtool -S <if_name>.
- Modify dim timer stop logic to make it dependent on cfg and run flags
- While setting the ring parameter restore the rx, vlan configuration and
set rx mode
- Indentation fix
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Initialize rx_id to 0 for bnad_cleanup_rx
- Return -ENOMEM in case if bna_rx_create fails
- Count the Rx buffer allocation failures in bnad_alloc_n_post_rxbufs()
- Remove unnecessary initialization of using_dac to false in bnad_pci_probe
- Release lock if error while doing bna_num_txq_set in bnad_pci_probe
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Print log messages when running with reduced number of MSI-X vectors
and when defaulting to INTx mode.
- Remove BUG_ONs and header file inclusion that are not needed
- Comments addition/cleanup
- Unused code cleanup
- Add New Line to Print msg in bfa_sm_fault
- Formatting fix
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Add bnad_pci_unmap_skb() API to unmap skb from transmit path and update the
unmap index. Add more checks for illegal skbs in transmit path. Add tx_skb
counters for dropped skbs.
- The unmap_cons index used in bnad_free_txbufs() is incorrectly declared as
u16. It quickly wraps around and accesses null sk_buff ptr. So using u32 to
handle unmap_array.
- Disable and enable interrupts from the same polling context to prevent
reordering in Rx path.
- Add Rx NAPI debug counters.
- Make NAPI budget check more generic.
- Modify dim timer stop logic to make it dependent on cfg and run flags
- Handle reduced MSI-X vectors case in bnad_enable_msix.
- Check for single frame TSO skbs and send them out as non-TSO.
- Put memory barrier after bna_txq_prod_indx_doorbell().
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Remove unnecessary ccb check from bnad_poll_cq
- Add bnad pointer to rx_ctrl structure, so that bnad can be accessed directly
from rx_ctrl in the NAPI poll routines, even if ccb is NULL
- Validate ccb before referencing to it in bnad_msix_rx and bnad_napi_poll_rx
- Fix the order of NAPI init / uninit in Tx / Rx setup / teardown path:
a. Kill bnad tx free tasklet ahead of call to bna_tx_destroy()
b. Call NAPI disable only after call to Rx free_irq(). This makes sure Rx
interrupt does not schedule a poll when NAPI is already disabled
- NAPI poll runs before the h/w has completed configuration. This causes a
crash. Delay enabling NAPI till after bna_rx_enable(). Split NAPI
initialization into 2 steps, bnad_napi_init() & bnad_napi_enable().
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Return error as -EIO if bnad_res_alloc fails
- Release the configuration lock before registering with net_device layer.
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Naming changes: rename devid, BNAD_MAX_TXS, BNAD_MAX_RXS,
BNAD_MAX_RXPS_PER_RX to device, BNAD_MAX_TX, BNAD_MAX_RX,
BNAD_MAX_RXP_PER_RX respectively and change all the references.
Macro Fix: Add ioc_isr_mod_set check to bfa_nw_ioc_mbox_regisr macro
Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Rasesh Mody <rmody@brocade.com>
Cc: Debashis Dutt <ddutt@brocade.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on finds for Stephen Rothwell, where current defconfig's
enable a ethernet driver and it is not compiled due to the newly
added NET_VENDOR_* component of Kconfig.
This patch enables all the "new" Kconfig options so that current
defconfig's will continue to compile the expected drivers. In
addition, by enabling all the new Kconfig options does not add
any un-expected options.
CC: Stephen Rothwll <sfc@canb.auug.org.au>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The rspq->rsphdlr[] array has BFI_MC_MAX elements, so this test was
off by one.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We introduced a new lock here, so there was error path which needs
an unlock now.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Removec bfi_ll.h bna_hw.h bna_ctrl.c and bna_txrx.c due to ENET, MSGQ
and TXRX changes for new FW Driver interface and TX RX re-design.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
This patch contains additional structure and function definition changes
that are required to enable the new msgq/enet/txrx redesign introduced
by the previous 4 patches.
- structure and function definition changes to header files as a result
of Ethport, Enet, IOCEth, Tx, Rx redesign.
- ethtool changes to use new enet function and definitions
- Set number of Tx and Rx queues bassed on underlying hardware. Define
separate macros for maximum and supported numbers of Tx and Rx queues
based on underlying hardware. Take VLAN header into account for MTU
calculation. Default to INTx mode when pci_enable_msix() fails. Set a
bit in Rx poll routine, check and wait for that bit to be cleared in
the cleanup routine before proceeding.
- The TX and Rx coalesce settings are programmed in steps of 5 us. The value
that are not divisible by 5 are rounded to the next lower number. This was
causing the value os 1 to 4 to be rounded to 0, which is an invalid setting.
When creating Rx and Tx object, we are currently assigning the default
values of Rx and Tx coalescing_timeo. If these values are changed in the
driver to a different value, the change is lost during such operations as
MTU change. In order to avoid that, pass the configured value of
coalescing_timeo before Rx and Tx object creation. Fix
bnad_tx_coalescing_timeo_set() so it applies to all the Tx objects.
- Reorg uninitialization path in case of pci_probe failure.
- Hardware clock setup changes to pass asic generation, port modes and
asic mode as part firmware boot parameters to firmware.
- FW mailbox interface changes to defined asic specific mailbox interfaces.
h/w mailbox interfaces take 8-bit FIDs and 2-bit port id for owner. Cleaned
up mailbox definitions and usage for new and old HW. Eliminated usage of
ASIC ID. MSI-X vector assignment and programming done by firmware. Fixed
host offsets for CPE/RME queue registers.
- Implement polling mechanism for FW ready to have poll mechanism replaces
the current interrupt based FW READY method. The timer based poll routine
in IOC will query the ioc_fwstate register to see if there is a state
change in FW, and sends the READY event. Removed infrastructure needed to
support mbox READY event from fw as well as IOC code.
- Move FW init to HW init. Handle the case where PCI mapping goes away when
IOCPF state machine is waiting for semaphore.
- Add IOC mbox call back to client indicating that the command is sent.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Add new file bna_hw_defs.h to support new code MSGQ, ENET and TX RX redign.
This makes bna_hw.h obsolete and is removed in a later patch. bna_hw_defs.h
removes all unused HW register definition that were part of bna_hw.h.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- This patch contains the changes as a result of redesigning of Tx, Rx data
path setup. In the old design, setting up Txqs, Rxqs were done in the driver.
With the new design, most of the hardware setup steps for the Txq, Rxqs are
moved to FW. Host driver issues commands to FW through the message queue to
setup/teardown tx, rx data path. FW performs necessary steps and responds
back to the driver with a status.
- As a result of this redesign, the state machine implementation for Tx, Rx
objects have changed significantly. Instead of doing the raw register access,
these state machines mostly send a command to FW and wait for response and
take the next action. In addition to tx, rx datapath setup, this patch also
deals with rx filter configuration - such as unicast address, multicast
address, vlan filter, promiscuous mode etc.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- This patch contains the messages, opcodes and structure format for the
messages and responses exchanged between driver and the FW. In addition
this patch contains the state machine implementation for Ethport, Enet,
IOCEth.
- Ethport object is responsible for receiving link state events, sending
port enable/disable commands to FW.
- Enet object is responsible for synchronizing initialization/teardown of
tx & rx datapath configuration.
- IOCEth object is responsible for init/un-init of IO Controller in the
adapter which runs the FW.
- This patch also contains code for initialization and resource assignment
for Ethport, Enet, IOCEth, Tx, Rx objects.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change details:
- Currently modules communicate with the FW using 32 byte command and
response register. This limits the size of the command and response
messages exchanged with the FW to 32 bytes. We need a mechanism to
exchange the comamnds and responses exchange with FW that exceeds 32 bytes.
- MSGQ implementation provides that facility. It removes the assumption that
command/response queue size is precisely calculated to accommodate all
concurrent FW commands/responses. The queue depth is made variable now, defined
by a macro. A waiting command list is implemented to hold all the commands
when there is no place in the command queue. Callback is implemented for
each command entry to invoke the module posting the command, when there is
space in the command queue and the command was finally posted to the queue.
Module/Object information is embedded in the response for tracking purpose.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Moves the Brocade driver into drivers/net/ethernet/brocade/ and make
the necessary Kconfig and Makefile changes.
CC: Rasesh Mody <rmody@brocade.com>
CC: Debashis Dutt <ddutt@brocade.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Rasesh Mody