| Age | Commit message (Collapse) | Author |
|---|
|
Added flow control between TMU queues and PFE Linux driver,
based on TMU credits availability.
Added tx_qos module parameter to control this behavior.
Use queue-0 as default queue to transmit packets.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Akhila Kavi <akhila.kavi@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
|
|
__hif_lib_update_credit function is used to update the tmu credits.
If tx_qos is set, tmu credit is updated based on the number of packets
transmitted by tmu.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
|
|
PCS initialization sequence for 2.5G SGMII interface governs
auto negotiation to be in disabled mode
Signed-off-by: Bhaskar Upadhaya <Bhaskar.Upadhaya@nxp.com>
|
|
when we opearate in clause 45 mode, we need to call
the function get_phy_device() with its 3rd argument as
"true" and then the resultant phy device needs to be
register with phy layer via phy_device_register()
Signed-off-by: Bhaskar Upadhaya <Bhaskar.Upadhaya@nxp.com>
|
|
Stop polling and unregister / disconnect the phydev structure on DPAA2
mac driver unbind.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
- move dpaa2_mac_open and dpaa2_mac_stop out of
CONFIG_FSL_DPAA2_MAC_NETDEVS, since their implementation is necessary
regardless of it
- reorder ndo ops to match function implementation order
- update comment to describe the phy connection mode that's to be used -
it no longer depends on DPC, but on the device tree
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
free_netdev (put_device) already handles freeing the private data
structure, and KASAN will complain due to a free after free if we
explicitly do the same afterwards.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Add a Kconfig option for enabling Data Center Bridging (DCB)
support in the Ethernet driver. This is needed for priority
flow control support.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
|
|
According to the hardware ArchDef, the PTV1 field in FD[CTRL]
is ignored by WRIOP, so setting it for Tx FDs is pointless.
Remove all references to it from the code.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
While in NAPI context, free skbs by calling napi_consume_skb()
instead of dev_kfree_skb(), to take advantage of the bulk freeing
mechanism.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
When releasing buffers to the hardware pool, don't call
cpu_relax() unless we are actually waiting for the QBMan
portal to finish the previous command.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
The network stack will take care to reserve space for HH_DATA_MOD when
building the skb, so there's no need to account for it in the netdevice
needed headroom. Based on that, decrease the frame buffer headroom for
ingress frames.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
Instead of the old opaque value, define the Rx buffer headroom size
based on other frame buffer fields, in accordance to how we expect them
to be configured by MC.
Based on DPNI_BUF_LAYOUT_OPT_TIMESTAMP, MC will enable an additional
field in the Tx hardware annotation area, increasing its size to 128.
Update the driver value to reflect this.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
When configuring the Tx buffer layout, the software annotation size is
mentioned, and MC accounts for it when configuring the frame
tx_data_offset. No need to handle it in the driver as well.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
Since adding support for multiple buffer thresholds (taildrop, flow
control pause frames, PFC pause frames), it's useful to have the
per-channel buffer count available in the debug stats.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Make sure the NAPI instances are enabled only after all other
device initialization is done, to avoid accessing uninit vars.
Mirror that change on device remove routine.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
On Rx path, if we fail to build an skb from the incoming FD,
we still need to update the channel buffer count accordingly,
otherwise we risk depleting the pool while the software counter
still sees available buffers.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
We incorrectly assumed that dpaa2_io_release() can only
return -EBUSY as an error code, when in fact it can also
fail in case some of its arguments don't have valid values.
Make sure we only retry the operation while the portal is
busy and abort for all other error cases, otherwise we risk
entering an endless loop.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Integrated-by: Guanhua Gao <guanhua.gao@nxp.com>
|
|
Fix compiling errors for backporting dash-lts.
Signed-off-by: Guanhua Gao <guanhua.gao>
|
|
The dpaa2-eth PFC configuration is kept in a private driver variable.
When getting PFC configuration via ieee_getpfc, get it straight from the
Management Complex based on the Rx congestion notification setup.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Implement the following:
- Enable PFC in DPNI options.
- Set queue taildrops for priorities (traffic classes) that are not PFC
enabled (pause frames are not sent in case of congestion). Since the
buffer pool is shared, we can't risk other traffic to consume all
available buffers.
- Configure Rx congestion notifications for PFC enabled priorities.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
With the introduction of PFC support, there are multiple ways taildrop
is configured, based on the DPNI pause frame configuration:
- no pause frames - per-queue taildrop thresholds.
- pause frames - no taildrop thresholds (pause frames sent based on
buffer depletion).
- PFC pause frames - per-group (traffic class) taildrop thresholds for
non-PFC enabled priorities (pause frames sent based on Rx congestion
notifications).
Refactor the fq taildrop setup area.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Configure static ingress classification based on VLAN PCP, required by
PFC.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Add initial DCB ops in the DPAA2 Ethernet driver. Currently they are
dummy calls.
Tested integration with userspace lldpad:
- lldpad daemon running
- bring interface up
- lldptool -L -i niX adminStatus=rxtx
- lldptool -T -i niX -V PFC -c enabled={prio comma separated list}
- lldptool -t -i niX -V PFC -c enabled
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
The v2 dpni_set_pools command format allows sending an additional
parameter, the priority mask for the buffer pool. This can be used to
configure multiple buffer pools for the same dpni, to serve different
priorities.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
The current MC (10.3.1) design uses a separate flow steering table for
each traffic class. When adding a flow steering rule, add it for each
traffic class. The location of the rule is the same in all tables.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
The driver only handles frame workqueues associated with traffic class
0. Add support for multiple traffic classes for the same flowid, based
on the tc field in the fq struct, and the num_tcs parameter in the dpni
atttributes. Create a separate fq for each tc x flowid. Display the
information in debugfs.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Make return value void since function never return meaningful value
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
|
|
Invalidate the cache for the software portals before using them
since the portals are non coherent. This ensures that the core
using the portal is seeing the most up to date information in
case the cache contained older data. This is important during
the cleanup phase if cleanup occurs on a differnt core than
what the application was using.
Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
|
|
Use the management commmand response registers to determine
the next expected valid bit when initializing a software
portal. This avoids using the wrong valid bit in cases
where a command was partially written but then not
completed.
Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
|
|
While fixing checkpatch errors some of the index increments
were commented out. They are enabled.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
|
|
Remove unused functions hif_xmit_pkt & hif_lib_xmit_pkt.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
|
|
Recently logic to enable RGMII tx delay was changed by
below patch.
https://patchwork.kernel.org/patch/9447581/
Based on the patch, appropriate change is made in PFE driver.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
|
|
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
|
|
[pfe part]
This patch introduces Linux support for NXP's LS1012A Packet
Forwarding Engine (pfe_eth). LS1012A uses hardware packet forwarding
engine to provide high performance Ethernet interfaces. The device
includes two Ethernet ports.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
Integrated-by: Zhao Qiang <qiang.zhao@nxp.com>
|
|
This patch has all pfe header files.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
|
|
[config part]
This patch introduces Linux support for NXP's LS1012A Packet
Forwarding Engine (pfe_eth). LS1012A uses hardware packet forwarding
engine to provide high performance Ethernet interfaces. The device
includes two Ethernet ports.
Signed-off-by: Calvin Johnson <calvin.johnson@nxp.com>
Signed-off-by: Anjaneyulu Jagarlmudi <anji.jagarlmudi@nxp.com>
Integrated-by: Zhao Qiang <qiang.zhao@nxp.com>
|
|
Use the cpu affine DPIO unless there isn't one which can happen
if less DPIOs than cores are assign to the kernel.
Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
|
|
For some dpio functions, a negative cpu id parameter value is
valid and means "any". But when trying to validate this param
value against an upper limit, in this case num_possible_cpus(),
we risk obtaining the wrong result due to an implicit cast.
Avoid an incorrect check result when cpu id is negative, by
explicitly stating the comparison is between signed values.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
Previous commits:
6e2387e8f19e ("staging: fsl-dpaa2/eth: Add Freescale DPAA2 Ethernet driver")
39163c0ce0f4 ("staging: fsl-dpaa2/eth: Errors checking update")
have added bits that are not specific to the WRIOP accelerator.
Move these where they belong (in DPIO) such that other accelerators
can make use of them.
While here, fix the values of FD_CTRL_FSE and FD_CTRL_FAERR, which
were shifted off by one bit.
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
|
|
Previous commits:
6e2387e8f19e ("staging: fsl-dpaa2/eth: Add Freescale DPAA2 Ethernet driver")
39163c0ce0f4 ("staging: fsl-dpaa2/eth: Errors checking update")
have added bits that are not specific to the WRIOP accelerator.
Move these where they belong (in DPIO) such that other accelerators
can make use of them.
While here, fix the values of FD_CTRL_FSE and FD_CTRL_FAERR, which
were shifted off by one bit.
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
|
|
We used to set pass_timestamp parameter only for Tx
confirmation buffer layout config, when in fact MC
also looks at the Tx buffer layout when deciding whether
to enable Tx timestamping support or not.
Older versions of MC (< MC.1.x) enabled this support regardless
of user config, so the issue was hidden.
Copy the pass_timestamp setting to Tx buffer layout configuration.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
Make sure the memory area where the hardware writes the
congestion state is properly synchronized before reading it.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
If build_frag_skb() fails, we risk accessing the skb
pointer before checking it's not NULL.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
When inserting a flow steering rule through ethtool, the user
may specify a location index. Ethtool manual says the expected
behaviour is to overwrite any rule present in that location.
What currently happens is that the driver passes on the location
to MC with no further verifications and if there's already
a rule present at that index, MC returns an error.
Fix this by making sure the location is available, and if not
free the current entry before inserting the new one.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
Once a Tx frame descriptor is enqueued, an interrupt might be triggered
to process the Tx confirmation and free the skb, hitting a memory use
after free when updating the tx_bytes statistic based on skb->len.
Use the frame descriptor length instead.
Signed-off-by: Bogdan Purcareata <bogdan.purcareata@nxp.com>
|
|
Private structure fields "hash_fields" and "num_hash_fields" were
used in check_cls_support() even though they were initialized
later.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
|
|
'type' and 'flags' fields were missing from dprc_rsp_get_obj_region
structure therefore the MC Bus driver was not receiving proper flags
from MC like DPRC_REGION_CACHEABLE.
Signed-off-by: Cristian Sovaiala <cristian.sovaiala@freescale.com>
|
