This feature can be enabled runtime using sysfs interface.
Support is added to dump and clear the histogram stats.
The lower delay regions has to be more granular to indicate any
medium related issues for time sensitive XR applications.
Change-Id: I0a44a54d12d92ce016de349810cb2bedebaf9a58
CRs-Fixed: 2981006
get_log_timestamp API is being called unconditionally when passed as
an argument to dp_pkt_add_timestamp (even for empty calls) from
dp_tx_hw_enqueue_li. This adds significant load on CPU, hence avoid
get_log_timestamp call by using preprocessor directive.
Change-Id: I1d988597e28b755a767a267d1b578b2bcedc324e
CRs-Fixed: 3150795
Add support to change the tx_desc value to 65536. Some changes
to make the function argument as u32 type us made
Change-Id: I7cbde1b7ed4ab4e278c25c1ecfa94b7f673197f2
CRs-Fixed: 3130833
Currently HW TX queue HP value is updated for every TSO segment enqueue,
as a part of optimization update HP only for the last segment.
Change-Id: Ibe349a1e6f55932bf780e1f755f13841078a493f
CRs-Fixed: 3138142
Add feature support to tag first packet that wakes up HOST from WoW.
rx_pkt_tlv.rx_msdu_end.reserved_1a field is used by TARGET to meet
such request.
Change-Id: I3d37e13e8cff49bc4f622d3070a19e4c4be56417
CRs-Fixed: 3137621
Ageout flush does not happen for WBM2SW4 if there
is only one TX completion pending in FIFO and all the
other WBM release rings are not active. This is due to
an issue in HW and this prevents suspend to happen due
to pending tx completions.
Fix is to avoid using WBM2SW4 release ring and instead
reuse WBM2SW0.
Change-Id: I250d8c9d460895449939212ebdb7abd62edb0234
CRs-Fixed: 3124733
Initialize txrx_ref_handle for whunt bypass,
Which is complaining use of uninitialized txrx_ref_handle,
as it is not able to detect that txrx_ref_handle is
populated from another API.
Change-Id: I0f69cf4c8442a5655559622a5825d7af35b9ae5e
CRs-Fixed: 3127788
Move the stats parameters from the dp_peer which are used
in per packet path to txrx_peer
Change-Id: Ieb68b6950740791b37bbd2bfdc4815c3d7bc4834
CRs-Fixed: 3095637
Add support to update tsf timestamp on driver entry and
exit in data packet. This helps debug latency issue in
XR usecases
Change-Id: I9c966c1b8cb09dc5eab6104fdad36c19a1d68045
CRs-Fixed: 3090108
simple Alloc is being used in RX path which avoids
certain debug logic. during free of nbuf we should
avoid this debug logic else it will report it as
double free, this will be triggered only when debug
is enabled
Change-Id: Iadb40071fb733cc4de3291784df5075d5a099a8e
In WIN BE chipsets, replace the REO tid
queue programming in FW via WMI with writing to a
Host managed table shared by HW and SW. REO HW will
pick the tid queue address from the table indexed by
peer id and tid number.
Change-Id: I8107ca5116425538329b11ae3519f02b32573bac
Enable the 4th Tx. completion ring to save CPU load
Initialization and interrupt handling for 4th completion ring
is done here.
Change-Id: I2db27218a3c3e14d719d012f03454a6a7aa647fe
1. avoid accessing CFG parameter in per pkt path.
2. Access peer, vdev and pdev only when current peer_id
is different from previous peer_id.
3. Avoid HKv1 and v2 checks in WDS learning path.
Change-Id: Id7fb7a29c642480f8ca58738dbb0c92130e44ecd
Tx descriptor is not being freed and put back in tx desc
pool for a long time. As a result, system is not going
into suspended state.
To find this descriptor, during dp_bus_suspend and
dp_runtime_suspend, parse through the descriptor pool to
check for any descriptors not freed for a long and trigger
self recovery when found.
Change-Id: Id97c5c8537c9bec922f4e254b5bf094505ee61ff
CRs-Fixed: 3109868
use dma_inv_range call instead of dma map and unmap
APIs during replenish and reap. This complete code is
under a compile time macro.
before change: 4160Mbps @ 96.5% (core 3)
after change: 4160Mbps @ 90.5% (core 3)
Change-Id: I61bca349a369ace06cd86f353880108cee013cb1
The following issues are fixed with this change-
1. Sojourn stats were printing incorrect values
2. RSSI values were junk numbers
Change-Id: Ibed73cf370a309bf9db5baad37a0e8261c58463b
- Fixes for compilation issues after enabling
monitor 2.0 support.
- change copyright year for all files in the chain.
Change-Id: I885e257bd8ca83850656d8a1f408c1bc34920d7a
CRs-Fixed: 3086483
Prefetch TX HW desc, SW desc and SKB in pipeline fashion in Tx.
completion path.
This improves the UDP DL CPU idle% by ~4.5%
Change-Id: I48096e996cd835321ce2681d3981fa94c7189f54
Use the tracepoints to trace delayed register write, ce
tasklet scheduling latency, tx, and rx packets.
Change-Id: I63a89276177a9d0466dcb0c831eeb8e938a2bf79
CRs-Fixed: 3081870
Use chip ID and destination peer to determine the target soc
and partner vdev for Intra-BSS in MLO case.
Change-Id: I709c52e74426c5e81b50c8063cad7669c0e7002d
in some corner cases (ex: SG frames are received) we
do not decrement quota or num_pending and go and fetch
the next descriptor, chances are ring may not have valid
descriptors at that point of time hence added this sanity
check.
Change-Id: I6994cc69c35de6bed7a71adca0a1d66ed5c94d3c
Following peer stats are updated in per packet Rx path:
to_stack
multicast
broadcast
In BE architecture, HW provides the support for basic vdev stats and
hence per packet stats updation of above parameters can be done only
when enhanced stats is enabled or HW vdev offload stats is disabled.
Avoiding per packet stats updation reduces CPU load and improves KPI.
Change-Id: Id7c11c025a8464951b615a7f7b006ce61db487fc
CRs-Fixed: 3067843
In the case of RAW mode, VAP parameters encap type, dscp_to_tid map id
and cipher are not updating in bank register.
Added a API to update vdev param.
Change-Id: I702bee563e7451f403fa32292bf20680cd66e213
CRs-Fixed: 3078687
Rx patch changes for multichip MLO
1. Create ini for rx ring mask for each chip
2. Configure hash based routing for each chip based
on lmac_peer_id_msb
3. Peer setup changes to configure lmac_peer_id_msb
to enable hash based routing
4. Rx Replenish changes to provide buffers back to owner
SOC of reo ring
Change-Id: Ibbe6e81f9e62d88d9bb289a082dd14b4362252c4
Changes needed for MLO soc attach to pass chip_id,
dp_ml_context from upper layer.
This change also takes care of assigning appropriate
RBM id for IDLE link descriptors based on chip_id.
Change-Id: I8f5f08c524d91942e6e458f048700b7bdd900107
Prefetch RX HW desc, SW desc and SKB in pipeline
fasion in the first loop of RX processing.
This has improved TPUT by 200Mbps and provided a
10% gain in CPU (single core)
PINE with other optimizations: 3960Mbps @ 100% core-3
PINE + pipeline prefetch: 4130Mbps @ 90% core-3
Change-Id: I47f351601b264eb3a2b50e4154229d55da738724
Added an API to do a batch invalidation of REO descs
saw an improvement of 40 to 45 Mbps.
Note: this change is applicable only for cached
descriptors
PINE with Default driver: 3189 @ 100% core-3
PINE with skb prefetch: 3469 @ 100% core-3
PINE with skb pre + batch inv: 3506 @ 100% core-3
Change-Id: Ic2cf294972acfe5765448a18bed7e903562836c3
For IPQ products, there is 1 refill ring which is of hardware type
and host replenishes the buffers onto this ring so that hardware can
use these buffers for Rx.
In IPA offload mode, the buffer replenishment model is different from
the one mentioned above. There are 3 refill rings, out of which,
2 are software refill rings (1 for host and 1 for IPA), and last ring
is hardware ring given to FW.
Ring given to IPA is to refill the buffers after processing the
regular Rx packets and ring given to host is to refill the buffers
after processing of exception packets. Since there are 2 entities to
refill the buffers, the hardware ring given to FW multiplexes these 2
software rings and provides the buffers to hardware.
Make changes to follow above replenishment model for SDX+Pine
integration.
Change-Id: I0d9e4ec811a3023a258e0a6b9ee22ccdffcebafa
CRs-Fixed: 3049633
Tid in RX frame header may be larger than MAX TID allowed
value, this will lead a out of boundary array access and
lead to kernel crash at last. Change is aimed to do a TID
check and discard such frame when necessary.
Change-Id: Ie9e7a1816d197d05cf845e81251ef7772721b849
CRs-Fixed: 3071743
For DA MC/BC RX frame, host should not use da_idx to access soc->ast_table
as this da_idx is not valid, OB accessing might happen.
Change-Id: I5c78d869e32536effe2635e561cb6881cdc97c38
CRs-Fixed: 3072841
Currently, to set checksum enable flags in Tx descriptor
we are checking the csum_enabled flag along with stack checksum
offload request. In case of roaming from latency-critical connection
to non-latency critical connection we request stack to calculate
checksum and set csum_enabled flag to 0. For some packets which
are already queued where the stack has not calculated checksum and
csum_enabled flag is set to 0, we will not set the checksum enable
flags in Tx descriptors for those packets.
To fix the issue remove the csum_enabled flag check and directly
check if the stack has calculated checksum for those packets or not.
Change-Id: I8d7754afc3d0a33315b85b0113cd3062e5783e28
CRs-Fixed: 3070858
