For non-NSS platform, update no of rx packets being
sent from wifi driver to network stack in case of
vow traffic.
Change-Id: If16a5b9c37a16374d4217369b1f02360c62155a9
CRs-Fixed: 2371429
Splitting the wds srcport learn function to add:
1. A wrapper function where host extracts the required fields from
nbuf cb and rx_tlv header.
2. A common function which can be called from both host path
as well as offload path.
Change-Id: I2f2c0580c049f48395a3e0a265e3fb5d8aed6774
Add cdp api to check if tx desc pool available descriptor
threshold has reached.
Change-Id: Ie542d03dd865d32aa6e01da00328aa51728b4276
CRs-Fixed: 2369218
The last_ack_rssi value is made to get updated with the
correct value and the code fragment to reset it to 0 has
been removed.
Change-Id: I87f9ca788c92ae6ffc05b10faeb82e03024050ce
Current driver doesn't check for any TX pending flag before doing
bus suspend.
Add a logic which is similar to existing helium platform.
Change-Id: I49d078c3b86fc0d9659fbbc2f3c1a604a79a9dff
CRs-Fixed: 2360189
Stats are updated periodically and
are categorized to soc, pdev, and
vdev
While the stats are getting updated
and if pdev, or vdev gets detached
in the same time, the stats handlers
will deference a NULL pointer
Hence Add basic sanity checks to DP
layer for soc, pdev and vdev inside
stat handlers
Change-Id: Ic4919b9c205679e1d6e7c571c577339be65c1bad
Change the QDF log level during update peer statistics. It is not
necessary because during roaming it is possible the peer is not
there when we update statistics. Also use QDF_TRACE_RL to replace
QDF_TRACE to avoid massive log prints.
CRs-Fixed: 2353709
Change-Id: I2896e6874b2802ee1d8bf05c6daf064e1953a5d5
Add a missing NULL check for peer object in rx_da_learn API
Also add missing elements in pdev_stats_cmn_tlv structure.
CRs-Fixed: 2357352
Change-Id: Ie54fd0a6a45f00d60420c391098f22b0097de3ac
dp_reo_desc_free function can be executed in SIRQ context, so it
should not be expected that INFO printing which might cause locks
holding too long.
CRs-Fixed: 2367454
Change-Id: Ic77126fa62329547494ec6b672111c6183fadac2
Add support for GRO functionality for TCP packets.
- Pass GRO config to DP layer as a part of dp_update_config_parameters
- Add API to read the number of active RX contexts from DP layer
cdp_get_num_rx_contexts
- Fill GRO info into skb->cb from rx_tlv during dp_rx_process
CRs-Fixed: 2346995
Change-Id: I1c143d8ce2c7522ba2c76142fb6cc82193df5608
Change with fix memory leak observed in case where
smart monitor is enabled and monitor direct bit is not set
Change-Id: Ib556482b9067dbd7418cbe226a19bbbd5844a075
In dp_rx_mon_status_buffers_replenish after replenish of all
mon status ring rx buffers there is log in which function
local desc list is accessed. But after successful replenish of
all the available free desc, free desc list will be NULL. So
accessing the free desc list in the debug log leads to NULL
pointer dereference. In this change remove the debug log as
before that enough logging is in place to check how many buffers
are replenished.
Change-Id: I306aa52f6f25d9b43afb8548b8c03467aa9f27de
CRs-Fixed: 2366585
During dp_rx_pdev_attach and dp_rx_pdev_mon_buf_attach we are allocating
three times as many rx descriptors as there are entries in the ring for
AP usecase. This is not needed for MCL and may need to kmalloc
failures for allocating that much contiguous memory
Allocate only as many RX descriptors as there are ring entries for MCL
usecase.
Change-Id: I8b559a85c3899bcbdc520e71ba5da409314db160
CRs-Fixed: 2342957
1. Reduce the log level for every buffer received
from firmware
2. Set the default pktlog buffer size to 10MB.
Change-Id: Iad8827084579592fcfa69056dcfc9d57095151e7
CRs-Fixed: 2364380
Current driver doesn't clean-up all the peers because the TARGET
state is not marked as reset and due to which nobody is there to
mimic fake peer unmap events to reduce the reference count.
Also, there is no peer delete API being called as well in case of SSR.
Identify the TARGET state in-case of SSR by calling HDD callback
and mimic the unmap events and call peer delete API as well.
CRs-Fixed: 2350285
Change-Id: I878b7a15701c01ac9ef80010086f7e868a631a74
In existing code, 132 bytes of OTA mgmt. and control payload is
provided to upper layers by subscribing to packet header TLV.
The requirement in M-copy mode though is to get complete payload for
mgmt. and ctrl. packets. Add change to configure monitor destination
ring and add mgmt. and ctrl. packet filter to receive full mgmt. and
ctrl. packets.
Change-Id: I0014837b51cac6f0143dcc5cb624ea2086cf4486
CRs-Fixed: 2256159
Currently there is only tso segment and tso number segment
descriptor free operation when tso preparation failed, do dma
unmap for each segment as well.
Change-Id: I9a507f27341877dc062528ce77e02970d16aa098
CRs-Fixed: 2347279
Add support for smart monitor feature on HKv2.
For HKv2, NAC filtering is done by HW and NAC is
programmed directly in the AST table. The NAC entries
in AST table have monitor direct bit set and this
is obtained by host via monitor status ring.
host subscribes for packet header in MD packets.
To these filtered packets radiotap header is added
and delivered to the stack.
Change-Id: I123f986531943e376ac5c492540e01f0b03348c4
For MCL function dp_rx_process_invalid_peer, there is corner case the
nbuf is not data and freed, but invalid_peer_head_msdu pointe to the
same nbuf in pdev is not reset to NULL, then invalid_peer_head_msdu is
reused again.
Reset invalid peer msdu list to NULL when nbuf is freed.
Change-Id: I73058afe9656c34971ea083faa19ef9d90b7f747
CRs-Fixed: 2356531
During SSR test, TX skb sent to target will not get TX completion,
so skb unmap and free is missed. later when driver is unloaded,
nbuf unmap leak is detected then panic happened.
Take care to unmap and free the corresponding skb of pending TX
descriptor during vdev detach.
Change-Id: I9f4e6443682097ec76632c96a0188ffa2c1a5fcc
CRs-Fixed: 2347770
Pass vdev mac-addr and vdev-mode, instead of vdev-id.
The vdev-id that was passed earlier, was used to get
the mac-addr and vdev-mode from the vdev structure.
But there are cases when this vdev structure may be
freed, when this API is called.
So pass the mac-addr and vdev-mode from the dp structures.
Change-Id: I2148e7ecef433f4363385eeaec2e18f3ec131ab8
CRs-Fixed: 2356315
Link descriptors should be returned to WBM release ring if
fragments are not reinjected due to defrag errors.
Change-Id: Ia37db9f195f6848092918cf7cc221dc50e827ac5
If there is anything wrong, the caller of dp_tx_send_msdu_single
will only handle skb free, but not skb unmapping. So inside of
dp_tx_send_msdu_single, it is needed to make sure unmap operation
is performed after map operation.
CRs-Fixed: 2358266
Change-Id: I6379305e84f70b5b978e9504b758c7a6ccf328ee
During vdev attachment in monitor mode, the ini_timer is not
initialzied which will cause FW assertion. Make sure int_timer
is initialized properly.
CRs-Fixed: 2359908
Change-Id: I48c97161345b17a0eb3e8a939499eabeae683bd2
When from control path peer is deleted, peer deletion in progress
is set in dp for that peer. So any caller tries to get the peer
using peer id the return peer reference will be NULL as deletion
is in progress. In this case during peer deletion in progress at
tx completion process status console logs are flooded with invalid
peer logs in soft irq context. This is leading to scheduler watch
dog bite. In this change reduce the severity to prevent excessive logs.
Change-Id: Ife8f9e496b81b19666ef9e36db674106111a7d43
CRs-fixed: 2357807
This FR is to enhance existing pktlog debug tool
This feature will allow to capture pktlog for particular
peer mac address.
Change-Id: I3676095536185f25b0d498e03f70246260a324fd
Eanable monitor mode low water mark interrupt.The monitor
mode processing is stuck in the monitor mode buffer ring full
if monitor mode low water mark interrupt is not enabled.This
is because there is no packet come in when monitor buffer
ring is full and hence there is no ppdu end interrupt.The
monitor is not processing the monitor ring and buffer ring
full condition is not remoevd.
Change-Id: I28b3a4b408db62873c17512d20e63c9844e8d4ea
Because of reference counter decrement is not triggered for every
condition path, it is possible that peer reference counter leaking
happens under some perticular circumstance. Make sure the peer ref
count unreference is called for each possible path.
CRs-Fixed: 2352433
Change-Id: I60bc5a5210519c26c57cdc563d0d1b02d799e090
WAPI cert AP sends rekey frames as unencrypted. Thus
RXDMA will report unencrypted frame error and error
frames are then sent to release ring as configured.
With current implementation, host just drops WAPI
error frames.
To pass WAPI cert case, fix is to deliever unencrypted
WAPI frames to stack.
Change-Id: I6dfd1b5e6958d729ec3da0a0d1f1e6c227b0597f
CRs-Fixed: 2350485
Check if the ast_entry peer is same as the peer received
in MEC event before removing the wds entry. This check is
required for wds repeater scenarios. When a multicast packet
goes from station to the rootap via the repeater, a mec event
will be received in host but the entry should not be replaced.
Change-Id: Ia6fae9f08b411b481fc6d7c3f829bd127f92caec
Assign dp_ath_rate_lpf return value to avg_tx_rate,
which is further used to compute ppdu_tx_rate
Change-Id: Ib51be85c2436bef8feffeb611982314720399a69
CRs-Fixed: 2356831
int_ctx variable is not defined but it is being used which creates
compilation error. Define this variable as part of the fix.
CRs-Fixed: 2359507
Change-Id: I17e57663eb8daa449616ee99d5eb9455e3b7cee2
Need to use tid_lock to protect the rx_tid data structure in the
dp_rx_reorder_flush_frag function. Or else there will be race
condition issues.
CRs-Fixed: 2357226
Change-Id: I860a64b529f0548eced7b537c4180a2c57175a56
MCL DP initialization path is running into issues due to change
I5732453f617bdc16995fda916b645c41845c3ecb. Fix the same.
Move flow control lock variables to re-use region. This is needed since
flow control deinit happens in dp_soc_detach, while in dp_soc_deinit, the
soc memory is being partially reset.
De-init HTT memory in dp_soc_deinit and free HTT handle
in dp_soc_detach. The existing code was not freeing HTT
packets for MCL case and freeing the HTT handle twice.
CRs-Fixed: 2359409
Change-Id: I52dcccc0444d4ce139a29ffcb884598bbacda88f
Initialize the PN to the value which is given by upper layer
and plumb it to HW.
Change-Id: If6b14d2b9b92e4571a85cb20dc445608d5f56961
CRs-Fixed: 2350954
Avoid memory fragmentation that happens during
attach-detach flow.
- Reuse transmit allocated static pool memory across soc up/down.
These memories are allocated during soc attach.
- Reuse DP source ring memory, DP soc context, DP pdev context
across soc up/down.
- Reorganise structure members of DP soc and DP pdev so that
we can zero out structure members across soc up/down
- Add cdp soc init/deinit and cdp pdev init/deinit that
will be active across soc up/down
Change-Id: I5732453f617bdc16995fda916b645c41845c3ecb
ASE entry aging for MEC entry type was same as
WDS type for 120 secs, this creates an issue when
device roams from MEC to WDS connection by not
aging out till 120secs. So now we reduce the MEC
aging time to 1sec
Change-Id: I6ab3b095a85c6d09f9fcd95b3b39f7152064d297
After gerrit#2436786, vdev in monitor mode is removed from vdev
list of pdev. So it is needed to add one more cdp API to get vdev
in monitor mode.
CRs-Fixed: 2343290
Change-Id: I9d74302a464280881ce0cf25dbc8c4f9d8b8d48d
When reap data from REO2SW ring and if the TP catches up to cached
HP, current logic will update HP again and give a second chance to
reap data. but the TP is not updated to ring register, if more data
is received on this ring then HP easily catches up to TP and the ring
is full.
Update TP as well if HP get chance be updated.
Change-Id: I92f381d2a31f3be987866a829e05b5578098649a
CRs-Fixed: 2334435
The ack_rssi_valid flag to set to 1 only when completion_status
value is 0 which shows that the TX was successful.
Change-Id: I9e520422995e374d7d35b9db0c060d1fce2f9067
