For some target like ROME, CE try to read more data than expected as
prefetch. For example in UDP TX case, when CE read a tx_desc(44 bytes)
located in host memory with SMMU mapped address 0x1025FFD4~0x1025FFFF,
SMMU will detect CE read beyond 0x1025FFFF to access 0x10260000. SMMU
fault will happen if 0x10260000 is not mapped to host memory.
To fix this issue, allocate contiguous dma buffer for all 1056 tx_descs,
which will map to contiguous SMMU address region.
CRs-Fixed: 3588459
Change-Id: Id0287b051f792f18d746baf39e1c66d076c9be3c
Prepare and Send WMI command to fw to set the rf path. Also, add module
ID for rf path switch for qdf log prints.
CRs-Fixed: 3568221
Change-Id: I7ef0ffe66eaed837424699641d3e5b8537cd0304
Use the VDEV response timer infrastructure to hold MAC address update
requests sent to FW on VDEV which is in link switch progress.
As post disconnect and sending set MAC address request, the scheduler
thread is exited and to notify MLO mgr about set MAC address update
complete (or either timer expiry due to FW response timeout) and
further the link switch process based on response.
This code is only applicable for Link switch VDEV MAC address update
path. For MAC update request from userspace still use the same path
without starting the timer as for those wait event is already in place.
Change-Id: Ice3e6f7b00f0d9d08d6aa62ee9c1e4d183142358
CRs-Fixed: 3556517
Currently, there are no statistics whatsoever present in WBUFF module.
To start with, add some basic statistics to the WBUFF module.
Change-Id: I63c06cfb8b1430b2e86539f88025e6d74a00012b
CRs-Fixed: 3548509
SKBs that are allocated using __netdev_alloc_skb() share a single
page frag cache that is maintained in the network layer and is
common for all network drivers/devices. There is always a chance
where memory can be fragmented when the WLAN driver shares a page
frag cache with other slow network devices. Over the time, lot of
memory would be wasted due to fragmentation eventually resulting
in SKB allocation failures due to OOM.
To circumvent this, define a QDF NBUF allocation API which accepts
custom page frag cache maintained in the driver as an input and
allocates memory for skb->head from the custom page frag cache.
Such an API will be of great help when used for allocating reusable
SKBs in the driver. This also avoids the aforementioned memory
fragmentation issue.
Change-Id: I33f3096bba4057fd06ef55bbed5dc7a3f0f5c759
CRs-Fixed: 3543424
Currently in driver, two macros are defined to specify the
mac address printing format. Since both macros achieve
the same result, replace all instances of QDF_FULL_MAC_FMT
with QDF_MAC_ADDR_FMT.
Change-Id: I195448267ef8e77a7fd5b232ffbf2cfb0ded1be4
CRs-Fixed: 3559488
qdf_walt_get_cpus_taken under WALT_GET_CPU_TAKEN_SUPPORT feature
flag to avoid compilation failure on branch where changes is not
present.
Change-Id: Iddcc777a476e41cd20fa0ff40ad2dd80af0748ef
CRs-Fixed: 3559247
It adds a set of frag way allocation APIs explicitly.
IPA Tx nbufs change to frag way allocation API, which saves memory.
Current __qdf_nbuf_alloc changes to non-frag way allocation by
default if SKB RECYLCER not defined.
Change-Id: I5f87ffac54c49f9af920775c13b6dfdd147476dd
CRs-Fixed: 3534452
Add support for filling fast_xmit in SKB for platform
which support skb_recycler feature.
Add a QDF abstraction to update fast_xmit in skb, when it is supported.
Change-Id: Ibc562023ddeb0bf5c2709bbb60b596676c932b5a
CRs-Fixed: 3522028
The TX_NBUF_CNT never decreases, which is an incorrect debug info.
We should decrease dp_tx_skb_count when nbuf unmap.
Change-Id: I3be63047f47305c5d24c5c40696e1c6db69ec4d6
CRs-Fixed: 3508530
Currently, opt_wifi_dp filter reserve and release
notification to IPA, is sent in a tasklet
context. As QMI calls tend to acquire mutexes, this
will cause an assert as we cannot sleep in a tasklet
context. To resolve this, WLAN will schedule the
notification call in a workqueue before calling
into IPA. This change ensures IPA calls from WLAN are
not made in an interrupt context.
Change-Id: Ic716e14340651db76d7dd7dc216c595c01732ee8
CRs-Fixed: 3492465
Currently connectivity logging is done for all opmode
but logging should be done only for STA mode.
So to fix the issue add check for opmode and skip
connectivity logging if opmode is not STA
Change-Id: Ie6175a02c8167f5ebfd81553837206640431f701
CRs-Fixed: 3515365
qdf_get_log_timestamp_lightweight function is defined only when
ENHANCED_OS_ABSTRACTION is not enabled, defining it when
ENHANCED_OS_ABSTRACTION is enabled as well.
Change-Id: Ia146feed5536a7b99653d7690f3204e6cbaf3f4c
CRs-Fixed: 3471529
Move prealloc DP descriptor types to QDF so that
the macros can be used in HIF layer.
Change-Id: I3de60876735e5aa37d80e9e698a86503b18574c1
CRs-Fixed: 3502615
Add new QDF API qdf_walt_get_cpus_taken to check if
walt_get_cpus_taken API is supported.
Change-Id: I8ed8e5608c35c45b5e630ee39a497bca31b3502d
CRs-Fixed: 3508584
The number of perf CPU clusters can vary across
different targets. Currently only the second CPU
cluster is considered as the perf cluster, whereas
there can be more than 1 perf clusters.
Fix this perf cluster derivation for interrupt
affinity.
Change-Id: Ie4407064ac5124fc050715fd75400ca516c9c6b8
CRs-Fixed: 3498024
On some third-party platforms, we observe the memory physical
address below 0x2000 is allocated will cause HW/FW NOC error,
so this region memory should not use by host.
This change will hold the memory if the allocated memory physical
address below 0x2000 until driver unload.
Change-Id: I37b6abc98033230dc4f572dafb849101497f6e93
CRs-Fixed: 3419648
PMAC is giving fatal when mgmt packet is getting queued
with non-zero version in frame control
Adding a check for version before queueing mgmt packets.
Change-Id: I606513b6ce027f1126e9cd40a9909886441de240
CRs-Fixed: 3422973
qdf assert always, assert without debug information.
Added support to dump needed information before assert.
Change-Id: I19898ec11c1bade4340cf1737a760d9507012f12
CRs-Fixed: 3480596
Add new QDF APIs to handle gpio, irqf and timer related
handling in TSF feature.
Change-Id: Iff6f85c6debe351c5533906559400b4a51333d4d
CRs-Fixed: 3469020
Fix the following 2 memory leaks in Rx monitor path:
1. When MPDU queue is empty and MON_BUF_ADDR_TLV is received,
then free page fragment memory.
2. In case of small size packets, 1 MPDU can have more than
2*QDF_NBUF_MAX_FRAGS fragments wherein each nbuf can have maximum of
QDF_NBUF_MAX_FRAGS frags. In this case, add the frags to nbuf in the
following way.
parent_nbuf (QDF_NBUF_MAX_FRAGS frags attached)
|
| (fraglist)
|
----> tmp_nbuf1 (QDF_NBUF_MAX_FRAGS frags attached) ----> tmp_nbuf2
(next)
Change-Id: I54e8162bf0b9da8629a3c80d123421fbeaf8df11
CRs-Fixed: 3453676
Add an API qdf_file_read_bytes to read a file from within the driver
and return the contents of it along with its size.
Change-Id: If777e1b9c610e8cc7dd35be42fa9ced2bb1c6560
CRs-Fixed: 3456115
qdf assert always, assert without debug information.
Added support to dump needed information before assert.
Change-Id: I2b53b7fa6a600b2a1aafdbc608b95b75feb2133d
CRs-Fixed: 3464724
As part of TX Mon 2.0, WIN has IEEE80211 specific macros only in WIN side
header file, but now Tx Mon 2.0 is required to be supported on MCC side
and these macros are not available. Hence extend by adding the required
type/subtype macros in qdf common header file.
Change-Id: I76cf0a0c226664c34c9c8bd10e280a500b3c61a9
CRs-Fixed: 3415715
On some third-party platforms, we observe the memory physical
address below 0x2000 is allocated will cause HW/FW NOC error,
so this region memory should not use by host.
This change will hold the memory if the allocated memory physical
address below 0x2000 until driver unload.
Change-Id: I9d5df20cbdf23c230d7c910cce9b9489dd9dd614
CRs-Fixed: 3419648
Some IOT APs only allow to connect if last 3 bytes of
BSSID and self MAC is same. They create a new bssid on
receiving unicast probe/auth req from STA and allow STA to
connect to this matching BSSID only. So boost the matching BSSID
to try to connect to this BSSID.
And add logic to refresh the candidate list before next
candidate try when the last candidate connect fail.
Change-Id: I482e122c8c9febbab42f64013fbb78c266f49655
CRs-Fixed: 3432618
IPA_OFFLOAD path will be enabled by default for MSM,
inorder to support optional wifi datapath feature.
This change is a WAR to disable IPA offload during
compile time, if the feature is not enabled.
This is done using the IPA_WDI_OPT_DPATH feature flag
from IPA. If the flag is not defined in the IPA test
module file, IPA offload path and optional wifi dp
feature will be disabled.
This is a WAR, and will be fixed once a Kernel config is
available from IPA to enable or disable the optional
wifi datapath feature.
Change-Id: Ic1b961656f52fa009ab4c3b3d8d3cac305c2c7c4
CRs-Fixed: 3431972
To support splitphy and MLO operation when multiple
TX ring are present under single soc, then to differentiate
the TX lmac, pmac id in per packet is require. As pmac_id in
bank register is always constant for pinnacle case, so need
to pass per packet in TX desc.
Change-Id: Ic6d8e14e8c1948e218c52c8eeaf99fcd6c437753
CRs-Fixed: 3356413
