The kernel-doc script identified a large number of documentation
issues in dp/wifi3.0/dp_tx_desc.[ch], so fix those issues. In
addition, there are a number of instances where public functions have
their implementation documented instead of having their interface
documented, so move that documentation.
Change-Id: I349f0e9d9336ae632fb31e88ddc34ecacc4a1f68
CRs-Fixed: 3373161
In case of MLO same adapter maps to both vdevs. Tx is done with
primary link vdev pool. In link switch case vdev0 flow pool is
deleted and traffic will continue to flow with same vdev0 flow_id
This is resulting in packets drops within the driver due non-availibilty
of flow pool.
Fix this by skipping flow pool deletion in case of STA MLO connection.
Change-Id: Ie56ea221f7f254484bd31ec4880f2155779fb513
CRs-Fixed: 3292979
DP uses multi page allocation for tx/rx descriptor.
ID and offset mask of decriptor is based on 4096 which
couples with Kernel's MMU PAGE_SIZE. This cause trouble
when deploy driver on none-4K page size kernel.
Set qdf_dp_blockmem_size to 4096 so that DP won't
depend on kernel page size.
Change-Id: I17f5c10b394e8709e6b4b153f3dd094cf792787f
CRs-Fixed: 3235246
-Wimplicit-fallthrough is being enabled by default. Some compilers
such as clang require the attribute instead of just a fallthrough comment.
Change-Id: I443da8d7f5e1771dceb3386c4458b0da6a5e9476
CRs-Fixed: 3218236
With restructuring in HIF runtime PM module, modules are
required to register with the HIF runtime PM module. Also,
changes are done in functions of allowing and preventing
runtime PM suspend as part of restructuring.
This change registers DP, HTC and HAL internal modules
with runtime PM module and update HIF runtime PM function
calls with the restructured code of HIF runtime PM module.
Change-Id: I8899a1d3b92a90a05c5eaf4df7609f4008f739f8
CRs-Fixed: 3169372
Use microsecond timestamping for the tx descriptors
for capturing the latencies more accurately. It will help
to determine the appropriate bucket.
Change-Id: I216e083a7c89b01b6f2f384c1c0a85ca323d3a3e
CRs-Fixed: 3165153
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
Race condition is observed when runtime suspend is aborted
due to pending tx completions, in this context all tx desc
are checked for corruption or leakage. In parallel tx completion
is received and descriptor timestamp is set to null. As a
result, while checking for leakage in runtime suspend context,
time difference for the tx completion is current jiffies which
is huge, resulting in triggering recovery.
To fix this, tx_desc timestamp is checked before calculating
time latency. Also, printing tx descriptor id for which
completion is not received.
Change-Id: I38b487a47170af374f43c44a96e2a0753f5d57ef
CRs-Fixed: 3117549
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
In flow control code switch logic break statements are not used
intentionally, but compiler is treating this as error.
So to avoid compilation error add fall through comment.
Change-Id: Ic79b75c48554182fba7c4c0f3885e3a44347e2e7
CRs-Fixed: 3090182
Due to DP flow control network queues are stuck in
paused state and not getting reset even after new connection.
This is causing DHCP packets to be dropped and resulting
immediate disconnection after every connection.
Fix this by properly pausing and unpausing the AC flow control
based network subqueues during disconnection/connection events.
Change-Id: I280e704d5a01b19d180c32aaa272c0cb731f8c0e
CRs-Fixed: 3078745
In tx desc store vdev_id to avoid unprotected access of
vdev memory.
In tx path pass VDEV as argument to all APIs
In completion path get vdev from peer as vdev reference
in peer is protected with reference, in places where peer
is invalid get vdev object using vdev_id stored in tx desc
with a reference count
Change-Id: I364ebb51143218872c0bc6f856772ab1df00bb17
Add debug support for VDEV refcount to take
refcount by module id and decrement corresponding
refcount with same module id
Change-Id: I15c075816994ba70155fefbc0bce208b20fb9a59
Add new API dp_vdev_get_ref_by_id() which will return vdev
pointer by holding a reference. Caller of this API has
to ensure that this reference is released by calling
dp_vdev_unref_delete() API
New lock soc->vdev_map_lock is introduced to protect
vdev id to object array
Change-Id: I883e328932e35ef31254125492dbae20cebe0e00
Current dp_tx_desc structure size is 136 bytes,
re-arranged the fields to fit in exactly 128 bytes to
align with number of descriptor per page and page offset
bit maps
Also add a compilation assert on tx desc size
Change-Id: Ifcb18a9da637cb71c0427b56ad4054d2206ddfbb
Currently when the last tx descriptor in any descriptor
pool is freed, it will also free that particular tx
descriptor pool. But before freeing the tx descriptor
pool, the pool is not marked as inactive.
Due to this, the next time when the tx descriptor pool
is tried to be allocated, since its not inactive, the same
pool is reattached, but without any memory allocated.
Deinitialize the tx desc pool before freeing it, to make
sure that the tx descriptor pool status is marked as
inactive when the decriptor pool memory is freed.
Change-Id: I52d792d115d215561c72f203abef6d327d126c01
CRs-Fixed: 2680397
Do a logical split of dp_soc_attach and
dp_pdev_attach into Allocation and initialization
and dp_soc_detach and dp_pdev_detach into
de-initialization and free routines
Change-Id: I23bdca0ca86db42a4d0b2554cd60d99bb207a647
Lot of checks in Tx completion path are for special handlings
such as when ol stats are enabled some protective debug checks etc..
Add a logic to fast free of buffer at transmit completion.
If extended stats not enabled(Typically needed for enterprise cases)
do the buffer free faster. This is controlled through a flag.
Change-Id: I04873b5e3643d8e93e5b248fcaf23504dcb7624f
This is to enhance statics for runtime put/get, which is
to detect if there is mismatch for usage_count.
Change-Id: I24cddb9d10e4cb675c8375cbd0f589c7718bd680
CRs-Fixed: 2647972
Currently the cdp apis are given pdev/vdev/peer
handle as its arguments, which is directly
accessed in those APIs. This can cause a
race-condition in access of the respective
handles if it has been deleted in parallel.
Hence as a part of cdp convergence, pass only
the pdev/vdev id or peer mac address, which will be
used to get the respective handles, and hence
avoiding the unwanted access of the handles if
it has been deleted.
Converged flowctl_ops
- flow_pool_map_handler
- flow_pool_unmap_handler
- dump_flow_pool_info
- tx_desc_thresh_reached
CRs-Fixed: 2539812
Change-Id: I084d3878df84778622340e87bddf90acd3e669d6
In some corner case, host has done Vdev destroy and TX desc flush
since there is pending outstanding TX data on this vdev, but later
FW/HW still indicate the TX completion to host and which likely
these TX completion get delayed, conflict happened due to same TX
Desc accessing.
To fix it:
(1) Only reset TX desc Vdev to NULL in dp_tx_vdev_detach().
(2) Do more enhancement in dp_tx_comp_handler() to avoid
invalid Vdev accessing.
(3) Do TX desc flush in dp_tx_pdev_detach().
Change-Id: I44cce9451e4f07fed6e21a9cd3fc0eaefa3d54c2
CRs-Fixed: 2441455
When the status is FLOW_POOL_ACTIVE_PAUSED, all queues are expected to
be paused. Still some on the fly packet can make it to the driver. In
such a case, the function dp_tx_desc_alloc takes a default action of
WLAN_WAKE_NON_PRIORITY_QUEUE. This is incorrect, since it will wake all
non priority queues, when they should be paused.
Make default action as WLAN_NETIF_ACTION_TYPE_NONE. If this is the
action to be taken, donot call the pause_cb from dp_tx_desc_alloc.
Rate limit log levels in case dp_tx_desc_alloc fails.
Change-Id: I1ef3018e90576d2c3aaa0d10d56e9b155681271b
CRs-Fixed: 2421813
Delay counters per TID have been implemented for following types:
1. Linux stack to hw enqueue delay
2. HW enqueue delay to tx completion delay
3. TX interframe delay
4. RX interframe delay
5. RX frame delay from ring reap to networking stack
Change-Id: I836596cbd878a43955c18b4981cb5b7b43d4df5e
Add cdp api to check if tx desc pool available descriptor
threshold has reached.
Change-Id: Ie542d03dd865d32aa6e01da00328aa51728b4276
CRs-Fixed: 2369218
There are several duplicated spin unlock function calling in the
AC based TX flow control path. Remove this redundant codes.
CRs-Fixed: 2329973
Change-Id: I483bf97612b8e56f745360a3e84fbdff8c357ea9
Tx descriptors that belong to a particular vdev are released in
vdev detach path. Since DP soc is not detached yet, interrupts
are not disabled, so it is possible that host gets completions
for same tx descriptor and it tries to process it again.
Add a check for vdev in tx completion path to avoid duplicate
processing of tx descriptors
Change-Id: I5a62ef4d981dbfd0a5ca7483acf4270145d016be
Implement per access category flow control. Make sure lower AC
traffic, such as BE, will not starve higher AC, such as VI and VO,
traffic inside a single vdev.
This change is part of FR49094
Change-Id: I6aa15832eb48e8f3ada00eb29a7bc00999edaafd
CRs-Fixed: 2304436
Tie increment and decrement PM usage count to descriptor allocation
and free instead of HW enqueue and TX completion since allocation
and free will be the most accurate places to track if the packet
has been sent or not.
Change-Id: Ia61c5bb26386d20590e87149d0fe88dba02d0fb7
CRs-fixed: 2302214
Clean up datapath component prints by correcting trace levels for
regularly occurring prints and removing newlines from converged
print APIs since qdf_trace_msh appends them by default.
Change-Id: Ie8fe319fcb737720f7400a165e134b6a18bd15b5
CRs-Fixed: 2243843
If host receive the tx completion with invalid tx_desc ID,
panic happened. add the tx desc ID check both on dp_tx_hw_enqueue
and dp_tx_comp_handler to know that how this invalid tx_desc ID
is generated by which reason of below two:
1. host tx desc is corrupted and fill in invalid id to TCL_DATA_CMD.
2. FW or HW generate invalid TX compl to host.
Change-Id: Id2b070697199b19f83e5d9fd89eadcfe0423d617
CRs-Fixed: 2243455
The avail_desc is not set to 0 when the tx desc pool is
freed, later if still try to allocate tx desc by judgement
of avail_desc, panic happens.
Change-Id: Ia4565da1caa6898c6d4293e2658cf4ccf89563fa
CRs-Fixed: 2246328
Enable multiqueue on VAP with 4 Tx and Rx queues in lithium.
In Rx path set the rx queue of skb based on the reo ring id on which it
is received.
In Tx path use the queue_mapping of skb to map to the hardware Tx ring
on which it has to be transmitted.
Change-Id: I103a21e91d1ed5c0e1d8441863d4fcd273b7bed9
Make sure to clear the tx_desc->flags field before putting the tx_desc to
free pool. This was not takencare in V2 flow_control implementation.
Change-Id: I666b9b09de62d52cffc45e2b6db3a6be06d688e8
CRs-Fixed: 2186884
As many as 5 vdevs can be created, which is defined by MAX_SME_SESSIONS.
But flow control pools are only valid for active TxRx capable vdevs.
which is 3 active vdevs. So attach flow pool creation to vdev_start
instead of vdev_attach.
Change-Id: I137676ba618aa3a6684c7d17ef7d9263c4df6efe
CRs-Fixed: 2220079
Replace internal spin_lock macros with qdf_spin_lock APIs for lock
operating on TSO/EXT descriptor pools.
TX_DESC_LOCK_* APIs are used only by the flow_control V2 support
to by pass them.
Change-Id: I926c279de3878b6f48efd798194fa896072c1f6d
CRs-Fixed: 2142815
Increase the number of bits allocated for page mask to
accomodate upto 16K descriptors needed in Tx path
Change-Id: I9a1a2757be353c3bd9f78773b87cca7c86041b9c
CRs-Fixed: 2004658
Adds support for flow control on convergence branch.
Allocate Tx descriptors dynamically when vdev comes up.
Tx queue is paused and unpaused internally in host based on the
stop and start thresholds.
Changes are added under compilation flag QCA_LL_TX_FLOW_CONTROL_V2.
Change-Id: I0ccb80b0099f39efad52ccd7d47f2709fdee2a93
CRs-Fixed: 2040457
