Restrict DMA Map/UnMap upto buffer size for packets in rx process.
This gives 2-3% cpu gain in peak throughput.
Change-Id: Iaf5e9f6f734d80b6d2c234bd8e679cf2a81c7e2c
CRs-Fixed: 2660698
Decouple IPA TCLring and IPA WBM completion ring size from regular
Tx TCLring and WBM completion ring size configuration.
This is required as there is limitation on IPA GSI rin size
configuration.
Change-Id: I689d0e8ca72f069c5b68dc1789358e091c554d30
CRs-Fixed: 2665010
Size of the TLVs have changed across generation of chipsets
Offset values need to be configured into DMA register for preheader DMA
Added APIs to get offsets of each TLV based on chip type
Change-Id: Ic011332cbf3a1017f324f246e47c9e2c91441c70
Validate number of wds entries deleted during peer unmap
handling with the number which firmware has sent in
peer unmap message, which indicates the number of
wds ast entries deleted by firmware after peer delete.
Change-Id: I09e1c41bab19cd023e7a83baf1e90d51aab4229e
CRs-fixed: 2667445
1) Modified MSI interrupt mask for QCN9000 so that rx and tx
interrupts can be decoupled.
2) Removed lmac interrupts from polling mode and switched
them to msi interrupts. Added MSI masks for lmac rings.
3) Enable monitor mode LWM interrupt. This was already enabled in
integrated ahb interrupts but missing in msi. Replenish buffers in
RXDMA refill ring based on low threshold interrupts in addition to
regular Rx processing. Also made interrupt batch counter threshold as 8
for monitor status ring since ppdu end interrupts are not available in
PCI chipset and require srng msi interrupts to reap monitor status ring
Change-Id: I5c84b14d6b0a9c26fb3f0d67c349e79751a60861
In current monitor status ring implementation,
on pdev_attach, (srng->num_entires – 1) entries
(to keep one entry slot between hp and tp)
are replenished and last entry is not replenished to HW.
With qcn9000 monitor mode HW enhancements, status and destination ring
can be made lock-stepped.
for qcn9000 lock step is achieved by making monitor status ring
follow the monitor destination reap for a PPDU
However in existing flow during attach monitor status replenish logic
do not fill last entry but is filled up during first subsequent reap.
for first ppdu, i.e. after reaping destination ring,
when status ring is reaped, as first entry (hp = srng->num_entires – 1)
in status ring is NULL, so lock-stepping is not achieved.
To address this issue for qcn9000 as well as HK:
a. Replenish last entry in monitor status ring during attach
b. Modify src srng peek API to peek it from hp+1 entry
c. Introduce new HAL API get cur desc and move next
d. Remove WAR to skip status ring entries if DMA is not done
Change-Id: I60b8e7c075253d37e6b849a9b24f473c5afce82c
CRs-Fixed: 2626049
- For directly connected peers, during peer de-auth,
a Single Peer delete request to the FW from host
should delete all AST entries attached to the peer.
- This optimization is aimed at optimizing host-fw
handshake messages
Change-Id: Iaebe2022f90ef9a10a6a0f37b21c409cf9e9ea39
CRs-fixed: 2647242
Enable monitor destination ring interrupt to fix
packet is getting stuck in monitor mode.
Fix and add monitor mode debug count
Change-Id: I74efdcf6a4373dd73e373285c8cd3aa5757cd0b2
During dp_tx_hw_enqueue writing into HW SRNG,
if devcie is already runtime suspended, set the
SRNG_FLUSH_EVENT, so that it will be flushed
as part of dp runtime resume callback.
Change-Id: I95e74844cf74eb4f73862da3da9d93422f1eea8c
CRs-Fixed: 2664660
The cleanup done in case of pdev attach failure
does not deinitialize the IPA tx and rx buffers.
Also the tx descriptors are not freed in this
cleanup path.
Due to this, the tx descriptors and IPA tx/rx
buffers memory is neither freed not unmapped.
Fix this pdev attach failure cleanup path to
avoid memory leaks.
Change-Id: I570e3618f0f48f56749466fce5bcba5a4ef1c161
CRs-Fixed: 2663036
Retrieve peer from mac address to avoid accessing peer
even if peer is deleted.
Change-Id: I1c9740fc5d3090ef296c4dc4a5afe7022e83dca5
CRs-Fixed: 2663836
The peer rx packets should be flushed when
deleting the peer and its state should be reset
to DISCONNECT, when deleting the peer.
If the state of peer is not set to DISCONNECT,
then the flushing of rx packets for the peer
which is being deleted will call the rx callback
and submit these packet to the stack, which can
cause unwanted behaviour.
This way the UMAC does not need to specifically
call clear peer before deleting the peer.
Change-Id: I3b5a737126350a361d968f6349aef6291b2e3f56
CRs-Fixed: 2659629
For wlan stop modules any pending vdevs flush happens as part
of pdev deinit. If in case of any peer ref count leak, driver
is stuck in dp_pdev_flush_pending_vdevs as the vdev is not getting
cleaned up. This is due to vdev's peer is not deleted. So during
force flush of vdev's peers set the ref count to one before peer
unmap handler.
Change-Id: Ie54996522eac870d00ea7a75c8db40a1cce09d2d
CRs-Fixed: 2658032
If tx capture, sniffer are not enabled, mgmt nbuf is freed.
It is causing use-after-free in bpr enabled case
Added change to free only when bpr is disabled
Change-Id: Ia56254ca371cd9fbc21f5d58fac7ea96792d0bee
CRs-Fixed: 2663098
For 4 core CPU system, Tx access can be made lockless by associating
one Tx ring per CPU. Three Tx DATA rings and one SW2TX command ring
can be used for sending data to Tx.
Change-Id: I6e38cad52dffa4e46ff2b736cd6b949d061fa2a4
Invalid peer handler does double free of nbuf and leads to crash.
To avoid this make nbuf->next NULL before calling dp_rx_null_q_desc_handle.
Change-Id: Ia3f92cb5136dbdbfeb1e9cda8a52c474456a4e0c
CRs-Fixed: 2655792
Read ini param 'full_mon_mode' and enable full
monitor mode support for qcn9000
CRs-Fixed: 2655293
Change-Id: I6550765246f6541453006fe2db28e01c8e06670c
As part of componentization of packet capture feature
removed cdp ops of packet capture.
Change-Id: Iea07cbdc5782dee58cd1c87377363cc26479464f
CRs-Fixed: 2659439
Remove lock to access REO destination rings because 4 rings are
accessed in 4 individual cores.
Change-Id: Ia3f92cc5136dbdbeea1e9cda8d52b474356a3e1a
CRs-Fixed: 2660901
In pdev deinit path tx_desc->vdev can be NULL
in dp_tx_comp_free_buf use soc->osdev to unmap instead
of vdev
Change-Id: I8b390656d482fed3c8c85d928860222a88325ca1
During the packet transfer from HOST to Transmit Classifier Layer,
if NAWDS is enabled, send peer id of NAWDS peer.
This is to avoid re-injection from FW to Host.
Change-Id: Ic0e5c46fff3bbe1a3e8ea3a01986478c925452f5
CRs-Fixed: 2628703
Update QCA_NAPI_DEF_SCALE_BIN_SHIFT for defconfig builds to
the same value as the perf build.
Move the WLAN ext irqs to gold cores for defconfig builds.
Disable cpu isolation before moving the IRQs to gold cores.
Add the ability to move IRQs to gold cores when the cpus
hotplug in.
Change-Id: I4cfecd02a1a2200dc99adee9a324b90c877c13fd
CRs-Fixed: 2638820
The dp_should_timer_irq_yield should return the
status DP_TIMER_NO_YIELD when no yield condition
has been met. Currently its returning the status
DP_TIMER_TIME_EXHAUST which causes the timer to yield
before the expected time.
Fix the return status from dp_should_timer_irq_yield
Change-Id: Ia57aae635789f34e79285c27f5312ea38e07df5c
CRs-Fixed: 2659900
Return DP_SW2HW_MACID if per_pdev_lmac is set.
This is needed when we need to get target pdev_id without
doing pdev lookup.
Change-Id: If83f29bbcf23175708ba7eaf58e01e040ccb5a45
This mode is supported through hw_mode_id 7. A special mode for
IPQ8074 platform alone to bring up single pdev alone on 2G mode.
FW would advertise the mode 7 support through the WMI service
ready message and this mode will be the default mode for AP-HK10 alone.
Added preferred_hw_mode and target_pdev_id into dp structures so that
they can be used to configure the dp rings for this particular mode.
Also changed the lmac_id to target_pdev_id mapping for mode 7.
Change-Id: I3aae46fea13132eebaf7d37cff68ef517253d229
Enable SW2TCL_CMD ring for data on QCN9000, QCA8074 V2/V1
and IPQ6018 targets.
Enabled 4th Tx ring for data enqueue to HW.
Transmit completions for packets from CPU0 and CPU3 are routed to
WBM2SW Completion ring 2.
WB2SW completion ring2 is mapped to CPU3.
Change-Id: Ied4c4704e1f8623e909ad45c547a611de26c7ec5
Use QDF_BUG instead of the qdf_assert. Since the qdf_assert is not
defined anymore.
Change-Id: I2bd46def51823829604b699321f9a02fcd3b07c6
CRs-Fixed: 2658545
Currently, in monitor mode, once the status ring is reaped
we do not check again if there are new entries added in the
ring during the monitor ring descriptor processing. This
can cause backpressure on monitor rings when there is very
high throughput or a bursty inflow of packets.
Add the logic to continue reaping the monitor rings, until
the work budget or a time quota is not exhausted.
Change-Id: Iee95ac8aec92eabb3a8037b1dd9cf0c6e1dbe10b
CRs-Fixed: 2650480
Currently the timer handler, in monitor mode, is
try to process all the rings based on the mask. In
monitor mode only LMAC rings are to be processed
and the UMAC ring processing can be avoided.
Process only the LMAC rings for monitor mode.
Change-Id: Ie48722d80722350129b5c079b67ede06945f5849
CRs-Fixed: 2650480
In monitor mode, the link descriptor return failure
can occur many times, due to the timer interval.
Logging each failure on the console floods it with
too many unnecessary logs.
Ratelimit the link descriptor return failure in monitor
mode and also change its log level to info.
Change-Id: I5709a33087c6d9271d4246f04ef1acd274fba0f1
CRs-Fixed: 2650480
The current time interval for monitor mode timer is
10 milliseconds. This time interval is a bit too high
in case of high throughput sniffing, thereby causes
the monitor rings to be full before the timer interrupt
comes and reaps the ring.
Decrease the timer interval to 5 milliseconds
Change-Id: Ib074ef0c09fde995a7481c056ad70eb809c5a89e
CRs-Fixed: 2650480
In mon dest ring, for m_copy mode, MGMT packets are received with
FCS error.Due to which multiple skbs (Head_skb with
attached frag_list) for single MPDU are sent to upper stack.
Add fix to drop FCS error frames in mcopy mode.
Change-Id: I6fbf2ea2bfdc489b56a412ece3c644ba2ce87929
Decouple software tx_descs from TCLring descriptors and Increase
TCLring descriptors max INI config limit to 2k to 32k.
Change-Id: I18d014cdb0c1e26b7661ff65aaa8dffe791d7163
CRs-Fixed: 2657056
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
Is some cases, FW wants to adjust BA session parameters
FW sends a HTT event and HOST helps to send out DELBA
and modify BA win size in the next round of ADDBA req/resp
exchange to reduce following RX AMPDU aggr-size
Change-Id: I97777ff59a18346f2cd4e2e562c8ae6acdcd69d7
CRs-Fixed: 2644641
MCL path add the desc back to freelist when FLUSH
failed. It may cause the CMD pending in free list
is even larger than REO_CMD_RING max size and lead CMD
flood then cause REO HW in an unexpected condition.
So it's needed to limit the number REO cmds in a batch
operation.
Change-Id: Ib6b97f0693feb7bb5829bda4683248b495b6c7b3
CRs-Fixed: 2642516
There might be aggressive error logging in the dmesg if wbm_desc_rel_ring
is hang. So use the dp_info_rl instead of QDF_TRACE to reduce aggressive
logging to avaoid spin lock huge delay when processing the
reo_exception_ring.
Change-Id: I4622f28cd6d8771cf27643a867892b62860d1ddc
CRs-Fixed: 2648149
Pointer rx_msdu_link_desc returned from call to function
dp_rx_cookie_2_mon_link_desc which may be NULL and may be
de-referenced latter
CR-Fixed: 2645199
Change-Id: I9ccba61df9571fcc99c5d5493194d5ae43a71a7f
For ACK completion, update ba_size, ba_bitmap, ba_seqno
only if these are not updated earlier.
This fix updates ba_size, ba_bitmap, ba_seqno for QOS TID frames
in ACK completion handler, if these are not updated earlier.
CRs-Fixed: 2650482
Change-Id: I1d5f4bdb29a3d5194cd4a1544b39d6aa42903b1b
Add mutex protection while freeing pkts in HTC misc list to prevent
double unmap of netbuf.
Change-Id: I5c52324e382f8047037ca1d1a50367bc6052b737
CRs-Fixed: 2646771
Dp_rx_vdev_detach() will flush all gro skb to make sure the
net_dev stored in skb is valid before this vdev is down, but
dp_rx_vdev_detach() is only invoked at the end of function
dp_vdev_detach_wifi3(). for some case, if there is still dp_peer
pending when execute dp_vdev_detach_wifi3(), dp_rx_vdev_detach()
won't be invoked.
Move dp_rx_vdev_detach() before pending peer detection in
dp_vdev_detach_wifi3() so that gro queue skb flushing can be
executed each time.
Change-Id: I0feea5b77de7efab884c0e5822c8671119892f7d
CRs-fixed: 2649385
Support RX 2K jump/OOR frame handling from REO2TCL ring.
(a) configure REO error destination ring register to route 2K jump
/OOR frame to REO2TCL ring.
(b) for 2K jump RX frame, only accept ARP frame and drop others,
meanwhile, send delba action frame to remote peer once receive first
2K jump data.
(c) for OOR RX frame, accept ARP/EAPOL/DHCP/IPV6_DHCP frame, otherwise
drop it.
Change-Id: I7cb33279a8ba543686da4eba547e40f86813e057
CRs-Fixed: 2631949
