NAPI poll can be done in threaded context along with soft irq
context. Threaded context can be scheduled efficiently, thus
creating less of bottleneck during Rx processing. This patch is
to enable threaded NAPI on ath10k driver.
Based on testing, it was observed that on WCN3990, the CPU0 reaches
100% utilization when napi runs in softirq context. At the same
time the other CPUs are at low consumption percentage. This
does not allow device to reach its maximum throughput potential.
After enabling threaded napi, CPU load is balanced across all CPUs
and following improvments were observed:
- UDP_RX increase by ~22-25%
- TCP_RX increase by ~15%
Here are some of the additional raw data with and without threaded napi:
==================================================
udp_rx(Without threaded NAPI)
435.98+-5.16 : Channel 44
439.06+-0.66 : Channel 157
udp_rx(With threaded NAPI)
509.73+-41.03 : Channel 44
549.97+-7.62 : Channel 157
===================================================
udp_tx(Without threaded NAPI)
461.31+-0.69 : Channel 44
461.46+-0.78 : Channel 157
udp_tx(With threaded NAPI)
459.20+-0.77 : Channel 44
459.78+-1.08 : Channel 157
===================================================
tcp_rx(Without threaded NAPI)
472.63+-2.35 : Channel 44
469.29+-6.31 : Channel 157
tcp_rx(With threaded NAPI)
498.49+-2.44 : Channel 44
541.14+-40.65 : Channel 157
===================================================
tcp_tx(Without threaded NAPI)
317.34+-2.37 : Channel 44
317.01+-2.56 : Channel 157
tcp_tx(With threaded NAPI)
371.34+-2.36 : Channel 44
376.95+-9.40 : Channel 157
===================================================
Tested-on: WCN3990 hw1.0 SNOC WLAN.HL.3.2.2-00696-QCAHLSWMTPL-1
Signed-off-by: Abhishek Kumar <kuabhs@chromium.org>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20230203000116.v2.1.I5bb9c164a2d2025655dee810b983e01ecd81c14e@changeid
This loop checks that i < max at the start of loop but then it does
i++ which could put it past the end of the array. It's harmless to
check again and prevent a potential out of bounds.
Fixes: 1048643ea9 ("ath5k: Clean up eeprom parsing and add missing calibration data")
Signed-off-by: Dan Carpenter <error27@gmail.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/Y+D9hPQrHfWBJhXz@kili
As of commit a1a2b7125e ("of/platform: Drop static setup of IRQ
resource from DT core"), we need to use platform_get_irq() instead of
platform_get_resource() to get our IRQs because
platform_get_resource() simply won't get them anymore.
This was already fixed in several other Atheros WiFi drivers,
apparently in response to Zeal Robot reports. An example of another
fix is commit 9503a1fc12 ("ath9k: Use platform_get_irq() to get the
interrupt"). ath5k seems to have been missed in this effort, though.
Fixes: a1a2b7125e ("of/platform: Drop static setup of IRQ resource from DT core")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20230201084131.v2.2.Ic4f8542b0588d7eb4bc6e322d4af3d2064e84ff0@changeid
As of commit a1a2b7125e ("of/platform: Drop static setup of IRQ
resource from DT core"), we need to use platform_get_irq() instead of
platform_get_resource() to get our IRQs because
platform_get_resource() simply won't get them anymore.
This was already fixed in several other Atheros WiFi drivers,
apparently in response to Zeal Robot reports. An example of another
fix is commit 9503a1fc12 ("ath9k: Use platform_get_irq() to get the
interrupt"). ath11k seems to have been missed in this effort, though.
Without this change, WiFi wasn't coming up on my Qualcomm sc7280-based
hardware. Specifically, "platform_get_resource(pdev, IORESOURCE_IRQ,
i)" was failing even for i=0.
Tested-on: WCN6750 hw1.0 AHB WLAN.MSL.1.0.1-00887-QCAMSLSWPLZ-1
Fixes: a1a2b7125e ("of/platform: Drop static setup of IRQ resource from DT core")
Fixes: 00402f49d2 ("ath11k: Add support for WCN6750 device")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Tested-by: Jun Yu <junyuu@chromium.org>
Reviewed-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20230201084131.v2.1.I69cf3d56c97098287fe3a70084ee515098390b70@changeid
Wakeup/release MHI is not needed before pci_read/write for QCN9274.
Since wakeup & release MHI is enabled for all QCN9274 and
WCN7850, below MHI assert is seen in QCN9274
[ 784.906613] BUG: sleeping function called from invalid context at drivers/bus/mhi/host/pm.c:989
[ 784.906633] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 0, name: swapper/3
[ 784.906637] preempt_count: 503, expected: 0
[ 784.906641] RCU nest depth: 0, expected: 0
[ 784.906644] 2 locks held by swapper/3/0:
[ 784.906646] #0: ffff8ed348e429e0 (&ab->ce.ce_lock){+.-.}-{2:2}, at: ath12k_ce_recv_process_cb+0xb3/0x2f0 [ath12k]
[ 784.906664] #1: ffff8ed348e491f0 (&srng->lock_key#3){+.-.}-{2:2}, at: ath12k_ce_recv_process_cb+0xfb/0x2f0 [ath12k]
[ 784.906678] Preemption disabled at:
[ 784.906680] [<0000000000000000>] 0x0
[ 784.906686] CPU: 3 PID: 0 Comm: swapper/3 Tainted: G W O 6.1.0-rc2+ #3
[ 784.906688] Hardware name: Intel(R) Client Systems NUC8i7HVK/NUC8i7HVB, BIOS HNKBLi70.86A.0056.2019.0506.1527 05/06/2019
[ 784.906690] Call Trace:
[ 784.906691] <IRQ>
[ 784.906693] dump_stack_lvl+0x56/0x7b
[ 784.906698] __might_resched+0x21c/0x270
[ 784.906704] __mhi_device_get_sync+0x7d/0x1c0 [mhi]
[ 784.906714] mhi_device_get_sync+0xd/0x20 [mhi]
[ 784.906719] ath12k_pci_write32+0x75/0x170 [ath12k]
[ 784.906729] ath12k_hal_srng_access_end+0x55/0xc0 [ath12k]
[ 784.906737] ath12k_ce_recv_process_cb+0x1f3/0x2f0 [ath12k]
[ 784.906776] ? ath12k_pci_ce_tasklet+0x11/0x30 [ath12k]
[ 784.906788] ath12k_pci_ce_tasklet+0x11/0x30 [ath12k]
[ 784.906813] tasklet_action_common.isra.18+0xb7/0xe0
[ 784.906820] __do_softirq+0xd0/0x4c9
[ 784.906826] irq_exit_rcu+0x88/0xe0
[ 784.906828] common_interrupt+0xa5/0xc0
[ 784.906831] </IRQ>
[ 784.906832] <TASK>
Adding function callbacks for MHI wakeup and release operations.
QCN9274 does not need wakeup/release, function callbacks are initialized
to NULL. In case of WCN7850, shadow registers are used to access rings.
Since, shadow register's offset is less than ACCESS_ALWAYS_OFF,
mhi_device_get_sync() or mhi_device_put() to wakeup
and release mhi will not be called during service ring accesses.
Tested-on: QCN9274 hw2.0 PCI WLAN.WBE.1.0-03171-QCAHKSWPL_SILICONZ-1
Tested-on: WCN7850 hw2.0 PCI WLAN.HMT.1.0-03427-QCAHMTSWPL_V1.0_V2.0_SILICONZ-1.15378.4
Signed-off-by: Ramya Gnanasekar <quic_rgnanase@quicinc.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20230123095141.5310-1-quic_rgnanase@quicinc.com
hif_dev->remain_skb is allocated and used exclusively in
ath9k_hif_usb_rx_stream(). It is implied that an allocated remain_skb is
processed and subsequently freed (in error paths) only during the next
call of ath9k_hif_usb_rx_stream().
So, if the urbs are deallocated between those two calls due to the device
deinitialization or suspend, it is possible that ath9k_hif_usb_rx_stream()
is not called next time and the allocated remain_skb is leaked. Our local
Syzkaller instance was able to trigger that.
remain_skb makes sense when receiving two consecutive urbs which are
logically linked together, i.e. a specific data field from the first skb
indicates a cached skb to be allocated, memcpy'd with some data and
subsequently processed in the next call to ath9k_hif_usb_rx_stream(). Urbs
deallocation supposedly makes that link irrelevant so we need to free the
cached skb in those cases.
Fix the leak by introducing a function to explicitly free remain_skb (if
it is not NULL) when the rx urbs have been deallocated. remain_skb is NULL
when it has not been allocated at all (hif_dev struct is kzalloced) or
when it has been processed in next call to ath9k_hif_usb_rx_stream().
Found by Linux Verification Center (linuxtesting.org) with Syzkaller.
Fixes: fb9987d0f7 ("ath9k_htc: Support for AR9271 chipset.")
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20230216192301.171225-1-pchelkin@ispras.ru
Set the WMI_SCAN_FLAG_EXT_PASSIVE_SCAN_START_TIME_ENHANCE flag
while sending the scan command. If this flag is enabled when the
incoming scan request comes with a strict start time and its duration
overlaps with next TBTT, then target adjust the start time accordingly
for passive scan. Target supporting this feature will advertise
WMI_TLV_SERVICE_PASSIVE_SCAN_START_TIME_ENHANCE.
Tested-on: IPQ8074 hw2.0 AHB WLAN.HK.2.4.0.1-01467-QCAHKSWPL_SILICONZ-1
Signed-off-by: Tamizh Chelvam Raja <quic_tamizhr@quicinc.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20221222131720.11368-1-quic_tamizhr@quicinc.com
This documentation wasn't added in the original patch,
add it now.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Fixes: 6e4c0d0460 ("wifi: mac80211: add a workaround for receiving non-standard mesh A-MSDU")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When we have multiple interfaces receiving the same frame,
such as a multicast frame, one interface might have a sta
and the other not. In this case, link_sta would be set but
not cleared again.
Always set link_sta, so we keep an invariant that link_sta
and sta are either both set or both not set.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
There's at least one case in ieee80211_rx_for_interface()
where we might pass &((struct sta_info *)NULL)->sta to it
only to then do container_of(), and then checking the
result for NULL, but checking the result of container_of()
for NULL looks really odd.
Fix this by just passing the struct sta_info * instead.
Fixes: e66b7920aa ("wifi: mac80211: fix initialization of rx->link and rx->link_sta")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When a connection was established without going through
NL80211_CMD_CONNECT, the ssid was never set in the wireless_dev struct.
Now we set it in __cfg80211_connect_result() when it is not already set.
When using a userspace configuration that does not call
cfg80211_connect() (can be checked with breakpoints in the kernel),
this patch should allow `networkctl status device_name` to output the
SSID instead of null.
Cc: stable@vger.kernel.org
Reported-by: Yohan Prod'homme <kernel@zoddo.fr>
Fixes: 7b0a0e3c3a (wifi: cfg80211: do some rework towards MLO link APIs)
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216711
Signed-off-by: Marc Bornand <dev.mbornand@systemb.ch>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The convention for find_first_bit() is 0-based, while ffs()
is 1-based, so this is now off-by-one. I cannot reproduce the
gcc-9 problem, but since the -1 is now removed, I'm hoping it
will still avoid the original issue.
Reported-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Fixes: 1d8d4af434 ("wifi: mac80211: avoid u32_encode_bits() warning")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The ieee80211_accept_frame() function performs a number of early checks
to decide whether or not further processing needs to be done on a frame.
One of those checks is the ieee80211_is_robust_mgmt_frame() function.
It requires to peek into the frame payload, but because defragmentation
does not occur until later on in the receive path, this peek is invalid
for any fragment other than the first one. Also, in this scenario there
is no STA and so the fragmented frame will be dropped later on in the
process and will not reach the upper stack. This can happen with large
action frames at low rates, for example, we see issues with DPP on S1G.
This change will only check if the frame is robust if it's the first
fragment. Invalid fragmented packets will be discarded later after
defragmentation is completed.
Signed-off-by: Gilad Itzkovitch <gilad.itzkovitch@morsemicro.com>
Link: https://lore.kernel.org/r/20221124005336.1618411-1-gilad.itzkovitch@morsemicro.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
gcc-9 triggers a false-postive warning in ieee80211_mlo_multicast_tx()
for u32_encode_bits(ffs(links) - 1, ...), since ffs() can return zero
on an empty bitmask, and the negative argument to u32_encode_bits()
is then out of range:
In file included from include/linux/ieee80211.h:21,
from include/net/cfg80211.h:23,
from net/mac80211/tx.c:23:
In function 'u32_encode_bits',
inlined from 'ieee80211_mlo_multicast_tx' at net/mac80211/tx.c:4437:17,
inlined from 'ieee80211_subif_start_xmit' at net/mac80211/tx.c:4485:3:
include/linux/bitfield.h:177:3: error: call to '__field_overflow' declared with attribute error: value doesn't fit into mask
177 | __field_overflow(); \
| ^~~~~~~~~~~~~~~~~~
include/linux/bitfield.h:197:2: note: in expansion of macro '____MAKE_OP'
197 | ____MAKE_OP(u##size,u##size,,)
| ^~~~~~~~~~~
include/linux/bitfield.h:200:1: note: in expansion of macro '__MAKE_OP'
200 | __MAKE_OP(32)
| ^~~~~~~~~
Newer compiler versions do not cause problems with the zero argument
because they do not consider this a __builtin_constant_p().
It's also harmless since the hweight16() check already guarantees
that this cannot be 0.
Replace the ffs() with an equivalent find_first_bit() check that
matches the later for_each_set_bit() style and avoids the warning.
Fixes: 963d0e8d08 ("wifi: mac80211: optionally implement MLO multicast TX")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20230214132025.1532147-1-arnd@kernel.org
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently the regulatory driver does not call the regulatory callback
reg_notifier for self managed wiphys. Sometimes driver needs cfg80211
to calculate the info of ieee80211_channel such as flags and power,
and driver needs to get the info of ieee80211_channel after hint of
driver, but driver does not know when calculation of the info of
ieee80211_channel become finished, so add notify to driver in
reg_process_self_managed_hint() from cfg80211 is a good way, then
driver could get the correct info in callback of reg_notifier.
Signed-off-by: Wen Gong <quic_wgong@quicinc.com>
Link: https://lore.kernel.org/r/20230201065313.27203-1-quic_wgong@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Wi-Fi Aware R4 specification defines NAN Pairing which uses PASN handshake
to authenticate the peer and generate keys. Hence allow to register and transmit
the PASN authentication frames on NAN interface and set the keys to driver or
underlying modules on NAN interface.
The driver needs to configure the feature flag NL80211_EXT_FEATURE_SECURE_NAN,
which also helps userspace modules to know if the driver supports secure NAN.
Signed-off-by: Vinay Gannevaram <quic_vganneva@quicinc.com>
Link: https://lore.kernel.org/r/1675519179-24174-1-git-send-email-quic_vganneva@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Non-MLO station frames are dropped in Rx path due to the condition
check in ieee80211_rx_is_valid_sta_link_id(). In multi-link AP scenario,
non-MLO stations try to connect in any of the valid links in the ML AP,
where the station valid_links and link_id params are valid in the
ieee80211_sta object. But ieee80211_rx_is_valid_sta_link_id() always
return false for the non-MLO stations by the assumption taken is
valid_links and link_id are not valid in non-MLO stations object
(ieee80211_sta), this assumption is wrong. Due to this assumption,
non-MLO station frames are dropped which leads to failure in association.
Fix it by removing the condition check and allow the link validation
check for the non-MLO stations.
Fixes: e66b7920aa ("wifi: mac80211: fix initialization of rx->link and rx->link_sta")
Signed-off-by: Karthikeyan Periyasamy <quic_periyasa@quicinc.com>
Link: https://lore.kernel.org/r/20230206160330.1613-1-quic_periyasa@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Stations can update bandwidth/NSS change in
VHT action frame with action type Operating Mode Notification.
(IEEE Std 802.11-2020 - 9.4.1.53 Operating Mode field)
For Operating Mode Notification, an RX NSS change to a value
greater than AP's maximum NSS should not be allowed.
During fuzz testing, by forcefully sending VHT Op. mode notif.
frames from STA with random rx_nss values, it is found that AP
accepts rx_nss values greater that APs maximum NSS instead of
discarding such NSS change.
Hence allow NSS change only up to maximum NSS that is negotiated
and capped to AP's capability during association.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Link: https://lore.kernel.org/r/20230207114146.10567-1-quic_ramess@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
At least ath10k and ath11k supported hardware (maybe more) does not implement
mesh A-MSDU aggregation in a standard compliant way.
802.11-2020 9.3.2.2.2 declares that the Mesh Control field is part of the
A-MSDU header (and little-endian).
As such, its length must not be included in the subframe length field.
Hardware affected by this bug treats the mesh control field as part of the
MSDU data and sets the length accordingly.
In order to avoid packet loss, keep track of which stations are affected
by this and take it into account when converting A-MSDU to 802.3 + mesh control
packets.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20230213100855.34315-5-nbd@nbd.name
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The current mac80211 mesh A-MSDU receive path fails to parse A-MSDU packets
on mesh interfaces, because it assumes that the Mesh Control field is always
directly after the 802.11 header.
802.11-2020 9.3.2.2.2 Figure 9-70 shows that the Mesh Control field is
actually part of the A-MSDU subframe header.
This makes more sense, since it allows packets for multiple different
destinations to be included in the same A-MSDU, as long as RA and TID are
still the same.
Another issue is the fact that the A-MSDU subframe length field was apparently
accidentally defined as little-endian in the standard.
In order to fix this, the mesh forwarding path needs happen at a different
point in the receive path.
ieee80211_data_to_8023_exthdr is changed to ignore the mesh control field
and leave it in after the ethernet header. This also affects the source/dest
MAC address fields, which now in the case of mesh point to the mesh SA/DA.
ieee80211_amsdu_to_8023s is changed to deal with the endian difference and
to add the Mesh Control length to the subframe length, since it's not covered
by the MSDU length field.
With these changes, the mac80211 will get the same packet structure for
converted regular data packets and unpacked A-MSDU subframes.
The mesh forwarding checks are now only performed after the A-MSDU decap.
For locally received packets, the Mesh Control header is stripped away.
For forwarded packets, a new 802.11 header gets added.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20230213100855.34315-4-nbd@nbd.name
[fix fortify build error]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently action frames TX only with ML address as A3(BSSID) are
allowed in an ML AP, but TX for a non-ML Station can happen in any
link of an ML BSS with link BSS address as A3.
In case of an MLD, if User-space has provided a valid link_id in
action frame TX request, allow transmission of the frame in that link.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Link: https://lore.kernel.org/r/20230201061602.3918-1-quic_ramess@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
- New feature flag, NL80211_EXT_FEATURE_PUNCT, to advertise
driver support for preamble puncturing in AP mode.
- New attribute, NL80211_ATTR_PUNCT_BITMAP, to receive a puncturing
bitmap from the userspace during AP bring up (NL80211_CMD_START_AP)
and channel switch (NL80211_CMD_CHANNEL_SWITCH) operations. Each bit
corresponds to a 20 MHz channel in the operating bandwidth, lowest
bit for the lowest channel. Bit set to 1 indicates that the channel
is punctured. Higher 16 bits are reserved.
- New members added to structures cfg80211_ap_settings and
cfg80211_csa_settings to propagate the bitmap to the driver after
validation.
Signed-off-by: Aloka Dixit <quic_alokad@quicinc.com>
Signed-off-by: Muna Sinada <quic_msinada@quicinc.com>
Link: https://lore.kernel.org/r/20230131001227.25014-3-quic_alokad@quicinc.com
[move validation against 0xffff into policy]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
- Move ieee80211_valid_disable_subchannel_bitmap() from mlme.c to
chan.c, rename it as cfg80211_valid_disable_subchannel_bitmap()
and export it.
- Modify the prototype to include struct cfg80211_chan_def instead
of only bandwidth to support a check which returns false if the
primary channel is punctured.
Signed-off-by: Aloka Dixit <quic_alokad@quicinc.com>
Link: https://lore.kernel.org/r/20230131001227.25014-2-quic_alokad@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
