The bootloader can support zImage linux kernel which can decrease
the firmware image size.
Signed-off-by: Shiji Yang <yangshiji66@qq.com>
Link: https://github.com/openwrt/openwrt/pull/16662
Signed-off-by: Robert Marko <robimarko@gmail.com>
There are instances where `skbuff_recycle` messes up skb headers causing
various issues in network related activities. This was initially meant
to resolve issues with memory usage issues on 6.1.
Introduce a proc option to enable or disable skbuff_recycle. The default
is to have it enabled. Users on 512M platforms should test if memory
leaks still occur with this option disabled.
echo 0 > /proc/net/skb_recycler/skb_recycler_enable
echo 1 > /proc/net/skb_recycler/skb_recycler_enable
NOTE: There is a new script in `qca-nss-drv` that will automatically
manage this as an UCI configurable option.
commit 217a4fa43ba2ad0a02e8080dfde11ad23534aa8c (HEAD -> NSS-12.5-K6.x)
Author: Sean Khan <datapronix@protonmail.com>
AuthorDate: Sat Jul 13 18:43:52 2024 -0400
Commit: Sean Khan <datapronix@protonmail.com>
CommitDate: Sat Jul 13 18:43:52 2024 -0400
nss-drv: add skb_recycler init script
Signed-off-by: Sean Khan <datapronix@protonmail.com>
These packages are required for basic NSS offloading. So ensure they're
selected for all qualcommax targets.
* kmod-qca-nss-ecm (Enhanced Connection Manager)
* kmod-qca-nss-drv (NSS Driver)
* kmod-qca-nss-drv-bridge-mgr (NSS Manager for Bridges)
Signed-off-by: Sean Khan <datapronix@protonmail.com>
Cleaned up erronous settings for disabled features. The syntax should be
"# CONFIG_SOME_OPTION is not set", rather than simply commenting it out.
Removed left over config options from 5.15, and enabled features
recommended by QSDK:
```
CONFIG_IRQ_TIME_ACCOUNTING=y
CONFIG_MULTI_IRQ_HANDLER=y
CONFIG_HANDLE_DOMAIN_IRQ=y
CONFIG_HZ_FIXED=0
CONFIG_NO_HZ=y
```
Signed-off-by: Sean Khan <datapronix@protonmail.com>
Revert to using the default kernel timer (CONFIG_HZ_100).
Setting the timer to 1000hz is unecessary and just adds CPU overhead. It
also causes atomic scheduling issues in various threaded tasks,
especially when using threaded NAPI.
When the kernel timer frequency is set to 1000 Hz, it generates interrupts
1000 times per second. This increased frequency can lead to more
frequent context switches and interrupt handling, which can
interfere with other kernel activities.
Since network packet processing is offloaded to the NSS cores, the main CPU
does not need high-resolution timer interrupts to manage network
traffic efficiently. The NSS cores handle this independently.
This is the setting used and recommended by Qualcomm for all their IPQ
platforms.
Signed-off-by: Sean Khan <datapronix@protonmail.com>
NSS patches should always get applied last as to avoid fuzzing issues
with upstream patches.
Also cleanup patches that introduced spaces as indents vs. tabs
Signed-off-by: Sean Khan <datapronix@protonmail.com>
ath11k_nss: drop upstreamed patch
A variation of this patch 'wifi: mac80211: Use flexible array in struct ieee80211_tim_ie'
was upstreamed in commit 'e6e4d90b69027d74ba7e9a0a586b95a5714e10c0'.
Remove it to avoid compilation issues.
Signed-off-by: Sean Khan <datapronix@protonmail.com>
ath11k_nss: Fix patch conflict
Signed-off-by: Sean Khan <datapronix@protonmail.com>
1. Fix function to check for bridge master status while checking
for Bridge VLAN filter feature is enabled on bridge slave ports.
2. Disable default PVID for bridges during device registration in
the system.
Change-Id: Ibea6559c1b0700a2300b60e20d57b7818e23a8a8
Signed-off-by: Vishnu Vardhan Bantanahal <quic_vishvard@quicinc.com>
bridge: Fix Bridge VLAN stats update
This patch fixes Bridge VLAN stats update for both bridge master
and bridge slave.
Change-Id: Ia26f4c71e83e27dd83336815cda5c05c8c3f24ff
Signed-off-by: Vishnu Vardhan Bantanahal <quic_vishvard@quicinc.com>
bridge: Add bridge VLAN filter APIs for offload for 6.1 Kernel
Change-Id: I54e44c26664f86ae024f54605a032713a9a3eee5
Signed-off-by: Vishnu Vardhan Bantanahal <quic_vishvard@quicinc.com>
Signed-off-by: Sean Khan <datapronix@protonmail.com>
Disable software input output translation lookaside buffer (swiotlb) as
it wastes memory on low memory platforms (512m or less)
qualcommax: only disable swiotlb for platforms <= 512M
Disabling swiotlb is only required for platforms with 512M or less to save
64mb.
The performance of Qualcomm Crypto Engine is 10x slower than kernel
based encryption via ARMv8 crypto extensions. Rather than building it
into the kernel allow users to select it as module which can be tested with
other HW based crypto APIs via cryptodev-linux.
QSDK NSS builds utilize skbuff recycling for better handling of memory.
On a Dynalink DL-WRX36 (pbuf script should be set to 'auto') a significant drop in
memory usage was observed as well consistent sustained RX/TX speeds.
BEFORE:
echo 3 >! /proc/sys/vm/drop_caches
free -m
total used free shared buff/cache available
Mem: 867 338 547 90 101 528
Swap: 0 0 0
AFTER:
total used free shared buff/cache available
Mem: 867 242 594 1 81 624
Swap: 0 0 0
NOTE:
For 512MB platforms, users need to test with the following scenarios,
as the patch `999-233-ath11k-Disable-rx_header-tlv-for-2K-SKB.patch` is
really only testable on platforms with 512M or less RAM.
1.) Explicitly setting 'ATH11K_MEM_PROFILE_512M' on and see if system
crashes on boot.
2.) Explicitly setting 'ATH11K_MEM_PROFILE_1G'
3.) Remove patches
999-233-ath11k-Disable-rx_header-tlv-for-2K-SKB.patch
999-311-ath11k-configure-nss-thread-priority-during-pdev_ini.patch
And re-test with #1 and #2
It was incorrectly assumed that setting a 512M for 1G platforms would save
memory, instead it needs to be explicitly set to know proper memory
regions, otherwise it would cause fw crash.
Signed-off-by: Sean Khan <datapronix@protonmail.com>
Attempt at getting NSS WiFi Offload working on kernel 6.1 (backport 6.5)
These patches are a mix of @ansuel branch for 5.15 (ipq807x-5.15-ecm-wifi)
3ecaee768e
And QUIC's upstream NSS patches on [NHSS.QSDK.12.4.5.r3](https://git.codelinaro.org/clo/qsdk/oss/system/feeds/wlan-open.git)
For the actual package setup (Makfiles,ath.mk) Qualcomm's branch is under 'wlan-open/NHSS.QSDK.12.4.5.r3'
Looks like they've swtiched to using backports 6.5-rc3, and figured to
use a mix of their patches on 'NHSS.QSDK.12.4.5.r3' (nss)
'wlan-open/NHSS.QSDK.12.4.5.r3' (syncing with openwrt main).
Booting with "nss_offload=1 frame_mode=2"
Logs show:
[ 16.606282] WARNING: CPU: 2 PID: 3524 at ath11k_nss_tx+0x1d4/0x1e0 [ath11k]
[ 16.611060] Modules linked in: ecm pppoe ppp_async nft_fib_inet nf_flow_table_inet ath11k_ahb ath11k pptp pppox ppp_generic nft_reject_ipv6 nft_reject_ipv4 nft_reject_inet nft_reject nft_redir nft_quota nft_objref nft_numgen nft_nat nft_masq nft_log nft_limit nft_hash nft_fullcone nft_flow_offload nft_fib_ipv6 nft_fib_ipv4 nft_fib nft_ct nft_compat nft_chain_nat nf_tables nf_nat nf_flow_table nf_conntrack_netlink nf_conntrack mac80211 iptable_mangle iptable_filter ipt_REJECT ipt_ECN ip_tables cfg80211 xt_time xt_tcpudp xt_tcpmss xt_statistic xt_multiport xt_mark xt_mac xt_limit xt_length xt_hl xt_ecn xt_dscp xt_comment xt_TCPMSS xt_LOG xt_HL xt_DSCP xt_CLASSIFY x_tables wireguard slhc sch_cake qrtr_smd qrtr qmi_helpers nfnetlink nf_reject_ipv6 nf_reject_ipv4 nf_log_syslog nf_defrag_ipv4 mhi libcrc32c libchacha20poly1305 compat sch_tbf sch_ingress sch_htb sch_hfsc em_u32 cls_u32 cls_route cls_matchall cls_fw cls_flow cls_basic act_skbedit act_mirred act_gact qca_nss_cfi_cryptoapi
[ 16.611215] qca_nss_crypto qca_nss_qdisc qca_nss_wifi_meshmgr qca_nss_gre ledtrig_usbport qca_mcs bonding ip6_gre ip_gre gre ifb nat46 nf_defrag_ipv6 sit qca_nss_drv ip6_tunnel tunnel6 tunnel4 nls_utf8 nls_iso8859_1 nls_cp437 vxlan seqiv michael_mic uas usb_storage leds_gpio xhci_plat_hcd xhci_pci xhci_hcd dwc3 dwc3_qcom qca_nss_dp qca_ssdk ramoops reed_solomon pstore gpio_button_hotplug ext4 mbcache jbd2 aquantia hwmon crc_ccitt crc32c_generic
[ 16.721723] CPU: 2 PID: 3524 Comm: hostapd Not tainted 6.1.68 #0
[ 16.743958] Hardware name: Dynalink DL-WRX36 (DT)
[ 16.750032] pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 16.754635] pc : ath11k_nss_tx+0x1d4/0x1e0 [ath11k]
[ 16.761403] lr : ath11k_nss_tx+0x1bc/0x1e0 [ath11k]
[ 16.766264] sp : ffffffc00b1db790
[ 16.771122] x29: ffffffc00b1db790 x28: 0000000000000038 x27: 0000000000000000
[ 16.774602] x26: ffffff8006dd7430 x25: ffffff8007ea9dc8 x24: 0000000060000012
[ 16.781719] x23: ffffff8005c6a060 x22: ffffff8005c6a060 x21: ffffff80072f12d8
[ 16.788837] x20: ffffff8007ea9dc8 x19: ffffff8006082800 x18: 0000000000000005
[ 16.795957] x17: 6976312065707974 x16: 207061636e652062 x15: 6b73207874207373
[ 16.803075] x14: ffffffc00a0dd518 x13: 000000000000018b x12: 000000000000018b
[ 16.810192] x11: 00000000ffffffea x10: ffffffc00a135518 x9 : 0000000000000001
[ 16.817310] x8 : 0000000000000001 x7 : 0000000000017fe8 x6 : c0000000ffffefff
[ 16.824429] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
[ 16.831547] x2 : ffffff80072f1370 x1 : 0000000000000000 x0 : 0000000000000001
[ 16.838665] Call trace:
[ 16.845774] ath11k_nss_tx+0x1d4/0x1e0 [ath11k]
[ 16.848035] ath11k_mac_tx_mgmt_pending_free+0x3034/0x9600 [ath11k]
[ 16.852551] ieee80211_handle_wake_tx_queue+0x68/0xb10 [mac80211]
[ 16.858801] ieee80211_probereq_get+0xca4/0x11f0 [mac80211]
[ 16.865049] ieee80211_tx_prepare_skb+0x1dc/0x240 [mac80211]
[ 16.870432] ieee80211_xmit+0xcc/0x120 [mac80211]
[ 16.876333] __ieee80211_subif_start_xmit+0x2b8/0x380 [mac80211]
[ 16.880937] ieee80211_subif_start_xmit+0x40/0x3b0 [mac80211]
[ 16.887013] ieee80211_subif_start_xmit_8023+0xb8/0x430 [mac80211]
[ 16.892656] dev_hard_start_xmit+0x8c/0x110
[ 16.898728] __dev_queue_xmit+0x1f0/0xbc0
[ 16.902808] dev_queue_xmit+0x14/0x20
[ 16.906975] packet_sendmsg+0x768/0x1260
[ 16.910620] __sys_sendto+0xdc/0x140
[ 16.914613] __arm64_sys_sendto+0x28/0x40
[ 16.918174] invoke_syscall.constprop.0+0x5c/0x110
[ 16.922081] do_el0_svc+0x58/0x170
[ 16.926765] el0_svc+0x18/0x60
[ 16.930149] el0t_64_sync_handler+0x114/0x120
[ 16.933190] el0t_64_sync+0x174/0x178
[ 16.933197] ---[ end trace 0000000000000000 ]---
[ 16.933230] ath11k c000000.wifi: failed to transmit frame -22
[ 17.041495] ath11k c000000.wifi: encap mismatch in nss tx skb encap type 1vif encap type 2
[ 17.041578] ath11k c000000.wifi: failed to transmit frame -22
Which is confusing since the parameter is definitley set to '2'.
➤ cat /sys/module/ath11k/parameters/frame_mode
2
Booting with either "nss_offload=1 frame_mode=3" (RAW), or "nss_offload=1 frame_mode=1" (Native Wifi)
Results in:
[ 15.644742] ath11k c000000.wifi: peer not found for nss peer delete
[ 15.744742] ath11k c000000.wifi: peer not found for nss peer delete
[ 15.745742] ath11k c000000.wifi: peer not found for nss peer delete
[ 15.746682] ath11k c000000.wifi: peer not found for nss peer delete
Clients connect and then are immediately kicked off, stuck in a loop.
ath11k-nss: Wifi offloading working (unstable)
Wifi offloading seems to be working, however it is not stable. Logs will
initially show a lot of flooding.
```
[Sun Dec 17 02:43:25 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:28 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:30 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:32 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:34 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:40 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:41 2023] ath11k c000000.wifi: peer not found for nss peer delete
[Sun Dec 17 02:43:46 2023] ath11k c000000.wifi: peer not found for nss peer delete
```
But eventually ends up subsiding. Needs heavy testing...
ath11k_nss: Add mac address to debug `nss peer delete` warnings
also delete ath10k patch, and refresh.
ath11k_nss: FIX `nss peer delete` + encap errors
This should hopefullly fix the following errors:
```
[Sun Dec 24 22:03:21 2023] ath11k c000000.wifi: encap mismatch in nss tx skb encap type 1vif encap type 2
[Sun Dec 24 22:03:21 2023] ath11k c000000.wifi: failed to transmit frame -22
[Sun Dec 24 22:08:25 2023] ath11k c000000.wifi: encap mismatch in nss tx skb encap type 1vif encap type 2
[Sun Dec 24 22:08:25 2023] ath11k c000000.wifi: failed to transmit frame -22
```
Clients should also be able to switch between 2Ghz and 5Ghz AP, as well
as join using password (no longer required to boot with NSS off first,
connect the client to acquire PSK, and reboot back with nss wifi offload
enabled).
Uptime has been 5 hours and so far no issues.
ath11k_nss: Remove unecessary clang-tidy formatting
ath11k_nss: add 'debug_mode' flag quiet warnings
Will properly handle:
```
[Mon Dec 25 16:51:34 2023] ath11k c000000.wifi: encap mismatch in nss tx skb encap type 1 vif encap type 2
```
should take the path for native wifi encap
ath11k_nss: Rework depends, make `nss_redirect` optional
1.) Added a reworked `qca-nss-pbuf` init script from QSDK that will set sysctl
NSS `n2hcfg` wifi options based on board type and available
RAM.
2.) ath11k is fully capable of offloading wifi, it does not need
mac80211 to create any NSS vifs. This lowers overhead and has actually
shown considerable speed improvements for CPU programs on the router
side (i.e. SMB). I believe ax3600's IoT (ath10k) would require it
for offloading.
I've added 2 new makemenu options, `MAC80211_NSS_SUPPORT` which enables
NSS related features, and `MAC80211_NSS_REDIRECT` which provides the option to
autoload the module on boot with `nss_redirect=1`.
3.) Reverted the option that deafulted `ATH11K_MEM_PROFILE_512M` to
true. This was not required for platforms with 1GB or more memory
(which is most of them). The default is 1GB. Select for platforms
like Xiaomi AX3600.
ath11k_nss: remove requirement for '/sys/kernel/debug/ath11k'
ath11k_nss: set default values lower to avoid eating up memory
ath11k_nss: Add NSS VLAN support + improvements
1) tweaked qca-nss-pbuf script to better handle different board configs
2) added some QSDK NSS patches for dynamic vlan, WDS.
3) added QSDK fixes for beacon, and bt-coex
For devices that keep getting disconnected (mostly on 2G). It is
recommended to use the following settings for wifi. Change for your
country.
config wifi-device 'radio0'
option type 'mac80211'
option path 'platform/soc/c000000.wifi'
option band '5g'
option txpower '24'
option country 'PA'
option channel '100'
option htmode 'HE160'
option cell_density '0'
config wifi-iface 'default_radio0'
option device 'radio0'
option network 'lan'
option mode 'ap'
option encryption 'psk2'
option key 'SOME_PASSWORD'
option ssid 'EXAMPLE_5G'
option dtim_period '3'
option max_inactivity '86400'
option disassoc_low_ack '0'
config wifi-device 'radio1'
option type 'mac80211'
option path 'platform/soc/c000000.wifi+1'
option band '2g'
option htmode 'HT20'
option channel '1'
option txpower '36'
option country 'PA'
option cell_density '0'
config wifi-iface 'default_radio1'
option device 'radio1'
option network 'lan'
option mode 'ap'
option ssid 'EXAMPLE_2G'
option encryption 'psk2'
option key 'SOME_PASSWORD'
option max_inactivity '86400'
option dtim_period '3'
option disassoc_low_ack '0'
WPA3 should be disabled, and left as WPA2. Experimenting with max_inactivity, dtim_period, and disassoc_low_ack to hopefully resolve the deauth disconnects.
Signed-off-by: bitthief <bitthief@protonmail.ch>
qualcommax: dts: provide label for NSS reserved-memory
Provide a label for the NSS reserved-memory node so it can be easily passed
to the NSS DRV instead of having to global match by name which is fragile.
Signed-off-by: bitthief <bitthief@protonmail.ch>
