Cleaned up formatting and changed the way console messages are printed.
The message will now properly show up in dmesg output with color, and
date formatting rather than a non descriptive message like the following:
Before:
```
/lib/firmware/qca-nss0-retail.bin 33
/lib/firmware/qca-nss1-retail.bin 33
```
After:
```
[Mon Jul 8 07:21:12 2024] hotplug: symlinking /lib/firmware/qca-nss0.bin to /lib/firmware/qca-nss0-retail.bin
[Mon Jul 8 07:21:12 2024] hotplug: symlinking /lib/firmware/qca-nss1.bin to /lib/firmware/qca-nss1-retail.bin
```
Signed-off-by: Sean Khan <datapronix@protonmail.com>
Reworked flawed smp_affinity logic. It was assumed that all 'nss_*' IRQs
needed to be equally balanced across all CPUs. However, what ends up
happening most of the time is messing with scheduling, thrashing, missed
cache and high context switching.
For each core:
1. nss_empty_buf_sos
2. nss_empty_buf_queue
3. nss_queue0
The IRQs should be kept on the same CPU. Doing so allows NSS cores to
properly feed and empty the queues.
Signed-off-by: Sean Khan <datapronix@protonmail.com>
allow users to specify specific metric to view, rather than
everything
Available stats objects:
cpu
crypto_cmn
dma
drv
dynamic_if
edma
eth_rx
ipv4
ipv6
n2h
unaligned
virt_if
wifi_ext_vdev
wifi
mesh
Usage: nss_stats obj1 obj2 ...
Default with no parameters shows all
Signed-off-by: Sean Khan <datapronix@protonmail.com>
1.) Rather than modify upstream `smp_affinity` script, move changes into
this init script.
Primarily: "xhci-hcd:usb1", "ppdu-end-interrupts-mac1", "ppdu-end-interrupts-mac2"
2.) Move logging into separate function, and better account for NSS core
specific output.
3.) Set the defaults for `enable_rps` and `enable_log` to true ('1').
4.) Change startup to sequence to `94` [after `smp_affinity` (93)]
Signed-off-by: Qosmio <datapronix@protonmail.com>
1.) Moved `set_affinity` function outside of `enable_rps`
2.) Added 2 new functions `bitmask_to_cpus` and `cpus_to_bitmask`
a.) bitmask_to_cpus - Takes a bitmask of CPUs and returns a
list of CPU numbers. (i.e. `bitmask_to_cpus "f"` -> 0,1,2,3)
b.) cpus_to_bitmask - Takes a comma/space or range list of CPUs and returns
a bitmask. (i.e. `cpus_to_bitmask "2,3"` -> c)
3.) Changed logic of `set_affinity` to now use physical CPUs rather than
knowing the bitmask. This should make it more straight forward when
testing changes
4.) Added an UCI option `enable_log` to enable/disable logging output to
`logger`. Default is `1` (on)
5.) Removed unused UCI options `nss_firmware *`
6.) Changed the way UCI options are retrieved to account for missing
options.
7.) Moved '/lib/debug/qca-nss-drv' to '/usr/bin/nss_stats' so it's in
$PATH and has a more intuitive name
Using the previous version, not all nss related interrupts were being balanced across the available CPU cores.
This makes sure they are as well as moving a bit more off cpu0.
To keep fork as closely synced with upstream, move NSS packages back
into repository. Not sure why they were moved out from my original fork.
* nss-firmware
* qca-nss-crypto
* qca-nss-cfi
Removed the following:
* mhz (already available in packages repo)
* qrtr (unecessary, and has been broken for years)
Also moved packages out of `qca` and back into root directory.
