I’m using irqbalance in 24.10.5, but is it still necessary to use it in the new version 25.12?
Not:I'm using SQM and I get an A in the bufferblock when I don't install this package, but an A+ when I do.
Youreself claims it is necessary. In general kernel relies on irqbalance for irq balancing, itself doing nothing about it.
Why? It's not neccessary in 25.12?
If it helps you, just use it.
But Irqbalance is not that good in recognising properly the network related IRQs in non-x86 processors, so some targets/devices have manual IRQ allocations scripts in use. E.g. qualcommax/ipq807x:
https://github.com/openwrt/openwrt/blob/main/target/linux/qualcommax/ipq807x/base-files/etc/init.d/smp_affinity
4 Likes
For generic defects it does not detect drivers which balance by themselves already like nvme or mtk crypto
So, it shouldn't cause any problems in the new version; I was using it for my AX3000T xaomi and I can continue using it.
is there a script like this you sent for qualcomm/ipq40xx?
netgear orbi rbr50v1 I want the ethernet irqs on one core the wifi on other and adguard home on final core and the pci wifi chip won't move does this make sense?? is it actually a good idea or bad I have packet steering right now on on all cpu's with steering flow set to 128
Sorry, but no knowledge about ipq40xx.
The script currently used in ipq807x evolved from attempts like this:
What I did was not sure correct but here is what I used and added this to /etc/rc.local
# Pin Ethernet hardware queues to CPU1
for irq in 33 37 41 45 49 51 53 55; do
echo 2 > /proc/irq/$irq/smp_affinity
done
# Pin ath10k_ahb WiFi queues to CPU2 (Skipping PCI-MSI IRQ 66)
for irq in 67 68; do
echo 4 > /proc/irq/$irq/smp_affinity
done
this is the irq snapshot its recent snapshot there is alot of power outages here so I can't give a day uptime stats but I think it works? let me know if what I did was correct or not that's why when i saw your script I got surprised that what I'm using is so simple and you had a big script to do these manual irq pining
CPU0 CPU1 CPU2 CPU3
26: 243827 239436 264549 212939 GIC-0 20 Level arch_timer
29: 5 0 0 0 GIC-0 270 Level bam_dma
30: 0 0 0 0 GIC-0 239 Level bam_dma
31: 5 0 0 0 GIC-0 139 Level msm_serial0
33: 10 119782 0 0 GIC-0 97 Edge c080000.ethernet:txq0
37: 13 119566 0 0 GIC-0 101 Edge c080000.ethernet:txq4
41: 0 198340 0 0 GIC-0 105 Edge c080000.ethernet:txq8
45: 2 135304 0 0 GIC-0 109 Edge c080000.ethernet:txq12
49: 14 663016 0 0 GIC-0 272 Edge c080000.ethernet:rxq0
51: 24 23260 0 0 GIC-0 274 Edge c080000.ethernet:rxq2
53: 9 36592 0 0 GIC-0 276 Edge c080000.ethernet:rxq4
55: 12 40624 0 0 GIC-0 278 Edge c080000.ethernet:rxq6
57: 31 0 0 0 GIC-0 129 Level i2c_qup
58: 2607 0 0 0 GIC-0 155 Level mmc0
59: 2 0 0 0 GIC-0 170 Level 7824900.mmc
61: 0 0 0 0 PCI-MSI 0 Edge aerdrv
62: 0 0 0 0 msmgpio 22 Edge 7824900.mmc cd
63: 0 0 0 0 msmgpio 18 Edge keys
64: 0 0 0 0 msmgpio 49 Edge keys
65: 0 0 0 0 GIC-0 168 Level xhci-hcd:usb1
66: 308895 0 0 0 PCI-MSI 524288 Edge ath10k_pci
67: 24 0 403915 0 GIC-0 200 Level ath10k_ahb
68: 22 0 0 0 GIC-0 201 Level ath10k_ahb
IPI0: 0 0 0 0 CPU wakeup interrupts
IPI1: 0 0 0 0 Timer broadcast interrupts
IPI2: 2737 2963 2859 2936 Rescheduling interrupts
IPI3: 287103 22666 270386 322491 Function call interrupts
IPI4: 0 0 0 0 CPU stop interrupts
IPI5: 2 0 0 0 IRQ work interrupts
IPI6: 0 0 0 0 completion interrupts
Err: 0
You dont show baseline - ie no irqbalance no script. It would go better to spread eth queues one of type per core. It is network router after all...
ohh yeah sure I can show you that wait
1 Like
This was the baseline before I saved it in a text beforehand
CPU0 CPU1 CPU2 CPU3
26: 2994759 2602121 2157948 2240521 GIC-0 20 Level arch_timer
29: 5 0 0 0 GIC-0 270 Level bam_dma
30: 0 0 0 0 GIC-0 239 Level bam_dma
31: 5 0 0 0 GIC-0 139 Level msm_serial0
33: 715099 0 2049 2700 GIC-0 97 Edge c080000.ethernet:txq0
37: 449511 0 1367 6839 GIC-0 101 Edge c080000.ethernet:txq4
41: 546931 12 1005 1493 GIC-0 105 Edge c080000.ethernet:txq8
45: 4649119 0 132926 2560 GIC-0 109 Edge c080000.ethernet:txq12
49: 11336949 0 0 0 GIC-0 272 Edge c080000.ethernet:rxq0
51: 261855 0 59 2005 GIC-0 274 Edge c080000.ethernet:rxq2
53: 225113 0 855 16141 GIC-0 276 Edge c080000.ethernet:rxq4
55: 361264 618 2926 8221 GIC-0 278 Edge c080000.ethernet:rxq6
57: 34 0 0 0 GIC-0 129 Level i2c_qup
58: 8320 0 149 18 GIC-0 155 Level mmc0
59: 2 0 0 0 GIC-0 170 Level 7824900.mmc
61: 0 0 0 0 PCI-MSI 0 Edge aerdrv
62: 0 0 0 0 msmgpio 22 Edge 7824900.mmc cd
63: 0 0 0 0 msmgpio 18 Edge keys
64: 0 0 0 0 msmgpio 49 Edge keys
65: 0 0 0 0 GIC-0 168 Level xhci-hcd:usb1
66: 9042101 0 0 0 PCI-MSI 524288 Edge ath10k_pci
67: 2805044 2094 20596 3007 GIC-0 200 Level ath10k_ahb
68: 18 0 0 0 GIC-0 201 Level ath10k_ahb
IPI0: 0 0 0 0 CPU wakeup interrupts
IPI1: 0 0 0 0 Timer broadcast interrupts
IPI2: 21899 25645 25515 19680 Rescheduling interrupts
IPI3: 913878 5788607 3362563 15937409 Function call interrupts
IPI4: 0 0 0 0 CPU stop interrupts
IPI5: 6370 6376 6511 6515 IRQ work interrupts
IPI6: 0 0 0 0 completion interrupts
Err: 0
You mean spread the tx queues on one core and the rx on other? did I understand you correctly?
See upstream https://www.kernel.org/doc/html/latest/networking/scaling.html#rss-irq-configuration
You can make irqbalance do automatic irq assignments and exclude ones you tune manually. As much as network (or video card) irqs are concerned i'd trust standard tools to do the work.
1 Like
I know that you explained it completely fine for a person who has more knowledge but can you explain to me in a simpler way 
what I did was correct the manual IRQs tuning script I have and from the snapshot of the IRQs is everything looking good?
1 Like
I would guess irqbalance would spread 4tc 4rx irqs to 4 cores. So there is softirqd to process network traffic on each.
You probably should not use packet steering along with irqbalance.
2 Likes