[PATCH v11 4/4] stmmac: s32: enable support for Multi-IRQ mode

[Jared Kangas]

Hi Jan,

On Thu, Mar 12, 2026 at 09:55:30AM +0100, Jan Petrous via B4 Relay wrote:
> From: "Jan Petrous (OSS)" <jan.petrous at oss.nxp.com>
> 
> Based on previous changes in platform driver, the vendor
> glue driver can enable Multi-IRQ mode, if needed.
> 
> [...]
> 
> If those prerequisites are met, the driver switches to Multi-IRQ mode,
> using per-queue IRQs for rx/tx data pathr:
> 
> [    1.387045] s32-dwmac 4033c000.ethernet: Multi-IRQ mode (per queue IRQs) selected
> 
> Now the driver owns all queues IRQs:
> 
> root at s32g399aevb3:~# grep eth /proc/interrupts
>  29:    0    0    0    0    0    0    0    0    GICv3  89 Level   eth0:mac
>  30:    0    0    0    0    0    0    0    0    GICv3  91 Level   eth0:rx-0
>  31:    0    0    0    0    0    0    0    0    GICv3  93 Level   eth0:rx-1
>  32:    0    0    0    0    0    0    0    0    GICv3  95 Level   eth0:rx-2
>  33:    0    0    0    0    0    0    0    0    GICv3  97 Level   eth0:rx-3
>  34:    0    0    0    0    0    0    0    0    GICv3  99 Level   eth0:rx-4
>  35:    0    0    0    0    0    0    0    0    GICv3  90 Level   eth0:tx-0
>  36:    0    0    0    0    0    0    0    0    GICv3  92 Level   eth0:tx-1
>  37:    0    0    0    0    0    0    0    0    GICv3  94 Level   eth0:tx-2
>  38:    0    0    0    0    0    0    0    0    GICv3  96 Level   eth0:tx-3
>  39:    0    0    0    0    0    0    0    0    GICv3  98 Level   eth0:tx-4

I ran this series' changes on an NXP S32G-VNP-RDB3 (dwmac-s32) and
confirmed multichannel TX by doing a basic iperf3 throughput test:

    # dmesg | grep Multi-IRQ
    [   37.463467] s32-dwmac 4033c000.ethernet: Multi-IRQ mode (per queue IRQs) selected
    # iperf3 -s
    [connection logs snipped]
    # grep end0 /proc/interrupts | column -t
    29:  0      0  0  0  0  0  0  0  GICv3  89  Level  end0:mac
    30:  968    0  0  0  0  0  0  0  GICv3  90  Level  end0:tx-0
    31:  0      3  0  0  0  0  0  0  GICv3  92  Level  end0:tx-1
    32:  0      0  3  0  0  0  0  0  GICv3  94  Level  end0:tx-2
    33:  0      0  0  3  0  0  0  0  GICv3  96  Level  end0:tx-3
    34:  0      0  0  0  3  0  0  0  GICv3  98  Level  end0:tx-4
    35:  67302  0  0  0  0  0  0  0  GICv3  91  Level  end0:rx-0
    36:  0      0  0  0  0  0  0  0  GICv3  93  Level  end0:rx-1
    37:  0      0  0  0  0  0  0  0  GICv3  95  Level  end0:rx-2
    38:  0      0  0  0  0  0  0  0  GICv3  97  Level  end0:rx-3
    39:  0      0  0  0  0  0  0  0  GICv3  99  Level  end0:rx-4

Also tried out multichannel RX by adding 'snps,route-multi-broad' to
rx-queues-config/queue2 in the devicetree, which showed activity on
the corresponding rx-2 entry:

    # grep end0 /proc/interrupts | column -t
    29:  0   0  0   0  0  0  0  0  GICv3  89  Level  end0:mac
    30:  4   0  0   0  0  0  0  0  GICv3  90  Level  end0:tx-0
    31:  0   1  0   0  0  0  0  0  GICv3  92  Level  end0:tx-1
    32:  0   0  1   0  0  0  0  0  GICv3  94  Level  end0:tx-2
    33:  0   0  0   0  0  0  0  0  GICv3  96  Level  end0:tx-3
    34:  0   0  0   0  1  0  0  0  GICv3  98  Level  end0:tx-4
    35:  68  0  0   0  0  0  0  0  GICv3  91  Level  end0:rx-0
    36:  0   0  0   0  0  0  0  0  GICv3  93  Level  end0:rx-1
    37:  0   0  91  0  0  0  0  0  GICv3  95  Level  end0:rx-2
    38:  0   0  0   0  0  0  0  0  GICv3  97  Level  end0:rx-3
    39:  0   0  0   0  0  0  0  0  GICv3  99  Level  end0:rx-4

I didn't see any regressions with light network usage, and both TX/RX
appear to function as expected.

Tested-by: Jared Kangas <jkangas at redhat.com>