Getting Bullseye/Bookworm kernel running on Loongson 3A boards

[Aurelien Jarno]

Dear MIPS porters,

DSA would like to upgrade the remaining mips*el buildds to bullseye (and
later to bookworm), however we have trouble running the Bullseye or
Bookworm kernel on the LS3A-RS780-1w ones.

Both Bullseye and Bookworm kernels boot fine up to the userland, but the
network card does not work. It *seems* to an interrupt issue as the
corresponding counter in /proc/interrupts does not change, and in
addition the kernel logs show the following warnings:

[ 5.077846] pcieport 0000:00:02.0: of_irq_parse_pci: failed with rc=-22
[ 5.084744] pcieport 0000:00:03.0: of_irq_parse_pci: failed with rc=-22
[ 5.170772] ahci 0000:00:11.0: of_irq_parse_pci: failed with rc=-22
[ 7.376554] pci 0000:00:07.0: of_irq_parse_pci: failed with rc=-22
[ 7.444769] ata_generic 0000:00:14.1: of_irq_parse_pci: failed with rc=-22
[ 7.495993] ehci-pci 0000:00:12.2: of_irq_parse_pci: failed with rc=-22
[ 7.624868] ehci-pci 0000:00:13.2: of_irq_parse_pci: failed with rc=-22
[ 7.753230] ohci-pci 0000:00:12.0: of_irq_parse_pci: failed with rc=-22
[ 7.886362] ohci-pci 0000:00:12.1: of_irq_parse_pci: failed with rc=-22
[ 8.018045] ohci-pci 0000:00:13.0: of_irq_parse_pci: failed with rc=-22
[ 8.150168] ohci-pci 0000:00:13.1: of_irq_parse_pci: failed with rc=-22
[ 8.294391] ohci-pci 0000:00:14.5: of_irq_parse_pci: failed with rc=-22
[ 18.398222] snd_hda_intel 0000:00:14.2: of_irq_parse_pci: failed with rc=-22
[ 18.547932] pci 0000:00:01.0: of_irq_parse_pci: failed with rc=-22

Any idea how to solve this issue? I have attached a full 6.1 kernel boot
log for reference.

Thanks,
Aurelien

--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aure...@aurel32.net http://aurel32.net

[Aurelien Jarno]

Any news about that?

[Adrian Bunk]

I am not a MIPS porter, but what about just reriring them?

These are 1/3 of our currently running MIPS buildds but they are only
~ 10% of the computing power, and being the slowest and buggiest hardware
of the current buildds they are also causing many build failures not
present on other buildds.

cu
Adrian

[Aurelien Jarno]

That's indeed an option, and we'll definitely do that if there is no way
to upgrade them. However it's also means almost loosing redundancy, as
the only buildd not hosted in OSUOSL would be mipsel-aql-03.

Note that we also have (different) issues upgrading the kernel on the
Octeon ones.

[Thomas Bogendoerfer]

asking on linux...@vger.kernel.org might get more feedback on the issue...
>From the quick look at kernel code and device tree of the board IMHO
the DTS is missing interrupt mapping entries for the PCI bus.

Could someone from Loongson have a look at the issue ?

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]

[YunQiang Su]

it seems about loongson 3a1000. i guess it is time for us to replace them with newer hardware. 

i get some loongson 3a3000 and cavium nic.

i will send them to replace these 1000 ones.

Aurelien Jarno <aur...@debian.org> 于 2023年7月24日周一 00:17写道：

[Jiaxun Yang]

Hi,

This error meesage is expected because we are not using devicetree to parse
interrupt, it is falling back to legacy PCI interrupt on i8259.

```

static int loongson_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
{
    int irq;
    u8 val;

    irq = of_irq_parse_and_map_pci(dev, slot, pin);
    if (irq > 0)
        return irq;

    /* Care i8259 legacy systems */
    pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &val);
    /* i8259 only have 15 IRQs */
    if (val > 15)
        return 0;

    return val;
}
```

Is there any other error message regarding NIC in dmesg?

Thanks
- Jiaxun

>
> Thomas.
>