Re: arm gicv3 its command time out / patch

On 26.11.2025 5:10, Warner Losh wrote:
> Also, a minor nit in the patch:
> 
> +	/* Register this device as a interrupt controller */
> 
> should not have moved. Finding the xref doesn't register the interrupt 
> controller, but
> 
>   	sc->sc_pic = intr_pic_register(dev, xref);
> 
> does and is the real start of the process.
> 
> Anyway, I'm looking forward to learning more about this issue.
> 
> Warner
> 
> On Tue, Nov 25, 2025 at 7:33 PM Warner Losh <[email protected] 
> <mailto:[email protected]>> wrote:
> 
>     What hardware does this fix booting on? What's the setup? The patch
>     itself looks OK, but I need to know the environment it runs in to
>     understand and evaluate it fully. Thanks!
> 
>     Warner
> 
>     On Tue, Nov 25, 2025 at 3:03 PM titus <[email protected]
>     <mailto:[email protected]>> wrote:
> 
>           cubie kernel: its0: <ARM GIC Interrupt Translation Service>
>         mem 0x3440000-0x345ffff on gic0
>           cubie kernel: its0: Timeout while waiting for CMD completion.
> 
>         patch (not ignoring dt dma-noncoherent flag) (adapted from linux
>         kernel)
> 

I am preparing a fix for this. One issue is that ITS may sit on a 
non-coherent bus. Another issue is that it can sit on a 32-bit bus, so 
we must also limit the address range when allocating various 
memory-mapped buffers and tables.

Another issue is that the "dma-noncoherent" property was only added in 
Linux 6.15. Therefore, we must also use another method to identify the 
necessary quirk if we want to MFC this fix.

Mainly Rockchip SoCs are affected.

Michal