Re: bug: audiohda broken in qemu

[Anthony Martin <ality@pbrane.org>]

Nick Owens <mischief@offblast.org> once said:
> i looked a bit at this 'rirb count reached' message and surmised that
> we needed to change RINTCNT. i looked at the linux driver and they set
> it to 1, and to 0xC0 for some 'quirky' cards. i tried 1 but that just
> makes subsequent commands after Getparam/Vendorid fail, but setting it
> to 0xC0 made all commands work and the device is enumerated, and
> #A/audio works.
>
> diff --git a/sys/src/9/pc/audiohda.c b/sys/src/9/pc/audiohda.c
> --- a/sys/src/9/pc/audiohda.c
> +++ b/sys/src/9/pc/audiohda.c
> @@ -1732,6 +1732,7 @@ hdastart(Ctlr *ctlr)
>      csr32(ctlr, Rirblbase) = pa;
>      csr32(ctlr, Rirbubase) = pa >> 32;
>      csr16(ctlr, Rirbwp) = Rirbptrrst;
> +    csr16(ctlr, Rintcnt) = 0xC0;
>      csr8(ctlr, Rirbctl) = Rirbdma;
>      waitup8(ctlr, Rirbctl, Rirbdma, Rirbdma);
>
> note that i have no idea what i am doing.

That's weird. There are two ways to figure out if there are responses
in the response input ring buffer (RIRB): either enable interrupts or
poll the write pointer. We do the latter and set the Rirbint field to
zero in the Rirbctl register. The value of Rintcnt should not matter
at all if RIRB interrupts are disabled.

Which method are they using in Linux?

  Anthony