Re: [Qemu-ppc] [PATCH RFC] openbios-devel/arch/ppc/qemu/init.c

[Alexander Graf]

On 16.04.14 18:35, BALATON Zoltan wrote:
> On Wed, 16 Apr 2014, Alexander Graf wrote:
> > > > > @@ -269,6 +269,18 @@
> > > > >       push_str("i-cache-block-size");
> > > > >       fword("property");
> > > > >
> > > > > +    if(cpu->tlb_size) {
> > > > > +        PUSH(cpu->tlb_sets);
> > > > > +        fword("encode-int");
> > > > > +        push_str("tlb-sets");
> > > > > +        fword("property");
> > > > > +
> > > > > +        PUSH(cpu->tlb_size);
> > > > > +        fword("encode-int");
> > > > > +        push_str("tlb-size");
> > > > > +        fword("property");
> > > >
> > > > Please just find the correct values and add the properties to the 
> > > > cpu nodes always.
> > >
> > > Where can I find the correct values? Maybe QEMU should send it via 
> > > FW_CFG because OpenBIOS only seems to have data for a few CPUs and 
> > > most of them have a comment saying that the values are not correct 
> > > anyway. But I could not find where is this data available in QEMU.
> >
> > That'd work for me. I also wouldn't be opposed to move OpenBIOS to a 
> > scheme similar to SLOF and u-boot where we pass the device tree to 
> > firmware and use that instead of fw_cfg to enumerate devices in our 
> > system.
>
> The problem with this is that OpenBIOS is used for a lot more 
> architectures than just mac emulation and such a change would also 
> affect others (like sparc). Also the device tree is partly created in 
> Forth and then patched in C so using a whole device tree from QEMU 
> would need modifications to the Forth part as well. So this is no 
> small task and one that's too much for me. If someone can take on this 
> task it may be done but the most I can do is adding these to FW_CFG 
> and even for that someone should tell me how to find these values in 
> QEMU.
>
> > > > > +    }
> > > > > +
> > > > >       PUSH(fw_cfg_read_i32(FW_CFG_PPC_TBFREQ));
> > > > >       fword("encode-int");
> > > > >       push_str("timebase-frequency");
> > > > > @@ -279,6 +291,11 @@
> > > > >       push_str("clock-frequency");
> > > > >       fword("property");
> > > > >
> > > > > +    PUSH(100 * 1000 * 1000);
> > > > > +    fword("encode-int");
> > > > > +    push_str("bus-frequency");
> > > > > +    fword("property");
> > > >
> > > > I'm fairly sure this is wrong.
> > >
> > > What is the good value then? (Some other data e.g. the clock 
> > > frequency is also a made up value to make the clients happy.)
> >
> > Don't we send the bus frequency via fw_cfg? I remember we had a 
> > pretty long thread about bus frequency and how it kept breaking Mac 
> > OS X.
>
> Wasn't that the clock-frequency? That is sent via FW_CFG but the 
> bus-frequency isn't and was just missing from the CPU properties.

Ah, it was the clock frequency vs tb frequency. Jeez - we have way too 
many of these.

Yes, please add it as a value in fw_cfg.


Alex