Re: UltraSPARC III/IIe cpufreq drivers, please test

Martin Husemann a écrit :
> On Wed, Oct 01, 2014 at 03:42:51PM +0200, BERTRAND Joël wrote:
>> Martin Husemann a écrit :
>>> On Mon, Sep 29, 2014 at 02:19:00PM +1000, matthew green wrote:
>>>> this is some what tricky.  ideally, it should be derived from
>>>> the cpu device, and that should find and talk to the schizo
>>>> device that controls it.
>>>
>>> Since mainbus device detection is mostly hardcoded, can't we rely on all
>>> cpus to attach their autoconfig device node before any pci root hub is
>>> attached?
>>
>> 	Why not. But I don't know autoconf process enough to write it. Do
>> 	you have some ideas to start ?
>
> Ok, looking at it again in more detail, I think we should do it differently.
>
> Unfortunately this looks like it would need copy&pasting to all pci root
> implementations (ideas to avoid that welcome).
>
> Let's for example take schizo_attach:
>
>   - go up one node (OF_parent(ma->ma_node))
>   - basically clone the loop used to get the number of cpus from the
>     firmware (get_ncpus) and just count, which index (from OF's point
>     of view) each gets
>   - within the same loop, count the PCI's, for each read the first word
>     from the "reg" property to find out wether it is bus A or B (same
>     condition we use in the attach code to set sp_bus_a). If bus is A,
>     count the node
>
> Now we should get the same node index for cpu and schizo, so if the
> schizo node is the same node as in our attach args, we have found our
> cpu and can break out of the loop.
>
> Once we know the node of the CPU we are looking for, we can do a loop over
> all CPUs and compare it against ci->ci_node.
>
> This should work completely independent of all kernel config contents.

	Hello,

	I'm trying to write you suggestion in schizo_attach() but it doesn't 
work as expected. I suppose I haven't understand what you suggested.

	But I don't understand why we need to write this piece of code. If I 
understand, cpufreq is only registered once with only once struct 
cpu_info as all CPU has to share same parameters. Thus, I can take the 
first elements of *cpus (defined in cpu.c).

	Last question : schizo_set_cpufreq() is called by xcall(). Thus, if I 
understand, schizo_set_cpufreq() should control cpufreq of CPU that runs 
this function.

	In dmesg, I only obtain :
...
Feb 25 22:43:14 hilbert /netbsd: register safari cpufreq driver
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 375 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 750 MHz (cpu 0)
Feb 25 22:43:14 hilbert /netbsd: schizo_set_cpufreq 23 MHz (cpu 0)
...
Feb 25 22:43:14 hilbert /netbsd: cd0 at scsibus0 target 6 lun 0: 
<TOSHIBA, DVD-ROM SD-M1401, 1009> cdrom removable
Feb 25 22:43:14 hilbert /netbsd: cd0: sync (50.00ns offset 16), 8-bit 
(20.000MB/s) transfers
Feb 25 22:43:14 hilbert /netbsd: sd0 at scsibus5 target 0 lun 0: 
<SEAGATE, ST3146807FC, 0006> disk fixed
Feb 25 22:43:14 hilbert /netbsd: sd0: 136 GB, 49855 cyl, 8 head, 718 
sec, 512 bytes/sect x 286749488 sectors
Feb 25 22:43:14 hilbert /netbsd: sd1 at scsibus5 target 1 lun 0: 
<SEAGATE, ST3146807FC, 0006> disk fixed
Feb 25 22:43:14 hilbert /netbsd: sd1: 136 GB, 49855 cyl, 8 head, 718 
sec, 512 bytes/sect x 286749488 sectors

	Output are written inside schizo_set_cpufreq() and cpu number is value 
returned by cpu_number().

	I don't understand why only CPU0 frequency is controlled by cpufreq and 
why no cpufreq debug line is written in log or console after "cd0 at 
scsibus0 target 6 lun 0:".

	I will send as soon as possible a patch.

	Regards,

	JKB