Re: Tyan S2885

[YhLu]

Eric,

Great, it works. Even normal scan it can find out 8151.

But the bus number is funny, I may need to update bus number mptable.c and
irq_table.c and reset.c

I wonder if 8151 on CPU 1 and 8131/8111 on CPU0. is the amdk8_scan_root_bus
called two times? If so, you may keep the
>     for(reg = 0xe0; reg <= 0xec; reg += 4) { 
>             value = f1_read_config32(reg);
>             printk_debug("amdk8_scan_root_bus %0x =  %0x  max =

%0x\n",
> reg,value,max);
>                 if((value>>24)>max)
>                         f1_write_config32(reg, 0);
>         }

Regards

YH.

-----邮件原件-----
发件人: [e-mail] [mailto:[e-mail]] 
发送时间: 2003年12月5日 15:26
收件人: YhLu
抄送: ron minnich; Stefan Reinauer; [e-mail]
主题: Re: Tyan S2885

YhLu  writes:
> Eric,
>
> I change amdk8_scan_root_bus to:
>
> unsigned int amdk8_scan_root_bus(device_t root, unsigned int max)
> {               
>         unsigned reg;   
>         uint32_t value; 
>         /* Unmap all of the other pci busses */
>         printk_debug("\nbefore pci_scan_bus\n");
>         for(reg = 0xe0; reg <= 0xec; reg += 4) { 
>                 value = f1_read_config32(reg);
>                 printk_debug("amdk8_scan_root_bus %0x =  %0x  max =

%0x\n",
> reg,value,max);
>                 if((value>>24)>max)
>                         f1_write_config32(reg, 0);
>         }       
>         for(reg = 0xe0; reg <= 0xec; reg += 4) { 
>                 value = f1_read_config32(reg);
>                 printk_debug("amdk8_scan_root_bus %0x =  %0x  max =

%0x\n",
> reg,value,max);
>         }
>     max = pci_scan_bus(&root->link[0], PCI_DEVFN(0x18, 0), 0xff, max);
>
>     printk_debug("after pci_scan_bus\n");
>     for(reg = 0xe0; reg <= 0xec; reg += 4) {
>             value = f1_read_config32(reg);
>             printk_debug("amdk8_scan_root_bus %0x =  %0x  max =

%0x\n",
> reg,value,max);
>         }
>     return max;
>
> }
>
> I think as least you only need to clear the regs that contain bigger bus
> number than the next bus for scanning.
>
> The result will be in the end.
>
> The problems:
> It can not access 8151 on CPU link0 and so can not scan it
> The HT scan seems can not figure out and set the link and read/write
>
> enable bit in e0. It set the e0 to 0x05050000

Doh.   The Opteron is too forgiving when you only have one link.  It
makes this kind of debugging a pain.  What I am doing should not work
but it does in that case...
My apologies, for missing this.

In amdk8_scan_chains() there is the snippet:

config_busses &= 0x0000ffff;
    	config_busses |= ((dev->link[link].secondary) << 16) |
    		((dev->link[link].subordinate) << 24);
    	f1_write_config32(config_reg, config_busses);

Which is the offending piece of code.  It neither sets the link that
we are supposed to go out, nor does it set the r/w enable bits.  Ouch.
It only works if those registers are pre programmed.
So I think this should be:

config_busses &= 0x000fc88;
    	config_busses |= 
    		(3 << 0) |  /* rw enable, no device compare */
    		(( nodeid & 7) << 4) | 
    		(( link & 3 ) << 8) |  
    		((dev->link[link].secondary) << 16) |
    		((dev->link[link].subordinate) << 24);
    	f1_write_config32(config_reg, config_busses);

Eric

Tyan S2885

YhLu  writes:
> Eric,
>
> Great, it works. Even normal scan it can find out 8151.
>
> But the bus number is funny, I may need to update bus number mptable.c and
> irq_table.c and reset.c
>
> I wonder if 8151 on CPU 1 and 8131/8111 on CPU0. is the amdk8_scan_root_bus
> called two times? If so, you may keep the

amdk8_scan_root_bus is always only called once.  It just finds the cpus.

It recurses into amdk8_scan_chains which actually finds all of the busses.

So it is ok to zap everything in amdk8_scan_root_bus.  But it may
be a little more elegant to zap just what we need in amdk8_scan_root_bus.

Eric