Re: Broken SCSI on SPARCclassic X

[Hamish Greig <hgreig@bigpond.net.au>]

The classic and the classic X differ by only a few lines on the prom, 
there's a HOWTO, to change the Classic X to a working Classic. Google 
for it and then everything will work. I have done this in the past and 
had debian working on a classic X with a disk.

Hamish

Markus Dahms wrote:
> Hi,
>
> Trying to install Etch on a Classic X (sun4m, 4/10) I encountered some 
> problems with the SCSI controller. After booting from the SCSI disk in 
> question (the actual installation I did in an SPARCstation 5, but it's
> the same for netboot or CD-ROM installer) the esp module oopses:
>
> [...]
> SCSI subsystem initialized                                                      
> esp0: IRQ 0 SCSI ID 7 Clk 40MHz CCYC=25000 CCF=8 TOut 167 <1>Unable to handle k0
> tsk->{mm,active_mm}->context = 0000003f                                         
> tsk->{mm,active_mm}->pgd = fc015000                                             
>               \|/ ____ \|/                                                      
>               "@'/ ,. \`@"                                                      
>               /_| \__/ |_\                                                      
>                  \__U_/                                                         
> modprobe(761): Oops [#1]                                                        
> PSR: 419000c6 PC: fe61e840 NPC: fe61e844 Y: 01c00000    Not tainted             
> PC: <esp_sbus_probe+0x5b4/0xae8 [esp]>                                          
> %G: 000000ff 0000000a  fd114000 41400fe4  f003046c 00004000  f020a000 00000000  
> %O: 0000001f fe620740  000000ff 000061a8  00000008 000000a7  f020bc10 fe61e7d8  
> RPC: <esp_sbus_probe+0x54c/0xae8 [esp]>                                         
> %L: 0000001f 000000ff  fe620400 f32e1000  f32e1800 f595e800  f32e1000 f32e1008  
> %I: fffffff4 f595ea9c  00000020 f019f190  f020a000 f0228d20  f020bc78 f001bb94  
> Caller[f001bb94]: of_device_probe+0x58/0x74                                     
> Caller[f0117068]: driver_probe_device+0x60/0xb8                                 
> Caller[f01171c8]: __driver_attach+0x70/0xc4                                     
> Caller[f0116950]: bus_for_each_dev+0x40/0x74                                    
> Caller[f01165a0]: bus_add_driver+0x6c/0x134                                     
> Caller[f004c97c]: sys_init_module+0x1610/0x1778                                 
> Caller[f0011634]: syscall_is_too_hard+0x3c/0x40                                 
> Caller[000133b4]: 0x133bc                                                       
> Instruction DUMP: 8210200a  c228a02c  c4064000 <c208a02c> 8208601f  80a0600a  0 
> Killed
>
> Being alerted by IRQ 0 and the fact that probe-scsi in OpenBoot doesn't
> work as expected I looked a bit deeper into the OpenBoot device tree and
> found what probably is a bug in the PROM. As this machine is used diskless
> in most cases it might not be noticed that the "scsi" devalias is wrong
> (at least in my version of the PROM - Rom Rev: 2.12).
>
> Note the swapped addresses (or is it something else?) for espdma and esp:
>
> ok show-devs
> [...]
> /iommu@0,10000000/sbus@0,10001000/espdma@4,8800000                              
> [...]
> /iommu@0,10000000/sbus@0,10001000/espdma@4,8800000/esp@4,8400000                
> /iommu@0,10000000/sbus@0,10001000/espdma@4,8800000/esp@4,8400000/st             
> /iommu@0,10000000/sbus@0,10001000/espdma@4,8800000/esp@4,8400000/sd             
> [...]
>
> ok devalias                                                                     
> [...]
> scsi    /iommu/sbus/espdma@4,8400000/esp@4,8800000                              
> [...]
> disk    /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@3,0                       
> cdrom   /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@6,0:d                     
> tape    /iommu/sbus/espdma@4,8400000/esp@4,8800000/st@4,0                       
> tape0   /iommu/sbus/espdma@4,8400000/esp@4,8800000/st@4,0                       
> tape1   /iommu/sbus/espdma@4,8400000/esp@4,8800000/st@5,0                       
> disk3   /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@3,0                       
> disk2   /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@2,0                       
> disk1   /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@1,0                       
> disk0   /iommu/sbus/espdma@4,8400000/esp@4,8800000/sd@0,0
>
> As the PROM is read-only one can only prepend device aliases to the
> list. Adding a correct (or not) alias for scsi the kernel halts after
> PROM initialization:
>
> ok devalias scsi /iommu/sbus/espdma@4,8800000/esp@4,8400000                     
> scsi isn't unique                                                               
>
> ok boot                                                                         
> SILO Version 1.4.13                                                             
> boot:                                                                           
> Uncompressing image...                                                          
> Loaded kernel version 2.6.18                                                    
> Loading initial ramdisk (3142737 bytes at 0x3000000 phys, 0x60000000 virt)...   
> PROMLIB: obio_ranges 1                                                          
>
> That's all. No more output... Is there any other way to correct the
> scsi alias or Linux' behaviour? The problem might be an alias which is
> not unique. Without this initial devalias command the kernel boots fine
> up to the esp oops shown above.
>
> Btw: How do I build Sparc netboot images to try booting a newer kernel?
>
> Greetings,
>
> 	Markus