Re: Stability problems on current running on acorn32

[christos%astron.com@localhost (Christos Zoulas)]

In article <4ED290B9.2080701%black-star.demon.co.uk@localhost>,
Mike Pumford  <mpumford%black-star.demon.co.uk@localhost> wrote:
>Mike Pumford wrote:
>Forgot a couple of things here:
>
>> uvm_fault(0xf27e36a0, deadd000, 1) -> e
>> Fatal kernel mode data abort: 'Tranlation Fault (S)'
>> trapframe: 0xf2c4bd58
>> FSR=18529005,FAR=deaddec5 SPSR=a0000013
>>
>> r0 =f037e128, r1 =0901a8c0, r2 =deaddead, r3 =0101a8c0
>> r4 =00003500, r5 =f1179658, r6 =0000faff, r7 =00000000
>> r8 =8150020a, r9 =c0a80109, r10=f1179658, r11=f2c4bde0
>> r12=00000000, ssp=f2c4bda4, slr=f02448a0, pc =f00be848
>>
>
>> in_pcblookup_connect+??
>This was actually pcblookup_connect+0xd4
>Which disassembles to:  ldr     r5, [r2, #24]
>
>If I'm reading the pool code correctly this means that the inpcb code 
>was referencing a freed pool entry! I've seen other DIAGNOSTIC traces 
>with the same failure. It also matches behaviour where I've seen an SSH 
>session forced closed when stopping a listening TCP server on the 
>machine from another SSH session.
>
>> I've attached the patches I've made to the MD code just in case I've
>> done something stupid in one of those changes. I've also attached a
>> dmesg taken before the system panics to give full details of the
>> hardware in the system.
>>
>And a description of the patches might be useful:
>
>vidcvideo.patch: Use the RI_NO_AUTO flag when bringing up the console. 
>Stops the system hanging before even starting to boot. Modelled on a 
>similar change in igsfb driver.
>
>iomd_clock.patch: Use the IOMD timer to provide a more accurate delay() 
>implementation. Previous implementation was becoming increasingly 
>optimised away by newer GCC versions to the point that the keyboard 
>driver no longer worked correctly. I'm not sure whether the SPL based 
>locking is better than the disable interrupt approach or if I should be 
>doing something else entirely here.

If the bus space functions don't work properly, then they should be fixed
as opposed to open-coding them.

christos