Unknown error codes

[Chris Wagner]

In article <1378 at mira.cs.nps.navy.mil>, pratt at mira.cs.nps.navy.mil
(david pratt) writes:
> Good Morning to all!
> 
> I hope you had a better weekend then I did.  My code (4 threads on an 
> Iris 4D/120 GTX running 3.3) gives me the following errors:
> 
> Bus Error
> MP Status Register: 0x580:id 0 FP MP Bus Error:IOEACK
> MP Bus Error Register 0 (Read Address): 0x17300004
> MP Bus Error Register 1: 0xA4FEB0
>         <Write Back Address 0xA4XXXX WRITE PIO Requestor:CPU id=0>
> Bus Error
> MP Status Register: 0x580:id 0 FP MP Bus Error:IOEACK
> MP Bus Error Register 0 (Read Address): 0x17302608
> MP Bus Error Register 1: 0x1FBFEB0
>         <Write Back Address 0x1FBXXXX WRITE PIO Requestor:CPU id=0>
> Bus Error
> MP Status Register: 0x580:id 0 FP MP Bus Error:IOEACK
> MP Bus Error Register 0 (Read Address): 0x17300004
> MP Bus Error Register 1: 0x78FEB0
>         <Write Back Address 0x78XXXX WRITE PIO Requestor:CPU id=0>
> 
> 

No, there is no TFM here - this is alas a bug in 3.3 to do with large
shared address processes - the exact conditions under which this ocurrs
is diffucult to explain - but a fix is in an upcoming maintenance release -
please contact your local rep for details.

Basically if a large ( > 7 distinct address spaces of < 2MB each) process
shrinks its space - by either sbrk(-xx) or be detaching say a mapped file
this problem can result.
In this example each part of your program - text, data, stack, graphics pipe,
new_server shd mem, mapped files, etc. count as a 'distinct space'.
SO for a 4 thread process linking with shared gl and shared libc:

text, data, stack1, stack2, stack3. stack4, gltext, gldada, libctest, libcdata
grpipe, gr shdmem - will be your spaces (clearly > 7) As long as you
don;t shrink any of them, I believe you will not get into any problems.

Note: there is NOTHING wrong with your hardware, etc. it is completely
a software generated problem..


Chris Wagner