[okl4-developer] Problem...

[Peter Nguyen]

Hi,

I have come across an implementation problem. More specifically, I  
have added some sampling facilities for OKL4 whereby samples are taken  
of the system and temporarily stored in the kernel. These samples are  
copied out to user space via a interrupt. The buffer for storing  
samples is derived from the trace buffer functionality, with the  
interrupt handler implemented in user-space, like the tracebuffer  
interrupt handler.

The problem that I have is the following: my sampling environment  
consists of a series of collection and processing threads, where there  
are buffers for each collection thread. Processing threads are able to  
read the data from these buffers to perform the required processing.  
In terms of execution, the processing threads should block when there  
is currently no data available for processing (which would apply when  
profiling begins). The interrupt handler copies the samples to the  
relevant buffers, and sends IPCs to the processing threads when the  
batch of samples have been completely copied. The IPCs to the  
processing threads allow for resumption of sample processing. However,  
every time I try to execute the profiling, i keep getting a "kernel  
access raised user pagefault" KDB message. Upon further investigation,  
there's something within the virtual interrupt that causes that  
message to occur. My profiling was tested while the iguana naming  
server was used for execution.

I'm not at all sure why it would be doing this. In the case where the  
processing threads don't block, the profiling actually executes to  
completion, although the results will be wrong because the samples  
will be attributed to the processing threads themselves. Furthermore,  
in testing the profiling against a simple iguana app (where this app  
consists of two threads that execute for a long time) with the  
blocking measures in the processing threads, the profiling also  
completes, with the samples attributed to the two threads in the  
simple iguana app.

In terms of my execution environment, I'm using an ia32 build of OKL4  
version 1.4.1.1, and have used a patch for the virtual interrupt from  
this link: http://wiki.ok-labs.com/forum#nabble-f9692

Any help on this would be greatly appreciated as it has been bugging  
me (no pun intended).

-----------------------------------------------------
Peter Nguyen
Jacaranda Research Group