problem with ending requests asynchronously in my block device driver

View: New views
1 Messages — Rating Filter:   Alert me  

problem with ending requests asynchronously in my block device driver

by cable_plug :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

Hi,

I've a block device driver which does the following,

Inside the request function I do something like this:
request(fn) {

 while ((req = elv_next_request(q)) != NULL) {
....................set up the request;
 spin_unlock_irq(q->queue_lock);
call the transfer(set_up_req) function;
spin_lock_irq(q->queue_lock);
}
       spin_unlock_irq (q->queue_lock);
        /* allow callback to execute as it needs the lock!!! */
        spin_lock_irq (q->queue_lock);


}
and the transfer function calls the scsi_execute_asyn(....) with the callback function doing the end request. So, the ending of the request is done like below:

callback(fn) {

 spin_lock_irqsave(q->queue_lock, flags);
        if (!end_that_request_first(set_up_req->req, cmpstatus,
                        set_up_req->req->nr_sectors)) {
                add_disk_randomness(...);
                end_that_request_last(set_up_req->req,0);
        }
        spin_unlock_irqrestore(q->queue_lock, flags);
}


This code works fine with most of the kernel versions, but fails on some like , Linux 2.6.18-8.el5-xen

Please help me to find out where I'm going wrong?

when I say 'fails' it just hangs without any error ....I'm using dt(Data test) to write to the disk. The logs show that all the requests that have been sent for processing, have completed sucessfully. Its just that new requests never enter the request function. So, the dt writes almost half the data and then simply hangs.

 
  The actual code does nothing but, call the scsi_execute_async , which
 later on calls the callback function which is used to end the request.
 So, both the callback function and the request function need to share
 the queue lock. So the code samples I sent cover all the aspects of my
 program. What I'm looking for here is how can I end the requests
 asynchronously much later after the request processing is done. Please note
 here that the asynchronous end requests is done by the callback function
 of the scsi_execute_async, which needs to share the queue lock with
 the request function.
   
  when I say 'fails' it just hangs without any error ....I'm using
 dt(Data
test) to write to the disk. The logs show that all the requests that
 have
been sent for processing, have completed successfully. Its just that new
requests never enter the request function. So, the dt writes almost
 half the
data and then simply hangs.


Thanks in advance for an early reply.
Anil P.
LightInTheBox - Buy quality products at wholesale price