Should I use memory barriers with wait_event ?

[John <jhnlmn@xxxxxxxxx>]

Hi,
In a kernel module I have a variable A, 
which is shared between processes and used
as a counter to protect access to other
variables, such as B.

All accesses to A are properly
synchronized with spin_lock_irqsave and 
spin_unlock_irqrestore, for example:

spin_lock_irqsave
if(A > 0)
{
  update B;
  A--;
}
spin_unlock_irqrestore


But in one place I also need to wait for A
to reach value 0 before reading B.

I was advised to use wait_event:
wait_event(wait_queue, A == 0);
Read and use B;

But I see that wait_event(wq, condition) is defined
as:
do {
  if (condition)
    break;
  __wait_event(wq, condition);
} while (0)

Now I started to worry because I see that
wait_event executes (A == 0) comparison without
any protection by lock and without any barrier.

So, I imagine that on SMP system with relaxed
operation ordering the following may happen:
CPU 1        CPU 2
             LOAD B
STORE B
STORE A = 0
             LOAD A->0

Then CPU 2 will call wait_event, find that A == 0
and proceed using the old value of B.

So, my question is whether my analysis is correct
and what is the right way to use wait_event
in this situation?

Thank you
John


             







 
____________________________________________________________________________________
Don't get soaked.  Take a quick peek at the forecast
with the Yahoo! Search weather shortcut.
http://tools.search.yahoo.com/shortcuts/#loc_weather

--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ