lkml.org 
[lkml]   [2000]   [Sep]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectBottom Handles/soft irqs/timer interrupts/SMP .....
From
Date

Hi Ingo & folks,
I am heard that you are the one that have set the timer
interrupt APIC by the 2.2+ linux kernel.

The Problem:
-----------

I am getting freezes with test9-pre7. The freezes happen while all
the processes that should be running are waiting in the semaphore:

cache_drain_sem

The process that holds that semaphore is waiting for all the CPUS to
run the function slab_drain_local_cache(), this function is called
from the timer interrupt.

1st Question: It is guaranteed that the timer interrupt will go to
all CPUs _soon_???

What the code tries to do:
-------------------------

The code tries to call the slab_drain_local_cache in all the CPUs and
wait for all the CPUs to call that function, and the last one to
wakeup the first one.

Problem: The variable slab_cache_drain_mask is used "quite happily"
without too much looking on it (i.e. it is read without any atomic
operation, I have changed it to be a volatile varible, but the problem
persists).


What we need to do:
------------------

We are trying that each CPU shrink its local caches, and that should
be done by the local CPU.

2nd Question: Is there a sane way to queue an operation to be done in
each specific CPU?

Something like tq_timer[cpu]?
It is that problem better solved with
soft_irqs/<whatever other mechanism>.

Notice that we are sleeping with the cache_drain_sem, but it should be
not necessary, if there is a way to queue an operation to do in each
CPU, with that, if I call drain_cpu_caches, I can always use a bitmap
to see if I have that interrupt pending, and in that case, does
nothing, if the bit was 0, I queue the operation.

If you need any more explanation, or a pointer to a place where I can
find how to setup a function to be called by each CPU, you are
welcome. I have only found smp_call_function, but I don't know if it
is run with local_irqs deactivated, or if it is the correct answer to
this problem.

Later, Juan.





--
In theory, practice and theory are the same, but in practice they
are different -- Larry McVoy
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 12:38    [W:1.455 / U:0.952 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site