lkml.org 
[lkml]   [2001]   [Sep]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: Deadlock on the mm->mmap_sem
Linus Torvalds wrote:
>
> If the choice is between a hack to do strange and incomprehensible things
> for a special case, and just making the semaphores do the same thing
> rw-spinlocks do and make the problem go away naturally, Ill take #2 any
> day. The patches already exist, after all.
>

I've attached a recursive semaphore patch against 2.4.10-pre11 - but it
makes mmap io unusable:

Testcase:
* file, mmaped, 300 MB, 128 MB Ram
* 2 threads: touch random pages
* third thread: calls mprotect(0xFFFF00000,0x1000, PAGE_READ)

Result:
mprotect hangs forever (minutes) with recursive semaphores.

With fair semaphores, mprotect returns after ~80 milliseconds with 5
worker threads, after ~380 milliseconds with 20 worker threads (slow IDE
disk).

One alternative to David's patch would be moving the locking into the
coredump handlers - would you prefer that?

--
Manfred--- 2.4/lib/rwsem-spinlock.c Sat Apr 28 10:37:27 2001
+++ build-2.4/lib/rwsem-spinlock.c Wed Sep 19 15:03:28 2001
@@ -115,7 +115,7 @@

spin_lock(&sem->wait_lock);

- if (sem->activity>=0 && list_empty(&sem->wait_list)) {
+ if (sem->activity>=0) {
/* granted */
sem->activity++;
spin_unlock(&sem->wait_lock);
--- 2.4/arch/i386/config.in Wed Sep 19 14:36:35 2001
+++ build-2.4/arch/i386/config.in Wed Sep 19 14:48:06 2001
@@ -59,8 +59,8 @@
define_bool CONFIG_X86_XADD y
define_bool CONFIG_X86_BSWAP y
define_bool CONFIG_X86_POPAD_OK y
- define_bool CONFIG_RWSEM_GENERIC_SPINLOCK n
- define_bool CONFIG_RWSEM_XCHGADD_ALGORITHM y
+ define_bool CONFIG_RWSEM_GENERIC_SPINLOCK y
+ define_bool CONFIG_RWSEM_XCHGADD_ALGORITHM n
fi
if [ "$CONFIG_M486" = "y" ]; then
define_int CONFIG_X86_L1_CACHE_SHIFT 4
\
 
 \ /
  Last update: 2005-03-22 13:03    [W:0.100 / U:0.244 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site