These days 4 kilobytes of stack just aren't enough for reliably operation,
and people using lots of threads have long switched to x86-64, so remove the
CONFIG_4KSTACKS option.
Signed-off-by: Christoph Hellwig <[email protected]>
Index: linux-2.6/arch/x86/Kconfig.debug
===================================================================
--- linux-2.6.orig/arch/x86/Kconfig.debug 2010-06-16 10:24:06.902272433 +0200
+++ linux-2.6/arch/x86/Kconfig.debug 2010-06-16 10:24:11.837068354 +0200
@@ -121,15 +121,6 @@ config DEBUG_NX_TEST
and the software setup of this feature.
If in doubt, say "N"
-config 4KSTACKS
- bool "Use 4Kb for kernel stacks instead of 8Kb"
- depends on X86_32
- ---help---
- If you say Y here the kernel will use a 4Kb stacksize for the
- kernel stack attached to each process/thread. This facilitates
- running more threads on a system and also reduces the pressure
- on the VM subsystem for higher order allocations.
-
config DOUBLEFAULT
default y
bool "Enable doublefault exception handler" if EMBEDDED
Index: linux-2.6/arch/x86/include/asm/module.h
===================================================================
--- linux-2.6.orig/arch/x86/include/asm/module.h 2010-06-16 10:24:06.922004449 +0200
+++ linux-2.6/arch/x86/include/asm/module.h 2010-06-16 10:24:23.255291640 +0200
@@ -60,12 +60,7 @@
#endif
#ifdef CONFIG_X86_32
-# ifdef CONFIG_4KSTACKS
-# define MODULE_STACKSIZE "4KSTACKS "
-# else
-# define MODULE_STACKSIZE ""
-# endif
-# define MODULE_ARCH_VERMAGIC MODULE_PROC_FAMILY MODULE_STACKSIZE
+# define MODULE_ARCH_VERMAGIC MODULE_PROC_FAMILY
#endif
#endif /* _ASM_X86_MODULE_H */
Index: linux-2.6/arch/x86/include/asm/page_32_types.h
===================================================================
--- linux-2.6.orig/arch/x86/include/asm/page_32_types.h 2010-06-16 10:24:06.933273132 +0200
+++ linux-2.6/arch/x86/include/asm/page_32_types.h 2010-06-16 10:24:30.251005496 +0200
@@ -15,11 +15,7 @@
*/
#define __PAGE_OFFSET _AC(CONFIG_PAGE_OFFSET, UL)
-#ifdef CONFIG_4KSTACKS
-#define THREAD_ORDER 0
-#else
#define THREAD_ORDER 1
-#endif
#define THREAD_SIZE (PAGE_SIZE << THREAD_ORDER)
#define STACKFAULT_STACK 0
On Wed, 16 Jun 2010 11:16:21 +0200
Christoph Hellwig <[email protected]> wrote:
> These days 4 kilobytes of stack just aren't enough for reliably operation,
> and people using lots of threads have long switched to x86-64, so remove the
> CONFIG_4KSTACKS option.
It's very useful for things like debugging and because of the split
kernel/irq stacks. It also seems to be quite adequate for most
configurations.
Perhaps XFS should select 8K stack instead ?
On Wed, Jun 16, 2010 at 11:21:36AM +0100, Alan Cox wrote:
> On Wed, 16 Jun 2010 11:16:21 +0200
> Christoph Hellwig <[email protected]> wrote:
>
> > These days 4 kilobytes of stack just aren't enough for reliably operation,
> > and people using lots of threads have long switched to x86-64, so remove the
> > CONFIG_4KSTACKS option.
>
> It's very useful for things like debugging and because of the split
> kernel/irq stacks.
Ingo just ACKed my previous patch to make irq stacks unconditional.
> It also seems to be quite adequate for most
> configurations.
See the direct reclaim flamewar. Currently we can end up in ->writepage
from af_unix poll code with over 3k stack alreay used. It's certainly
not anywhere near enough.
> Perhaps XFS should select 8K stack instead ?
or rather !CONFIG_INSANE. I don't care too much. No one sane should
ever enable it anymore, but Ingo asked me to send a patch to remove it.
On Wed, Jun 16, 2010 at 12:20:51PM +0200, Christoph Hellwig wrote:
> > Perhaps XFS should select 8K stack instead ?
>
> or rather !CONFIG_INSANE. I don't care too much. No one sane should
> ever enable it anymore, but Ingo asked me to send a patch to remove it.
>
Alan--
It's not just XFS, we were regularly seeing blown stacks on Fedora with
ext3/4 and encrypted disks using dm-crypt...
--Kyle
On 06/16/2010 12:23 PM, Kyle McMartin wrote:
> On Wed, Jun 16, 2010 at 12:20:51PM +0200, Christoph Hellwig wrote:
>>> Perhaps XFS should select 8K stack instead ?
>>
>> or rather !CONFIG_INSANE. I don't care too much. No one sane should
>> ever enable it anymore, but Ingo asked me to send a patch to remove it.
>>
>
> Alan--
>
> It's not just XFS, we were regularly seeing blown stacks on Fedora with
> ext3/4 and encrypted disks using dm-crypt...
Can you point me to report where 4k stacks is not enough for plain dm-crypt?
IIRC it was always XFS (maybe ext4) on top of it what caused problem...
(Anyway, removing 4k stacks is probably good thing:-)
Milan
On Wed, Jun 16, 2010 at 12:46:55PM +0200, Milan Broz wrote:
> Can you point me to report where 4k stacks is not enough for plain dm-crypt?
> IIRC it was always XFS (maybe ext4) on top of it what caused problem...
>
> (Anyway, removing 4k stacks is probably good thing:-)
>
LVM may have been involved as well, I can't recall. I'll try to dig up
the report from my email.
--Kyle