On Mon, 2009-06-22 at 14:10 +0100, Catalin Marinas wrote:
> On Mon, 2009-06-22 at 14:49 +0200, Ingo Molnar wrote:
> > * Vegard Nossum <[email protected]> wrote:
> > > Seems to be CONFIG_DEBUG_SLAB=y that is the culprit in this case. Hm,
> > > is Kconfig busted?
> > >
> > > lib/Kconfig.debug:301:config DEBUG_SLAB
> > > lib/Kconfig.debug-302- bool "Debug slab memory allocations"
> > > lib/Kconfig.debug:303: depends on DEBUG_KERNEL && SLAB && !KMEMCHECK
> > >
> > > fujita-config:1475:CONFIG_DEBUG_SLAB=y
> > > fujita-config:1558:CONFIG_KMEMCHECK=y
> > >
> > > ...what gives? Pekka?
> >
> > Kmemleak introduced this piece of not so nice solution recently:
> >
> > +config DEBUG_KMEMLEAK
> > + bool "Kernel memory leak detector"
> > + depends on DEBUG_KERNEL && EXPERIMENTAL && (X86 || ARM) && \
> > + !MEMORY_HOTPLUG
> > + select DEBUG_SLAB if SLAB
> > + select SLUB_DEBUG if SLUB
> > + select DEBUG_FS if SYSFS
> > + select STACKTRACE if STACKTRACE_SUPPORT
> > + select KALLSYMS
> >
> > that should be a depends line, not a select line.
>
> Kmemleak doesn't strictly need DEBUG_SLAB to make it a dependency. But
> enabling it may reduce (in theory) the false negatives by poisoning the
> allocated objects (and hence clearing any possible pointers to other
> objects). But I don't have any figures to show this is the case. I'll
> post a patch to drop those selects.
>
> BTW, wouldn't it be feasible for kbuild to ignore the select statements
> if the selected config has unmet dependencies?
Hmm, no idea, lets cc some relevant people here. But can we remove the
select and add a config option help text to kmemleak as a short-term
solution?
Pekka
On Mon, 2009-06-22 at 16:22 +0300, Pekka Enberg wrote:
> On Mon, 2009-06-22 at 14:10 +0100, Catalin Marinas wrote:
> > On Mon, 2009-06-22 at 14:49 +0200, Ingo Molnar wrote:
> > > * Vegard Nossum <[email protected]> wrote:
> > > > Seems to be CONFIG_DEBUG_SLAB=y that is the culprit in this case. Hm,
> > > > is Kconfig busted?
> > > >
> > > > lib/Kconfig.debug:301:config DEBUG_SLAB
> > > > lib/Kconfig.debug-302- bool "Debug slab memory allocations"
> > > > lib/Kconfig.debug:303: depends on DEBUG_KERNEL && SLAB && !KMEMCHECK
> > > >
> > > > fujita-config:1475:CONFIG_DEBUG_SLAB=y
> > > > fujita-config:1558:CONFIG_KMEMCHECK=y
> > > >
> > > > ...what gives? Pekka?
> > >
> > > Kmemleak introduced this piece of not so nice solution recently:
> > >
> > > +config DEBUG_KMEMLEAK
> > > + bool "Kernel memory leak detector"
> > > + depends on DEBUG_KERNEL && EXPERIMENTAL && (X86 || ARM) && \
> > > + !MEMORY_HOTPLUG
> > > + select DEBUG_SLAB if SLAB
> > > + select SLUB_DEBUG if SLUB
> > > + select DEBUG_FS if SYSFS
> > > + select STACKTRACE if STACKTRACE_SUPPORT
> > > + select KALLSYMS
> > >
> > > that should be a depends line, not a select line.
> >
> > Kmemleak doesn't strictly need DEBUG_SLAB to make it a dependency. But
> > enabling it may reduce (in theory) the false negatives by poisoning the
> > allocated objects (and hence clearing any possible pointers to other
> > objects). But I don't have any figures to show this is the case. I'll
> > post a patch to drop those selects.
> >
> > BTW, wouldn't it be feasible for kbuild to ignore the select statements
> > if the selected config has unmet dependencies?
>
> Hmm, no idea, lets cc some relevant people here. But can we remove the
> select and add a config option help text to kmemleak as a short-term
> solution?
Here it is:
kmemleak: Do not force the slab debugging Kconfig options
From: Catalin Marinas <[email protected]>
Selecting DEBUG_SLAB or SLUB_DEBUG by the KMEMLEAK menu entry may cause
issues with other dependencies (KMEMCHECK). These configuration options
aren't strictly needed by kmemleak but they may increase the chances of
finding leaks. This patch also updates the KMEMLEAK config entry help
text.
Signed-off-by: Catalin Marinas <[email protected]>
---
lib/Kconfig.debug | 5 +++--
1 files changed, 3 insertions(+), 2 deletions(-)
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 23067ab..4c32b1a 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -340,8 +340,6 @@ config DEBUG_KMEMLEAK
bool "Kernel memory leak detector"
depends on DEBUG_KERNEL && EXPERIMENTAL && (X86 || ARM) && \
!MEMORY_HOTPLUG
- select DEBUG_SLAB if SLAB
- select SLUB_DEBUG if SLUB
select DEBUG_FS if SYSFS
select STACKTRACE if STACKTRACE_SUPPORT
select KALLSYMS
@@ -355,6 +353,9 @@ config DEBUG_KMEMLEAK
allocations. See Documentation/kmemleak.txt for more
details.
+ Enabling DEBUG_SLAB or SLUB_DEBUG may increase the chances
+ of finding leaks due to the slab objects poisoning.
+
In order to access the kmemleak file, debugfs needs to be
mounted (usually at /sys/kernel/debug).
--
Catalin
Catalin Marinas wrote:
> kmemleak: Do not force the slab debugging Kconfig options
>
> From: Catalin Marinas <[email protected]>
>
> Selecting DEBUG_SLAB or SLUB_DEBUG by the KMEMLEAK menu entry may cause
> issues with other dependencies (KMEMCHECK). These configuration options
> aren't strictly needed by kmemleak but they may increase the chances of
> finding leaks. This patch also updates the KMEMLEAK config entry help
> text.
>
> Signed-off-by: Catalin Marinas <[email protected]>
Acked-by: Pekka Enberg <[email protected]>